diff --git a/.github/workflows/test_python.yml b/.github/workflows/test_python.yml index a91fb553..97419b46 100644 --- a/.github/workflows/test_python.yml +++ b/.github/workflows/test_python.yml @@ -1,30 +1,135 @@ -name: Test Python -on: [push, pull_request] +name: Python + +on: # yamllint disable-line rule:truthy + push: + pull_request: + workflow_call: + workflow_dispatch: jobs: - test: - runs-on: ${{ matrix.runner }} - strategy: - fail-fast: false - matrix: - runner: - - ubuntu-latest - - macos-latest - python-version: - - "3.10" - - "3.x" - steps: - - - uses: actions/checkout@v6 - with: - submodules: recursive - - - uses: actions/setup-python@v6 - with: - python-version: ${{ matrix.python-version }} - - - name: Install package for testing - run: python -m pip install ".[test]" - - - name: Run tests - run: pytest + + sdist: + runs-on: ubuntu-latest + steps: + + - uses: actions/checkout@v6 + + - uses: actions/setup-python@v6 + with: + python-version: 3 + pip-install: build twine + + - name: Build sdist + run: python -m build --sdist + + - name: Check sdist + run: python -m twine check dist/*.tar.gz + + - name: Upload sdist + uses: actions/upload-artifact@v5 + with: + name: sdist + path: dist/*.tar.gz + retention-days: 1 + + wheel: + runs-on: ${{ matrix.runner }} + strategy: + fail-fast: false + matrix: + runner: + - ubuntu-24.04 + - ubuntu-24.04-arm + - macos-15 + - macos-15-intel + python: + - "cp310" + - "cp311" + - "cp312" + env: + pysabi: ${{ matrix.python == 'cp312' && matrix.python || '' }} + steps: + + - name: Checkout + uses: actions/checkout@v6 + + - name: Setup Python + uses: actions/setup-python@v6 + with: + python-version: 3 + pip-install: abi3audit + + - name: Setup environment + run: | + # Reproducible Builds + echo SOURCE_DATE_EPOCH=$(git log -1 --pretty=%at) >> $GITHUB_ENV + if [ ${{ runner.os }} = macOS ]; then + echo ZERO_AR_DATE=YES >> $GITHUB_ENV; fi + # macOS Deployment Target + if [ ${{ runner.os }}-${{ runner.arch }} = macOS-X64 ]; then + echo MACOSX_DEPLOYMENT_TARGET=10.13 >> $GITHUB_ENV; fi + + - name: Build wheel + uses: pypa/cibuildwheel@v3.3.0 + with: + extras: uv + output-dir: dist + env: + CIBW_BUILD: "${{ matrix.python }}-*" + CIBW_SKIP: "*musllinux*" + CIBW_BUILD_FRONTEND: "build[uv]" + CIBW_ENVIRONMENT: >- + SKBUILD_WHEEL_PY_API=${{ env.pysabi }} + + - name: Check wheel + run: python -m abi3audit -Ss dist/*.whl + if: ${{ env.pysabi != '' }} + + - name: Upload wheel + uses: actions/upload-artifact@v5 + with: + name: wheel-${{ runner.os }}-${{ runner.arch }}-${{ matrix.python }} + path: dist/*.whl + retention-days: 1 + + test: + needs: [wheel] + runs-on: ${{ matrix.runner }} + strategy: + fail-fast: false + matrix: + runner: + - ubuntu-24.04 + - ubuntu-24.04-arm + - macos-15 + - macos-15-intel + python: + - "3.10" + - "3.11" + - "3.12" + - "3.13" + - "3.14" + - "3.x" + steps: + + - name: Checkout + uses: actions/checkout@v6 + + - name: Setup Python + uses: actions/setup-python@v6 + with: + python-version: ${{ matrix.python }} + pip-install: pytest + + - name: Download wheel + uses: actions/download-artifact@v6 + with: + path: dist + pattern: wheel-${{ runner.os }}-${{ runner.arch }}-* + merge-multiple: true + + - name: Install wheel + run: python -m pip install mutationpp --find-links=dist + + - name: Test wheel + run: pytest diff --git a/.github/workflows/wheels.yml b/.github/workflows/wheels.yml index 0113e745..aa681ac1 100644 --- a/.github/workflows/wheels.yml +++ b/.github/workflows/wheels.yml @@ -1,157 +1,72 @@ name: Python wheels -on: + +on: # yamllint disable-line rule:truthy push: tags: - 'v*.*.*' + workflow_dispatch: + inputs: + publish-pypi: + description: 'Publish to PyPI' + required: false + type: boolean + default: false jobs: - test_linux: - runs-on: ubuntu-latest - steps: - - uses: actions/checkout@v6 - with: - submodules: recursive - - - uses: actions/setup-python@v6 - with: - python-version: "3.x" - - - name: Install package for testing - run: python -m pip install ".[test]" - - - name: Run tests - run: pytest - - test_mac: - runs-on: macos-latest - steps: - - uses: actions/checkout@v6 - with: - submodules: recursive - - - uses: actions/setup-python@v6 - with: - python-version: "3.x" - - - name: Install package for testing - run: python -m pip install ".[test]" - - - name: Run tests - run: pytest - - linux_wheels: - strategy: - matrix: - python-version: - - cp310-cp310 - - cp311-cp311 - - cp312-cp312 - - cp313-cp313 - - cp314-cp314 - - needs: test_linux - runs-on: ubuntu-latest - steps: - - uses: actions/checkout@v6 - with: - submodules: recursive - - - uses: actions/setup-python@v6 - with: - python-version: 3.14 - - - name: Build manylinux Python wheels - uses: RalfG/python-wheels-manylinux-build@v0.3.4-manylinux2010_x86_64 - with: - python-versions: ${{ matrix.python-version }} - - - uses: actions/upload-artifact@v2 - with: - name: linux_wheels - path: "dist/*-manylinux*.whl" - retention-days: 1 - - mac_wheels: - strategy: - matrix: - python-version: - - "3.10" - - "3.11" - - "3.12" - - "3.13" - - "3.14" - needs: test_mac - runs-on: macos-latest - steps: - - uses: actions/checkout@v6 - with: - submodules: recursive - - - uses: actions/setup-python@v6 - with: - python-version: ${{ matrix.python-version }} - - - name: Install build - run: python -m pip install build - - - name: Build wheel - run: python -m build --whell - - - uses: actions/upload-artifact@v2 - with: - name: mac_wheels - path: "dist/*.whl" - retention-days: 1 - - sdist: - needs: test_linux - runs-on: ubuntu-latest - steps: - - uses: actions/checkout@v6 - with: - submodules: recursive - - - uses: actions/setup-python@v6 - with: - python-version: "3.x" - - - name: Install build - run: python -m pip install build - - - name: Build sdist - run: python -m build --sdist - - - uses: actions/upload-artifact@v2 - with: - name: sdist - path: "dist/*.tar.gz" - retention-days: 1 - - upload_wheels: - runs-on: ubuntu-latest - needs: [linux_wheels, mac_wheels] - steps: - - uses: actions/download-artifact@v2 - with: - name: linux_wheels - path: dist - - - uses: actions/download-artifact@v2 - with: - name: mac_wheels - path: dist - - - uses: actions/download-artifact@v2 - with: - name: sdist - path: dist - - - uses: actions/setup-python@v6 - with: - python-version: "3.x" - - - name: Install twine - run: python -m pip install twine - - name: Upload wheels - run: twine upload -u __token__ -p "${{ secrets.TESTPYPI_TOKEN }}" --repository testpypi dist/* + build: + uses: ./.github/workflows/test_python.yml + + publish-pypi: + # TODO: create GitHub environment + # environment: + # name: pypi + # url: https://pypi.org/project/mutationpp + needs: [build] + runs-on: ubuntu-latest + permissions: + contents: read + id-token: write + attestations: write + steps: + + - name: Download sdist artifact + uses: actions/download-artifact@v6 + with: + path: dist + pattern: sdist + merge-multiple: true + + - name: Download wheel artifacts + uses: actions/download-artifact@v6 + with: + path: dist + pattern: wheel-* + merge-multiple: true + + - name: Checksum artifacts + run: | + sha256sum -b * + echo '```' >> "$GITHUB_STEP_SUMMARY" + sha256sum -b * >> "$GITHUB_STEP_SUMMARY" + echo '```' >> "$GITHUB_STEP_SUMMARY" + working-directory: dist + + - if: | + inputs.publish-pypi || + github.ref_type == 'tag' + name: Attest artifacts + uses: actions/attest-build-provenance@v3 + with: + subject-path: dist/* + + - if: | + inputs.publish-pypi || + github.ref_type == 'tag' + name: Publish to PyPI + uses: pypa/gh-action-pypi-publish@release/v1 + with: + # TODO: remove next line to use the main PyPI index + repository-url: https://test.pypi.org/legacy/ + # TODO: setup Trusted Publising on PyPI, then remove next line + password: ${{ secrets.TESTPYPI_TOKEN }} diff --git a/CMakeLists.txt b/CMakeLists.txt index d4327cc7..49ea2e46 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -47,7 +47,7 @@ if (BUILD_FORTRAN_WRAPPER) endif() ####################################################################### -# Python bindings built with scikit-build-core + pybind11 # +# Python bindings built with scikit-build-core + nanobind # ####################################################################### if (SKBUILD) diff --git a/interface/python/CMakeLists.txt b/interface/python/CMakeLists.txt index 803aa71c..ac9f10ce 100644 --- a/interface/python/CMakeLists.txt +++ b/interface/python/CMakeLists.txt @@ -1,8 +1,13 @@ -find_package(Python COMPONENTS Interpreter Development.Module REQUIRED) +find_package(Python COMPONENTS Interpreter Development.Module ${SKBUILD_SABI_COMPONENT} REQUIRED) -find_package(pybind11 CONFIG REQUIRED) +find_package(nanobind CONFIG REQUIRED) -pybind11_add_module(_mutationpp +if(NOT "${SKBUILD_SABI_VERSION}" STREQUAL "") + set(STABLE_ABI STABLE_ABI) +endif() + +nanobind_add_module(_mutationpp + ${STABLE_ABI} src/mutationpp_python.cpp src/pyGlobalOptions.cpp src/pyMixtureOptions.cpp diff --git a/interface/python/src/mutationpp_python.cpp b/interface/python/src/mutationpp_python.cpp index b1f76fc0..b7253505 100644 --- a/interface/python/src/mutationpp_python.cpp +++ b/interface/python/src/mutationpp_python.cpp @@ -1,4 +1,7 @@ -#include +#include + +#define PY_MODULE NB_MODULE +namespace py = nanobind; /** * This is the module definition for the Mutation++ Python wrapper. @@ -6,13 +9,11 @@ * form minimal maintenance effort. */ -namespace py = pybind11; - -void py_export_GlobalOptions(py::module &); -void py_export_MixtureOptions(py::module &); -void py_export_Mixture(py::module &); +void py_export_GlobalOptions(py::module_ &); +void py_export_MixtureOptions(py::module_ &); +void py_export_Mixture(py::module_ &); -PYBIND11_MODULE(_mutationpp, m) { +PY_MODULE(_mutationpp, m) { m.doc() = "Mutation++ Python bindings"; py_export_GlobalOptions(m); py_export_MixtureOptions(m); diff --git a/interface/python/src/pyGlobalOptions.cpp b/interface/python/src/pyGlobalOptions.cpp index dfba9f14..73888004 100644 --- a/interface/python/src/pyGlobalOptions.cpp +++ b/interface/python/src/pyGlobalOptions.cpp @@ -1,7 +1,8 @@ #include -#include +#include +#include -namespace py = pybind11; +namespace py = nanobind; /** * Python wrapper definition for the GlobalOptions class. @@ -17,13 +18,13 @@ namespace { } -void py_export_GlobalOptions(py::module &m) { +void py_export_GlobalOptions(py::module_ &m) { - const py::object resources = py::module_::import("importlib.resources"); + const py::object resources = py::module_::import_("importlib.resources"); const py::object pkgname = m.attr("__spec__").attr("parent"); const py::object rootdir = resources.attr("files")(pkgname); const py::object datadir = rootdir / py::str("data"); - const std::string data_directory = py::str(datadir).cast(); + const std::string data_directory = py::cast(py::str(datadir)); setDefaultDataDirectory(data_directory); diff --git a/interface/python/src/pyMixture.cpp b/interface/python/src/pyMixture.cpp index d3e83824..808f421b 100644 --- a/interface/python/src/pyMixture.cpp +++ b/interface/python/src/pyMixture.cpp @@ -1,13 +1,13 @@ #include #include -#include -#include -#include -#include -namespace py = pybind11; +#include +#include +#include +#include +#include -// Todo return the values with py::array_t so it returns as numpy type +namespace py = nanobind; /** * Python wrapper definition for the Mixture class. All member @@ -16,53 +16,53 @@ namespace py = pybind11; * ones which do not are kept unexposed. */ -void py_export_Mixture(py::module &m) { +void py_export_Mixture(py::module_ &m) { py::class_(m, "Mixture") .def(py::init()) .def(py::init()) - .def("NA", + .def("NA", [](const Mutation::Mixture &self) { return Mutation::NA; }, "Returns the Avogadro's number (molecule/mol)." ) - .def("KB", + .def("KB", [](const Mutation::Mixture &self) { return Mutation::KB; - }, + }, "Returns the Boltzmann's constant (J/molecule-K)") - .def("RU", + .def("RU", [](const Mutation::Mixture &self) { return Mutation::RU; }, - "Returns the Universal Gas constant (J/mole-K)") + "Returns the Universal Gas constant (J/mole-K)") - .def("HP", + .def("HP", [](const Mutation::Mixture &self) { return Mutation::HP; }, "Returns the Planck's constant (J-s)") - - .def("C0", + + .def("C0", [](const Mutation::Mixture &self) { return Mutation::C0; }, "Returns the Speed of light in vacuum (m/s)") - .def("ONEATM", + .def("ONEATM", [](const Mutation::Mixture &self) { return Mutation::ONEATM; }, - "Returns the 1 atm in Pa") + "Returns the 1 atm in Pa") - .def("SB", + .def("SB", [](const Mutation::Mixture &self) { return Mutation::SB; }, - "Returns the Stefan-Boltzmann constant (W/m^2-K^4)") + "Returns the Stefan-Boltzmann constant (W/m^2-K^4)") .def("nElements", &Mutation::Mixture::nElements, "Returns the number of elements considered in the mixture.") @@ -227,7 +227,7 @@ void py_export_Mixture(py::module &m) { std::vector x_e(self.nElements()); self.getComposition(mixture_composition, x_e.data(), Mutation::Thermodynamics::Composition::MASS); - return py::array(py::cast(x_e)); + return x_e; }, "Gets the element mass fractions associated with a named composition " "in the mixture.") @@ -238,7 +238,7 @@ void py_export_Mixture(py::module &m) { std::vector x_e(self.nElements()); self.getComposition(mixture_composition, x_e.data(), Mutation::Thermodynamics::Composition::MOLE); - return py::array(py::cast(x_e)); + return x_e; }, "Gets the element mole fractions associated with a named composition " "in the mixture.") @@ -282,7 +282,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector rho_i(self.nSpecies()); self.densities(rho_i.data()); - return py::array(py::cast(rho_i)); + return rho_i; }, "Returns the current species densities.") @@ -291,7 +291,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector cp_i(self.nSpecies()); self.speciesCpOverR(cp_i.data()); - return py::array(py::cast(cp_i)); + return cp_i; }, "Returns the unitless vector of species specific heats at constant " "pressure \f$ C_{p,i} / R_u \f$.") @@ -301,7 +301,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self, const double T) { std::vector cp_i(self.nSpecies()); self.speciesCpOverR(T, cp_i.data()); - return py::array(py::cast(cp_i)); + return cp_i; }, "Returns the unitless vector of species specific heats at constant " "pressure " @@ -356,7 +356,7 @@ void py_export_Mixture(py::module &m) { "X", [](const Mutation::Mixture &self) { std::vector x(self.X(), self.X() + self.nSpecies()); - return py::array(py::cast(x)); + return x; }, "Returns the current species mole fractions.") @@ -364,7 +364,7 @@ void py_export_Mixture(py::module &m) { "Y", [](const Mutation::Mixture &self) { std::vector y(self.Y(), self.Y() + self.nSpecies()); - return py::array(py::cast(y)); + return y; }, "Returns the current species mass fractions.") @@ -397,7 +397,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self, double T, double P) { std::vector x(self.nSpecies()); self.equilibriumComposition(T, P, x.data()); - return py::array(py::cast(x)); + return x; }, "Computes the equilibrium composition of the mixture at the given " "fixed" @@ -409,7 +409,7 @@ void py_export_Mixture(py::module &m) { std::vector xe) { std::vector x(self.nSpecies()); self.equilibriumComposition(T, P, xe.data(), x.data()); - return py::array(py::cast(x)); + return x; }, "Computes the equilibrium composition of the mixture at the given " "fixed" @@ -429,7 +429,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector h_i(self.nSpecies()); self.speciesHOverRT(h_i.data()); - return py::array(py::cast(h_i)); + return h_i; }, "Computes the unitless species enthalpies and can optionally fill vectors" "for each energy mode.") @@ -439,7 +439,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self, double T) { std::vector h_i(self.nSpecies()); self.speciesHOverRT(T, h_i.data()); - return py::array(py::cast(h_i)); + return h_i; }, "Returns the unitless vector of species enthalpies \f$ H_i / R_u T " "\f$.") @@ -488,7 +488,7 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nSpecies()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::RHO_TO_Y>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, "Converts species densities to mass fraction.") @@ -498,7 +498,7 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nSpecies()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::RHO_TO_X>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, "Converts species densities to mole fraction.") @@ -508,7 +508,7 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nSpecies()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::Y_TO_X>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, "Converts species mass to mole fraction.") @@ -518,7 +518,7 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nSpecies()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::X_TO_Y>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, "Converts species mole to mass fraction.") @@ -528,7 +528,7 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nElements()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::X_TO_XE>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, "Converts species mole to elemental mole fraction.") @@ -538,7 +538,7 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nElements()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::Y_TO_YE>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, "Converts species mass to elemental mass fraction.") @@ -548,7 +548,7 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nElements()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::XE_TO_YE>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, "Converts elemental mole fraction to elemental mass fractions.") @@ -558,9 +558,9 @@ void py_export_Mixture(py::module &m) { std::vector y(self.nElements()); self.Mutation::Mixture::Thermodynamics::convert< Mutation::Thermodynamics::YE_TO_XE>(x.data(), y.data()); - return py::array(py::cast(y)); + return y; }, - "Converts elemental mass fraction to elemental mole fractions.") + "Converts elemental mass fraction to elemental mole fractions.") .def("dRhodP", &Mutation::Mixture::dRhodP, "Returns the density derivative with respect to pressure for the " @@ -571,7 +571,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector value(self.nSpecies()); self.Mutation::Mixture::dXidT(value.data()); - return py::array(py::cast(value)); + return value; }, "Returns the derivative of mole fraction w.t.r temperature for the " "given equilibrium mixture." @@ -582,7 +582,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector value(self.nSpecies()); self.Mutation::Mixture::dXidP(value.data()); - return py::array(py::cast(value)); + return value; }, "Returns the species derivatives of mole fraction w.r.t. pressure " "for the" @@ -598,7 +598,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector value(self.nReactions()); self.Mutation::Mixture::forwardRateCoefficients(value.data()); - return py::array(py::cast(value)); + return value; }, "Fills the vector kf with the forward rate coefficients \f$ k_{f,j} " "\f$" @@ -611,7 +611,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector value(self.nReactions()); self.Mutation::Mixture::forwardRatesOfProgress(value.data()); - return py::array(py::cast(value)); + return value; }, "Fills the vector ropf with the forward rate of progress variables " "for each reaction.") @@ -621,7 +621,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector value(self.nReactions()); self.Mutation::Mixture::backwardRateCoefficients(value.data()); - return py::array(py::cast(value)); + return value; }, "Fills the vector kb with the backward rate coefficients for each " "reaction.") @@ -631,7 +631,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector value(self.nReactions()); self.Mutation::Mixture::backwardRatesOfProgress(value.data()); - return py::array(py::cast(value)); + return value; }, "Fills the vector ropb with the backward rates of progress variables " "for each reaction.") @@ -641,7 +641,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector value(self.nReactions()); self.Mutation::Mixture::netRatesOfProgress(value.data()); - return py::array(py::cast(value)); + return value; }, "Fills the vector rop with the net rates of progress for each " "reaction") @@ -651,7 +651,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector value(self.nSpecies()); self.Mutation::Mixture::netProductionRates(value.data()); - return py::array(py::cast(value)); + return value; }, "Fills the vector wdot with the net species production rates due to " "the chemical reactions.") @@ -683,7 +683,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector lambda_i(self.nEnergyEqns()); self.frozenThermalConductivityVector(lambda_i.data()); - return py::array(py::cast(lambda_i)); + return lambda_i; }, "Returns the mixture thermal conductivity vector for a frozen " "mixture according to the state model.") @@ -743,7 +743,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector diff_i(self.nSpecies()); self.heavyThermalDiffusionRatios(diff_i.data()); - return py::array(py::cast(diff_i)); + return diff_i; }, "Returns the heavy thermal diffusion ratios for each species.") @@ -800,7 +800,7 @@ void py_export_Mixture(py::module &m) { self.stefanMaxwell(grad_x.data(), diff_velocities.data(), electricField); - return py::make_tuple(py::array(py::cast(diff_velocities)), electricField); + return py::make_tuple(diff_velocities, electricField); }, "Computes the species diffusion velocities and ambipolar electric " "field" @@ -812,7 +812,7 @@ void py_export_Mixture(py::module &m) { [](Mutation::Mixture &self) { std::vector array(self.nSpecies()); self.averageDiffusionCoeffs(array.data()); - return py::array(py::cast(array)); + return array; }, "Returns the average diffusion coefficients.") @@ -821,17 +821,17 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector T_i(self.nEnergyEqns()); self.getTemperatures(T_i.data()); - return py::array(py::cast(T_i)); + return T_i; }, "Fills temperature array with tempertures according to the used " "StateModel.") - + .def( "getEnergiesMass", [](const Mutation::Mixture &self) { std::vector e_i(self.nSpecies()); self.Mutation::Mixture::getEnergiesMass(e_i.data()); - return py::array(py::cast(e_i)); + return e_i; }, "Fills energy per mass array with energies according to the used " "StateModel (total + internal for each species).") @@ -841,7 +841,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector h_i(self.nSpecies()); self.Mutation::Mixture::getEnthalpiesMass(h_i.data()); - return py::array(py::cast(h_i)); + return h_i; }, "Fills enthalpy per mass array with enthalpy according to the used " "StateModel (total + internal for each species).") @@ -851,7 +851,7 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector cp_i(self.nSpecies()); self.Mutation::Mixture::getCpsMass(cp_i.data()); - return py::array(py::cast(cp_i)); + return cp_i; }, "Fills the constant pressure specific heat according to the used " "StateModel") @@ -861,41 +861,41 @@ void py_export_Mixture(py::module &m) { [](const Mutation::Mixture &self) { std::vector cv_i(self.nSpecies()); self.Mutation::Mixture::getCvsMass(cv_i.data()); - return py::array(py::cast(cv_i)); + return cv_i; }, "Fills the constant volume specific heat according to the used " - "StateModel") + "StateModel") + - .def( "getGibbsMass", [](const Mutation::Mixture &self) { Eigen::ArrayXd gibbs_i(self.nSpecies()); self.Mutation::Mixture::speciesGOverRT(gibbs_i.data()); - gibbs_i*=self.T()* Mutation::RU/self.speciesMw(); - return py::array(py::cast(gibbs_i));; + gibbs_i *= self.T() * Mutation::RU/self.speciesMw(); + return gibbs_i; }, - "Return an array of gibbs free energy for each species per unit mass") + "Return an array of gibbs free energy for each species per unit mass") .def( "getGibbsMass", [](const Mutation::Mixture &self, double T, double P) { Eigen::ArrayXd gibbs_i(self.nSpecies()); self.Mutation::Mixture::speciesGOverRT(T, P, gibbs_i.data()); - gibbs_i*=T* Mutation::RU/self.speciesMw(); - return py::array(py::cast(gibbs_i));; + gibbs_i *= T * Mutation::RU/self.speciesMw(); + return gibbs_i; }, - "Return an array of gibbs free energy for each species per unit mass") + "Return an array of gibbs free energy for each species per unit mass") .def( "getSTGibbsMass", [](const Mutation::Mixture &self, double T) { Eigen::ArrayXd gibbs_i(self.nSpecies()); self.Mutation::Mixture::speciesSTGOverRT(T, gibbs_i.data()); - gibbs_i*=T* Mutation::RU/self.speciesMw(); - return py::array(py::cast(gibbs_i));; + gibbs_i *= T * Mutation::RU/self.speciesMw(); + return gibbs_i; }, - "Return an array of gibbs free energy for each species per unit mass at standard pressure") - + "Return an array of gibbs free energy for each species per unit mass at standard pressure") + ; -} \ No newline at end of file +} diff --git a/interface/python/src/pyMixtureOptions.cpp b/interface/python/src/pyMixtureOptions.cpp index 8eb32476..24ba66b0 100644 --- a/interface/python/src/pyMixtureOptions.cpp +++ b/interface/python/src/pyMixtureOptions.cpp @@ -1,7 +1,9 @@ #include -#include -namespace py = pybind11; +#include +#include + +namespace py = nanobind; /** * Python wrapper definition for the MixtureOptions class. All member @@ -10,7 +12,7 @@ namespace py = pybind11; * ones which do not are kept unexposed. */ -void py_export_MixtureOptions(py::module &m) { +void py_export_MixtureOptions(py::module_ &m) { /** * Overloaded member functions wrappers */ @@ -49,4 +51,4 @@ void py_export_MixtureOptions(py::module &m) { &Mutation::MixtureOptions::setDefaultComposition) .def("hasDefaultComposition", &Mutation::MixtureOptions::hasDefaultComposition); -} \ No newline at end of file +} diff --git a/pyproject.toml b/pyproject.toml index 9f0141e8..61c33e5a 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -33,7 +33,7 @@ Downloads = "https://github.com/mutationpp/Mutationpp/releases" test = ["pytest"] [build-system] -requires = ["scikit-build-core>=0.11", "pybind11"] +requires = ["scikit-build-core>=0.11", "nanobind"] build-backend = "scikit_build_core.build" [tool.scikit-build] diff --git a/thirdparty/eigen/.clang-format b/thirdparty/eigen/.clang-format new file mode 100644 index 00000000..1f33cbed --- /dev/null +++ b/thirdparty/eigen/.clang-format @@ -0,0 +1,19 @@ +--- +BasedOnStyle: Google +ColumnLimit: 120 +--- +Language: Cpp +BasedOnStyle: Google +ColumnLimit: 120 +StatementMacros: + - EIGEN_STATIC_ASSERT + - EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED + - EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN +SortIncludes: false +AttributeMacros: +- EIGEN_STRONG_INLINE +- EIGEN_ALWAYS_INLINE +- EIGEN_DEVICE_FUNC +- EIGEN_DONT_INLINE +- EIGEN_DEPRECATED +- EIGEN_UNUSED diff --git a/thirdparty/eigen/.hgignore b/thirdparty/eigen/.gitignore similarity index 55% rename from thirdparty/eigen/.hgignore rename to thirdparty/eigen/.gitignore index e33ba2e9..19dfac9e 100644 --- a/thirdparty/eigen/.hgignore +++ b/thirdparty/eigen/.gitignore @@ -1,4 +1,3 @@ -syntax: glob qrc_*cxx *.orig *.pyc @@ -13,7 +12,7 @@ core core.* *.bak *~ -build* +*build* *.moc.* *.moc ui_* @@ -28,5 +27,15 @@ activity.png *.rej log patch +*.patch a a.* +lapack/testing +lapack/reference +.*project +.settings +Makefile +!ci/build.gitlab-ci.yml +!scripts/buildtests.in +!Eigen/Core +!Eigen/src/Core diff --git a/thirdparty/eigen/.gitlab-ci.yml b/thirdparty/eigen/.gitlab-ci.yml new file mode 100644 index 00000000..68cd6804 --- /dev/null +++ b/thirdparty/eigen/.gitlab-ci.yml @@ -0,0 +1,34 @@ +# This file is part of Eigen, a lightweight C++ template library +# for linear algebra. +# +# Copyright (C) 2023, The Eigen Authors +# +# This Source Code Form is subject to the terms of the Mozilla +# Public License v. 2.0. If a copy of the MPL was not distributed +# with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +stages: + - checkformat + - build + - test + - deploy + +variables: + # CMake build directory. + EIGEN_CI_BUILDDIR: .build + # Specify the CMake build target. + EIGEN_CI_BUILD_TARGET: "" + # If a test regex is specified, that will be selected. + # Otherwise, we will try a label if specified. + EIGEN_CI_CTEST_REGEX: "" + EIGEN_CI_CTEST_LABEL: "" + EIGEN_CI_CTEST_ARGS: "" + +include: + - "/ci/checkformat.gitlab-ci.yml" + - "/ci/common.gitlab-ci.yml" + - "/ci/build.linux.gitlab-ci.yml" + - "/ci/build.windows.gitlab-ci.yml" + - "/ci/test.linux.gitlab-ci.yml" + - "/ci/test.windows.gitlab-ci.yml" + - "/ci/deploy.gitlab-ci.yml" diff --git a/thirdparty/eigen/.gitlab/issue_templates/Bug Report.md b/thirdparty/eigen/.gitlab/issue_templates/Bug Report.md new file mode 100644 index 00000000..0c49b0fe --- /dev/null +++ b/thirdparty/eigen/.gitlab/issue_templates/Bug Report.md @@ -0,0 +1,69 @@ + + +### Summary + + +### Environment + +- **Operating System** : Windows/Linux +- **Architecture** : x64/Arm64/PowerPC ... +- **Eigen Version** : 3.3.9 +- **Compiler Version** : Gcc7.0 +- **Compile Flags** : -O3 -march=native +- **Vector Extension** : SSE/AVX/NEON ... + +### Minimal Example + + +```cpp +//show your code here +``` + +### Steps to reproduce + + +1. first step +2. second step +3. ... + +### What is the current *bug* behavior? + + +### What is the expected *correct* behavior? + + +### Relevant logs + + + +### Warning Messages + + + +### Benchmark scripts and results + + +### Anything else that might help + + +- [ ] Have a plan to fix this issue. diff --git a/thirdparty/eigen/.gitlab/issue_templates/Feature Request.md b/thirdparty/eigen/.gitlab/issue_templates/Feature Request.md new file mode 100644 index 00000000..2c6f908a --- /dev/null +++ b/thirdparty/eigen/.gitlab/issue_templates/Feature Request.md @@ -0,0 +1,7 @@ +### Describe the feature you would like to be implemented. + +### Would such a feature be useful for other users? Why? + +### Any hints on how to implement the requested feature? + +### Additional resources diff --git a/thirdparty/eigen/.gitlab/merge_request_templates/Merge Request Template.md b/thirdparty/eigen/.gitlab/merge_request_templates/Merge Request Template.md new file mode 100644 index 00000000..3fe963af --- /dev/null +++ b/thirdparty/eigen/.gitlab/merge_request_templates/Merge Request Template.md @@ -0,0 +1,26 @@ + + +### Reference issue + + +### What does this implement/fix? + + +### Additional information + diff --git a/thirdparty/eigen/.gitrepo b/thirdparty/eigen/.gitrepo index 0b631af3..da05f8a8 100644 --- a/thirdparty/eigen/.gitrepo +++ b/thirdparty/eigen/.gitrepo @@ -4,8 +4,9 @@ ; git-subrepo command. See https://github.com/git-commands/git-subrepo#readme ; [subrepo] - remote = https://github.com/RLovelett/eigen.git - branch = 3.2.10 - commit = c50d586b9a43395a83e51821ebd1759d1c4668e4 - parent = 849bc10baa20b04cc360e81c8331da5c815f277f - cmdver = 0.3.1 + remote = https://gitlab.com/libeigen/eigen.git + branch = 3.4 + commit = 28ded8800c26864e537852658428ab44c8399e87 + parent = 9d7a3ce9369d417147e791151a106706e289da72 + cmdver = 0.4.9 + method = merge diff --git a/thirdparty/eigen/.hgeol b/thirdparty/eigen/.hgeol index 423676d3..5327df16 100644 --- a/thirdparty/eigen/.hgeol +++ b/thirdparty/eigen/.hgeol @@ -1,6 +1,9 @@ [patterns] +*.sh = LF +*.MINPACK = CRLF scripts/*.in = LF debug/msvc/*.dat = CRLF +debug/msvc/*.natvis = CRLF unsupported/test/mpreal/*.* = CRLF ** = native diff --git a/thirdparty/eigen/.hgtags b/thirdparty/eigen/.hgtags deleted file mode 100644 index c8312857..00000000 --- a/thirdparty/eigen/.hgtags +++ /dev/null @@ -1,35 +0,0 @@ -2db9468678c6480c9633b6272ff0e3599d1e11a3 2.0-beta3 -375224817dce669b6fa31d920d4c895a63fabf32 2.0-beta1 -3b8120f077865e2a072e10f5be33e1d942b83a06 2.0-rc1 -19dfc0e7666bcee26f7a49eb42f39a0280a3485e 2.0-beta5 -7a7d8a9526f003ffa2430dfb0c2c535b5add3023 2.0-beta4 -7d14ad088ac23769c349518762704f0257f6a39b 2.0.1 -b9d48561579fd7d4c05b2aa42235dc9de6484bf2 2.0-beta6 -e17630a40408243cb1a51ad0fe3a99beb75b7450 before-hg-migration -eda654d4cda2210ce80719addcf854773e6dec5a 2.0.0 -ee9a7c468a9e73fab12f38f02bac24b07f29ed71 2.0-beta2 -d49097c25d8049e730c254a2fed725a240ce4858 after-hg-migration -655348878731bcb5d9bbe0854077b052e75e5237 actual-start-from-scratch -12a658962d4e6dfdc9a1c350fe7b69e36e70675c 3.0-beta1 -5c4180ad827b3f869b13b1d82f5a6ce617d6fcee 3.0-beta2 -7ae24ca6f3891d5ac58ddc7db60ad413c8d6ec35 3.0-beta3 -c40708b9088d622567fecc9208ad4a426621d364 3.0-beta4 -b6456624eae74f49ae8683d8e7b2882a2ca0342a 3.0-rc1 -a810d5dbab47acfe65b3350236efdd98f67d4d8a 3.1.0-alpha1 -304c88ca3affc16dd0b008b1104873986edd77af 3.1.0-alpha2 -920fc730b5930daae0a6dbe296d60ce2e3808215 3.1.0-beta1 -8383e883ebcc6f14695ff0b5e20bb631abab43fb 3.1.0-rc1 -bf4cb8c934fa3a79f45f1e629610f0225e93e493 3.1.0-rc2 -da195914abcc1d739027cbee7c52077aab30b336 3.2-beta1 -4b687cad1d23066f66863f4f87298447298443df 3.2-rc1 -1eeda7b1258bcd306018c0738e2b6a8543661141 3.2-rc2 -ffa86ffb557094721ca71dcea6aed2651b9fd610 3.2.0 -6b38706d90a9fe182e66ab88477b3dbde34b9f66 3.2.1 -1306d75b4a21891e59ff9bd96678882cf831e39f 3.2.2 -36fd1ba04c120cfdd90f3e4cede47f43b21d19ad 3.2.3 -10219c95fe653d4962aa9db4946f6fbea96dd740 3.2.4 -bdd17ee3b1b3a166cd5ec36dcad4fc1f3faf774a 3.2.5 -c58038c56923e0fd86de3ded18e03df442e66dfb 3.2.6 -b30b87236a1b1552af32ac34075ee5696a9b5a33 3.2.7 -07105f7124f9aef00a68c85e0fc606e65d3d6c15 3.2.8 -dc6cfdf9bcec5efc7b6593bddbbb3d675de53524 3.2.9 diff --git a/thirdparty/eigen/CMakeLists.txt b/thirdparty/eigen/CMakeLists.txt index 77e9f2d3..94be8721 100644 --- a/thirdparty/eigen/CMakeLists.txt +++ b/thirdparty/eigen/CMakeLists.txt @@ -1,31 +1,101 @@ -project(Eigen) -cmake_minimum_required(VERSION 2.8.5) +cmake_minimum_required(VERSION 3.10.0) + +#============================================================================== +# CMake Policy issues. +#============================================================================== +# Allow overriding options in a parent project via `set` before including Eigen. +if (POLICY CMP0077) + cmake_policy (SET CMP0077 NEW) +endif (POLICY CMP0077) + +# NOTE Remove setting the policy once the minimum required CMake version is +# increased to at least 3.15. Retain enabling the export to package registry. +if (POLICY CMP0090) + # The export command does not populate package registry by default + cmake_policy (SET CMP0090 NEW) + # Unless otherwise specified, always export to package registry to ensure + # backwards compatibility. + if (NOT DEFINED CMAKE_EXPORT_PACKAGE_REGISTRY) + set (CMAKE_EXPORT_PACKAGE_REGISTRY ON) + endif (NOT DEFINED CMAKE_EXPORT_PACKAGE_REGISTRY) +endif (POLICY CMP0090) + +# Disable warning about find_package(CUDA). +# CUDA language support is lacking for clang as the CUDA compiler +# until at least cmake version 3.18. Even then, there seems to be +# issues on Windows+Ninja in passing build flags. Continue using +# the "old" way for now. +if (POLICY CMP0146) + cmake_policy(SET CMP0146 OLD) +endif () + +#============================================================================== +# CMake Project. +#============================================================================== + +project(Eigen3) + +# Remove this block after bumping CMake to v3.21.0 +# PROJECT_IS_TOP_LEVEL is defined then by default +if(CMAKE_VERSION VERSION_LESS 3.21.0) + if(CMAKE_SOURCE_DIR STREQUAL CMAKE_CURRENT_SOURCE_DIR) + set(PROJECT_IS_TOP_LEVEL ON) + else() + set(PROJECT_IS_TOP_LEVEL OFF) + endif() +endif() -# guard against in-source builds +#============================================================================== +# Build ON/OFF Settings. +#============================================================================== +# Determine if we should build tests. +include(CMakeDependentOption) +cmake_dependent_option(BUILD_TESTING "Enable creation of tests." ON "PROJECT_IS_TOP_LEVEL" OFF) +option(EIGEN_BUILD_TESTING "Enable creation of Eigen tests." ${BUILD_TESTING}) +option(EIGEN_LEAVE_TEST_IN_ALL_TARGET "Leaves tests in the all target, needed by ctest for automatic building." OFF) + +# Determine if we should build BLAS/LAPACK implementations. +option(EIGEN_BUILD_BLAS "Toggles the building of the Eigen Blas library" ${PROJECT_IS_TOP_LEVEL}) +option(EIGEN_BUILD_LAPACK "Toggles the building of the included Eigen LAPACK library" ${PROJECT_IS_TOP_LEVEL}) +if (EIGEN_BUILD_BLAS OR EIGEN_BUILD_LAPACK) + # BLAS and LAPACK currently need a fortran compiler. + include(CMakeDetermineFortranCompiler) + if (NOT CMAKE_Fortran_COMPILER) + set(EIGEN_BUILD_BLAS OFF) + set(EIGEN_BUILD_LAPACK OFF) + else() + # Determine if we should build shared libraries for BLAS/LAPACK on this platform. + get_cmake_property(EIGEN_BUILD_SHARED_LIBS TARGET_SUPPORTS_SHARED_LIBS) + endif() +endif() -if(${CMAKE_SOURCE_DIR} STREQUAL ${CMAKE_BINARY_DIR}) - message(FATAL_ERROR "In-source builds not allowed. Please make a new directory (called a build directory) and run CMake from there. You may need to remove CMakeCache.txt. ") +option(EIGEN_BUILD_BTL "Build benchmark suite" OFF) +option(EIGEN_BUILD_SPBENCH "Build sparse benchmark suite" OFF) +# Avoid building docs if included from another project. +# Building documentation requires creating and running executables on the host +# platform. We shouldn't do this if cross-compiling. +if (PROJECT_IS_TOP_LEVEL AND NOT CMAKE_CROSSCOMPILING) + set(EIGEN_BUILD_DOC_DEFAULT ON) endif() +option(EIGEN_BUILD_DOC "Enable creation of Eigen documentation" ${EIGEN_BUILD_DOC_DEFAULT}) -# guard against bad build-type strings +option(EIGEN_BUILD_DEMOS "Toggles the building of the Eigen demos" ${PROJECT_IS_TOP_LEVEL}) -if (NOT CMAKE_BUILD_TYPE) - set(CMAKE_BUILD_TYPE "Release") +# Disable pkgconfig only for native Windows builds +if(NOT WIN32 OR NOT CMAKE_HOST_SYSTEM_NAME MATCHES Windows) + option(EIGEN_BUILD_PKGCONFIG "Build pkg-config .pc file for Eigen" ${PROJECT_IS_TOP_LEVEL}) endif() +option(EIGEN_BUILD_CMAKE_PACKAGE "Enables the creation of EigenConfig.cmake and related files" ${PROJECT_IS_TOP_LEVEL}) -string(TOLOWER "${CMAKE_BUILD_TYPE}" cmake_build_type_tolower) -if( NOT cmake_build_type_tolower STREQUAL "debug" - AND NOT cmake_build_type_tolower STREQUAL "release" - AND NOT cmake_build_type_tolower STREQUAL "relwithdebinfo") - message(FATAL_ERROR "Unknown build type \"${CMAKE_BUILD_TYPE}\". Allowed values are Debug, Release, RelWithDebInfo (case-insensitive).") +if (EIGEN_BUILD_TESTING OR EIGEN_BUILD_BLAS OR EIGEN_BUILD_LAPACK OR EIGEN_BUILT_BTL OR EIGEN_BUILD_BTL OR EIGEN_BUILD_SPBENCH OR EIGEN_BUILD_DOC OR EIGEN_BUILD_DEMOS) + set(EIGEN_IS_BUILDING_ ON) endif() +#============================================================================== +# Version Info. +#============================================================================== -############################################################################# -# retrieve version infomation # -############################################################################# - -# automatically parse the version number +# Automatically parse the version number from header files. file(READ "${PROJECT_SOURCE_DIR}/Eigen/src/Core/util/Macros.h" _eigen_version_header) string(REGEX MATCH "define[ \t]+EIGEN_WORLD_VERSION[ \t]+([0-9]+)" _eigen_world_version_match "${_eigen_version_header}") set(EIGEN_WORLD_VERSION "${CMAKE_MATCH_1}") @@ -35,378 +105,666 @@ string(REGEX MATCH "define[ \t]+EIGEN_MINOR_VERSION[ \t]+([0-9]+)" _eigen_minor_ set(EIGEN_MINOR_VERSION "${CMAKE_MATCH_1}") set(EIGEN_VERSION_NUMBER ${EIGEN_WORLD_VERSION}.${EIGEN_MAJOR_VERSION}.${EIGEN_MINOR_VERSION}) -# if the mercurial program is absent, this will leave the EIGEN_HG_CHANGESET string empty, -# but won't stop CMake. -execute_process(COMMAND hg tip -R ${CMAKE_SOURCE_DIR} OUTPUT_VARIABLE EIGEN_HGTIP_OUTPUT) -execute_process(COMMAND hg branch -R ${CMAKE_SOURCE_DIR} OUTPUT_VARIABLE EIGEN_BRANCH_OUTPUT) +# If we are in a git repo, extract a changeset. +if(IS_DIRECTORY ${CMAKE_SOURCE_DIR}/.git) + # if the git program is absent or this will leave the EIGEN_GIT_REVNUM string empty, + # but won't stop CMake. + execute_process(COMMAND git ls-remote -q ${CMAKE_SOURCE_DIR} HEAD OUTPUT_VARIABLE EIGEN_GIT_OUTPUT) +endif() -# if this is the default (aka development) branch, extract the mercurial changeset number from the hg tip output... -if(EIGEN_BRANCH_OUTPUT MATCHES "default") -string(REGEX MATCH "^changeset: *[0-9]*:([0-9;a-f]+).*" EIGEN_HG_CHANGESET_MATCH "${EIGEN_HGTIP_OUTPUT}") -set(EIGEN_HG_CHANGESET "${CMAKE_MATCH_1}") -endif(EIGEN_BRANCH_OUTPUT MATCHES "default") +# extract the git rev number from the git output... +if(EIGEN_GIT_OUTPUT) +string(REGEX MATCH "^([0-9;a-f]+).*" EIGEN_GIT_CHANGESET_MATCH "${EIGEN_GIT_OUTPUT}") +set(EIGEN_GIT_REVNUM "${CMAKE_MATCH_1}") +endif() #...and show it next to the version number -if(EIGEN_HG_CHANGESET) - set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER} (mercurial changeset ${EIGEN_HG_CHANGESET})") -else(EIGEN_HG_CHANGESET) +if(EIGEN_GIT_REVNUM) + set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER} (git rev ${EIGEN_GIT_REVNUM})") +else() set(EIGEN_VERSION "${EIGEN_VERSION_NUMBER}") -endif(EIGEN_HG_CHANGESET) - +endif() -include(CheckCXXCompilerFlag) -include(GNUInstallDirs) +#============================================================================== +# Install Path Configuration. +#============================================================================== -set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake) +# Unconditionally allow install of targets to support nested dependency +# installations. +# +# Note: projects that depend on Eigen should _probably_ exclude installing +# Eigen by default (e.g. by using EXCLUDE_FROM_ALL when using +# FetchContent_Declare or add_subdirectory) to avoid overwriting a previous +# installation. -############################################################################# -# find how to link to the standard libraries # -############################################################################# +include(GNUInstallDirs) +# Backward compatibility support for EIGEN_INCLUDE_INSTALL_DIR +if(EIGEN_INCLUDE_INSTALL_DIR) + message(WARNING "EIGEN_INCLUDE_INSTALL_DIR is deprecated. Use INCLUDE_INSTALL_DIR instead.") +endif() -find_package(StandardMathLibrary) +if(EIGEN_INCLUDE_INSTALL_DIR AND NOT INCLUDE_INSTALL_DIR) + set(INCLUDE_INSTALL_DIR ${EIGEN_INCLUDE_INSTALL_DIR} + CACHE PATH "The directory relative to CMAKE_INSTALL_PREFIX where Eigen header files are installed") +else() + set(INCLUDE_INSTALL_DIR + "${CMAKE_INSTALL_INCLUDEDIR}/eigen3" + CACHE PATH "The directory relative to CMAKE_INSTALL_PREFIX where Eigen header files are installed" + ) +endif() +set(CMAKEPACKAGE_INSTALL_DIR + "${CMAKE_INSTALL_DATADIR}/eigen3/cmake" + CACHE PATH "The directory relative to CMAKE_INSTALL_PREFIX where Eigen3Config.cmake is installed" + ) +set(PKGCONFIG_INSTALL_DIR + "${CMAKE_INSTALL_DATADIR}/pkgconfig" + CACHE PATH "The directory relative to CMAKE_INSTALL_PREFIX where eigen3.pc is installed" + ) +foreach(var INCLUDE_INSTALL_DIR CMAKEPACKAGE_INSTALL_DIR PKGCONFIG_INSTALL_DIR) + # If an absolute path is specified, make it relative to "{CMAKE_INSTALL_PREFIX}". + if(IS_ABSOLUTE "${${var}}") + file(RELATIVE_PATH "${var}" "${CMAKE_INSTALL_PREFIX}" "${${var}}") + endif() +endforeach() + +#============================================================================== +# Eigen Library. +#============================================================================== + +set ( EIGEN_VERSION_STRING ${EIGEN_VERSION_NUMBER} ) +set ( EIGEN_VERSION_MAJOR ${EIGEN_WORLD_VERSION} ) +set ( EIGEN_VERSION_MINOR ${EIGEN_MAJOR_VERSION} ) +set ( EIGEN_VERSION_PATCH ${EIGEN_MINOR_VERSION} ) +set ( EIGEN_DEFINITIONS "") +set ( EIGEN_INCLUDE_DIR "${CMAKE_INSTALL_PREFIX}/${INCLUDE_INSTALL_DIR}" ) +set ( EIGEN_ROOT_DIR ${CMAKE_INSTALL_PREFIX} ) + +# Alias Eigen_*_DIR to Eigen3_*_DIR: +set(Eigen_SOURCE_DIR ${Eigen3_SOURCE_DIR}) +set(Eigen_BINARY_DIR ${Eigen3_BINARY_DIR}) + +# Imported target support +add_library (eigen INTERFACE) +add_library (Eigen3::Eigen ALIAS eigen) +target_include_directories (eigen INTERFACE + $ + $ +) + +# Export as title case Eigen +set_target_properties (eigen PROPERTIES EXPORT_NAME Eigen) + +#============================================================================== +# Install Rule Configuration. +#============================================================================== -set(EIGEN_TEST_CUSTOM_LINKER_FLAGS "" CACHE STRING "Additional linker flags when linking unit tests.") -set(EIGEN_TEST_CUSTOM_CXX_FLAGS "" CACHE STRING "Additional compiler flags when compiling unit tests.") +install(FILES + signature_of_eigen3_matrix_library + DESTINATION ${INCLUDE_INSTALL_DIR} COMPONENT Devel + ) -set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "") +if(EIGEN_BUILD_PKGCONFIG) + configure_file(eigen3.pc.in eigen3.pc @ONLY) + install(FILES ${CMAKE_CURRENT_BINARY_DIR}/eigen3.pc + DESTINATION ${PKGCONFIG_INSTALL_DIR}) +endif() -if(NOT STANDARD_MATH_LIBRARY_FOUND) +install(DIRECTORY Eigen DESTINATION ${INCLUDE_INSTALL_DIR} COMPONENT Devel) - message(FATAL_ERROR - "Can't link to the standard math library. Please report to the Eigen developers, telling them about your platform.") +install(TARGETS eigen EXPORT Eigen3Targets) -else() +if(EIGEN_BUILD_CMAKE_PACKAGE) + include (CMakePackageConfigHelpers) + configure_package_config_file ( + ${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3Config.cmake.in + ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake + PATH_VARS EIGEN_INCLUDE_DIR EIGEN_ROOT_DIR + INSTALL_DESTINATION ${CMAKEPACKAGE_INSTALL_DIR} + NO_SET_AND_CHECK_MACRO # Eigen does not provide legacy style defines + NO_CHECK_REQUIRED_COMPONENTS_MACRO # Eigen does not provide components + ) - if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) - set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO} ${STANDARD_MATH_LIBRARY}") - else() - set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "${STANDARD_MATH_LIBRARY}") + set(CVF_VERSION "${EIGEN_VERSION_NUMBER}") + configure_file("${CMAKE_CURRENT_SOURCE_DIR}/cmake/Eigen3ConfigVersion.cmake.in" + "Eigen3ConfigVersion.cmake" + @ONLY) + + # The Eigen target will be located in the Eigen3 namespace. Other CMake + # targets can refer to it using Eigen3::Eigen. + export (TARGETS eigen NAMESPACE Eigen3:: FILE Eigen3Targets.cmake) + # Export Eigen3 package to CMake registry such that it can be easily found by + # CMake even if it has not been installed to a standard directory. + export (PACKAGE Eigen3) + + install (EXPORT Eigen3Targets NAMESPACE Eigen3:: DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}) + + install (FILES ${CMAKE_CURRENT_SOURCE_DIR}/cmake/UseEigen3.cmake + ${CMAKE_CURRENT_BINARY_DIR}/Eigen3Config.cmake + ${CMAKE_CURRENT_BINARY_DIR}/Eigen3ConfigVersion.cmake + DESTINATION ${CMAKEPACKAGE_INSTALL_DIR}) + + # Add uninstall target + if(NOT TARGET uninstall) + add_custom_target ( uninstall + COMMAND ${CMAKE_COMMAND} -P ${CMAKE_CURRENT_SOURCE_DIR}/cmake/EigenUninstall.cmake) endif() +endif() + +#============================================================================== +# General Build Configuration. +#============================================================================== +# Guard against in-source builds +if(${CMAKE_SOURCE_DIR} STREQUAL ${CMAKE_BINARY_DIR}) + message(FATAL_ERROR "In-source builds not allowed. Please make a new directory (called a build directory) and run CMake from there. You may need to remove CMakeCache.txt. ") endif() -if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) - message(STATUS "Standard libraries to link to explicitly: ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}") -else() - message(STATUS "Standard libraries to link to explicitly: none") +# Guard against bad build-type strings +if (PROJECT_IS_TOP_LEVEL AND NOT CMAKE_BUILD_TYPE) + set(CMAKE_BUILD_TYPE "Release") endif() -option(EIGEN_BUILD_BTL "Build benchmark suite" OFF) -if(NOT WIN32) - option(EIGEN_BUILD_PKGCONFIG "Build pkg-config .pc file for Eigen" ON) -endif(NOT WIN32) +# Only try to figure out how to link the math library if we are building something. +# Otherwise, let the parent project deal with dependencies. +if (EIGEN_IS_BUILDING_) + # Use Eigen's cmake files. + set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake) -set(CMAKE_INCLUDE_CURRENT_DIR ON) + set(CMAKE_INCLUDE_CURRENT_DIR OFF) -option(EIGEN_SPLIT_LARGE_TESTS "Split large tests into smaller executables" ON) + find_package(StandardMathLibrary) + set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "") + if(NOT STANDARD_MATH_LIBRARY_FOUND) + message(FATAL_ERROR + "Can't link to the standard math library. Please report to the Eigen developers, telling them about your platform.") + else() + if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) + set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO} ${STANDARD_MATH_LIBRARY}") + else() + set(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO "${STANDARD_MATH_LIBRARY}") + endif() + endif() + if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) + message(STATUS "Standard libraries to link to explicitly: ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}") + else() + message(STATUS "Standard libraries to link to explicitly: none") + endif() -option(EIGEN_DEFAULT_TO_ROW_MAJOR "Use row-major as default matrix storage order" OFF) -if(EIGEN_DEFAULT_TO_ROW_MAJOR) - add_definitions("-DEIGEN_DEFAULT_TO_ROW_MAJOR") + # Default tests/examples/libraries to row-major. + option(EIGEN_DEFAULT_TO_ROW_MAJOR "Use row-major as default matrix storage order" OFF) + if(EIGEN_DEFAULT_TO_ROW_MAJOR) + add_definitions("-DEIGEN_DEFAULT_TO_ROW_MAJOR") + endif() endif() -set(EIGEN_TEST_MAX_SIZE "320" CACHE STRING "Maximal matrix/vector size, default is 320") +#============================================================================== +# Test Configuration. +#============================================================================== + +if (EIGEN_BUILD_TESTING) + function(ei_maybe_separate_arguments variable mode args) + # Use separate_arguments if the input is a single string containing a space. + # Otherwise, if it is already a list or doesn't have a space, just propagate + # the original value. This is to better support multi-argument lists. + list(LENGTH args list_length) + if (${list_length} EQUAL 1) + string(FIND "${args}" " " has_space) + if (${has_space} GREATER -1) + separate_arguments(args ${mode} "${args}") + endif() + endif() + set(${variable} ${args} PARENT_SCOPE) + endfunction(ei_maybe_separate_arguments) + + include(CheckCXXCompilerFlag) + macro(ei_add_cxx_compiler_flag FLAG) + string(REGEX REPLACE "-" "" SFLAG1 ${FLAG}) + string(REGEX REPLACE "\\+" "p" SFLAG ${SFLAG1}) + check_cxx_compiler_flag(${FLAG} COMPILER_SUPPORT_${SFLAG}) + if(COMPILER_SUPPORT_${SFLAG}) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${FLAG}") + endif() + endmacro() -macro(ei_add_cxx_compiler_flag FLAG) - string(REGEX REPLACE "-" "" SFLAG ${FLAG}) - check_cxx_compiler_flag(${FLAG} COMPILER_SUPPORT_${SFLAG}) - if(COMPILER_SUPPORT_${SFLAG}) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} ${FLAG}") - endif() -endmacro(ei_add_cxx_compiler_flag) - -if(NOT MSVC) - # We assume that other compilers are partly compatible with GNUCC - - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fexceptions") - set(CMAKE_CXX_FLAGS_DEBUG "-g3") - set(CMAKE_CXX_FLAGS_RELEASE "-g0 -O2") - - # clang outputs some warnings for unknwon flags that are not caught by check_cxx_compiler_flag - # adding -Werror turns such warnings into errors - check_cxx_compiler_flag("-Werror" COMPILER_SUPPORT_WERROR) - if(COMPILER_SUPPORT_WERROR) - set(CMAKE_REQUIRED_FLAGS "-Werror") - endif() - - ei_add_cxx_compiler_flag("-pedantic") - ei_add_cxx_compiler_flag("-Wall") - ei_add_cxx_compiler_flag("-Wextra") - #ei_add_cxx_compiler_flag("-Weverything") # clang - - ei_add_cxx_compiler_flag("-Wundef") - ei_add_cxx_compiler_flag("-Wcast-align") - ei_add_cxx_compiler_flag("-Wchar-subscripts") - ei_add_cxx_compiler_flag("-Wnon-virtual-dtor") - ei_add_cxx_compiler_flag("-Wunused-local-typedefs") - ei_add_cxx_compiler_flag("-Wpointer-arith") - ei_add_cxx_compiler_flag("-Wwrite-strings") - ei_add_cxx_compiler_flag("-Wformat-security") - - ei_add_cxx_compiler_flag("-Wno-psabi") - ei_add_cxx_compiler_flag("-Wno-variadic-macros") - ei_add_cxx_compiler_flag("-Wno-long-long") - - ei_add_cxx_compiler_flag("-fno-check-new") - ei_add_cxx_compiler_flag("-fno-common") - ei_add_cxx_compiler_flag("-fstrict-aliasing") - ei_add_cxx_compiler_flag("-wd981") # disable ICC's "operands are evaluated in unspecified order" remark - ei_add_cxx_compiler_flag("-wd2304") # disbale ICC's "warning #2304: non-explicit constructor with single argument may cause implicit type conversion" produced by -Wnon-virtual-dtor - - # The -ansi flag must be added last, otherwise it is also used as a linker flag by check_cxx_compiler_flag making it fails - # Moreover we should not set both -strict-ansi and -ansi - check_cxx_compiler_flag("-strict-ansi" COMPILER_SUPPORT_STRICTANSI) - ei_add_cxx_compiler_flag("-Qunused-arguments") # disable clang warning: argument unused during compilation: '-ansi' - - if(COMPILER_SUPPORT_STRICTANSI) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -strict-ansi") + check_cxx_compiler_flag("-std=c++11" EIGEN_COMPILER_SUPPORT_CPP11) + + option(EIGEN_TEST_CXX11 "Enable testing with C++11 and C++11 features (e.g. Tensor module)." ${EIGEN_COMPILER_SUPPORT_CPP11}) + if(EIGEN_TEST_CXX11) + set(CMAKE_CXX_STANDARD 11) + set(CMAKE_CXX_EXTENSIONS OFF) + if(EIGEN_COMPILER_SUPPORT_CPP11) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11") + endif() else() - ei_add_cxx_compiler_flag("-ansi") + ei_add_cxx_compiler_flag("-std=c++03") endif() - - set(CMAKE_REQUIRED_FLAGS "") - option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF) - if(EIGEN_TEST_SSE2) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse2") - message(STATUS "Enabling SSE2 in tests/examples") - endif() + set(EIGEN_TEST_CUSTOM_LINKER_FLAGS "" CACHE STRING "Additional linker flags when linking unit tests.") + set(EIGEN_TEST_CUSTOM_CXX_FLAGS "" CACHE STRING "Additional compiler flags when compiling unit tests.") + # Convert space-separated arguments into CMake lists for downstream consumption. + ei_maybe_separate_arguments(EIGEN_TEST_CUSTOM_LINKER_FLAGS NATIVE_COMMAND "${EIGEN_TEST_CUSTOM_LINKER_FLAGS}") + ei_maybe_separate_arguments(EIGEN_TEST_CUSTOM_CXX_FLAGS NATIVE_COMMAND "${EIGEN_TEST_CUSTOM_CXX_FLAGS}") - option(EIGEN_TEST_SSE3 "Enable/Disable SSE3 in tests/examples" OFF) - if(EIGEN_TEST_SSE3) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3") - message(STATUS "Enabling SSE3 in tests/examples") - endif() + option(EIGEN_SPLIT_LARGE_TESTS "Split large tests into smaller executables" ON) + set(EIGEN_TEST_MAX_SIZE "320" CACHE STRING "Maximal matrix/vector size, default is 320") - option(EIGEN_TEST_SSSE3 "Enable/Disable SSSE3 in tests/examples" OFF) - if(EIGEN_TEST_SSSE3) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mssse3") - message(STATUS "Enabling SSSE3 in tests/examples") - endif() + # Flags for tests. + if(NOT MSVC) + # We assume that other compilers are partly compatible with GNUCC - option(EIGEN_TEST_SSE4_1 "Enable/Disable SSE4.1 in tests/examples" OFF) - if(EIGEN_TEST_SSE4_1) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.1") - message(STATUS "Enabling SSE4.1 in tests/examples") - endif() + # clang outputs some warnings for unknown flags that are not caught by check_cxx_compiler_flag + # adding -Werror turns such warnings into errors + check_cxx_compiler_flag("-Werror" COMPILER_SUPPORT_WERROR) + if(COMPILER_SUPPORT_WERROR) + set(CMAKE_REQUIRED_FLAGS "-Werror") + endif() + ei_add_cxx_compiler_flag("-pedantic") + ei_add_cxx_compiler_flag("-Wall") + ei_add_cxx_compiler_flag("-Wextra") + # ei_add_cxx_compiler_flag("-Weverything") # clang + ei_add_cxx_compiler_flag("-Wundef") + ei_add_cxx_compiler_flag("-Wcast-align") + ei_add_cxx_compiler_flag("-Wchar-subscripts") + ei_add_cxx_compiler_flag("-Wnon-virtual-dtor") + ei_add_cxx_compiler_flag("-Wunused-local-typedefs") + ei_add_cxx_compiler_flag("-Wpointer-arith") + ei_add_cxx_compiler_flag("-Wwrite-strings") + ei_add_cxx_compiler_flag("-Wformat-security") + ei_add_cxx_compiler_flag("-Wshorten-64-to-32") + ei_add_cxx_compiler_flag("-Wlogical-op") + ei_add_cxx_compiler_flag("-Wenum-conversion") + ei_add_cxx_compiler_flag("-Wc++11-extensions") + ei_add_cxx_compiler_flag("-Wdouble-promotion") + # ei_add_cxx_compiler_flag("-Wconversion") + ei_add_cxx_compiler_flag("-Wshadow") + ei_add_cxx_compiler_flag("-Wno-psabi") + ei_add_cxx_compiler_flag("-Wno-variadic-macros") + ei_add_cxx_compiler_flag("-Wno-long-long") + ei_add_cxx_compiler_flag("-fno-check-new") + ei_add_cxx_compiler_flag("-fno-common") + ei_add_cxx_compiler_flag("-fstrict-aliasing") + ei_add_cxx_compiler_flag("-wd981") # disable ICC's "operands are evaluated in unspecified order" remark + ei_add_cxx_compiler_flag("-wd2304") # disable ICC's "warning #2304: non-explicit constructor with single argument may cause implicit type conversion" produced by -Wnon-virtual-dtor + + if(ANDROID_NDK) + ei_add_cxx_compiler_flag("-pie") + ei_add_cxx_compiler_flag("-fPIE") + endif() - option(EIGEN_TEST_SSE4_2 "Enable/Disable SSE4.2 in tests/examples" OFF) - if(EIGEN_TEST_SSE4_2) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.2") - message(STATUS "Enabling SSE4.2 in tests/examples") - endif() + set(CMAKE_REQUIRED_FLAGS "") - option(EIGEN_TEST_ALTIVEC "Enable/Disable AltiVec in tests/examples" OFF) - if(EIGEN_TEST_ALTIVEC) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec") - message(STATUS "Enabling AltiVec in tests/examples") - endif() + option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF) + if(EIGEN_TEST_SSE2) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse2") + message(STATUS "Enabling SSE2 in tests/examples") + endif() - option(EIGEN_TEST_NEON "Enable/Disable Neon in tests/examples" OFF) - if(EIGEN_TEST_NEON) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon -mcpu=cortex-a8") - message(STATUS "Enabling NEON in tests/examples") - endif() + option(EIGEN_TEST_SSE3 "Enable/Disable SSE3 in tests/examples" OFF) + if(EIGEN_TEST_SSE3) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse3") + message(STATUS "Enabling SSE3 in tests/examples") + endif() - check_cxx_compiler_flag("-fopenmp" COMPILER_SUPPORT_OPENMP) - if(COMPILER_SUPPORT_OPENMP) - option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF) - if(EIGEN_TEST_OPENMP) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp") - message(STATUS "Enabling OpenMP in tests/examples") + option(EIGEN_TEST_SSSE3 "Enable/Disable SSSE3 in tests/examples" OFF) + if(EIGEN_TEST_SSSE3) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mssse3") + message(STATUS "Enabling SSSE3 in tests/examples") + endif() + + option(EIGEN_TEST_SSE4_1 "Enable/Disable SSE4.1 in tests/examples" OFF) + if(EIGEN_TEST_SSE4_1) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.1") + message(STATUS "Enabling SSE4.1 in tests/examples") endif() - endif() -else(NOT MSVC) + option(EIGEN_TEST_SSE4_2 "Enable/Disable SSE4.2 in tests/examples" OFF) + if(EIGEN_TEST_SSE4_2) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse4.2") + message(STATUS "Enabling SSE4.2 in tests/examples") + endif() - # C4127 - conditional expression is constant - # C4714 - marked as __forceinline not inlined (I failed to deactivate it selectively) - # We can disable this warning in the unit tests since it is clear that it occurs - # because we are oftentimes returning objects that have a destructor or may - # throw exceptions - in particular in the unit tests we are throwing extra many - # exceptions to cover indexing errors. - # C4505 - unreferenced local function has been removed (impossible to deactive selectively) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc /wd4127 /wd4505 /wd4714") + option(EIGEN_TEST_AVX "Enable/Disable AVX in tests/examples" OFF) + if(EIGEN_TEST_AVX) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx") + message(STATUS "Enabling AVX in tests/examples") + endif() - # replace all /Wx by /W4 - string(REGEX REPLACE "/W[0-9]" "/W4" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + option(EIGEN_TEST_FMA "Enable/Disable FMA in tests/examples" OFF) + if(EIGEN_TEST_FMA AND NOT EIGEN_TEST_NEON) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfma") + message(STATUS "Enabling FMA in tests/examples") + endif() - check_cxx_compiler_flag("/openmp" COMPILER_SUPPORT_OPENMP) - if(COMPILER_SUPPORT_OPENMP) - option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF) - if(EIGEN_TEST_OPENMP) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /openmp") - message(STATUS "Enabling OpenMP in tests/examples") + option(EIGEN_TEST_AVX2 "Enable/Disable AVX2 in tests/examples" OFF) + if(EIGEN_TEST_AVX2) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx2 -mfma") + message(STATUS "Enabling AVX2 in tests/examples") endif() - endif() - option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF) - if(EIGEN_TEST_SSE2) - if(NOT CMAKE_CL_64) - # arch is not supported on 64 bit systems, SSE is enabled automatically. - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:SSE2") - endif(NOT CMAKE_CL_64) - message(STATUS "Enabling SSE2 in tests/examples") - endif(EIGEN_TEST_SSE2) -endif(NOT MSVC) - -option(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION "Disable explicit vectorization in tests/examples" OFF) -option(EIGEN_TEST_X87 "Force using X87 instructions. Implies no vectorization." OFF) -option(EIGEN_TEST_32BIT "Force generating 32bit code." OFF) - -if(EIGEN_TEST_X87) - set(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION ON) - if(CMAKE_COMPILER_IS_GNUCXX) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpmath=387") - message(STATUS "Forcing use of x87 instructions in tests/examples") - else() - message(STATUS "EIGEN_TEST_X87 ignored on your compiler") - endif() -endif() + option(EIGEN_TEST_AVX512 "Enable/Disable AVX512 in tests/examples" OFF) + if(EIGEN_TEST_AVX512) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx512f -mfma") + message(STATUS "Enabling AVX512 in tests/examples") + endif() + + option(EIGEN_TEST_AVX512DQ "Enable/Disable AVX512DQ in tests/examples" OFF) + if(EIGEN_TEST_AVX512DQ) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx512dq -mfma") + message(STATUS "Enabling AVX512DQ in tests/examples") + endif() + + option(EIGEN_TEST_AVX512FP16 "Enable/Disable AVX512-FP16 in tests/examples" OFF) + if(EIGEN_TEST_AVX512FP16) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mavx512f -mfma -mavx512vl -mavx512fp16") + message(STATUS "Enabling AVX512-FP16 in tests/examples") + endif() + + option(EIGEN_TEST_F16C "Enable/Disable F16C in tests/examples" OFF) + if(EIGEN_TEST_F16C) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mf16c") + message(STATUS "Enabling F16C in tests/examples") + endif() + + option(EIGEN_TEST_ALTIVEC "Enable/Disable AltiVec in tests/examples" OFF) + if(EIGEN_TEST_ALTIVEC) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -maltivec -mabi=altivec") + message(STATUS "Enabling AltiVec in tests/examples") + endif() + + option(EIGEN_TEST_VSX "Enable/Disable VSX in tests/examples" OFF) + if(EIGEN_TEST_VSX) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m64 -mvsx") + message(STATUS "Enabling VSX in tests/examples") + endif() + + option(EIGEN_TEST_MSA "Enable/Disable MSA in tests/examples" OFF) + if(EIGEN_TEST_MSA) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mmsa") + message(STATUS "Enabling MSA in tests/examples") + endif() + + option(EIGEN_TEST_NEON "Enable/Disable Neon in tests/examples" OFF) + if(EIGEN_TEST_NEON) + if(EIGEN_TEST_FMA) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon-vfpv4") + else() + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpu=neon") + endif() + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfloat-abi=hard") + message(STATUS "Enabling NEON in tests/examples") + endif() + + option(EIGEN_TEST_NEON64 "Enable/Disable Neon in tests/examples" OFF) + if(EIGEN_TEST_NEON64) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + message(STATUS "Enabling NEON in tests/examples") + endif() + + option(EIGEN_TEST_Z13 "Enable/Disable S390X(zEC13) ZVECTOR in tests/examples" OFF) + if(EIGEN_TEST_Z13) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z13 -mzvector") + message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples") + endif() + + option(EIGEN_TEST_Z14 "Enable/Disable S390X(zEC14) ZVECTOR in tests/examples" OFF) + if(EIGEN_TEST_Z14) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -march=z14 -mzvector") + message(STATUS "Enabling S390X(zEC13) ZVECTOR in tests/examples") + endif() + + check_cxx_compiler_flag("-fopenmp" COMPILER_SUPPORT_OPENMP) + if(COMPILER_SUPPORT_OPENMP) + option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF) + if(EIGEN_TEST_OPENMP) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fopenmp") + message(STATUS "Enabling OpenMP in tests/examples") + endif() + endif() -if(EIGEN_TEST_32BIT) - if(CMAKE_COMPILER_IS_GNUCXX) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32") - message(STATUS "Forcing generation of 32-bit code in tests/examples") else() - message(STATUS "EIGEN_TEST_32BIT ignored on your compiler") - endif() -endif() + # C4127 - conditional expression is constant + # C4714 - marked as __forceinline not inlined (I failed to deactivate it selectively) + # We can disable this warning in the unit tests since it is clear that it occurs + # because we are oftentimes returning objects that have a destructor or may + # throw exceptions - in particular in the unit tests we are throwing extra many + # exceptions to cover indexing errors. + # C4505 - unreferenced local function has been removed (impossible to deactivate selectively) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /EHsc /wd4127 /wd4505 /wd4714") + + # replace all /Wx by /W4 + string(REGEX REPLACE "/W[0-9]" "/W4" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + + check_cxx_compiler_flag("/openmp" COMPILER_SUPPORT_OPENMP) + if(COMPILER_SUPPORT_OPENMP) + option(EIGEN_TEST_OPENMP "Enable/Disable OpenMP in tests/examples" OFF) + if(EIGEN_TEST_OPENMP) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /openmp") + message(STATUS "Enabling OpenMP in tests/examples") + endif() + endif() -if(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION) - add_definitions(-DEIGEN_DONT_VECTORIZE=1) - message(STATUS "Disabling vectorization in tests/examples") -endif() + option(EIGEN_TEST_SSE2 "Enable/Disable SSE2 in tests/examples" OFF) + if(EIGEN_TEST_SSE2) + if(NOT CMAKE_CL_64) + # arch is not supported on 64 bit systems, SSE is enabled automatically. + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:SSE2") + endif() + message(STATUS "Enabling SSE2 in tests/examples") + endif() -option(EIGEN_TEST_NO_EXPLICIT_ALIGNMENT "Disable explicit alignment (hence vectorization) in tests/examples" OFF) -if(EIGEN_TEST_NO_EXPLICIT_ALIGNMENT) - add_definitions(-DEIGEN_DONT_ALIGN=1) - message(STATUS "Disabling alignment in tests/examples") -endif() + option(EIGEN_TEST_AVX "Enable/Disable AVX in tests/examples" OFF) + if(EIGEN_TEST_AVX) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX") + message(STATUS "Enabling AVX in tests/examples") + endif() -option(EIGEN_TEST_C++0x "Enables all C++0x features." OFF) + option(EIGEN_TEST_FMA "Enable/Disable FMA/AVX2 in tests/examples" OFF) + option(EIGEN_TEST_AVX2 "Enable/Disable FMA/AVX2 in tests/examples" OFF) + if((EIGEN_TEST_FMA AND NOT EIGEN_TEST_NEON) OR EIGEN_TEST_AVX2) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX2") + message(STATUS "Enabling FMA/AVX2 in tests/examples") + endif() -include_directories(${CMAKE_CURRENT_SOURCE_DIR} ${CMAKE_CURRENT_BINARY_DIR}) + option(EIGEN_TEST_AVX512 "Enable/Disable AVX512 in tests/examples" OFF) + option(EIGEN_TEST_AVX512DQ "Enable/Disable AVX512DQ in tests/examples" OFF) + if(EIGEN_TEST_AVX512 OR EIGEN_TEST_AVX512DQ) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:AVX512") + message(STATUS "Enabling AVX512 in tests/examples") + endif() -# Backward compatibility support for EIGEN_INCLUDE_INSTALL_DIR -if(EIGEN_INCLUDE_INSTALL_DIR AND NOT INCLUDE_INSTALL_DIR) - set(INCLUDE_INSTALL_DIR ${EIGEN_INCLUDE_INSTALL_DIR} - CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed") -else() - set(INCLUDE_INSTALL_DIR - "${CMAKE_INSTALL_INCLUDEDIR}/eigen3" - CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen header files are installed" - ) -endif() + endif(NOT MSVC) -set(CMAKEPACKAGE_INSTALL_DIR - "${CMAKE_INSTALL_LIBDIR}/cmake/eigen3" - CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where Eigen3Config.cmake is installed" - ) -set(PKGCONFIG_INSTALL_DIR - "${CMAKE_INSTALL_DATADIR}/pkgconfig" - CACHE PATH "The directory relative to CMAKE_PREFIX_PATH where eigen3.pc is installed" - ) + option(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION "Disable explicit vectorization in tests/examples" OFF) + option(EIGEN_TEST_X87 "Force using X87 instructions. Implies no vectorization." OFF) + option(EIGEN_TEST_32BIT "Force generating 32bit code." OFF) -# similar to set_target_properties but append the property instead of overwriting it -macro(ei_add_target_property target prop value) + if(EIGEN_TEST_X87) + set(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION ON) + if(CMAKE_COMPILER_IS_GNUCXX) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -mfpmath=387") + message(STATUS "Forcing use of x87 instructions in tests/examples") + else() + message(STATUS "EIGEN_TEST_X87 ignored on your compiler") + endif() + endif() - get_target_property(previous ${target} ${prop}) - # if the property wasn't previously set, ${previous} is now "previous-NOTFOUND" which cmake allows catching with plain if() - if(NOT previous) - set(previous "") - endif(NOT previous) - set_target_properties(${target} PROPERTIES ${prop} "${previous} ${value}") -endmacro(ei_add_target_property) + if(EIGEN_TEST_32BIT) + if(CMAKE_COMPILER_IS_GNUCXX) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -m32") + message(STATUS "Forcing generation of 32-bit code in tests/examples") + else() + message(STATUS "EIGEN_TEST_32BIT ignored on your compiler") + endif() + endif() -install(FILES - signature_of_eigen3_matrix_library - DESTINATION ${INCLUDE_INSTALL_DIR} COMPONENT Devel - ) + if(EIGEN_TEST_NO_EXPLICIT_VECTORIZATION) + add_definitions(-DEIGEN_DONT_VECTORIZE=1) + message(STATUS "Disabling vectorization in tests/examples") + endif() -if(EIGEN_BUILD_PKGCONFIG) - configure_file(eigen3.pc.in eigen3.pc @ONLY) - install(FILES ${CMAKE_CURRENT_BINARY_DIR}/eigen3.pc - DESTINATION ${PKGCONFIG_INSTALL_DIR} - ) -endif(EIGEN_BUILD_PKGCONFIG) + option(EIGEN_TEST_NO_EXPLICIT_ALIGNMENT "Disable explicit alignment (hence vectorization) in tests/examples" OFF) + if(EIGEN_TEST_NO_EXPLICIT_ALIGNMENT) + add_definitions(-DEIGEN_DONT_ALIGN=1) + message(STATUS "Disabling alignment in tests/examples") + endif() + + option(EIGEN_TEST_NO_EXCEPTIONS "Disables C++ exceptions" OFF) + if(EIGEN_TEST_NO_EXCEPTIONS) + ei_add_cxx_compiler_flag("-fno-exceptions") + message(STATUS "Disabling exceptions in tests/examples") + endif() -add_subdirectory(Eigen) + set(EIGEN_CUDA_CXX_FLAGS "" CACHE STRING "Additional flags to pass to the cuda compiler.") + set(EIGEN_CUDA_COMPUTE_ARCH 30 CACHE STRING "The CUDA compute architecture(s) to target when compiling CUDA code") + + option(EIGEN_TEST_SYCL "Add Sycl support." OFF) + if(EIGEN_TEST_SYCL) + option(EIGEN_SYCL_DPCPP "Use the DPCPP Sycl implementation (DPCPP is default SYCL-Compiler)." ON) + option(EIGEN_SYCL_TRISYCL "Use the triSYCL Sycl implementation." OFF) + option(EIGEN_SYCL_ComputeCpp "Use the ComputeCPP Sycl implementation." OFF) + + # Building options + # https://developer.codeplay.com/products/computecpp/ce/2.11.0/guides/eigen-overview/options-for-building-eigen-sycl + option(EIGEN_SYCL_USE_DEFAULT_SELECTOR "Use sycl default selector to select the preferred device." OFF) + option(EIGEN_SYCL_NO_LOCAL_MEM "Build for devices without dedicated shared memory." OFF) + option(EIGEN_SYCL_LOCAL_MEM "Allow the use of local memory (enabled by default)." ON) + option(EIGEN_SYCL_LOCAL_THREAD_DIM0 "Set work group size for dimension 0." 16) + option(EIGEN_SYCL_LOCAL_THREAD_DIM1 "Set work group size for dimension 1." 16) + option(EIGEN_SYCL_ASYNC_EXECUTION "Allow asynchronous execution (enabled by default)." ON) + option(EIGEN_SYCL_DISABLE_SKINNY "Disable optimization for tall/skinny matrices." OFF) + option(EIGEN_SYCL_DISABLE_DOUBLE_BUFFER "Disable double buffer." OFF) + option(EIGEN_SYCL_DISABLE_SCALAR "Disable scalar contraction." OFF) + option(EIGEN_SYCL_DISABLE_GEMV "Disable GEMV and create a single kernel to calculate contraction instead." OFF) + + set(EIGEN_SYCL ON) + set(CMAKE_CXX_STANDARD 17) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-deprecated-declarations -Wno-shorten-64-to-32 -Wno-cast-align") + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-deprecated-copy-with-user-provided-copy -Wno-unused-variable") + set (CMAKE_MODULE_PATH "${CMAKE_ROOT}/Modules" "cmake/Modules/" "${CMAKE_MODULE_PATH}") + find_package(Threads REQUIRED) + if(EIGEN_SYCL_TRISYCL) + message(STATUS "Using triSYCL") + include(FindTriSYCL) + elseif(EIGEN_SYCL_ComputeCpp) + message(STATUS "Using ComputeCPP SYCL") + include(FindComputeCpp) + set(COMPUTECPP_DRIVER_DEFAULT_VALUE OFF) + if (NOT MSVC) + set(COMPUTECPP_DRIVER_DEFAULT_VALUE ON) + endif() + option(COMPUTECPP_USE_COMPILER_DRIVER + "Use ComputeCpp driver instead of a 2 steps compilation" + ${COMPUTECPP_DRIVER_DEFAULT_VALUE} + ) + else() #Default SYCL compiler is DPCPP (EIGEN_SYCL_DPCPP) + set(DPCPP_SYCL_TARGET "spir64" CACHE STRING "Default target for Intel CPU/GPU") + message(STATUS "Using DPCPP") + find_package(DPCPP) + add_definitions(-DSYCL_COMPILER_IS_DPCPP) + endif(EIGEN_SYCL_TRISYCL) + if(EIGEN_DONT_VECTORIZE_SYCL) + message(STATUS "Disabling SYCL vectorization in tests/examples") + # When disabling SYCL vectorization, also disable Eigen default vectorization + add_definitions(-DEIGEN_DONT_VECTORIZE=1) + add_definitions(-DEIGEN_DONT_VECTORIZE_SYCL=1) + endif() + endif() -add_subdirectory(doc EXCLUDE_FROM_ALL) + include(EigenConfigureTesting) -include(EigenConfigureTesting) + if(EIGEN_LEAVE_TEST_IN_ALL_TARGET) + # CTest automatic test building relies on the "all" target. + add_subdirectory(test) + add_subdirectory(failtest) + else() + add_subdirectory(test EXCLUDE_FROM_ALL) + add_subdirectory(failtest EXCLUDE_FROM_ALL) + endif() -# fixme, not sure this line is still needed: -enable_testing() # must be called from the root CMakeLists, see man page + ei_testing_print_summary() + if (EIGEN_SPLIT_TESTSUITE) + ei_split_testsuite("${EIGEN_SPLIT_TESTSUITE}") + endif() +endif(EIGEN_BUILD_TESTING) -if(EIGEN_LEAVE_TEST_IN_ALL_TARGET) - add_subdirectory(test) # can't do EXCLUDE_FROM_ALL here, breaks CTest -else() - add_subdirectory(test EXCLUDE_FROM_ALL) -endif() +#============================================================================== +# Other Build Configurations. +#============================================================================== +add_subdirectory(unsupported) -if(EIGEN_LEAVE_TEST_IN_ALL_TARGET) +if(EIGEN_BUILD_BLAS) add_subdirectory(blas) - add_subdirectory(lapack) -else() - add_subdirectory(blas EXCLUDE_FROM_ALL) - add_subdirectory(lapack EXCLUDE_FROM_ALL) endif() -add_subdirectory(unsupported) - -add_subdirectory(demos EXCLUDE_FROM_ALL) +if (EIGEN_BUILD_LAPACK) + add_subdirectory(lapack) +endif() -# must be after test and unsupported, for configuring buildtests.in -add_subdirectory(scripts EXCLUDE_FROM_ALL) +if(EIGEN_BUILD_DOC) + add_subdirectory(doc EXCLUDE_FROM_ALL) +endif() # TODO: consider also replacing EIGEN_BUILD_BTL by a custom target "make btl"? if(EIGEN_BUILD_BTL) add_subdirectory(bench/btl EXCLUDE_FROM_ALL) -endif(EIGEN_BUILD_BTL) +endif() -if(NOT WIN32) +if(NOT WIN32 AND EIGEN_BUILD_SPBENCH) add_subdirectory(bench/spbench EXCLUDE_FROM_ALL) -endif(NOT WIN32) - -configure_file(scripts/cdashtesting.cmake.in cdashtesting.cmake @ONLY) - -ei_testing_print_summary() +endif() -message(STATUS "") -message(STATUS "Configured Eigen ${EIGEN_VERSION_NUMBER}") -message(STATUS "") +if (EIGEN_BUILD_DEMOS) + add_subdirectory(demos EXCLUDE_FROM_ALL) +endif() -option(EIGEN_FAILTEST "Enable failtests." OFF) -if(EIGEN_FAILTEST) - add_subdirectory(failtest) +if (PROJECT_IS_TOP_LEVEL) + # must be after test and unsupported, for configuring buildtests.in + add_subdirectory(scripts EXCLUDE_FROM_ALL) + configure_file(scripts/cdashtesting.cmake.in cdashtesting.cmake @ONLY) endif() -string(TOLOWER "${CMAKE_GENERATOR}" cmake_generator_tolower) -if(cmake_generator_tolower MATCHES "makefile") - message(STATUS "Some things you can do now:") - message(STATUS "--------------+--------------------------------------------------------------") - message(STATUS "Command | Description") - message(STATUS "--------------+--------------------------------------------------------------") - message(STATUS "make install | Install Eigen. Headers will be installed to:") - message(STATUS " | /") - message(STATUS " | Using the following values:") - message(STATUS " | CMAKE_INSTALL_PREFIX: ${CMAKE_INSTALL_PREFIX}") - message(STATUS " | INCLUDE_INSTALL_DIR: ${INCLUDE_INSTALL_DIR}") - message(STATUS " | Change the install location of Eigen headers using:") - message(STATUS " | cmake . -DCMAKE_INSTALL_PREFIX=yourprefix") - message(STATUS " | Or:") - message(STATUS " | cmake . -DINCLUDE_INSTALL_DIR=yourdir") - message(STATUS "make doc | Generate the API documentation, requires Doxygen & LaTeX") - message(STATUS "make check | Build and run the unit-tests. Read this page:") - message(STATUS " | http://eigen.tuxfamily.org/index.php?title=Tests") - message(STATUS "make blas | Build BLAS library (not the same thing as Eigen)") - message(STATUS "--------------+--------------------------------------------------------------") -else() - message(STATUS "To build/run the unit tests, read this page:") - message(STATUS " http://eigen.tuxfamily.org/index.php?title=Tests") +#============================================================================== +# Summary. +#============================================================================== + +if(PROJECT_IS_TOP_LEVEL) + string(TOLOWER "${CMAKE_GENERATOR}" cmake_generator_tolower) + if(cmake_generator_tolower MATCHES "makefile") + message(STATUS "Available targets (use: make TARGET):") + else() + message(STATUS "Available targets (use: cmake --build . --target TARGET):") + endif() + message(STATUS "---------+--------------------------------------------------------------") + message(STATUS "Target | Description") + message(STATUS "---------+--------------------------------------------------------------") + message(STATUS "install | Install Eigen. Headers will be installed to:") + message(STATUS " | /") + message(STATUS " | Using the following values:") + message(STATUS " | CMAKE_INSTALL_PREFIX: ${CMAKE_INSTALL_PREFIX}") + message(STATUS " | INCLUDE_INSTALL_DIR: ${INCLUDE_INSTALL_DIR}") + message(STATUS " | Change the install location of Eigen headers using:") + message(STATUS " | cmake . -DCMAKE_INSTALL_PREFIX=yourprefix") + message(STATUS " | Or:") + message(STATUS " | cmake . -DINCLUDE_INSTALL_DIR=yourdir") + message(STATUS "uninstall| Remove files installed by the install target") + if (EIGEN_BUILD_DOC) + message(STATUS "doc | Generate the API documentation, requires Doxygen & LaTeX") + endif() + if(EIGEN_BUILD_TESTING) + message(STATUS "check | Build and run the unit-tests. Read this page:") + message(STATUS " | http://eigen.tuxfamily.org/index.php?title=Tests") + endif() + if (EIGEN_BUILD_BLAS) + message(STATUS "blas | Build BLAS library (not the same thing as Eigen)") + endif() + if (EIGEN_BUILD_LAPACK) + message(STATUS "lapack | Build LAPACK subset library (not the same thing as Eigen)") + endif() + message(STATUS "---------+--------------------------------------------------------------") + message(STATUS "") endif() message(STATUS "") +message(STATUS "Configured Eigen ${EIGEN_VERSION_NUMBER}") +message(STATUS "") diff --git a/thirdparty/eigen/COPYING.APACHE b/thirdparty/eigen/COPYING.APACHE new file mode 100644 index 00000000..61e948d2 --- /dev/null +++ b/thirdparty/eigen/COPYING.APACHE @@ -0,0 +1,203 @@ +/* + Apache License + Version 2.0, January 2004 + http://www.apache.org/licenses/ + + TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION + + 1. Definitions. + + "License" shall mean the terms and conditions for use, reproduction, + and distribution as defined by Sections 1 through 9 of this document. + + "Licensor" shall mean the copyright owner or entity authorized by + the copyright owner that is granting the License. + + "Legal Entity" shall mean the union of the acting entity and all + other entities that control, are controlled by, or are under common + control with that entity. For the purposes of this definition, + "control" means (i) the power, direct or indirect, to cause the + direction or management of such entity, whether by contract or + otherwise, or (ii) ownership of fifty percent (50%) or more of the + outstanding shares, or (iii) beneficial ownership of such entity. + + "You" (or "Your") shall mean an individual or Legal Entity + exercising permissions granted by this License. + + "Source" form shall mean the preferred form for making modifications, + including but not limited to software source code, documentation + source, and configuration files. + + "Object" form shall mean any form resulting from mechanical + transformation or translation of a Source form, including but + not limited to compiled object code, generated documentation, + and conversions to other media types. + + "Work" shall mean the work of authorship, whether in Source or + Object form, made available under the License, as indicated by a + copyright notice that is included in or attached to the work + (an example is provided in the Appendix below). + + "Derivative Works" shall mean any work, whether in Source or Object + form, that is based on (or derived from) the Work and for which the + editorial revisions, annotations, elaborations, or other modifications + represent, as a whole, an original work of authorship. For the purposes + of this License, Derivative Works shall not include works that remain + separable from, or merely link (or bind by name) to the interfaces of, + the Work and Derivative Works thereof. + + "Contribution" shall mean any work of authorship, including + the original version of the Work and any modifications or additions + to that Work or Derivative Works thereof, that is intentionally + submitted to Licensor for inclusion in the Work by the copyright owner + or by an individual or Legal Entity authorized to submit on behalf of + the copyright owner. For the purposes of this definition, "submitted" + means any form of electronic, verbal, or written communication sent + to the Licensor or its representatives, including but not limited to + communication on electronic mailing lists, source code control systems, + and issue tracking systems that are managed by, or on behalf of, the + Licensor for the purpose of discussing and improving the Work, but + excluding communication that is conspicuously marked or otherwise + designated in writing by the copyright owner as "Not a Contribution." + + "Contributor" shall mean Licensor and any individual or Legal Entity + on behalf of whom a Contribution has been received by Licensor and + subsequently incorporated within the Work. + + 2. Grant of Copyright License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + copyright license to reproduce, prepare Derivative Works of, + publicly display, publicly perform, sublicense, and distribute the + Work and such Derivative Works in Source or Object form. + + 3. Grant of Patent License. Subject to the terms and conditions of + this License, each Contributor hereby grants to You a perpetual, + worldwide, non-exclusive, no-charge, royalty-free, irrevocable + (except as stated in this section) patent license to make, have made, + use, offer to sell, sell, import, and otherwise transfer the Work, + where such license applies only to those patent claims licensable + by such Contributor that are necessarily infringed by their + Contribution(s) alone or by combination of their Contribution(s) + with the Work to which such Contribution(s) was submitted. If You + institute patent litigation against any entity (including a + cross-claim or counterclaim in a lawsuit) alleging that the Work + or a Contribution incorporated within the Work constitutes direct + or contributory patent infringement, then any patent licenses + granted to You under this License for that Work shall terminate + as of the date such litigation is filed. + + 4. Redistribution. You may reproduce and distribute copies of the + Work or Derivative Works thereof in any medium, with or without + modifications, and in Source or Object form, provided that You + meet the following conditions: + + (a) You must give any other recipients of the Work or + Derivative Works a copy of this License; and + + (b) You must cause any modified files to carry prominent notices + stating that You changed the files; and + + (c) You must retain, in the Source form of any Derivative Works + that You distribute, all copyright, patent, trademark, and + attribution notices from the Source form of the Work, + excluding those notices that do not pertain to any part of + the Derivative Works; and + + (d) If the Work includes a "NOTICE" text file as part of its + distribution, then any Derivative Works that You distribute must + include a readable copy of the attribution notices contained + within such NOTICE file, excluding those notices that do not + pertain to any part of the Derivative Works, in at least one + of the following places: within a NOTICE text file distributed + as part of the Derivative Works; within the Source form or + documentation, if provided along with the Derivative Works; or, + within a display generated by the Derivative Works, if and + wherever such third-party notices normally appear. The contents + of the NOTICE file are for informational purposes only and + do not modify the License. You may add Your own attribution + notices within Derivative Works that You distribute, alongside + or as an addendum to the NOTICE text from the Work, provided + that such additional attribution notices cannot be construed + as modifying the License. + + You may add Your own copyright statement to Your modifications and + may provide additional or different license terms and conditions + for use, reproduction, or distribution of Your modifications, or + for any such Derivative Works as a whole, provided Your use, + reproduction, and distribution of the Work otherwise complies with + the conditions stated in this License. + + 5. Submission of Contributions. Unless You explicitly state otherwise, + any Contribution intentionally submitted for inclusion in the Work + by You to the Licensor shall be under the terms and conditions of + this License, without any additional terms or conditions. + Notwithstanding the above, nothing herein shall supersede or modify + the terms of any separate license agreement you may have executed + with Licensor regarding such Contributions. + + 6. Trademarks. This License does not grant permission to use the trade + names, trademarks, service marks, or product names of the Licensor, + except as required for reasonable and customary use in describing the + origin of the Work and reproducing the content of the NOTICE file. + + 7. Disclaimer of Warranty. Unless required by applicable law or + agreed to in writing, Licensor provides the Work (and each + Contributor provides its Contributions) on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or + implied, including, without limitation, any warranties or conditions + of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A + PARTICULAR PURPOSE. You are solely responsible for determining the + appropriateness of using or redistributing the Work and assume any + risks associated with Your exercise of permissions under this License. + + 8. Limitation of Liability. In no event and under no legal theory, + whether in tort (including negligence), contract, or otherwise, + unless required by applicable law (such as deliberate and grossly + negligent acts) or agreed to in writing, shall any Contributor be + liable to You for damages, including any direct, indirect, special, + incidental, or consequential damages of any character arising as a + result of this License or out of the use or inability to use the + Work (including but not limited to damages for loss of goodwill, + work stoppage, computer failure or malfunction, or any and all + other commercial damages or losses), even if such Contributor + has been advised of the possibility of such damages. + + 9. Accepting Warranty or Additional Liability. While redistributing + the Work or Derivative Works thereof, You may choose to offer, + and charge a fee for, acceptance of support, warranty, indemnity, + or other liability obligations and/or rights consistent with this + License. However, in accepting such obligations, You may act only + on Your own behalf and on Your sole responsibility, not on behalf + of any other Contributor, and only if You agree to indemnify, + defend, and hold each Contributor harmless for any liability + incurred by, or claims asserted against, such Contributor by reason + of your accepting any such warranty or additional liability. + + END OF TERMS AND CONDITIONS + + APPENDIX: How to apply the Apache License to your work. + + To apply the Apache License to your work, attach the following + boilerplate notice, with the fields enclosed by brackets "[]" + replaced with your own identifying information. (Don't include + the brackets!) The text should be enclosed in the appropriate + comment syntax for the file format. We also recommend that a + file or class name and description of purpose be included on the + same "printed page" as the copyright notice for easier + identification within third-party archives. + + Copyright [yyyy] [name of copyright owner] + + Licensed under the Apache License, Version 2.0 (the "License"); + you may not use this file except in compliance with the License. + You may obtain a copy of the License at + + http://www.apache.org/licenses/LICENSE-2.0 + + Unless required by applicable law or agreed to in writing, software + distributed under the License is distributed on an "AS IS" BASIS, + WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + See the License for the specific language governing permissions and + limitations under the License. +*/ \ No newline at end of file diff --git a/thirdparty/eigen/COPYING.BSD b/thirdparty/eigen/COPYING.BSD index 11971ffe..8964ddfd 100644 --- a/thirdparty/eigen/COPYING.BSD +++ b/thirdparty/eigen/COPYING.BSD @@ -23,4 +23,4 @@ ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -*/ \ No newline at end of file +*/ diff --git a/thirdparty/eigen/COPYING.MINPACK b/thirdparty/eigen/COPYING.MINPACK index 11d8a9a6..132cc3f3 100644 --- a/thirdparty/eigen/COPYING.MINPACK +++ b/thirdparty/eigen/COPYING.MINPACK @@ -1,52 +1,51 @@ -Minpack Copyright Notice (1999) University of Chicago. All rights reserved - -Redistribution and use in source and binary forms, with or -without modification, are permitted provided that the -following conditions are met: - -1. Redistributions of source code must retain the above -copyright notice, this list of conditions and the following -disclaimer. - -2. Redistributions in binary form must reproduce the above -copyright notice, this list of conditions and the following -disclaimer in the documentation and/or other materials -provided with the distribution. - -3. The end-user documentation included with the -redistribution, if any, must include the following -acknowledgment: - - "This product includes software developed by the - University of Chicago, as Operator of Argonne National - Laboratory. - -Alternately, this acknowledgment may appear in the software -itself, if and wherever such third-party acknowledgments -normally appear. - -4. WARRANTY DISCLAIMER. THE SOFTWARE IS SUPPLIED "AS IS" -WITHOUT WARRANTY OF ANY KIND. THE COPYRIGHT HOLDER, THE -UNITED STATES, THE UNITED STATES DEPARTMENT OF ENERGY, AND -THEIR EMPLOYEES: (1) DISCLAIM ANY WARRANTIES, EXPRESS OR -IMPLIED, INCLUDING BUT NOT LIMITED TO ANY IMPLIED WARRANTIES -OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE -OR NON-INFRINGEMENT, (2) DO NOT ASSUME ANY LEGAL LIABILITY -OR RESPONSIBILITY FOR THE ACCURACY, COMPLETENESS, OR -USEFULNESS OF THE SOFTWARE, (3) DO NOT REPRESENT THAT USE OF -THE SOFTWARE WOULD NOT INFRINGE PRIVATELY OWNED RIGHTS, (4) -DO NOT WARRANT THAT THE SOFTWARE WILL FUNCTION -UNINTERRUPTED, THAT IT IS ERROR-FREE OR THAT ANY ERRORS WILL -BE CORRECTED. - -5. LIMITATION OF LIABILITY. IN NO EVENT WILL THE COPYRIGHT -HOLDER, THE UNITED STATES, THE UNITED STATES DEPARTMENT OF -ENERGY, OR THEIR EMPLOYEES: BE LIABLE FOR ANY INDIRECT, -INCIDENTAL, CONSEQUENTIAL, SPECIAL OR PUNITIVE DAMAGES OF -ANY KIND OR NATURE, INCLUDING BUT NOT LIMITED TO LOSS OF -PROFITS OR LOSS OF DATA, FOR ANY REASON WHATSOEVER, WHETHER -SUCH LIABILITY IS ASSERTED ON THE BASIS OF CONTRACT, TORT -(INCLUDING NEGLIGENCE OR STRICT LIABILITY), OR OTHERWISE, -EVEN IF ANY OF SAID PARTIES HAS BEEN WARNED OF THE -POSSIBILITY OF SUCH LOSS OR DAMAGES. - +Minpack Copyright Notice (1999) University of Chicago. All rights reserved + +Redistribution and use in source and binary forms, with or +without modification, are permitted provided that the +following conditions are met: + +1. Redistributions of source code must retain the above +copyright notice, this list of conditions and the following +disclaimer. + +2. Redistributions in binary form must reproduce the above +copyright notice, this list of conditions and the following +disclaimer in the documentation and/or other materials +provided with the distribution. + +3. The end-user documentation included with the +redistribution, if any, must include the following +acknowledgment: + + "This product includes software developed by the + University of Chicago, as Operator of Argonne National + Laboratory. + +Alternately, this acknowledgment may appear in the software +itself, if and wherever such third-party acknowledgments +normally appear. + +4. WARRANTY DISCLAIMER. THE SOFTWARE IS SUPPLIED "AS IS" +WITHOUT WARRANTY OF ANY KIND. THE COPYRIGHT HOLDER, THE +UNITED STATES, THE UNITED STATES DEPARTMENT OF ENERGY, AND +THEIR EMPLOYEES: (1) DISCLAIM ANY WARRANTIES, EXPRESS OR +IMPLIED, INCLUDING BUT NOT LIMITED TO ANY IMPLIED WARRANTIES +OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, TITLE +OR NON-INFRINGEMENT, (2) DO NOT ASSUME ANY LEGAL LIABILITY +OR RESPONSIBILITY FOR THE ACCURACY, COMPLETENESS, OR +USEFULNESS OF THE SOFTWARE, (3) DO NOT REPRESENT THAT USE OF +THE SOFTWARE WOULD NOT INFRINGE PRIVATELY OWNED RIGHTS, (4) +DO NOT WARRANT THAT THE SOFTWARE WILL FUNCTION +UNINTERRUPTED, THAT IT IS ERROR-FREE OR THAT ANY ERRORS WILL +BE CORRECTED. + +5. LIMITATION OF LIABILITY. IN NO EVENT WILL THE COPYRIGHT +HOLDER, THE UNITED STATES, THE UNITED STATES DEPARTMENT OF +ENERGY, OR THEIR EMPLOYEES: BE LIABLE FOR ANY INDIRECT, +INCIDENTAL, CONSEQUENTIAL, SPECIAL OR PUNITIVE DAMAGES OF +ANY KIND OR NATURE, INCLUDING BUT NOT LIMITED TO LOSS OF +PROFITS OR LOSS OF DATA, FOR ANY REASON WHATSOEVER, WHETHER +SUCH LIABILITY IS ASSERTED ON THE BASIS OF CONTRACT, TORT +(INCLUDING NEGLIGENCE OR STRICT LIABILITY), OR OTHERWISE, +EVEN IF ANY OF SAID PARTIES HAS BEEN WARNED OF THE +POSSIBILITY OF SUCH LOSS OR DAMAGES. diff --git a/thirdparty/eigen/CTestConfig.cmake b/thirdparty/eigen/CTestConfig.cmake index 0557c491..0ea24b8e 100644 --- a/thirdparty/eigen/CTestConfig.cmake +++ b/thirdparty/eigen/CTestConfig.cmake @@ -2,12 +2,16 @@ ## Then modify the CMakeLists.txt file in the root directory of your ## project to incorporate the testing dashboard. ## # The following are required to uses Dart and the Cdash dashboard -## ENABLE_TESTING() -## INCLUDE(CTest) -set(CTEST_PROJECT_NAME "Eigen3.2") +## enable_testing() +## include(CTest) +set(CTEST_PROJECT_NAME "Eigen") set(CTEST_NIGHTLY_START_TIME "00:00:00 UTC") set(CTEST_DROP_METHOD "http") -set(CTEST_DROP_SITE "manao.inria.fr") -set(CTEST_DROP_LOCATION "/CDash/submit.php?project=Eigen3.2") +set(CTEST_DROP_SITE "my.cdash.org") +set(CTEST_DROP_LOCATION "/submit.php?project=Eigen") set(CTEST_DROP_SITE_CDASH TRUE) +#set(CTEST_PROJECT_SUBPROJECTS +#Official +#Unsupported +#) diff --git a/thirdparty/eigen/CTestCustom.cmake.in b/thirdparty/eigen/CTestCustom.cmake.in index 9fed9d32..89e487f0 100644 --- a/thirdparty/eigen/CTestCustom.cmake.in +++ b/thirdparty/eigen/CTestCustom.cmake.in @@ -1,3 +1,4 @@ set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_WARNINGS "2000") set(CTEST_CUSTOM_MAXIMUM_NUMBER_OF_ERRORS "2000") +list(APPEND CTEST_CUSTOM_ERROR_EXCEPTION @EIGEN_CTEST_ERROR_EXCEPTION@) diff --git a/thirdparty/eigen/Eigen/Array b/thirdparty/eigen/Eigen/Array deleted file mode 100644 index 3d004fb6..00000000 --- a/thirdparty/eigen/Eigen/Array +++ /dev/null @@ -1,11 +0,0 @@ -#ifndef EIGEN_ARRAY_MODULE_H -#define EIGEN_ARRAY_MODULE_H - -// include Core first to handle Eigen2 support macros -#include "Core" - -#ifndef EIGEN2_SUPPORT - #error The Eigen/Array header does no longer exist in Eigen3. All that functionality has moved to Eigen/Core. -#endif - -#endif // EIGEN_ARRAY_MODULE_H diff --git a/thirdparty/eigen/Eigen/CMakeLists.txt b/thirdparty/eigen/Eigen/CMakeLists.txt deleted file mode 100644 index a92dd6f6..00000000 --- a/thirdparty/eigen/Eigen/CMakeLists.txt +++ /dev/null @@ -1,19 +0,0 @@ -include(RegexUtils) -test_escape_string_as_regex() - -file(GLOB Eigen_directory_files "*") - -escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}") - -foreach(f ${Eigen_directory_files}) - if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/src") - list(APPEND Eigen_directory_files_to_install ${f}) - endif() -endforeach(f ${Eigen_directory_files}) - -install(FILES - ${Eigen_directory_files_to_install} - DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen COMPONENT Devel - ) - -add_subdirectory(src) diff --git a/thirdparty/eigen/Eigen/Cholesky b/thirdparty/eigen/Eigen/Cholesky index f727f5d8..a318ceb7 100644 --- a/thirdparty/eigen/Eigen/Cholesky +++ b/thirdparty/eigen/Eigen/Cholesky @@ -1,7 +1,15 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_CHOLESKY_MODULE_H #define EIGEN_CHOLESKY_MODULE_H #include "Core" +#include "Jacobi" #include "src/Core/util/DisableStupidWarnings.h" @@ -10,23 +18,28 @@ * * * This module provides two variants of the Cholesky decomposition for selfadjoint (hermitian) matrices. - * Those decompositions are accessible via the following MatrixBase methods: - * - MatrixBase::llt(), + * Those decompositions are also accessible via the following methods: + * - MatrixBase::llt() * - MatrixBase::ldlt() + * - SelfAdjointView::llt() + * - SelfAdjointView::ldlt() * * \code * #include * \endcode */ -#include "src/misc/Solve.h" #include "src/Cholesky/LLT.h" #include "src/Cholesky/LDLT.h" #ifdef EIGEN_USE_LAPACKE -#include "src/Cholesky/LLT_MKL.h" +#ifdef EIGEN_USE_MKL +#include "mkl_lapacke.h" +#else +#include "src/misc/lapacke.h" +#endif +#include "src/Cholesky/LLT_LAPACKE.h" #endif #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_CHOLESKY_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/CholmodSupport b/thirdparty/eigen/Eigen/CholmodSupport index 88c29a64..1037bd55 100644 --- a/thirdparty/eigen/Eigen/CholmodSupport +++ b/thirdparty/eigen/Eigen/CholmodSupport @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_CHOLMODSUPPORT_MODULE_H #define EIGEN_CHOLMODSUPPORT_MODULE_H @@ -15,7 +22,7 @@ extern "C" { * This module provides an interface to the Cholmod library which is part of the suitesparse package. * It provides the two following main factorization classes: * - class CholmodSupernodalLLT: a supernodal LLT Cholesky factorization. - * - class CholmodDecomposiiton: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial). + * - class CholmodDecomposition: a general L(D)LT Cholesky factorization with automatic or explicit runtime selection of the underlying factorization method (supernodal or simplicial). * * For the sake of completeness, this module also propose the two following classes: * - class CholmodSimplicialLLT @@ -33,12 +40,8 @@ extern "C" { * */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" - #include "src/CholmodSupport/CholmodSupport.h" - #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_CHOLMODSUPPORT_MODULE_H diff --git a/thirdparty/eigen/Eigen/Core b/thirdparty/eigen/Eigen/Core index 509c529e..bb16c86f 100644 --- a/thirdparty/eigen/Eigen/Core +++ b/thirdparty/eigen/Eigen/Core @@ -11,123 +11,55 @@ #ifndef EIGEN_CORE_H #define EIGEN_CORE_H -// first thing Eigen does: stop the compiler from committing suicide +// first thing Eigen does: stop the compiler from reporting useless warnings. #include "src/Core/util/DisableStupidWarnings.h" // then include this file where all our macros are defined. It's really important to do it first because -// it's where we do all the alignment settings (platform detection and honoring the user's will if he -// defined e.g. EIGEN_DONT_ALIGN) so it needs to be done before we do anything with vectorization. +// it's where we do all the compiler/OS/arch detections and define most defaults. #include "src/Core/util/Macros.h" +// This detects SSE/AVX/NEON/etc. and configure alignment settings +#include "src/Core/util/ConfigureVectorization.h" + +// We need cuda_runtime.h/hip_runtime.h to ensure that +// the EIGEN_USING_STD macro works properly on the device side +#if defined(EIGEN_CUDACC) + #include +#elif defined(EIGEN_HIPCC) + #include +#endif + + +#ifdef EIGEN_EXCEPTIONS + #include +#endif + // Disable the ipa-cp-clone optimization flag with MinGW 6.x or newer (enabled by default with -O3) // See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=556 for details. -#if defined(__MINGW32__) && EIGEN_GNUC_AT_LEAST(4,6) +#if EIGEN_COMP_MINGW && EIGEN_GNUC_AT_LEAST(4,6) && EIGEN_GNUC_AT_MOST(5,5) #pragma GCC optimize ("-fno-ipa-cp-clone") #endif +// Prevent ICC from specializing std::complex operators that silently fail +// on device. This allows us to use our own device-compatible specializations +// instead. +#if defined(EIGEN_COMP_ICC) && defined(EIGEN_GPU_COMPILE_PHASE) \ + && !defined(_OVERRIDE_COMPLEX_SPECIALIZATION_) +#define _OVERRIDE_COMPLEX_SPECIALIZATION_ 1 +#endif #include // this include file manages BLAS and MKL related macros // and inclusion of their respective header files #include "src/Core/util/MKL_support.h" -// if alignment is disabled, then disable vectorization. Note: EIGEN_ALIGN is the proper check, it takes into -// account both the user's will (EIGEN_DONT_ALIGN) and our own platform checks -#if !EIGEN_ALIGN - #ifndef EIGEN_DONT_VECTORIZE - #define EIGEN_DONT_VECTORIZE - #endif -#endif -#ifdef _MSC_VER - #include // for _aligned_malloc -- need it regardless of whether vectorization is enabled - #if (_MSC_VER >= 1500) // 2008 or later - // Remember that usage of defined() in a #define is undefined by the standard. - // a user reported that in 64-bit mode, MSVC doesn't care to define _M_IX86_FP. - #if (defined(_M_IX86_FP) && (_M_IX86_FP >= 2)) || defined(_M_X64) - #define EIGEN_SSE2_ON_MSVC_2008_OR_LATER - #endif - #endif -#else - // Remember that usage of defined() in a #define is undefined by the standard - #if (defined __SSE2__) && ( (!defined __GNUC__) || (defined __INTEL_COMPILER) || EIGEN_GNUC_AT_LEAST(4,2) ) - #define EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC - #endif +#if defined(EIGEN_HAS_CUDA_FP16) || defined(EIGEN_HAS_HIP_FP16) + #define EIGEN_HAS_GPU_FP16 #endif -#ifndef EIGEN_DONT_VECTORIZE - - #if defined (EIGEN_SSE2_ON_NON_MSVC_BUT_NOT_OLD_GCC) || defined(EIGEN_SSE2_ON_MSVC_2008_OR_LATER) - - // Defines symbols for compile-time detection of which instructions are - // used. - // EIGEN_VECTORIZE_YY is defined if and only if the instruction set YY is used - #define EIGEN_VECTORIZE - #define EIGEN_VECTORIZE_SSE - #define EIGEN_VECTORIZE_SSE2 - - // Detect sse3/ssse3/sse4: - // gcc and icc defines __SSE3__, ... - // there is no way to know about this on msvc. You can define EIGEN_VECTORIZE_SSE* if you - // want to force the use of those instructions with msvc. - #ifdef __SSE3__ - #define EIGEN_VECTORIZE_SSE3 - #endif - #ifdef __SSSE3__ - #define EIGEN_VECTORIZE_SSSE3 - #endif - #ifdef __SSE4_1__ - #define EIGEN_VECTORIZE_SSE4_1 - #endif - #ifdef __SSE4_2__ - #define EIGEN_VECTORIZE_SSE4_2 - #endif - - // include files - - // This extern "C" works around a MINGW-w64 compilation issue - // https://sourceforge.net/tracker/index.php?func=detail&aid=3018394&group_id=202880&atid=983354 - // In essence, intrin.h is included by windows.h and also declares intrinsics (just as emmintrin.h etc. below do). - // However, intrin.h uses an extern "C" declaration, and g++ thus complains of duplicate declarations - // with conflicting linkage. The linkage for intrinsics doesn't matter, but at that stage the compiler doesn't know; - // so, to avoid compile errors when windows.h is included after Eigen/Core, ensure intrinsics are extern "C" here too. - // notice that since these are C headers, the extern "C" is theoretically needed anyways. - extern "C" { - // In theory we should only include immintrin.h and not the other *mmintrin.h header files directly. - // Doing so triggers some issues with ICC. However old gcc versions seems to not have this file, thus: - #if defined(__INTEL_COMPILER) && __INTEL_COMPILER >= 1110 - #include - #else - #include - #include - #ifdef EIGEN_VECTORIZE_SSE3 - #include - #endif - #ifdef EIGEN_VECTORIZE_SSSE3 - #include - #endif - #ifdef EIGEN_VECTORIZE_SSE4_1 - #include - #endif - #ifdef EIGEN_VECTORIZE_SSE4_2 - #include - #endif - #endif - } // end extern "C" - #elif defined __ALTIVEC__ - #define EIGEN_VECTORIZE - #define EIGEN_VECTORIZE_ALTIVEC - #include - // We need to #undef all these ugly tokens defined in - // => use __vector instead of vector - #undef bool - #undef vector - #undef pixel - #elif defined __ARM_NEON - #define EIGEN_VECTORIZE - #define EIGEN_VECTORIZE_NEON - #include - #endif +#if defined(EIGEN_HAS_CUDA_BF16) || defined(EIGEN_HAS_HIP_BF16) + #define EIGEN_HAS_GPU_BF16 #endif #if (defined _OPENMP) && (!defined EIGEN_DONT_PARALLELIZE) @@ -139,7 +71,7 @@ #endif // MSVC for windows mobile does not have the errno.h file -#if !(defined(_MSC_VER) && defined(_WIN32_WCE)) && !defined(__ARMCC_VERSION) +#if !(EIGEN_COMP_MSVC && EIGEN_OS_WINCE) && !EIGEN_COMP_ARM #define EIGEN_HAS_ERRNO #endif @@ -151,7 +83,10 @@ #include #include #include -#include +#ifndef EIGEN_NO_IO + #include + #include +#endif #include #include #include @@ -159,85 +94,61 @@ // for min/max: #include +#if EIGEN_HAS_CXX11 +#include +#endif + +// for std::is_nothrow_move_assignable +#ifdef EIGEN_INCLUDE_TYPE_TRAITS +#include +#endif + // for outputting debug info #ifdef EIGEN_DEBUG_ASSIGN #include #endif // required for __cpuid, needs to be included after cmath -#if defined(_MSC_VER) && (defined(_M_IX86)||defined(_M_X64)) && (!defined(_WIN32_WCE)) +// also required for _BitScanReverse on Windows on ARM +#if EIGEN_COMP_MSVC && (EIGEN_ARCH_i386_OR_x86_64 || EIGEN_ARCH_ARM64) && !EIGEN_OS_WINCE #include #endif -#if defined(_CPPUNWIND) || defined(__EXCEPTIONS) - #define EIGEN_EXCEPTIONS -#endif - -#ifdef EIGEN_EXCEPTIONS - #include +#if defined(EIGEN_USE_SYCL) + #undef min + #undef max + #undef isnan + #undef isinf + #undef isfinite + #include + #include + #include + #include + #include + #ifndef EIGEN_SYCL_LOCAL_THREAD_DIM0 + #define EIGEN_SYCL_LOCAL_THREAD_DIM0 16 + #endif + #ifndef EIGEN_SYCL_LOCAL_THREAD_DIM1 + #define EIGEN_SYCL_LOCAL_THREAD_DIM1 16 + #endif #endif -/** \brief Namespace containing all symbols from the %Eigen library. */ -namespace Eigen { -inline static const char *SimdInstructionSetsInUse(void) { -#if defined(EIGEN_VECTORIZE_SSE4_2) - return "SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2"; -#elif defined(EIGEN_VECTORIZE_SSE4_1) - return "SSE, SSE2, SSE3, SSSE3, SSE4.1"; -#elif defined(EIGEN_VECTORIZE_SSSE3) - return "SSE, SSE2, SSE3, SSSE3"; -#elif defined(EIGEN_VECTORIZE_SSE3) - return "SSE, SSE2, SSE3"; -#elif defined(EIGEN_VECTORIZE_SSE2) - return "SSE, SSE2"; -#elif defined(EIGEN_VECTORIZE_ALTIVEC) - return "AltiVec"; -#elif defined(EIGEN_VECTORIZE_NEON) - return "ARM NEON"; -#else - return "None"; -#endif -} - -} // end namespace Eigen - -#define STAGE10_FULL_EIGEN2_API 10 -#define STAGE20_RESOLVE_API_CONFLICTS 20 -#define STAGE30_FULL_EIGEN3_API 30 -#define STAGE40_FULL_EIGEN3_STRICTNESS 40 -#define STAGE99_NO_EIGEN2_SUPPORT 99 - -#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE40_FULL_EIGEN3_STRICTNESS -#elif defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API -#elif defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE20_RESOLVE_API_CONFLICTS -#elif defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API - #define EIGEN2_SUPPORT - #define EIGEN2_SUPPORT_STAGE STAGE10_FULL_EIGEN2_API -#elif defined EIGEN2_SUPPORT - // default to stage 3, that's what it's always meant - #define EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API - #define EIGEN2_SUPPORT_STAGE STAGE30_FULL_EIGEN3_API -#else - #define EIGEN2_SUPPORT_STAGE STAGE99_NO_EIGEN2_SUPPORT +#if defined EIGEN2_SUPPORT_STAGE40_FULL_EIGEN3_STRICTNESS || defined EIGEN2_SUPPORT_STAGE30_FULL_EIGEN3_API || defined EIGEN2_SUPPORT_STAGE20_RESOLVE_API_CONFLICTS || defined EIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API || defined EIGEN2_SUPPORT +// This will generate an error message: +#error Eigen2-support is only available up to version 3.2. Please go to "http://eigen.tuxfamily.org/index.php?title=Eigen2" for further information #endif -#ifdef EIGEN2_SUPPORT -#undef minor -#endif +namespace Eigen { -// we use size_t frequently and we'll never remember to prepend it with std:: everytime just to +// we use size_t frequently and we'll never remember to prepend it with std:: every time just to // ensure QNX/QCC support using std::size_t; -// gcc 4.6.0 wants std:: for ptrdiff_t +// gcc 4.6.0 wants std:: for ptrdiff_t using std::ptrdiff_t; +} + /** \defgroup Core_Module Core module * This is the main module of Eigen providing dense matrix and vector support * (both fixed and dynamic size) with all the features corresponding to a BLAS library @@ -249,50 +160,142 @@ using std::ptrdiff_t; */ #include "src/Core/util/Constants.h" -#include "src/Core/util/ForwardDeclarations.h" #include "src/Core/util/Meta.h" +#include "src/Core/util/ForwardDeclarations.h" #include "src/Core/util/StaticAssert.h" #include "src/Core/util/XprHelper.h" #include "src/Core/util/Memory.h" +#include "src/Core/util/IntegralConstant.h" +#include "src/Core/util/SymbolicIndex.h" #include "src/Core/NumTraits.h" #include "src/Core/MathFunctions.h" #include "src/Core/GenericPacketMath.h" - -#if defined EIGEN_VECTORIZE_SSE +#include "src/Core/MathFunctionsImpl.h" +#include "src/Core/arch/Default/ConjHelper.h" +// Generic half float support +#include "src/Core/arch/Default/Half.h" +#include "src/Core/arch/Default/BFloat16.h" +#include "src/Core/arch/Default/TypeCasting.h" +#include "src/Core/arch/Default/GenericPacketMathFunctionsFwd.h" + +#if defined EIGEN_VECTORIZE_AVX512 #include "src/Core/arch/SSE/PacketMath.h" + #include "src/Core/arch/SSE/TypeCasting.h" + #include "src/Core/arch/SSE/Complex.h" + #include "src/Core/arch/AVX/PacketMath.h" + #include "src/Core/arch/AVX/TypeCasting.h" + #include "src/Core/arch/AVX/Complex.h" + #include "src/Core/arch/AVX512/PacketMath.h" + #include "src/Core/arch/AVX512/TypeCasting.h" + #include "src/Core/arch/AVX512/Complex.h" #include "src/Core/arch/SSE/MathFunctions.h" + #include "src/Core/arch/AVX/MathFunctions.h" + #include "src/Core/arch/AVX512/MathFunctions.h" +#elif defined EIGEN_VECTORIZE_AVX + // Use AVX for floats and doubles, SSE for integers + #include "src/Core/arch/SSE/PacketMath.h" + #include "src/Core/arch/SSE/TypeCasting.h" #include "src/Core/arch/SSE/Complex.h" -#elif defined EIGEN_VECTORIZE_ALTIVEC + #include "src/Core/arch/AVX/PacketMath.h" + #include "src/Core/arch/AVX/TypeCasting.h" + #include "src/Core/arch/AVX/Complex.h" + #include "src/Core/arch/SSE/MathFunctions.h" + #include "src/Core/arch/AVX/MathFunctions.h" +#elif defined EIGEN_VECTORIZE_SSE + #include "src/Core/arch/SSE/PacketMath.h" + #include "src/Core/arch/SSE/TypeCasting.h" + #include "src/Core/arch/SSE/MathFunctions.h" + #include "src/Core/arch/SSE/Complex.h" +#elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX) #include "src/Core/arch/AltiVec/PacketMath.h" + #include "src/Core/arch/AltiVec/MathFunctions.h" #include "src/Core/arch/AltiVec/Complex.h" #elif defined EIGEN_VECTORIZE_NEON #include "src/Core/arch/NEON/PacketMath.h" + #include "src/Core/arch/NEON/TypeCasting.h" + #include "src/Core/arch/NEON/MathFunctions.h" #include "src/Core/arch/NEON/Complex.h" +#elif defined EIGEN_VECTORIZE_SVE + #include "src/Core/arch/SVE/PacketMath.h" + #include "src/Core/arch/SVE/TypeCasting.h" + #include "src/Core/arch/SVE/MathFunctions.h" +#elif defined EIGEN_VECTORIZE_ZVECTOR + #include "src/Core/arch/ZVector/PacketMath.h" + #include "src/Core/arch/ZVector/MathFunctions.h" + #include "src/Core/arch/ZVector/Complex.h" +#elif defined EIGEN_VECTORIZE_MSA + #include "src/Core/arch/MSA/PacketMath.h" + #include "src/Core/arch/MSA/MathFunctions.h" + #include "src/Core/arch/MSA/Complex.h" +#endif + +#if defined EIGEN_VECTORIZE_GPU + #include "src/Core/arch/GPU/PacketMath.h" + #include "src/Core/arch/GPU/MathFunctions.h" + #include "src/Core/arch/GPU/TypeCasting.h" +#endif + +#if defined(EIGEN_USE_SYCL) + #include "src/Core/arch/SYCL/SyclMemoryModel.h" + #include "src/Core/arch/SYCL/InteropHeaders.h" +#if !defined(EIGEN_DONT_VECTORIZE_SYCL) + #include "src/Core/arch/SYCL/PacketMath.h" + #include "src/Core/arch/SYCL/MathFunctions.h" + #include "src/Core/arch/SYCL/TypeCasting.h" +#endif #endif #include "src/Core/arch/Default/Settings.h" +// This file provides generic implementations valid for scalar as well +#include "src/Core/arch/Default/GenericPacketMathFunctions.h" + +#include "src/Core/functors/TernaryFunctors.h" +#include "src/Core/functors/BinaryFunctors.h" +#include "src/Core/functors/UnaryFunctors.h" +#include "src/Core/functors/NullaryFunctors.h" +#include "src/Core/functors/StlFunctors.h" +#include "src/Core/functors/AssignmentFunctors.h" + +// Specialized functors to enable the processing of complex numbers +// on CUDA devices +#ifdef EIGEN_CUDACC +#include "src/Core/arch/CUDA/Complex.h" +#endif -#include "src/Core/Functors.h" +#include "src/Core/util/IndexedViewHelper.h" +#include "src/Core/util/ReshapedHelper.h" +#include "src/Core/ArithmeticSequence.h" +#ifndef EIGEN_NO_IO + #include "src/Core/IO.h" +#endif #include "src/Core/DenseCoeffsBase.h" #include "src/Core/DenseBase.h" #include "src/Core/MatrixBase.h" #include "src/Core/EigenBase.h" +#include "src/Core/Product.h" +#include "src/Core/CoreEvaluators.h" +#include "src/Core/AssignEvaluator.h" + #ifndef EIGEN_PARSED_BY_DOXYGEN // work around Doxygen bug triggered by Assign.h r814874 // at least confirmed with Doxygen 1.5.5 and 1.5.6 #include "src/Core/Assign.h" #endif +#include "src/Core/ArrayBase.h" #include "src/Core/util/BlasUtil.h" #include "src/Core/DenseStorage.h" #include "src/Core/NestByValue.h" -#include "src/Core/ForceAlignedAccess.h" + +// #include "src/Core/ForceAlignedAccess.h" + #include "src/Core/ReturnByValue.h" #include "src/Core/NoAlias.h" #include "src/Core/PlainObjectBase.h" #include "src/Core/Matrix.h" #include "src/Core/Array.h" +#include "src/Core/CwiseTernaryOp.h" #include "src/Core/CwiseBinaryOp.h" #include "src/Core/CwiseUnaryOp.h" #include "src/Core/CwiseNullaryOp.h" @@ -300,32 +303,34 @@ using std::ptrdiff_t; #include "src/Core/SelfCwiseBinaryOp.h" #include "src/Core/Dot.h" #include "src/Core/StableNorm.h" -#include "src/Core/MapBase.h" #include "src/Core/Stride.h" +#include "src/Core/MapBase.h" #include "src/Core/Map.h" +#include "src/Core/Ref.h" #include "src/Core/Block.h" #include "src/Core/VectorBlock.h" -#include "src/Core/Ref.h" +#include "src/Core/IndexedView.h" +#include "src/Core/Reshaped.h" #include "src/Core/Transpose.h" #include "src/Core/DiagonalMatrix.h" #include "src/Core/Diagonal.h" #include "src/Core/DiagonalProduct.h" -#include "src/Core/PermutationMatrix.h" -#include "src/Core/Transpositions.h" #include "src/Core/Redux.h" #include "src/Core/Visitor.h" #include "src/Core/Fuzzy.h" -#include "src/Core/IO.h" #include "src/Core/Swap.h" #include "src/Core/CommaInitializer.h" -#include "src/Core/Flagged.h" -#include "src/Core/ProductBase.h" #include "src/Core/GeneralProduct.h" +#include "src/Core/Solve.h" +#include "src/Core/Inverse.h" +#include "src/Core/SolverBase.h" +#include "src/Core/PermutationMatrix.h" +#include "src/Core/Transpositions.h" #include "src/Core/TriangularMatrix.h" #include "src/Core/SelfAdjointView.h" #include "src/Core/products/GeneralBlockPanelKernel.h" #include "src/Core/products/Parallelizer.h" -#include "src/Core/products/CoeffBasedProduct.h" +#include "src/Core/ProductEvaluators.h" #include "src/Core/products/GeneralMatrixVector.h" #include "src/Core/products/GeneralMatrixMatrix.h" #include "src/Core/SolveTriangular.h" @@ -340,25 +345,33 @@ using std::ptrdiff_t; #include "src/Core/products/TriangularSolverVector.h" #include "src/Core/BandMatrix.h" #include "src/Core/CoreIterators.h" +#include "src/Core/ConditionEstimator.h" + +#if defined(EIGEN_VECTORIZE_VSX) + #include "src/Core/arch/AltiVec/MatrixProduct.h" +#elif defined EIGEN_VECTORIZE_NEON + #include "src/Core/arch/NEON/GeneralBlockPanelKernel.h" +#endif #include "src/Core/BooleanRedux.h" #include "src/Core/Select.h" #include "src/Core/VectorwiseOp.h" +#include "src/Core/PartialReduxEvaluator.h" #include "src/Core/Random.h" #include "src/Core/Replicate.h" #include "src/Core/Reverse.h" -#include "src/Core/ArrayBase.h" #include "src/Core/ArrayWrapper.h" +#include "src/Core/StlIterators.h" #ifdef EIGEN_USE_BLAS -#include "src/Core/products/GeneralMatrixMatrix_MKL.h" -#include "src/Core/products/GeneralMatrixVector_MKL.h" -#include "src/Core/products/GeneralMatrixMatrixTriangular_MKL.h" -#include "src/Core/products/SelfadjointMatrixMatrix_MKL.h" -#include "src/Core/products/SelfadjointMatrixVector_MKL.h" -#include "src/Core/products/TriangularMatrixMatrix_MKL.h" -#include "src/Core/products/TriangularMatrixVector_MKL.h" -#include "src/Core/products/TriangularSolverMatrix_MKL.h" +#include "src/Core/products/GeneralMatrixMatrix_BLAS.h" +#include "src/Core/products/GeneralMatrixVector_BLAS.h" +#include "src/Core/products/GeneralMatrixMatrixTriangular_BLAS.h" +#include "src/Core/products/SelfadjointMatrixMatrix_BLAS.h" +#include "src/Core/products/SelfadjointMatrixVector_BLAS.h" +#include "src/Core/products/TriangularMatrixMatrix_BLAS.h" +#include "src/Core/products/TriangularMatrixVector_BLAS.h" +#include "src/Core/products/TriangularSolverMatrix_BLAS.h" #endif // EIGEN_USE_BLAS #ifdef EIGEN_USE_MKL_VML @@ -369,8 +382,4 @@ using std::ptrdiff_t; #include "src/Core/util/ReenableStupidWarnings.h" -#ifdef EIGEN2_SUPPORT -#include "Eigen2Support" -#endif - #endif // EIGEN_CORE_H diff --git a/thirdparty/eigen/Eigen/Eigen b/thirdparty/eigen/Eigen/Eigen index 19b40ea4..654c8dc6 100644 --- a/thirdparty/eigen/Eigen/Eigen +++ b/thirdparty/eigen/Eigen/Eigen @@ -1,2 +1,2 @@ #include "Dense" -//#include "Sparse" +#include "Sparse" diff --git a/thirdparty/eigen/Eigen/Eigen2Support b/thirdparty/eigen/Eigen/Eigen2Support deleted file mode 100644 index 6aa009d2..00000000 --- a/thirdparty/eigen/Eigen/Eigen2Support +++ /dev/null @@ -1,95 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2009 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN2SUPPORT_H -#define EIGEN2SUPPORT_H - -#if (!defined(EIGEN2_SUPPORT)) || (!defined(EIGEN_CORE_H)) -#error Eigen2 support must be enabled by defining EIGEN2_SUPPORT before including any Eigen header -#endif - -#ifndef EIGEN_NO_EIGEN2_DEPRECATED_WARNING - -#if defined(__GNUC__) || defined(__INTEL_COMPILER) || defined(__clang__) -#warning "Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3. (Define EIGEN_NO_EIGEN2_DEPRECATED_WARNING to disable this warning)" -#else -#pragma message ("Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3. (Define EIGEN_NO_EIGEN2_DEPRECATED_WARNING to disable this warning)") -#endif - -#endif // EIGEN_NO_EIGEN2_DEPRECATED_WARNING - -#include "src/Core/util/DisableStupidWarnings.h" - -/** \ingroup Support_modules - * \defgroup Eigen2Support_Module Eigen2 support module - * - * \warning Eigen2 support is deprecated in Eigen 3.2.x and it will be removed in Eigen 3.3. - * - * This module provides a couple of deprecated functions improving the compatibility with Eigen2. - * - * To use it, define EIGEN2_SUPPORT before including any Eigen header - * \code - * #define EIGEN2_SUPPORT - * \endcode - * - */ - -#include "src/Eigen2Support/Macros.h" -#include "src/Eigen2Support/Memory.h" -#include "src/Eigen2Support/Meta.h" -#include "src/Eigen2Support/Lazy.h" -#include "src/Eigen2Support/Cwise.h" -#include "src/Eigen2Support/CwiseOperators.h" -#include "src/Eigen2Support/TriangularSolver.h" -#include "src/Eigen2Support/Block.h" -#include "src/Eigen2Support/VectorBlock.h" -#include "src/Eigen2Support/Minor.h" -#include "src/Eigen2Support/MathFunctions.h" - - -#include "src/Core/util/ReenableStupidWarnings.h" - -// Eigen2 used to include iostream -#include - -#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \ -using Eigen::Matrix##SizeSuffix##TypeSuffix; \ -using Eigen::Vector##SizeSuffix##TypeSuffix; \ -using Eigen::RowVector##SizeSuffix##TypeSuffix; - -#define EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(TypeSuffix) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 2) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 3) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, 4) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, X) \ - -#define EIGEN_USING_MATRIX_TYPEDEFS \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(i) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(f) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(d) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cf) \ -EIGEN_USING_MATRIX_TYPEDEFS_FOR_TYPE(cd) - -#define USING_PART_OF_NAMESPACE_EIGEN \ -EIGEN_USING_MATRIX_TYPEDEFS \ -using Eigen::Matrix; \ -using Eigen::MatrixBase; \ -using Eigen::ei_random; \ -using Eigen::ei_real; \ -using Eigen::ei_imag; \ -using Eigen::ei_conj; \ -using Eigen::ei_abs; \ -using Eigen::ei_abs2; \ -using Eigen::ei_sqrt; \ -using Eigen::ei_exp; \ -using Eigen::ei_log; \ -using Eigen::ei_sin; \ -using Eigen::ei_cos; - -#endif // EIGEN2SUPPORT_H diff --git a/thirdparty/eigen/Eigen/Eigenvalues b/thirdparty/eigen/Eigen/Eigenvalues index 53c5a73a..5467a2e7 100644 --- a/thirdparty/eigen/Eigen/Eigenvalues +++ b/thirdparty/eigen/Eigen/Eigenvalues @@ -1,16 +1,23 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_EIGENVALUES_MODULE_H #define EIGEN_EIGENVALUES_MODULE_H #include "Core" -#include "src/Core/util/DisableStupidWarnings.h" - #include "Cholesky" #include "Jacobi" #include "Householder" #include "LU" #include "Geometry" +#include "src/Core/util/DisableStupidWarnings.h" + /** \defgroup Eigenvalues_Module Eigenvalues module * * @@ -25,6 +32,7 @@ * \endcode */ +#include "src/misc/RealSvd2x2.h" #include "src/Eigenvalues/Tridiagonalization.h" #include "src/Eigenvalues/RealSchur.h" #include "src/Eigenvalues/EigenSolver.h" @@ -37,12 +45,16 @@ #include "src/Eigenvalues/GeneralizedEigenSolver.h" #include "src/Eigenvalues/MatrixBaseEigenvalues.h" #ifdef EIGEN_USE_LAPACKE -#include "src/Eigenvalues/RealSchur_MKL.h" -#include "src/Eigenvalues/ComplexSchur_MKL.h" -#include "src/Eigenvalues/SelfAdjointEigenSolver_MKL.h" +#ifdef EIGEN_USE_MKL +#include "mkl_lapacke.h" +#else +#include "src/misc/lapacke.h" +#endif +#include "src/Eigenvalues/RealSchur_LAPACKE.h" +#include "src/Eigenvalues/ComplexSchur_LAPACKE.h" +#include "src/Eigenvalues/SelfAdjointEigenSolver_LAPACKE.h" #endif #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_EIGENVALUES_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/Geometry b/thirdparty/eigen/Eigen/Geometry index efd9d450..bc78110a 100644 --- a/thirdparty/eigen/Eigen/Geometry +++ b/thirdparty/eigen/Eigen/Geometry @@ -1,29 +1,32 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_GEOMETRY_MODULE_H #define EIGEN_GEOMETRY_MODULE_H #include "Core" -#include "src/Core/util/DisableStupidWarnings.h" - #include "SVD" #include "LU" #include -#ifndef M_PI -#define M_PI 3.14159265358979323846 -#endif +#include "src/Core/util/DisableStupidWarnings.h" /** \defgroup Geometry_Module Geometry module - * - * * * This module provides support for: * - fixed-size homogeneous transformations * - translation, scaling, 2D and 3D rotations - * - quaternions - * - \ref MatrixBase::cross() "cross product" - * - \ref MatrixBase::unitOrthogonal() "orthognal vector generation" - * - some linear components: parametrized-lines and hyperplanes + * - \link Quaternion quaternions \endlink + * - cross products (\ref MatrixBase::cross, \ref MatrixBase::cross3) + * - orthognal vector generation (\ref MatrixBase::unitOrthogonal) + * - some linear components: \link ParametrizedLine parametrized-lines \endlink and \link Hyperplane hyperplanes \endlink + * - \link AlignedBox axis aligned bounding boxes \endlink + * - \link umeyama least-square transformation fitting \endlink * * \code * #include @@ -33,31 +36,24 @@ #include "src/Geometry/OrthoMethods.h" #include "src/Geometry/EulerAngles.h" -#if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS - #include "src/Geometry/Homogeneous.h" - #include "src/Geometry/RotationBase.h" - #include "src/Geometry/Rotation2D.h" - #include "src/Geometry/Quaternion.h" - #include "src/Geometry/AngleAxis.h" - #include "src/Geometry/Transform.h" - #include "src/Geometry/Translation.h" - #include "src/Geometry/Scaling.h" - #include "src/Geometry/Hyperplane.h" - #include "src/Geometry/ParametrizedLine.h" - #include "src/Geometry/AlignedBox.h" - #include "src/Geometry/Umeyama.h" - - #if defined EIGEN_VECTORIZE_SSE - #include "src/Geometry/arch/Geometry_SSE.h" - #endif -#endif - -#ifdef EIGEN2_SUPPORT -#include "src/Eigen2Support/Geometry/All.h" +#include "src/Geometry/Homogeneous.h" +#include "src/Geometry/RotationBase.h" +#include "src/Geometry/Rotation2D.h" +#include "src/Geometry/Quaternion.h" +#include "src/Geometry/AngleAxis.h" +#include "src/Geometry/Transform.h" +#include "src/Geometry/Translation.h" +#include "src/Geometry/Scaling.h" +#include "src/Geometry/Hyperplane.h" +#include "src/Geometry/ParametrizedLine.h" +#include "src/Geometry/AlignedBox.h" +#include "src/Geometry/Umeyama.h" + +// Use the SSE optimized version whenever possible. +#if (defined EIGEN_VECTORIZE_SSE) || (defined EIGEN_VECTORIZE_NEON) +#include "src/Geometry/arch/Geometry_SIMD.h" #endif #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_GEOMETRY_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ - diff --git a/thirdparty/eigen/Eigen/Householder b/thirdparty/eigen/Eigen/Householder index 6e348db5..f2fa7996 100644 --- a/thirdparty/eigen/Eigen/Householder +++ b/thirdparty/eigen/Eigen/Householder @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_HOUSEHOLDER_MODULE_H #define EIGEN_HOUSEHOLDER_MODULE_H @@ -20,4 +27,3 @@ #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_HOUSEHOLDER_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/IterativeLinearSolvers b/thirdparty/eigen/Eigen/IterativeLinearSolvers index 0f4159dc..957d5750 100644 --- a/thirdparty/eigen/Eigen/IterativeLinearSolvers +++ b/thirdparty/eigen/Eigen/IterativeLinearSolvers @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_ITERATIVELINEARSOLVERS_MODULE_H #define EIGEN_ITERATIVELINEARSOLVERS_MODULE_H @@ -12,28 +19,29 @@ * This module currently provides iterative methods to solve problems of the form \c A \c x = \c b, where \c A is a squared matrix, usually very large and sparse. * Those solvers are accessible via the following classes: * - ConjugateGradient for selfadjoint (hermitian) matrices, + * - LeastSquaresConjugateGradient for rectangular least-square problems, * - BiCGSTAB for general square matrices. * * These iterative solvers are associated with some preconditioners: * - IdentityPreconditioner - not really useful - * - DiagonalPreconditioner - also called JAcobi preconditioner, work very well on diagonal dominant matrices. - * - IncompleteILUT - incomplete LU factorization with dual thresholding + * - DiagonalPreconditioner - also called Jacobi preconditioner, work very well on diagonal dominant matrices. + * - IncompleteLUT - incomplete LU factorization with dual thresholding * * Such problems can also be solved using the direct sparse decomposition modules: SparseCholesky, CholmodSupport, UmfPackSupport, SuperLUSupport. * - * \code - * #include - * \endcode + \code + #include + \endcode */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" - +#include "src/IterativeLinearSolvers/SolveWithGuess.h" #include "src/IterativeLinearSolvers/IterativeSolverBase.h" #include "src/IterativeLinearSolvers/BasicPreconditioners.h" #include "src/IterativeLinearSolvers/ConjugateGradient.h" +#include "src/IterativeLinearSolvers/LeastSquareConjugateGradient.h" #include "src/IterativeLinearSolvers/BiCGSTAB.h" #include "src/IterativeLinearSolvers/IncompleteLUT.h" +#include "src/IterativeLinearSolvers/IncompleteCholesky.h" #include "src/Core/util/ReenableStupidWarnings.h" diff --git a/thirdparty/eigen/Eigen/Jacobi b/thirdparty/eigen/Eigen/Jacobi index ba8a4dc3..43edc7a1 100644 --- a/thirdparty/eigen/Eigen/Jacobi +++ b/thirdparty/eigen/Eigen/Jacobi @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_JACOBI_MODULE_H #define EIGEN_JACOBI_MODULE_H @@ -22,5 +29,4 @@ #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_JACOBI_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/KLUSupport b/thirdparty/eigen/Eigen/KLUSupport new file mode 100644 index 00000000..b23d9053 --- /dev/null +++ b/thirdparty/eigen/Eigen/KLUSupport @@ -0,0 +1,41 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_KLUSUPPORT_MODULE_H +#define EIGEN_KLUSUPPORT_MODULE_H + +#include + +#include + +extern "C" { +#include +#include + } + +/** \ingroup Support_modules + * \defgroup KLUSupport_Module KLUSupport module + * + * This module provides an interface to the KLU library which is part of the suitesparse package. + * It provides the following factorization class: + * - class KLU: a sparse LU factorization, well-suited for circuit simulation. + * + * \code + * #include + * \endcode + * + * In order to use this module, the klu and btf headers must be accessible from the include paths, and your binary must be linked to the klu library and its dependencies. + * The dependencies depend on how umfpack has been compiled. + * For a cmake based project, you can use our FindKLU.cmake module to help you in this task. + * + */ + +#include "src/KLUSupport/KLUSupport.h" + +#include + +#endif // EIGEN_KLUSUPPORT_MODULE_H diff --git a/thirdparty/eigen/Eigen/LU b/thirdparty/eigen/Eigen/LU index db579550..1236ceb0 100644 --- a/thirdparty/eigen/Eigen/LU +++ b/thirdparty/eigen/Eigen/LU @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_LU_MODULE_H #define EIGEN_LU_MODULE_H @@ -16,26 +23,25 @@ * \endcode */ -#include "src/misc/Solve.h" #include "src/misc/Kernel.h" #include "src/misc/Image.h" #include "src/LU/FullPivLU.h" #include "src/LU/PartialPivLU.h" #ifdef EIGEN_USE_LAPACKE -#include "src/LU/PartialPivLU_MKL.h" +#ifdef EIGEN_USE_MKL +#include "mkl_lapacke.h" +#else +#include "src/misc/lapacke.h" #endif -#include "src/LU/Determinant.h" -#include "src/LU/Inverse.h" - -#if defined EIGEN_VECTORIZE_SSE - #include "src/LU/arch/Inverse_SSE.h" +#include "src/LU/PartialPivLU_LAPACKE.h" #endif +#include "src/LU/Determinant.h" +#include "src/LU/InverseImpl.h" -#ifdef EIGEN2_SUPPORT - #include "src/Eigen2Support/LU.h" +#if defined EIGEN_VECTORIZE_SSE || defined EIGEN_VECTORIZE_NEON + #include "src/LU/arch/InverseSize4.h" #endif #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_LU_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/LeastSquares b/thirdparty/eigen/Eigen/LeastSquares deleted file mode 100644 index 35137c25..00000000 --- a/thirdparty/eigen/Eigen/LeastSquares +++ /dev/null @@ -1,32 +0,0 @@ -#ifndef EIGEN_REGRESSION_MODULE_H -#define EIGEN_REGRESSION_MODULE_H - -#ifndef EIGEN2_SUPPORT -#error LeastSquares is only available in Eigen2 support mode (define EIGEN2_SUPPORT) -#endif - -// exclude from normal eigen3-only documentation -#ifdef EIGEN2_SUPPORT - -#include "Core" - -#include "src/Core/util/DisableStupidWarnings.h" - -#include "Eigenvalues" -#include "Geometry" - -/** \defgroup LeastSquares_Module LeastSquares module - * This module provides linear regression and related features. - * - * \code - * #include - * \endcode - */ - -#include "src/Eigen2Support/LeastSquares.h" - -#include "src/Core/util/ReenableStupidWarnings.h" - -#endif // EIGEN2_SUPPORT - -#endif // EIGEN_REGRESSION_MODULE_H diff --git a/thirdparty/eigen/Eigen/MetisSupport b/thirdparty/eigen/Eigen/MetisSupport index 6a113f7a..85c41bf3 100644 --- a/thirdparty/eigen/Eigen/MetisSupport +++ b/thirdparty/eigen/Eigen/MetisSupport @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_METISSUPPORT_MODULE_H #define EIGEN_METISSUPPORT_MODULE_H diff --git a/thirdparty/eigen/Eigen/OrderingMethods b/thirdparty/eigen/Eigen/OrderingMethods index 7c0f1fff..29691a62 100644 --- a/thirdparty/eigen/Eigen/OrderingMethods +++ b/thirdparty/eigen/Eigen/OrderingMethods @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_ORDERINGMETHODS_MODULE_H #define EIGEN_ORDERINGMETHODS_MODULE_H @@ -56,10 +63,7 @@ * \endcode */ -#ifndef EIGEN_MPL2_ONLY #include "src/OrderingMethods/Amd.h" -#endif - #include "src/OrderingMethods/Ordering.h" #include "src/Core/util/ReenableStupidWarnings.h" diff --git a/thirdparty/eigen/Eigen/PaStiXSupport b/thirdparty/eigen/Eigen/PaStiXSupport index 7c616ee5..234619ac 100644 --- a/thirdparty/eigen/Eigen/PaStiXSupport +++ b/thirdparty/eigen/Eigen/PaStiXSupport @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_PASTIXSUPPORT_MODULE_H #define EIGEN_PASTIXSUPPORT_MODULE_H @@ -5,7 +12,6 @@ #include "src/Core/util/DisableStupidWarnings.h" -#include extern "C" { #include #include @@ -30,17 +36,14 @@ extern "C" { * \endcode * * In order to use this module, the PaSTiX headers must be accessible from the include paths, and your binary must be linked to the PaSTiX library and its dependencies. + * This wrapper resuires PaStiX version 5.x compiled without MPI support. * The dependencies depend on how PaSTiX has been compiled. * For a cmake based project, you can use our FindPaSTiX.cmake module to help you in this task. * */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" - #include "src/PaStiXSupport/PaStiXSupport.h" - #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_PASTIXSUPPORT_MODULE_H diff --git a/thirdparty/eigen/Eigen/PardisoSupport b/thirdparty/eigen/Eigen/PardisoSupport index 99330ce7..340edf51 100644 --- a/thirdparty/eigen/Eigen/PardisoSupport +++ b/thirdparty/eigen/Eigen/PardisoSupport @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_PARDISOSUPPORT_MODULE_H #define EIGEN_PARDISOSUPPORT_MODULE_H @@ -7,8 +14,6 @@ #include -#include - /** \ingroup Support_modules * \defgroup PardisoSupport_Module PardisoSupport module * diff --git a/thirdparty/eigen/Eigen/QR b/thirdparty/eigen/Eigen/QR index ac5b0269..8465b62c 100644 --- a/thirdparty/eigen/Eigen/QR +++ b/thirdparty/eigen/Eigen/QR @@ -1,45 +1,50 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_QR_MODULE_H #define EIGEN_QR_MODULE_H #include "Core" -#include "src/Core/util/DisableStupidWarnings.h" - #include "Cholesky" #include "Jacobi" #include "Householder" +#include "src/Core/util/DisableStupidWarnings.h" + /** \defgroup QR_Module QR module * * * * This module provides various QR decompositions * This module also provides some MatrixBase methods, including: - * - MatrixBase::qr(), + * - MatrixBase::householderQr() + * - MatrixBase::colPivHouseholderQr() + * - MatrixBase::fullPivHouseholderQr() * * \code * #include * \endcode */ -#include "src/misc/Solve.h" #include "src/QR/HouseholderQR.h" #include "src/QR/FullPivHouseholderQR.h" #include "src/QR/ColPivHouseholderQR.h" +#include "src/QR/CompleteOrthogonalDecomposition.h" #ifdef EIGEN_USE_LAPACKE -#include "src/QR/HouseholderQR_MKL.h" -#include "src/QR/ColPivHouseholderQR_MKL.h" +#ifdef EIGEN_USE_MKL +#include "mkl_lapacke.h" +#else +#include "src/misc/lapacke.h" #endif - -#ifdef EIGEN2_SUPPORT -#include "src/Eigen2Support/QR.h" +#include "src/QR/HouseholderQR_LAPACKE.h" +#include "src/QR/ColPivHouseholderQR_LAPACKE.h" #endif #include "src/Core/util/ReenableStupidWarnings.h" -#ifdef EIGEN2_SUPPORT -#include "Eigenvalues" -#endif - #endif // EIGEN_QR_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/QtAlignedMalloc b/thirdparty/eigen/Eigen/QtAlignedMalloc index 46f7d83b..6fe82374 100644 --- a/thirdparty/eigen/Eigen/QtAlignedMalloc +++ b/thirdparty/eigen/Eigen/QtAlignedMalloc @@ -1,3 +1,9 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. #ifndef EIGEN_QTMALLOC_MODULE_H #define EIGEN_QTMALLOC_MODULE_H @@ -8,7 +14,7 @@ #include "src/Core/util/DisableStupidWarnings.h" -void *qMalloc(size_t size) +void *qMalloc(std::size_t size) { return Eigen::internal::aligned_malloc(size); } @@ -18,10 +24,10 @@ void qFree(void *ptr) Eigen::internal::aligned_free(ptr); } -void *qRealloc(void *ptr, size_t size) +void *qRealloc(void *ptr, std::size_t size) { void* newPtr = Eigen::internal::aligned_malloc(size); - memcpy(newPtr, ptr, size); + std::memcpy(newPtr, ptr, size); Eigen::internal::aligned_free(ptr); return newPtr; } @@ -31,4 +37,3 @@ void *qRealloc(void *ptr, size_t size) #endif #endif // EIGEN_QTMALLOC_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/SPQRSupport b/thirdparty/eigen/Eigen/SPQRSupport index 7f1eb477..f70390c1 100644 --- a/thirdparty/eigen/Eigen/SPQRSupport +++ b/thirdparty/eigen/Eigen/SPQRSupport @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_SPQRSUPPORT_MODULE_H #define EIGEN_SPQRSUPPORT_MODULE_H @@ -21,8 +28,6 @@ * */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" #include "src/CholmodSupport/CholmodSupport.h" #include "src/SPQRSupport/SuiteSparseQRSupport.h" diff --git a/thirdparty/eigen/Eigen/SVD b/thirdparty/eigen/Eigen/SVD index fd310017..34517949 100644 --- a/thirdparty/eigen/Eigen/SVD +++ b/thirdparty/eigen/Eigen/SVD @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_SVD_MODULE_H #define EIGEN_SVD_MODULE_H @@ -12,26 +19,32 @@ * * * This module provides SVD decomposition for matrices (both real and complex). - * This decomposition is accessible via the following MatrixBase method: + * Two decomposition algorithms are provided: + * - JacobiSVD implementing two-sided Jacobi iterations is numerically very accurate, fast for small matrices, but very slow for larger ones. + * - BDCSVD implementing a recursive divide & conquer strategy on top of an upper-bidiagonalization which remains fast for large problems. + * These decompositions are accessible via the respective classes and following MatrixBase methods: * - MatrixBase::jacobiSvd() + * - MatrixBase::bdcSvd() * * \code * #include * \endcode */ -#include "src/misc/Solve.h" +#include "src/misc/RealSvd2x2.h" +#include "src/SVD/UpperBidiagonalization.h" +#include "src/SVD/SVDBase.h" #include "src/SVD/JacobiSVD.h" +#include "src/SVD/BDCSVD.h" #if defined(EIGEN_USE_LAPACKE) && !defined(EIGEN_USE_LAPACKE_STRICT) -#include "src/SVD/JacobiSVD_MKL.h" +#ifdef EIGEN_USE_MKL +#include "mkl_lapacke.h" +#else +#include "src/misc/lapacke.h" #endif -#include "src/SVD/UpperBidiagonalization.h" - -#ifdef EIGEN2_SUPPORT -#include "src/Eigen2Support/SVD.h" +#include "src/SVD/JacobiSVD_LAPACKE.h" #endif #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_SVD_MODULE_H -/* vim: set filetype=cpp et sw=2 ts=2 ai: */ diff --git a/thirdparty/eigen/Eigen/Sparse b/thirdparty/eigen/Eigen/Sparse index 7cc9c091..a2ef7a66 100644 --- a/thirdparty/eigen/Eigen/Sparse +++ b/thirdparty/eigen/Eigen/Sparse @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_SPARSE_MODULE_H #define EIGEN_SPARSE_MODULE_H @@ -11,9 +18,9 @@ * - \ref SparseQR_Module * - \ref IterativeLinearSolvers_Module * - * \code - * #include - * \endcode + \code + #include + \endcode */ #include "SparseCore" diff --git a/thirdparty/eigen/Eigen/SparseCholesky b/thirdparty/eigen/Eigen/SparseCholesky index 9f5056aa..d2b1f127 100644 --- a/thirdparty/eigen/Eigen/SparseCholesky +++ b/thirdparty/eigen/Eigen/SparseCholesky @@ -30,18 +30,8 @@ * \endcode */ -#ifdef EIGEN_MPL2_ONLY -#error The SparseCholesky module has nothing to offer in MPL2 only mode -#endif - -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" #include "src/SparseCholesky/SimplicialCholesky.h" - -#ifndef EIGEN_MPL2_ONLY #include "src/SparseCholesky/SimplicialCholesky_impl.h" -#endif - #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_SPARSECHOLESKY_MODULE_H diff --git a/thirdparty/eigen/Eigen/SparseCore b/thirdparty/eigen/Eigen/SparseCore index 24bcf015..76966c4c 100644 --- a/thirdparty/eigen/Eigen/SparseCore +++ b/thirdparty/eigen/Eigen/SparseCore @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_SPARSECORE_MODULE_H #define EIGEN_SPARSECORE_MODULE_H @@ -26,37 +33,35 @@ * This module depends on: Core. */ -namespace Eigen { - -/** The type used to identify a general sparse storage. */ -struct Sparse {}; - -} - #include "src/SparseCore/SparseUtil.h" #include "src/SparseCore/SparseMatrixBase.h" +#include "src/SparseCore/SparseAssign.h" #include "src/SparseCore/CompressedStorage.h" #include "src/SparseCore/AmbiVector.h" +#include "src/SparseCore/SparseCompressedBase.h" #include "src/SparseCore/SparseMatrix.h" +#include "src/SparseCore/SparseMap.h" #include "src/SparseCore/MappedSparseMatrix.h" #include "src/SparseCore/SparseVector.h" -#include "src/SparseCore/SparseBlock.h" -#include "src/SparseCore/SparseTranspose.h" +#include "src/SparseCore/SparseRef.h" #include "src/SparseCore/SparseCwiseUnaryOp.h" #include "src/SparseCore/SparseCwiseBinaryOp.h" +#include "src/SparseCore/SparseTranspose.h" +#include "src/SparseCore/SparseBlock.h" #include "src/SparseCore/SparseDot.h" -#include "src/SparseCore/SparsePermutation.h" #include "src/SparseCore/SparseRedux.h" -#include "src/SparseCore/SparseFuzzy.h" +#include "src/SparseCore/SparseView.h" +#include "src/SparseCore/SparseDiagonalProduct.h" #include "src/SparseCore/ConservativeSparseSparseProduct.h" #include "src/SparseCore/SparseSparseProductWithPruning.h" #include "src/SparseCore/SparseProduct.h" #include "src/SparseCore/SparseDenseProduct.h" -#include "src/SparseCore/SparseDiagonalProduct.h" -#include "src/SparseCore/SparseTriangularView.h" #include "src/SparseCore/SparseSelfAdjointView.h" +#include "src/SparseCore/SparseTriangularView.h" #include "src/SparseCore/TriangularSolver.h" -#include "src/SparseCore/SparseView.h" +#include "src/SparseCore/SparsePermutation.h" +#include "src/SparseCore/SparseFuzzy.h" +#include "src/SparseCore/SparseSolverBase.h" #include "src/Core/util/ReenableStupidWarnings.h" diff --git a/thirdparty/eigen/Eigen/SparseLU b/thirdparty/eigen/Eigen/SparseLU index 8527a49b..047cf0dc 100644 --- a/thirdparty/eigen/Eigen/SparseLU +++ b/thirdparty/eigen/Eigen/SparseLU @@ -20,13 +20,10 @@ * Please, see the documentation of the SparseLU class for more details. */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" - // Ordering interface #include "OrderingMethods" -#include "src/SparseLU/SparseLU_gemm_kernel.h" +#include "src/Core/util/DisableStupidWarnings.h" #include "src/SparseLU/SparseLU_Structs.h" #include "src/SparseLU/SparseLU_SupernodalMatrix.h" @@ -46,4 +43,6 @@ #include "src/SparseLU/SparseLU_Utils.h" #include "src/SparseLU/SparseLU.h" +#include "src/Core/util/ReenableStupidWarnings.h" + #endif // EIGEN_SPARSELU_MODULE_H diff --git a/thirdparty/eigen/Eigen/SparseQR b/thirdparty/eigen/Eigen/SparseQR index 4ee42065..f5fc5fa7 100644 --- a/thirdparty/eigen/Eigen/SparseQR +++ b/thirdparty/eigen/Eigen/SparseQR @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_SPARSEQR_MODULE_H #define EIGEN_SPARSEQR_MODULE_H @@ -21,10 +28,6 @@ * */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" - -#include "OrderingMethods" #include "src/SparseCore/SparseColEtree.h" #include "src/SparseQR/SparseQR.h" diff --git a/thirdparty/eigen/Eigen/StdDeque b/thirdparty/eigen/Eigen/StdDeque index f2723477..bc68397b 100644 --- a/thirdparty/eigen/Eigen/StdDeque +++ b/thirdparty/eigen/Eigen/StdDeque @@ -14,7 +14,7 @@ #include "Core" #include -#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */ +#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */ #define EIGEN_DEFINE_STL_DEQUE_SPECIALIZATION(...) diff --git a/thirdparty/eigen/Eigen/StdList b/thirdparty/eigen/Eigen/StdList index 225c1e18..4c6262c0 100644 --- a/thirdparty/eigen/Eigen/StdList +++ b/thirdparty/eigen/Eigen/StdList @@ -13,7 +13,7 @@ #include "Core" #include -#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */ +#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */ #define EIGEN_DEFINE_STL_LIST_SPECIALIZATION(...) diff --git a/thirdparty/eigen/Eigen/StdVector b/thirdparty/eigen/Eigen/StdVector index 6b22627f..0c4697ad 100644 --- a/thirdparty/eigen/Eigen/StdVector +++ b/thirdparty/eigen/Eigen/StdVector @@ -14,7 +14,7 @@ #include "Core" #include -#if (defined(_MSC_VER) && defined(_WIN64)) /* MSVC auto aligns in 64 bit builds */ +#if EIGEN_COMP_MSVC && EIGEN_OS_WIN64 && (EIGEN_MAX_STATIC_ALIGN_BYTES<=16) /* MSVC auto aligns up to 16 bytes in 64 bit builds */ #define EIGEN_DEFINE_STL_VECTOR_SPECIALIZATION(...) diff --git a/thirdparty/eigen/Eigen/SuperLUSupport b/thirdparty/eigen/Eigen/SuperLUSupport index 575e14fb..59312a82 100644 --- a/thirdparty/eigen/Eigen/SuperLUSupport +++ b/thirdparty/eigen/Eigen/SuperLUSupport @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_SUPERLUSUPPORT_MODULE_H #define EIGEN_SUPERLUSUPPORT_MODULE_H @@ -36,6 +43,8 @@ namespace Eigen { struct SluMatrix; } * - class SuperLU: a supernodal sequential LU factorization. * - class SuperILU: a supernodal sequential incomplete LU factorization (to be used as a preconditioner for iterative methods). * + * \warning This wrapper requires at least versions 4.0 of SuperLU. The 3.x versions are not supported. + * * \warning When including this module, you have to use SUPERLU_EMPTY instead of EMPTY which is no longer defined because it is too polluting. * * \code @@ -48,12 +57,8 @@ namespace Eigen { struct SluMatrix; } * */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" - #include "src/SuperLUSupport/SuperLUSupport.h" - #include "src/Core/util/ReenableStupidWarnings.h" #endif // EIGEN_SUPERLUSUPPORT_MODULE_H diff --git a/thirdparty/eigen/Eigen/UmfPackSupport b/thirdparty/eigen/Eigen/UmfPackSupport index 7b1b6606..00eec808 100644 --- a/thirdparty/eigen/Eigen/UmfPackSupport +++ b/thirdparty/eigen/Eigen/UmfPackSupport @@ -1,3 +1,10 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #ifndef EIGEN_UMFPACKSUPPORT_MODULE_H #define EIGEN_UMFPACKSUPPORT_MODULE_H @@ -26,9 +33,6 @@ extern "C" { * */ -#include "src/misc/Solve.h" -#include "src/misc/SparseSolve.h" - #include "src/UmfPackSupport/UmfPackSupport.h" #include "src/Core/util/ReenableStupidWarnings.h" diff --git a/thirdparty/eigen/Eigen/src/CMakeLists.txt b/thirdparty/eigen/Eigen/src/CMakeLists.txt deleted file mode 100644 index c326f374..00000000 --- a/thirdparty/eigen/Eigen/src/CMakeLists.txt +++ /dev/null @@ -1,7 +0,0 @@ -file(GLOB Eigen_src_subdirectories "*") -escape_string_as_regex(ESCAPED_CMAKE_CURRENT_SOURCE_DIR "${CMAKE_CURRENT_SOURCE_DIR}") -foreach(f ${Eigen_src_subdirectories}) - if(NOT f MATCHES "\\.txt" AND NOT f MATCHES "${ESCAPED_CMAKE_CURRENT_SOURCE_DIR}/[.].+" ) - add_subdirectory(${f}) - endif() -endforeach() diff --git a/thirdparty/eigen/Eigen/src/Cholesky/CMakeLists.txt b/thirdparty/eigen/Eigen/src/Cholesky/CMakeLists.txt deleted file mode 100644 index d01488b4..00000000 --- a/thirdparty/eigen/Eigen/src/Cholesky/CMakeLists.txt +++ /dev/null @@ -1,6 +0,0 @@ -FILE(GLOB Eigen_Cholesky_SRCS "*.h") - -INSTALL(FILES - ${Eigen_Cholesky_SRCS} - DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Cholesky COMPONENT Devel - ) diff --git a/thirdparty/eigen/Eigen/src/Cholesky/LDLT.h b/thirdparty/eigen/Eigen/src/Cholesky/LDLT.h index abd30bd9..1013ca04 100644 --- a/thirdparty/eigen/Eigen/src/Cholesky/LDLT.h +++ b/thirdparty/eigen/Eigen/src/Cholesky/LDLT.h @@ -13,9 +13,18 @@ #ifndef EIGEN_LDLT_H #define EIGEN_LDLT_H -namespace Eigen { +namespace Eigen { namespace internal { + template struct traits > + : traits<_MatrixType> + { + typedef MatrixXpr XprKind; + typedef SolverStorage StorageKind; + typedef int StorageIndex; + enum { Flags = 0 }; + }; + template struct LDLT_Traits; // PositiveSemiDef means positive semi-definite and non-zero; same for NegativeSemiDef @@ -28,39 +37,40 @@ namespace internal { * * \brief Robust Cholesky decomposition of a matrix with pivoting * - * \param MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition - * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper. + * \tparam _MatrixType the type of the matrix of which to compute the LDL^T Cholesky decomposition + * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper. * The other triangular part won't be read. * * Perform a robust Cholesky decomposition of a positive semidefinite or negative semidefinite * matrix \f$ A \f$ such that \f$ A = P^TLDL^*P \f$, where P is a permutation matrix, L * is lower triangular with a unit diagonal and D is a diagonal matrix. * - * The decomposition uses pivoting to ensure stability, so that L will have + * The decomposition uses pivoting to ensure stability, so that D will have * zeros in the bottom right rank(A) - n submatrix. Avoiding the square root * on D also stabilizes the computation. * * Remember that Cholesky decompositions are not rank-revealing. Also, do not use a Cholesky * decomposition to determine whether a system of equations has a solution. * - * \sa MatrixBase::ldlt(), class LLT + * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism. + * + * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt(), class LLT */ template class LDLT + : public SolverBase > { public: typedef _MatrixType MatrixType; + typedef SolverBase Base; + friend class SolverBase; + + EIGEN_GENERIC_PUBLIC_INTERFACE(LDLT) enum { - RowsAtCompileTime = MatrixType::RowsAtCompileTime, - ColsAtCompileTime = MatrixType::ColsAtCompileTime, - Options = MatrixType::Options & ~RowMajorBit, // these are the options for the TmpMatrixType, we need a ColMajor matrix here! MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime, MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime, UpLo = _UpLo }; - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef typename MatrixType::Index Index; - typedef Matrix TmpMatrixType; + typedef Matrix TmpMatrixType; typedef Transpositions TranspositionType; typedef PermutationMatrix PermutationType; @@ -72,11 +82,11 @@ template class LDLT * The default constructor is useful in cases in which the user intends to * perform decompositions via LDLT::compute(const MatrixType&). */ - LDLT() - : m_matrix(), - m_transpositions(), + LDLT() + : m_matrix(), + m_transpositions(), m_sign(internal::ZeroSign), - m_isInitialized(false) + m_isInitialized(false) {} /** \brief Default Constructor with memory preallocation @@ -85,7 +95,7 @@ template class LDLT * according to the specified problem \a size. * \sa LDLT() */ - LDLT(Index size) + explicit LDLT(Index size) : m_matrix(size, size), m_transpositions(size), m_temporary(size), @@ -96,16 +106,35 @@ template class LDLT /** \brief Constructor with decomposition * * This calculates the decomposition for the input \a matrix. + * * \sa LDLT(Index size) */ - LDLT(const MatrixType& matrix) + template + explicit LDLT(const EigenBase& matrix) : m_matrix(matrix.rows(), matrix.cols()), m_transpositions(matrix.rows()), m_temporary(matrix.rows()), m_sign(internal::ZeroSign), m_isInitialized(false) { - compute(matrix); + compute(matrix.derived()); + } + + /** \brief Constructs a LDLT factorization from a given matrix + * + * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when \c MatrixType is a Eigen::Ref. + * + * \sa LDLT(const EigenBase&) + */ + template + explicit LDLT(EigenBase& matrix) + : m_matrix(matrix.derived()), + m_transpositions(matrix.rows()), + m_temporary(matrix.rows()), + m_sign(internal::ZeroSign), + m_isInitialized(false) + { + compute(matrix.derived()); } /** Clear any existing decomposition @@ -151,13 +180,6 @@ template class LDLT eigen_assert(m_isInitialized && "LDLT is not initialized."); return m_sign == internal::PositiveSemiDef || m_sign == internal::ZeroSign; } - - #ifdef EIGEN2_SUPPORT - inline bool isPositiveDefinite() const - { - return isPositive(); - } - #endif /** \returns true if the matrix is negative (semidefinite) */ inline bool isNegative(void) const @@ -166,6 +188,7 @@ template class LDLT return m_sign == internal::NegativeSemiDef || m_sign == internal::ZeroSign; } + #ifdef EIGEN_PARSED_BY_DOXYGEN /** \returns a solution x of \f$ A x = b \f$ using the current decomposition of A. * * This function also supports in-place solves using the syntax x = decompositionObject.solve(x) . @@ -173,37 +196,33 @@ template class LDLT * \note_about_checking_solutions * * More precisely, this method solves \f$ A x = b \f$ using the decomposition \f$ A = P^T L D L^* P \f$ - * by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$, + * by solving the systems \f$ P^T y_1 = b \f$, \f$ L y_2 = y_1 \f$, \f$ D y_3 = y_2 \f$, * \f$ L^* y_4 = y_3 \f$ and \f$ P x = y_4 \f$ in succession. If the matrix \f$ A \f$ is singular, then * \f$ D \f$ will also be singular (all the other matrices are invertible). In that case, the * least-square solution of \f$ D y_3 = y_2 \f$ is computed. This does not mean that this function - * computes the least-square solution of \f$ A x = b \f$ is \f$ A \f$ is singular. + * computes the least-square solution of \f$ A x = b \f$ if \f$ A \f$ is singular. * - * \sa MatrixBase::ldlt() + * \sa MatrixBase::ldlt(), SelfAdjointView::ldlt() */ template - inline const internal::solve_retval - solve(const MatrixBase& b) const - { - eigen_assert(m_isInitialized && "LDLT is not initialized."); - eigen_assert(m_matrix.rows()==b.rows() - && "LDLT::solve(): invalid number of rows of the right hand side matrix b"); - return internal::solve_retval(*this, b.derived()); - } - - #ifdef EIGEN2_SUPPORT - template - bool solve(const MatrixBase& b, ResultType *result) const - { - *result = this->solve(b); - return true; - } + inline const Solve + solve(const MatrixBase& b) const; #endif template bool solveInPlace(MatrixBase &bAndX) const; - LDLT& compute(const MatrixType& matrix); + template + LDLT& compute(const EigenBase& matrix); + + /** \returns an estimate of the reciprocal condition number of the matrix of + * which \c *this is the LDLT decomposition. + */ + RealScalar rcond() const + { + eigen_assert(m_isInitialized && "LDLT is not initialized."); + return internal::rcond_estimate_helper(m_l1_norm, *this); + } template LDLT& rankUpdate(const MatrixBase& w, const RealScalar& alpha=1); @@ -220,22 +239,37 @@ template class LDLT MatrixType reconstructedMatrix() const; - inline Index rows() const { return m_matrix.rows(); } - inline Index cols() const { return m_matrix.cols(); } + /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint. + * + * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as: + * \code x = decomposition.adjoint().solve(b) \endcode + */ + const LDLT& adjoint() const { return *this; }; + + EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); } + EIGEN_DEVICE_FUNC inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); } /** \brief Reports whether previous computation was successful. * - * \returns \c Success if computation was succesful, - * \c NumericalIssue if the matrix.appears to be negative. + * \returns \c Success if computation was successful, + * \c NumericalIssue if the factorization failed because of a zero pivot. */ ComputationInfo info() const { eigen_assert(m_isInitialized && "LDLT is not initialized."); - return Success; + return m_info; } + #ifndef EIGEN_PARSED_BY_DOXYGEN + template + void _solve_impl(const RhsType &rhs, DstType &dst) const; + + template + void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const; + #endif + protected: - + static void check_template_parameters() { EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar); @@ -248,10 +282,12 @@ template class LDLT * is not stored), and the diagonal entries correspond to D. */ MatrixType m_matrix; + RealScalar m_l1_norm; TranspositionType m_transpositions; TmpMatrixType m_temporary; internal::SignMatrix m_sign; bool m_isInitialized; + ComputationInfo m_info; }; namespace internal { @@ -266,15 +302,18 @@ template<> struct ldlt_inplace using std::abs; typedef typename MatrixType::Scalar Scalar; typedef typename MatrixType::RealScalar RealScalar; - typedef typename MatrixType::Index Index; + typedef typename TranspositionType::StorageIndex IndexType; eigen_assert(mat.rows()==mat.cols()); const Index size = mat.rows(); + bool found_zero_pivot = false; + bool ret = true; if (size <= 1) { transpositions.setIdentity(); - if (numext::real(mat.coeff(0,0)) > 0) sign = PositiveSemiDef; - else if (numext::real(mat.coeff(0,0)) < 0) sign = NegativeSemiDef; + if(size==0) sign = ZeroSign; + else if (numext::real(mat.coeff(0,0)) > static_cast(0) ) sign = PositiveSemiDef; + else if (numext::real(mat.coeff(0,0)) < static_cast(0)) sign = NegativeSemiDef; else sign = ZeroSign; return true; } @@ -286,7 +325,7 @@ template<> struct ldlt_inplace mat.diagonal().tail(size-k).cwiseAbs().maxCoeff(&index_of_biggest_in_corner); index_of_biggest_in_corner += k; - transpositions.coeffRef(k) = index_of_biggest_in_corner; + transpositions.coeffRef(k) = IndexType(index_of_biggest_in_corner); if(k != index_of_biggest_in_corner) { // apply the transposition while taking care to consider only @@ -295,7 +334,7 @@ template<> struct ldlt_inplace mat.row(k).head(k).swap(mat.row(index_of_biggest_in_corner).head(k)); mat.col(k).tail(s).swap(mat.col(index_of_biggest_in_corner).tail(s)); std::swap(mat.coeffRef(k,k),mat.coeffRef(index_of_biggest_in_corner,index_of_biggest_in_corner)); - for(int i=k+1;i struct ldlt_inplace if(rs>0) A21.noalias() -= A20 * temp.head(k); } - + // In some previous versions of Eigen (e.g., 3.2.1), the scaling was omitted if the pivot - // was smaller than the cutoff value. However, soince LDLT is not rank-revealing - // we should only make sure we do not introduce INF or NaN values. - // LAPACK also uses 0 as the cutoff value. + // was smaller than the cutoff value. However, since LDLT is not rank-revealing + // we should only make sure that we do not introduce INF or NaN values. + // Remark that LAPACK also uses 0 as the cutoff value. RealScalar realAkk = numext::real(mat.coeffRef(k,k)); - if((rs>0) && (abs(realAkk) > RealScalar(0))) + bool pivot_is_valid = (abs(realAkk) > RealScalar(0)); + + if(k==0 && !pivot_is_valid) + { + // The entire diagonal is zero, there is nothing more to do + // except filling the transpositions, and checking whether the matrix is zero. + sign = ZeroSign; + for(Index j = 0; j0) && pivot_is_valid) A21 /= realAkk; + else if(rs>0) + ret = ret && (A21.array()==Scalar(0)).all(); + + if(found_zero_pivot && pivot_is_valid) ret = false; // factorization failed + else if(!pivot_is_valid) found_zero_pivot = true; if (sign == PositiveSemiDef) { - if (realAkk < 0) sign = Indefinite; + if (realAkk < static_cast(0)) sign = Indefinite; } else if (sign == NegativeSemiDef) { - if (realAkk > 0) sign = Indefinite; + if (realAkk > static_cast(0)) sign = Indefinite; } else if (sign == ZeroSign) { - if (realAkk > 0) sign = PositiveSemiDef; - else if (realAkk < 0) sign = NegativeSemiDef; + if (realAkk > static_cast(0)) sign = PositiveSemiDef; + else if (realAkk < static_cast(0)) sign = NegativeSemiDef; } } - return true; + return ret; } // Reference for the algorithm: Davis and Hager, "Multiple Rank @@ -356,7 +415,6 @@ template<> struct ldlt_inplace using numext::isfinite; typedef typename MatrixType::Scalar Scalar; typedef typename MatrixType::RealScalar RealScalar; - typedef typename MatrixType::Index Index; const Index size = mat.rows(); eigen_assert(mat.cols() == size && w.size()==size); @@ -420,16 +478,16 @@ template struct LDLT_Traits { typedef const TriangularView MatrixL; typedef const TriangularView MatrixU; - static inline MatrixL getL(const MatrixType& m) { return m; } - static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); } + static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); } + static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); } }; template struct LDLT_Traits { typedef const TriangularView MatrixL; typedef const TriangularView MatrixU; - static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); } - static inline MatrixU getU(const MatrixType& m) { return m; } + static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); } + static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); } }; } // end namespace internal @@ -437,21 +495,35 @@ template struct LDLT_Traits /** Compute / recompute the LDLT decomposition A = L D L^* = U^* D U of \a matrix */ template -LDLT& LDLT::compute(const MatrixType& a) +template +LDLT& LDLT::compute(const EigenBase& a) { check_template_parameters(); - + eigen_assert(a.rows()==a.cols()); const Index size = a.rows(); - m_matrix = a; + m_matrix = a.derived(); + + // Compute matrix L1 norm = max abs column sum. + m_l1_norm = RealScalar(0); + // TODO move this code to SelfAdjointView + for (Index col = 0; col < size; ++col) { + RealScalar abs_col_sum; + if (_UpLo == Lower) + abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>(); + else + abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>(); + if (abs_col_sum > m_l1_norm) + m_l1_norm = abs_col_sum; + } m_transpositions.resize(size); m_isInitialized = false; m_temporary.resize(size); m_sign = internal::ZeroSign; - internal::ldlt_inplace::unblocked(m_matrix, m_transpositions, m_temporary, m_sign); + m_info = internal::ldlt_inplace::unblocked(m_matrix, m_transpositions, m_temporary, m_sign) ? Success : NumericalIssue; m_isInitialized = true; return *this; @@ -466,18 +538,19 @@ template template LDLT& LDLT::rankUpdate(const MatrixBase& w, const typename LDLT::RealScalar& sigma) { + typedef typename TranspositionType::StorageIndex IndexType; const Index size = w.rows(); if (m_isInitialized) { eigen_assert(m_matrix.rows()==size); } else - { + { m_matrix.resize(size,size); m_matrix.setZero(); m_transpositions.resize(size); for (Index i = 0; i < size; i++) - m_transpositions.coeffRef(i) = i; + m_transpositions.coeffRef(i) = IndexType(i); m_temporary.resize(size); m_sign = sigma>=0 ? internal::PositiveSemiDef : internal::NegativeSemiDef; m_isInitialized = true; @@ -488,53 +561,55 @@ LDLT& LDLT::rankUpdate(const MatrixBase -struct solve_retval, Rhs> - : solve_retval_base, Rhs> +#ifndef EIGEN_PARSED_BY_DOXYGEN +template +template +void LDLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const { - typedef LDLT<_MatrixType,_UpLo> LDLTType; - EIGEN_MAKE_SOLVE_HELPERS(LDLTType,Rhs) + _solve_impl_transposed(rhs, dst); +} - template void evalTo(Dest& dst) const +template +template +void LDLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const +{ + // dst = P b + dst = m_transpositions * rhs; + + // dst = L^-1 (P b) + // dst = L^-*T (P b) + matrixL().template conjugateIf().solveInPlace(dst); + + // dst = D^-* (L^-1 P b) + // dst = D^-1 (L^-*T P b) + // more precisely, use pseudo-inverse of D (see bug 241) + using std::abs; + const typename Diagonal::RealReturnType vecD(vectorD()); + // In some previous versions, tolerance was set to the max of 1/highest (or rather numeric_limits::min()) + // and the maximal diagonal entry * epsilon as motivated by LAPACK's xGELSS: + // RealScalar tolerance = numext::maxi(vecD.array().abs().maxCoeff() * NumTraits::epsilon(),RealScalar(1) / NumTraits::highest()); + // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest + // diagonal element is not well justified and leads to numerical issues in some cases. + // Moreover, Lapack's xSYTRS routines use 0 for the tolerance. + // Using numeric_limits::min() gives us more robustness to denormals. + RealScalar tolerance = (std::numeric_limits::min)(); + for (Index i = 0; i < vecD.size(); ++i) { - eigen_assert(rhs().rows() == dec().matrixLDLT().rows()); - // dst = P b - dst = dec().transpositionsP() * rhs(); - - // dst = L^-1 (P b) - dec().matrixL().solveInPlace(dst); - - // dst = D^-1 (L^-1 P b) - // more precisely, use pseudo-inverse of D (see bug 241) - using std::abs; - using std::max; - typedef typename LDLTType::MatrixType MatrixType; - typedef typename LDLTType::RealScalar RealScalar; - const typename Diagonal::RealReturnType vectorD(dec().vectorD()); - // In some previous versions, tolerance was set to the max of 1/highest and the maximal diagonal entry * epsilon - // as motivated by LAPACK's xGELSS: - // RealScalar tolerance = (max)(vectorD.array().abs().maxCoeff() *NumTraits::epsilon(),RealScalar(1) / NumTraits::highest()); - // However, LDLT is not rank revealing, and so adjusting the tolerance wrt to the highest - // diagonal element is not well justified and to numerical issues in some cases. - // Moreover, Lapack's xSYTRS routines use 0 for the tolerance. - RealScalar tolerance = RealScalar(1) / NumTraits::highest(); - - for (Index i = 0; i < vectorD.size(); ++i) { - if(abs(vectorD(i)) > tolerance) - dst.row(i) /= vectorD(i); - else - dst.row(i).setZero(); - } + if(abs(vecD(i)) > tolerance) + dst.row(i) /= vecD(i); + else + dst.row(i).setZero(); + } - // dst = L^-T (D^-1 L^-1 P b) - dec().matrixU().solveInPlace(dst); + // dst = L^-* (D^-* L^-1 P b) + // dst = L^-T (D^-1 L^-*T P b) + matrixL().transpose().template conjugateIf().solveInPlace(dst); - // dst = P^-1 (L^-T D^-1 L^-1 P b) = A^-1 b - dst = dec().transpositionsP().transpose() * dst; - } -}; + // dst = P^T (L^-* D^-* L^-1 P b) = A^-1 b + // dst = P^-T (L^-T D^-1 L^-*T P b) = A^-1 b + dst = m_transpositions.transpose() * dst; } +#endif /** \internal use x = ldlt_object.solve(x); * @@ -588,6 +663,7 @@ MatrixType LDLT::reconstructedMatrix() const /** \cholesky_module * \returns the Cholesky decomposition with full pivoting without square root of \c *this + * \sa MatrixBase::ldlt() */ template inline const LDLT::PlainObject, UpLo> @@ -598,6 +674,7 @@ SelfAdjointView::ldlt() const /** \cholesky_module * \returns the Cholesky decomposition with full pivoting without square root of \c *this + * \sa SelfAdjointView::ldlt() */ template inline const LDLT::PlainObject> diff --git a/thirdparty/eigen/Eigen/src/Cholesky/LLT.h b/thirdparty/eigen/Eigen/src/Cholesky/LLT.h index 7c11a2dc..8c9b2b39 100644 --- a/thirdparty/eigen/Eigen/src/Cholesky/LLT.h +++ b/thirdparty/eigen/Eigen/src/Cholesky/LLT.h @@ -10,9 +10,19 @@ #ifndef EIGEN_LLT_H #define EIGEN_LLT_H -namespace Eigen { +namespace Eigen { namespace internal{ + +template struct traits > + : traits<_MatrixType> +{ + typedef MatrixXpr XprKind; + typedef SolverStorage StorageKind; + typedef int StorageIndex; + enum { Flags = 0 }; +}; + template struct LLT_Traits; } @@ -22,9 +32,9 @@ template struct LLT_Traits; * * \brief Standard Cholesky decomposition (LL^T) of a matrix and associated features * - * \param MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition - * \param UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper. - * The other triangular part won't be read. + * \tparam _MatrixType the type of the matrix of which we are computing the LL^T Cholesky decomposition + * \tparam _UpLo the triangular part that will be used for the decompositon: Lower (default) or Upper. + * The other triangular part won't be read. * * This class performs a LL^T Cholesky decomposition of a symmetric, positive definite * matrix A such that A = LL^* = U^*U, where L is lower triangular. @@ -40,26 +50,31 @@ template struct LLT_Traits; * * Example: \include LLT_example.cpp * Output: \verbinclude LLT_example.out - * - * \sa MatrixBase::llt(), class LDLT - */ - /* HEY THIS DOX IS DISABLED BECAUSE THERE's A BUG EITHER HERE OR IN LDLT ABOUT THAT (OR BOTH) - * Note that during the decomposition, only the upper triangular part of A is considered. Therefore, - * the strict lower part does not have to store correct values. + * + * \b Performance: for best performance, it is recommended to use a column-major storage format + * with the Lower triangular part (the default), or, equivalently, a row-major storage format + * with the Upper triangular part. Otherwise, you might get a 20% slowdown for the full factorization + * step, and rank-updates can be up to 3 times slower. + * + * This class supports the \link InplaceDecomposition inplace decomposition \endlink mechanism. + * + * Note that during the decomposition, only the lower (or upper, as defined by _UpLo) triangular part of A is considered. + * Therefore, the strict lower part does not have to store correct values. + * + * \sa MatrixBase::llt(), SelfAdjointView::llt(), class LDLT */ template class LLT + : public SolverBase > { public: typedef _MatrixType MatrixType; + typedef SolverBase Base; + friend class SolverBase; + + EIGEN_GENERIC_PUBLIC_INTERFACE(LLT) enum { - RowsAtCompileTime = MatrixType::RowsAtCompileTime, - ColsAtCompileTime = MatrixType::ColsAtCompileTime, - Options = MatrixType::Options, MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime }; - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef typename MatrixType::Index Index; enum { PacketSize = internal::packet_traits::size, @@ -83,14 +98,30 @@ template class LLT * according to the specified problem \a size. * \sa LLT() */ - LLT(Index size) : m_matrix(size, size), + explicit LLT(Index size) : m_matrix(size, size), m_isInitialized(false) {} - LLT(const MatrixType& matrix) + template + explicit LLT(const EigenBase& matrix) : m_matrix(matrix.rows(), matrix.cols()), m_isInitialized(false) { - compute(matrix); + compute(matrix.derived()); + } + + /** \brief Constructs a LLT factorization from a given matrix + * + * This overloaded constructor is provided for \link InplaceDecomposition inplace decomposition \endlink when + * \c MatrixType is a Eigen::Ref. + * + * \sa LLT(const EigenBase&) + */ + template + explicit LLT(EigenBase& matrix) + : m_matrix(matrix.derived()), + m_isInitialized(false) + { + compute(matrix.derived()); } /** \returns a view of the upper triangular matrix U */ @@ -107,6 +138,7 @@ template class LLT return Traits::getL(m_matrix); } + #ifdef EIGEN_PARSED_BY_DOXYGEN /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A. * * Since this LLT class assumes anyway that the matrix A is invertible, the solution @@ -115,33 +147,28 @@ template class LLT * Example: \include LLT_solve.cpp * Output: \verbinclude LLT_solve.out * - * \sa solveInPlace(), MatrixBase::llt() + * \sa solveInPlace(), MatrixBase::llt(), SelfAdjointView::llt() */ template - inline const internal::solve_retval - solve(const MatrixBase& b) const - { - eigen_assert(m_isInitialized && "LLT is not initialized."); - eigen_assert(m_matrix.rows()==b.rows() - && "LLT::solve(): invalid number of rows of the right hand side matrix b"); - return internal::solve_retval(*this, b.derived()); - } - - #ifdef EIGEN2_SUPPORT - template - bool solve(const MatrixBase& b, ResultType *result) const - { - *result = this->solve(b); - return true; - } - - bool isPositiveDefinite() const { return true; } + inline const Solve + solve(const MatrixBase& b) const; #endif template - void solveInPlace(MatrixBase &bAndX) const; + void solveInPlace(const MatrixBase &bAndX) const; - LLT& compute(const MatrixType& matrix); + template + LLT& compute(const EigenBase& matrix); + + /** \returns an estimate of the reciprocal condition number of the matrix of + * which \c *this is the Cholesky decomposition. + */ + RealScalar rcond() const + { + eigen_assert(m_isInitialized && "LLT is not initialized."); + eigen_assert(m_info == Success && "LLT failed because matrix appears to be negative"); + return internal::rcond_estimate_helper(m_l1_norm, *this); + } /** \returns the LLT decomposition matrix * @@ -158,8 +185,8 @@ template class LLT /** \brief Reports whether previous computation was successful. * - * \returns \c Success if computation was succesful, - * \c NumericalIssue if the matrix.appears to be negative. + * \returns \c Success if computation was successful, + * \c NumericalIssue if the matrix.appears not to be positive definite. */ ComputationInfo info() const { @@ -167,24 +194,40 @@ template class LLT return m_info; } - inline Index rows() const { return m_matrix.rows(); } - inline Index cols() const { return m_matrix.cols(); } + /** \returns the adjoint of \c *this, that is, a const reference to the decomposition itself as the underlying matrix is self-adjoint. + * + * This method is provided for compatibility with other matrix decompositions, thus enabling generic code such as: + * \code x = decomposition.adjoint().solve(b) \endcode + */ + const LLT& adjoint() const EIGEN_NOEXCEPT { return *this; }; + + inline EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); } + inline EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); } template - LLT rankUpdate(const VectorType& vec, const RealScalar& sigma = 1); + LLT & rankUpdate(const VectorType& vec, const RealScalar& sigma = 1); + + #ifndef EIGEN_PARSED_BY_DOXYGEN + template + void _solve_impl(const RhsType &rhs, DstType &dst) const; + + template + void _solve_impl_transposed(const RhsType &rhs, DstType &dst) const; + #endif protected: - + static void check_template_parameters() { EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar); } - + /** \internal * Used to compute and store L * The strict upper part is not used and even not initialized. */ MatrixType m_matrix; + RealScalar m_l1_norm; bool m_isInitialized; ComputationInfo m_info; }; @@ -194,12 +237,11 @@ namespace internal { template struct llt_inplace; template -static typename MatrixType::Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) +static Index llt_rank_update_lower(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) { using std::sqrt; typedef typename MatrixType::Scalar Scalar; typedef typename MatrixType::RealScalar RealScalar; - typedef typename MatrixType::Index Index; typedef typename MatrixType::ColXpr ColXpr; typedef typename internal::remove_all::type ColXprCleaned; typedef typename ColXprCleaned::SegmentReturnType ColXprSegment; @@ -268,11 +310,10 @@ template struct llt_inplace { typedef typename NumTraits::Real RealScalar; template - static typename MatrixType::Index unblocked(MatrixType& mat) + static Index unblocked(MatrixType& mat) { using std::sqrt; - typedef typename MatrixType::Index Index; - + eigen_assert(mat.rows()==mat.cols()); const Index size = mat.rows(); for(Index k = 0; k < size; ++k) @@ -295,9 +336,8 @@ template struct llt_inplace } template - static typename MatrixType::Index blocked(MatrixType& m) + static Index blocked(MatrixType& m) { - typedef typename MatrixType::Index Index; eigen_assert(m.rows()==m.cols()); Index size = m.rows(); if(size<32) @@ -322,36 +362,36 @@ template struct llt_inplace Index ret; if((ret=unblocked(A11))>=0) return k+ret; if(rs>0) A11.adjoint().template triangularView().template solveInPlace(A21); - if(rs>0) A22.template selfadjointView().rankUpdate(A21,-1); // bottleneck + if(rs>0) A22.template selfadjointView().rankUpdate(A21,typename NumTraits::Literal(-1)); // bottleneck } return -1; } template - static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma) + static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma) { return Eigen::internal::llt_rank_update_lower(mat, vec, sigma); } }; - + template struct llt_inplace { typedef typename NumTraits::Real RealScalar; template - static EIGEN_STRONG_INLINE typename MatrixType::Index unblocked(MatrixType& mat) + static EIGEN_STRONG_INLINE Index unblocked(MatrixType& mat) { Transpose matt(mat); return llt_inplace::unblocked(matt); } template - static EIGEN_STRONG_INLINE typename MatrixType::Index blocked(MatrixType& mat) + static EIGEN_STRONG_INLINE Index blocked(MatrixType& mat) { Transpose matt(mat); return llt_inplace::blocked(matt); } template - static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma) + static Index rankUpdate(MatrixType& mat, const VectorType& vec, const RealScalar& sigma) { Transpose matt(mat); return llt_inplace::rankUpdate(matt, vec.conjugate(), sigma); @@ -362,8 +402,8 @@ template struct LLT_Traits { typedef const TriangularView MatrixL; typedef const TriangularView MatrixU; - static inline MatrixL getL(const MatrixType& m) { return m; } - static inline MatrixU getU(const MatrixType& m) { return m.adjoint(); } + static inline MatrixL getL(const MatrixType& m) { return MatrixL(m); } + static inline MatrixU getU(const MatrixType& m) { return MatrixU(m.adjoint()); } static bool inplace_decomposition(MatrixType& m) { return llt_inplace::blocked(m)==-1; } }; @@ -372,8 +412,8 @@ template struct LLT_Traits { typedef const TriangularView MatrixL; typedef const TriangularView MatrixU; - static inline MatrixL getL(const MatrixType& m) { return m.adjoint(); } - static inline MatrixU getU(const MatrixType& m) { return m; } + static inline MatrixL getL(const MatrixType& m) { return MatrixL(m.adjoint()); } + static inline MatrixU getU(const MatrixType& m) { return MatrixU(m); } static bool inplace_decomposition(MatrixType& m) { return llt_inplace::blocked(m)==-1; } }; @@ -388,14 +428,29 @@ template struct LLT_Traits * Output: \verbinclude TutorialLinAlgComputeTwice.out */ template -LLT& LLT::compute(const MatrixType& a) +template +LLT& LLT::compute(const EigenBase& a) { check_template_parameters(); - + eigen_assert(a.rows()==a.cols()); const Index size = a.rows(); m_matrix.resize(size, size); - m_matrix = a; + if (!internal::is_same_dense(m_matrix, a.derived())) + m_matrix = a.derived(); + + // Compute matrix L1 norm = max abs column sum. + m_l1_norm = RealScalar(0); + // TODO move this code to SelfAdjointView + for (Index col = 0; col < size; ++col) { + RealScalar abs_col_sum; + if (_UpLo == Lower) + abs_col_sum = m_matrix.col(col).tail(size - col).template lpNorm<1>() + m_matrix.row(col).head(col).template lpNorm<1>(); + else + abs_col_sum = m_matrix.col(col).head(col).template lpNorm<1>() + m_matrix.row(col).tail(size - col).template lpNorm<1>(); + if (abs_col_sum > m_l1_norm) + m_l1_norm = abs_col_sum; + } m_isInitialized = true; bool ok = Traits::inplace_decomposition(m_matrix); @@ -411,7 +466,7 @@ LLT& LLT::compute(const MatrixType& a) */ template template -LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma) +LLT<_MatrixType,_UpLo> & LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, const RealScalar& sigma) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(VectorType); eigen_assert(v.size()==m_matrix.cols()); @@ -423,39 +478,42 @@ LLT<_MatrixType,_UpLo> LLT<_MatrixType,_UpLo>::rankUpdate(const VectorType& v, c return *this; } - -namespace internal { -template -struct solve_retval, Rhs> - : solve_retval_base, Rhs> + +#ifndef EIGEN_PARSED_BY_DOXYGEN +template +template +void LLT<_MatrixType,_UpLo>::_solve_impl(const RhsType &rhs, DstType &dst) const +{ + _solve_impl_transposed(rhs, dst); +} + +template +template +void LLT<_MatrixType,_UpLo>::_solve_impl_transposed(const RhsType &rhs, DstType &dst) const { - typedef LLT<_MatrixType,UpLo> LLTType; - EIGEN_MAKE_SOLVE_HELPERS(LLTType,Rhs) + dst = rhs; - template void evalTo(Dest& dst) const - { - dst = rhs(); - dec().solveInPlace(dst); - } -}; + matrixL().template conjugateIf().solveInPlace(dst); + matrixU().template conjugateIf().solveInPlace(dst); } +#endif /** \internal use x = llt_object.solve(x); - * + * * This is the \em in-place version of solve(). * * \param bAndX represents both the right-hand side matrix b and result x. * - * \returns true always! If you need to check for existence of solutions, use another decomposition like LU, QR, or SVD. + * This version avoids a copy when the right hand side matrix b is not needed anymore. * - * This version avoids a copy when the right hand side matrix b is not - * needed anymore. + * \warning The parameter is only marked 'const' to make the C++ compiler accept a temporary expression here. + * This function will const_cast it, so constness isn't honored here. * * \sa LLT::solve(), MatrixBase::llt() */ template template -void LLT::solveInPlace(MatrixBase &bAndX) const +void LLT::solveInPlace(const MatrixBase &bAndX) const { eigen_assert(m_isInitialized && "LLT is not initialized."); eigen_assert(m_matrix.rows()==bAndX.rows()); @@ -475,6 +533,7 @@ MatrixType LLT::reconstructedMatrix() const /** \cholesky_module * \returns the LLT decomposition of \c *this + * \sa SelfAdjointView::llt() */ template inline const LLT::PlainObject> @@ -485,6 +544,7 @@ MatrixBase::llt() const /** \cholesky_module * \returns the LLT decomposition of \c *this + * \sa SelfAdjointView::llt() */ template inline const LLT::PlainObject, UpLo> diff --git a/thirdparty/eigen/Eigen/src/Cholesky/LLT_MKL.h b/thirdparty/eigen/Eigen/src/Cholesky/LLT_LAPACKE.h similarity index 71% rename from thirdparty/eigen/Eigen/src/Cholesky/LLT_MKL.h rename to thirdparty/eigen/Eigen/src/Cholesky/LLT_LAPACKE.h index 66675d74..bc6489e6 100644 --- a/thirdparty/eigen/Eigen/src/Cholesky/LLT_MKL.h +++ b/thirdparty/eigen/Eigen/src/Cholesky/LLT_LAPACKE.h @@ -25,41 +25,38 @@ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. ******************************************************************************** - * Content : Eigen bindings to Intel(R) MKL + * Content : Eigen bindings to LAPACKe * LLt decomposition based on LAPACKE_?potrf function. ******************************************************************************** */ -#ifndef EIGEN_LLT_MKL_H -#define EIGEN_LLT_MKL_H - -#include "Eigen/src/Core/util/MKL_support.h" -#include +#ifndef EIGEN_LLT_LAPACKE_H +#define EIGEN_LLT_LAPACKE_H namespace Eigen { namespace internal { -template struct mkl_llt; +template struct lapacke_llt; -#define EIGEN_MKL_LLT(EIGTYPE, MKLTYPE, MKLPREFIX) \ -template<> struct mkl_llt \ +#define EIGEN_LAPACKE_LLT(EIGTYPE, BLASTYPE, LAPACKE_PREFIX) \ +template<> struct lapacke_llt \ { \ template \ - static inline typename MatrixType::Index potrf(MatrixType& m, char uplo) \ + static inline Index potrf(MatrixType& m, char uplo) \ { \ lapack_int matrix_order; \ lapack_int size, lda, info, StorageOrder; \ EIGTYPE* a; \ eigen_assert(m.rows()==m.cols()); \ /* Set up parameters for ?potrf */ \ - size = m.rows(); \ + size = convert_index(m.rows()); \ StorageOrder = MatrixType::Flags&RowMajorBit?RowMajor:ColMajor; \ matrix_order = StorageOrder==RowMajor ? LAPACK_ROW_MAJOR : LAPACK_COL_MAJOR; \ a = &(m.coeffRef(0,0)); \ - lda = m.outerStride(); \ + lda = convert_index(m.outerStride()); \ \ - info = LAPACKE_##MKLPREFIX##potrf( matrix_order, uplo, size, (MKLTYPE*)a, lda ); \ + info = LAPACKE_##LAPACKE_PREFIX##potrf( matrix_order, uplo, size, (BLASTYPE*)a, lda ); \ info = (info==0) ? -1 : info>0 ? info-1 : size; \ return info; \ } \ @@ -67,36 +64,36 @@ template<> struct mkl_llt \ template<> struct llt_inplace \ { \ template \ - static typename MatrixType::Index blocked(MatrixType& m) \ + static Index blocked(MatrixType& m) \ { \ - return mkl_llt::potrf(m, 'L'); \ + return lapacke_llt::potrf(m, 'L'); \ } \ template \ - static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \ + static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \ { return Eigen::internal::llt_rank_update_lower(mat, vec, sigma); } \ }; \ template<> struct llt_inplace \ { \ template \ - static typename MatrixType::Index blocked(MatrixType& m) \ + static Index blocked(MatrixType& m) \ { \ - return mkl_llt::potrf(m, 'U'); \ + return lapacke_llt::potrf(m, 'U'); \ } \ template \ - static typename MatrixType::Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \ + static Index rankUpdate(MatrixType& mat, const VectorType& vec, const typename MatrixType::RealScalar& sigma) \ { \ Transpose matt(mat); \ return llt_inplace::rankUpdate(matt, vec.conjugate(), sigma); \ } \ }; -EIGEN_MKL_LLT(double, double, d) -EIGEN_MKL_LLT(float, float, s) -EIGEN_MKL_LLT(dcomplex, MKL_Complex16, z) -EIGEN_MKL_LLT(scomplex, MKL_Complex8, c) +EIGEN_LAPACKE_LLT(double, double, d) +EIGEN_LAPACKE_LLT(float, float, s) +EIGEN_LAPACKE_LLT(dcomplex, lapack_complex_double, z) +EIGEN_LAPACKE_LLT(scomplex, lapack_complex_float, c) } // end namespace internal } // end namespace Eigen -#endif // EIGEN_LLT_MKL_H +#endif // EIGEN_LLT_LAPACKE_H diff --git a/thirdparty/eigen/Eigen/src/CholmodSupport/CMakeLists.txt b/thirdparty/eigen/Eigen/src/CholmodSupport/CMakeLists.txt deleted file mode 100644 index 814dfa61..00000000 --- a/thirdparty/eigen/Eigen/src/CholmodSupport/CMakeLists.txt +++ /dev/null @@ -1,6 +0,0 @@ -FILE(GLOB Eigen_CholmodSupport_SRCS "*.h") - -INSTALL(FILES - ${Eigen_CholmodSupport_SRCS} - DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/CholmodSupport COMPONENT Devel - ) diff --git a/thirdparty/eigen/Eigen/src/CholmodSupport/CholmodSupport.h b/thirdparty/eigen/Eigen/src/CholmodSupport/CholmodSupport.h index 99dbe171..adaf5285 100644 --- a/thirdparty/eigen/Eigen/src/CholmodSupport/CholmodSupport.h +++ b/thirdparty/eigen/Eigen/src/CholmodSupport/CholmodSupport.h @@ -10,50 +10,56 @@ #ifndef EIGEN_CHOLMODSUPPORT_H #define EIGEN_CHOLMODSUPPORT_H -namespace Eigen { +namespace Eigen { namespace internal { -template -void cholmod_configure_matrix(CholmodType& mat) -{ - if (internal::is_same::value) - { - mat.xtype = CHOLMOD_REAL; - mat.dtype = CHOLMOD_SINGLE; - } - else if (internal::is_same::value) - { +template struct cholmod_configure_matrix; + +template<> struct cholmod_configure_matrix { + template + static void run(CholmodType& mat) { mat.xtype = CHOLMOD_REAL; mat.dtype = CHOLMOD_DOUBLE; } - else if (internal::is_same >::value) - { - mat.xtype = CHOLMOD_COMPLEX; - mat.dtype = CHOLMOD_SINGLE; - } - else if (internal::is_same >::value) - { +}; + +template<> struct cholmod_configure_matrix > { + template + static void run(CholmodType& mat) { mat.xtype = CHOLMOD_COMPLEX; mat.dtype = CHOLMOD_DOUBLE; } - else - { - eigen_assert(false && "Scalar type not supported by CHOLMOD"); - } -} +}; + +// Other scalar types are not yet supported by Cholmod +// template<> struct cholmod_configure_matrix { +// template +// static void run(CholmodType& mat) { +// mat.xtype = CHOLMOD_REAL; +// mat.dtype = CHOLMOD_SINGLE; +// } +// }; +// +// template<> struct cholmod_configure_matrix > { +// template +// static void run(CholmodType& mat) { +// mat.xtype = CHOLMOD_COMPLEX; +// mat.dtype = CHOLMOD_SINGLE; +// } +// }; } // namespace internal /** Wraps the Eigen sparse matrix \a mat into a Cholmod sparse matrix object. * Note that the data are shared. */ -template -cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat) +template +cholmod_sparse viewAsCholmod(Ref > mat) { cholmod_sparse res; res.nzmax = mat.nonZeros(); - res.nrow = mat.rows();; + res.nrow = mat.rows(); res.ncol = mat.cols(); res.p = mat.outerIndexPtr(); res.i = mat.innerIndexPtr(); @@ -73,12 +79,12 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat) res.dtype = 0; res.stype = -1; - - if (internal::is_same<_Index,int>::value) + + if (internal::is_same<_StorageIndex,int>::value) { res.itype = CHOLMOD_INT; } - else if (internal::is_same<_Index,SuiteSparse_long>::value) + else if (internal::is_same<_StorageIndex,SuiteSparse_long>::value) { res.itype = CHOLMOD_LONG; } @@ -88,29 +94,39 @@ cholmod_sparse viewAsCholmod(SparseMatrix<_Scalar,_Options,_Index>& mat) } // setup res.xtype - internal::cholmod_configure_matrix<_Scalar>(res); - + internal::cholmod_configure_matrix<_Scalar>::run(res); + res.stype = 0; - + return res; } template const cholmod_sparse viewAsCholmod(const SparseMatrix<_Scalar,_Options,_Index>& mat) { - cholmod_sparse res = viewAsCholmod(mat.const_cast_derived()); + cholmod_sparse res = viewAsCholmod(Ref >(mat.const_cast_derived())); + return res; +} + +template +const cholmod_sparse viewAsCholmod(const SparseVector<_Scalar,_Options,_Index>& mat) +{ + cholmod_sparse res = viewAsCholmod(Ref >(mat.const_cast_derived())); return res; } /** Returns a view of the Eigen sparse matrix \a mat as Cholmod sparse matrix. * The data are not copied but shared. */ template -cholmod_sparse viewAsCholmod(const SparseSelfAdjointView, UpLo>& mat) +cholmod_sparse viewAsCholmod(const SparseSelfAdjointView, UpLo>& mat) { - cholmod_sparse res = viewAsCholmod(mat.matrix().const_cast_derived()); - + cholmod_sparse res = viewAsCholmod(Ref >(mat.matrix().const_cast_derived())); + if(UpLo==Upper) res.stype = 1; if(UpLo==Lower) res.stype = -1; + // swap stype for rowmajor matrices (only works for real matrices) + EIGEN_STATIC_ASSERT((_Options & RowMajorBit) == 0 || NumTraits<_Scalar>::IsComplex == 0, THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES); + if(_Options & RowMajorBit) res.stype *=-1; return res; } @@ -131,21 +147,59 @@ cholmod_dense viewAsCholmod(MatrixBase& mat) res.x = (void*)(mat.derived().data()); res.z = 0; - internal::cholmod_configure_matrix(res); + internal::cholmod_configure_matrix::run(res); return res; } /** Returns a view of the Cholmod sparse matrix \a cm as an Eigen sparse matrix. * The data are not copied but shared. */ -template -MappedSparseMatrix viewAsEigen(cholmod_sparse& cm) +template +MappedSparseMatrix viewAsEigen(cholmod_sparse& cm) { - return MappedSparseMatrix - (cm.nrow, cm.ncol, static_cast(cm.p)[cm.ncol], - static_cast(cm.p), static_cast(cm.i),static_cast(cm.x) ); + return MappedSparseMatrix + (cm.nrow, cm.ncol, static_cast(cm.p)[cm.ncol], + static_cast(cm.p), static_cast(cm.i),static_cast(cm.x) ); } +namespace internal { + +// template specializations for int and long that call the correct cholmod method + +#define EIGEN_CHOLMOD_SPECIALIZE0(ret, name) \ + template inline ret cm_ ## name (cholmod_common &Common) { return cholmod_ ## name (&Common); } \ + template<> inline ret cm_ ## name (cholmod_common &Common) { return cholmod_l_ ## name (&Common); } + +#define EIGEN_CHOLMOD_SPECIALIZE1(ret, name, t1, a1) \ + template inline ret cm_ ## name (t1& a1, cholmod_common &Common) { return cholmod_ ## name (&a1, &Common); } \ + template<> inline ret cm_ ## name (t1& a1, cholmod_common &Common) { return cholmod_l_ ## name (&a1, &Common); } + +EIGEN_CHOLMOD_SPECIALIZE0(int, start) +EIGEN_CHOLMOD_SPECIALIZE0(int, finish) + +EIGEN_CHOLMOD_SPECIALIZE1(int, free_factor, cholmod_factor*, L) +EIGEN_CHOLMOD_SPECIALIZE1(int, free_dense, cholmod_dense*, X) +EIGEN_CHOLMOD_SPECIALIZE1(int, free_sparse, cholmod_sparse*, A) + +EIGEN_CHOLMOD_SPECIALIZE1(cholmod_factor*, analyze, cholmod_sparse, A) + +template inline cholmod_dense* cm_solve (int sys, cholmod_factor& L, cholmod_dense& B, cholmod_common &Common) { return cholmod_solve (sys, &L, &B, &Common); } +template<> inline cholmod_dense* cm_solve (int sys, cholmod_factor& L, cholmod_dense& B, cholmod_common &Common) { return cholmod_l_solve (sys, &L, &B, &Common); } + +template inline cholmod_sparse* cm_spsolve (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_spsolve (sys, &L, &B, &Common); } +template<> inline cholmod_sparse* cm_spsolve (int sys, cholmod_factor& L, cholmod_sparse& B, cholmod_common &Common) { return cholmod_l_spsolve (sys, &L, &B, &Common); } + +template +inline int cm_factorize_p (cholmod_sparse* A, double beta[2], _StorageIndex* fset, std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_factorize_p (A, beta, fset, fsize, L, &Common); } +template<> +inline int cm_factorize_p (cholmod_sparse* A, double beta[2], SuiteSparse_long* fset, std::size_t fsize, cholmod_factor* L, cholmod_common &Common) { return cholmod_l_factorize_p (A, beta, fset, fsize, L, &Common); } + +#undef EIGEN_CHOLMOD_SPECIALIZE0 +#undef EIGEN_CHOLMOD_SPECIALIZE1 + +} // namespace internal + + enum CholmodMode { CholmodAuto, CholmodSimplicialLLt, CholmodSupernodalLLt, CholmodLDLt }; @@ -157,49 +211,56 @@ enum CholmodMode { * \sa class CholmodSupernodalLLT, class CholmodSimplicialLDLT, class CholmodSimplicialLLT */ template -class CholmodBase : internal::noncopyable +class CholmodBase : public SparseSolverBase { + protected: + typedef SparseSolverBase Base; + using Base::derived; + using Base::m_isInitialized; public: typedef _MatrixType MatrixType; enum { UpLo = _UpLo }; typedef typename MatrixType::Scalar Scalar; typedef typename MatrixType::RealScalar RealScalar; typedef MatrixType CholMatrixType; - typedef typename MatrixType::Index Index; + typedef typename MatrixType::StorageIndex StorageIndex; + enum { + ColsAtCompileTime = MatrixType::ColsAtCompileTime, + MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime + }; public: CholmodBase() - : m_cholmodFactor(0), m_info(Success), m_isInitialized(false) + : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false) { - m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0); - cholmod_start(&m_cholmod); + EIGEN_STATIC_ASSERT((internal::is_same::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY); + m_shiftOffset[0] = m_shiftOffset[1] = 0.0; + internal::cm_start(m_cholmod); } - CholmodBase(const MatrixType& matrix) - : m_cholmodFactor(0), m_info(Success), m_isInitialized(false) + explicit CholmodBase(const MatrixType& matrix) + : m_cholmodFactor(0), m_info(Success), m_factorizationIsOk(false), m_analysisIsOk(false) { - m_shiftOffset[0] = m_shiftOffset[1] = RealScalar(0.0); - cholmod_start(&m_cholmod); + EIGEN_STATIC_ASSERT((internal::is_same::value), CHOLMOD_SUPPORTS_DOUBLE_PRECISION_ONLY); + m_shiftOffset[0] = m_shiftOffset[1] = 0.0; + internal::cm_start(m_cholmod); compute(matrix); } ~CholmodBase() { if(m_cholmodFactor) - cholmod_free_factor(&m_cholmodFactor, &m_cholmod); - cholmod_finish(&m_cholmod); + internal::cm_free_factor(m_cholmodFactor, m_cholmod); + internal::cm_finish(m_cholmod); } - - inline Index cols() const { return m_cholmodFactor->n; } - inline Index rows() const { return m_cholmodFactor->n; } - - Derived& derived() { return *static_cast(this); } - const Derived& derived() const { return *static_cast(this); } - + + inline StorageIndex cols() const { return internal::convert_index(m_cholmodFactor->n); } + inline StorageIndex rows() const { return internal::convert_index(m_cholmodFactor->n); } + /** \brief Reports whether previous computation was successful. * - * \returns \c Success if computation was succesful, + * \returns \c Success if computation was successful, * \c NumericalIssue if the matrix.appears to be negative. */ ComputationInfo info() const @@ -215,57 +276,29 @@ class CholmodBase : internal::noncopyable factorize(matrix); return derived(); } - - /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A. - * - * \sa compute() - */ - template - inline const internal::solve_retval - solve(const MatrixBase& b) const - { - eigen_assert(m_isInitialized && "LLT is not initialized."); - eigen_assert(rows()==b.rows() - && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b"); - return internal::solve_retval(*this, b.derived()); - } - - /** \returns the solution x of \f$ A x = b \f$ using the current decomposition of A. - * - * \sa compute() - */ - template - inline const internal::sparse_solve_retval - solve(const SparseMatrixBase& b) const - { - eigen_assert(m_isInitialized && "LLT is not initialized."); - eigen_assert(rows()==b.rows() - && "CholmodDecomposition::solve(): invalid number of rows of the right hand side matrix b"); - return internal::sparse_solve_retval(*this, b.derived()); - } - + /** Performs a symbolic decomposition on the sparsity pattern of \a matrix. * * This function is particularly useful when solving for several problems having the same structure. - * + * * \sa factorize() */ void analyzePattern(const MatrixType& matrix) { if(m_cholmodFactor) { - cholmod_free_factor(&m_cholmodFactor, &m_cholmod); + internal::cm_free_factor(m_cholmodFactor, m_cholmod); m_cholmodFactor = 0; } cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView()); - m_cholmodFactor = cholmod_analyze(&A, &m_cholmod); - + m_cholmodFactor = internal::cm_analyze(A, m_cholmod); + this->m_isInitialized = true; this->m_info = Success; m_analysisIsOk = true; m_factorizationIsOk = false; } - + /** Performs a numeric decomposition of \a matrix * * The given matrix must have the same sparsity pattern as the matrix on which the symbolic decomposition has been performed. @@ -276,43 +309,46 @@ class CholmodBase : internal::noncopyable { eigen_assert(m_analysisIsOk && "You must first call analyzePattern()"); cholmod_sparse A = viewAsCholmod(matrix.template selfadjointView()); - cholmod_factorize_p(&A, m_shiftOffset, 0, 0, m_cholmodFactor, &m_cholmod); - + internal::cm_factorize_p(&A, m_shiftOffset, 0, 0, m_cholmodFactor, m_cholmod); + // If the factorization failed, minor is the column at which it did. On success minor == n. this->m_info = (m_cholmodFactor->minor == m_cholmodFactor->n ? Success : NumericalIssue); m_factorizationIsOk = true; } - + /** Returns a reference to the Cholmod's configuration structure to get a full control over the performed operations. * See the Cholmod user guide for details. */ cholmod_common& cholmod() { return m_cholmod; } - + #ifndef EIGEN_PARSED_BY_DOXYGEN /** \internal */ template - void _solve(const MatrixBase &b, MatrixBase &dest) const + void _solve_impl(const MatrixBase &b, MatrixBase &dest) const { eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()"); const Index size = m_cholmodFactor->n; EIGEN_UNUSED_VARIABLE(size); eigen_assert(size==b.rows()); - // note: cd stands for Cholmod Dense - Rhs& b_ref(b.const_cast_derived()); + // Cholmod needs column-major storage without inner-stride, which corresponds to the default behavior of Ref. + Ref > b_ref(b.derived()); + cholmod_dense b_cd = viewAsCholmod(b_ref); - cholmod_dense* x_cd = cholmod_solve(CHOLMOD_A, m_cholmodFactor, &b_cd, &m_cholmod); + cholmod_dense* x_cd = internal::cm_solve(CHOLMOD_A, *m_cholmodFactor, b_cd, m_cholmod); if(!x_cd) { this->m_info = NumericalIssue; + return; } // TODO optimize this copy by swapping when possible (be careful with alignment, etc.) + // NOTE Actually, the copy can be avoided by calling cholmod_solve2 instead of cholmod_solve dest = Matrix::Map(reinterpret_cast(x_cd->x),b.rows(),b.cols()); - cholmod_free_dense(&x_cd, &m_cholmod); + internal::cm_free_dense(x_cd, m_cholmod); } - + /** \internal */ - template - void _solve(const SparseMatrix &b, SparseMatrix &dest) const + template + void _solve_impl(const SparseMatrixBase &b, SparseMatrixBase &dest) const { eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()"); const Index size = m_cholmodFactor->n; @@ -320,19 +356,22 @@ class CholmodBase : internal::noncopyable eigen_assert(size==b.rows()); // note: cs stands for Cholmod Sparse - cholmod_sparse b_cs = viewAsCholmod(b); - cholmod_sparse* x_cs = cholmod_spsolve(CHOLMOD_A, m_cholmodFactor, &b_cs, &m_cholmod); + Ref > b_ref(b.const_cast_derived()); + cholmod_sparse b_cs = viewAsCholmod(b_ref); + cholmod_sparse* x_cs = internal::cm_spsolve(CHOLMOD_A, *m_cholmodFactor, b_cs, m_cholmod); if(!x_cs) { this->m_info = NumericalIssue; + return; } // TODO optimize this copy by swapping when possible (be careful with alignment, etc.) - dest = viewAsEigen(*x_cs); - cholmod_free_sparse(&x_cs, &m_cholmod); + // NOTE cholmod_spsolve in fact just calls the dense solver for blocks of 4 columns at a time (similar to Eigen's sparse solver) + dest.derived() = viewAsEigen(*x_cs); + internal::cm_free_sparse(x_cs, m_cholmod); } #endif // EIGEN_PARSED_BY_DOXYGEN - - + + /** Sets the shift parameter that will be used to adjust the diagonal coefficients during the numerical factorization. * * During the numerical factorization, an offset term is added to the diagonal coefficients:\n @@ -344,20 +383,70 @@ class CholmodBase : internal::noncopyable */ Derived& setShift(const RealScalar& offset) { - m_shiftOffset[0] = offset; + m_shiftOffset[0] = double(offset); return derived(); } - + + /** \returns the determinant of the underlying matrix from the current factorization */ + Scalar determinant() const + { + using std::exp; + return exp(logDeterminant()); + } + + /** \returns the log determinant of the underlying matrix from the current factorization */ + Scalar logDeterminant() const + { + using std::log; + using numext::real; + eigen_assert(m_factorizationIsOk && "The decomposition is not in a valid state for solving, you must first call either compute() or symbolic()/numeric()"); + + RealScalar logDet = 0; + Scalar *x = static_cast(m_cholmodFactor->x); + if (m_cholmodFactor->is_super) + { + // Supernodal factorization stored as a packed list of dense column-major blocs, + // as described by the following structure: + + // super[k] == index of the first column of the j-th super node + StorageIndex *super = static_cast(m_cholmodFactor->super); + // pi[k] == offset to the description of row indices + StorageIndex *pi = static_cast(m_cholmodFactor->pi); + // px[k] == offset to the respective dense block + StorageIndex *px = static_cast(m_cholmodFactor->px); + + Index nb_super_nodes = m_cholmodFactor->nsuper; + for (Index k=0; k < nb_super_nodes; ++k) + { + StorageIndex ncols = super[k + 1] - super[k]; + StorageIndex nrows = pi[k + 1] - pi[k]; + + Map, 0, InnerStride<> > sk(x + px[k], ncols, InnerStride<>(nrows+1)); + logDet += sk.real().log().sum(); + } + } + else + { + // Simplicial factorization stored as standard CSC matrix. + StorageIndex *p = static_cast(m_cholmodFactor->p); + Index size = m_cholmodFactor->n; + for (Index k=0; kis_ll) + logDet *= 2.0; + return logDet; + }; + template void dumpMemory(Stream& /*s*/) {} - + protected: mutable cholmod_common m_cholmod; cholmod_factor* m_cholmodFactor; - RealScalar m_shiftOffset[2]; + double m_shiftOffset[2]; mutable ComputationInfo m_info; - bool m_isInitialized; int m_factorizationIsOk; int m_analysisIsOk; }; @@ -376,26 +465,30 @@ class CholmodBase : internal::noncopyable * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower * or Upper. Default is Lower. * + * \implsparsesolverconcept + * * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed. * - * \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLLT + * \warning Only double precision real and complex scalar types are supported by Cholmod. + * + * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLLT */ template class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT<_MatrixType, _UpLo> > { typedef CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLLT> Base; using Base::m_cholmod; - + public: - + typedef _MatrixType MatrixType; - + CholmodSimplicialLLT() : Base() { init(); } CholmodSimplicialLLT(const MatrixType& matrix) : Base() { init(); - Base::compute(matrix); + this->compute(matrix); } ~CholmodSimplicialLLT() {} @@ -423,26 +516,30 @@ class CholmodSimplicialLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimpl * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower * or Upper. Default is Lower. * + * \implsparsesolverconcept + * * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed. * - * \sa \ref TutorialSparseDirectSolvers, class CholmodSupernodalLLT, class SimplicialLDLT + * \warning Only double precision real and complex scalar types are supported by Cholmod. + * + * \sa \ref TutorialSparseSolverConcept, class CholmodSupernodalLLT, class SimplicialLDLT */ template class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT<_MatrixType, _UpLo> > { typedef CholmodBase<_MatrixType, _UpLo, CholmodSimplicialLDLT> Base; using Base::m_cholmod; - + public: - + typedef _MatrixType MatrixType; - + CholmodSimplicialLDLT() : Base() { init(); } CholmodSimplicialLDLT(const MatrixType& matrix) : Base() { init(); - Base::compute(matrix); + this->compute(matrix); } ~CholmodSimplicialLDLT() {} @@ -468,26 +565,30 @@ class CholmodSimplicialLDLT : public CholmodBase<_MatrixType, _UpLo, CholmodSimp * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower * or Upper. Default is Lower. * + * \implsparsesolverconcept + * * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed. * - * \sa \ref TutorialSparseDirectSolvers + * \warning Only double precision real and complex scalar types are supported by Cholmod. + * + * \sa \ref TutorialSparseSolverConcept */ template class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT<_MatrixType, _UpLo> > { typedef CholmodBase<_MatrixType, _UpLo, CholmodSupernodalLLT> Base; using Base::m_cholmod; - + public: - + typedef _MatrixType MatrixType; - + CholmodSupernodalLLT() : Base() { init(); } CholmodSupernodalLLT(const MatrixType& matrix) : Base() { init(); - Base::compute(matrix); + this->compute(matrix); } ~CholmodSupernodalLLT() {} @@ -515,30 +616,34 @@ class CholmodSupernodalLLT : public CholmodBase<_MatrixType, _UpLo, CholmodSuper * \tparam _UpLo the triangular part that will be used for the computations. It can be Lower * or Upper. Default is Lower. * + * \implsparsesolverconcept + * * This class supports all kind of SparseMatrix<>: row or column major; upper, lower, or both; compressed or non compressed. * - * \sa \ref TutorialSparseDirectSolvers + * \warning Only double precision real and complex scalar types are supported by Cholmod. + * + * \sa \ref TutorialSparseSolverConcept */ template class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecomposition<_MatrixType, _UpLo> > { typedef CholmodBase<_MatrixType, _UpLo, CholmodDecomposition> Base; using Base::m_cholmod; - + public: - + typedef _MatrixType MatrixType; - + CholmodDecomposition() : Base() { init(); } CholmodDecomposition(const MatrixType& matrix) : Base() { init(); - Base::compute(matrix); + this->compute(matrix); } ~CholmodDecomposition() {} - + void setMode(CholmodMode mode) { switch(mode) @@ -572,36 +677,6 @@ class CholmodDecomposition : public CholmodBase<_MatrixType, _UpLo, CholmodDecom } }; -namespace internal { - -template -struct solve_retval, Rhs> - : solve_retval_base, Rhs> -{ - typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec; - EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs) - - template void evalTo(Dest& dst) const - { - dec()._solve(rhs(),dst); - } -}; - -template -struct sparse_solve_retval, Rhs> - : sparse_solve_retval_base, Rhs> -{ - typedef CholmodBase<_MatrixType,_UpLo,Derived> Dec; - EIGEN_MAKE_SPARSE_SOLVE_HELPERS(Dec,Rhs) - - template void evalTo(Dest& dst) const - { - dec()._solve(rhs(),dst); - } -}; - -} // end namespace internal - } // end namespace Eigen #endif // EIGEN_CHOLMODSUPPORT_H diff --git a/thirdparty/eigen/Eigen/src/Core/ArithmeticSequence.h b/thirdparty/eigen/Eigen/src/Core/ArithmeticSequence.h new file mode 100644 index 00000000..d04f726d --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/ArithmeticSequence.h @@ -0,0 +1,406 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2017 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_ARITHMETIC_SEQUENCE_H +#define EIGEN_ARITHMETIC_SEQUENCE_H + +namespace Eigen { + +namespace internal { + +#if (!EIGEN_HAS_CXX11) || !((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48) +template struct aseq_negate {}; + +template<> struct aseq_negate { + typedef Index type; +}; + +template struct aseq_negate > { + typedef FixedInt<-N> type; +}; + +// Compilation error in the following case: +template<> struct aseq_negate > {}; + +template::value, + bool SizeIsSymbolic =symbolic::is_symbolic::value> +struct aseq_reverse_first_type { + typedef Index type; +}; + +template +struct aseq_reverse_first_type { + typedef symbolic::AddExpr > >, + symbolic::ValueExpr > + > type; +}; + +template +struct aseq_reverse_first_type_aux { + typedef Index type; +}; + +template +struct aseq_reverse_first_type_aux::type> { + typedef FixedInt<(SizeType::value-1)*IncrType::value> type; +}; + +template +struct aseq_reverse_first_type { + typedef typename aseq_reverse_first_type_aux::type Aux; + typedef symbolic::AddExpr > type; +}; + +template +struct aseq_reverse_first_type { + typedef symbolic::AddExpr > >, + symbolic::ValueExpr >, + symbolic::ValueExpr<> > type; +}; +#endif + +// Helper to cleanup the type of the increment: +template struct cleanup_seq_incr { + typedef typename cleanup_index_type::type type; +}; + +} + +//-------------------------------------------------------------------------------- +// seq(first,last,incr) and seqN(first,size,incr) +//-------------------------------------------------------------------------------- + +template > +class ArithmeticSequence; + +template +ArithmeticSequence::type, + typename internal::cleanup_index_type::type, + typename internal::cleanup_seq_incr::type > +seqN(FirstType first, SizeType size, IncrType incr); + +/** \class ArithmeticSequence + * \ingroup Core_Module + * + * This class represents an arithmetic progression \f$ a_0, a_1, a_2, ..., a_{n-1}\f$ defined by + * its \em first value \f$ a_0 \f$, its \em size (aka length) \em n, and the \em increment (aka stride) + * that is equal to \f$ a_{i+1}-a_{i}\f$ for any \em i. + * + * It is internally used as the return type of the Eigen::seq and Eigen::seqN functions, and as the input arguments + * of DenseBase::operator()(const RowIndices&, const ColIndices&), and most of the time this is the + * only way it is used. + * + * \tparam FirstType type of the first element, usually an Index, + * but internally it can be a symbolic expression + * \tparam SizeType type representing the size of the sequence, usually an Index + * or a compile time integral constant. Internally, it can also be a symbolic expression + * \tparam IncrType type of the increment, can be a runtime Index, or a compile time integral constant (default is compile-time 1) + * + * \sa Eigen::seq, Eigen::seqN, DenseBase::operator()(const RowIndices&, const ColIndices&), class IndexedView + */ +template +class ArithmeticSequence +{ +public: + ArithmeticSequence(FirstType first, SizeType size) : m_first(first), m_size(size) {} + ArithmeticSequence(FirstType first, SizeType size, IncrType incr) : m_first(first), m_size(size), m_incr(incr) {} + + enum { + SizeAtCompileTime = internal::get_fixed_value::value, + IncrAtCompileTime = internal::get_fixed_value::value + }; + + /** \returns the size, i.e., number of elements, of the sequence */ + Index size() const { return m_size; } + + /** \returns the first element \f$ a_0 \f$ in the sequence */ + Index first() const { return m_first; } + + /** \returns the value \f$ a_i \f$ at index \a i in the sequence. */ + Index operator[](Index i) const { return m_first + i * m_incr; } + + const FirstType& firstObject() const { return m_first; } + const SizeType& sizeObject() const { return m_size; } + const IncrType& incrObject() const { return m_incr; } + +protected: + FirstType m_first; + SizeType m_size; + IncrType m_incr; + +public: + +#if EIGEN_HAS_CXX11 && ((!EIGEN_COMP_GNUC) || EIGEN_COMP_GNUC>=48) + auto reverse() const -> decltype(Eigen::seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr)) { + return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr); + } +#else +protected: + typedef typename internal::aseq_negate::type ReverseIncrType; + typedef typename internal::aseq_reverse_first_type::type ReverseFirstType; +public: + ArithmeticSequence + reverse() const { + return seqN(m_first+(m_size+fix<-1>())*m_incr,m_size,-m_incr); + } +#endif +}; + +/** \returns an ArithmeticSequence starting at \a first, of length \a size, and increment \a incr + * + * \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */ +template +ArithmeticSequence::type,typename internal::cleanup_index_type::type,typename internal::cleanup_seq_incr::type > +seqN(FirstType first, SizeType size, IncrType incr) { + return ArithmeticSequence::type,typename internal::cleanup_index_type::type,typename internal::cleanup_seq_incr::type>(first,size,incr); +} + +/** \returns an ArithmeticSequence starting at \a first, of length \a size, and unit increment + * + * \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType) */ +template +ArithmeticSequence::type,typename internal::cleanup_index_type::type > +seqN(FirstType first, SizeType size) { + return ArithmeticSequence::type,typename internal::cleanup_index_type::type>(first,size); +} + + +#if EIGEN_HAS_CXX11 + +/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and with positive (or negative) increment \a incr + * + * It is essentially an alias to: + * \code + * seqN(f, (l-f+incr)/incr, incr); + * \endcode + * + * \sa seqN(FirstType,SizeType,IncrType), seq(FirstType,LastType) + */ +template +auto seq(FirstType f, LastType l) -> decltype(seqN(typename internal::cleanup_index_type::type(f), + ( typename internal::cleanup_index_type::type(l) + - typename internal::cleanup_index_type::type(f)+fix<1>()))) +{ + return seqN(typename internal::cleanup_index_type::type(f), + (typename internal::cleanup_index_type::type(l) + -typename internal::cleanup_index_type::type(f)+fix<1>())); +} + +/** \returns an ArithmeticSequence starting at \a f, up (or down) to \a l, and unit increment + * + * It is essentially an alias to: + * \code + * seqN(f,l-f+1); + * \endcode + * + * \sa seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) + */ +template +auto seq(FirstType f, LastType l, IncrType incr) + -> decltype(seqN(typename internal::cleanup_index_type::type(f), + ( typename internal::cleanup_index_type::type(l) + - typename internal::cleanup_index_type::type(f)+typename internal::cleanup_seq_incr::type(incr) + ) / typename internal::cleanup_seq_incr::type(incr), + typename internal::cleanup_seq_incr::type(incr))) +{ + typedef typename internal::cleanup_seq_incr::type CleanedIncrType; + return seqN(typename internal::cleanup_index_type::type(f), + ( typename internal::cleanup_index_type::type(l) + -typename internal::cleanup_index_type::type(f)+CleanedIncrType(incr)) / CleanedIncrType(incr), + CleanedIncrType(incr)); +} + +#else // EIGEN_HAS_CXX11 + +template +typename internal::enable_if::value || symbolic::is_symbolic::value), + ArithmeticSequence::type,Index> >::type +seq(FirstType f, LastType l) +{ + return seqN(typename internal::cleanup_index_type::type(f), + Index((typename internal::cleanup_index_type::type(l)-typename internal::cleanup_index_type::type(f)+fix<1>()))); +} + +template +typename internal::enable_if::value, + ArithmeticSequence,symbolic::ValueExpr<> >, + symbolic::ValueExpr > > > >::type +seq(const symbolic::BaseExpr &f, LastType l) +{ + return seqN(f.derived(),(typename internal::cleanup_index_type::type(l)-f.derived()+fix<1>())); +} + +template +typename internal::enable_if::value, + ArithmeticSequence::type, + symbolic::AddExpr >, + symbolic::ValueExpr > > > >::type +seq(FirstType f, const symbolic::BaseExpr &l) +{ + return seqN(typename internal::cleanup_index_type::type(f),(l.derived()-typename internal::cleanup_index_type::type(f)+fix<1>())); +} + +template +ArithmeticSequence >,symbolic::ValueExpr > > > +seq(const symbolic::BaseExpr &f, const symbolic::BaseExpr &l) +{ + return seqN(f.derived(),(l.derived()-f.derived()+fix<1>())); +} + + +template +typename internal::enable_if::value || symbolic::is_symbolic::value), + ArithmeticSequence::type,Index,typename internal::cleanup_seq_incr::type> >::type +seq(FirstType f, LastType l, IncrType incr) +{ + typedef typename internal::cleanup_seq_incr::type CleanedIncrType; + return seqN(typename internal::cleanup_index_type::type(f), + Index((typename internal::cleanup_index_type::type(l)-typename internal::cleanup_index_type::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr)), incr); +} + +template +typename internal::enable_if::value, + ArithmeticSequence, + symbolic::ValueExpr<> >, + symbolic::ValueExpr::type> >, + symbolic::ValueExpr::type> >, + typename internal::cleanup_seq_incr::type> >::type +seq(const symbolic::BaseExpr &f, LastType l, IncrType incr) +{ + typedef typename internal::cleanup_seq_incr::type CleanedIncrType; + return seqN(f.derived(),(typename internal::cleanup_index_type::type(l)-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr); +} + +template +typename internal::enable_if::value, + ArithmeticSequence::type, + symbolic::QuotientExpr >, + symbolic::ValueExpr::type> >, + symbolic::ValueExpr::type> >, + typename internal::cleanup_seq_incr::type> >::type +seq(FirstType f, const symbolic::BaseExpr &l, IncrType incr) +{ + typedef typename internal::cleanup_seq_incr::type CleanedIncrType; + return seqN(typename internal::cleanup_index_type::type(f), + (l.derived()-typename internal::cleanup_index_type::type(f)+CleanedIncrType(incr))/CleanedIncrType(incr), incr); +} + +template +ArithmeticSequence >, + symbolic::ValueExpr::type> >, + symbolic::ValueExpr::type> >, + typename internal::cleanup_seq_incr::type> +seq(const symbolic::BaseExpr &f, const symbolic::BaseExpr &l, IncrType incr) +{ + typedef typename internal::cleanup_seq_incr::type CleanedIncrType; + return seqN(f.derived(),(l.derived()-f.derived()+CleanedIncrType(incr))/CleanedIncrType(incr), incr); +} +#endif // EIGEN_HAS_CXX11 + +#if EIGEN_HAS_CXX11 +/** \cpp11 + * \returns a symbolic ArithmeticSequence representing the last \a size elements with a unit increment. + * + * \anchor indexing_lastN + * + * It is a shortcut for: \code seq(last+fix<1>-size, last) \endcode + * + * \sa lastN(SizeType,IncrType, seqN(FirstType,SizeType), seq(FirstType,LastType) */ +template +auto lastN(SizeType size) +-> decltype(seqN(Eigen::last+fix<1>()-size, size)) +{ + return seqN(Eigen::last+fix<1>()-size, size); +} + +/** \cpp11 + * \returns a symbolic ArithmeticSequence representing the last \a size elements with increment \a incr. + * + * \anchor indexing_lastN_with_incr + * + * It is a shortcut for: \code seqN(last-(size-fix<1>)*incr, size, incr) \endcode + * + * \sa lastN(SizeType), seqN(FirstType,SizeType), seq(FirstType,LastType,IncrType) */ +template +auto lastN(SizeType size, IncrType incr) +-> decltype(seqN(Eigen::last-(size-fix<1>())*incr, size, incr)) +{ + return seqN(Eigen::last-(size-fix<1>())*incr, size, incr); +} +#endif + +namespace internal { + +// Convert a symbolic span into a usable one (i.e., remove last/end "keywords") +template +struct make_size_type { + typedef typename internal::conditional::value, Index, T>::type type; +}; + +template +struct IndexedViewCompatibleType, XprSize> { + typedef ArithmeticSequence::type,IncrType> type; +}; + +template +ArithmeticSequence::type,IncrType> +makeIndexedViewCompatible(const ArithmeticSequence& ids, Index size,SpecializedType) { + return ArithmeticSequence::type,IncrType>( + eval_expr_given_size(ids.firstObject(),size),eval_expr_given_size(ids.sizeObject(),size),ids.incrObject()); +} + +template +struct get_compile_time_incr > { + enum { value = get_fixed_value::value }; +}; + +} // end namespace internal + +/** \namespace Eigen::indexing + * \ingroup Core_Module + * + * The sole purpose of this namespace is to be able to import all functions + * and symbols that are expected to be used within operator() for indexing + * and slicing. If you already imported the whole Eigen namespace: + * \code using namespace Eigen; \endcode + * then you are already all set. Otherwise, if you don't want/cannot import + * the whole Eigen namespace, the following line: + * \code using namespace Eigen::indexing; \endcode + * is equivalent to: + * \code + using Eigen::all; + using Eigen::seq; + using Eigen::seqN; + using Eigen::lastN; // c++11 only + using Eigen::last; + using Eigen::lastp1; + using Eigen::fix; + \endcode + */ +namespace indexing { + using Eigen::all; + using Eigen::seq; + using Eigen::seqN; + #if EIGEN_HAS_CXX11 + using Eigen::lastN; + #endif + using Eigen::last; + using Eigen::lastp1; + using Eigen::fix; +} + +} // end namespace Eigen + +#endif // EIGEN_ARITHMETIC_SEQUENCE_H diff --git a/thirdparty/eigen/Eigen/src/Core/Array.h b/thirdparty/eigen/Eigen/src/Core/Array.h index 0b9c38c8..6d50ea44 100644 --- a/thirdparty/eigen/Eigen/src/Core/Array.h +++ b/thirdparty/eigen/Eigen/src/Core/Array.h @@ -12,7 +12,16 @@ namespace Eigen { -/** \class Array +namespace internal { +template +struct traits > : traits > +{ + typedef ArrayXpr XprKind; + typedef ArrayBase > XprBase; +}; +} + +/** \class Array * \ingroup Core_Module * * \brief General-purpose arrays with easy API for coefficient-wise operations @@ -24,20 +33,14 @@ namespace Eigen { * API for the %Matrix class provides easy access to linear-algebra * operations. * + * See documentation of class Matrix for detailed information on the template parameters + * storage layout. + * * This class can be extended with the help of the plugin mechanism described on the page - * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN. + * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAY_PLUGIN. * - * \sa \ref TutorialArrayClass, \ref TopicClassHierarchy + * \sa \blank \ref TutorialArrayClass, \ref TopicClassHierarchy */ -namespace internal { -template -struct traits > : traits > -{ - typedef ArrayXpr XprKind; - typedef ArrayBase > XprBase; -}; -} - template class Array : public PlainObjectBase > @@ -69,11 +72,27 @@ class Array * the usage of 'using'. This should be done only for operator=. */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Array& operator=(const EigenBase &other) { return Base::operator=(other); } + /** Set all the entries to \a value. + * \sa DenseBase::setConstant(), DenseBase::fill() + */ + /* This overload is needed because the usage of + * using Base::operator=; + * fails on MSVC. Since the code below is working with GCC and MSVC, we skipped + * the usage of 'using'. This should be done only for operator=. + */ + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Array& operator=(const Scalar &value) + { + Base::setConstant(value); + return *this; + } + /** Copies the value of the expression \a other into \c *this with automatic resizing. * * *this might be resized to match the dimensions of \a other. If *this was a null matrix (not already initialized), @@ -84,7 +103,8 @@ class Array * remain row-vectors and vectors remain vectors. */ template - EIGEN_STRONG_INLINE Array& operator=(const ArrayBase& other) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Array& operator=(const DenseBase& other) { return Base::_set(other); } @@ -92,6 +112,7 @@ class Array /** This is a special case of the templated operator=. Its purpose is to * prevent a default operator= from hiding the templated operator=. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Array& operator=(const Array& other) { return Base::_set(other); @@ -107,6 +128,7 @@ class Array * * \sa resize(Index,Index) */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Array() : Base() { Base::_check_template_params(); @@ -116,6 +138,7 @@ class Array #ifndef EIGEN_PARSED_BY_DOXYGEN // FIXME is it still needed ?? /** \internal */ + EIGEN_DEVICE_FUNC Array(internal::constructor_without_unaligned_array_assert) : Base(internal::constructor_without_unaligned_array_assert()) { @@ -124,56 +147,114 @@ class Array } #endif -#ifdef EIGEN_HAVE_RVALUE_REFERENCES - Array(Array&& other) +#if EIGEN_HAS_RVALUE_REFERENCES + EIGEN_DEVICE_FUNC + Array(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible::value) : Base(std::move(other)) { Base::_check_template_params(); - if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic) - Base::_set_noalias(other); } - Array& operator=(Array&& other) + EIGEN_DEVICE_FUNC + Array& operator=(Array&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable::value) { - other.swap(*this); + Base::operator=(std::move(other)); return *this; } #endif - /** Constructs a vector or row-vector with given dimension. \only_for_vectors + #if EIGEN_HAS_CXX11 + /** \brief Construct a row of column vector with fixed size from an arbitrary number of coefficients. \cpp11 + * + * \only_for_vectors + * + * This constructor is for 1D array or vectors with more than 4 coefficients. + * There exists C++98 analogue constructors for fixed-size array/vector having 1, 2, 3, or 4 coefficients. + * + * \warning To construct a column (resp. row) vector of fixed length, the number of values passed to this + * constructor must match the the fixed number of rows (resp. columns) of \c *this. + * + * Example: \include Array_variadic_ctor_cxx11.cpp + * Output: \verbinclude Array_variadic_ctor_cxx11.out + * + * \sa Array(const std::initializer_list>&) + * \sa Array(const Scalar&), Array(const Scalar&,const Scalar&) + */ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) + : Base(a0, a1, a2, a3, args...) {} + + /** \brief Constructs an array and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11 * - * Note that this is only useful for dynamic-size vectors. For fixed-size vectors, - * it is redundant to pass the dimension here, so it makes more sense to use the default - * constructor Matrix() instead. + * In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients: + * + * Example: \include Array_initializer_list_23_cxx11.cpp + * Output: \verbinclude Array_initializer_list_23_cxx11.out + * + * Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered. + * + * In the case of a compile-time column 1D array, implicit transposition from a single row is allowed. + * Therefore Array{{1,2,3,4,5}} is legal and the more verbose syntax + * Array{{1},{2},{3},{4},{5}} can be avoided: + * + * Example: \include Array_initializer_list_vector_cxx11.cpp + * Output: \verbinclude Array_initializer_list_vector_cxx11.out + * + * In the case of fixed-sized arrays, the initializer list sizes must exactly match the array sizes, + * and implicit transposition is allowed for compile-time 1D arrays only. + * + * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) */ - EIGEN_STRONG_INLINE explicit Array(Index dim) - : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Array(const std::initializer_list>& list) : Base(list) {} + #endif // end EIGEN_HAS_CXX11 + + #ifndef EIGEN_PARSED_BY_DOXYGEN + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE explicit Array(const T& x) { Base::_check_template_params(); - EIGEN_STATIC_ASSERT_VECTOR_ONLY(Array) - eigen_assert(dim >= 0); - eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim); - EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED + Base::template _init1(x); } - #ifndef EIGEN_PARSED_BY_DOXYGEN template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Array(const T0& val0, const T1& val1) { Base::_check_template_params(); this->template _init2(val0, val1); } + #else - /** constructs an uninitialized matrix with \a rows rows and \a cols columns. + /** \brief Constructs a fixed-sized array initialized with coefficients starting at \a data */ + EIGEN_DEVICE_FUNC explicit Array(const Scalar *data); + /** Constructs a vector or row-vector with given dimension. \only_for_vectors * - * This is useful for dynamic-size matrices. For fixed-size matrices, + * Note that this is only useful for dynamic-size vectors. For fixed-size vectors, + * it is redundant to pass the dimension here, so it makes more sense to use the default + * constructor Array() instead. + */ + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE explicit Array(Index dim); + /** constructs an initialized 1x1 Array with the given coefficient + * \sa const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args */ + Array(const Scalar& value); + /** constructs an uninitialized array with \a rows rows and \a cols columns. + * + * This is useful for dynamic-size arrays. For fixed-size arrays, * it is redundant to pass these parameters, so one should use the default constructor - * Matrix() instead. */ + * Array() instead. */ Array(Index rows, Index cols); - /** constructs an initialized 2D vector with given coefficients */ + /** constructs an initialized 2D vector with given coefficients + * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) */ Array(const Scalar& val0, const Scalar& val1); - #endif + #endif // end EIGEN_PARSED_BY_DOXYGEN - /** constructs an initialized 3D vector with given coefficients */ + /** constructs an initialized 3D vector with given coefficients + * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) + */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2) { Base::_check_template_params(); @@ -182,7 +263,10 @@ class Array m_storage.data()[1] = val1; m_storage.data()[2] = val2; } - /** constructs an initialized 4D vector with given coefficients */ + /** constructs an initialized 4D vector with given coefficients + * \sa Array(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) + */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Array(const Scalar& val0, const Scalar& val1, const Scalar& val2, const Scalar& val3) { Base::_check_template_params(); @@ -193,51 +277,29 @@ class Array m_storage.data()[3] = val3; } - explicit Array(const Scalar *data); - - /** Constructor copying the value of the expression \a other */ - template - EIGEN_STRONG_INLINE Array(const ArrayBase& other) - : Base(other.rows() * other.cols(), other.rows(), other.cols()) - { - Base::_check_template_params(); - Base::_set_noalias(other); - } /** Copy constructor */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Array(const Array& other) - : Base(other.rows() * other.cols(), other.rows(), other.cols()) - { - Base::_check_template_params(); - Base::_set_noalias(other); - } - /** Copy constructor with in-place evaluation */ - template - EIGEN_STRONG_INLINE Array(const ReturnByValue& other) - { - Base::_check_template_params(); - Base::resize(other.rows(), other.cols()); - other.evalTo(*this); - } + : Base(other) + { } - /** \sa MatrixBase::operator=(const EigenBase&) */ - template - EIGEN_STRONG_INLINE Array(const EigenBase &other) - : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols()) - { - Base::_check_template_params(); - Base::_resize_to_match(other); - *this = other; - } + private: + struct PrivateType {}; + public: - /** Override MatrixBase::swap() since for dynamic-sized matrices of same type it is enough to swap the - * data pointers. - */ + /** \sa MatrixBase::operator=(const EigenBase&) */ template - void swap(ArrayBase const & other) - { this->_swap(other.derived()); } - - inline Index innerStride() const { return 1; } - inline Index outerStride() const { return this->innerSize(); } + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Array(const EigenBase &other, + typename internal::enable_if::value, + PrivateType>::type = PrivateType()) + : Base(other.derived()) + { } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const EIGEN_NOEXCEPT{ return 1; } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); } #ifdef EIGEN_ARRAY_PLUGIN #include EIGEN_ARRAY_PLUGIN @@ -252,7 +314,7 @@ class Array /** \defgroup arraytypedefs Global array typedefs * \ingroup Core_Module * - * Eigen defines several typedef shortcuts for most common 1D and 2D array types. + * %Eigen defines several typedef shortcuts for most common 1D and 2D array types. * * The general patterns are the following: * @@ -265,6 +327,12 @@ class Array * There are also \c ArraySizeType which are self-explanatory. For example, \c Array4cf is * a fixed-size 1D array of 4 complex floats. * + * With \cpp11, template alias are also defined for common sizes. + * They follow the same pattern as above except that the scalar type suffix is replaced by a + * template parameter, i.e.: + * - `ArrayRowsCols` where `Rows` and `Cols` can be \c 2,\c 3,\c 4, or \c X for fixed or dynamic size. + * - `ArraySize` where `Size` can be \c 2,\c 3,\c 4 or \c X for fixed or dynamic size 1D arrays. + * * \sa class Array */ @@ -297,8 +365,42 @@ EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES(std::complex, cd) #undef EIGEN_MAKE_ARRAY_TYPEDEFS_ALL_SIZES #undef EIGEN_MAKE_ARRAY_TYPEDEFS +#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS + +#if EIGEN_HAS_CXX11 + +#define EIGEN_MAKE_ARRAY_TYPEDEFS(Size, SizeSuffix) \ +/** \ingroup arraytypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Array##SizeSuffix##SizeSuffix = Array; \ +/** \ingroup arraytypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Array##SizeSuffix = Array; + +#define EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(Size) \ +/** \ingroup arraytypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Array##Size##X = Array; \ +/** \ingroup arraytypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Array##X##Size = Array; + +EIGEN_MAKE_ARRAY_TYPEDEFS(2, 2) +EIGEN_MAKE_ARRAY_TYPEDEFS(3, 3) +EIGEN_MAKE_ARRAY_TYPEDEFS(4, 4) +EIGEN_MAKE_ARRAY_TYPEDEFS(Dynamic, X) +EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(2) +EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(3) +EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS(4) + +#undef EIGEN_MAKE_ARRAY_TYPEDEFS +#undef EIGEN_MAKE_ARRAY_FIXED_TYPEDEFS -#undef EIGEN_MAKE_ARRAY_TYPEDEFS_LARGE +#endif // EIGEN_HAS_CXX11 #define EIGEN_USING_ARRAY_TYPEDEFS_FOR_TYPE_AND_SIZE(TypeSuffix, SizeSuffix) \ using Eigen::Matrix##SizeSuffix##TypeSuffix; \ diff --git a/thirdparty/eigen/Eigen/src/Core/ArrayBase.h b/thirdparty/eigen/Eigen/src/Core/ArrayBase.h index 33ff5537..ea3dd1c3 100644 --- a/thirdparty/eigen/Eigen/src/Core/ArrayBase.h +++ b/thirdparty/eigen/Eigen/src/Core/ArrayBase.h @@ -32,7 +32,7 @@ template class MatrixWrapper; * \tparam Derived is the derived type, e.g., an array or an expression type. * * This class can be extended with the help of the plugin mechanism described on the page - * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN. + * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_ARRAYBASE_PLUGIN. * * \sa class MatrixBase, \ref TopicClassHierarchy */ @@ -47,13 +47,11 @@ template class ArrayBase typedef ArrayBase Eigen_BaseClassForSpecializationOfGlobalMathFuncImpl; typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; typedef typename internal::packet_traits::type PacketScalar; typedef typename NumTraits::Real RealScalar; typedef DenseBase Base; - using Base::operator*; using Base::RowsAtCompileTime; using Base::ColsAtCompileTime; using Base::SizeAtCompileTime; @@ -62,8 +60,7 @@ template class ArrayBase using Base::MaxSizeAtCompileTime; using Base::IsVectorAtCompileTime; using Base::Flags; - using Base::CoeffReadCost; - + using Base::derived; using Base::const_cast_derived; using Base::rows; @@ -72,6 +69,7 @@ template class ArrayBase using Base::coeff; using Base::coeffRef; using Base::lazyAssign; + using Base::operator-; using Base::operator=; using Base::operator+=; using Base::operator-=; @@ -83,26 +81,14 @@ template class ArrayBase #endif // not EIGEN_PARSED_BY_DOXYGEN #ifndef EIGEN_PARSED_BY_DOXYGEN - /** \internal the plain matrix type corresponding to this expression. Note that is not necessarily - * exactly the return type of eval(): in the case of plain matrices, the return type of eval() is a const - * reference to a matrix, not a matrix! It is however guaranteed that the return type of eval() is either - * PlainObject or const PlainObject&. - */ - typedef Array::Scalar, - internal::traits::RowsAtCompileTime, - internal::traits::ColsAtCompileTime, - AutoAlign | (internal::traits::Flags&RowMajorBit ? RowMajor : ColMajor), - internal::traits::MaxRowsAtCompileTime, - internal::traits::MaxColsAtCompileTime - > PlainObject; - + typedef typename Base::PlainObject PlainObject; /** \internal Represents a matrix with all coefficients equal to one another*/ - typedef CwiseNullaryOp,Derived> ConstantReturnType; + typedef CwiseNullaryOp,PlainObject> ConstantReturnType; #endif // not EIGEN_PARSED_BY_DOXYGEN #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::ArrayBase -# include "../plugins/CommonCwiseUnaryOps.h" +#define EIGEN_DOC_UNARY_ADDONS(X,Y) # include "../plugins/MatrixCwiseUnaryOps.h" # include "../plugins/ArrayCwiseUnaryOps.h" # include "../plugins/CommonCwiseBinaryOps.h" @@ -112,45 +98,63 @@ template class ArrayBase # include EIGEN_ARRAYBASE_PLUGIN # endif #undef EIGEN_CURRENT_STORAGE_BASE_CLASS +#undef EIGEN_DOC_UNARY_ADDONS /** Special case of the template operator=, in order to prevent the compiler * from generating a default operator= (issue hit with g++ 4.1) */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const ArrayBase& other) { - return internal::assign_selector::run(derived(), other.derived()); + internal::call_assignment(derived(), other.derived()); + return derived(); } - - Derived& operator+=(const Scalar& scalar) - { return *this = derived() + scalar; } - Derived& operator-=(const Scalar& scalar) - { return *this = derived() - scalar; } + + /** Set all the entries to \a value. + * \sa DenseBase::setConstant(), DenseBase::fill() */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Derived& operator=(const Scalar &value) + { Base::setConstant(value); return derived(); } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Derived& operator+=(const Scalar& scalar); + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Derived& operator-=(const Scalar& scalar); template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator+=(const ArrayBase& other); template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator-=(const ArrayBase& other); template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator*=(const ArrayBase& other); template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator/=(const ArrayBase& other); public: + EIGEN_DEVICE_FUNC ArrayBase& array() { return *this; } + EIGEN_DEVICE_FUNC const ArrayBase& array() const { return *this; } /** \returns an \link Eigen::MatrixBase Matrix \endlink expression of this array * \sa MatrixBase::array() */ - MatrixWrapper matrix() { return derived(); } - const MatrixWrapper matrix() const { return derived(); } + EIGEN_DEVICE_FUNC + MatrixWrapper matrix() { return MatrixWrapper(derived()); } + EIGEN_DEVICE_FUNC + const MatrixWrapper matrix() const { return MatrixWrapper(derived()); } // template // inline void evalTo(Dest& dst) const { dst = matrix(); } protected: - ArrayBase() : Base() {} + EIGEN_DEFAULT_COPY_CONSTRUCTOR(ArrayBase) + EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(ArrayBase) private: explicit ArrayBase(Index); @@ -171,11 +175,10 @@ template class ArrayBase */ template template -EIGEN_STRONG_INLINE Derived & +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived & ArrayBase::operator-=(const ArrayBase &other) { - SelfCwiseBinaryOp, Derived, OtherDerived> tmp(derived()); - tmp = other.derived(); + call_assignment(derived(), other.derived(), internal::sub_assign_op()); return derived(); } @@ -185,11 +188,10 @@ ArrayBase::operator-=(const ArrayBase &other) */ template template -EIGEN_STRONG_INLINE Derived & +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived & ArrayBase::operator+=(const ArrayBase& other) { - SelfCwiseBinaryOp, Derived, OtherDerived> tmp(derived()); - tmp = other.derived(); + call_assignment(derived(), other.derived(), internal::add_assign_op()); return derived(); } @@ -199,11 +201,10 @@ ArrayBase::operator+=(const ArrayBase& other) */ template template -EIGEN_STRONG_INLINE Derived & +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived & ArrayBase::operator*=(const ArrayBase& other) { - SelfCwiseBinaryOp, Derived, OtherDerived> tmp(derived()); - tmp = other.derived(); + call_assignment(derived(), other.derived(), internal::mul_assign_op()); return derived(); } @@ -213,11 +214,10 @@ ArrayBase::operator*=(const ArrayBase& other) */ template template -EIGEN_STRONG_INLINE Derived & +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived & ArrayBase::operator/=(const ArrayBase& other) { - SelfCwiseBinaryOp, Derived, OtherDerived> tmp(derived()); - tmp = other.derived(); + call_assignment(derived(), other.derived(), internal::div_assign_op()); return derived(); } diff --git a/thirdparty/eigen/Eigen/src/Core/ArrayWrapper.h b/thirdparty/eigen/Eigen/src/Core/ArrayWrapper.h index b4641e2a..2e9555b5 100644 --- a/thirdparty/eigen/Eigen/src/Core/ArrayWrapper.h +++ b/thirdparty/eigen/Eigen/src/Core/ArrayWrapper.h @@ -10,7 +10,7 @@ #ifndef EIGEN_ARRAYWRAPPER_H #define EIGEN_ARRAYWRAPPER_H -namespace Eigen { +namespace Eigen { /** \class ArrayWrapper * \ingroup Core_Module @@ -32,7 +32,8 @@ struct traits > // Let's remove NestByRefBit enum { Flags0 = traits::type >::Flags, - Flags = Flags0 & ~NestByRefBit + LvalueBitFlag = is_lvalue::value ? LvalueBit : 0, + Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag }; }; } @@ -44,6 +45,7 @@ class ArrayWrapper : public ArrayBase > typedef ArrayBase Base; EIGEN_DENSE_PUBLIC_INTERFACE(ArrayWrapper) EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ArrayWrapper) + typedef typename internal::remove_all::type NestedExpression; typedef typename internal::conditional< internal::is_lvalue::value, @@ -51,87 +53,58 @@ class ArrayWrapper : public ArrayBase > const Scalar >::type ScalarWithConstIfNotLvalue; - typedef typename internal::nested::type NestedExpressionType; + typedef typename internal::ref_selector::non_const_type NestedExpressionType; - inline ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {} + using Base::coeffRef; - inline Index rows() const { return m_expression.rows(); } - inline Index cols() const { return m_expression.cols(); } - inline Index outerStride() const { return m_expression.outerStride(); } - inline Index innerStride() const { return m_expression.innerStride(); } + EIGEN_DEVICE_FUNC + explicit EIGEN_STRONG_INLINE ArrayWrapper(ExpressionType& matrix) : m_expression(matrix) {} - inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); } - inline const Scalar* data() const { return m_expression.data(); } - - inline CoeffReturnType coeff(Index rowId, Index colId) const - { - return m_expression.coeff(rowId, colId); - } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); } - inline Scalar& coeffRef(Index rowId, Index colId) - { - return m_expression.const_cast_derived().coeffRef(rowId, colId); - } + EIGEN_DEVICE_FUNC + inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); } + EIGEN_DEVICE_FUNC + inline const Scalar* data() const { return m_expression.data(); } + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index rowId, Index colId) const { - return m_expression.const_cast_derived().coeffRef(rowId, colId); - } - - inline CoeffReturnType coeff(Index index) const - { - return m_expression.coeff(index); - } - - inline Scalar& coeffRef(Index index) - { - return m_expression.const_cast_derived().coeffRef(index); + return m_expression.coeffRef(rowId, colId); } + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index index) const { - return m_expression.const_cast_derived().coeffRef(index); - } - - template - inline const PacketScalar packet(Index rowId, Index colId) const - { - return m_expression.template packet(rowId, colId); - } - - template - inline void writePacket(Index rowId, Index colId, const PacketScalar& val) - { - m_expression.const_cast_derived().template writePacket(rowId, colId, val); - } - - template - inline const PacketScalar packet(Index index) const - { - return m_expression.template packet(index); - } - - template - inline void writePacket(Index index, const PacketScalar& val) - { - m_expression.const_cast_derived().template writePacket(index, val); + return m_expression.coeffRef(index); } template + EIGEN_DEVICE_FUNC inline void evalTo(Dest& dst) const { dst = m_expression; } - const typename internal::remove_all::type& - nestedExpression() const + EIGEN_DEVICE_FUNC + const typename internal::remove_all::type& + nestedExpression() const { return m_expression; } /** Forwards the resizing request to the nested expression * \sa DenseBase::resize(Index) */ - void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); } + EIGEN_DEVICE_FUNC + void resize(Index newSize) { m_expression.resize(newSize); } /** Forwards the resizing request to the nested expression * \sa DenseBase::resize(Index,Index)*/ - void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); } + EIGEN_DEVICE_FUNC + void resize(Index rows, Index cols) { m_expression.resize(rows,cols); } protected: NestedExpressionType m_expression; @@ -157,7 +130,8 @@ struct traits > // Let's remove NestByRefBit enum { Flags0 = traits::type >::Flags, - Flags = Flags0 & ~NestByRefBit + LvalueBitFlag = is_lvalue::value ? LvalueBit : 0, + Flags = (Flags0 & ~(NestByRefBit | LvalueBit)) | LvalueBitFlag }; }; } @@ -169,6 +143,7 @@ class MatrixWrapper : public MatrixBase > typedef MatrixBase > Base; EIGEN_DENSE_PUBLIC_INTERFACE(MatrixWrapper) EIGEN_INHERIT_ASSIGNMENT_OPERATORS(MatrixWrapper) + typedef typename internal::remove_all::type NestedExpression; typedef typename internal::conditional< internal::is_lvalue::value, @@ -176,84 +151,54 @@ class MatrixWrapper : public MatrixBase > const Scalar >::type ScalarWithConstIfNotLvalue; - typedef typename internal::nested::type NestedExpressionType; + typedef typename internal::ref_selector::non_const_type NestedExpressionType; - inline MatrixWrapper(ExpressionType& a_matrix) : m_expression(a_matrix) {} + using Base::coeffRef; - inline Index rows() const { return m_expression.rows(); } - inline Index cols() const { return m_expression.cols(); } - inline Index outerStride() const { return m_expression.outerStride(); } - inline Index innerStride() const { return m_expression.innerStride(); } + EIGEN_DEVICE_FUNC + explicit inline MatrixWrapper(ExpressionType& matrix) : m_expression(matrix) {} - inline ScalarWithConstIfNotLvalue* data() { return m_expression.const_cast_derived().data(); } - inline const Scalar* data() const { return m_expression.data(); } - - inline CoeffReturnType coeff(Index rowId, Index colId) const - { - return m_expression.coeff(rowId, colId); - } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); } - inline Scalar& coeffRef(Index rowId, Index colId) - { - return m_expression.const_cast_derived().coeffRef(rowId, colId); - } + EIGEN_DEVICE_FUNC + inline ScalarWithConstIfNotLvalue* data() { return m_expression.data(); } + EIGEN_DEVICE_FUNC + inline const Scalar* data() const { return m_expression.data(); } + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index rowId, Index colId) const { return m_expression.derived().coeffRef(rowId, colId); } - inline CoeffReturnType coeff(Index index) const - { - return m_expression.coeff(index); - } - - inline Scalar& coeffRef(Index index) - { - return m_expression.const_cast_derived().coeffRef(index); - } - + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index index) const { - return m_expression.const_cast_derived().coeffRef(index); - } - - template - inline const PacketScalar packet(Index rowId, Index colId) const - { - return m_expression.template packet(rowId, colId); - } - - template - inline void writePacket(Index rowId, Index colId, const PacketScalar& val) - { - m_expression.const_cast_derived().template writePacket(rowId, colId, val); - } - - template - inline const PacketScalar packet(Index index) const - { - return m_expression.template packet(index); - } - - template - inline void writePacket(Index index, const PacketScalar& val) - { - m_expression.const_cast_derived().template writePacket(index, val); + return m_expression.coeffRef(index); } - const typename internal::remove_all::type& - nestedExpression() const + EIGEN_DEVICE_FUNC + const typename internal::remove_all::type& + nestedExpression() const { return m_expression; } /** Forwards the resizing request to the nested expression * \sa DenseBase::resize(Index) */ - void resize(Index newSize) { m_expression.const_cast_derived().resize(newSize); } + EIGEN_DEVICE_FUNC + void resize(Index newSize) { m_expression.resize(newSize); } /** Forwards the resizing request to the nested expression * \sa DenseBase::resize(Index,Index)*/ - void resize(Index nbRows, Index nbCols) { m_expression.const_cast_derived().resize(nbRows,nbCols); } + EIGEN_DEVICE_FUNC + void resize(Index rows, Index cols) { m_expression.resize(rows,cols); } protected: NestedExpressionType m_expression; diff --git a/thirdparty/eigen/Eigen/src/Core/Assign.h b/thirdparty/eigen/Eigen/src/Core/Assign.h index f4817317..655412ef 100644 --- a/thirdparty/eigen/Eigen/src/Core/Assign.h +++ b/thirdparty/eigen/Eigen/src/Core/Assign.h @@ -14,481 +14,9 @@ namespace Eigen { -namespace internal { - -/*************************************************************************** -* Part 1 : the logic deciding a strategy for traversal and unrolling * -***************************************************************************/ - -template -struct assign_traits -{ -public: - enum { - DstIsAligned = Derived::Flags & AlignedBit, - DstHasDirectAccess = Derived::Flags & DirectAccessBit, - SrcIsAligned = OtherDerived::Flags & AlignedBit, - JointAlignment = bool(DstIsAligned) && bool(SrcIsAligned) ? Aligned : Unaligned - }; - -private: - enum { - InnerSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::SizeAtCompileTime) - : int(Derived::Flags)&RowMajorBit ? int(Derived::ColsAtCompileTime) - : int(Derived::RowsAtCompileTime), - InnerMaxSize = int(Derived::IsVectorAtCompileTime) ? int(Derived::MaxSizeAtCompileTime) - : int(Derived::Flags)&RowMajorBit ? int(Derived::MaxColsAtCompileTime) - : int(Derived::MaxRowsAtCompileTime), - MaxSizeAtCompileTime = Derived::SizeAtCompileTime, - PacketSize = packet_traits::size - }; - - enum { - StorageOrdersAgree = (int(Derived::IsRowMajor) == int(OtherDerived::IsRowMajor)), - MightVectorize = StorageOrdersAgree - && (int(Derived::Flags) & int(OtherDerived::Flags) & ActualPacketAccessBit), - MayInnerVectorize = MightVectorize && int(InnerSize)!=Dynamic && int(InnerSize)%int(PacketSize)==0 - && int(DstIsAligned) && int(SrcIsAligned), - MayLinearize = StorageOrdersAgree && (int(Derived::Flags) & int(OtherDerived::Flags) & LinearAccessBit), - MayLinearVectorize = MightVectorize && MayLinearize && DstHasDirectAccess - && (DstIsAligned || MaxSizeAtCompileTime == Dynamic), - /* If the destination isn't aligned, we have to do runtime checks and we don't unroll, - so it's only good for large enough sizes. */ - MaySliceVectorize = MightVectorize && DstHasDirectAccess - && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=3*PacketSize) - /* slice vectorization can be slow, so we only want it if the slices are big, which is - indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block - in a fixed-size matrix */ - }; - -public: - enum { - Traversal = int(MayInnerVectorize) ? int(InnerVectorizedTraversal) - : int(MayLinearVectorize) ? int(LinearVectorizedTraversal) - : int(MaySliceVectorize) ? int(SliceVectorizedTraversal) - : int(MayLinearize) ? int(LinearTraversal) - : int(DefaultTraversal), - Vectorized = int(Traversal) == InnerVectorizedTraversal - || int(Traversal) == LinearVectorizedTraversal - || int(Traversal) == SliceVectorizedTraversal - }; - -private: - enum { - UnrollingLimit = EIGEN_UNROLLING_LIMIT * (Vectorized ? int(PacketSize) : 1), - MayUnrollCompletely = int(Derived::SizeAtCompileTime) != Dynamic - && int(OtherDerived::CoeffReadCost) != Dynamic - && int(Derived::SizeAtCompileTime) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit), - MayUnrollInner = int(InnerSize) != Dynamic - && int(OtherDerived::CoeffReadCost) != Dynamic - && int(InnerSize) * int(OtherDerived::CoeffReadCost) <= int(UnrollingLimit) - }; - -public: - enum { - Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal)) - ? ( - int(MayUnrollCompletely) ? int(CompleteUnrolling) - : int(MayUnrollInner) ? int(InnerUnrolling) - : int(NoUnrolling) - ) - : int(Traversal) == int(LinearVectorizedTraversal) - ? ( bool(MayUnrollCompletely) && bool(DstIsAligned) ? int(CompleteUnrolling) : int(NoUnrolling) ) - : int(Traversal) == int(LinearTraversal) - ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) : int(NoUnrolling) ) - : int(NoUnrolling) - }; - -#ifdef EIGEN_DEBUG_ASSIGN - static void debug() - { - EIGEN_DEBUG_VAR(DstIsAligned) - EIGEN_DEBUG_VAR(SrcIsAligned) - EIGEN_DEBUG_VAR(JointAlignment) - EIGEN_DEBUG_VAR(InnerSize) - EIGEN_DEBUG_VAR(InnerMaxSize) - EIGEN_DEBUG_VAR(PacketSize) - EIGEN_DEBUG_VAR(StorageOrdersAgree) - EIGEN_DEBUG_VAR(MightVectorize) - EIGEN_DEBUG_VAR(MayLinearize) - EIGEN_DEBUG_VAR(MayInnerVectorize) - EIGEN_DEBUG_VAR(MayLinearVectorize) - EIGEN_DEBUG_VAR(MaySliceVectorize) - EIGEN_DEBUG_VAR(Traversal) - EIGEN_DEBUG_VAR(UnrollingLimit) - EIGEN_DEBUG_VAR(MayUnrollCompletely) - EIGEN_DEBUG_VAR(MayUnrollInner) - EIGEN_DEBUG_VAR(Unrolling) - } -#endif -}; - -/*************************************************************************** -* Part 2 : meta-unrollers -***************************************************************************/ - -/************************ -*** Default traversal *** -************************/ - -template -struct assign_DefaultTraversal_CompleteUnrolling -{ - enum { - outer = Index / Derived1::InnerSizeAtCompileTime, - inner = Index % Derived1::InnerSizeAtCompileTime - }; - - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - dst.copyCoeffByOuterInner(outer, inner, src); - assign_DefaultTraversal_CompleteUnrolling::run(dst, src); - } -}; - -template -struct assign_DefaultTraversal_CompleteUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {} -}; - -template -struct assign_DefaultTraversal_InnerUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer) - { - dst.copyCoeffByOuterInner(outer, Index, src); - assign_DefaultTraversal_InnerUnrolling::run(dst, src, outer); - } -}; - -template -struct assign_DefaultTraversal_InnerUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {} -}; - -/*********************** -*** Linear traversal *** -***********************/ - -template -struct assign_LinearTraversal_CompleteUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - dst.copyCoeff(Index, src); - assign_LinearTraversal_CompleteUnrolling::run(dst, src); - } -}; - -template -struct assign_LinearTraversal_CompleteUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {} -}; - -/************************** -*** Inner vectorization *** -**************************/ - -template -struct assign_innervec_CompleteUnrolling -{ - enum { - outer = Index / Derived1::InnerSizeAtCompileTime, - inner = Index % Derived1::InnerSizeAtCompileTime, - JointAlignment = assign_traits::JointAlignment - }; - - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - dst.template copyPacketByOuterInner(outer, inner, src); - assign_innervec_CompleteUnrolling::size, Stop>::run(dst, src); - } -}; - -template -struct assign_innervec_CompleteUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &) {} -}; - -template -struct assign_innervec_InnerUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src, typename Derived1::Index outer) - { - dst.template copyPacketByOuterInner(outer, Index, src); - assign_innervec_InnerUnrolling::size, Stop>::run(dst, src, outer); - } -}; - -template -struct assign_innervec_InnerUnrolling -{ - static EIGEN_STRONG_INLINE void run(Derived1 &, const Derived2 &, typename Derived1::Index) {} -}; - -/*************************************************************************** -* Part 3 : implementation of all cases -***************************************************************************/ - -template::Traversal, - int Unrolling = assign_traits::Unrolling, - int Version = Specialized> -struct assign_impl; - -/************************ -*** Default traversal *** -************************/ - -template -struct assign_impl -{ - static inline void run(Derived1 &, const Derived2 &) { } -}; - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static inline void run(Derived1 &dst, const Derived2 &src) - { - const Index innerSize = dst.innerSize(); - const Index outerSize = dst.outerSize(); - for(Index outer = 0; outer < outerSize; ++outer) - for(Index inner = 0; inner < innerSize; ++inner) - dst.copyCoeffByOuterInner(outer, inner, src); - } -}; - -template -struct assign_impl -{ - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - assign_DefaultTraversal_CompleteUnrolling - ::run(dst, src); - } -}; - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - const Index outerSize = dst.outerSize(); - for(Index outer = 0; outer < outerSize; ++outer) - assign_DefaultTraversal_InnerUnrolling - ::run(dst, src, outer); - } -}; - -/*********************** -*** Linear traversal *** -***********************/ - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static inline void run(Derived1 &dst, const Derived2 &src) - { - const Index size = dst.size(); - for(Index i = 0; i < size; ++i) - dst.copyCoeff(i, src); - } -}; - -template -struct assign_impl -{ - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - assign_LinearTraversal_CompleteUnrolling - ::run(dst, src); - } -}; - -/************************** -*** Inner vectorization *** -**************************/ - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static inline void run(Derived1 &dst, const Derived2 &src) - { - const Index innerSize = dst.innerSize(); - const Index outerSize = dst.outerSize(); - const Index packetSize = packet_traits::size; - for(Index outer = 0; outer < outerSize; ++outer) - for(Index inner = 0; inner < innerSize; inner+=packetSize) - dst.template copyPacketByOuterInner(outer, inner, src); - } -}; - -template -struct assign_impl -{ - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - assign_innervec_CompleteUnrolling - ::run(dst, src); - } -}; - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - const Index outerSize = dst.outerSize(); - for(Index outer = 0; outer < outerSize; ++outer) - assign_innervec_InnerUnrolling - ::run(dst, src, outer); - } -}; - -/*************************** -*** Linear vectorization *** -***************************/ - -template -struct unaligned_assign_impl -{ - template - static EIGEN_STRONG_INLINE void run(const Derived&, OtherDerived&, typename Derived::Index, typename Derived::Index) {} -}; - -template <> -struct unaligned_assign_impl -{ - // MSVC must not inline this functions. If it does, it fails to optimize the - // packet access path. -#ifdef _MSC_VER - template - static EIGEN_DONT_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end) -#else - template - static EIGEN_STRONG_INLINE void run(const Derived& src, OtherDerived& dst, typename Derived::Index start, typename Derived::Index end) -#endif - { - for (typename Derived::Index index = start; index < end; ++index) - dst.copyCoeff(index, src); - } -}; - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - const Index size = dst.size(); - typedef packet_traits PacketTraits; - enum { - packetSize = PacketTraits::size, - dstAlignment = PacketTraits::AlignedOnScalar ? Aligned : int(assign_traits::DstIsAligned) , - srcAlignment = assign_traits::JointAlignment - }; - const Index alignedStart = assign_traits::DstIsAligned ? 0 - : internal::first_aligned(&dst.coeffRef(0), size); - const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize; - - unaligned_assign_impl::DstIsAligned!=0>::run(src,dst,0,alignedStart); - - for(Index index = alignedStart; index < alignedEnd; index += packetSize) - { - dst.template copyPacket(index, src); - } - - unaligned_assign_impl<>::run(src,dst,alignedEnd,size); - } -}; - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static EIGEN_STRONG_INLINE void run(Derived1 &dst, const Derived2 &src) - { - enum { size = Derived1::SizeAtCompileTime, - packetSize = packet_traits::size, - alignedSize = (size/packetSize)*packetSize }; - - assign_innervec_CompleteUnrolling::run(dst, src); - assign_DefaultTraversal_CompleteUnrolling::run(dst, src); - } -}; - -/************************** -*** Slice vectorization *** -***************************/ - -template -struct assign_impl -{ - typedef typename Derived1::Index Index; - static inline void run(Derived1 &dst, const Derived2 &src) - { - typedef typename Derived1::Scalar Scalar; - typedef packet_traits PacketTraits; - enum { - packetSize = PacketTraits::size, - alignable = PacketTraits::AlignedOnScalar, - dstIsAligned = assign_traits::DstIsAligned, - dstAlignment = alignable ? Aligned : int(dstIsAligned), - srcAlignment = assign_traits::JointAlignment - }; - const Scalar *dst_ptr = &dst.coeffRef(0,0); - if((!bool(dstIsAligned)) && (size_t(dst_ptr) % sizeof(Scalar))>0) - { - // the pointer is not aligend-on scalar, so alignment is not possible - return assign_impl::run(dst, src); - } - const Index packetAlignedMask = packetSize - 1; - const Index innerSize = dst.innerSize(); - const Index outerSize = dst.outerSize(); - const Index alignedStep = alignable ? (packetSize - dst.outerStride() % packetSize) & packetAlignedMask : 0; - Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned(dst_ptr, innerSize); - - for(Index outer = 0; outer < outerSize; ++outer) - { - const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask); - // do the non-vectorizable part of the assignment - for(Index inner = 0; inner(outer, inner, src); - - // do the non-vectorizable part of the assignment - for(Index inner = alignedEnd; inner((alignedStart+alignedStep)%packetSize, innerSize); - } - } -}; - -} // end namespace internal - -/*************************************************************************** -* Part 4 : implementation of DenseBase methods -***************************************************************************/ - template template -EIGEN_STRONG_INLINE Derived& DenseBase +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase ::lazyAssign(const DenseBase& other) { enum{ @@ -499,90 +27,62 @@ EIGEN_STRONG_INLINE Derived& DenseBase EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Derived,OtherDerived) EIGEN_STATIC_ASSERT(SameType,YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) -#ifdef EIGEN_DEBUG_ASSIGN - internal::assign_traits::debug(); -#endif eigen_assert(rows() == other.rows() && cols() == other.cols()); - internal::assign_impl::Traversal) - : int(InvalidTraversal)>::run(derived(),other.derived()); -#ifndef EIGEN_NO_DEBUG - checkTransposeAliasing(other.derived()); -#endif + internal::call_assignment_no_alias(derived(),other.derived()); + return derived(); } -namespace internal { - -template::Flags) & EvalBeforeAssigningBit) != 0, - bool NeedToTranspose = ((int(Derived::RowsAtCompileTime) == 1 && int(OtherDerived::ColsAtCompileTime) == 1) - | // FIXME | instead of || to please GCC 4.4.0 stupid warning "suggest parentheses around &&". - // revert to || as soon as not needed anymore. - (int(Derived::ColsAtCompileTime) == 1 && int(OtherDerived::RowsAtCompileTime) == 1)) - && int(Derived::SizeAtCompileTime) != 1> -struct assign_selector; - -template -struct assign_selector { - static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.derived()); } - template - static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { other.evalTo(dst); return dst; } -}; -template -struct assign_selector { - static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.eval()); } -}; -template -struct assign_selector { - static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose()); } - template - static EIGEN_STRONG_INLINE Derived& evalTo(ActualDerived& dst, const ActualOtherDerived& other) { Transpose dstTrans(dst); other.evalTo(dstTrans); return dst; } -}; -template -struct assign_selector { - static EIGEN_STRONG_INLINE Derived& run(Derived& dst, const OtherDerived& other) { return dst.lazyAssign(other.transpose().eval()); } -}; - -} // end namespace internal - template template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::operator=(const DenseBase& other) { - return internal::assign_selector::run(derived(), other.derived()); + internal::call_assignment(derived(), other.derived()); + return derived(); } template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::operator=(const DenseBase& other) { - return internal::assign_selector::run(derived(), other.derived()); + internal::call_assignment(derived(), other.derived()); + return derived(); } template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const MatrixBase& other) { - return internal::assign_selector::run(derived(), other.derived()); + internal::call_assignment(derived(), other.derived()); + return derived(); } template template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const DenseBase& other) { - return internal::assign_selector::run(derived(), other.derived()); + internal::call_assignment(derived(), other.derived()); + return derived(); } template template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const EigenBase& other) { - return internal::assign_selector::evalTo(derived(), other.derived()); + internal::call_assignment(derived(), other.derived()); + return derived(); } template template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::operator=(const ReturnByValue& other) { - return internal::assign_selector::evalTo(derived(), other.derived()); + other.derived().evalTo(derived()); + return derived(); } } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/AssignEvaluator.h b/thirdparty/eigen/Eigen/src/Core/AssignEvaluator.h new file mode 100644 index 00000000..7d76f0c2 --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/AssignEvaluator.h @@ -0,0 +1,1010 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2011 Benoit Jacob +// Copyright (C) 2011-2014 Gael Guennebaud +// Copyright (C) 2011-2012 Jitse Niesen +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_ASSIGN_EVALUATOR_H +#define EIGEN_ASSIGN_EVALUATOR_H + +namespace Eigen { + +// This implementation is based on Assign.h + +namespace internal { + +/*************************************************************************** +* Part 1 : the logic deciding a strategy for traversal and unrolling * +***************************************************************************/ + +// copy_using_evaluator_traits is based on assign_traits + +template +struct copy_using_evaluator_traits +{ + typedef typename DstEvaluator::XprType Dst; + typedef typename Dst::Scalar DstScalar; + + enum { + DstFlags = DstEvaluator::Flags, + SrcFlags = SrcEvaluator::Flags + }; + +public: + enum { + DstAlignment = DstEvaluator::Alignment, + SrcAlignment = SrcEvaluator::Alignment, + DstHasDirectAccess = (DstFlags & DirectAccessBit) == DirectAccessBit, + JointAlignment = EIGEN_PLAIN_ENUM_MIN(DstAlignment,SrcAlignment) + }; + +private: + enum { + InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime) + : int(DstFlags)&RowMajorBit ? int(Dst::ColsAtCompileTime) + : int(Dst::RowsAtCompileTime), + InnerMaxSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::MaxSizeAtCompileTime) + : int(DstFlags)&RowMajorBit ? int(Dst::MaxColsAtCompileTime) + : int(Dst::MaxRowsAtCompileTime), + RestrictedInnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(InnerSize,MaxPacketSize), + RestrictedLinearSize = EIGEN_SIZE_MIN_PREFER_FIXED(Dst::SizeAtCompileTime,MaxPacketSize), + OuterStride = int(outer_stride_at_compile_time::ret), + MaxSizeAtCompileTime = Dst::SizeAtCompileTime + }; + + // TODO distinguish between linear traversal and inner-traversals + typedef typename find_best_packet::type LinearPacketType; + typedef typename find_best_packet::type InnerPacketType; + + enum { + LinearPacketSize = unpacket_traits::size, + InnerPacketSize = unpacket_traits::size + }; + +public: + enum { + LinearRequiredAlignment = unpacket_traits::alignment, + InnerRequiredAlignment = unpacket_traits::alignment + }; + +private: + enum { + DstIsRowMajor = DstFlags&RowMajorBit, + SrcIsRowMajor = SrcFlags&RowMajorBit, + StorageOrdersAgree = (int(DstIsRowMajor) == int(SrcIsRowMajor)), + MightVectorize = bool(StorageOrdersAgree) + && (int(DstFlags) & int(SrcFlags) & ActualPacketAccessBit) + && bool(functor_traits::PacketAccess), + MayInnerVectorize = MightVectorize + && int(InnerSize)!=Dynamic && int(InnerSize)%int(InnerPacketSize)==0 + && int(OuterStride)!=Dynamic && int(OuterStride)%int(InnerPacketSize)==0 + && (EIGEN_UNALIGNED_VECTORIZE || int(JointAlignment)>=int(InnerRequiredAlignment)), + MayLinearize = bool(StorageOrdersAgree) && (int(DstFlags) & int(SrcFlags) & LinearAccessBit), + MayLinearVectorize = bool(MightVectorize) && bool(MayLinearize) && bool(DstHasDirectAccess) + && (EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment)) || MaxSizeAtCompileTime == Dynamic), + /* If the destination isn't aligned, we have to do runtime checks and we don't unroll, + so it's only good for large enough sizes. */ + MaySliceVectorize = bool(MightVectorize) && bool(DstHasDirectAccess) + && (int(InnerMaxSize)==Dynamic || int(InnerMaxSize)>=(EIGEN_UNALIGNED_VECTORIZE?InnerPacketSize:(3*InnerPacketSize))) + /* slice vectorization can be slow, so we only want it if the slices are big, which is + indicated by InnerMaxSize rather than InnerSize, think of the case of a dynamic block + in a fixed-size matrix + However, with EIGEN_UNALIGNED_VECTORIZE and unrolling, slice vectorization is still worth it */ + }; + +public: + enum { + Traversal = int(Dst::SizeAtCompileTime) == 0 ? int(AllAtOnceTraversal) // If compile-size is zero, traversing will fail at compile-time. + : (int(MayLinearVectorize) && (LinearPacketSize>InnerPacketSize)) ? int(LinearVectorizedTraversal) + : int(MayInnerVectorize) ? int(InnerVectorizedTraversal) + : int(MayLinearVectorize) ? int(LinearVectorizedTraversal) + : int(MaySliceVectorize) ? int(SliceVectorizedTraversal) + : int(MayLinearize) ? int(LinearTraversal) + : int(DefaultTraversal), + Vectorized = int(Traversal) == InnerVectorizedTraversal + || int(Traversal) == LinearVectorizedTraversal + || int(Traversal) == SliceVectorizedTraversal + }; + + typedef typename conditional::type PacketType; + +private: + enum { + ActualPacketSize = int(Traversal)==LinearVectorizedTraversal ? LinearPacketSize + : Vectorized ? InnerPacketSize + : 1, + UnrollingLimit = EIGEN_UNROLLING_LIMIT * ActualPacketSize, + MayUnrollCompletely = int(Dst::SizeAtCompileTime) != Dynamic + && int(Dst::SizeAtCompileTime) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit), + MayUnrollInner = int(InnerSize) != Dynamic + && int(InnerSize) * (int(DstEvaluator::CoeffReadCost)+int(SrcEvaluator::CoeffReadCost)) <= int(UnrollingLimit) + }; + +public: + enum { + Unrolling = (int(Traversal) == int(InnerVectorizedTraversal) || int(Traversal) == int(DefaultTraversal)) + ? ( + int(MayUnrollCompletely) ? int(CompleteUnrolling) + : int(MayUnrollInner) ? int(InnerUnrolling) + : int(NoUnrolling) + ) + : int(Traversal) == int(LinearVectorizedTraversal) + ? ( bool(MayUnrollCompletely) && ( EIGEN_UNALIGNED_VECTORIZE || (int(DstAlignment)>=int(LinearRequiredAlignment))) + ? int(CompleteUnrolling) + : int(NoUnrolling) ) + : int(Traversal) == int(LinearTraversal) + ? ( bool(MayUnrollCompletely) ? int(CompleteUnrolling) + : int(NoUnrolling) ) +#if EIGEN_UNALIGNED_VECTORIZE + : int(Traversal) == int(SliceVectorizedTraversal) + ? ( bool(MayUnrollInner) ? int(InnerUnrolling) + : int(NoUnrolling) ) +#endif + : int(NoUnrolling) + }; + +#ifdef EIGEN_DEBUG_ASSIGN + static void debug() + { + std::cerr << "DstXpr: " << typeid(typename DstEvaluator::XprType).name() << std::endl; + std::cerr << "SrcXpr: " << typeid(typename SrcEvaluator::XprType).name() << std::endl; + std::cerr.setf(std::ios::hex, std::ios::basefield); + std::cerr << "DstFlags" << " = " << DstFlags << " (" << demangle_flags(DstFlags) << " )" << std::endl; + std::cerr << "SrcFlags" << " = " << SrcFlags << " (" << demangle_flags(SrcFlags) << " )" << std::endl; + std::cerr.unsetf(std::ios::hex); + EIGEN_DEBUG_VAR(DstAlignment) + EIGEN_DEBUG_VAR(SrcAlignment) + EIGEN_DEBUG_VAR(LinearRequiredAlignment) + EIGEN_DEBUG_VAR(InnerRequiredAlignment) + EIGEN_DEBUG_VAR(JointAlignment) + EIGEN_DEBUG_VAR(InnerSize) + EIGEN_DEBUG_VAR(InnerMaxSize) + EIGEN_DEBUG_VAR(LinearPacketSize) + EIGEN_DEBUG_VAR(InnerPacketSize) + EIGEN_DEBUG_VAR(ActualPacketSize) + EIGEN_DEBUG_VAR(StorageOrdersAgree) + EIGEN_DEBUG_VAR(MightVectorize) + EIGEN_DEBUG_VAR(MayLinearize) + EIGEN_DEBUG_VAR(MayInnerVectorize) + EIGEN_DEBUG_VAR(MayLinearVectorize) + EIGEN_DEBUG_VAR(MaySliceVectorize) + std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl; + EIGEN_DEBUG_VAR(SrcEvaluator::CoeffReadCost) + EIGEN_DEBUG_VAR(DstEvaluator::CoeffReadCost) + EIGEN_DEBUG_VAR(Dst::SizeAtCompileTime) + EIGEN_DEBUG_VAR(UnrollingLimit) + EIGEN_DEBUG_VAR(MayUnrollCompletely) + EIGEN_DEBUG_VAR(MayUnrollInner) + std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl; + std::cerr << std::endl; + } +#endif +}; + +/*************************************************************************** +* Part 2 : meta-unrollers +***************************************************************************/ + +/************************ +*** Default traversal *** +************************/ + +template +struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling +{ + // FIXME: this is not very clean, perhaps this information should be provided by the kernel? + typedef typename Kernel::DstEvaluatorType DstEvaluatorType; + typedef typename DstEvaluatorType::XprType DstXprType; + + enum { + outer = Index / DstXprType::InnerSizeAtCompileTime, + inner = Index % DstXprType::InnerSizeAtCompileTime + }; + + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + kernel.assignCoeffByOuterInner(outer, inner); + copy_using_evaluator_DefaultTraversal_CompleteUnrolling::run(kernel); + } +}; + +template +struct copy_using_evaluator_DefaultTraversal_CompleteUnrolling +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { } +}; + +template +struct copy_using_evaluator_DefaultTraversal_InnerUnrolling +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer) + { + kernel.assignCoeffByOuterInner(outer, Index_); + copy_using_evaluator_DefaultTraversal_InnerUnrolling::run(kernel, outer); + } +}; + +template +struct copy_using_evaluator_DefaultTraversal_InnerUnrolling +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index) { } +}; + +/*********************** +*** Linear traversal *** +***********************/ + +template +struct copy_using_evaluator_LinearTraversal_CompleteUnrolling +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel& kernel) + { + kernel.assignCoeff(Index); + copy_using_evaluator_LinearTraversal_CompleteUnrolling::run(kernel); + } +}; + +template +struct copy_using_evaluator_LinearTraversal_CompleteUnrolling +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { } +}; + +/************************** +*** Inner vectorization *** +**************************/ + +template +struct copy_using_evaluator_innervec_CompleteUnrolling +{ + // FIXME: this is not very clean, perhaps this information should be provided by the kernel? + typedef typename Kernel::DstEvaluatorType DstEvaluatorType; + typedef typename DstEvaluatorType::XprType DstXprType; + typedef typename Kernel::PacketType PacketType; + + enum { + outer = Index / DstXprType::InnerSizeAtCompileTime, + inner = Index % DstXprType::InnerSizeAtCompileTime, + SrcAlignment = Kernel::AssignmentTraits::SrcAlignment, + DstAlignment = Kernel::AssignmentTraits::DstAlignment + }; + + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + kernel.template assignPacketByOuterInner(outer, inner); + enum { NextIndex = Index + unpacket_traits::size }; + copy_using_evaluator_innervec_CompleteUnrolling::run(kernel); + } +}; + +template +struct copy_using_evaluator_innervec_CompleteUnrolling +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&) { } +}; + +template +struct copy_using_evaluator_innervec_InnerUnrolling +{ + typedef typename Kernel::PacketType PacketType; + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, Index outer) + { + kernel.template assignPacketByOuterInner(outer, Index_); + enum { NextIndex = Index_ + unpacket_traits::size }; + copy_using_evaluator_innervec_InnerUnrolling::run(kernel, outer); + } +}; + +template +struct copy_using_evaluator_innervec_InnerUnrolling +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &, Index) { } +}; + +/*************************************************************************** +* Part 3 : implementation of all cases +***************************************************************************/ + +// dense_assignment_loop is based on assign_impl + +template +struct dense_assignment_loop; + +/************************ +***** Special Cases ***** +************************/ + +// Zero-sized assignment is a no-op. +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel& /*kernel*/) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + EIGEN_STATIC_ASSERT(int(DstXprType::SizeAtCompileTime) == 0, + EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT) + } +}; + +/************************ +*** Default traversal *** +************************/ + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static void EIGEN_STRONG_INLINE run(Kernel &kernel) + { + for(Index outer = 0; outer < kernel.outerSize(); ++outer) { + for(Index inner = 0; inner < kernel.innerSize(); ++inner) { + kernel.assignCoeffByOuterInner(outer, inner); + } + } + } +}; + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + copy_using_evaluator_DefaultTraversal_CompleteUnrolling::run(kernel); + } +}; + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + + const Index outerSize = kernel.outerSize(); + for(Index outer = 0; outer < outerSize; ++outer) + copy_using_evaluator_DefaultTraversal_InnerUnrolling::run(kernel, outer); + } +}; + +/*************************** +*** Linear vectorization *** +***************************/ + + +// The goal of unaligned_dense_assignment_loop is simply to factorize the handling +// of the non vectorizable beginning and ending parts + +template +struct unaligned_dense_assignment_loop +{ + // if IsAligned = true, then do nothing + template + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel&, Index, Index) {} +}; + +template <> +struct unaligned_dense_assignment_loop +{ + // MSVC must not inline this functions. If it does, it fails to optimize the + // packet access path. + // FIXME check which version exhibits this issue +#if EIGEN_COMP_MSVC + template + static EIGEN_DONT_INLINE void run(Kernel &kernel, + Index start, + Index end) +#else + template + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel, + Index start, + Index end) +#endif + { + for (Index index = start; index < end; ++index) + kernel.assignCoeff(index); + } +}; + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + const Index size = kernel.size(); + typedef typename Kernel::Scalar Scalar; + typedef typename Kernel::PacketType PacketType; + enum { + requestedAlignment = Kernel::AssignmentTraits::LinearRequiredAlignment, + packetSize = unpacket_traits::size, + dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment), + dstAlignment = packet_traits::AlignedOnScalar ? int(requestedAlignment) + : int(Kernel::AssignmentTraits::DstAlignment), + srcAlignment = Kernel::AssignmentTraits::JointAlignment + }; + const Index alignedStart = dstIsAligned ? 0 : internal::first_aligned(kernel.dstDataPtr(), size); + const Index alignedEnd = alignedStart + ((size-alignedStart)/packetSize)*packetSize; + + unaligned_dense_assignment_loop::run(kernel, 0, alignedStart); + + for(Index index = alignedStart; index < alignedEnd; index += packetSize) + kernel.template assignPacket(index); + + unaligned_dense_assignment_loop<>::run(kernel, alignedEnd, size); + } +}; + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + typedef typename Kernel::PacketType PacketType; + + enum { size = DstXprType::SizeAtCompileTime, + packetSize =unpacket_traits::size, + alignedSize = (int(size)/packetSize)*packetSize }; + + copy_using_evaluator_innervec_CompleteUnrolling::run(kernel); + copy_using_evaluator_DefaultTraversal_CompleteUnrolling::run(kernel); + } +}; + +/************************** +*** Inner vectorization *** +**************************/ + +template +struct dense_assignment_loop +{ + typedef typename Kernel::PacketType PacketType; + enum { + SrcAlignment = Kernel::AssignmentTraits::SrcAlignment, + DstAlignment = Kernel::AssignmentTraits::DstAlignment + }; + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + const Index innerSize = kernel.innerSize(); + const Index outerSize = kernel.outerSize(); + const Index packetSize = unpacket_traits::size; + for(Index outer = 0; outer < outerSize; ++outer) + for(Index inner = 0; inner < innerSize; inner+=packetSize) + kernel.template assignPacketByOuterInner(outer, inner); + } +}; + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + copy_using_evaluator_innervec_CompleteUnrolling::run(kernel); + } +}; + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + typedef typename Kernel::AssignmentTraits Traits; + const Index outerSize = kernel.outerSize(); + for(Index outer = 0; outer < outerSize; ++outer) + copy_using_evaluator_innervec_InnerUnrolling::run(kernel, outer); + } +}; + +/*********************** +*** Linear traversal *** +***********************/ + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + const Index size = kernel.size(); + for(Index i = 0; i < size; ++i) + kernel.assignCoeff(i); + } +}; + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + copy_using_evaluator_LinearTraversal_CompleteUnrolling::run(kernel); + } +}; + +/************************** +*** Slice vectorization *** +***************************/ + +template +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::Scalar Scalar; + typedef typename Kernel::PacketType PacketType; + enum { + packetSize = unpacket_traits::size, + requestedAlignment = int(Kernel::AssignmentTraits::InnerRequiredAlignment), + alignable = packet_traits::AlignedOnScalar || int(Kernel::AssignmentTraits::DstAlignment)>=sizeof(Scalar), + dstIsAligned = int(Kernel::AssignmentTraits::DstAlignment)>=int(requestedAlignment), + dstAlignment = alignable ? int(requestedAlignment) + : int(Kernel::AssignmentTraits::DstAlignment) + }; + const Scalar *dst_ptr = kernel.dstDataPtr(); + if((!bool(dstIsAligned)) && (UIntPtr(dst_ptr) % sizeof(Scalar))>0) + { + // the pointer is not aligned-on scalar, so alignment is not possible + return dense_assignment_loop::run(kernel); + } + const Index packetAlignedMask = packetSize - 1; + const Index innerSize = kernel.innerSize(); + const Index outerSize = kernel.outerSize(); + const Index alignedStep = alignable ? (packetSize - kernel.outerStride() % packetSize) & packetAlignedMask : 0; + Index alignedStart = ((!alignable) || bool(dstIsAligned)) ? 0 : internal::first_aligned(dst_ptr, innerSize); + + for(Index outer = 0; outer < outerSize; ++outer) + { + const Index alignedEnd = alignedStart + ((innerSize-alignedStart) & ~packetAlignedMask); + // do the non-vectorizable part of the assignment + for(Index inner = 0; inner(outer, inner); + + // do the non-vectorizable part of the assignment + for(Index inner = alignedEnd; inner +struct dense_assignment_loop +{ + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void run(Kernel &kernel) + { + typedef typename Kernel::DstEvaluatorType::XprType DstXprType; + typedef typename Kernel::PacketType PacketType; + + enum { innerSize = DstXprType::InnerSizeAtCompileTime, + packetSize =unpacket_traits::size, + vectorizableSize = (int(innerSize) / int(packetSize)) * int(packetSize), + size = DstXprType::SizeAtCompileTime }; + + for(Index outer = 0; outer < kernel.outerSize(); ++outer) + { + copy_using_evaluator_innervec_InnerUnrolling::run(kernel, outer); + copy_using_evaluator_DefaultTraversal_InnerUnrolling::run(kernel, outer); + } + } +}; +#endif + + +/*************************************************************************** +* Part 4 : Generic dense assignment kernel +***************************************************************************/ + +// This class generalize the assignment of a coefficient (or packet) from one dense evaluator +// to another dense writable evaluator. +// It is parametrized by the two evaluators, and the actual assignment functor. +// This abstraction level permits to keep the evaluation loops as simple and as generic as possible. +// One can customize the assignment using this generic dense_assignment_kernel with different +// functors, or by completely overloading it, by-passing a functor. +template +class generic_dense_assignment_kernel +{ +protected: + typedef typename DstEvaluatorTypeT::XprType DstXprType; + typedef typename SrcEvaluatorTypeT::XprType SrcXprType; +public: + + typedef DstEvaluatorTypeT DstEvaluatorType; + typedef SrcEvaluatorTypeT SrcEvaluatorType; + typedef typename DstEvaluatorType::Scalar Scalar; + typedef copy_using_evaluator_traits AssignmentTraits; + typedef typename AssignmentTraits::PacketType PacketType; + + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + generic_dense_assignment_kernel(DstEvaluatorType &dst, const SrcEvaluatorType &src, const Functor &func, DstXprType& dstExpr) + : m_dst(dst), m_src(src), m_functor(func), m_dstExpr(dstExpr) + { + #ifdef EIGEN_DEBUG_ASSIGN + AssignmentTraits::debug(); + #endif + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index size() const EIGEN_NOEXCEPT { return m_dstExpr.size(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index innerSize() const EIGEN_NOEXCEPT { return m_dstExpr.innerSize(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index outerSize() const EIGEN_NOEXCEPT { return m_dstExpr.outerSize(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_dstExpr.rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_dstExpr.cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index outerStride() const EIGEN_NOEXCEPT { return m_dstExpr.outerStride(); } + + EIGEN_DEVICE_FUNC DstEvaluatorType& dstEvaluator() EIGEN_NOEXCEPT { return m_dst; } + EIGEN_DEVICE_FUNC const SrcEvaluatorType& srcEvaluator() const EIGEN_NOEXCEPT { return m_src; } + + /// Assign src(row,col) to dst(row,col) through the assignment functor. + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index row, Index col) + { + m_functor.assignCoeff(m_dst.coeffRef(row,col), m_src.coeff(row,col)); + } + + /// \sa assignCoeff(Index,Index) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeff(Index index) + { + m_functor.assignCoeff(m_dst.coeffRef(index), m_src.coeff(index)); + } + + /// \sa assignCoeff(Index,Index) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignCoeffByOuterInner(Index outer, Index inner) + { + Index row = rowIndexByOuterInner(outer, inner); + Index col = colIndexByOuterInner(outer, inner); + assignCoeff(row, col); + } + + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index row, Index col) + { + m_functor.template assignPacket(&m_dst.coeffRef(row,col), m_src.template packet(row,col)); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacket(Index index) + { + m_functor.template assignPacket(&m_dst.coeffRef(index), m_src.template packet(index)); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void assignPacketByOuterInner(Index outer, Index inner) + { + Index row = rowIndexByOuterInner(outer, inner); + Index col = colIndexByOuterInner(outer, inner); + assignPacket(row, col); + } + + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner) + { + typedef typename DstEvaluatorType::ExpressionTraits Traits; + return int(Traits::RowsAtCompileTime) == 1 ? 0 + : int(Traits::ColsAtCompileTime) == 1 ? inner + : int(DstEvaluatorType::Flags)&RowMajorBit ? outer + : inner; + } + + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner) + { + typedef typename DstEvaluatorType::ExpressionTraits Traits; + return int(Traits::ColsAtCompileTime) == 1 ? 0 + : int(Traits::RowsAtCompileTime) == 1 ? inner + : int(DstEvaluatorType::Flags)&RowMajorBit ? inner + : outer; + } + + EIGEN_DEVICE_FUNC const Scalar* dstDataPtr() const + { + return m_dstExpr.data(); + } + +protected: + DstEvaluatorType& m_dst; + const SrcEvaluatorType& m_src; + const Functor &m_functor; + // TODO find a way to avoid the needs of the original expression + DstXprType& m_dstExpr; +}; + +// Special kernel used when computing small products whose operands have dynamic dimensions. It ensures that the +// PacketSize used is no larger than 4, thereby increasing the chance that vectorized instructions will be used +// when computing the product. + +template +class restricted_packet_dense_assignment_kernel : public generic_dense_assignment_kernel +{ +protected: + typedef generic_dense_assignment_kernel Base; + public: + typedef typename Base::Scalar Scalar; + typedef typename Base::DstXprType DstXprType; + typedef copy_using_evaluator_traits AssignmentTraits; + typedef typename AssignmentTraits::PacketType PacketType; + + EIGEN_DEVICE_FUNC restricted_packet_dense_assignment_kernel(DstEvaluatorTypeT &dst, const SrcEvaluatorTypeT &src, const Functor &func, DstXprType& dstExpr) + : Base(dst, src, func, dstExpr) + { + } + }; + +/*************************************************************************** +* Part 5 : Entry point for dense rectangular assignment +***************************************************************************/ + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const Functor &/*func*/) +{ + EIGEN_ONLY_USED_FOR_DEBUG(dst); + EIGEN_ONLY_USED_FOR_DEBUG(src); + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); +} + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void resize_if_allowed(DstXprType &dst, const SrcXprType& src, const internal::assign_op &/*func*/) +{ + Index dstRows = src.rows(); + Index dstCols = src.cols(); + if(((dst.rows()!=dstRows) || (dst.cols()!=dstCols))) + dst.resize(dstRows, dstCols); + eigen_assert(dst.rows() == dstRows && dst.cols() == dstCols); +} + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src, const Functor &func) +{ + typedef evaluator DstEvaluatorType; + typedef evaluator SrcEvaluatorType; + + SrcEvaluatorType srcEvaluator(src); + + // NOTE To properly handle A = (A*A.transpose())/s with A rectangular, + // we need to resize the destination after the source evaluator has been created. + resize_if_allowed(dst, src, func); + + DstEvaluatorType dstEvaluator(dst); + + typedef generic_dense_assignment_kernel Kernel; + Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived()); + + dense_assignment_loop::run(kernel); +} + +// Specialization for filling the destination with a constant value. +#ifndef EIGEN_GPU_COMPILE_PHASE +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const Eigen::CwiseNullaryOp, DstXprType>& src, const internal::assign_op& func) +{ + resize_if_allowed(dst, src, func); + std::fill_n(dst.data(), dst.size(), src.functor()()); +} +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void call_dense_assignment_loop(DstXprType& dst, const SrcXprType& src) +{ + call_dense_assignment_loop(dst, src, internal::assign_op()); +} + +/*************************************************************************** +* Part 6 : Generic assignment +***************************************************************************/ + +// Based on the respective shapes of the destination and source, +// the class AssignmentKind determine the kind of assignment mechanism. +// AssignmentKind must define a Kind typedef. +template struct AssignmentKind; + +// Assignment kind defined in this file: +struct Dense2Dense {}; +struct EigenBase2EigenBase {}; + +template struct AssignmentKind { typedef EigenBase2EigenBase Kind; }; +template<> struct AssignmentKind { typedef Dense2Dense Kind; }; + +// This is the main assignment class +template< typename DstXprType, typename SrcXprType, typename Functor, + typename Kind = typename AssignmentKind< typename evaluator_traits::Shape , typename evaluator_traits::Shape >::Kind, + typename EnableIf = void> +struct Assignment; + + +// The only purpose of this call_assignment() function is to deal with noalias() / "assume-aliasing" and automatic transposition. +// Indeed, I (Gael) think that this concept of "assume-aliasing" was a mistake, and it makes thing quite complicated. +// So this intermediate function removes everything related to "assume-aliasing" such that Assignment +// does not has to bother about these annoying details. + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment(Dst& dst, const Src& src) +{ + call_assignment(dst, src, internal::assign_op()); +} +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment(const Dst& dst, const Src& src) +{ + call_assignment(dst, src, internal::assign_op()); +} + +// Deal with "assume-aliasing" +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if< evaluator_assume_aliasing::value, void*>::type = 0) +{ + typename plain_matrix_type::type tmp(src); + call_assignment_no_alias(dst, tmp, func); +} + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment(Dst& dst, const Src& src, const Func& func, typename enable_if::value, void*>::type = 0) +{ + call_assignment_no_alias(dst, src, func); +} + +// by-pass "assume-aliasing" +// When there is no aliasing, we require that 'dst' has been properly resized +template class StorageBase, typename Src, typename Func> +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment(NoAlias& dst, const Src& src, const Func& func) +{ + call_assignment_no_alias(dst.expression(), src, func); +} + + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment_no_alias(Dst& dst, const Src& src, const Func& func) +{ + enum { + NeedToTranspose = ( (int(Dst::RowsAtCompileTime) == 1 && int(Src::ColsAtCompileTime) == 1) + || (int(Dst::ColsAtCompileTime) == 1 && int(Src::RowsAtCompileTime) == 1) + ) && int(Dst::SizeAtCompileTime) != 1 + }; + + typedef typename internal::conditional, Dst>::type ActualDstTypeCleaned; + typedef typename internal::conditional, Dst&>::type ActualDstType; + ActualDstType actualDst(dst); + + // TODO check whether this is the right place to perform these checks: + EIGEN_STATIC_ASSERT_LVALUE(Dst) + EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(ActualDstTypeCleaned,Src) + EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename ActualDstTypeCleaned::Scalar,typename Src::Scalar); + + Assignment::run(actualDst, src, func); +} + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_restricted_packet_assignment_no_alias(Dst& dst, const Src& src, const Func& func) +{ + typedef evaluator DstEvaluatorType; + typedef evaluator SrcEvaluatorType; + typedef restricted_packet_dense_assignment_kernel Kernel; + + EIGEN_STATIC_ASSERT_LVALUE(Dst) + EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar); + + SrcEvaluatorType srcEvaluator(src); + resize_if_allowed(dst, src, func); + + DstEvaluatorType dstEvaluator(dst); + Kernel kernel(dstEvaluator, srcEvaluator, func, dst.const_cast_derived()); + + dense_assignment_loop::run(kernel); +} + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment_no_alias(Dst& dst, const Src& src) +{ + call_assignment_no_alias(dst, src, internal::assign_op()); +} + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src, const Func& func) +{ + // TODO check whether this is the right place to perform these checks: + EIGEN_STATIC_ASSERT_LVALUE(Dst) + EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Dst,Src) + EIGEN_CHECK_BINARY_COMPATIBILIY(Func,typename Dst::Scalar,typename Src::Scalar); + + Assignment::run(dst, src, func); +} +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +void call_assignment_no_alias_no_transpose(Dst& dst, const Src& src) +{ + call_assignment_no_alias_no_transpose(dst, src, internal::assign_op()); +} + +// forward declaration +template void check_for_aliasing(const Dst &dst, const Src &src); + +// Generic Dense to Dense assignment +// Note that the last template argument "Weak" is needed to make it possible to perform +// both partial specialization+SFINAE without ambiguous specialization +template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak> +struct Assignment +{ + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const Functor &func) + { +#ifndef EIGEN_NO_DEBUG + internal::check_for_aliasing(dst, src); +#endif + + call_dense_assignment_loop(dst, src, func); + } +}; + +// Generic assignment through evalTo. +// TODO: not sure we have to keep that one, but it helps porting current code to new evaluator mechanism. +// Note that the last template argument "Weak" is needed to make it possible to perform +// both partial specialization+SFINAE without ambiguous specialization +template< typename DstXprType, typename SrcXprType, typename Functor, typename Weak> +struct Assignment +{ + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op &/*func*/) + { + Index dstRows = src.rows(); + Index dstCols = src.cols(); + if((dst.rows()!=dstRows) || (dst.cols()!=dstCols)) + dst.resize(dstRows, dstCols); + + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); + src.evalTo(dst); + } + + // NOTE The following two functions are templated to avoid their instantiation if not needed + // This is needed because some expressions supports evalTo only and/or have 'void' as scalar type. + template + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op &/*func*/) + { + Index dstRows = src.rows(); + Index dstCols = src.cols(); + if((dst.rows()!=dstRows) || (dst.cols()!=dstCols)) + dst.resize(dstRows, dstCols); + + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); + src.addTo(dst); + } + + template + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op &/*func*/) + { + Index dstRows = src.rows(); + Index dstCols = src.cols(); + if((dst.rows()!=dstRows) || (dst.cols()!=dstCols)) + dst.resize(dstRows, dstCols); + + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); + src.subTo(dst); + } +}; + +} // namespace internal + +} // end namespace Eigen + +#endif // EIGEN_ASSIGN_EVALUATOR_H diff --git a/thirdparty/eigen/Eigen/src/Core/Assign_MKL.h b/thirdparty/eigen/Eigen/src/Core/Assign_MKL.h old mode 100644 new mode 100755 index 7772951b..c6140d18 --- a/thirdparty/eigen/Eigen/src/Core/Assign_MKL.h +++ b/thirdparty/eigen/Eigen/src/Core/Assign_MKL.h @@ -1,6 +1,7 @@ /* Copyright (c) 2011, Intel Corporation. All rights reserved. - + Copyright (C) 2015 Gael Guennebaud + Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met: @@ -37,17 +38,13 @@ namespace Eigen { namespace internal { -template struct vml_call -{ enum { IsSupported = 0 }; }; - -template +template class vml_assign_traits { private: enum { DstHasDirectAccess = Dst::Flags & DirectAccessBit, SrcHasDirectAccess = Src::Flags & DirectAccessBit, - StorageOrdersAgree = (int(Dst::IsRowMajor) == int(Src::IsRowMajor)), InnerSize = int(Dst::IsVectorAtCompileTime) ? int(Dst::SizeAtCompileTime) : int(Dst::Flags)&RowMajorBit ? int(Dst::ColsAtCompileTime) @@ -57,165 +54,122 @@ class vml_assign_traits : int(Dst::MaxRowsAtCompileTime), MaxSizeAtCompileTime = Dst::SizeAtCompileTime, - MightEnableVml = vml_call::IsSupported && StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess - && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1, + MightEnableVml = StorageOrdersAgree && DstHasDirectAccess && SrcHasDirectAccess && Src::InnerStrideAtCompileTime==1 && Dst::InnerStrideAtCompileTime==1, MightLinearize = MightEnableVml && (int(Dst::Flags) & int(Src::Flags) & LinearAccessBit), VmlSize = MightLinearize ? MaxSizeAtCompileTime : InnerMaxSize, - LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD, - MayEnableVml = MightEnableVml && LargeEnough, - MayLinearize = MayEnableVml && MightLinearize + LargeEnough = VmlSize==Dynamic || VmlSize>=EIGEN_MKL_VML_THRESHOLD }; public: enum { - Traversal = MayLinearize ? LinearVectorizedTraversal - : MayEnableVml ? InnerVectorizedTraversal - : DefaultTraversal + EnableVml = MightEnableVml && LargeEnough, + Traversal = MightLinearize ? LinearTraversal : DefaultTraversal }; }; -template::Traversal > -struct vml_assign_impl - : assign_impl,Traversal,Unrolling,BuiltIn> -{ -}; - -template -struct vml_assign_impl -{ - typedef typename Derived1::Scalar Scalar; - typedef typename Derived1::Index Index; - static inline void run(Derived1& dst, const CwiseUnaryOp& src) - { - // in case we want to (or have to) skip VML at runtime we can call: - // assign_impl,Traversal,Unrolling,BuiltIn>::run(dst,src); - const Index innerSize = dst.innerSize(); - const Index outerSize = dst.outerSize(); - for(Index outer = 0; outer < outerSize; ++outer) { - const Scalar *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) : - &(src.nestedExpression().coeffRef(0, outer)); - Scalar *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer)); - vml_call::run(src.functor(), innerSize, src_ptr, dst_ptr ); - } - } -}; - -template -struct vml_assign_impl -{ - static inline void run(Derived1& dst, const CwiseUnaryOp& src) - { - // in case we want to (or have to) skip VML at runtime we can call: - // assign_impl,Traversal,Unrolling,BuiltIn>::run(dst,src); - vml_call::run(src.functor(), dst.size(), src.nestedExpression().data(), dst.data() ); - } -}; - -// Macroses - -#define EIGEN_MKL_VML_SPECIALIZE_ASSIGN(TRAVERSAL,UNROLLING) \ - template \ - struct assign_impl, TRAVERSAL, UNROLLING, Specialized> { \ - static inline void run(Derived1 &dst, const Eigen::CwiseUnaryOp &src) { \ - vml_assign_impl::run(dst, src); \ - } \ - }; - -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,NoUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,CompleteUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(DefaultTraversal,InnerUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,NoUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearTraversal,CompleteUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,NoUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,CompleteUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(InnerVectorizedTraversal,InnerUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,CompleteUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(LinearVectorizedTraversal,NoUnrolling) -EIGEN_MKL_VML_SPECIALIZE_ASSIGN(SliceVectorizedTraversal,NoUnrolling) - - +#define EIGEN_PP_EXPAND(ARG) ARG #if !defined (EIGEN_FAST_MATH) || (EIGEN_FAST_MATH != 1) -#define EIGEN_MKL_VML_MODE VML_HA +#define EIGEN_VMLMODE_EXPAND_xLA , VML_HA #else -#define EIGEN_MKL_VML_MODE VML_LA +#define EIGEN_VMLMODE_EXPAND_xLA , VML_LA #endif -#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE) \ - template<> struct vml_call< scalar_##EIGENOP##_op > { \ - enum { IsSupported = 1 }; \ - static inline void run( const scalar_##EIGENOP##_op& /*func*/, \ - int size, const EIGENTYPE* src, EIGENTYPE* dst) { \ - VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst); \ - } \ +#define EIGEN_VMLMODE_EXPAND_x_ + +#define EIGEN_VMLMODE_PREFIX_xLA vm +#define EIGEN_VMLMODE_PREFIX_x_ v +#define EIGEN_VMLMODE_PREFIX(VMLMODE) EIGEN_CAT(EIGEN_VMLMODE_PREFIX_x,VMLMODE) + +#define EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE) \ + template< typename DstXprType, typename SrcXprNested> \ + struct Assignment, SrcXprNested>, assign_op, \ + Dense2Dense, typename enable_if::EnableVml>::type> { \ + typedef CwiseUnaryOp, SrcXprNested> SrcXprType; \ + static void run(DstXprType &dst, const SrcXprType &src, const assign_op &func) { \ + resize_if_allowed(dst, src, func); \ + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \ + if(vml_assign_traits::Traversal==LinearTraversal) { \ + VMLOP(dst.size(), (const VMLTYPE*)src.nestedExpression().data(), \ + (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE) ); \ + } else { \ + const Index outerSize = dst.outerSize(); \ + for(Index outer = 0; outer < outerSize; ++outer) { \ + const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.nestedExpression().coeffRef(outer,0)) : \ + &(src.nestedExpression().coeffRef(0, outer)); \ + EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer)); \ + VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, \ + (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE)); \ + } \ + } \ + } \ + }; \ + + +#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE) \ + EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),s##VMLOP), float, float, VMLMODE) \ + EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),d##VMLOP), double, double, VMLMODE) + +#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE) \ + EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),c##VMLOP), scomplex, MKL_Complex8, VMLMODE) \ + EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, EIGEN_CAT(EIGEN_VMLMODE_PREFIX(VMLMODE),z##VMLOP), dcomplex, MKL_Complex16, VMLMODE) + +#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP, VMLMODE) \ + EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP, VMLMODE) \ + EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(EIGENOP, VMLOP, VMLMODE) + + +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sin, Sin, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(asin, Asin, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sinh, Sinh, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cos, Cos, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(acos, Acos, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(cosh, Cosh, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tan, Tan, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(atan, Atan, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(tanh, Tanh, LA) +// EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs, Abs, _) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(exp, Exp, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log, Ln, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(log10, Log10, LA) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS(sqrt, Sqrt, _) + +EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr, _) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS_CPLX(arg, Arg, _) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(round, Round, _) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(floor, Floor, _) +EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(ceil, Ceil, _) + +#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE, VMLMODE) \ + template< typename DstXprType, typename SrcXprNested, typename Plain> \ + struct Assignment, SrcXprNested, \ + const CwiseNullaryOp,Plain> >, assign_op, \ + Dense2Dense, typename enable_if::EnableVml>::type> { \ + typedef CwiseBinaryOp, SrcXprNested, \ + const CwiseNullaryOp,Plain> > SrcXprType; \ + static void run(DstXprType &dst, const SrcXprType &src, const assign_op &func) { \ + resize_if_allowed(dst, src, func); \ + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); \ + VMLTYPE exponent = reinterpret_cast(src.rhs().functor().m_other); \ + if(vml_assign_traits::Traversal==LinearTraversal) \ + { \ + VMLOP( dst.size(), (const VMLTYPE*)src.lhs().data(), exponent, \ + (VMLTYPE*)dst.data() EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE) ); \ + } else { \ + const Index outerSize = dst.outerSize(); \ + for(Index outer = 0; outer < outerSize; ++outer) { \ + const EIGENTYPE *src_ptr = src.IsRowMajor ? &(src.lhs().coeffRef(outer,0)) : \ + &(src.lhs().coeffRef(0, outer)); \ + EIGENTYPE *dst_ptr = dst.IsRowMajor ? &(dst.coeffRef(outer,0)) : &(dst.coeffRef(0, outer)); \ + VMLOP( dst.innerSize(), (const VMLTYPE*)src_ptr, exponent, \ + (VMLTYPE*)dst_ptr EIGEN_PP_EXPAND(EIGEN_VMLMODE_EXPAND_x##VMLMODE)); \ + } \ + } \ + } \ }; - -#define EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE) \ - template<> struct vml_call< scalar_##EIGENOP##_op > { \ - enum { IsSupported = 1 }; \ - static inline void run( const scalar_##EIGENOP##_op& /*func*/, \ - int size, const EIGENTYPE* src, EIGENTYPE* dst) { \ - MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE; \ - VMLOP(size, (const VMLTYPE*)src, (VMLTYPE*)dst, vmlMode); \ - } \ - }; - -#define EIGEN_MKL_VML_DECLARE_POW_CALL(EIGENOP, VMLOP, EIGENTYPE, VMLTYPE) \ - template<> struct vml_call< scalar_##EIGENOP##_op > { \ - enum { IsSupported = 1 }; \ - static inline void run( const scalar_##EIGENOP##_op& func, \ - int size, const EIGENTYPE* src, EIGENTYPE* dst) { \ - EIGENTYPE exponent = func.m_exponent; \ - MKL_INT64 vmlMode = EIGEN_MKL_VML_MODE; \ - VMLOP(&size, (const VMLTYPE*)src, (const VMLTYPE*)&exponent, \ - (VMLTYPE*)dst, &vmlMode); \ - } \ - }; - -#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vs##VMLOP, float, float) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vd##VMLOP, double, double) - -#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vc##VMLOP, scomplex, MKL_Complex8) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL(EIGENOP, vz##VMLOP, dcomplex, MKL_Complex16) - -#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX(EIGENOP, VMLOP) - - -#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vms##VMLOP, float, float) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmd##VMLOP, double, double) - -#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmc##VMLOP, scomplex, MKL_Complex8) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALL_LA(EIGENOP, vmz##VMLOP, dcomplex, MKL_Complex16) - -#define EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL_LA(EIGENOP, VMLOP) \ - EIGEN_MKL_VML_DECLARE_UNARY_CALLS_COMPLEX_LA(EIGENOP, VMLOP) - - -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sin, Sin) -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(asin, Asin) -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(cos, Cos) -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(acos, Acos) -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(tan, Tan) -//EIGEN_MKL_VML_DECLARE_UNARY_CALLS(abs, Abs) -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(exp, Exp) -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(log, Ln) -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_LA(sqrt, Sqrt) - -EIGEN_MKL_VML_DECLARE_UNARY_CALLS_REAL(square, Sqr) - -// The vm*powx functions are not avaibale in the windows version of MKL. -#ifndef _WIN32 -EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmspowx_, float, float) -EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdpowx_, double, double) -EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcpowx_, scomplex, MKL_Complex8) -EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzpowx_, dcomplex, MKL_Complex16) -#endif + +EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmsPowx, float, float, LA) +EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmdPowx, double, double, LA) +EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmcPowx, scomplex, MKL_Complex8, LA) +EIGEN_MKL_VML_DECLARE_POW_CALL(pow, vmzPowx, dcomplex, MKL_Complex16, LA) } // end namespace internal diff --git a/thirdparty/eigen/Eigen/src/Core/BandMatrix.h b/thirdparty/eigen/Eigen/src/Core/BandMatrix.h index ffd7fe8b..878c0240 100644 --- a/thirdparty/eigen/Eigen/src/Core/BandMatrix.h +++ b/thirdparty/eigen/Eigen/src/Core/BandMatrix.h @@ -10,7 +10,7 @@ #ifndef EIGEN_BANDMATRIX_H #define EIGEN_BANDMATRIX_H -namespace Eigen { +namespace Eigen { namespace internal { @@ -32,7 +32,7 @@ class BandMatrixBase : public EigenBase }; typedef typename internal::traits::Scalar Scalar; typedef Matrix DenseMatrixType; - typedef typename DenseMatrixType::Index Index; + typedef typename DenseMatrixType::StorageIndex StorageIndex; typedef typename internal::traits::CoefficientsType CoefficientsType; typedef EigenBase Base; @@ -45,7 +45,7 @@ class BandMatrixBase : public EigenBase }; public: - + using Base::derived; using Base::rows; using Base::cols; @@ -55,10 +55,10 @@ class BandMatrixBase : public EigenBase /** \returns the number of sub diagonals */ inline Index subs() const { return derived().subs(); } - + /** \returns an expression of the underlying coefficient matrix */ inline const CoefficientsType& coeffs() const { return derived().coeffs(); } - + /** \returns an expression of the underlying coefficient matrix */ inline CoefficientsType& coeffs() { return derived().coeffs(); } @@ -67,7 +67,7 @@ class BandMatrixBase : public EigenBase * \warning the internal storage must be column major. */ inline Block col(Index i) { - EIGEN_STATIC_ASSERT((Options&RowMajor)==0,THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES); + EIGEN_STATIC_ASSERT((int(Options) & int(RowMajor)) == 0, THIS_METHOD_IS_ONLY_FOR_COLUMN_MAJOR_MATRICES); Index start = 0; Index len = coeffs().rows(); if (i<=supers()) @@ -90,7 +90,7 @@ class BandMatrixBase : public EigenBase template struct DiagonalIntReturnType { enum { - ReturnOpposite = (Options&SelfAdjoint) && (((Index)>0 && Supers==0) || ((Index)<0 && Subs==0)), + ReturnOpposite = (int(Options) & int(SelfAdjoint)) && (((Index) > 0 && Supers == 0) || ((Index) < 0 && Subs == 0)), Conjugate = ReturnOpposite && NumTraits::IsComplex, ActualIndex = ReturnOpposite ? -Index : Index, DiagonalSize = (RowsAtCompileTime==Dynamic || ColsAtCompileTime==Dynamic) @@ -130,7 +130,7 @@ class BandMatrixBase : public EigenBase eigen_assert((i<0 && -i<=subs()) || (i>=0 && i<=supers())); return Block(coeffs(), supers()-i, std::max(0,i), 1, diagonalLength(i)); } - + template inline void evalTo(Dest& dst) const { dst.resize(rows(),cols()); @@ -161,15 +161,15 @@ class BandMatrixBase : public EigenBase * * \brief Represents a rectangular matrix with a banded storage * - * \param _Scalar Numeric type, i.e. float, double, int - * \param Rows Number of rows, or \b Dynamic - * \param Cols Number of columns, or \b Dynamic - * \param Supers Number of super diagonal - * \param Subs Number of sub diagonal - * \param _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint - * The former controls \ref TopicStorageOrders "storage order", and defaults to - * column-major. The latter controls whether the matrix represents a selfadjoint - * matrix in which case either Supers of Subs have to be null. + * \tparam _Scalar Numeric type, i.e. float, double, int + * \tparam _Rows Number of rows, or \b Dynamic + * \tparam _Cols Number of columns, or \b Dynamic + * \tparam _Supers Number of super diagonal + * \tparam _Subs Number of sub diagonal + * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of \b #SelfAdjoint + * The former controls \ref TopicStorageOrders "storage order", and defaults to + * column-major. The latter controls whether the matrix represents a selfadjoint + * matrix in which case either Supers of Subs have to be null. * * \sa class TridiagonalMatrix */ @@ -179,7 +179,7 @@ struct traits > { typedef _Scalar Scalar; typedef Dense StorageKind; - typedef DenseIndex Index; + typedef Eigen::Index StorageIndex; enum { CoeffReadCost = NumTraits::ReadCost, RowsAtCompileTime = _Rows, @@ -192,7 +192,7 @@ struct traits > Options = _Options, DataRowsAtCompileTime = ((Supers!=Dynamic) && (Subs!=Dynamic)) ? 1 + Supers + Subs : Dynamic }; - typedef Matrix CoefficientsType; + typedef Matrix CoefficientsType; }; template @@ -201,26 +201,26 @@ class BandMatrix : public BandMatrixBase::Scalar Scalar; - typedef typename internal::traits::Index Index; + typedef typename internal::traits::StorageIndex StorageIndex; typedef typename internal::traits::CoefficientsType CoefficientsType; - inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs) + explicit inline BandMatrix(Index rows=Rows, Index cols=Cols, Index supers=Supers, Index subs=Subs) : m_coeffs(1+supers+subs,cols), m_rows(rows), m_supers(supers), m_subs(subs) { } /** \returns the number of columns */ - inline Index rows() const { return m_rows.value(); } + inline EIGEN_CONSTEXPR Index rows() const { return m_rows.value(); } /** \returns the number of rows */ - inline Index cols() const { return m_coeffs.cols(); } + inline EIGEN_CONSTEXPR Index cols() const { return m_coeffs.cols(); } /** \returns the number of super diagonals */ - inline Index supers() const { return m_supers.value(); } + inline EIGEN_CONSTEXPR Index supers() const { return m_supers.value(); } /** \returns the number of sub diagonals */ - inline Index subs() const { return m_subs.value(); } + inline EIGEN_CONSTEXPR Index subs() const { return m_subs.value(); } inline const CoefficientsType& coeffs() const { return m_coeffs; } inline CoefficientsType& coeffs() { return m_coeffs; } @@ -241,7 +241,7 @@ struct traits::CoeffReadCost, RowsAtCompileTime = _Rows, @@ -264,9 +264,9 @@ class BandMatrixWrapper : public BandMatrixBase::Scalar Scalar; typedef typename internal::traits::CoefficientsType CoefficientsType; - typedef typename internal::traits::Index Index; + typedef typename internal::traits::StorageIndex StorageIndex; - inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs) + explicit inline BandMatrixWrapper(const CoefficientsType& coeffs, Index rows=_Rows, Index cols=_Cols, Index supers=_Supers, Index subs=_Subs) : m_coeffs(coeffs), m_rows(rows), m_supers(supers), m_subs(subs) { @@ -275,16 +275,16 @@ class BandMatrixWrapper : public BandMatrixBase class TridiagonalMatrix : public BandMatrix { typedef BandMatrix Base; - typedef typename Base::Index Index; + typedef typename Base::StorageIndex StorageIndex; public: - TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {} + explicit TridiagonalMatrix(Index size = Size) : Base(size,size,Options&SelfAdjoint?0:1,1) {} inline typename Base::template DiagonalIntReturnType<1>::Type super() { return Base::template diagonal<1>(); } @@ -327,6 +327,25 @@ class TridiagonalMatrix : public BandMatrix +struct evaluator_traits > + : public evaluator_traits_base > +{ + typedef BandShape Shape; +}; + +template +struct evaluator_traits > + : public evaluator_traits_base > +{ + typedef BandShape Shape; +}; + +template<> struct AssignmentKind { typedef EigenBase2EigenBase Kind; }; + } // end namespace internal } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/Block.h b/thirdparty/eigen/Eigen/src/Core/Block.h index 87bedfa4..9d89b60c 100644 --- a/thirdparty/eigen/Eigen/src/Core/Block.h +++ b/thirdparty/eigen/Eigen/src/Core/Block.h @@ -11,39 +11,7 @@ #ifndef EIGEN_BLOCK_H #define EIGEN_BLOCK_H -namespace Eigen { - -/** \class Block - * \ingroup Core_Module - * - * \brief Expression of a fixed-size or dynamic-size block - * - * \param XprType the type of the expression in which we are taking a block - * \param BlockRows the number of rows of the block we are taking at compile time (optional) - * \param BlockCols the number of columns of the block we are taking at compile time (optional) - * - * This class represents an expression of either a fixed-size or dynamic-size block. It is the return - * type of DenseBase::block(Index,Index,Index,Index) and DenseBase::block(Index,Index) and - * most of the time this is the only way it is used. - * - * However, if you want to directly maniputate block expressions, - * for instance if you want to write a function returning such an expression, you - * will need to use this class. - * - * Here is an example illustrating the dynamic case: - * \include class_Block.cpp - * Output: \verbinclude class_Block.out - * - * \note Even though this expression has dynamic size, in the case where \a XprType - * has fixed size, this expression inherits a fixed maximal size which means that evaluating - * it does not cause a dynamic memory allocation. - * - * Here is an example illustrating the fixed-size case: - * \include class_FixedBlock.cpp - * Output: \verbinclude class_FixedBlock.out - * - * \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock - */ +namespace Eigen { namespace internal { template @@ -52,7 +20,7 @@ struct traits > : traits::Scalar Scalar; typedef typename traits::StorageKind StorageKind; typedef typename traits::XprKind XprKind; - typedef typename nested::type XprTypeNested; + typedef typename ref_selector::type XprTypeNested; typedef typename remove_reference::type _XprTypeNested; enum{ MatrixRows = traits::RowsAtCompileTime, @@ -65,6 +33,7 @@ struct traits > : traits::MaxColsAtCompileTime), + XprTypeIsRowMajor = (int(traits::Flags)&RowMajorBit) != 0, IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1 : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0 @@ -77,28 +46,60 @@ struct traits > : traits::ret) : int(inner_stride_at_compile_time::ret), - MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits::size) == 0) - && (InnerStrideAtCompileTime == 1) - ? PacketAccessBit : 0, - MaskAlignedBit = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0, - FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (traits::Flags&LinearAccessBit))) ? LinearAccessBit : 0, + + // FIXME, this traits is rather specialized for dense object and it needs to be cleaned further FlagsLvalueBit = is_lvalue::value ? LvalueBit : 0, FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0, - Flags0 = traits::Flags & ( (HereditaryBits & ~RowMajorBit) | - DirectAccessBit | - MaskPacketAccessBit | - MaskAlignedBit), - Flags = Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit + Flags = (traits::Flags & (DirectAccessBit | (InnerPanel?CompressedAccessBit:0))) | FlagsLvalueBit | FlagsRowMajorBit, + // FIXME DirectAccessBit should not be handled by expressions + // + // Alignment is needed by MapBase's assertions + // We can sefely set it to false here. Internal alignment errors will be detected by an eigen_internal_assert in the respective evaluator + Alignment = 0 }; }; template::ret> class BlockImpl_dense; - + } // end namespace internal template class BlockImpl; +/** \class Block + * \ingroup Core_Module + * + * \brief Expression of a fixed-size or dynamic-size block + * + * \tparam XprType the type of the expression in which we are taking a block + * \tparam BlockRows the number of rows of the block we are taking at compile time (optional) + * \tparam BlockCols the number of columns of the block we are taking at compile time (optional) + * \tparam InnerPanel is true, if the block maps to a set of rows of a row major matrix or + * to set of columns of a column major matrix (optional). The parameter allows to determine + * at compile time whether aligned access is possible on the block expression. + * + * This class represents an expression of either a fixed-size or dynamic-size block. It is the return + * type of DenseBase::block(Index,Index,Index,Index) and DenseBase::block(Index,Index) and + * most of the time this is the only way it is used. + * + * However, if you want to directly maniputate block expressions, + * for instance if you want to write a function returning such an expression, you + * will need to use this class. + * + * Here is an example illustrating the dynamic case: + * \include class_Block.cpp + * Output: \verbinclude class_Block.out + * + * \note Even though this expression has dynamic size, in the case where \a XprType + * has fixed size, this expression inherits a fixed maximal size which means that evaluating + * it does not cause a dynamic memory allocation. + * + * Here is an example illustrating the fixed-size case: + * \include class_FixedBlock.cpp + * Output: \verbinclude class_FixedBlock.out + * + * \sa DenseBase::block(Index,Index,Index,Index), DenseBase::block(Index,Index), class VectorBlock + */ template class Block : public BlockImpl::StorageKind> { @@ -108,10 +109,13 @@ template class typedef Impl Base; EIGEN_GENERIC_PUBLIC_INTERFACE(Block) EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Block) - + + typedef typename internal::remove_all::type NestedExpression; + /** Column or Row constructor */ - inline Block(XprType& xpr, Index i) : Impl(xpr,i) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Block(XprType& xpr, Index i) : Impl(xpr,i) { eigen_assert( (i>=0) && ( ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && i class /** Fixed-size constructor */ - inline Block(XprType& xpr, Index a_startRow, Index a_startCol) - : Impl(xpr, a_startRow, a_startCol) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Block(XprType& xpr, Index startRow, Index startCol) + : Impl(xpr, startRow, startCol) { EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE) - eigen_assert(a_startRow >= 0 && BlockRows >= 1 && a_startRow + BlockRows <= xpr.rows() - && a_startCol >= 0 && BlockCols >= 1 && a_startCol + BlockCols <= xpr.cols()); + eigen_assert(startRow >= 0 && BlockRows >= 0 && startRow + BlockRows <= xpr.rows() + && startCol >= 0 && BlockCols >= 0 && startCol + BlockCols <= xpr.cols()); } /** Dynamic-size constructor */ - inline Block(XprType& xpr, - Index a_startRow, Index a_startCol, + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Block(XprType& xpr, + Index startRow, Index startCol, Index blockRows, Index blockCols) - : Impl(xpr, a_startRow, a_startCol, blockRows, blockCols) + : Impl(xpr, startRow, startCol, blockRows, blockCols) { eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==blockRows) && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==blockCols)); - eigen_assert(a_startRow >= 0 && blockRows >= 0 && a_startRow <= xpr.rows() - blockRows - && a_startCol >= 0 && blockCols >= 0 && a_startCol <= xpr.cols() - blockCols); + eigen_assert(startRow >= 0 && blockRows >= 0 && startRow <= xpr.rows() - blockRows + && startCol >= 0 && blockCols >= 0 && startCol <= xpr.cols() - blockCols); } }; - + // The generic default implementation for dense block simplu forward to the internal::BlockImpl_dense // that must be specialized for direct and non-direct access... template @@ -149,14 +155,15 @@ class BlockImpl : public internal::BlockImpl_dense { typedef internal::BlockImpl_dense Impl; - typedef typename XprType::Index Index; + typedef typename XprType::StorageIndex StorageIndex; public: typedef Impl Base; EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl) - inline BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {} - inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol) : Impl(xpr, a_startRow, a_startCol) {} - inline BlockImpl(XprType& xpr, Index a_startRow, Index a_startCol, Index blockRows, Index blockCols) - : Impl(xpr, a_startRow, a_startCol, blockRows, blockCols) {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index i) : Impl(xpr,i) {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol) : Impl(xpr, startRow, startCol) {} + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE BlockImpl(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols) + : Impl(xpr, startRow, startCol, blockRows, blockCols) {} }; namespace internal { @@ -166,16 +173,18 @@ template >::type { typedef Block BlockType; + typedef typename internal::ref_selector::non_const_type XprTypeNested; public: typedef typename internal::dense_xpr_base::type Base; EIGEN_DENSE_PUBLIC_INTERFACE(BlockType) EIGEN_INHERIT_ASSIGNMENT_OPERATORS(BlockImpl_dense) - class InnerIterator; + // class InnerIterator; // FIXME apparently never used /** Column or Row constructor */ + EIGEN_DEVICE_FUNC inline BlockImpl_dense(XprType& xpr, Index i) : m_xpr(xpr), // It is a row if and only if BlockRows==1 and BlockCols==XprType::ColsAtCompileTime, @@ -190,79 +199,80 @@ template - inline PacketScalar packet(Index rowId, Index colId) const + EIGEN_DEVICE_FUNC inline PacketScalar packet(Index rowId, Index colId) const { - return m_xpr.template packet - (rowId + m_startRow.value(), colId + m_startCol.value()); + return m_xpr.template packet(rowId + m_startRow.value(), colId + m_startCol.value()); } template - inline void writePacket(Index rowId, Index colId, const PacketScalar& val) + EIGEN_DEVICE_FUNC inline void writePacket(Index rowId, Index colId, const PacketScalar& val) { - m_xpr.const_cast_derived().template writePacket - (rowId + m_startRow.value(), colId + m_startCol.value(), val); + m_xpr.template writePacket(rowId + m_startRow.value(), colId + m_startCol.value(), val); } template - inline PacketScalar packet(Index index) const + EIGEN_DEVICE_FUNC inline PacketScalar packet(Index index) const { return m_xpr.template packet (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index), @@ -270,42 +280,48 @@ template - inline void writePacket(Index index, const PacketScalar& val) + EIGEN_DEVICE_FUNC inline void writePacket(Index index, const PacketScalar& val) { - m_xpr.const_cast_derived().template writePacket + m_xpr.template writePacket (m_startRow.value() + (RowsAtCompileTime == 1 ? 0 : index), m_startCol.value() + (RowsAtCompileTime == 1 ? index : 0), val); } #ifdef EIGEN_PARSED_BY_DOXYGEN /** \sa MapBase::data() */ - inline const Scalar* data() const; - inline Index innerStride() const; - inline Index outerStride() const; + EIGEN_DEVICE_FUNC inline const Scalar* data() const; + EIGEN_DEVICE_FUNC inline Index innerStride() const; + EIGEN_DEVICE_FUNC inline Index outerStride() const; #endif - const typename internal::remove_all::type& nestedExpression() const - { - return m_xpr; + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const typename internal::remove_all::type& nestedExpression() const + { + return m_xpr; } - - Index startRow() const - { - return m_startRow.value(); + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + XprType& nestedExpression() { return m_xpr; } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + StorageIndex startRow() const EIGEN_NOEXCEPT + { + return m_startRow.value(); } - - Index startCol() const - { - return m_startCol.value(); + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + StorageIndex startCol() const EIGEN_NOEXCEPT + { + return m_startCol.value(); } protected: - const typename XprType::Nested m_xpr; - const internal::variable_if_dynamic m_startRow; - const internal::variable_if_dynamic m_startCol; - const internal::variable_if_dynamic m_blockRows; - const internal::variable_if_dynamic m_blockCols; + XprTypeNested m_xpr; + const internal::variable_if_dynamic m_startRow; + const internal::variable_if_dynamic m_startCol; + const internal::variable_if_dynamic m_blockRows; + const internal::variable_if_dynamic m_blockCols; }; /** \internal Internal implementation of dense Blocks in the direct access case.*/ @@ -314,6 +330,21 @@ class BlockImpl_dense : public MapBase > { typedef Block BlockType; + typedef typename internal::ref_selector::non_const_type XprTypeNested; + enum { + XprTypeIsRowMajor = (int(traits::Flags)&RowMajorBit) != 0 + }; + + /** \internal Returns base+offset (unless base is null, in which case returns null). + * Adding an offset to nullptr is undefined behavior, so we must avoid it. + */ + template + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE + static Scalar* add_to_nullable_pointer(Scalar* base, Index offset) + { + return base != NULL ? base+offset : NULL; + } + public: typedef MapBase Base; @@ -322,43 +353,57 @@ class BlockImpl_dense /** Column or Row constructor */ - inline BlockImpl_dense(XprType& xpr, Index i) - : Base(internal::const_cast_ptr(&xpr.coeffRef( - (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0, - (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0)), + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + BlockImpl_dense(XprType& xpr, Index i) + : Base((BlockRows == 0 || BlockCols == 0) ? NULL : add_to_nullable_pointer(xpr.data(), + i * ( ((BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) && (!XprTypeIsRowMajor)) + || ((BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) && ( XprTypeIsRowMajor)) ? xpr.innerStride() : xpr.outerStride())), BlockRows==1 ? 1 : xpr.rows(), BlockCols==1 ? 1 : xpr.cols()), - m_xpr(xpr) + m_xpr(xpr), + m_startRow( (BlockRows==1) && (BlockCols==XprType::ColsAtCompileTime) ? i : 0), + m_startCol( (BlockRows==XprType::RowsAtCompileTime) && (BlockCols==1) ? i : 0) { init(); } /** Fixed-size constructor */ - inline BlockImpl_dense(XprType& xpr, Index startRow, Index startCol) - : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol))), m_xpr(xpr) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + BlockImpl_dense(XprType& xpr, Index startRow, Index startCol) + : Base((BlockRows == 0 || BlockCols == 0) ? NULL : add_to_nullable_pointer(xpr.data(), + xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol))), + m_xpr(xpr), m_startRow(startRow), m_startCol(startCol) { init(); } /** Dynamic-size constructor */ - inline BlockImpl_dense(XprType& xpr, + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + BlockImpl_dense(XprType& xpr, Index startRow, Index startCol, Index blockRows, Index blockCols) - : Base(internal::const_cast_ptr(&xpr.coeffRef(startRow,startCol)), blockRows, blockCols), - m_xpr(xpr) + : Base((blockRows == 0 || blockCols == 0) ? NULL : add_to_nullable_pointer(xpr.data(), + xpr.innerStride()*(XprTypeIsRowMajor?startCol:startRow) + xpr.outerStride()*(XprTypeIsRowMajor?startRow:startCol)), + blockRows, blockCols), + m_xpr(xpr), m_startRow(startRow), m_startCol(startCol) { init(); } - const typename internal::remove_all::type& nestedExpression() const - { - return m_xpr; + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const typename internal::remove_all::type& nestedExpression() const EIGEN_NOEXCEPT + { + return m_xpr; } - + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + XprType& nestedExpression() { return m_xpr; } + /** \sa MapBase::innerStride() */ - inline Index innerStride() const + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index innerStride() const EIGEN_NOEXCEPT { return internal::traits::HasSameStorageOrderAsXprType ? m_xpr.innerStride() @@ -366,11 +411,20 @@ class BlockImpl_dense } /** \sa MapBase::outerStride() */ - inline Index outerStride() const + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index outerStride() const EIGEN_NOEXCEPT { - return m_outerStride; + return internal::traits::HasSameStorageOrderAsXprType + ? m_xpr.outerStride() + : m_xpr.innerStride(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + StorageIndex startRow() const EIGEN_NOEXCEPT { return m_startRow.value(); } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + StorageIndex startCol() const EIGEN_NOEXCEPT { return m_startCol.value(); } + #ifndef __SUNPRO_CC // FIXME sunstudio is not friendly with the above friend... // META-FIXME there is no 'friend' keyword around here. Is this obsolete? @@ -379,7 +433,8 @@ class BlockImpl_dense #ifndef EIGEN_PARSED_BY_DOXYGEN /** \internal used by allowAligned() */ - inline BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + BlockImpl_dense(XprType& xpr, const Scalar* data, Index blockRows, Index blockCols) : Base(data, blockRows, blockCols), m_xpr(xpr) { init(); @@ -387,6 +442,7 @@ class BlockImpl_dense #endif protected: + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void init() { m_outerStride = internal::traits::HasSameStorageOrderAsXprType @@ -394,7 +450,9 @@ class BlockImpl_dense : m_xpr.innerStride(); } - typename XprType::Nested m_xpr; + XprTypeNested m_xpr; + const internal::variable_if_dynamic m_startRow; + const internal::variable_if_dynamic m_startCol; Index m_outerStride; }; diff --git a/thirdparty/eigen/Eigen/src/Core/BooleanRedux.h b/thirdparty/eigen/Eigen/src/Core/BooleanRedux.h index be9f48a8..fa4d7c33 100644 --- a/thirdparty/eigen/Eigen/src/Core/BooleanRedux.h +++ b/thirdparty/eigen/Eigen/src/Core/BooleanRedux.h @@ -14,56 +14,58 @@ namespace Eigen { namespace internal { -template +template struct all_unroller { enum { - col = (UnrollCount-1) / Derived::RowsAtCompileTime, - row = (UnrollCount-1) % Derived::RowsAtCompileTime + IsRowMajor = (int(Derived::Flags) & int(RowMajor)), + i = (UnrollCount-1) / InnerSize, + j = (UnrollCount-1) % InnerSize }; - static inline bool run(const Derived &mat) + EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat) { - return all_unroller::run(mat) && mat.coeff(row, col); + return all_unroller::run(mat) && mat.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i); } }; -template -struct all_unroller +template +struct all_unroller { - static inline bool run(const Derived &/*mat*/) { return true; } + EIGEN_DEVICE_FUNC static inline bool run(const Derived &/*mat*/) { return true; } }; -template -struct all_unroller +template +struct all_unroller { - static inline bool run(const Derived &) { return false; } + EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; } }; -template +template struct any_unroller { enum { - col = (UnrollCount-1) / Derived::RowsAtCompileTime, - row = (UnrollCount-1) % Derived::RowsAtCompileTime + IsRowMajor = (int(Derived::Flags) & int(RowMajor)), + i = (UnrollCount-1) / InnerSize, + j = (UnrollCount-1) % InnerSize }; - static inline bool run(const Derived &mat) + EIGEN_DEVICE_FUNC static inline bool run(const Derived &mat) { - return any_unroller::run(mat) || mat.coeff(row, col); + return any_unroller::run(mat) || mat.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i); } }; -template -struct any_unroller +template +struct any_unroller { - static inline bool run(const Derived & /*mat*/) { return false; } + EIGEN_DEVICE_FUNC static inline bool run(const Derived & /*mat*/) { return false; } }; -template -struct any_unroller +template +struct any_unroller { - static inline bool run(const Derived &) { return false; } + EIGEN_DEVICE_FUNC static inline bool run(const Derived &) { return false; } }; } // end namespace internal @@ -76,21 +78,21 @@ struct any_unroller * \sa any(), Cwise::operator<() */ template -inline bool DenseBase::all() const +EIGEN_DEVICE_FUNC inline bool DenseBase::all() const { + typedef internal::evaluator Evaluator; enum { unroll = SizeAtCompileTime != Dynamic - && CoeffReadCost != Dynamic - && NumTraits::AddCost != Dynamic - && SizeAtCompileTime * (CoeffReadCost + NumTraits::AddCost) <= EIGEN_UNROLLING_LIMIT + && SizeAtCompileTime * (int(Evaluator::CoeffReadCost) + int(NumTraits::AddCost)) <= EIGEN_UNROLLING_LIMIT }; + Evaluator evaluator(derived()); if(unroll) - return internal::all_unroller::run(derived()); + return internal::all_unroller::run(evaluator); else { - for(Index j = 0; j < cols(); ++j) - for(Index i = 0; i < rows(); ++i) - if (!coeff(i, j)) return false; + for(Index i = 0; i < derived().outerSize(); ++i) + for(Index j = 0; j < derived().innerSize(); ++j) + if (!evaluator.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i)) return false; return true; } } @@ -100,21 +102,21 @@ inline bool DenseBase::all() const * \sa all() */ template -inline bool DenseBase::any() const +EIGEN_DEVICE_FUNC inline bool DenseBase::any() const { + typedef internal::evaluator Evaluator; enum { unroll = SizeAtCompileTime != Dynamic - && CoeffReadCost != Dynamic - && NumTraits::AddCost != Dynamic - && SizeAtCompileTime * (CoeffReadCost + NumTraits::AddCost) <= EIGEN_UNROLLING_LIMIT + && SizeAtCompileTime * (int(Evaluator::CoeffReadCost) + int(NumTraits::AddCost)) <= EIGEN_UNROLLING_LIMIT }; + Evaluator evaluator(derived()); if(unroll) - return internal::any_unroller::run(derived()); + return internal::any_unroller::run(evaluator); else { - for(Index j = 0; j < cols(); ++j) - for(Index i = 0; i < rows(); ++i) - if (coeff(i, j)) return true; + for(Index i = 0; i < derived().outerSize(); ++i) + for(Index j = 0; j < derived().innerSize(); ++j) + if (evaluator.coeff(IsRowMajor ? i : j, IsRowMajor ? j : i)) return true; return false; } } @@ -124,7 +126,7 @@ inline bool DenseBase::any() const * \sa all(), any() */ template -inline typename DenseBase::Index DenseBase::count() const +EIGEN_DEVICE_FUNC inline Eigen::Index DenseBase::count() const { return derived().template cast().template cast().sum(); } @@ -136,7 +138,11 @@ inline typename DenseBase::Index DenseBase::count() const template inline bool DenseBase::hasNaN() const { +#if EIGEN_COMP_MSVC || (defined __FAST_MATH__) + return derived().array().isNaN().any(); +#else return !((derived().array()==derived().array()).all()); +#endif } /** \returns true if \c *this contains only finite numbers, i.e., no NaN and no +/-INF values. @@ -146,9 +152,13 @@ inline bool DenseBase::hasNaN() const template inline bool DenseBase::allFinite() const { +#if EIGEN_COMP_MSVC || (defined __FAST_MATH__) + return derived().array().isFinite().all(); +#else return !((derived()-derived()).hasNaN()); +#endif } - + } // end namespace Eigen #endif // EIGEN_ALLANDANY_H diff --git a/thirdparty/eigen/Eigen/src/Core/CMakeLists.txt b/thirdparty/eigen/Eigen/src/Core/CMakeLists.txt deleted file mode 100644 index 2346fc2b..00000000 --- a/thirdparty/eigen/Eigen/src/Core/CMakeLists.txt +++ /dev/null @@ -1,10 +0,0 @@ -FILE(GLOB Eigen_Core_SRCS "*.h") - -INSTALL(FILES - ${Eigen_Core_SRCS} - DESTINATION ${INCLUDE_INSTALL_DIR}/Eigen/src/Core COMPONENT Devel - ) - -ADD_SUBDIRECTORY(products) -ADD_SUBDIRECTORY(util) -ADD_SUBDIRECTORY(arch) diff --git a/thirdparty/eigen/Eigen/src/Core/CommaInitializer.h b/thirdparty/eigen/Eigen/src/Core/CommaInitializer.h index 5dd3adea..c0e29c75 100644 --- a/thirdparty/eigen/Eigen/src/Core/CommaInitializer.h +++ b/thirdparty/eigen/Eigen/src/Core/CommaInitializer.h @@ -22,30 +22,36 @@ namespace Eigen { * the return type of MatrixBase::operator<<, and most of the time this is the only * way it is used. * - * \sa \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished() + * \sa \blank \ref MatrixBaseCommaInitRef "MatrixBase::operator<<", CommaInitializer::finished() */ template struct CommaInitializer { typedef typename XprType::Scalar Scalar; - typedef typename XprType::Index Index; + EIGEN_DEVICE_FUNC inline CommaInitializer(XprType& xpr, const Scalar& s) : m_xpr(xpr), m_row(0), m_col(1), m_currentBlockRows(1) { + eigen_assert(m_xpr.rows() > 0 && m_xpr.cols() > 0 + && "Cannot comma-initialize a 0x0 matrix (operator<<)"); m_xpr.coeffRef(0,0) = s; } template + EIGEN_DEVICE_FUNC inline CommaInitializer(XprType& xpr, const DenseBase& other) : m_xpr(xpr), m_row(0), m_col(other.cols()), m_currentBlockRows(other.rows()) { + eigen_assert(m_xpr.rows() >= other.rows() && m_xpr.cols() >= other.cols() + && "Cannot comma-initialize a 0x0 matrix (operator<<)"); m_xpr.block(0, 0, other.rows(), other.cols()) = other; } /* Copy/Move constructor which transfers ownership. This is crucial in * absence of return value optimization to avoid assertions during destruction. */ // FIXME in C++11 mode this could be replaced by a proper RValue constructor + EIGEN_DEVICE_FUNC inline CommaInitializer(const CommaInitializer& o) : m_xpr(o.m_xpr), m_row(o.m_row), m_col(o.m_col), m_currentBlockRows(o.m_currentBlockRows) { // Mark original object as finished. In absence of R-value references we need to const_cast: @@ -55,6 +61,7 @@ struct CommaInitializer } /* inserts a scalar value in the target matrix */ + EIGEN_DEVICE_FUNC CommaInitializer& operator,(const Scalar& s) { if (m_col==m_xpr.cols()) @@ -74,6 +81,7 @@ struct CommaInitializer /* inserts a matrix expression in the target matrix */ template + EIGEN_DEVICE_FUNC CommaInitializer& operator,(const DenseBase& other) { if (m_col==m_xpr.cols() && (other.cols()!=0 || other.rows()!=m_currentBlockRows)) @@ -93,9 +101,13 @@ struct CommaInitializer return *this; } + EIGEN_DEVICE_FUNC inline ~CommaInitializer() +#if defined VERIFY_RAISES_ASSERT && (!defined EIGEN_NO_ASSERTION_CHECKING) && defined EIGEN_EXCEPTIONS + EIGEN_EXCEPTION_SPEC(Eigen::eigen_assert_exception) +#endif { - finished(); + finished(); } /** \returns the built matrix once all its coefficients have been set. @@ -105,6 +117,7 @@ struct CommaInitializer * quaternion.fromRotationMatrix((Matrix3f() << axis0, axis1, axis2).finished()); * \endcode */ + EIGEN_DEVICE_FUNC inline XprType& finished() { eigen_assert(((m_row+m_currentBlockRows) == m_xpr.rows() || m_xpr.cols() == 0) && m_col == m_xpr.cols() @@ -112,7 +125,7 @@ struct CommaInitializer return m_xpr; } - XprType& m_xpr; // target expression + XprType& m_xpr; // target expression Index m_row; // current row id Index m_col; // current col id Index m_currentBlockRows; // current block height @@ -132,7 +145,7 @@ struct CommaInitializer * \sa CommaInitializer::finished(), class CommaInitializer */ template -inline CommaInitializer DenseBase::operator<< (const Scalar& s) +EIGEN_DEVICE_FUNC inline CommaInitializer DenseBase::operator<< (const Scalar& s) { return CommaInitializer(*static_cast(this), s); } @@ -140,7 +153,7 @@ inline CommaInitializer DenseBase::operator<< (const Scalar& s /** \sa operator<<(const Scalar&) */ template template -inline CommaInitializer +EIGEN_DEVICE_FUNC inline CommaInitializer DenseBase::operator<<(const DenseBase& other) { return CommaInitializer(*static_cast(this), other); diff --git a/thirdparty/eigen/Eigen/src/Core/ConditionEstimator.h b/thirdparty/eigen/Eigen/src/Core/ConditionEstimator.h new file mode 100644 index 00000000..51a2e5f1 --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/ConditionEstimator.h @@ -0,0 +1,175 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2016 Rasmus Munk Larsen (rmlarsen@google.com) +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_CONDITIONESTIMATOR_H +#define EIGEN_CONDITIONESTIMATOR_H + +namespace Eigen { + +namespace internal { + +template +struct rcond_compute_sign { + static inline Vector run(const Vector& v) { + const RealVector v_abs = v.cwiseAbs(); + return (v_abs.array() == static_cast(0)) + .select(Vector::Ones(v.size()), v.cwiseQuotient(v_abs)); + } +}; + +// Partial specialization to avoid elementwise division for real vectors. +template +struct rcond_compute_sign { + static inline Vector run(const Vector& v) { + return (v.array() < static_cast(0)) + .select(-Vector::Ones(v.size()), Vector::Ones(v.size())); + } +}; + +/** + * \returns an estimate of ||inv(matrix)||_1 given a decomposition of + * \a matrix that implements .solve() and .adjoint().solve() methods. + * + * This function implements Algorithms 4.1 and 5.1 from + * http://www.maths.manchester.ac.uk/~higham/narep/narep135.pdf + * which also forms the basis for the condition number estimators in + * LAPACK. Since at most 10 calls to the solve method of dec are + * performed, the total cost is O(dims^2), as opposed to O(dims^3) + * needed to compute the inverse matrix explicitly. + * + * The most common usage is in estimating the condition number + * ||matrix||_1 * ||inv(matrix)||_1. The first term ||matrix||_1 can be + * computed directly in O(n^2) operations. + * + * Supports the following decompositions: FullPivLU, PartialPivLU, LDLT, and + * LLT. + * + * \sa FullPivLU, PartialPivLU, LDLT, LLT. + */ +template +typename Decomposition::RealScalar rcond_invmatrix_L1_norm_estimate(const Decomposition& dec) +{ + typedef typename Decomposition::MatrixType MatrixType; + typedef typename Decomposition::Scalar Scalar; + typedef typename Decomposition::RealScalar RealScalar; + typedef typename internal::plain_col_type::type Vector; + typedef typename internal::plain_col_type::type RealVector; + const bool is_complex = (NumTraits::IsComplex != 0); + + eigen_assert(dec.rows() == dec.cols()); + const Index n = dec.rows(); + if (n == 0) + return 0; + + // Disable Index to float conversion warning +#ifdef __INTEL_COMPILER + #pragma warning push + #pragma warning ( disable : 2259 ) +#endif + Vector v = dec.solve(Vector::Ones(n) / Scalar(n)); +#ifdef __INTEL_COMPILER + #pragma warning pop +#endif + + // lower_bound is a lower bound on + // ||inv(matrix)||_1 = sup_v ||inv(matrix) v||_1 / ||v||_1 + // and is the objective maximized by the ("super-") gradient ascent + // algorithm below. + RealScalar lower_bound = v.template lpNorm<1>(); + if (n == 1) + return lower_bound; + + // Gradient ascent algorithm follows: We know that the optimum is achieved at + // one of the simplices v = e_i, so in each iteration we follow a + // super-gradient to move towards the optimal one. + RealScalar old_lower_bound = lower_bound; + Vector sign_vector(n); + Vector old_sign_vector; + Index v_max_abs_index = -1; + Index old_v_max_abs_index = v_max_abs_index; + for (int k = 0; k < 4; ++k) + { + sign_vector = internal::rcond_compute_sign::run(v); + if (k > 0 && !is_complex && sign_vector == old_sign_vector) { + // Break if the solution stagnated. + break; + } + // v_max_abs_index = argmax |real( inv(matrix)^T * sign_vector )| + v = dec.adjoint().solve(sign_vector); + v.real().cwiseAbs().maxCoeff(&v_max_abs_index); + if (v_max_abs_index == old_v_max_abs_index) { + // Break if the solution stagnated. + break; + } + // Move to the new simplex e_j, where j = v_max_abs_index. + v = dec.solve(Vector::Unit(n, v_max_abs_index)); // v = inv(matrix) * e_j. + lower_bound = v.template lpNorm<1>(); + if (lower_bound <= old_lower_bound) { + // Break if the gradient step did not increase the lower_bound. + break; + } + if (!is_complex) { + old_sign_vector = sign_vector; + } + old_v_max_abs_index = v_max_abs_index; + old_lower_bound = lower_bound; + } + // The following calculates an independent estimate of ||matrix||_1 by + // multiplying matrix by a vector with entries of slowly increasing + // magnitude and alternating sign: + // v_i = (-1)^{i} (1 + (i / (dim-1))), i = 0,...,dim-1. + // This improvement to Hager's algorithm above is due to Higham. It was + // added to make the algorithm more robust in certain corner cases where + // large elements in the matrix might otherwise escape detection due to + // exact cancellation (especially when op and op_adjoint correspond to a + // sequence of backsubstitutions and permutations), which could cause + // Hager's algorithm to vastly underestimate ||matrix||_1. + Scalar alternating_sign(RealScalar(1)); + for (Index i = 0; i < n; ++i) { + // The static_cast is needed when Scalar is a complex and RealScalar implements expression templates + v[i] = alternating_sign * static_cast(RealScalar(1) + (RealScalar(i) / (RealScalar(n - 1)))); + alternating_sign = -alternating_sign; + } + v = dec.solve(v); + const RealScalar alternate_lower_bound = (2 * v.template lpNorm<1>()) / (3 * RealScalar(n)); + return numext::maxi(lower_bound, alternate_lower_bound); +} + +/** \brief Reciprocal condition number estimator. + * + * Computing a decomposition of a dense matrix takes O(n^3) operations, while + * this method estimates the condition number quickly and reliably in O(n^2) + * operations. + * + * \returns an estimate of the reciprocal condition number + * (1 / (||matrix||_1 * ||inv(matrix)||_1)) of matrix, given ||matrix||_1 and + * its decomposition. Supports the following decompositions: FullPivLU, + * PartialPivLU, LDLT, and LLT. + * + * \sa FullPivLU, PartialPivLU, LDLT, LLT. + */ +template +typename Decomposition::RealScalar +rcond_estimate_helper(typename Decomposition::RealScalar matrix_norm, const Decomposition& dec) +{ + typedef typename Decomposition::RealScalar RealScalar; + eigen_assert(dec.rows() == dec.cols()); + if (dec.rows() == 0) return NumTraits::infinity(); + if (matrix_norm == RealScalar(0)) return RealScalar(0); + if (dec.rows() == 1) return RealScalar(1); + const RealScalar inverse_matrix_norm = rcond_invmatrix_L1_norm_estimate(dec); + return (inverse_matrix_norm == RealScalar(0) ? RealScalar(0) + : (RealScalar(1) / inverse_matrix_norm) / matrix_norm); +} + +} // namespace internal + +} // namespace Eigen + +#endif diff --git a/thirdparty/eigen/Eigen/src/Core/CoreEvaluators.h b/thirdparty/eigen/Eigen/src/Core/CoreEvaluators.h new file mode 100644 index 00000000..0ff8c8de --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/CoreEvaluators.h @@ -0,0 +1,1741 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2011 Benoit Jacob +// Copyright (C) 2011-2014 Gael Guennebaud +// Copyright (C) 2011-2012 Jitse Niesen +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + + +#ifndef EIGEN_COREEVALUATORS_H +#define EIGEN_COREEVALUATORS_H + +namespace Eigen { + +namespace internal { + +// This class returns the evaluator kind from the expression storage kind. +// Default assumes index based accessors +template +struct storage_kind_to_evaluator_kind { + typedef IndexBased Kind; +}; + +// This class returns the evaluator shape from the expression storage kind. +// It can be Dense, Sparse, Triangular, Diagonal, SelfAdjoint, Band, etc. +template struct storage_kind_to_shape; + +template<> struct storage_kind_to_shape { typedef DenseShape Shape; }; +template<> struct storage_kind_to_shape { typedef SolverShape Shape; }; +template<> struct storage_kind_to_shape { typedef PermutationShape Shape; }; +template<> struct storage_kind_to_shape { typedef TranspositionsShape Shape; }; + +// Evaluators have to be specialized with respect to various criteria such as: +// - storage/structure/shape +// - scalar type +// - etc. +// Therefore, we need specialization of evaluator providing additional template arguments for each kind of evaluators. +// We currently distinguish the following kind of evaluators: +// - unary_evaluator for expressions taking only one arguments (CwiseUnaryOp, CwiseUnaryView, Transpose, MatrixWrapper, ArrayWrapper, Reverse, Replicate) +// - binary_evaluator for expression taking two arguments (CwiseBinaryOp) +// - ternary_evaluator for expression taking three arguments (CwiseTernaryOp) +// - product_evaluator for linear algebra products (Product); special case of binary_evaluator because it requires additional tags for dispatching. +// - mapbase_evaluator for Map, Block, Ref +// - block_evaluator for Block (special dispatching to a mapbase_evaluator or unary_evaluator) + +template< typename T, + typename Arg1Kind = typename evaluator_traits::Kind, + typename Arg2Kind = typename evaluator_traits::Kind, + typename Arg3Kind = typename evaluator_traits::Kind, + typename Arg1Scalar = typename traits::Scalar, + typename Arg2Scalar = typename traits::Scalar, + typename Arg3Scalar = typename traits::Scalar> struct ternary_evaluator; + +template< typename T, + typename LhsKind = typename evaluator_traits::Kind, + typename RhsKind = typename evaluator_traits::Kind, + typename LhsScalar = typename traits::Scalar, + typename RhsScalar = typename traits::Scalar> struct binary_evaluator; + +template< typename T, + typename Kind = typename evaluator_traits::Kind, + typename Scalar = typename T::Scalar> struct unary_evaluator; + +// evaluator_traits contains traits for evaluator + +template +struct evaluator_traits_base +{ + // by default, get evaluator kind and shape from storage + typedef typename storage_kind_to_evaluator_kind::StorageKind>::Kind Kind; + typedef typename storage_kind_to_shape::StorageKind>::Shape Shape; +}; + +// Default evaluator traits +template +struct evaluator_traits : public evaluator_traits_base +{ +}; + +template::Shape > +struct evaluator_assume_aliasing { + static const bool value = false; +}; + +// By default, we assume a unary expression: +template +struct evaluator : public unary_evaluator +{ + typedef unary_evaluator Base; + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const T& xpr) : Base(xpr) {} +}; + + +// TODO: Think about const-correctness +template +struct evaluator + : evaluator +{ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const T& xpr) : evaluator(xpr) {} +}; + +// ---------- base class for all evaluators ---------- + +template +struct evaluator_base +{ + // TODO that's not very nice to have to propagate all these traits. They are currently only needed to handle outer,inner indices. + typedef traits ExpressionTraits; + + enum { + Alignment = 0 + }; + // noncopyable: + // Don't make this class inherit noncopyable as this kills EBO (Empty Base Optimization) + // and make complex evaluator much larger than then should do. + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE evaluator_base() {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ~evaluator_base() {} +private: + EIGEN_DEVICE_FUNC evaluator_base(const evaluator_base&); + EIGEN_DEVICE_FUNC const evaluator_base& operator=(const evaluator_base&); +}; + +// -------------------- Matrix and Array -------------------- +// +// evaluator is a common base class for the +// Matrix and Array evaluators. +// Here we directly specialize evaluator. This is not really a unary expression, and it is, by definition, dense, +// so no need for more sophisticated dispatching. + +// this helper permits to completely eliminate m_outerStride if it is known at compiletime. +template class plainobjectbase_evaluator_data { +public: + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + plainobjectbase_evaluator_data(const Scalar* ptr, Index outerStride) : data(ptr) + { +#ifndef EIGEN_INTERNAL_DEBUGGING + EIGEN_UNUSED_VARIABLE(outerStride); +#endif + eigen_internal_assert(outerStride==OuterStride); + } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index outerStride() const EIGEN_NOEXCEPT { return OuterStride; } + const Scalar *data; +}; + +template class plainobjectbase_evaluator_data { +public: + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + plainobjectbase_evaluator_data(const Scalar* ptr, Index outerStride) : data(ptr), m_outerStride(outerStride) {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Index outerStride() const { return m_outerStride; } + const Scalar *data; +protected: + Index m_outerStride; +}; + +template +struct evaluator > + : evaluator_base +{ + typedef PlainObjectBase PlainObjectType; + typedef typename PlainObjectType::Scalar Scalar; + typedef typename PlainObjectType::CoeffReturnType CoeffReturnType; + + enum { + IsRowMajor = PlainObjectType::IsRowMajor, + IsVectorAtCompileTime = PlainObjectType::IsVectorAtCompileTime, + RowsAtCompileTime = PlainObjectType::RowsAtCompileTime, + ColsAtCompileTime = PlainObjectType::ColsAtCompileTime, + + CoeffReadCost = NumTraits::ReadCost, + Flags = traits::EvaluatorFlags, + Alignment = traits::Alignment + }; + enum { + // We do not need to know the outer stride for vectors + OuterStrideAtCompileTime = IsVectorAtCompileTime ? 0 + : int(IsRowMajor) ? ColsAtCompileTime + : RowsAtCompileTime + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + evaluator() + : m_d(0,OuterStrideAtCompileTime) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const PlainObjectType& m) + : m_d(m.data(),IsVectorAtCompileTime ? 0 : m.outerStride()) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + if (IsRowMajor) + return m_d.data[row * m_d.outerStride() + col]; + else + return m_d.data[row + col * m_d.outerStride()]; + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_d.data[index]; + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + if (IsRowMajor) + return const_cast(m_d.data)[row * m_d.outerStride() + col]; + else + return const_cast(m_d.data)[row + col * m_d.outerStride()]; + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + return const_cast(m_d.data)[index]; + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + if (IsRowMajor) + return ploadt(m_d.data + row * m_d.outerStride() + col); + else + return ploadt(m_d.data + row + col * m_d.outerStride()); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + return ploadt(m_d.data + index); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index row, Index col, const PacketType& x) + { + if (IsRowMajor) + return pstoret + (const_cast(m_d.data) + row * m_d.outerStride() + col, x); + else + return pstoret + (const_cast(m_d.data) + row + col * m_d.outerStride(), x); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index index, const PacketType& x) + { + return pstoret(const_cast(m_d.data) + index, x); + } + +protected: + + plainobjectbase_evaluator_data m_d; +}; + +template +struct evaluator > + : evaluator > > +{ + typedef Matrix XprType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + evaluator() {} + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const XprType& m) + : evaluator >(m) + { } +}; + +template +struct evaluator > + : evaluator > > +{ + typedef Array XprType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + evaluator() {} + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const XprType& m) + : evaluator >(m) + { } +}; + +// -------------------- Transpose -------------------- + +template +struct unary_evaluator, IndexBased> + : evaluator_base > +{ + typedef Transpose XprType; + + enum { + CoeffReadCost = evaluator::CoeffReadCost, + Flags = evaluator::Flags ^ RowMajorBit, + Alignment = evaluator::Alignment + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit unary_evaluator(const XprType& t) : m_argImpl(t.nestedExpression()) {} + + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_argImpl.coeff(col, row); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_argImpl.coeff(index); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + return m_argImpl.coeffRef(col, row); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + typename XprType::Scalar& coeffRef(Index index) + { + return m_argImpl.coeffRef(index); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + return m_argImpl.template packet(col, row); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + return m_argImpl.template packet(index); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index row, Index col, const PacketType& x) + { + m_argImpl.template writePacket(col, row, x); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index index, const PacketType& x) + { + m_argImpl.template writePacket(index, x); + } + +protected: + evaluator m_argImpl; +}; + +// -------------------- CwiseNullaryOp -------------------- +// Like Matrix and Array, this is not really a unary expression, so we directly specialize evaluator. +// Likewise, there is not need to more sophisticated dispatching here. + +template::value, + bool has_unary = has_unary_operator::value, + bool has_binary = has_binary_operator::value> +struct nullary_wrapper +{ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const { return op(i,j); } + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); } + + template EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const { return op.template packetOp(i,j); } + template EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp(i); } +}; + +template +struct nullary_wrapper +{ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType=0, IndexType=0) const { return op(); } + template EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType=0, IndexType=0) const { return op.template packetOp(); } +}; + +template +struct nullary_wrapper +{ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j=0) const { return op(i,j); } + template EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j=0) const { return op.template packetOp(i,j); } +}; + +// We need the following specialization for vector-only functors assigned to a runtime vector, +// for instance, using linspace and assigning a RowVectorXd to a MatrixXd or even a row of a MatrixXd. +// In this case, i==0 and j is used for the actual iteration. +template +struct nullary_wrapper +{ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const { + eigen_assert(i==0 || j==0); + return op(i+j); + } + template EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const { + eigen_assert(i==0 || j==0); + return op.template packetOp(i+j); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { return op(i); } + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { return op.template packetOp(i); } +}; + +template +struct nullary_wrapper {}; + +#if 0 && EIGEN_COMP_MSVC>0 +// Disable this ugly workaround. This is now handled in traits::match, +// but this piece of code might still become handly if some other weird compilation +// erros pop up again. + +// MSVC exhibits a weird compilation error when +// compiling: +// Eigen::MatrixXf A = MatrixXf::Random(3,3); +// Ref R = 2.f*A; +// and that has_*ary_operator> have not been instantiated yet. +// The "problem" is that evaluator<2.f*A> is instantiated by traits::match<2.f*A> +// and at that time has_*ary_operator returns true regardless of T. +// Then nullary_wrapper is badly instantiated as nullary_wrapper<.,.,true,true,true>. +// The trick is thus to defer the proper instantiation of nullary_wrapper when coeff(), +// and packet() are really instantiated as implemented below: + +// This is a simple wrapper around Index to enforce the re-instantiation of +// has_*ary_operator when needed. +template struct nullary_wrapper_workaround_msvc { + nullary_wrapper_workaround_msvc(const T&); + operator T()const; +}; + +template +struct nullary_wrapper +{ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i, IndexType j) const { + return nullary_wrapper >::value, + has_unary_operator >::value, + has_binary_operator >::value>().operator()(op,i,j); + } + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar operator()(const NullaryOp& op, IndexType i) const { + return nullary_wrapper >::value, + has_unary_operator >::value, + has_binary_operator >::value>().operator()(op,i); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i, IndexType j) const { + return nullary_wrapper >::value, + has_unary_operator >::value, + has_binary_operator >::value>().template packetOp(op,i,j); + } + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE T packetOp(const NullaryOp& op, IndexType i) const { + return nullary_wrapper >::value, + has_unary_operator >::value, + has_binary_operator >::value>().template packetOp(op,i); + } +}; +#endif // MSVC workaround + +template +struct evaluator > + : evaluator_base > +{ + typedef CwiseNullaryOp XprType; + typedef typename internal::remove_all::type PlainObjectTypeCleaned; + + enum { + CoeffReadCost = internal::functor_traits::Cost, + + Flags = (evaluator::Flags + & ( HereditaryBits + | (functor_has_linear_access::ret ? LinearAccessBit : 0) + | (functor_traits::PacketAccess ? PacketAccessBit : 0))) + | (functor_traits::IsRepeatable ? 0 : EvalBeforeNestingBit), + Alignment = AlignedMax + }; + + EIGEN_DEVICE_FUNC explicit evaluator(const XprType& n) + : m_functor(n.functor()), m_wrapper() + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::CoeffReturnType CoeffReturnType; + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(IndexType row, IndexType col) const + { + return m_wrapper(m_functor, row, col); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(IndexType index) const + { + return m_wrapper(m_functor,index); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(IndexType row, IndexType col) const + { + return m_wrapper.template packetOp(m_functor, row, col); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(IndexType index) const + { + return m_wrapper.template packetOp(m_functor, index); + } + +protected: + const NullaryOp m_functor; + const internal::nullary_wrapper m_wrapper; +}; + +// -------------------- CwiseUnaryOp -------------------- + +template +struct unary_evaluator, IndexBased > + : evaluator_base > +{ + typedef CwiseUnaryOp XprType; + + enum { + CoeffReadCost = int(evaluator::CoeffReadCost) + int(functor_traits::Cost), + + Flags = evaluator::Flags + & (HereditaryBits | LinearAccessBit | (functor_traits::PacketAccess ? PacketAccessBit : 0)), + Alignment = evaluator::Alignment + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit unary_evaluator(const XprType& op) : m_d(op) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits::Cost); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_d.func()(m_d.argImpl.coeff(row, col)); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_d.func()(m_d.argImpl.coeff(index)); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + return m_d.func().packetOp(m_d.argImpl.template packet(row, col)); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + return m_d.func().packetOp(m_d.argImpl.template packet(index)); + } + +protected: + + // this helper permits to completely eliminate the functor if it is empty + struct Data + { + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Data(const XprType& xpr) : op(xpr.functor()), argImpl(xpr.nestedExpression()) {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const UnaryOp& func() const { return op; } + UnaryOp op; + evaluator argImpl; + }; + + Data m_d; +}; + +// -------------------- CwiseTernaryOp -------------------- + +// this is a ternary expression +template +struct evaluator > + : public ternary_evaluator > +{ + typedef CwiseTernaryOp XprType; + typedef ternary_evaluator > Base; + + EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : Base(xpr) {} +}; + +template +struct ternary_evaluator, IndexBased, IndexBased> + : evaluator_base > +{ + typedef CwiseTernaryOp XprType; + + enum { + CoeffReadCost = int(evaluator::CoeffReadCost) + int(evaluator::CoeffReadCost) + int(evaluator::CoeffReadCost) + int(functor_traits::Cost), + + Arg1Flags = evaluator::Flags, + Arg2Flags = evaluator::Flags, + Arg3Flags = evaluator::Flags, + SameType = is_same::value && is_same::value, + StorageOrdersAgree = (int(Arg1Flags)&RowMajorBit)==(int(Arg2Flags)&RowMajorBit) && (int(Arg1Flags)&RowMajorBit)==(int(Arg3Flags)&RowMajorBit), + Flags0 = (int(Arg1Flags) | int(Arg2Flags) | int(Arg3Flags)) & ( + HereditaryBits + | (int(Arg1Flags) & int(Arg2Flags) & int(Arg3Flags) & + ( (StorageOrdersAgree ? LinearAccessBit : 0) + | (functor_traits::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0) + ) + ) + ), + Flags = (Flags0 & ~RowMajorBit) | (Arg1Flags & RowMajorBit), + Alignment = EIGEN_PLAIN_ENUM_MIN( + EIGEN_PLAIN_ENUM_MIN(evaluator::Alignment, evaluator::Alignment), + evaluator::Alignment) + }; + + EIGEN_DEVICE_FUNC explicit ternary_evaluator(const XprType& xpr) : m_d(xpr) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits::Cost); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_d.func()(m_d.arg1Impl.coeff(row, col), m_d.arg2Impl.coeff(row, col), m_d.arg3Impl.coeff(row, col)); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_d.func()(m_d.arg1Impl.coeff(index), m_d.arg2Impl.coeff(index), m_d.arg3Impl.coeff(index)); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + return m_d.func().packetOp(m_d.arg1Impl.template packet(row, col), + m_d.arg2Impl.template packet(row, col), + m_d.arg3Impl.template packet(row, col)); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + return m_d.func().packetOp(m_d.arg1Impl.template packet(index), + m_d.arg2Impl.template packet(index), + m_d.arg3Impl.template packet(index)); + } + +protected: + // this helper permits to completely eliminate the functor if it is empty + struct Data + { + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Data(const XprType& xpr) : op(xpr.functor()), arg1Impl(xpr.arg1()), arg2Impl(xpr.arg2()), arg3Impl(xpr.arg3()) {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const TernaryOp& func() const { return op; } + TernaryOp op; + evaluator arg1Impl; + evaluator arg2Impl; + evaluator arg3Impl; + }; + + Data m_d; +}; + +// -------------------- CwiseBinaryOp -------------------- + +// this is a binary expression +template +struct evaluator > + : public binary_evaluator > +{ + typedef CwiseBinaryOp XprType; + typedef binary_evaluator > Base; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const XprType& xpr) : Base(xpr) {} +}; + +template +struct binary_evaluator, IndexBased, IndexBased> + : evaluator_base > +{ + typedef CwiseBinaryOp XprType; + + enum { + CoeffReadCost = int(evaluator::CoeffReadCost) + int(evaluator::CoeffReadCost) + int(functor_traits::Cost), + + LhsFlags = evaluator::Flags, + RhsFlags = evaluator::Flags, + SameType = is_same::value, + StorageOrdersAgree = (int(LhsFlags)&RowMajorBit)==(int(RhsFlags)&RowMajorBit), + Flags0 = (int(LhsFlags) | int(RhsFlags)) & ( + HereditaryBits + | (int(LhsFlags) & int(RhsFlags) & + ( (StorageOrdersAgree ? LinearAccessBit : 0) + | (functor_traits::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0) + ) + ) + ), + Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit), + Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator::Alignment,evaluator::Alignment) + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit binary_evaluator(const XprType& xpr) : m_d(xpr) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits::Cost); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_d.func()(m_d.lhsImpl.coeff(row, col), m_d.rhsImpl.coeff(row, col)); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_d.func()(m_d.lhsImpl.coeff(index), m_d.rhsImpl.coeff(index)); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + return m_d.func().packetOp(m_d.lhsImpl.template packet(row, col), + m_d.rhsImpl.template packet(row, col)); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + return m_d.func().packetOp(m_d.lhsImpl.template packet(index), + m_d.rhsImpl.template packet(index)); + } + +protected: + + // this helper permits to completely eliminate the functor if it is empty + struct Data + { + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Data(const XprType& xpr) : op(xpr.functor()), lhsImpl(xpr.lhs()), rhsImpl(xpr.rhs()) {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const BinaryOp& func() const { return op; } + BinaryOp op; + evaluator lhsImpl; + evaluator rhsImpl; + }; + + Data m_d; +}; + +// -------------------- CwiseUnaryView -------------------- + +template +struct unary_evaluator, IndexBased> + : evaluator_base > +{ + typedef CwiseUnaryView XprType; + + enum { + CoeffReadCost = int(evaluator::CoeffReadCost) + int(functor_traits::Cost), + + Flags = (evaluator::Flags & (HereditaryBits | LinearAccessBit | DirectAccessBit)), + + Alignment = 0 // FIXME it is not very clear why alignment is necessarily lost... + }; + + EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& op) : m_d(op) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(functor_traits::Cost); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_d.func()(m_d.argImpl.coeff(row, col)); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_d.func()(m_d.argImpl.coeff(index)); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + return m_d.func()(m_d.argImpl.coeffRef(row, col)); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + return m_d.func()(m_d.argImpl.coeffRef(index)); + } + +protected: + + // this helper permits to completely eliminate the functor if it is empty + struct Data + { + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Data(const XprType& xpr) : op(xpr.functor()), argImpl(xpr.nestedExpression()) {} + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const UnaryOp& func() const { return op; } + UnaryOp op; + evaluator argImpl; + }; + + Data m_d; +}; + +// -------------------- Map -------------------- + +// FIXME perhaps the PlainObjectType could be provided by Derived::PlainObject ? +// but that might complicate template specialization +template +struct mapbase_evaluator; + +template +struct mapbase_evaluator : evaluator_base +{ + typedef Derived XprType; + typedef typename XprType::PointerType PointerType; + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + enum { + IsRowMajor = XprType::RowsAtCompileTime, + ColsAtCompileTime = XprType::ColsAtCompileTime, + CoeffReadCost = NumTraits::ReadCost + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit mapbase_evaluator(const XprType& map) + : m_data(const_cast(map.data())), + m_innerStride(map.innerStride()), + m_outerStride(map.outerStride()) + { + EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(evaluator::Flags&PacketAccessBit, internal::inner_stride_at_compile_time::ret==1), + PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_data[col * colStride() + row * rowStride()]; + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_data[index * m_innerStride.value()]; + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + return m_data[col * colStride() + row * rowStride()]; + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + return m_data[index * m_innerStride.value()]; + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + PointerType ptr = m_data + row * rowStride() + col * colStride(); + return internal::ploadt(ptr); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + return internal::ploadt(m_data + index * m_innerStride.value()); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index row, Index col, const PacketType& x) + { + PointerType ptr = m_data + row * rowStride() + col * colStride(); + return internal::pstoret(ptr, x); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index index, const PacketType& x) + { + internal::pstoret(m_data + index * m_innerStride.value(), x); + } +protected: + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rowStride() const EIGEN_NOEXCEPT { + return XprType::IsRowMajor ? m_outerStride.value() : m_innerStride.value(); + } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index colStride() const EIGEN_NOEXCEPT { + return XprType::IsRowMajor ? m_innerStride.value() : m_outerStride.value(); + } + + PointerType m_data; + const internal::variable_if_dynamic m_innerStride; + const internal::variable_if_dynamic m_outerStride; +}; + +template +struct evaluator > + : public mapbase_evaluator, PlainObjectType> +{ + typedef Map XprType; + typedef typename XprType::Scalar Scalar; + // TODO: should check for smaller packet types once we can handle multi-sized packet types + typedef typename packet_traits::type PacketScalar; + + enum { + InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0 + ? int(PlainObjectType::InnerStrideAtCompileTime) + : int(StrideType::InnerStrideAtCompileTime), + OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0 + ? int(PlainObjectType::OuterStrideAtCompileTime) + : int(StrideType::OuterStrideAtCompileTime), + HasNoInnerStride = InnerStrideAtCompileTime == 1, + HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0, + HasNoStride = HasNoInnerStride && HasNoOuterStride, + IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic, + + PacketAccessMask = bool(HasNoInnerStride) ? ~int(0) : ~int(PacketAccessBit), + LinearAccessMask = bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime) ? ~int(0) : ~int(LinearAccessBit), + Flags = int( evaluator::Flags) & (LinearAccessMask&PacketAccessMask), + + Alignment = int(MapOptions)&int(AlignedMask) + }; + + EIGEN_DEVICE_FUNC explicit evaluator(const XprType& map) + : mapbase_evaluator(map) + { } +}; + +// -------------------- Ref -------------------- + +template +struct evaluator > + : public mapbase_evaluator, PlainObjectType> +{ + typedef Ref XprType; + + enum { + Flags = evaluator >::Flags, + Alignment = evaluator >::Alignment + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const XprType& ref) + : mapbase_evaluator(ref) + { } +}; + +// -------------------- Block -------------------- + +template::ret> struct block_evaluator; + +template +struct evaluator > + : block_evaluator +{ + typedef Block XprType; + typedef typename XprType::Scalar Scalar; + // TODO: should check for smaller packet types once we can handle multi-sized packet types + typedef typename packet_traits::type PacketScalar; + + enum { + CoeffReadCost = evaluator::CoeffReadCost, + + RowsAtCompileTime = traits::RowsAtCompileTime, + ColsAtCompileTime = traits::ColsAtCompileTime, + MaxRowsAtCompileTime = traits::MaxRowsAtCompileTime, + MaxColsAtCompileTime = traits::MaxColsAtCompileTime, + + ArgTypeIsRowMajor = (int(evaluator::Flags)&RowMajorBit) != 0, + IsRowMajor = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? 1 + : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0 + : ArgTypeIsRowMajor, + HasSameStorageOrderAsArgType = (IsRowMajor == ArgTypeIsRowMajor), + InnerSize = IsRowMajor ? int(ColsAtCompileTime) : int(RowsAtCompileTime), + InnerStrideAtCompileTime = HasSameStorageOrderAsArgType + ? int(inner_stride_at_compile_time::ret) + : int(outer_stride_at_compile_time::ret), + OuterStrideAtCompileTime = HasSameStorageOrderAsArgType + ? int(outer_stride_at_compile_time::ret) + : int(inner_stride_at_compile_time::ret), + MaskPacketAccessBit = (InnerStrideAtCompileTime == 1 || HasSameStorageOrderAsArgType) ? PacketAccessBit : 0, + + FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1 || (InnerPanel && (evaluator::Flags&LinearAccessBit))) ? LinearAccessBit : 0, + FlagsRowMajorBit = XprType::Flags&RowMajorBit, + Flags0 = evaluator::Flags & ( (HereditaryBits & ~RowMajorBit) | + DirectAccessBit | + MaskPacketAccessBit), + Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit, + + PacketAlignment = unpacket_traits::alignment, + Alignment0 = (InnerPanel && (OuterStrideAtCompileTime!=Dynamic) + && (OuterStrideAtCompileTime!=0) + && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % int(PacketAlignment)) == 0)) ? int(PacketAlignment) : 0, + Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator::Alignment, Alignment0) + }; + typedef block_evaluator block_evaluator_type; + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const XprType& block) : block_evaluator_type(block) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } +}; + +// no direct-access => dispatch to a unary evaluator +template +struct block_evaluator + : unary_evaluator > +{ + typedef Block XprType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit block_evaluator(const XprType& block) + : unary_evaluator(block) + {} +}; + +template +struct unary_evaluator, IndexBased> + : evaluator_base > +{ + typedef Block XprType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit unary_evaluator(const XprType& block) + : m_argImpl(block.nestedExpression()), + m_startRow(block.startRow()), + m_startCol(block.startCol()), + m_linear_offset(ForwardLinearAccess?(ArgType::IsRowMajor ? block.startRow()*block.nestedExpression().cols() + block.startCol() : block.startCol()*block.nestedExpression().rows() + block.startRow()):0) + { } + + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + enum { + RowsAtCompileTime = XprType::RowsAtCompileTime, + ForwardLinearAccess = (InnerPanel || int(XprType::IsRowMajor)==int(ArgType::IsRowMajor)) && bool(evaluator::Flags&LinearAccessBit) + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_argImpl.coeff(m_startRow.value() + row, m_startCol.value() + col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return linear_coeff_impl(index, bool_constant()); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + return m_argImpl.coeffRef(m_startRow.value() + row, m_startCol.value() + col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + return linear_coeffRef_impl(index, bool_constant()); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + return m_argImpl.template packet(m_startRow.value() + row, m_startCol.value() + col); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + if (ForwardLinearAccess) + return m_argImpl.template packet(m_linear_offset.value() + index); + else + return packet(RowsAtCompileTime == 1 ? 0 : index, + RowsAtCompileTime == 1 ? index : 0); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index row, Index col, const PacketType& x) + { + return m_argImpl.template writePacket(m_startRow.value() + row, m_startCol.value() + col, x); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index index, const PacketType& x) + { + if (ForwardLinearAccess) + return m_argImpl.template writePacket(m_linear_offset.value() + index, x); + else + return writePacket(RowsAtCompileTime == 1 ? 0 : index, + RowsAtCompileTime == 1 ? index : 0, + x); + } + +protected: + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType linear_coeff_impl(Index index, internal::true_type /* ForwardLinearAccess */) const + { + return m_argImpl.coeff(m_linear_offset.value() + index); + } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType linear_coeff_impl(Index index, internal::false_type /* not ForwardLinearAccess */) const + { + return coeff(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& linear_coeffRef_impl(Index index, internal::true_type /* ForwardLinearAccess */) + { + return m_argImpl.coeffRef(m_linear_offset.value() + index); + } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& linear_coeffRef_impl(Index index, internal::false_type /* not ForwardLinearAccess */) + { + return coeffRef(RowsAtCompileTime == 1 ? 0 : index, RowsAtCompileTime == 1 ? index : 0); + } + + evaluator m_argImpl; + const variable_if_dynamic m_startRow; + const variable_if_dynamic m_startCol; + const variable_if_dynamic m_linear_offset; +}; + +// TODO: This evaluator does not actually use the child evaluator; +// all action is via the data() as returned by the Block expression. + +template +struct block_evaluator + : mapbase_evaluator, + typename Block::PlainObject> +{ + typedef Block XprType; + typedef typename XprType::Scalar Scalar; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit block_evaluator(const XprType& block) + : mapbase_evaluator(block) + { + // TODO: for the 3.3 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime + eigen_assert(((internal::UIntPtr(block.data()) % EIGEN_PLAIN_ENUM_MAX(1,evaluator::Alignment)) == 0) && "data is not aligned"); + } +}; + + +// -------------------- Select -------------------- +// NOTE shall we introduce a ternary_evaluator? + +// TODO enable vectorization for Select +template +struct evaluator > + : evaluator_base > +{ + typedef Select XprType; + enum { + CoeffReadCost = evaluator::CoeffReadCost + + EIGEN_PLAIN_ENUM_MAX(evaluator::CoeffReadCost, + evaluator::CoeffReadCost), + + Flags = (unsigned int)evaluator::Flags & evaluator::Flags & HereditaryBits, + + Alignment = EIGEN_PLAIN_ENUM_MIN(evaluator::Alignment, evaluator::Alignment) + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const XprType& select) + : m_conditionImpl(select.conditionMatrix()), + m_thenImpl(select.thenMatrix()), + m_elseImpl(select.elseMatrix()) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + if (m_conditionImpl.coeff(row, col)) + return m_thenImpl.coeff(row, col); + else + return m_elseImpl.coeff(row, col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + if (m_conditionImpl.coeff(index)) + return m_thenImpl.coeff(index); + else + return m_elseImpl.coeff(index); + } + +protected: + evaluator m_conditionImpl; + evaluator m_thenImpl; + evaluator m_elseImpl; +}; + + +// -------------------- Replicate -------------------- + +template +struct unary_evaluator > + : evaluator_base > +{ + typedef Replicate XprType; + typedef typename XprType::CoeffReturnType CoeffReturnType; + enum { + Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor + }; + typedef typename internal::nested_eval::type ArgTypeNested; + typedef typename internal::remove_all::type ArgTypeNestedCleaned; + + enum { + CoeffReadCost = evaluator::CoeffReadCost, + LinearAccessMask = XprType::IsVectorAtCompileTime ? LinearAccessBit : 0, + Flags = (evaluator::Flags & (HereditaryBits|LinearAccessMask) & ~RowMajorBit) | (traits::Flags & RowMajorBit), + + Alignment = evaluator::Alignment + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit unary_evaluator(const XprType& replicate) + : m_arg(replicate.nestedExpression()), + m_argImpl(m_arg), + m_rows(replicate.nestedExpression().rows()), + m_cols(replicate.nestedExpression().cols()) + {} + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + // try to avoid using modulo; this is a pure optimization strategy + const Index actual_row = internal::traits::RowsAtCompileTime==1 ? 0 + : RowFactor==1 ? row + : row % m_rows.value(); + const Index actual_col = internal::traits::ColsAtCompileTime==1 ? 0 + : ColFactor==1 ? col + : col % m_cols.value(); + + return m_argImpl.coeff(actual_row, actual_col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + // try to avoid using modulo; this is a pure optimization strategy + const Index actual_index = internal::traits::RowsAtCompileTime==1 + ? (ColFactor==1 ? index : index%m_cols.value()) + : (RowFactor==1 ? index : index%m_rows.value()); + + return m_argImpl.coeff(actual_index); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + const Index actual_row = internal::traits::RowsAtCompileTime==1 ? 0 + : RowFactor==1 ? row + : row % m_rows.value(); + const Index actual_col = internal::traits::ColsAtCompileTime==1 ? 0 + : ColFactor==1 ? col + : col % m_cols.value(); + + return m_argImpl.template packet(actual_row, actual_col); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + const Index actual_index = internal::traits::RowsAtCompileTime==1 + ? (ColFactor==1 ? index : index%m_cols.value()) + : (RowFactor==1 ? index : index%m_rows.value()); + + return m_argImpl.template packet(actual_index); + } + +protected: + const ArgTypeNested m_arg; + evaluator m_argImpl; + const variable_if_dynamic m_rows; + const variable_if_dynamic m_cols; +}; + +// -------------------- MatrixWrapper and ArrayWrapper -------------------- +// +// evaluator_wrapper_base is a common base class for the +// MatrixWrapper and ArrayWrapper evaluators. + +template +struct evaluator_wrapper_base + : evaluator_base +{ + typedef typename remove_all::type ArgType; + enum { + CoeffReadCost = evaluator::CoeffReadCost, + Flags = evaluator::Flags, + Alignment = evaluator::Alignment + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator_wrapper_base(const ArgType& arg) : m_argImpl(arg) {} + + typedef typename ArgType::Scalar Scalar; + typedef typename ArgType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_argImpl.coeff(row, col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_argImpl.coeff(index); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + return m_argImpl.coeffRef(row, col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + return m_argImpl.coeffRef(index); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + return m_argImpl.template packet(row, col); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + return m_argImpl.template packet(index); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index row, Index col, const PacketType& x) + { + m_argImpl.template writePacket(row, col, x); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index index, const PacketType& x) + { + m_argImpl.template writePacket(index, x); + } + +protected: + evaluator m_argImpl; +}; + +template +struct unary_evaluator > + : evaluator_wrapper_base > +{ + typedef MatrixWrapper XprType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit unary_evaluator(const XprType& wrapper) + : evaluator_wrapper_base >(wrapper.nestedExpression()) + { } +}; + +template +struct unary_evaluator > + : evaluator_wrapper_base > +{ + typedef ArrayWrapper XprType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit unary_evaluator(const XprType& wrapper) + : evaluator_wrapper_base >(wrapper.nestedExpression()) + { } +}; + + +// -------------------- Reverse -------------------- + +// defined in Reverse.h: +template struct reverse_packet_cond; + +template +struct unary_evaluator > + : evaluator_base > +{ + typedef Reverse XprType; + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + enum { + IsRowMajor = XprType::IsRowMajor, + IsColMajor = !IsRowMajor, + ReverseRow = (Direction == Vertical) || (Direction == BothDirections), + ReverseCol = (Direction == Horizontal) || (Direction == BothDirections), + ReversePacket = (Direction == BothDirections) + || ((Direction == Vertical) && IsColMajor) + || ((Direction == Horizontal) && IsRowMajor), + + CoeffReadCost = evaluator::CoeffReadCost, + + // let's enable LinearAccess only with vectorization because of the product overhead + // FIXME enable DirectAccess with negative strides? + Flags0 = evaluator::Flags, + LinearAccess = ( (Direction==BothDirections) && (int(Flags0)&PacketAccessBit) ) + || ((ReverseRow && XprType::ColsAtCompileTime==1) || (ReverseCol && XprType::RowsAtCompileTime==1)) + ? LinearAccessBit : 0, + + Flags = int(Flags0) & (HereditaryBits | PacketAccessBit | LinearAccess), + + Alignment = 0 // FIXME in some rare cases, Alignment could be preserved, like a Vector4f. + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit unary_evaluator(const XprType& reverse) + : m_argImpl(reverse.nestedExpression()), + m_rows(ReverseRow ? reverse.nestedExpression().rows() : 1), + m_cols(ReverseCol ? reverse.nestedExpression().cols() : 1) + { } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + return m_argImpl.coeff(ReverseRow ? m_rows.value() - row - 1 : row, + ReverseCol ? m_cols.value() - col - 1 : col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_argImpl.coeff(m_rows.value() * m_cols.value() - index - 1); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + return m_argImpl.coeffRef(ReverseRow ? m_rows.value() - row - 1 : row, + ReverseCol ? m_cols.value() - col - 1 : col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + return m_argImpl.coeffRef(m_rows.value() * m_cols.value() - index - 1); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index row, Index col) const + { + enum { + PacketSize = unpacket_traits::size, + OffsetRow = ReverseRow && IsColMajor ? PacketSize : 1, + OffsetCol = ReverseCol && IsRowMajor ? PacketSize : 1 + }; + typedef internal::reverse_packet_cond reverse_packet; + return reverse_packet::run(m_argImpl.template packet( + ReverseRow ? m_rows.value() - row - OffsetRow : row, + ReverseCol ? m_cols.value() - col - OffsetCol : col)); + } + + template + EIGEN_STRONG_INLINE + PacketType packet(Index index) const + { + enum { PacketSize = unpacket_traits::size }; + return preverse(m_argImpl.template packet(m_rows.value() * m_cols.value() - index - PacketSize)); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index row, Index col, const PacketType& x) + { + // FIXME we could factorize some code with packet(i,j) + enum { + PacketSize = unpacket_traits::size, + OffsetRow = ReverseRow && IsColMajor ? PacketSize : 1, + OffsetCol = ReverseCol && IsRowMajor ? PacketSize : 1 + }; + typedef internal::reverse_packet_cond reverse_packet; + m_argImpl.template writePacket( + ReverseRow ? m_rows.value() - row - OffsetRow : row, + ReverseCol ? m_cols.value() - col - OffsetCol : col, + reverse_packet::run(x)); + } + + template + EIGEN_STRONG_INLINE + void writePacket(Index index, const PacketType& x) + { + enum { PacketSize = unpacket_traits::size }; + m_argImpl.template writePacket + (m_rows.value() * m_cols.value() - index - PacketSize, preverse(x)); + } + +protected: + evaluator m_argImpl; + + // If we do not reverse rows, then we do not need to know the number of rows; same for columns + // Nonetheless, in this case it is important to set to 1 such that the coeff(index) method works fine for vectors. + const variable_if_dynamic m_rows; + const variable_if_dynamic m_cols; +}; + + +// -------------------- Diagonal -------------------- + +template +struct evaluator > + : evaluator_base > +{ + typedef Diagonal XprType; + + enum { + CoeffReadCost = evaluator::CoeffReadCost, + + Flags = (unsigned int)(evaluator::Flags & (HereditaryBits | DirectAccessBit) & ~RowMajorBit) | LinearAccessBit, + + Alignment = 0 + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit evaluator(const XprType& diagonal) + : m_argImpl(diagonal.nestedExpression()), + m_index(diagonal.index()) + { } + + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index) const + { + return m_argImpl.coeff(row + rowOffset(), row + colOffset()); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index index) const + { + return m_argImpl.coeff(index + rowOffset(), index + colOffset()); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index) + { + return m_argImpl.coeffRef(row + rowOffset(), row + colOffset()); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + return m_argImpl.coeffRef(index + rowOffset(), index + colOffset()); + } + +protected: + evaluator m_argImpl; + const internal::variable_if_dynamicindex m_index; + +private: + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rowOffset() const { return m_index.value() > 0 ? 0 : -m_index.value(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index colOffset() const { return m_index.value() > 0 ? m_index.value() : 0; } +}; + + +//---------------------------------------------------------------------- +// deprecated code +//---------------------------------------------------------------------- + +// -------------------- EvalToTemp -------------------- + +// expression class for evaluating nested expression to a temporary + +template class EvalToTemp; + +template +struct traits > + : public traits +{ }; + +template +class EvalToTemp + : public dense_xpr_base >::type +{ + public: + + typedef typename dense_xpr_base::type Base; + EIGEN_GENERIC_PUBLIC_INTERFACE(EvalToTemp) + + explicit EvalToTemp(const ArgType& arg) + : m_arg(arg) + { } + + const ArgType& arg() const + { + return m_arg; + } + + EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT + { + return m_arg.rows(); + } + + EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT + { + return m_arg.cols(); + } + + private: + const ArgType& m_arg; +}; + +template +struct evaluator > + : public evaluator +{ + typedef EvalToTemp XprType; + typedef typename ArgType::PlainObject PlainObject; + typedef evaluator Base; + + EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) + : m_result(xpr.arg()) + { + ::new (static_cast(this)) Base(m_result); + } + + // This constructor is used when nesting an EvalTo evaluator in another evaluator + EIGEN_DEVICE_FUNC evaluator(const ArgType& arg) + : m_result(arg) + { + ::new (static_cast(this)) Base(m_result); + } + +protected: + PlainObject m_result; +}; + +} // namespace internal + +} // end namespace Eigen + +#endif // EIGEN_COREEVALUATORS_H diff --git a/thirdparty/eigen/Eigen/src/Core/CoreIterators.h b/thirdparty/eigen/Eigen/src/Core/CoreIterators.h index 6da4683d..b9671968 100644 --- a/thirdparty/eigen/Eigen/src/Core/CoreIterators.h +++ b/thirdparty/eigen/Eigen/src/Core/CoreIterators.h @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2008-2010 Gael Guennebaud +// Copyright (C) 2008-2014 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -15,47 +15,118 @@ namespace Eigen { /* This file contains the respective InnerIterator definition of the expressions defined in Eigen/Core */ -/** \ingroup SparseCore_Module - * \class InnerIterator - * \brief An InnerIterator allows to loop over the element of a sparse (or dense) matrix or expression - * - * todo +namespace internal { + +template +class inner_iterator_selector; + +} + +/** \class InnerIterator + * \brief An InnerIterator allows to loop over the element of any matrix expression. + * + * \warning To be used with care because an evaluator is constructed every time an InnerIterator iterator is constructed. + * + * TODO: add a usage example */ +template +class InnerIterator +{ +protected: + typedef internal::inner_iterator_selector::Kind> IteratorType; + typedef internal::evaluator EvaluatorType; + typedef typename internal::traits::Scalar Scalar; +public: + /** Construct an iterator over the \a outerId -th row or column of \a xpr */ + InnerIterator(const XprType &xpr, const Index &outerId) + : m_eval(xpr), m_iter(m_eval, outerId, xpr.innerSize()) + {} + + /// \returns the value of the current coefficient. + EIGEN_STRONG_INLINE Scalar value() const { return m_iter.value(); } + /** Increment the iterator \c *this to the next non-zero coefficient. + * Explicit zeros are not skipped over. To skip explicit zeros, see class SparseView + */ + EIGEN_STRONG_INLINE InnerIterator& operator++() { m_iter.operator++(); return *this; } + EIGEN_STRONG_INLINE InnerIterator& operator+=(Index i) { m_iter.operator+=(i); return *this; } + EIGEN_STRONG_INLINE InnerIterator operator+(Index i) + { InnerIterator result(*this); result+=i; return result; } + + + /// \returns the column or row index of the current coefficient. + EIGEN_STRONG_INLINE Index index() const { return m_iter.index(); } + /// \returns the row index of the current coefficient. + EIGEN_STRONG_INLINE Index row() const { return m_iter.row(); } + /// \returns the column index of the current coefficient. + EIGEN_STRONG_INLINE Index col() const { return m_iter.col(); } + /// \returns \c true if the iterator \c *this still references a valid coefficient. + EIGEN_STRONG_INLINE operator bool() const { return m_iter; } + +protected: + EvaluatorType m_eval; + IteratorType m_iter; +private: + // If you get here, then you're not using the right InnerIterator type, e.g.: + // SparseMatrix A; + // SparseMatrix::InnerIterator it(A,0); + template InnerIterator(const EigenBase&,Index outer); +}; + +namespace internal { + +// Generic inner iterator implementation for dense objects +template +class inner_iterator_selector +{ +protected: + typedef evaluator EvaluatorType; + typedef typename traits::Scalar Scalar; + enum { IsRowMajor = (XprType::Flags&RowMajorBit)==RowMajorBit }; + +public: + EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &innerSize) + : m_eval(eval), m_inner(0), m_outer(outerId), m_end(innerSize) + {} + + EIGEN_STRONG_INLINE Scalar value() const + { + return (IsRowMajor) ? m_eval.coeff(m_outer, m_inner) + : m_eval.coeff(m_inner, m_outer); + } + + EIGEN_STRONG_INLINE inner_iterator_selector& operator++() { m_inner++; return *this; } + + EIGEN_STRONG_INLINE Index index() const { return m_inner; } + inline Index row() const { return IsRowMajor ? m_outer : index(); } + inline Index col() const { return IsRowMajor ? index() : m_outer; } -// generic version for dense matrix and expressions -template class DenseBase::InnerIterator + EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; } + +protected: + const EvaluatorType& m_eval; + Index m_inner; + const Index m_outer; + const Index m_end; +}; + +// For iterator-based evaluator, inner-iterator is already implemented as +// evaluator<>::InnerIterator +template +class inner_iterator_selector + : public evaluator::InnerIterator { - protected: - typedef typename Derived::Scalar Scalar; - typedef typename Derived::Index Index; - - enum { IsRowMajor = (Derived::Flags&RowMajorBit)==RowMajorBit }; - public: - EIGEN_STRONG_INLINE InnerIterator(const Derived& expr, Index outer) - : m_expression(expr), m_inner(0), m_outer(outer), m_end(expr.innerSize()) - {} - - EIGEN_STRONG_INLINE Scalar value() const - { - return (IsRowMajor) ? m_expression.coeff(m_outer, m_inner) - : m_expression.coeff(m_inner, m_outer); - } - - EIGEN_STRONG_INLINE InnerIterator& operator++() { m_inner++; return *this; } - - EIGEN_STRONG_INLINE Index index() const { return m_inner; } - inline Index row() const { return IsRowMajor ? m_outer : index(); } - inline Index col() const { return IsRowMajor ? index() : m_outer; } - - EIGEN_STRONG_INLINE operator bool() const { return m_inner < m_end && m_inner>=0; } - - protected: - const Derived& m_expression; - Index m_inner; - const Index m_outer; - const Index m_end; +protected: + typedef typename evaluator::InnerIterator Base; + typedef evaluator EvaluatorType; + +public: + EIGEN_STRONG_INLINE inner_iterator_selector(const EvaluatorType &eval, const Index &outerId, const Index &/*innerSize*/) + : Base(eval, outerId) + {} }; +} // end namespace internal + } // end namespace Eigen #endif // EIGEN_COREITERATORS_H diff --git a/thirdparty/eigen/Eigen/src/Core/CwiseBinaryOp.h b/thirdparty/eigen/Eigen/src/Core/CwiseBinaryOp.h index 519a866e..2202b1cc 100644 --- a/thirdparty/eigen/Eigen/src/Core/CwiseBinaryOp.h +++ b/thirdparty/eigen/Eigen/src/Core/CwiseBinaryOp.h @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2008-2009 Gael Guennebaud +// Copyright (C) 2008-2014 Gael Guennebaud // Copyright (C) 2006-2008 Benoit Jacob // // This Source Code Form is subject to the terms of the Mozilla @@ -13,26 +13,6 @@ namespace Eigen { -/** \class CwiseBinaryOp - * \ingroup Core_Module - * - * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions - * - * \param BinaryOp template functor implementing the operator - * \param Lhs the type of the left-hand side - * \param Rhs the type of the right-hand side - * - * This class represents an expression where a coefficient-wise binary operator is applied to two expressions. - * It is the return type of binary operators, by which we mean only those binary operators where - * both the left-hand side and the right-hand side are Eigen expressions. - * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp. - * - * Most of the time, this is the only way that it is used, so you typically don't have to name - * CwiseBinaryOp types explicitly. - * - * \sa MatrixBase::binaryExpr(const MatrixBase &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp - */ - namespace internal { template struct traits > @@ -52,78 +32,82 @@ struct traits > // we still want to handle the case when the result type is different. typedef typename result_of< BinaryOp( - typename Lhs::Scalar, - typename Rhs::Scalar + const typename Lhs::Scalar&, + const typename Rhs::Scalar& ) >::type Scalar; - typedef typename promote_storage_type::StorageKind, - typename traits::StorageKind>::ret StorageKind; - typedef typename promote_index_type::Index, - typename traits::Index>::type Index; + typedef typename cwise_promote_storage_type::StorageKind, + typename traits::StorageKind, + BinaryOp>::ret StorageKind; + typedef typename promote_index_type::StorageIndex, + typename traits::StorageIndex>::type StorageIndex; typedef typename Lhs::Nested LhsNested; typedef typename Rhs::Nested RhsNested; typedef typename remove_reference::type _LhsNested; typedef typename remove_reference::type _RhsNested; enum { - LhsCoeffReadCost = _LhsNested::CoeffReadCost, - RhsCoeffReadCost = _RhsNested::CoeffReadCost, - LhsFlags = _LhsNested::Flags, - RhsFlags = _RhsNested::Flags, - SameType = is_same::value, - StorageOrdersAgree = (int(Lhs::Flags)&RowMajorBit)==(int(Rhs::Flags)&RowMajorBit), - Flags0 = (int(LhsFlags) | int(RhsFlags)) & ( - HereditaryBits - | (int(LhsFlags) & int(RhsFlags) & - ( AlignedBit - | (StorageOrdersAgree ? LinearAccessBit : 0) - | (functor_traits::PacketAccess && StorageOrdersAgree && SameType ? PacketAccessBit : 0) - ) - ) - ), - Flags = (Flags0 & ~RowMajorBit) | (LhsFlags & RowMajorBit), - Cost0 = EIGEN_ADD_COST(LhsCoeffReadCost,RhsCoeffReadCost), - CoeffReadCost = EIGEN_ADD_COST(Cost0,functor_traits::Cost) + Flags = cwise_promote_storage_order::StorageKind,typename traits::StorageKind,_LhsNested::Flags & RowMajorBit,_RhsNested::Flags & RowMajorBit>::value }; }; } // end namespace internal -// we require Lhs and Rhs to have the same scalar type. Currently there is no example of a binary functor -// that would take two operands of different types. If there were such an example, then this check should be -// moved to the BinaryOp functors, on a per-case basis. This would however require a change in the BinaryOp functors, as -// currently they take only one typename Scalar template parameter. -// It is tempting to always allow mixing different types but remember that this is often impossible in the vectorized paths. -// So allowing mixing different types gives very unexpected errors when enabling vectorization, when the user tries to -// add together a float matrix and a double matrix. -#define EIGEN_CHECK_BINARY_COMPATIBILIY(BINOP,LHS,RHS) \ - EIGEN_STATIC_ASSERT((internal::functor_is_product_like::ret \ - ? int(internal::scalar_product_traits::Defined) \ - : int(internal::is_same::value)), \ - YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) - template class CwiseBinaryOpImpl; -template -class CwiseBinaryOp : internal::no_assignment_operator, +/** \class CwiseBinaryOp + * \ingroup Core_Module + * + * \brief Generic expression where a coefficient-wise binary operator is applied to two expressions + * + * \tparam BinaryOp template functor implementing the operator + * \tparam LhsType the type of the left-hand side + * \tparam RhsType the type of the right-hand side + * + * This class represents an expression where a coefficient-wise binary operator is applied to two expressions. + * It is the return type of binary operators, by which we mean only those binary operators where + * both the left-hand side and the right-hand side are Eigen expressions. + * For example, the return type of matrix1+matrix2 is a CwiseBinaryOp. + * + * Most of the time, this is the only way that it is used, so you typically don't have to name + * CwiseBinaryOp types explicitly. + * + * \sa MatrixBase::binaryExpr(const MatrixBase &,const CustomBinaryOp &) const, class CwiseUnaryOp, class CwiseNullaryOp + */ +template +class CwiseBinaryOp : public CwiseBinaryOpImpl< - BinaryOp, Lhs, Rhs, - typename internal::promote_storage_type::StorageKind, - typename internal::traits::StorageKind>::ret> + BinaryOp, LhsType, RhsType, + typename internal::cwise_promote_storage_type::StorageKind, + typename internal::traits::StorageKind, + BinaryOp>::ret>, + internal::no_assignment_operator { public: + typedef typename internal::remove_all::type Functor; + typedef typename internal::remove_all::type Lhs; + typedef typename internal::remove_all::type Rhs; + typedef typename CwiseBinaryOpImpl< - BinaryOp, Lhs, Rhs, - typename internal::promote_storage_type::StorageKind, - typename internal::traits::StorageKind>::ret>::Base Base; + BinaryOp, LhsType, RhsType, + typename internal::cwise_promote_storage_type::StorageKind, + typename internal::traits::StorageKind, + BinaryOp>::ret>::Base Base; EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseBinaryOp) - typedef typename internal::nested::type LhsNested; - typedef typename internal::nested::type RhsNested; + typedef typename internal::ref_selector::type LhsNested; + typedef typename internal::ref_selector::type RhsNested; typedef typename internal::remove_reference::type _LhsNested; typedef typename internal::remove_reference::type _RhsNested; - EIGEN_STRONG_INLINE CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp()) +#if EIGEN_COMP_MSVC && EIGEN_HAS_CXX11 + //Required for Visual Studio or the Copy constructor will probably not get inlined! + EIGEN_STRONG_INLINE + CwiseBinaryOp(const CwiseBinaryOp&) = default; +#endif + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CwiseBinaryOp(const Lhs& aLhs, const Rhs& aRhs, const BinaryOp& func = BinaryOp()) : m_lhs(aLhs), m_rhs(aRhs), m_functor(func) { EIGEN_CHECK_BINARY_COMPATIBILIY(BinaryOp,typename Lhs::Scalar,typename Rhs::Scalar); @@ -132,26 +116,25 @@ class CwiseBinaryOp : internal::no_assignment_operator, eigen_assert(aLhs.rows() == aRhs.rows() && aLhs.cols() == aRhs.cols()); } - EIGEN_STRONG_INLINE Index rows() const { + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rows() const EIGEN_NOEXCEPT { // return the fixed size type if available to enable compile time optimizations - if (internal::traits::type>::RowsAtCompileTime==Dynamic) - return m_rhs.rows(); - else - return m_lhs.rows(); + return internal::traits::type>::RowsAtCompileTime==Dynamic ? m_rhs.rows() : m_lhs.rows(); } - EIGEN_STRONG_INLINE Index cols() const { + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index cols() const EIGEN_NOEXCEPT { // return the fixed size type if available to enable compile time optimizations - if (internal::traits::type>::ColsAtCompileTime==Dynamic) - return m_rhs.cols(); - else - return m_lhs.cols(); + return internal::traits::type>::ColsAtCompileTime==Dynamic ? m_rhs.cols() : m_lhs.cols(); } /** \returns the left hand side nested expression */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const _LhsNested& lhs() const { return m_lhs; } /** \returns the right hand side nested expression */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const _RhsNested& rhs() const { return m_rhs; } /** \returns the functor representing the binary operation */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const BinaryOp& functor() const { return m_functor; } protected: @@ -160,41 +143,13 @@ class CwiseBinaryOp : internal::no_assignment_operator, const BinaryOp m_functor; }; -template -class CwiseBinaryOpImpl - : public internal::dense_xpr_base >::type +// Generic API dispatcher +template +class CwiseBinaryOpImpl + : public internal::generic_xpr_base >::type { - typedef CwiseBinaryOp Derived; - public: - - typedef typename internal::dense_xpr_base >::type Base; - EIGEN_DENSE_PUBLIC_INTERFACE( Derived ) - - EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const - { - return derived().functor()(derived().lhs().coeff(rowId, colId), - derived().rhs().coeff(rowId, colId)); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const - { - return derived().functor().packetOp(derived().lhs().template packet(rowId, colId), - derived().rhs().template packet(rowId, colId)); - } - - EIGEN_STRONG_INLINE const Scalar coeff(Index index) const - { - return derived().functor()(derived().lhs().coeff(index), - derived().rhs().coeff(index)); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index index) const - { - return derived().functor().packetOp(derived().lhs().template packet(index), - derived().rhs().template packet(index)); - } +public: + typedef typename internal::generic_xpr_base >::type Base; }; /** replaces \c *this by \c *this - \a other. @@ -203,11 +158,10 @@ class CwiseBinaryOpImpl */ template template -EIGEN_STRONG_INLINE Derived & +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived & MatrixBase::operator-=(const MatrixBase &other) { - SelfCwiseBinaryOp, Derived, OtherDerived> tmp(derived()); - tmp = other.derived(); + call_assignment(derived(), other.derived(), internal::sub_assign_op()); return derived(); } @@ -217,11 +171,10 @@ MatrixBase::operator-=(const MatrixBase &other) */ template template -EIGEN_STRONG_INLINE Derived & +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived & MatrixBase::operator+=(const MatrixBase& other) { - SelfCwiseBinaryOp, Derived, OtherDerived> tmp(derived()); - tmp = other.derived(); + call_assignment(derived(), other.derived(), internal::add_assign_op()); return derived(); } diff --git a/thirdparty/eigen/Eigen/src/Core/CwiseNullaryOp.h b/thirdparty/eigen/Eigen/src/Core/CwiseNullaryOp.h index a93bab2d..ba07e71e 100644 --- a/thirdparty/eigen/Eigen/src/Core/CwiseNullaryOp.h +++ b/thirdparty/eigen/Eigen/src/Core/CwiseNullaryOp.h @@ -12,13 +12,24 @@ namespace Eigen { +namespace internal { +template +struct traits > : traits +{ + enum { + Flags = traits::Flags & RowMajorBit + }; +}; + +} // namespace internal + /** \class CwiseNullaryOp * \ingroup Core_Module * * \brief Generic expression of a matrix where all coefficients are defined by a functor * - * \param NullaryOp template functor implementing the operator - * \param PlainObjectType the underlying plain matrix/array type + * \tparam NullaryOp template functor implementing the operator + * \tparam PlainObjectType the underlying plain matrix/array type * * This class represents an expression of a generic nullary operator. * It is the return type of the Ones(), Zero(), Constant(), Identity() and Random() methods, @@ -27,68 +38,49 @@ namespace Eigen { * However, if you want to write a function returning such an expression, you * will need to use this class. * - * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr() + * The functor NullaryOp must expose one of the following method: + + + + +
\c operator()() if the procedural generation does not depend on the coefficient entries (e.g., random numbers)
\c operator()(Index i)if the procedural generation makes sense for vectors only and that it depends on the coefficient index \c i (e.g., linspace)
\c operator()(Index i,Index j)if the procedural generation depends on the matrix coordinates \c i, \c j (e.g., to generate a checkerboard with 0 and 1)
+ * It is also possible to expose the last two operators if the generation makes sense for matrices but can be optimized for vectors. + * + * See DenseBase::NullaryExpr(Index,const CustomNullaryOp&) for an example binding + * C++11 random number generators. + * + * A nullary expression can also be used to implement custom sophisticated matrix manipulations + * that cannot be covered by the existing set of natively supported matrix manipulations. + * See this \ref TopicCustomizing_NullaryExpr "page" for some examples and additional explanations + * on the behavior of CwiseNullaryOp. + * + * \sa class CwiseUnaryOp, class CwiseBinaryOp, DenseBase::NullaryExpr */ - -namespace internal { template -struct traits > : traits -{ - enum { - Flags = (traits::Flags - & ( HereditaryBits - | (functor_has_linear_access::ret ? LinearAccessBit : 0) - | (functor_traits::PacketAccess ? PacketAccessBit : 0))) - | (functor_traits::IsRepeatable ? 0 : EvalBeforeNestingBit), - CoeffReadCost = functor_traits::Cost - }; -}; -} - -template -class CwiseNullaryOp : internal::no_assignment_operator, - public internal::dense_xpr_base< CwiseNullaryOp >::type +class CwiseNullaryOp : public internal::dense_xpr_base< CwiseNullaryOp >::type, internal::no_assignment_operator { public: typedef typename internal::dense_xpr_base::type Base; EIGEN_DENSE_PUBLIC_INTERFACE(CwiseNullaryOp) - CwiseNullaryOp(Index nbRows, Index nbCols, const NullaryOp& func = NullaryOp()) - : m_rows(nbRows), m_cols(nbCols), m_functor(func) - { - eigen_assert(nbRows >= 0 - && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows) - && nbCols >= 0 - && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols)); - } - - EIGEN_STRONG_INLINE Index rows() const { return m_rows.value(); } - EIGEN_STRONG_INLINE Index cols() const { return m_cols.value(); } - - EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const - { - return m_functor(rowId, colId); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const + EIGEN_DEVICE_FUNC + CwiseNullaryOp(Index rows, Index cols, const NullaryOp& func = NullaryOp()) + : m_rows(rows), m_cols(cols), m_functor(func) { - return m_functor.packetOp(rowId, colId); + eigen_assert(rows >= 0 + && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows) + && cols >= 0 + && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols)); } - EIGEN_STRONG_INLINE const Scalar coeff(Index index) const - { - return m_functor(index); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index index) const - { - return m_functor.packetOp(index); - } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rows() const { return m_rows.value(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index cols() const { return m_cols.value(); } /** \returns the functor representing the nullary operation */ + EIGEN_DEVICE_FUNC const NullaryOp& functor() const { return m_functor; } protected: @@ -113,10 +105,15 @@ class CwiseNullaryOp : internal::no_assignment_operator, */ template template -EIGEN_STRONG_INLINE const CwiseNullaryOp +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +#ifndef EIGEN_PARSED_BY_DOXYGEN +const CwiseNullaryOp::PlainObject> +#else +const CwiseNullaryOp +#endif DenseBase::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func) { - return CwiseNullaryOp(rows, cols, func); + return CwiseNullaryOp(rows, cols, func); } /** \returns an expression of a matrix defined by a custom functor \a func @@ -132,16 +129,24 @@ DenseBase::NullaryExpr(Index rows, Index cols, const CustomNullaryOp& f * * The template parameter \a CustomNullaryOp is the type of the functor. * + * Here is an example with C++11 random generators: \include random_cpp11.cpp + * Output: \verbinclude random_cpp11.out + * * \sa class CwiseNullaryOp */ template template -EIGEN_STRONG_INLINE const CwiseNullaryOp +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +#ifndef EIGEN_PARSED_BY_DOXYGEN +const CwiseNullaryOp::PlainObject> +#else +const CwiseNullaryOp +#endif DenseBase::NullaryExpr(Index size, const CustomNullaryOp& func) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) - if(RowsAtCompileTime == 1) return CwiseNullaryOp(1, size, func); - else return CwiseNullaryOp(size, 1, func); + if(RowsAtCompileTime == 1) return CwiseNullaryOp(1, size, func); + else return CwiseNullaryOp(size, 1, func); } /** \returns an expression of a matrix defined by a custom functor \a func @@ -155,19 +160,24 @@ DenseBase::NullaryExpr(Index size, const CustomNullaryOp& func) */ template template -EIGEN_STRONG_INLINE const CwiseNullaryOp +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +#ifndef EIGEN_PARSED_BY_DOXYGEN +const CwiseNullaryOp::PlainObject> +#else +const CwiseNullaryOp +#endif DenseBase::NullaryExpr(const CustomNullaryOp& func) { - return CwiseNullaryOp(RowsAtCompileTime, ColsAtCompileTime, func); + return CwiseNullaryOp(RowsAtCompileTime, ColsAtCompileTime, func); } /** \returns an expression of a constant matrix of value \a value * - * The parameters \a nbRows and \a nbCols are the number of rows and of columns of + * The parameters \a rows and \a cols are the number of rows and of columns of * the returned matrix. Must be compatible with this DenseBase type. * * This variant is meant to be used for dynamic-size matrix types. For fixed-size types, - * it is redundant to pass \a nbRows and \a nbCols as arguments, so Zero() should be used + * it is redundant to pass \a rows and \a cols as arguments, so Zero() should be used * instead. * * The template parameter \a CustomNullaryOp is the type of the functor. @@ -175,10 +185,10 @@ DenseBase::NullaryExpr(const CustomNullaryOp& func) * \sa class CwiseNullaryOp */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType -DenseBase::Constant(Index nbRows, Index nbCols, const Scalar& value) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +DenseBase::Constant(Index rows, Index cols, const Scalar& value) { - return DenseBase::NullaryExpr(nbRows, nbCols, internal::scalar_constant_op(value)); + return DenseBase::NullaryExpr(rows, cols, internal::scalar_constant_op(value)); } /** \returns an expression of a constant matrix of value \a value @@ -197,7 +207,7 @@ DenseBase::Constant(Index nbRows, Index nbCols, const Scalar& value) * \sa class CwiseNullaryOp */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType DenseBase::Constant(Index size, const Scalar& value) { return DenseBase::NullaryExpr(size, internal::scalar_constant_op(value)); @@ -213,53 +223,45 @@ DenseBase::Constant(Index size, const Scalar& value) * \sa class CwiseNullaryOp */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType DenseBase::Constant(const Scalar& value) { EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived) return DenseBase::NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_constant_op(value)); } -/** - * \brief Sets a linearly space vector. - * - * The function generates 'size' equally spaced values in the closed interval [low,high]. - * This particular version of LinSpaced() uses sequential access, i.e. vector access is - * assumed to be a(0), a(1), ..., a(size). This assumption allows for better vectorization - * and yields faster code than the random access version. - * - * When size is set to 1, a vector of length 1 containing 'high' is returned. +/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(Index,const Scalar&,const Scalar&) * * \only_for_vectors * - * Example: \include DenseBase_LinSpaced_seq.cpp - * Output: \verbinclude DenseBase_LinSpaced_seq.out + * Example: \include DenseBase_LinSpaced_seq_deprecated.cpp + * Output: \verbinclude DenseBase_LinSpaced_seq_deprecated.out * - * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Index,Scalar,Scalar), CwiseNullaryOp + * \sa LinSpaced(Index,const Scalar&, const Scalar&), setLinSpaced(Index,const Scalar&,const Scalar&) */ template -EIGEN_STRONG_INLINE const typename DenseBase::SequentialLinSpacedReturnType +EIGEN_DEPRECATED EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::RandomAccessLinSpacedReturnType DenseBase::LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) - return DenseBase::NullaryExpr(size, internal::linspaced_op(low,high,size)); + return DenseBase::NullaryExpr(size, internal::linspaced_op(low,high,size)); } -/** - * \copydoc DenseBase::LinSpaced(Sequential_t, Index, const Scalar&, const Scalar&) - * Special version for fixed size types which does not require the size parameter. +/** \deprecated because of accuracy loss. In Eigen 3.3, it is an alias for LinSpaced(const Scalar&,const Scalar&) + * + * \sa LinSpaced(const Scalar&, const Scalar&) */ template -EIGEN_STRONG_INLINE const typename DenseBase::SequentialLinSpacedReturnType +EIGEN_DEPRECATED EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::RandomAccessLinSpacedReturnType DenseBase::LinSpaced(Sequential_t, const Scalar& low, const Scalar& high) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived) - return DenseBase::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op(low,high,Derived::SizeAtCompileTime)); + return DenseBase::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op(low,high,Derived::SizeAtCompileTime)); } /** - * \brief Sets a linearly space vector. + * \brief Sets a linearly spaced vector. * * The function generates 'size' equally spaced values in the closed interval [low,high]. * When size is set to 1, a vector of length 1 containing 'high' is returned. @@ -269,37 +271,48 @@ DenseBase::LinSpaced(Sequential_t, const Scalar& low, const Scalar& hig * Example: \include DenseBase_LinSpaced.cpp * Output: \verbinclude DenseBase_LinSpaced.out * - * \sa setLinSpaced(Index,const Scalar&,const Scalar&), LinSpaced(Sequential_t,Index,const Scalar&,const Scalar&,Index), CwiseNullaryOp + * For integer scalar types, an even spacing is possible if and only if the length of the range, + * i.e., \c high-low is a scalar multiple of \c size-1, or if \c size is a scalar multiple of the + * number of values \c high-low+1 (meaning each value can be repeated the same number of time). + * If one of these two considions is not satisfied, then \c high is lowered to the largest value + * satisfying one of this constraint. + * Here are some examples: + * + * Example: \include DenseBase_LinSpacedInt.cpp + * Output: \verbinclude DenseBase_LinSpacedInt.out + * + * \sa setLinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp */ template -EIGEN_STRONG_INLINE const typename DenseBase::RandomAccessLinSpacedReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::RandomAccessLinSpacedReturnType DenseBase::LinSpaced(Index size, const Scalar& low, const Scalar& high) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) - return DenseBase::NullaryExpr(size, internal::linspaced_op(low,high,size)); + return DenseBase::NullaryExpr(size, internal::linspaced_op(low,high,size)); } /** - * \copydoc DenseBase::LinSpaced(Index, const Scalar&, const Scalar&) + * \copydoc DenseBase::LinSpaced(Index, const DenseBase::Scalar&, const DenseBase::Scalar&) * Special version for fixed size types which does not require the size parameter. */ template -EIGEN_STRONG_INLINE const typename DenseBase::RandomAccessLinSpacedReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::RandomAccessLinSpacedReturnType DenseBase::LinSpaced(const Scalar& low, const Scalar& high) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived) - return DenseBase::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op(low,high,Derived::SizeAtCompileTime)); + return DenseBase::NullaryExpr(Derived::SizeAtCompileTime, internal::linspaced_op(low,high,Derived::SizeAtCompileTime)); } /** \returns true if all coefficients in this matrix are approximately equal to \a val, to within precision \a prec */ template -bool DenseBase::isApproxToConstant +EIGEN_DEVICE_FUNC bool DenseBase::isApproxToConstant (const Scalar& val, const RealScalar& prec) const { + typename internal::nested_eval::type self(derived()); for(Index j = 0; j < cols(); ++j) for(Index i = 0; i < rows(); ++i) - if(!internal::isApprox(this->coeff(i, j), val, prec)) + if(!internal::isApprox(self.coeff(i, j), val, prec)) return false; return true; } @@ -308,7 +321,7 @@ bool DenseBase::isApproxToConstant * * \returns true if all coefficients in this matrix are approximately equal to \a value, to within precision \a prec */ template -bool DenseBase::isConstant +EIGEN_DEVICE_FUNC bool DenseBase::isConstant (const Scalar& val, const RealScalar& prec) const { return isApproxToConstant(val, prec); @@ -319,22 +332,22 @@ bool DenseBase::isConstant * \sa setConstant(), Constant(), class CwiseNullaryOp */ template -EIGEN_STRONG_INLINE void DenseBase::fill(const Scalar& val) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void DenseBase::fill(const Scalar& val) { setConstant(val); } -/** Sets all coefficients in this expression to \a value. +/** Sets all coefficients in this expression to value \a val. * * \sa fill(), setConstant(Index,const Scalar&), setConstant(Index,Index,const Scalar&), setZero(), setOnes(), Constant(), class CwiseNullaryOp, setZero(), setOnes() */ template -EIGEN_STRONG_INLINE Derived& DenseBase::setConstant(const Scalar& val) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::setConstant(const Scalar& val) { return derived() = Constant(rows(), cols(), val); } -/** Resizes to the given \a size, and sets all coefficients in this expression to the given \a value. +/** Resizes to the given \a size, and sets all coefficients in this expression to the given value \a val. * * \only_for_vectors * @@ -344,17 +357,17 @@ EIGEN_STRONG_INLINE Derived& DenseBase::setConstant(const Scalar& val) * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&) */ template -EIGEN_STRONG_INLINE Derived& +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& PlainObjectBase::setConstant(Index size, const Scalar& val) { resize(size); return setConstant(val); } -/** Resizes to the given size, and sets all coefficients in this expression to the given \a value. +/** Resizes to the given size, and sets all coefficients in this expression to the given value \a val. * - * \param nbRows the new number of rows - * \param nbCols the new number of columns + * \param rows the new number of rows + * \param cols the new number of columns * \param val the value to which all coefficients are set * * Example: \include Matrix_setConstant_int_int.cpp @@ -363,15 +376,42 @@ PlainObjectBase::setConstant(Index size, const Scalar& val) * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&) */ template -EIGEN_STRONG_INLINE Derived& -PlainObjectBase::setConstant(Index nbRows, Index nbCols, const Scalar& val) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setConstant(Index rows, Index cols, const Scalar& val) { - resize(nbRows, nbCols); + resize(rows, cols); return setConstant(val); } +/** Resizes to the given size, changing only the number of columns, and sets all + * coefficients in this expression to the given value \a val. For the parameter + * of type NoChange_t, just pass the special value \c NoChange. + * + * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&) + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setConstant(NoChange_t, Index cols, const Scalar& val) +{ + return setConstant(rows(), cols, val); +} + +/** Resizes to the given size, changing only the number of rows, and sets all + * coefficients in this expression to the given value \a val. For the parameter + * of type NoChange_t, just pass the special value \c NoChange. + * + * \sa MatrixBase::setConstant(const Scalar&), setConstant(Index,const Scalar&), class CwiseNullaryOp, MatrixBase::Constant(const Scalar&) + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setConstant(Index rows, NoChange_t, const Scalar& val) +{ + return setConstant(rows, cols(), val); +} + + /** - * \brief Sets a linearly space vector. + * \brief Sets a linearly spaced vector. * * The function generates 'size' equally spaced values in the closed interval [low,high]. * When size is set to 1, a vector of length 1 containing 'high' is returned. @@ -381,27 +421,33 @@ PlainObjectBase::setConstant(Index nbRows, Index nbCols, const Scalar& * Example: \include DenseBase_setLinSpaced.cpp * Output: \verbinclude DenseBase_setLinSpaced.out * - * \sa CwiseNullaryOp + * For integer scalar types, do not miss the explanations on the definition + * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink. + * + * \sa LinSpaced(Index,const Scalar&,const Scalar&), CwiseNullaryOp */ template -EIGEN_STRONG_INLINE Derived& DenseBase::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::setLinSpaced(Index newSize, const Scalar& low, const Scalar& high) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) - return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op(low,high,newSize)); + return derived() = Derived::NullaryExpr(newSize, internal::linspaced_op(low,high,newSize)); } /** - * \brief Sets a linearly space vector. + * \brief Sets a linearly spaced vector. * - * The function fill *this with equally spaced values in the closed interval [low,high]. + * The function fills \c *this with equally spaced values in the closed interval [low,high]. * When size is set to 1, a vector of length 1 containing 'high' is returned. * * \only_for_vectors * - * \sa setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp + * For integer scalar types, do not miss the explanations on the definition + * of \link LinSpaced(Index,const Scalar&,const Scalar&) even spacing \endlink. + * + * \sa LinSpaced(Index,const Scalar&,const Scalar&), setLinSpaced(Index, const Scalar&, const Scalar&), CwiseNullaryOp */ template -EIGEN_STRONG_INLINE Derived& DenseBase::setLinSpaced(const Scalar& low, const Scalar& high) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::setLinSpaced(const Scalar& low, const Scalar& high) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) return setLinSpaced(size(), low, high); @@ -424,10 +470,10 @@ EIGEN_STRONG_INLINE Derived& DenseBase::setLinSpaced(const Scalar& low, * \sa Zero(), Zero(Index) */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType -DenseBase::Zero(Index nbRows, Index nbCols) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +DenseBase::Zero(Index rows, Index cols) { - return Constant(nbRows, nbCols, Scalar(0)); + return Constant(rows, cols, Scalar(0)); } /** \returns an expression of a zero vector. @@ -447,7 +493,7 @@ DenseBase::Zero(Index nbRows, Index nbCols) * \sa Zero(), Zero(Index,Index) */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType DenseBase::Zero(Index size) { return Constant(size, Scalar(0)); @@ -464,7 +510,7 @@ DenseBase::Zero(Index size) * \sa Zero(Index), Zero(Index,Index) */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType DenseBase::Zero() { return Constant(Scalar(0)); @@ -479,11 +525,12 @@ DenseBase::Zero() * \sa class CwiseNullaryOp, Zero() */ template -bool DenseBase::isZero(const RealScalar& prec) const +EIGEN_DEVICE_FUNC bool DenseBase::isZero(const RealScalar& prec) const { + typename internal::nested_eval::type self(derived()); for(Index j = 0; j < cols(); ++j) for(Index i = 0; i < rows(); ++i) - if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast(1), prec)) + if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast(1), prec)) return false; return true; } @@ -496,7 +543,7 @@ bool DenseBase::isZero(const RealScalar& prec) const * \sa class CwiseNullaryOp, Zero() */ template -EIGEN_STRONG_INLINE Derived& DenseBase::setZero() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::setZero() { return setConstant(Scalar(0)); } @@ -511,7 +558,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase::setZero() * \sa DenseBase::setZero(), setZero(Index,Index), class CwiseNullaryOp, DenseBase::Zero() */ template -EIGEN_STRONG_INLINE Derived& +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& PlainObjectBase::setZero(Index newSize) { resize(newSize); @@ -520,8 +567,8 @@ PlainObjectBase::setZero(Index newSize) /** Resizes to the given size, and sets all coefficients in this expression to zero. * - * \param nbRows the new number of rows - * \param nbCols the new number of columns + * \param rows the new number of rows + * \param cols the new number of columns * * Example: \include Matrix_setZero_int_int.cpp * Output: \verbinclude Matrix_setZero_int_int.out @@ -529,18 +576,44 @@ PlainObjectBase::setZero(Index newSize) * \sa DenseBase::setZero(), setZero(Index), class CwiseNullaryOp, DenseBase::Zero() */ template -EIGEN_STRONG_INLINE Derived& -PlainObjectBase::setZero(Index nbRows, Index nbCols) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setZero(Index rows, Index cols) { - resize(nbRows, nbCols); + resize(rows, cols); return setConstant(Scalar(0)); } +/** Resizes to the given size, changing only the number of columns, and sets all + * coefficients in this expression to zero. For the parameter of type NoChange_t, + * just pass the special value \c NoChange. + * + * \sa DenseBase::setZero(), setZero(Index), setZero(Index, Index), setZero(Index, NoChange_t), class CwiseNullaryOp, DenseBase::Zero() + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setZero(NoChange_t, Index cols) +{ + return setZero(rows(), cols); +} + +/** Resizes to the given size, changing only the number of rows, and sets all + * coefficients in this expression to zero. For the parameter of type NoChange_t, + * just pass the special value \c NoChange. + * + * \sa DenseBase::setZero(), setZero(Index), setZero(Index, Index), setZero(NoChange_t, Index), class CwiseNullaryOp, DenseBase::Zero() + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setZero(Index rows, NoChange_t) +{ + return setZero(rows, cols()); +} + // ones: /** \returns an expression of a matrix where all coefficients equal one. * - * The parameters \a nbRows and \a nbCols are the number of rows and of columns of + * The parameters \a rows and \a cols are the number of rows and of columns of * the returned matrix. Must be compatible with this MatrixBase type. * * This variant is meant to be used for dynamic-size matrix types. For fixed-size types, @@ -553,10 +626,10 @@ PlainObjectBase::setZero(Index nbRows, Index nbCols) * \sa Ones(), Ones(Index), isOnes(), class Ones */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType -DenseBase::Ones(Index nbRows, Index nbCols) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +DenseBase::Ones(Index rows, Index cols) { - return Constant(nbRows, nbCols, Scalar(1)); + return Constant(rows, cols, Scalar(1)); } /** \returns an expression of a vector where all coefficients equal one. @@ -576,7 +649,7 @@ DenseBase::Ones(Index nbRows, Index nbCols) * \sa Ones(), Ones(Index,Index), isOnes(), class Ones */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType DenseBase::Ones(Index newSize) { return Constant(newSize, Scalar(1)); @@ -593,7 +666,7 @@ DenseBase::Ones(Index newSize) * \sa Ones(Index), Ones(Index,Index), isOnes(), class Ones */ template -EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename DenseBase::ConstantReturnType DenseBase::Ones() { return Constant(Scalar(1)); @@ -608,7 +681,7 @@ DenseBase::Ones() * \sa class CwiseNullaryOp, Ones() */ template -bool DenseBase::isOnes +EIGEN_DEVICE_FUNC bool DenseBase::isOnes (const RealScalar& prec) const { return isApproxToConstant(Scalar(1), prec); @@ -622,7 +695,7 @@ bool DenseBase::isOnes * \sa class CwiseNullaryOp, Ones() */ template -EIGEN_STRONG_INLINE Derived& DenseBase::setOnes() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& DenseBase::setOnes() { return setConstant(Scalar(1)); } @@ -637,7 +710,7 @@ EIGEN_STRONG_INLINE Derived& DenseBase::setOnes() * \sa MatrixBase::setOnes(), setOnes(Index,Index), class CwiseNullaryOp, MatrixBase::Ones() */ template -EIGEN_STRONG_INLINE Derived& +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& PlainObjectBase::setOnes(Index newSize) { resize(newSize); @@ -646,8 +719,8 @@ PlainObjectBase::setOnes(Index newSize) /** Resizes to the given size, and sets all coefficients in this expression to one. * - * \param nbRows the new number of rows - * \param nbCols the new number of columns + * \param rows the new number of rows + * \param cols the new number of columns * * Example: \include Matrix_setOnes_int_int.cpp * Output: \verbinclude Matrix_setOnes_int_int.out @@ -655,18 +728,44 @@ PlainObjectBase::setOnes(Index newSize) * \sa MatrixBase::setOnes(), setOnes(Index), class CwiseNullaryOp, MatrixBase::Ones() */ template -EIGEN_STRONG_INLINE Derived& -PlainObjectBase::setOnes(Index nbRows, Index nbCols) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setOnes(Index rows, Index cols) { - resize(nbRows, nbCols); + resize(rows, cols); return setConstant(Scalar(1)); } +/** Resizes to the given size, changing only the number of rows, and sets all + * coefficients in this expression to one. For the parameter of type NoChange_t, + * just pass the special value \c NoChange. + * + * \sa MatrixBase::setOnes(), setOnes(Index), setOnes(Index, Index), setOnes(NoChange_t, Index), class CwiseNullaryOp, MatrixBase::Ones() + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setOnes(Index rows, NoChange_t) +{ + return setOnes(rows, cols()); +} + +/** Resizes to the given size, changing only the number of columns, and sets all + * coefficients in this expression to one. For the parameter of type NoChange_t, + * just pass the special value \c NoChange. + * + * \sa MatrixBase::setOnes(), setOnes(Index), setOnes(Index, Index), setOnes(Index, NoChange_t) class CwiseNullaryOp, MatrixBase::Ones() + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setOnes(NoChange_t, Index cols) +{ + return setOnes(rows(), cols); +} + // Identity: /** \returns an expression of the identity matrix (not necessarily square). * - * The parameters \a nbRows and \a nbCols are the number of rows and of columns of + * The parameters \a rows and \a cols are the number of rows and of columns of * the returned matrix. Must be compatible with this MatrixBase type. * * This variant is meant to be used for dynamic-size matrix types. For fixed-size types, @@ -679,10 +778,10 @@ PlainObjectBase::setOnes(Index nbRows, Index nbCols) * \sa Identity(), setIdentity(), isIdentity() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::IdentityReturnType -MatrixBase::Identity(Index nbRows, Index nbCols) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::IdentityReturnType +MatrixBase::Identity(Index rows, Index cols) { - return DenseBase::NullaryExpr(nbRows, nbCols, internal::scalar_identity_op()); + return DenseBase::NullaryExpr(rows, cols, internal::scalar_identity_op()); } /** \returns an expression of the identity matrix (not necessarily square). @@ -696,7 +795,7 @@ MatrixBase::Identity(Index nbRows, Index nbCols) * \sa Identity(Index,Index), setIdentity(), isIdentity() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::IdentityReturnType +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::IdentityReturnType MatrixBase::Identity() { EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived) @@ -716,18 +815,19 @@ template bool MatrixBase::isIdentity (const RealScalar& prec) const { + typename internal::nested_eval::type self(derived()); for(Index j = 0; j < cols(); ++j) { for(Index i = 0; i < rows(); ++i) { if(i == j) { - if(!internal::isApprox(this->coeff(i, j), static_cast(1), prec)) + if(!internal::isApprox(self.coeff(i, j), static_cast(1), prec)) return false; } else { - if(!internal::isMuchSmallerThan(this->coeff(i, j), static_cast(1), prec)) + if(!internal::isMuchSmallerThan(self.coeff(i, j), static_cast(1), prec)) return false; } } @@ -740,6 +840,7 @@ namespace internal { template=16)> struct setIdentity_impl { + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Derived& run(Derived& m) { return m = Derived::Identity(m.rows(), m.cols()); @@ -749,11 +850,11 @@ struct setIdentity_impl template struct setIdentity_impl { - typedef typename Derived::Index Index; + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE Derived& run(Derived& m) { m.setZero(); - const Index size = (std::min)(m.rows(), m.cols()); + const Index size = numext::mini(m.rows(), m.cols()); for(Index i = 0; i < size; ++i) m.coeffRef(i,i) = typename Derived::Scalar(1); return m; } @@ -769,15 +870,15 @@ struct setIdentity_impl * \sa class CwiseNullaryOp, Identity(), Identity(Index,Index), isIdentity() */ template -EIGEN_STRONG_INLINE Derived& MatrixBase::setIdentity() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::setIdentity() { return internal::setIdentity_impl::run(derived()); } /** \brief Resizes to the given size, and writes the identity expression (not necessarily square) into *this. * - * \param nbRows the new number of rows - * \param nbCols the new number of columns + * \param rows the new number of rows + * \param cols the new number of columns * * Example: \include Matrix_setIdentity_int_int.cpp * Output: \verbinclude Matrix_setIdentity_int_int.out @@ -785,9 +886,9 @@ EIGEN_STRONG_INLINE Derived& MatrixBase::setIdentity() * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Identity() */ template -EIGEN_STRONG_INLINE Derived& MatrixBase::setIdentity(Index nbRows, Index nbCols) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::setIdentity(Index rows, Index cols) { - derived().resize(nbRows, nbCols); + derived().resize(rows, cols); return setIdentity(); } @@ -798,7 +899,7 @@ EIGEN_STRONG_INLINE Derived& MatrixBase::setIdentity(Index nbRows, Inde * \sa MatrixBase::Unit(Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::Unit(Index newSize, Index i) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::Unit(Index newSize, Index i) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) return BasisReturnType(SquareMatrixType::Identity(newSize,newSize), i); @@ -813,7 +914,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBa * \sa MatrixBase::Unit(Index,Index), MatrixBase::UnitX(), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::Unit(Index i) +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::Unit(Index i) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) return BasisReturnType(SquareMatrixType::Identity(),i); @@ -826,7 +927,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBa * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitX() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitX() { return Derived::Unit(0); } /** \returns an expression of the Y axis unit vector (0,1{,0}^*) @@ -836,7 +937,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBa * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitY() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitY() { return Derived::Unit(1); } /** \returns an expression of the Z axis unit vector (0,0,1{,0}^*) @@ -846,7 +947,7 @@ EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBa * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitZ() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitZ() { return Derived::Unit(2); } /** \returns an expression of the W axis unit vector (0,0,0,1) @@ -856,9 +957,45 @@ EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBa * \sa MatrixBase::Unit(Index,Index), MatrixBase::Unit(Index), MatrixBase::UnitY(), MatrixBase::UnitZ(), MatrixBase::UnitW() */ template -EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitW() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::BasisReturnType MatrixBase::UnitW() { return Derived::Unit(3); } +/** \brief Set the coefficients of \c *this to the i-th unit (basis) vector + * + * \param i index of the unique coefficient to be set to 1 + * + * \only_for_vectors + * + * \sa MatrixBase::setIdentity(), class CwiseNullaryOp, MatrixBase::Unit(Index,Index) + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::setUnit(Index i) +{ + EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived); + eigen_assert(i +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& MatrixBase::setUnit(Index newSize, Index i) +{ + EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived); + eigen_assert(i +// Copyright (C) 2006-2008 Benoit Jacob +// Copyright (C) 2016 Eugene Brevdo +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_CWISE_TERNARY_OP_H +#define EIGEN_CWISE_TERNARY_OP_H + +namespace Eigen { + +namespace internal { +template +struct traits > { + // we must not inherit from traits since it has + // the potential to cause problems with MSVC + typedef typename remove_all::type Ancestor; + typedef typename traits::XprKind XprKind; + enum { + RowsAtCompileTime = traits::RowsAtCompileTime, + ColsAtCompileTime = traits::ColsAtCompileTime, + MaxRowsAtCompileTime = traits::MaxRowsAtCompileTime, + MaxColsAtCompileTime = traits::MaxColsAtCompileTime + }; + + // even though we require Arg1, Arg2, and Arg3 to have the same scalar type + // (see CwiseTernaryOp constructor), + // we still want to handle the case when the result type is different. + typedef typename result_of::type Scalar; + + typedef typename internal::traits::StorageKind StorageKind; + typedef typename internal::traits::StorageIndex StorageIndex; + + typedef typename Arg1::Nested Arg1Nested; + typedef typename Arg2::Nested Arg2Nested; + typedef typename Arg3::Nested Arg3Nested; + typedef typename remove_reference::type _Arg1Nested; + typedef typename remove_reference::type _Arg2Nested; + typedef typename remove_reference::type _Arg3Nested; + enum { Flags = _Arg1Nested::Flags & RowMajorBit }; +}; +} // end namespace internal + +template +class CwiseTernaryOpImpl; + +/** \class CwiseTernaryOp + * \ingroup Core_Module + * + * \brief Generic expression where a coefficient-wise ternary operator is + * applied to two expressions + * + * \tparam TernaryOp template functor implementing the operator + * \tparam Arg1Type the type of the first argument + * \tparam Arg2Type the type of the second argument + * \tparam Arg3Type the type of the third argument + * + * This class represents an expression where a coefficient-wise ternary + * operator is applied to three expressions. + * It is the return type of ternary operators, by which we mean only those + * ternary operators where + * all three arguments are Eigen expressions. + * For example, the return type of betainc(matrix1, matrix2, matrix3) is a + * CwiseTernaryOp. + * + * Most of the time, this is the only way that it is used, so you typically + * don't have to name + * CwiseTernaryOp types explicitly. + * + * \sa MatrixBase::ternaryExpr(const MatrixBase &, const + * MatrixBase &, const CustomTernaryOp &) const, class CwiseBinaryOp, + * class CwiseUnaryOp, class CwiseNullaryOp + */ +template +class CwiseTernaryOp : public CwiseTernaryOpImpl< + TernaryOp, Arg1Type, Arg2Type, Arg3Type, + typename internal::traits::StorageKind>, + internal::no_assignment_operator +{ + public: + typedef typename internal::remove_all::type Arg1; + typedef typename internal::remove_all::type Arg2; + typedef typename internal::remove_all::type Arg3; + + typedef typename CwiseTernaryOpImpl< + TernaryOp, Arg1Type, Arg2Type, Arg3Type, + typename internal::traits::StorageKind>::Base Base; + EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseTernaryOp) + + typedef typename internal::ref_selector::type Arg1Nested; + typedef typename internal::ref_selector::type Arg2Nested; + typedef typename internal::ref_selector::type Arg3Nested; + typedef typename internal::remove_reference::type _Arg1Nested; + typedef typename internal::remove_reference::type _Arg2Nested; + typedef typename internal::remove_reference::type _Arg3Nested; + + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE CwiseTernaryOp(const Arg1& a1, const Arg2& a2, + const Arg3& a3, + const TernaryOp& func = TernaryOp()) + : m_arg1(a1), m_arg2(a2), m_arg3(a3), m_functor(func) { + // require the sizes to match + EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg2) + EIGEN_STATIC_ASSERT_SAME_MATRIX_SIZE(Arg1, Arg3) + + // The index types should match + EIGEN_STATIC_ASSERT((internal::is_same< + typename internal::traits::StorageKind, + typename internal::traits::StorageKind>::value), + STORAGE_KIND_MUST_MATCH) + EIGEN_STATIC_ASSERT((internal::is_same< + typename internal::traits::StorageKind, + typename internal::traits::StorageKind>::value), + STORAGE_KIND_MUST_MATCH) + + eigen_assert(a1.rows() == a2.rows() && a1.cols() == a2.cols() && + a1.rows() == a3.rows() && a1.cols() == a3.cols()); + } + + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Index rows() const { + // return the fixed size type if available to enable compile time + // optimizations + if (internal::traits::type>:: + RowsAtCompileTime == Dynamic && + internal::traits::type>:: + RowsAtCompileTime == Dynamic) + return m_arg3.rows(); + else if (internal::traits::type>:: + RowsAtCompileTime == Dynamic && + internal::traits::type>:: + RowsAtCompileTime == Dynamic) + return m_arg2.rows(); + else + return m_arg1.rows(); + } + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Index cols() const { + // return the fixed size type if available to enable compile time + // optimizations + if (internal::traits::type>:: + ColsAtCompileTime == Dynamic && + internal::traits::type>:: + ColsAtCompileTime == Dynamic) + return m_arg3.cols(); + else if (internal::traits::type>:: + ColsAtCompileTime == Dynamic && + internal::traits::type>:: + ColsAtCompileTime == Dynamic) + return m_arg2.cols(); + else + return m_arg1.cols(); + } + + /** \returns the first argument nested expression */ + EIGEN_DEVICE_FUNC + const _Arg1Nested& arg1() const { return m_arg1; } + /** \returns the first argument nested expression */ + EIGEN_DEVICE_FUNC + const _Arg2Nested& arg2() const { return m_arg2; } + /** \returns the third argument nested expression */ + EIGEN_DEVICE_FUNC + const _Arg3Nested& arg3() const { return m_arg3; } + /** \returns the functor representing the ternary operation */ + EIGEN_DEVICE_FUNC + const TernaryOp& functor() const { return m_functor; } + + protected: + Arg1Nested m_arg1; + Arg2Nested m_arg2; + Arg3Nested m_arg3; + const TernaryOp m_functor; +}; + +// Generic API dispatcher +template +class CwiseTernaryOpImpl + : public internal::generic_xpr_base< + CwiseTernaryOp >::type { + public: + typedef typename internal::generic_xpr_base< + CwiseTernaryOp >::type Base; +}; + +} // end namespace Eigen + +#endif // EIGEN_CWISE_TERNARY_OP_H diff --git a/thirdparty/eigen/Eigen/src/Core/CwiseUnaryOp.h b/thirdparty/eigen/Eigen/src/Core/CwiseUnaryOp.h index f7ee60e9..e68c4f74 100644 --- a/thirdparty/eigen/Eigen/src/Core/CwiseUnaryOp.h +++ b/thirdparty/eigen/Eigen/src/Core/CwiseUnaryOp.h @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2008-2010 Gael Guennebaud +// Copyright (C) 2008-2014 Gael Guennebaud // Copyright (C) 2006-2008 Benoit Jacob // // This Source Code Form is subject to the terms of the Mozilla @@ -11,27 +11,7 @@ #ifndef EIGEN_CWISE_UNARY_OP_H #define EIGEN_CWISE_UNARY_OP_H -namespace Eigen { - -/** \class CwiseUnaryOp - * \ingroup Core_Module - * - * \brief Generic expression where a coefficient-wise unary operator is applied to an expression - * - * \param UnaryOp template functor implementing the operator - * \param XprType the type of the expression to which we are applying the unary operator - * - * This class represents an expression where a unary operator is applied to an expression. - * It is the return type of all operations taking exactly 1 input expression, regardless of the - * presence of other inputs such as scalars. For example, the operator* in the expression 3*matrix - * is considered unary, because only the right-hand side is an expression, and its - * return type is a specialization of CwiseUnaryOp. - * - * Most of the time, this is the only way that it is used, so you typically don't have to name - * CwiseUnaryOp types explicitly. - * - * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp - */ +namespace Eigen { namespace internal { template @@ -39,15 +19,12 @@ struct traits > : traits { typedef typename result_of< - UnaryOp(typename XprType::Scalar) + UnaryOp(const typename XprType::Scalar&) >::type Scalar; typedef typename XprType::Nested XprTypeNested; typedef typename remove_reference::type _XprTypeNested; enum { - Flags = _XprTypeNested::Flags & ( - HereditaryBits | LinearAccessBit | AlignedBit - | (functor_traits::PacketAccess ? PacketAccessBit : 0)), - CoeffReadCost = EIGEN_ADD_COST(_XprTypeNested::CoeffReadCost, functor_traits::Cost) + Flags = _XprTypeNested::Flags & RowMajorBit }; }; } @@ -55,70 +32,70 @@ struct traits > template class CwiseUnaryOpImpl; +/** \class CwiseUnaryOp + * \ingroup Core_Module + * + * \brief Generic expression where a coefficient-wise unary operator is applied to an expression + * + * \tparam UnaryOp template functor implementing the operator + * \tparam XprType the type of the expression to which we are applying the unary operator + * + * This class represents an expression where a unary operator is applied to an expression. + * It is the return type of all operations taking exactly 1 input expression, regardless of the + * presence of other inputs such as scalars. For example, the operator* in the expression 3*matrix + * is considered unary, because only the right-hand side is an expression, and its + * return type is a specialization of CwiseUnaryOp. + * + * Most of the time, this is the only way that it is used, so you typically don't have to name + * CwiseUnaryOp types explicitly. + * + * \sa MatrixBase::unaryExpr(const CustomUnaryOp &) const, class CwiseBinaryOp, class CwiseNullaryOp + */ template -class CwiseUnaryOp : internal::no_assignment_operator, - public CwiseUnaryOpImpl::StorageKind> +class CwiseUnaryOp : public CwiseUnaryOpImpl::StorageKind>, internal::no_assignment_operator { public: typedef typename CwiseUnaryOpImpl::StorageKind>::Base Base; EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryOp) + typedef typename internal::ref_selector::type XprTypeNested; + typedef typename internal::remove_all::type NestedExpression; - inline CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp()) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit CwiseUnaryOp(const XprType& xpr, const UnaryOp& func = UnaryOp()) : m_xpr(xpr), m_functor(func) {} - EIGEN_STRONG_INLINE Index rows() const { return m_xpr.rows(); } - EIGEN_STRONG_INLINE Index cols() const { return m_xpr.cols(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rows() const EIGEN_NOEXCEPT { return m_xpr.rows(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index cols() const EIGEN_NOEXCEPT { return m_xpr.cols(); } /** \returns the functor representing the unary operation */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const UnaryOp& functor() const { return m_functor; } /** \returns the nested expression */ - const typename internal::remove_all::type& + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const typename internal::remove_all::type& nestedExpression() const { return m_xpr; } /** \returns the nested expression */ - typename internal::remove_all::type& - nestedExpression() { return m_xpr.const_cast_derived(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + typename internal::remove_all::type& + nestedExpression() { return m_xpr; } protected: - typename XprType::Nested m_xpr; + XprTypeNested m_xpr; const UnaryOp m_functor; }; -// This is the generic implementation for dense storage. -// It can be used for any expression types implementing the dense concept. -template -class CwiseUnaryOpImpl - : public internal::dense_xpr_base >::type +// Generic API dispatcher +template +class CwiseUnaryOpImpl + : public internal::generic_xpr_base >::type { - public: - - typedef CwiseUnaryOp Derived; - typedef typename internal::dense_xpr_base >::type Base; - EIGEN_DENSE_PUBLIC_INTERFACE(Derived) - - EIGEN_STRONG_INLINE const Scalar coeff(Index rowId, Index colId) const - { - return derived().functor()(derived().nestedExpression().coeff(rowId, colId)); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index rowId, Index colId) const - { - return derived().functor().packetOp(derived().nestedExpression().template packet(rowId, colId)); - } - - EIGEN_STRONG_INLINE const Scalar coeff(Index index) const - { - return derived().functor()(derived().nestedExpression().coeff(index)); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index index) const - { - return derived().functor().packetOp(derived().nestedExpression().template packet(index)); - } +public: + typedef typename internal::generic_xpr_base >::type Base; }; } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/CwiseUnaryView.h b/thirdparty/eigen/Eigen/src/Core/CwiseUnaryView.h index f3b2ffeb..a06d7621 100644 --- a/thirdparty/eigen/Eigen/src/Core/CwiseUnaryView.h +++ b/thirdparty/eigen/Eigen/src/Core/CwiseUnaryView.h @@ -12,33 +12,19 @@ namespace Eigen { -/** \class CwiseUnaryView - * \ingroup Core_Module - * - * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector - * - * \param ViewOp template functor implementing the view - * \param MatrixType the type of the matrix we are applying the unary operator - * - * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector. - * It is the return type of real() and imag(), and most of the time this is the only way it is used. - * - * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp - */ - namespace internal { template struct traits > : traits { typedef typename result_of< - ViewOp(typename traits::Scalar) + ViewOp(const typename traits::Scalar&) >::type Scalar; typedef typename MatrixType::Nested MatrixTypeNested; typedef typename remove_all::type _MatrixTypeNested; enum { - Flags = (traits<_MatrixTypeNested>::Flags & (HereditaryBits | LvalueBit | LinearAccessBit | DirectAccessBit)), - CoeffReadCost = EIGEN_ADD_COST(traits<_MatrixTypeNested>::CoeffReadCost, functor_traits::Cost), + FlagsLvalueBit = is_lvalue::value ? LvalueBit : 0, + Flags = traits<_MatrixTypeNested>::Flags & (RowMajorBit | FlagsLvalueBit | DirectAccessBit), // FIXME DirectAccessBit should not be handled by expressions MatrixTypeInnerStride = inner_stride_at_compile_time::ret, // need to cast the sizeof's from size_t to int explicitly, otherwise: // "error: no integral type can represent all of the enumerator values @@ -55,6 +41,19 @@ struct traits > template class CwiseUnaryViewImpl; +/** \class CwiseUnaryView + * \ingroup Core_Module + * + * \brief Generic lvalue expression of a coefficient-wise unary operator of a matrix or a vector + * + * \tparam ViewOp template functor implementing the view + * \tparam MatrixType the type of the matrix we are applying the unary operator + * + * This class represents a lvalue expression of a generic unary view operator of a matrix or a vector. + * It is the return type of real() and imag(), and most of the time this is the only way it is used. + * + * \sa MatrixBase::unaryViewExpr(const CustomUnaryOp &) const, class CwiseUnaryOp + */ template class CwiseUnaryView : public CwiseUnaryViewImpl::StorageKind> { @@ -62,32 +61,44 @@ class CwiseUnaryView : public CwiseUnaryViewImpl::StorageKind>::Base Base; EIGEN_GENERIC_PUBLIC_INTERFACE(CwiseUnaryView) + typedef typename internal::ref_selector::non_const_type MatrixTypeNested; + typedef typename internal::remove_all::type NestedExpression; - inline CwiseUnaryView(const MatrixType& mat, const ViewOp& func = ViewOp()) + explicit EIGEN_DEVICE_FUNC inline CwiseUnaryView(MatrixType& mat, const ViewOp& func = ViewOp()) : m_matrix(mat), m_functor(func) {} EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryView) - EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); } - EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); } /** \returns the functor representing unary operation */ - const ViewOp& functor() const { return m_functor; } + EIGEN_DEVICE_FUNC const ViewOp& functor() const { return m_functor; } /** \returns the nested expression */ - const typename internal::remove_all::type& + EIGEN_DEVICE_FUNC const typename internal::remove_all::type& nestedExpression() const { return m_matrix; } /** \returns the nested expression */ - typename internal::remove_all::type& - nestedExpression() { return m_matrix.const_cast_derived(); } + EIGEN_DEVICE_FUNC typename internal::remove_reference::type& + nestedExpression() { return m_matrix; } protected: - // FIXME changed from MatrixType::Nested because of a weird compilation error with sun CC - typename internal::nested::type m_matrix; + MatrixTypeNested m_matrix; ViewOp m_functor; }; +// Generic API dispatcher +template +class CwiseUnaryViewImpl + : public internal::generic_xpr_base >::type +{ +public: + typedef typename internal::generic_xpr_base >::type Base; +}; + template class CwiseUnaryViewImpl : public internal::dense_xpr_base< CwiseUnaryView >::type @@ -99,39 +110,21 @@ class CwiseUnaryViewImpl EIGEN_DENSE_PUBLIC_INTERFACE(Derived) EIGEN_INHERIT_ASSIGNMENT_OPERATORS(CwiseUnaryViewImpl) - - inline Scalar* data() { return &coeffRef(0); } - inline const Scalar* data() const { return &coeff(0); } - inline Index innerStride() const + EIGEN_DEVICE_FUNC inline Scalar* data() { return &(this->coeffRef(0)); } + EIGEN_DEVICE_FUNC inline const Scalar* data() const { return &(this->coeff(0)); } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const { return derived().nestedExpression().innerStride() * sizeof(typename internal::traits::Scalar) / sizeof(Scalar); } - inline Index outerStride() const + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const { return derived().nestedExpression().outerStride() * sizeof(typename internal::traits::Scalar) / sizeof(Scalar); } - - EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const - { - return derived().functor()(derived().nestedExpression().coeff(row, col)); - } - - EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const - { - return derived().functor()(derived().nestedExpression().coeff(index)); - } - - EIGEN_STRONG_INLINE Scalar& coeffRef(Index row, Index col) - { - return derived().functor()(const_cast_derived().nestedExpression().coeffRef(row, col)); - } - - EIGEN_STRONG_INLINE Scalar& coeffRef(Index index) - { - return derived().functor()(const_cast_derived().nestedExpression().coeffRef(index)); - } + protected: + EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(CwiseUnaryViewImpl) }; } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/DenseBase.h b/thirdparty/eigen/Eigen/src/Core/DenseBase.h index 4b371b07..cdd0f5f1 100644 --- a/thirdparty/eigen/Eigen/src/Core/DenseBase.h +++ b/thirdparty/eigen/Eigen/src/Core/DenseBase.h @@ -14,15 +14,15 @@ namespace Eigen { namespace internal { - + // The index type defined by EIGEN_DEFAULT_DENSE_INDEX_TYPE must be a signed type. // This dummy function simply aims at checking that at compile time. static inline void check_DenseIndex_is_signed() { - EIGEN_STATIC_ASSERT(NumTraits::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE); + EIGEN_STATIC_ASSERT(NumTraits::IsSigned,THE_INDEX_TYPE_MUST_BE_A_SIGNED_TYPE) } } // end namespace internal - + /** \class DenseBase * \ingroup Core_Module * @@ -34,37 +34,45 @@ static inline void check_DenseIndex_is_signed() { * \tparam Derived is the derived type, e.g., a matrix type or an expression. * * This class can be extended with the help of the plugin mechanism described on the page - * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN. + * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_DENSEBASE_PLUGIN. * - * \sa \ref TopicClassHierarchy + * \sa \blank \ref TopicClassHierarchy */ template class DenseBase #ifndef EIGEN_PARSED_BY_DOXYGEN - : public internal::special_scalar_op_base::Scalar, - typename NumTraits::Scalar>::Real, - DenseCoeffsBase > + : public DenseCoeffsBase::value> #else - : public DenseCoeffsBase + : public DenseCoeffsBase #endif // not EIGEN_PARSED_BY_DOXYGEN { public: - class InnerIterator; + /** Inner iterator type to iterate over the coefficients of a row or column. + * \sa class InnerIterator + */ + typedef Eigen::InnerIterator InnerIterator; typedef typename internal::traits::StorageKind StorageKind; - /** \brief The type of indices - * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE. - * \sa \ref TopicPreprocessorDirectives. - */ - typedef typename internal::traits::Index Index; + /** + * \brief The type used to store indices + * \details This typedef is relevant for types that store multiple indices such as + * PermutationMatrix or Transpositions, otherwise it defaults to Eigen::Index + * \sa \blank \ref TopicPreprocessorDirectives, Eigen::Index, SparseMatrixBase. + */ + typedef typename internal::traits::StorageIndex StorageIndex; + /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex, etc. */ typedef typename internal::traits::Scalar Scalar; - typedef typename internal::packet_traits::type PacketScalar; + + /** The numeric type of the expression' coefficients, e.g. float, double, int or std::complex, etc. + * + * It is an alias for the Scalar type */ + typedef Scalar value_type; + typedef typename NumTraits::Real RealScalar; - typedef internal::special_scalar_op_base > Base; + typedef DenseCoeffsBase::value> Base; - using Base::operator*; using Base::derived; using Base::const_cast_derived; using Base::rows; @@ -74,16 +82,6 @@ template class DenseBase using Base::colIndexByOuterInner; using Base::coeff; using Base::coeffByOuterInner; - using Base::packet; - using Base::packetByOuterInner; - using Base::writePacket; - using Base::writePacketByOuterInner; - using Base::coeffRef; - using Base::coeffRefByOuterInner; - using Base::copyCoeff; - using Base::copyCoeffByOuterInner; - using Base::copyPacket; - using Base::copyPacketByOuterInner; using Base::operator(); using Base::operator[]; using Base::x; @@ -152,13 +150,18 @@ template class DenseBase * \sa SizeAtCompileTime, MaxRowsAtCompileTime, MaxColsAtCompileTime */ - IsVectorAtCompileTime = internal::traits::MaxRowsAtCompileTime == 1 - || internal::traits::MaxColsAtCompileTime == 1, + IsVectorAtCompileTime = internal::traits::RowsAtCompileTime == 1 + || internal::traits::ColsAtCompileTime == 1, /**< This is set to true if either the number of rows or the number of * columns is known at compile-time to be equal to 1. Indeed, in that case, * we are dealing with a column-vector (if there is only one column) or with * a row-vector (if there is only one row). */ + NumDimensions = int(MaxSizeAtCompileTime) == 1 ? 0 : bool(IsVectorAtCompileTime) ? 1 : 2, + /**< This value is equal to Tensor::NumDimensions, i.e. 0 for scalars, 1 for vectors, + * and 2 for matrices. + */ + Flags = internal::traits::Flags, /**< This stores expression \ref flags flags which may or may not be inherited by new expressions * constructed from this one. See the \ref flags "list of flags". @@ -169,19 +172,46 @@ template class DenseBase InnerSizeAtCompileTime = int(IsVectorAtCompileTime) ? int(SizeAtCompileTime) : int(IsRowMajor) ? int(ColsAtCompileTime) : int(RowsAtCompileTime), - CoeffReadCost = internal::traits::CoeffReadCost, - /**< This is a rough measure of how expensive it is to read one coefficient from - * this expression. - */ - InnerStrideAtCompileTime = internal::inner_stride_at_compile_time::ret, OuterStrideAtCompileTime = internal::outer_stride_at_compile_time::ret }; - enum { ThisConstantIsPrivateInPlainObjectBase }; + typedef typename internal::find_best_packet::type PacketScalar; + + enum { IsPlainObjectBase = 0 }; + + /** The plain matrix type corresponding to this expression. + * \sa PlainObject */ + typedef Matrix::Scalar, + internal::traits::RowsAtCompileTime, + internal::traits::ColsAtCompileTime, + AutoAlign | (internal::traits::Flags&RowMajorBit ? RowMajor : ColMajor), + internal::traits::MaxRowsAtCompileTime, + internal::traits::MaxColsAtCompileTime + > PlainMatrix; + + /** The plain array type corresponding to this expression. + * \sa PlainObject */ + typedef Array::Scalar, + internal::traits::RowsAtCompileTime, + internal::traits::ColsAtCompileTime, + AutoAlign | (internal::traits::Flags&RowMajorBit ? RowMajor : ColMajor), + internal::traits::MaxRowsAtCompileTime, + internal::traits::MaxColsAtCompileTime + > PlainArray; + + /** \brief The plain matrix or array type corresponding to this expression. + * + * This is not necessarily exactly the return type of eval(). In the case of plain matrices, + * the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed + * that the return type of eval() is either PlainObject or const PlainObject&. + */ + typedef typename internal::conditional::XprKind,MatrixXpr >::value, + PlainMatrix, PlainArray>::type PlainObject; /** \returns the number of nonzero coefficients which is in practice the number * of stored coefficients. */ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index nonZeros() const { return size(); } /** \returns the outer size. @@ -189,6 +219,7 @@ template class DenseBase * \note For a vector, this returns just 1. For a matrix (non-vector), this is the major dimension * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of columns for a * column-major matrix, and the number of rows for a row-major matrix. */ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index outerSize() const { return IsVectorAtCompileTime ? 1 @@ -198,8 +229,9 @@ template class DenseBase /** \returns the inner size. * * \note For a vector, this is just the size. For a matrix (non-vector), this is the minor dimension - * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a + * with respect to the \ref TopicStorageOrders "storage order", i.e., the number of rows for a * column-major matrix, and the number of columns for a row-major matrix. */ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index innerSize() const { return IsVectorAtCompileTime ? this->size() @@ -210,6 +242,7 @@ template class DenseBase * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does * nothing else. */ + EIGEN_DEVICE_FUNC void resize(Index newSize) { EIGEN_ONLY_USED_FOR_DEBUG(newSize); @@ -220,22 +253,22 @@ template class DenseBase * Matrix::resize() and Array::resize(). The present method only asserts that the new size equals the old size, and does * nothing else. */ - void resize(Index nbRows, Index nbCols) + EIGEN_DEVICE_FUNC + void resize(Index rows, Index cols) { - EIGEN_ONLY_USED_FOR_DEBUG(nbRows); - EIGEN_ONLY_USED_FOR_DEBUG(nbCols); - eigen_assert(nbRows == this->rows() && nbCols == this->cols() + EIGEN_ONLY_USED_FOR_DEBUG(rows); + EIGEN_ONLY_USED_FOR_DEBUG(cols); + eigen_assert(rows == this->rows() && cols == this->cols() && "DenseBase::resize() does not actually allow to resize."); } #ifndef EIGEN_PARSED_BY_DOXYGEN - /** \internal Represents a matrix with all coefficients equal to one another*/ - typedef CwiseNullaryOp,Derived> ConstantReturnType; - /** \internal Represents a vector with linearly spaced coefficients that allows sequential access only. */ - typedef CwiseNullaryOp,Derived> SequentialLinSpacedReturnType; + typedef CwiseNullaryOp,PlainObject> ConstantReturnType; + /** \internal \deprecated Represents a vector with linearly spaced coefficients that allows sequential access only. */ + EIGEN_DEPRECATED typedef CwiseNullaryOp,PlainObject> SequentialLinSpacedReturnType; /** \internal Represents a vector with linearly spaced coefficients that allows random access. */ - typedef CwiseNullaryOp,Derived> RandomAccessLinSpacedReturnType; + typedef CwiseNullaryOp,PlainObject> RandomAccessLinSpacedReturnType; /** \internal the return type of MatrixBase::eigenvalues() */ typedef Matrix::Scalar>::Real, internal::traits::ColsAtCompileTime, 1> EigenvaluesReturnType; @@ -243,120 +276,134 @@ template class DenseBase /** Copies \a other into *this. \returns a reference to *this. */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const DenseBase& other); /** Special case of the template operator=, in order to prevent the compiler * from generating a default operator= (issue hit with g++ 4.1) */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const DenseBase& other); template + EIGEN_DEVICE_FUNC Derived& operator=(const EigenBase &other); template + EIGEN_DEVICE_FUNC Derived& operator+=(const EigenBase &other); template + EIGEN_DEVICE_FUNC Derived& operator-=(const EigenBase &other); template + EIGEN_DEVICE_FUNC Derived& operator=(const ReturnByValue& func); - /** \internal Copies \a other into *this without evaluating other. \returns a reference to *this. */ + /** \internal + * Copies \a other into *this without evaluating other. \returns a reference to *this. */ template + /** \deprecated */ + EIGEN_DEPRECATED EIGEN_DEVICE_FUNC Derived& lazyAssign(const DenseBase& other); - /** \internal Evaluates \a other into *this. \returns a reference to *this. */ - template - Derived& lazyAssign(const ReturnByValue& other); - + EIGEN_DEVICE_FUNC CommaInitializer operator<< (const Scalar& s); template - const Flagged flagged() const; + /** \deprecated it now returns \c *this */ + EIGEN_DEPRECATED + const Derived& flagged() const + { return derived(); } template + EIGEN_DEVICE_FUNC CommaInitializer operator<< (const DenseBase& other); - Eigen::Transpose transpose(); - typedef typename internal::add_const >::type ConstTransposeReturnType; - ConstTransposeReturnType transpose() const; + typedef Transpose TransposeReturnType; + EIGEN_DEVICE_FUNC + TransposeReturnType transpose(); + typedef Transpose ConstTransposeReturnType; + EIGEN_DEVICE_FUNC + const ConstTransposeReturnType transpose() const; + EIGEN_DEVICE_FUNC void transposeInPlace(); -#ifndef EIGEN_NO_DEBUG - protected: - template - void checkTransposeAliasing(const OtherDerived& other) const; - public: -#endif - - static const ConstantReturnType + EIGEN_DEVICE_FUNC static const ConstantReturnType Constant(Index rows, Index cols, const Scalar& value); - static const ConstantReturnType + EIGEN_DEVICE_FUNC static const ConstantReturnType Constant(Index size, const Scalar& value); - static const ConstantReturnType + EIGEN_DEVICE_FUNC static const ConstantReturnType Constant(const Scalar& value); - static const SequentialLinSpacedReturnType + EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType LinSpaced(Sequential_t, Index size, const Scalar& low, const Scalar& high); - static const RandomAccessLinSpacedReturnType - LinSpaced(Index size, const Scalar& low, const Scalar& high); - static const SequentialLinSpacedReturnType + EIGEN_DEPRECATED EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType LinSpaced(Sequential_t, const Scalar& low, const Scalar& high); - static const RandomAccessLinSpacedReturnType + + EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType + LinSpaced(Index size, const Scalar& low, const Scalar& high); + EIGEN_DEVICE_FUNC static const RandomAccessLinSpacedReturnType LinSpaced(const Scalar& low, const Scalar& high); - template - static const CwiseNullaryOp + template EIGEN_DEVICE_FUNC + static const CwiseNullaryOp NullaryExpr(Index rows, Index cols, const CustomNullaryOp& func); - template - static const CwiseNullaryOp + template EIGEN_DEVICE_FUNC + static const CwiseNullaryOp NullaryExpr(Index size, const CustomNullaryOp& func); - template - static const CwiseNullaryOp + template EIGEN_DEVICE_FUNC + static const CwiseNullaryOp NullaryExpr(const CustomNullaryOp& func); - static const ConstantReturnType Zero(Index rows, Index cols); - static const ConstantReturnType Zero(Index size); - static const ConstantReturnType Zero(); - static const ConstantReturnType Ones(Index rows, Index cols); - static const ConstantReturnType Ones(Index size); - static const ConstantReturnType Ones(); - - void fill(const Scalar& value); - Derived& setConstant(const Scalar& value); - Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high); - Derived& setLinSpaced(const Scalar& low, const Scalar& high); - Derived& setZero(); - Derived& setOnes(); - Derived& setRandom(); - - template + EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index rows, Index cols); + EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(Index size); + EIGEN_DEVICE_FUNC static const ConstantReturnType Zero(); + EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index rows, Index cols); + EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(Index size); + EIGEN_DEVICE_FUNC static const ConstantReturnType Ones(); + + EIGEN_DEVICE_FUNC void fill(const Scalar& value); + EIGEN_DEVICE_FUNC Derived& setConstant(const Scalar& value); + EIGEN_DEVICE_FUNC Derived& setLinSpaced(Index size, const Scalar& low, const Scalar& high); + EIGEN_DEVICE_FUNC Derived& setLinSpaced(const Scalar& low, const Scalar& high); + EIGEN_DEVICE_FUNC Derived& setZero(); + EIGEN_DEVICE_FUNC Derived& setOnes(); + EIGEN_DEVICE_FUNC Derived& setRandom(); + + template EIGEN_DEVICE_FUNC bool isApprox(const DenseBase& other, const RealScalar& prec = NumTraits::dummy_precision()) const; + EIGEN_DEVICE_FUNC bool isMuchSmallerThan(const RealScalar& other, const RealScalar& prec = NumTraits::dummy_precision()) const; - template + template EIGEN_DEVICE_FUNC bool isMuchSmallerThan(const DenseBase& other, const RealScalar& prec = NumTraits::dummy_precision()) const; - bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits::dummy_precision()) const; - bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits::dummy_precision()) const; - bool isZero(const RealScalar& prec = NumTraits::dummy_precision()) const; - bool isOnes(const RealScalar& prec = NumTraits::dummy_precision()) const; - + EIGEN_DEVICE_FUNC bool isApproxToConstant(const Scalar& value, const RealScalar& prec = NumTraits::dummy_precision()) const; + EIGEN_DEVICE_FUNC bool isConstant(const Scalar& value, const RealScalar& prec = NumTraits::dummy_precision()) const; + EIGEN_DEVICE_FUNC bool isZero(const RealScalar& prec = NumTraits::dummy_precision()) const; + EIGEN_DEVICE_FUNC bool isOnes(const RealScalar& prec = NumTraits::dummy_precision()) const; + inline bool hasNaN() const; inline bool allFinite() const; - inline Derived& operator*=(const Scalar& other); - inline Derived& operator/=(const Scalar& other); + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Derived& operator*=(const Scalar& other); + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Derived& operator/=(const Scalar& other); typedef typename internal::add_const_on_value_type::type>::type EvalReturnType; /** \returns the matrix or vector obtained by evaluating this expression. * * Notice that in the case of a plain matrix or vector (not an expression) this function just returns * a const reference, in order to avoid a useless copy. + * + * \warning Be careful with eval() and the auto C++ keyword, as detailed in this \link TopicPitfalls_auto_keyword page \endlink. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EvalReturnType eval() const { // Even though MSVC does not honor strong inlining when the return type @@ -369,56 +416,113 @@ template class DenseBase * */ template - void swap(const DenseBase& other, - int = OtherDerived::ThisConstantIsPrivateInPlainObjectBase) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void swap(const DenseBase& other) { - SwapWrapper(derived()).lazyAssign(other.derived()); + EIGEN_STATIC_ASSERT(!OtherDerived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY); + eigen_assert(rows()==other.rows() && cols()==other.cols()); + call_assignment(derived(), other.const_cast_derived(), internal::swap_assign_op()); } /** swaps *this with the matrix or array \a other. * */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void swap(PlainObjectBase& other) { - SwapWrapper(derived()).lazyAssign(other.derived()); + eigen_assert(rows()==other.rows() && cols()==other.cols()); + call_assignment(derived(), other.derived(), internal::swap_assign_op()); } + EIGEN_DEVICE_FUNC inline const NestByValue nestByValue() const; + EIGEN_DEVICE_FUNC inline const ForceAlignedAccess forceAlignedAccess() const; + EIGEN_DEVICE_FUNC inline ForceAlignedAccess forceAlignedAccess(); + template EIGEN_DEVICE_FUNC + inline const typename internal::conditional,Derived&>::type forceAlignedAccessIf() const; + template EIGEN_DEVICE_FUNC + inline typename internal::conditional,Derived&>::type forceAlignedAccessIf(); - inline const NestByValue nestByValue() const; - inline const ForceAlignedAccess forceAlignedAccess() const; - inline ForceAlignedAccess forceAlignedAccess(); - template inline const typename internal::conditional,Derived&>::type forceAlignedAccessIf() const; - template inline typename internal::conditional,Derived&>::type forceAlignedAccessIf(); + EIGEN_DEVICE_FUNC Scalar sum() const; + EIGEN_DEVICE_FUNC Scalar mean() const; + EIGEN_DEVICE_FUNC Scalar trace() const; - Scalar sum() const; - Scalar mean() const; - Scalar trace() const; + EIGEN_DEVICE_FUNC Scalar prod() const; - Scalar prod() const; + template + EIGEN_DEVICE_FUNC typename internal::traits::Scalar minCoeff() const; + template + EIGEN_DEVICE_FUNC typename internal::traits::Scalar maxCoeff() const; - typename internal::traits::Scalar minCoeff() const; - typename internal::traits::Scalar maxCoeff() const; - template + // By default, the fastest version with undefined NaN propagation semantics is + // used. + // TODO(rmlarsen): Replace with default template argument when we move to + // c++11 or beyond. + EIGEN_DEVICE_FUNC inline typename internal::traits::Scalar minCoeff() const { + return minCoeff(); + } + EIGEN_DEVICE_FUNC inline typename internal::traits::Scalar maxCoeff() const { + return maxCoeff(); + } + + template + EIGEN_DEVICE_FUNC typename internal::traits::Scalar minCoeff(IndexType* row, IndexType* col) const; - template + template + EIGEN_DEVICE_FUNC typename internal::traits::Scalar maxCoeff(IndexType* row, IndexType* col) const; - template + template + EIGEN_DEVICE_FUNC typename internal::traits::Scalar minCoeff(IndexType* index) const; - template + template + EIGEN_DEVICE_FUNC typename internal::traits::Scalar maxCoeff(IndexType* index) const; + // TODO(rmlarsen): Replace these methods with a default template argument. + template + EIGEN_DEVICE_FUNC inline + typename internal::traits::Scalar minCoeff(IndexType* row, IndexType* col) const { + return minCoeff(row, col); + } + template + EIGEN_DEVICE_FUNC inline + typename internal::traits::Scalar maxCoeff(IndexType* row, IndexType* col) const { + return maxCoeff(row, col); + } + template + EIGEN_DEVICE_FUNC inline + typename internal::traits::Scalar minCoeff(IndexType* index) const { + return minCoeff(index); + } + template + EIGEN_DEVICE_FUNC inline + typename internal::traits::Scalar maxCoeff(IndexType* index) const { + return maxCoeff(index); + } + template - typename internal::result_of::Scalar)>::type - redux(const BinaryOp& func) const; + EIGEN_DEVICE_FUNC + Scalar redux(const BinaryOp& func) const; template + EIGEN_DEVICE_FUNC void visit(Visitor& func) const; - inline const WithFormat format(const IOFormat& fmt) const; + /** \returns a WithFormat proxy object allowing to print a matrix the with given + * format \a fmt. + * + * See class IOFormat for some examples. + * + * \sa class IOFormat, class WithFormat + */ + inline const WithFormat format(const IOFormat& fmt) const + { + return WithFormat(derived(), fmt); + } /** \returns the unique coefficient of a 1x1 expression */ + EIGEN_DEVICE_FUNC CoeffReturnType value() const { EIGEN_STATIC_ASSERT_SIZE_1x1(Derived) @@ -426,82 +530,158 @@ template class DenseBase return derived().coeff(0,0); } - bool all(void) const; - bool any(void) const; - Index count() const; + EIGEN_DEVICE_FUNC bool all() const; + EIGEN_DEVICE_FUNC bool any() const; + EIGEN_DEVICE_FUNC Index count() const; typedef VectorwiseOp RowwiseReturnType; typedef const VectorwiseOp ConstRowwiseReturnType; typedef VectorwiseOp ColwiseReturnType; typedef const VectorwiseOp ConstColwiseReturnType; - ConstRowwiseReturnType rowwise() const; - RowwiseReturnType rowwise(); - ConstColwiseReturnType colwise() const; - ColwiseReturnType colwise(); + /** \returns a VectorwiseOp wrapper of *this for broadcasting and partial reductions + * + * Example: \include MatrixBase_rowwise.cpp + * Output: \verbinclude MatrixBase_rowwise.out + * + * \sa colwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting + */ + //Code moved here due to a CUDA compiler bug + EIGEN_DEVICE_FUNC inline ConstRowwiseReturnType rowwise() const { + return ConstRowwiseReturnType(derived()); + } + EIGEN_DEVICE_FUNC RowwiseReturnType rowwise(); + + /** \returns a VectorwiseOp wrapper of *this broadcasting and partial reductions + * + * Example: \include MatrixBase_colwise.cpp + * Output: \verbinclude MatrixBase_colwise.out + * + * \sa rowwise(), class VectorwiseOp, \ref TutorialReductionsVisitorsBroadcasting + */ + EIGEN_DEVICE_FUNC inline ConstColwiseReturnType colwise() const { + return ConstColwiseReturnType(derived()); + } + EIGEN_DEVICE_FUNC ColwiseReturnType colwise(); - static const CwiseNullaryOp,Derived> Random(Index rows, Index cols); - static const CwiseNullaryOp,Derived> Random(Index size); - static const CwiseNullaryOp,Derived> Random(); + typedef CwiseNullaryOp,PlainObject> RandomReturnType; + static const RandomReturnType Random(Index rows, Index cols); + static const RandomReturnType Random(Index size); + static const RandomReturnType Random(); template - const Select + inline EIGEN_DEVICE_FUNC const Select select(const DenseBase& thenMatrix, const DenseBase& elseMatrix) const; template - inline const Select + inline EIGEN_DEVICE_FUNC const Select select(const DenseBase& thenMatrix, const typename ThenDerived::Scalar& elseScalar) const; template - inline const Select + inline EIGEN_DEVICE_FUNC const Select select(const typename ElseDerived::Scalar& thenScalar, const DenseBase& elseMatrix) const; template RealScalar lpNorm() const; template - inline const Replicate replicate() const; - - typedef Replicate ReplicateReturnType; - inline const ReplicateReturnType replicate(Index rowFacor,Index colFactor) const; + EIGEN_DEVICE_FUNC + const Replicate replicate() const; + /** + * \return an expression of the replication of \c *this + * + * Example: \include MatrixBase_replicate_int_int.cpp + * Output: \verbinclude MatrixBase_replicate_int_int.out + * + * \sa VectorwiseOp::replicate(), DenseBase::replicate(), class Replicate + */ + //Code moved here due to a CUDA compiler bug + EIGEN_DEVICE_FUNC + const Replicate replicate(Index rowFactor, Index colFactor) const + { + return Replicate(derived(), rowFactor, colFactor); + } typedef Reverse ReverseReturnType; typedef const Reverse ConstReverseReturnType; - ReverseReturnType reverse(); - ConstReverseReturnType reverse() const; - void reverseInPlace(); + EIGEN_DEVICE_FUNC ReverseReturnType reverse(); + /** This is the const version of reverse(). */ + //Code moved here due to a CUDA compiler bug + EIGEN_DEVICE_FUNC ConstReverseReturnType reverse() const + { + return ConstReverseReturnType(derived()); + } + EIGEN_DEVICE_FUNC void reverseInPlace(); + + #ifdef EIGEN_PARSED_BY_DOXYGEN + /** STL-like RandomAccessIterator + * iterator type as returned by the begin() and end() methods. + */ + typedef random_access_iterator_type iterator; + /** This is the const version of iterator (aka read-only) */ + typedef random_access_iterator_type const_iterator; + #else + typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit, + internal::pointer_based_stl_iterator, + internal::generic_randaccess_stl_iterator + >::type iterator_type; + + typedef typename internal::conditional< (Flags&DirectAccessBit)==DirectAccessBit, + internal::pointer_based_stl_iterator, + internal::generic_randaccess_stl_iterator + >::type const_iterator_type; + + // Stl-style iterators are supported only for vectors. + + typedef typename internal::conditional< IsVectorAtCompileTime, + iterator_type, + void + >::type iterator; + + typedef typename internal::conditional< IsVectorAtCompileTime, + const_iterator_type, + void + >::type const_iterator; + #endif + + inline iterator begin(); + inline const_iterator begin() const; + inline const_iterator cbegin() const; + inline iterator end(); + inline const_iterator end() const; + inline const_iterator cend() const; #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::DenseBase +#define EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL +#define EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF(COND) +#define EIGEN_DOC_UNARY_ADDONS(X,Y) +# include "../plugins/CommonCwiseUnaryOps.h" # include "../plugins/BlockMethods.h" +# include "../plugins/IndexedViewMethods.h" +# include "../plugins/ReshapedMethods.h" # ifdef EIGEN_DENSEBASE_PLUGIN # include EIGEN_DENSEBASE_PLUGIN # endif #undef EIGEN_CURRENT_STORAGE_BASE_CLASS - -#ifdef EIGEN2_SUPPORT - - Block corner(CornerType type, Index cRows, Index cCols); - const Block corner(CornerType type, Index cRows, Index cCols) const; - template - Block corner(CornerType type); - template - const Block corner(CornerType type) const; - -#endif // EIGEN2_SUPPORT - +#undef EIGEN_DOC_BLOCK_ADDONS_NOT_INNER_PANEL +#undef EIGEN_DOC_BLOCK_ADDONS_INNER_PANEL_IF +#undef EIGEN_DOC_UNARY_ADDONS // disable the use of evalTo for dense objects with a nice compilation error - template inline void evalTo(Dest& ) const + template + EIGEN_DEVICE_FUNC + inline void evalTo(Dest& ) const { EIGEN_STATIC_ASSERT((internal::is_same::value),THE_EVAL_EVALTO_FUNCTION_SHOULD_NEVER_BE_CALLED_FOR_DENSE_OBJECTS); } protected: + EIGEN_DEFAULT_COPY_CONSTRUCTOR(DenseBase) /** Default constructor. Do nothing. */ - DenseBase() + EIGEN_DEVICE_FUNC DenseBase() { /* Just checks for self-consistency of the flags. - * Only do it when debugging Eigen, as this borders on paranoiac and could slow compilation down + * Only do it when debugging Eigen, as this borders on paranoia and could slow compilation down */ #ifdef EIGEN_INTERNAL_DEBUGGING EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, int(IsRowMajor)) @@ -511,9 +691,9 @@ template class DenseBase } private: - explicit DenseBase(int); - DenseBase(int,int); - template explicit DenseBase(const DenseBase&); + EIGEN_DEVICE_FUNC explicit DenseBase(int); + EIGEN_DEVICE_FUNC DenseBase(int,int); + template EIGEN_DEVICE_FUNC explicit DenseBase(const DenseBase&); }; } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/DenseCoeffsBase.h b/thirdparty/eigen/Eigen/src/Core/DenseCoeffsBase.h index 3c890f21..37fcdb59 100644 --- a/thirdparty/eigen/Eigen/src/Core/DenseCoeffsBase.h +++ b/thirdparty/eigen/Eigen/src/Core/DenseCoeffsBase.h @@ -22,11 +22,12 @@ template struct add_const_on_value_type_if_arithmetic /** \brief Base class providing read-only coefficient access to matrices and arrays. * \ingroup Core_Module * \tparam Derived Type of the derived class - * \tparam #ReadOnlyAccessors Constant indicating read-only access + * + * \note #ReadOnlyAccessors Constant indicating read-only access * * This class defines the \c operator() \c const function and friends, which can be used to read specific * entries of a matrix or array. - * + * * \sa DenseCoeffsBase, DenseCoeffsBase, * \ref TopicClassHierarchy */ @@ -35,7 +36,6 @@ class DenseCoeffsBase : public EigenBase { public: typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; typedef typename internal::packet_traits::type PacketScalar; @@ -61,6 +61,7 @@ class DenseCoeffsBase : public EigenBase using Base::size; using Base::derived; + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index rowIndexByOuterInner(Index outer, Index inner) const { return int(Derived::RowsAtCompileTime) == 1 ? 0 @@ -69,6 +70,7 @@ class DenseCoeffsBase : public EigenBase : inner; } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Index colIndexByOuterInner(Index outer, Index inner) const { return int(Derived::ColsAtCompileTime) == 1 ? 0 @@ -91,13 +93,15 @@ class DenseCoeffsBase : public EigenBase * * \sa operator()(Index,Index) const, coeffRef(Index,Index), coeff(Index) const */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index row, Index col) const { eigen_internal_assert(row >= 0 && row < rows() - && col >= 0 && col < cols()); - return derived().coeff(row, col); + && col >= 0 && col < cols()); + return internal::evaluator(derived()).coeff(row,col); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeffByOuterInner(Index outer, Index inner) const { return coeff(rowIndexByOuterInner(outer, inner), @@ -108,11 +112,12 @@ class DenseCoeffsBase : public EigenBase * * \sa operator()(Index,Index), operator[](Index) */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType operator()(Index row, Index col) const { eigen_assert(row >= 0 && row < rows() && col >= 0 && col < cols()); - return derived().coeff(row, col); + return coeff(row, col); } /** Short version: don't use this function, use @@ -130,11 +135,14 @@ class DenseCoeffsBase : public EigenBase * \sa operator[](Index) const, coeffRef(Index), coeff(Index,Index) const */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType coeff(Index index) const { + EIGEN_STATIC_ASSERT(internal::evaluator::Flags & LinearAccessBit, + THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS) eigen_internal_assert(index >= 0 && index < size()); - return derived().coeff(index); + return internal::evaluator(derived()).coeff(index); } @@ -146,15 +154,14 @@ class DenseCoeffsBase : public EigenBase * z() const, w() const */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType operator[](Index index) const { - #ifndef EIGEN2_SUPPORT EIGEN_STATIC_ASSERT(Derived::IsVectorAtCompileTime, THE_BRACKET_OPERATOR_IS_ONLY_FOR_VECTORS__USE_THE_PARENTHESIS_OPERATOR_INSTEAD) - #endif eigen_assert(index >= 0 && index < size()); - return derived().coeff(index); + return coeff(index); } /** \returns the coefficient at given index. @@ -167,32 +174,49 @@ class DenseCoeffsBase : public EigenBase * z() const, w() const */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType operator()(Index index) const { eigen_assert(index >= 0 && index < size()); - return derived().coeff(index); + return coeff(index); } /** equivalent to operator[](0). */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType x() const { return (*this)[0]; } /** equivalent to operator[](1). */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType - y() const { return (*this)[1]; } + y() const + { + EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS); + return (*this)[1]; + } /** equivalent to operator[](2). */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType - z() const { return (*this)[2]; } + z() const + { + EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS); + return (*this)[2]; + } /** equivalent to operator[](3). */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE CoeffReturnType - w() const { return (*this)[3]; } + w() const + { + EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS); + return (*this)[3]; + } /** \internal * \returns the packet of coefficients starting at the given row and column. It is your responsibility @@ -207,9 +231,9 @@ class DenseCoeffsBase : public EigenBase template EIGEN_STRONG_INLINE PacketReturnType packet(Index row, Index col) const { - eigen_internal_assert(row >= 0 && row < rows() - && col >= 0 && col < cols()); - return derived().template packet(row,col); + typedef typename internal::packet_traits::type DefaultPacketType; + eigen_internal_assert(row >= 0 && row < rows() && col >= 0 && col < cols()); + return internal::evaluator(derived()).template packet(row,col); } @@ -234,8 +258,11 @@ class DenseCoeffsBase : public EigenBase template EIGEN_STRONG_INLINE PacketReturnType packet(Index index) const { + EIGEN_STATIC_ASSERT(internal::evaluator::Flags & LinearAccessBit, + THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS) + typedef typename internal::packet_traits::type DefaultPacketType; eigen_internal_assert(index >= 0 && index < size()); - return derived().template packet(index); + return internal::evaluator(derived()).template packet(index); } protected: @@ -262,12 +289,13 @@ class DenseCoeffsBase : public EigenBase /** \brief Base class providing read/write coefficient access to matrices and arrays. * \ingroup Core_Module * \tparam Derived Type of the derived class - * \tparam #WriteAccessors Constant indicating read/write access + * + * \note #WriteAccessors Constant indicating read/write access * * This class defines the non-const \c operator() function and friends, which can be used to write specific * entries of a matrix or array. This class inherits DenseCoeffsBase which * defines the const variant for reading specific entries. - * + * * \sa DenseCoeffsBase, \ref TopicClassHierarchy */ template @@ -278,7 +306,6 @@ class DenseCoeffsBase : public DenseCoeffsBase Base; typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; typedef typename internal::packet_traits::type PacketScalar; typedef typename NumTraits::Real RealScalar; @@ -311,13 +338,15 @@ class DenseCoeffsBase : public DenseCoeffsBase= 0 && row < rows() - && col >= 0 && col < cols()); - return derived().coeffRef(row, col); + && col >= 0 && col < cols()); + return internal::evaluator(derived()).coeffRef(row,col); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar& coeffRefByOuterInner(Index outer, Index inner) { @@ -330,12 +359,13 @@ class DenseCoeffsBase : public DenseCoeffsBase= 0 && row < rows() && col >= 0 && col < cols()); - return derived().coeffRef(row, col); + return coeffRef(row, col); } @@ -354,11 +384,14 @@ class DenseCoeffsBase : public DenseCoeffsBase::Flags & LinearAccessBit, + THIS_COEFFICIENT_ACCESSOR_TAKING_ONE_ACCESS_IS_ONLY_FOR_EXPRESSIONS_ALLOWING_LINEAR_ACCESS) eigen_internal_assert(index >= 0 && index < size()); - return derived().coeffRef(index); + return internal::evaluator(derived()).coeffRef(index); } /** \returns a reference to the coefficient at given index. @@ -368,15 +401,14 @@ class DenseCoeffsBase : public DenseCoeffsBase= 0 && index < size()); - return derived().coeffRef(index); + return coeffRef(index); } /** \returns a reference to the coefficient at given index. @@ -388,179 +420,62 @@ class DenseCoeffsBase : public DenseCoeffsBase= 0 && index < size()); - return derived().coeffRef(index); + return coeffRef(index); } /** equivalent to operator[](0). */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar& x() { return (*this)[0]; } /** equivalent to operator[](1). */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar& - y() { return (*this)[1]; } - - /** equivalent to operator[](2). */ - - EIGEN_STRONG_INLINE Scalar& - z() { return (*this)[2]; } - - /** equivalent to operator[](3). */ - - EIGEN_STRONG_INLINE Scalar& - w() { return (*this)[3]; } - - /** \internal - * Stores the given packet of coefficients, at the given row and column of this expression. It is your responsibility - * to ensure that a packet really starts there. This method is only available on expressions having the - * PacketAccessBit. - * - * The \a LoadMode parameter may have the value \a #Aligned or \a #Unaligned. Its effect is to select - * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets - * starting at an address which is a multiple of the packet size. - */ - - template - EIGEN_STRONG_INLINE void writePacket - (Index row, Index col, const typename internal::packet_traits::type& val) - { - eigen_internal_assert(row >= 0 && row < rows() - && col >= 0 && col < cols()); - derived().template writePacket(row,col,val); - } - - - /** \internal */ - template - EIGEN_STRONG_INLINE void writePacketByOuterInner - (Index outer, Index inner, const typename internal::packet_traits::type& val) - { - writePacket(rowIndexByOuterInner(outer, inner), - colIndexByOuterInner(outer, inner), - val); - } - - /** \internal - * Stores the given packet of coefficients, at the given index in this expression. It is your responsibility - * to ensure that a packet really starts there. This method is only available on expressions having the - * PacketAccessBit and the LinearAccessBit. - * - * The \a LoadMode parameter may have the value \a Aligned or \a Unaligned. Its effect is to select - * the appropriate vectorization instruction. Aligned access is faster, but is only possible for packets - * starting at an address which is a multiple of the packet size. - */ - template - EIGEN_STRONG_INLINE void writePacket - (Index index, const typename internal::packet_traits::type& val) + y() { - eigen_internal_assert(index >= 0 && index < size()); - derived().template writePacket(index,val); - } - -#ifndef EIGEN_PARSED_BY_DOXYGEN - - /** \internal Copies the coefficient at position (row,col) of other into *this. - * - * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code - * with usual assignments. - * - * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox. - */ - - template - EIGEN_STRONG_INLINE void copyCoeff(Index row, Index col, const DenseBase& other) - { - eigen_internal_assert(row >= 0 && row < rows() - && col >= 0 && col < cols()); - derived().coeffRef(row, col) = other.derived().coeff(row, col); - } - - /** \internal Copies the coefficient at the given index of other into *this. - * - * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code - * with usual assignments. - * - * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox. - */ - - template - EIGEN_STRONG_INLINE void copyCoeff(Index index, const DenseBase& other) - { - eigen_internal_assert(index >= 0 && index < size()); - derived().coeffRef(index) = other.derived().coeff(index); + EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=2, OUT_OF_RANGE_ACCESS); + return (*this)[1]; } + /** equivalent to operator[](2). */ - template - EIGEN_STRONG_INLINE void copyCoeffByOuterInner(Index outer, Index inner, const DenseBase& other) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Scalar& + z() { - const Index row = rowIndexByOuterInner(outer,inner); - const Index col = colIndexByOuterInner(outer,inner); - // derived() is important here: copyCoeff() may be reimplemented in Derived! - derived().copyCoeff(row, col, other); + EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=3, OUT_OF_RANGE_ACCESS); + return (*this)[2]; } - /** \internal Copies the packet at position (row,col) of other into *this. - * - * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code - * with usual assignments. - * - * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox. - */ - - template - EIGEN_STRONG_INLINE void copyPacket(Index row, Index col, const DenseBase& other) - { - eigen_internal_assert(row >= 0 && row < rows() - && col >= 0 && col < cols()); - derived().template writePacket(row, col, - other.derived().template packet(row, col)); - } - - /** \internal Copies the packet at the given index of other into *this. - * - * This method is overridden in SwapWrapper, allowing swap() assignments to share 99% of their code - * with usual assignments. - * - * Outside of this internal usage, this method has probably no usefulness. It is hidden in the public API dox. - */ - - template - EIGEN_STRONG_INLINE void copyPacket(Index index, const DenseBase& other) - { - eigen_internal_assert(index >= 0 && index < size()); - derived().template writePacket(index, - other.derived().template packet(index)); - } + /** equivalent to operator[](3). */ - /** \internal */ - template - EIGEN_STRONG_INLINE void copyPacketByOuterInner(Index outer, Index inner, const DenseBase& other) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Scalar& + w() { - const Index row = rowIndexByOuterInner(outer,inner); - const Index col = colIndexByOuterInner(outer,inner); - // derived() is important here: copyCoeff() may be reimplemented in Derived! - derived().template copyPacket< OtherDerived, StoreMode, LoadMode>(row, col, other); + EIGEN_STATIC_ASSERT(Derived::SizeAtCompileTime==-1 || Derived::SizeAtCompileTime>=4, OUT_OF_RANGE_ACCESS); + return (*this)[3]; } -#endif - }; /** \brief Base class providing direct read-only coefficient access to matrices and arrays. * \ingroup Core_Module * \tparam Derived Type of the derived class - * \tparam #DirectAccessors Constant indicating direct access + * + * \note #DirectAccessors Constant indicating direct access * * This class defines functions to work with strides which can be used to access entries directly. This class * inherits DenseCoeffsBase which defines functions to access entries read-only using * \c operator() . * - * \sa \ref TopicClassHierarchy + * \sa \blank \ref TopicClassHierarchy */ template class DenseCoeffsBase : public DenseCoeffsBase @@ -568,7 +483,6 @@ class DenseCoeffsBase : public DenseCoeffsBase Base; - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; typedef typename NumTraits::Real RealScalar; @@ -581,6 +495,7 @@ class DenseCoeffsBase : public DenseCoeffsBase : public DenseCoeffsBase : public DenseCoeffsBase : public DenseCoeffsBase : public DenseCoeffsBase which defines functions to access entries read/write using * \c operator(). * - * \sa \ref TopicClassHierarchy + * \sa \blank \ref TopicClassHierarchy */ template class DenseCoeffsBase @@ -639,7 +558,6 @@ class DenseCoeffsBase public: typedef DenseCoeffsBase Base; - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; typedef typename NumTraits::Real RealScalar; @@ -652,7 +570,8 @@ class DenseCoeffsBase * * \sa outerStride(), rowStride(), colStride() */ - inline Index innerStride() const + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const EIGEN_NOEXCEPT { return derived().innerStride(); } @@ -662,13 +581,14 @@ class DenseCoeffsBase * * \sa innerStride(), rowStride(), colStride() */ - inline Index outerStride() const + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const EIGEN_NOEXCEPT { return derived().outerStride(); } // FIXME shall we remove it ? - inline Index stride() const + EIGEN_CONSTEXPR inline Index stride() const EIGEN_NOEXCEPT { return Derived::IsVectorAtCompileTime ? innerStride() : outerStride(); } @@ -677,7 +597,8 @@ class DenseCoeffsBase * * \sa innerStride(), outerStride(), colStride() */ - inline Index rowStride() const + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rowStride() const EIGEN_NOEXCEPT { return Derived::IsRowMajor ? outerStride() : innerStride(); } @@ -686,7 +607,8 @@ class DenseCoeffsBase * * \sa innerStride(), outerStride(), rowStride() */ - inline Index colStride() const + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index colStride() const EIGEN_NOEXCEPT { return Derived::IsRowMajor ? innerStride() : outerStride(); } @@ -694,33 +616,42 @@ class DenseCoeffsBase namespace internal { -template +template struct first_aligned_impl { - static inline typename Derived::Index run(const Derived&) + static EIGEN_CONSTEXPR inline Index run(const Derived&) EIGEN_NOEXCEPT { return 0; } }; -template -struct first_aligned_impl +template +struct first_aligned_impl { - static inline typename Derived::Index run(const Derived& m) + static inline Index run(const Derived& m) { - return internal::first_aligned(&m.const_cast_derived().coeffRef(0,0), m.size()); + return internal::first_aligned(m.data(), m.size()); } }; -/** \internal \returns the index of the first element of the array that is well aligned for vectorization. +/** \internal \returns the index of the first element of the array stored by \a m that is properly aligned with respect to \a Alignment for vectorization. + * + * \tparam Alignment requested alignment in Bytes. * * There is also the variant first_aligned(const Scalar*, Integer) defined in Memory.h. See it for more * documentation. */ +template +static inline Index first_aligned(const DenseBase& m) +{ + enum { ReturnZero = (int(evaluator::Alignment) >= Alignment) || !(Derived::Flags & DirectAccessBit) }; + return first_aligned_impl::run(m.derived()); +} + template -static inline typename Derived::Index first_aligned(const Derived& m) +static inline Index first_default_aligned(const DenseBase& m) { - return first_aligned_impl - - ::run(m); + typedef typename Derived::Scalar Scalar; + typedef typename packet_traits::type DefaultPacketType; + return internal::first_aligned::alignment),Derived>(m); } template::ret> diff --git a/thirdparty/eigen/Eigen/src/Core/DenseStorage.h b/thirdparty/eigen/Eigen/src/Core/DenseStorage.h index 568493cb..08ef6c53 100644 --- a/thirdparty/eigen/Eigen/src/Core/DenseStorage.h +++ b/thirdparty/eigen/Eigen/src/Core/DenseStorage.h @@ -3,7 +3,7 @@ // // Copyright (C) 2008 Gael Guennebaud // Copyright (C) 2006-2009 Benoit Jacob -// Copyright (C) 2010 Hauke Heibel +// Copyright (C) 2010-2013 Hauke Heibel // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -13,9 +13,9 @@ #define EIGEN_MATRIXSTORAGE_H #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN - #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN EIGEN_DENSE_STORAGE_CTOR_PLUGIN; + #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) X; EIGEN_DENSE_STORAGE_CTOR_PLUGIN; #else - #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN + #define EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(X) #endif namespace Eigen { @@ -24,7 +24,9 @@ namespace internal { struct constructor_without_unaligned_array_assert {}; -template void check_static_allocation_size() +template +EIGEN_DEVICE_FUNC +void check_static_allocation_size() { // if EIGEN_STACK_ALLOCATION_LIMIT is defined to 0, then no limit #if EIGEN_STACK_ALLOCATION_LIMIT @@ -38,57 +40,117 @@ template void check_static_allocation_size() */ template + : compute_default_alignment::value > struct plain_array { T array[Size]; - plain_array() - { + EIGEN_DEVICE_FUNC + plain_array() + { check_static_allocation_size(); } - plain_array(constructor_without_unaligned_array_assert) - { + EIGEN_DEVICE_FUNC + plain_array(constructor_without_unaligned_array_assert) + { check_static_allocation_size(); } }; #if defined(EIGEN_DISABLE_UNALIGNED_ARRAY_ASSERT) #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) -#elif EIGEN_GNUC_AT_LEAST(4,7) - // GCC 4.7 is too aggressive in its optimizations and remove the alignement test based on the fact the array is declared to be aligned. +#elif EIGEN_GNUC_AT_LEAST(4,7) + // GCC 4.7 is too aggressive in its optimizations and remove the alignment test based on the fact the array is declared to be aligned. // See this bug report: http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53900 // Hiding the origin of the array pointer behind a function argument seems to do the trick even if the function is inlined: template EIGEN_ALWAYS_INLINE PtrType eigen_unaligned_array_assert_workaround_gcc47(PtrType array) { return array; } #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \ - eigen_assert((reinterpret_cast(eigen_unaligned_array_assert_workaround_gcc47(array)) & sizemask) == 0 \ + eigen_assert((internal::UIntPtr(eigen_unaligned_array_assert_workaround_gcc47(array)) & (sizemask)) == 0 \ && "this assertion is explained here: " \ "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \ " **** READ THIS WEB PAGE !!! ****"); #else #define EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(sizemask) \ - eigen_assert((reinterpret_cast(array) & sizemask) == 0 \ + eigen_assert((internal::UIntPtr(array) & (sizemask)) == 0 \ && "this assertion is explained here: " \ "http://eigen.tuxfamily.org/dox-devel/group__TopicUnalignedArrayAssert.html" \ " **** READ THIS WEB PAGE !!! ****"); #endif +template +struct plain_array +{ + EIGEN_ALIGN_TO_BOUNDARY(8) T array[Size]; + + EIGEN_DEVICE_FUNC + plain_array() + { + EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(7); + check_static_allocation_size(); + } + + EIGEN_DEVICE_FUNC + plain_array(constructor_without_unaligned_array_assert) + { + check_static_allocation_size(); + } +}; + template struct plain_array { - EIGEN_USER_ALIGN16 T array[Size]; + EIGEN_ALIGN_TO_BOUNDARY(16) T array[Size]; + + EIGEN_DEVICE_FUNC + plain_array() + { + EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(15); + check_static_allocation_size(); + } + + EIGEN_DEVICE_FUNC + plain_array(constructor_without_unaligned_array_assert) + { + check_static_allocation_size(); + } +}; + +template +struct plain_array +{ + EIGEN_ALIGN_TO_BOUNDARY(32) T array[Size]; - plain_array() - { - EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(0xf); + EIGEN_DEVICE_FUNC + plain_array() + { + EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(31); check_static_allocation_size(); } - plain_array(constructor_without_unaligned_array_assert) - { + EIGEN_DEVICE_FUNC + plain_array(constructor_without_unaligned_array_assert) + { + check_static_allocation_size(); + } +}; + +template +struct plain_array +{ + EIGEN_ALIGN_TO_BOUNDARY(64) T array[Size]; + + EIGEN_DEVICE_FUNC + plain_array() + { + EIGEN_MAKE_UNALIGNED_ARRAY_ASSERT(63); + check_static_allocation_size(); + } + + EIGEN_DEVICE_FUNC + plain_array(constructor_without_unaligned_array_assert) + { check_static_allocation_size(); } }; @@ -96,9 +158,33 @@ struct plain_array template struct plain_array { - EIGEN_USER_ALIGN16 T array[1]; - plain_array() {} - plain_array(constructor_without_unaligned_array_assert) {} + T array[1]; + EIGEN_DEVICE_FUNC plain_array() {} + EIGEN_DEVICE_FUNC plain_array(constructor_without_unaligned_array_assert) {} +}; + +struct plain_array_helper { + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + static void copy(const plain_array& src, const Eigen::Index size, + plain_array& dst) { + smart_copy(src.array, src.array + size, dst.array); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + static void swap(plain_array& a, const Eigen::Index a_size, + plain_array& b, const Eigen::Index b_size) { + if (a_size < b_size) { + std::swap_ranges(b.array, b.array + a_size, a.array); + smart_move(b.array + a_size, b.array + b_size, a.array + a_size); + } else if (a_size > b_size) { + std::swap_ranges(a.array, a.array + b_size, b.array); + smart_move(a.array + b_size, a.array + a_size, b.array + b_size); + } else { + std::swap_ranges(a.array, a.array + a_size, b.array); + } + } }; } // end namespace internal @@ -122,41 +208,81 @@ template class DenseSt { internal::plain_array m_data; public: - DenseStorage() {} - DenseStorage(internal::constructor_without_unaligned_array_assert) + EIGEN_DEVICE_FUNC DenseStorage() { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size) + } + EIGEN_DEVICE_FUNC + explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(internal::constructor_without_unaligned_array_assert()) {} - DenseStorage(const DenseStorage& other) : m_data(other.m_data) {} +#if !EIGEN_HAS_CXX11 || defined(EIGEN_DENSE_STORAGE_CTOR_PLUGIN) + EIGEN_DEVICE_FUNC + DenseStorage(const DenseStorage& other) : m_data(other.m_data) { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = Size) + } +#else + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) = default; +#endif +#if !EIGEN_HAS_CXX11 + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) { if (this != &other) m_data = other.m_data; return *this; } - DenseStorage(DenseIndex,DenseIndex,DenseIndex) {} - void swap(DenseStorage& other) { std::swap(m_data,other.m_data); } - static DenseIndex rows(void) {return _Rows;} - static DenseIndex cols(void) {return _Cols;} - void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {} - void resize(DenseIndex,DenseIndex,DenseIndex) {} - const T *data() const { return m_data.array; } - T *data() { return m_data.array; } +#else + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) = default; +#endif +#if EIGEN_HAS_RVALUE_REFERENCES +#if !EIGEN_HAS_CXX11 + EIGEN_DEVICE_FUNC DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT + : m_data(std::move(other.m_data)) + { + } + EIGEN_DEVICE_FUNC DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT + { + if (this != &other) + m_data = std::move(other.m_data); + return *this; + } +#else + EIGEN_DEVICE_FUNC DenseStorage(DenseStorage&&) = default; + EIGEN_DEVICE_FUNC DenseStorage& operator=(DenseStorage&&) = default; +#endif +#endif + EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({}) + eigen_internal_assert(size==rows*cols && rows==_Rows && cols==_Cols); + EIGEN_UNUSED_VARIABLE(size); + EIGEN_UNUSED_VARIABLE(rows); + EIGEN_UNUSED_VARIABLE(cols); + } + EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { + numext::swap(m_data, other.m_data); + } + EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;} + EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {return _Cols;} + EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {} + EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {} + EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; } + EIGEN_DEVICE_FUNC T *data() { return m_data.array; } }; // null matrix template class DenseStorage { public: - DenseStorage() {} - DenseStorage(internal::constructor_without_unaligned_array_assert) {} - DenseStorage(const DenseStorage&) {} - DenseStorage& operator=(const DenseStorage&) { return *this; } - DenseStorage(DenseIndex,DenseIndex,DenseIndex) {} - void swap(DenseStorage& ) {} - static DenseIndex rows(void) {return _Rows;} - static DenseIndex cols(void) {return _Cols;} - void conservativeResize(DenseIndex,DenseIndex,DenseIndex) {} - void resize(DenseIndex,DenseIndex,DenseIndex) {} - const T *data() const { return 0; } - T *data() { return 0; } + EIGEN_DEVICE_FUNC DenseStorage() {} + EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) {} + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage&) {} + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage&) { return *this; } + EIGEN_DEVICE_FUNC DenseStorage(Index,Index,Index) {} + EIGEN_DEVICE_FUNC void swap(DenseStorage& ) {} + EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;} + EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) EIGEN_NOEXCEPT {return _Cols;} + EIGEN_DEVICE_FUNC void conservativeResize(Index,Index,Index) {} + EIGEN_DEVICE_FUNC void resize(Index,Index,Index) {} + EIGEN_DEVICE_FUNC const T *data() const { return 0; } + EIGEN_DEVICE_FUNC T *data() { return 0; } }; // more specializations for null matrices; these are necessary to resolve ambiguities @@ -173,260 +299,352 @@ template class DenseStorage class DenseStorage { internal::plain_array m_data; - DenseIndex m_rows; - DenseIndex m_cols; + Index m_rows; + Index m_cols; public: - DenseStorage() : m_rows(0), m_cols(0) {} - DenseStorage(internal::constructor_without_unaligned_array_assert) + EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0), m_cols(0) {} + EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0), m_cols(0) {} - DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows), m_cols(other.m_cols) {} - DenseStorage& operator=(const DenseStorage& other) + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) + : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(other.m_rows), m_cols(other.m_cols) + { + internal::plain_array_helper::copy(other.m_data, m_rows * m_cols, m_data); + } + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) { if (this != &other) { - m_data = other.m_data; m_rows = other.m_rows; m_cols = other.m_cols; + internal::plain_array_helper::copy(other.m_data, m_rows * m_cols, m_data); } return *this; } - DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) : m_rows(nbRows), m_cols(nbCols) {} - void swap(DenseStorage& other) - { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); } - DenseIndex rows() const {return m_rows;} - DenseIndex cols() const {return m_cols;} - void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; } - void resize(DenseIndex, DenseIndex nbRows, DenseIndex nbCols) { m_rows = nbRows; m_cols = nbCols; } - const T *data() const { return m_data.array; } - T *data() { return m_data.array; } + EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index cols) : m_rows(rows), m_cols(cols) {} + EIGEN_DEVICE_FUNC void swap(DenseStorage& other) + { + internal::plain_array_helper::swap(m_data, m_rows * m_cols, other.m_data, other.m_rows * other.m_cols); + numext::swap(m_rows,other.m_rows); + numext::swap(m_cols,other.m_cols); + } + EIGEN_DEVICE_FUNC Index rows() const {return m_rows;} + EIGEN_DEVICE_FUNC Index cols() const {return m_cols;} + EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; } + EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index cols) { m_rows = rows; m_cols = cols; } + EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; } + EIGEN_DEVICE_FUNC T *data() { return m_data.array; } }; // dynamic-size matrix with fixed-size storage and fixed width template class DenseStorage { internal::plain_array m_data; - DenseIndex m_rows; + Index m_rows; public: - DenseStorage() : m_rows(0) {} - DenseStorage(internal::constructor_without_unaligned_array_assert) + EIGEN_DEVICE_FUNC DenseStorage() : m_rows(0) {} + EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(0) {} - DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_rows(other.m_rows) {} - DenseStorage& operator=(const DenseStorage& other) + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) + : m_data(internal::constructor_without_unaligned_array_assert()), m_rows(other.m_rows) + { + internal::plain_array_helper::copy(other.m_data, m_rows * _Cols, m_data); + } + + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) { if (this != &other) { - m_data = other.m_data; m_rows = other.m_rows; + internal::plain_array_helper::copy(other.m_data, m_rows * _Cols, m_data); } return *this; } - DenseStorage(DenseIndex, DenseIndex nbRows, DenseIndex) : m_rows(nbRows) {} - void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); } - DenseIndex rows(void) const {return m_rows;} - DenseIndex cols(void) const {return _Cols;} - void conservativeResize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; } - void resize(DenseIndex, DenseIndex nbRows, DenseIndex) { m_rows = nbRows; } - const T *data() const { return m_data.array; } - T *data() { return m_data.array; } + EIGEN_DEVICE_FUNC DenseStorage(Index, Index rows, Index) : m_rows(rows) {} + EIGEN_DEVICE_FUNC void swap(DenseStorage& other) + { + internal::plain_array_helper::swap(m_data, m_rows * _Cols, other.m_data, other.m_rows * _Cols); + numext::swap(m_rows, other.m_rows); + } + EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;} + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols(void) const EIGEN_NOEXCEPT {return _Cols;} + EIGEN_DEVICE_FUNC void conservativeResize(Index, Index rows, Index) { m_rows = rows; } + EIGEN_DEVICE_FUNC void resize(Index, Index rows, Index) { m_rows = rows; } + EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; } + EIGEN_DEVICE_FUNC T *data() { return m_data.array; } }; // dynamic-size matrix with fixed-size storage and fixed height template class DenseStorage { internal::plain_array m_data; - DenseIndex m_cols; + Index m_cols; public: - DenseStorage() : m_cols(0) {} - DenseStorage(internal::constructor_without_unaligned_array_assert) + EIGEN_DEVICE_FUNC DenseStorage() : m_cols(0) {} + EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(internal::constructor_without_unaligned_array_assert()), m_cols(0) {} - DenseStorage(const DenseStorage& other) : m_data(other.m_data), m_cols(other.m_cols) {} - DenseStorage& operator=(const DenseStorage& other) + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) + : m_data(internal::constructor_without_unaligned_array_assert()), m_cols(other.m_cols) + { + internal::plain_array_helper::copy(other.m_data, _Rows * m_cols, m_data); + } + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) { if (this != &other) { - m_data = other.m_data; m_cols = other.m_cols; + internal::plain_array_helper::copy(other.m_data, _Rows * m_cols, m_data); } return *this; } - DenseStorage(DenseIndex, DenseIndex, DenseIndex nbCols) : m_cols(nbCols) {} - void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); } - DenseIndex rows(void) const {return _Rows;} - DenseIndex cols(void) const {return m_cols;} - void conservativeResize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; } - void resize(DenseIndex, DenseIndex, DenseIndex nbCols) { m_cols = nbCols; } - const T *data() const { return m_data.array; } - T *data() { return m_data.array; } + EIGEN_DEVICE_FUNC DenseStorage(Index, Index, Index cols) : m_cols(cols) {} + EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { + internal::plain_array_helper::swap(m_data, _Rows * m_cols, other.m_data, _Rows * other.m_cols); + numext::swap(m_cols, other.m_cols); + } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows(void) const EIGEN_NOEXCEPT {return _Rows;} + EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;} + EIGEN_DEVICE_FUNC void conservativeResize(Index, Index, Index cols) { m_cols = cols; } + EIGEN_DEVICE_FUNC void resize(Index, Index, Index cols) { m_cols = cols; } + EIGEN_DEVICE_FUNC const T *data() const { return m_data.array; } + EIGEN_DEVICE_FUNC T *data() { return m_data.array; } }; // purely dynamic matrix. template class DenseStorage { T *m_data; - DenseIndex m_rows; - DenseIndex m_cols; + Index m_rows; + Index m_cols; public: - DenseStorage() : m_data(0), m_rows(0), m_cols(0) {} - DenseStorage(internal::constructor_without_unaligned_array_assert) + EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0), m_cols(0) {} + EIGEN_DEVICE_FUNC explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0), m_cols(0) {} - DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols) - : m_data(internal::conditional_aligned_new_auto(size)), m_rows(nbRows), m_cols(nbCols) - { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN } -#ifdef EIGEN_HAVE_RVALUE_REFERENCES - DenseStorage(DenseStorage&& other) + EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) + : m_data(internal::conditional_aligned_new_auto(size)), m_rows(rows), m_cols(cols) + { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({}) + eigen_internal_assert(size==rows*cols && rows>=0 && cols >=0); + } + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) + : m_data(internal::conditional_aligned_new_auto(other.m_rows*other.m_cols)) + , m_rows(other.m_rows) + , m_cols(other.m_cols) + { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*m_cols) + internal::smart_copy(other.m_data, other.m_data+other.m_rows*other.m_cols, m_data); + } + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) + { + if (this != &other) + { + DenseStorage tmp(other); + this->swap(tmp); + } + return *this; + } +#if EIGEN_HAS_RVALUE_REFERENCES + EIGEN_DEVICE_FUNC + DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT : m_data(std::move(other.m_data)) , m_rows(std::move(other.m_rows)) , m_cols(std::move(other.m_cols)) { other.m_data = nullptr; + other.m_rows = 0; + other.m_cols = 0; } - DenseStorage& operator=(DenseStorage&& other) + EIGEN_DEVICE_FUNC + DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT { - using std::swap; - swap(m_data, other.m_data); - swap(m_rows, other.m_rows); - swap(m_cols, other.m_cols); + numext::swap(m_data, other.m_data); + numext::swap(m_rows, other.m_rows); + numext::swap(m_cols, other.m_cols); return *this; } #endif - ~DenseStorage() { internal::conditional_aligned_delete_auto(m_data, m_rows*m_cols); } - void swap(DenseStorage& other) - { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); std::swap(m_cols,other.m_cols); } - DenseIndex rows(void) const {return m_rows;} - DenseIndex cols(void) const {return m_cols;} - void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols) + EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto(m_data, m_rows*m_cols); } + EIGEN_DEVICE_FUNC void swap(DenseStorage& other) + { + numext::swap(m_data,other.m_data); + numext::swap(m_rows,other.m_rows); + numext::swap(m_cols,other.m_cols); + } + EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;} + EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;} + void conservativeResize(Index size, Index rows, Index cols) { m_data = internal::conditional_aligned_realloc_new_auto(m_data, size, m_rows*m_cols); - m_rows = nbRows; - m_cols = nbCols; + m_rows = rows; + m_cols = cols; } - void resize(DenseIndex size, DenseIndex nbRows, DenseIndex nbCols) + EIGEN_DEVICE_FUNC void resize(Index size, Index rows, Index cols) { if(size != m_rows*m_cols) { internal::conditional_aligned_delete_auto(m_data, m_rows*m_cols); - if (size) + if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative m_data = internal::conditional_aligned_new_auto(size); else m_data = 0; - EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({}) } - m_rows = nbRows; - m_cols = nbCols; - } - const T *data() const { return m_data; } - T *data() { return m_data; } - private: - DenseStorage(const DenseStorage&); - DenseStorage& operator=(const DenseStorage&); + m_rows = rows; + m_cols = cols; + } + EIGEN_DEVICE_FUNC const T *data() const { return m_data; } + EIGEN_DEVICE_FUNC T *data() { return m_data; } }; // matrix with dynamic width and fixed height (so that matrix has dynamic size). template class DenseStorage { T *m_data; - DenseIndex m_cols; + Index m_cols; public: - DenseStorage() : m_data(0), m_cols(0) {} - DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {} - DenseStorage(DenseIndex size, DenseIndex, DenseIndex nbCols) : m_data(internal::conditional_aligned_new_auto(size)), m_cols(nbCols) - { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN } -#ifdef EIGEN_HAVE_RVALUE_REFERENCES - DenseStorage(DenseStorage&& other) + EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_cols(0) {} + explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_cols(0) {} + EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto(size)), m_cols(cols) + { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({}) + eigen_internal_assert(size==rows*cols && rows==_Rows && cols >=0); + EIGEN_UNUSED_VARIABLE(rows); + } + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) + : m_data(internal::conditional_aligned_new_auto(_Rows*other.m_cols)) + , m_cols(other.m_cols) + { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_cols*_Rows) + internal::smart_copy(other.m_data, other.m_data+_Rows*m_cols, m_data); + } + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) + { + if (this != &other) + { + DenseStorage tmp(other); + this->swap(tmp); + } + return *this; + } +#if EIGEN_HAS_RVALUE_REFERENCES + EIGEN_DEVICE_FUNC + DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT : m_data(std::move(other.m_data)) , m_cols(std::move(other.m_cols)) { other.m_data = nullptr; + other.m_cols = 0; } - DenseStorage& operator=(DenseStorage&& other) + EIGEN_DEVICE_FUNC + DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT { - using std::swap; - swap(m_data, other.m_data); - swap(m_cols, other.m_cols); + numext::swap(m_data, other.m_data); + numext::swap(m_cols, other.m_cols); return *this; } #endif - ~DenseStorage() { internal::conditional_aligned_delete_auto(m_data, _Rows*m_cols); } - void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_cols,other.m_cols); } - static DenseIndex rows(void) {return _Rows;} - DenseIndex cols(void) const {return m_cols;} - void conservativeResize(DenseIndex size, DenseIndex, DenseIndex nbCols) + EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto(m_data, _Rows*m_cols); } + EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { + numext::swap(m_data,other.m_data); + numext::swap(m_cols,other.m_cols); + } + EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index rows(void) EIGEN_NOEXCEPT {return _Rows;} + EIGEN_DEVICE_FUNC Index cols(void) const EIGEN_NOEXCEPT {return m_cols;} + EIGEN_DEVICE_FUNC void conservativeResize(Index size, Index, Index cols) { m_data = internal::conditional_aligned_realloc_new_auto(m_data, size, _Rows*m_cols); - m_cols = nbCols; + m_cols = cols; } - EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex, DenseIndex nbCols) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index, Index cols) { if(size != _Rows*m_cols) { internal::conditional_aligned_delete_auto(m_data, _Rows*m_cols); - if (size) + if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative m_data = internal::conditional_aligned_new_auto(size); else m_data = 0; - EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({}) } - m_cols = nbCols; + m_cols = cols; } - const T *data() const { return m_data; } - T *data() { return m_data; } - private: - DenseStorage(const DenseStorage&); - DenseStorage& operator=(const DenseStorage&); + EIGEN_DEVICE_FUNC const T *data() const { return m_data; } + EIGEN_DEVICE_FUNC T *data() { return m_data; } }; // matrix with dynamic height and fixed width (so that matrix has dynamic size). template class DenseStorage { T *m_data; - DenseIndex m_rows; + Index m_rows; public: - DenseStorage() : m_data(0), m_rows(0) {} - DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {} - DenseStorage(DenseIndex size, DenseIndex nbRows, DenseIndex) : m_data(internal::conditional_aligned_new_auto(size)), m_rows(nbRows) - { EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN } -#ifdef EIGEN_HAVE_RVALUE_REFERENCES - DenseStorage(DenseStorage&& other) + EIGEN_DEVICE_FUNC DenseStorage() : m_data(0), m_rows(0) {} + explicit DenseStorage(internal::constructor_without_unaligned_array_assert) : m_data(0), m_rows(0) {} + EIGEN_DEVICE_FUNC DenseStorage(Index size, Index rows, Index cols) : m_data(internal::conditional_aligned_new_auto(size)), m_rows(rows) + { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({}) + eigen_internal_assert(size==rows*cols && rows>=0 && cols == _Cols); + EIGEN_UNUSED_VARIABLE(cols); + } + EIGEN_DEVICE_FUNC DenseStorage(const DenseStorage& other) + : m_data(internal::conditional_aligned_new_auto(other.m_rows*_Cols)) + , m_rows(other.m_rows) + { + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN(Index size = m_rows*_Cols) + internal::smart_copy(other.m_data, other.m_data+other.m_rows*_Cols, m_data); + } + EIGEN_DEVICE_FUNC DenseStorage& operator=(const DenseStorage& other) + { + if (this != &other) + { + DenseStorage tmp(other); + this->swap(tmp); + } + return *this; + } +#if EIGEN_HAS_RVALUE_REFERENCES + EIGEN_DEVICE_FUNC + DenseStorage(DenseStorage&& other) EIGEN_NOEXCEPT : m_data(std::move(other.m_data)) , m_rows(std::move(other.m_rows)) { other.m_data = nullptr; + other.m_rows = 0; } - DenseStorage& operator=(DenseStorage&& other) + EIGEN_DEVICE_FUNC + DenseStorage& operator=(DenseStorage&& other) EIGEN_NOEXCEPT { - using std::swap; - swap(m_data, other.m_data); - swap(m_rows, other.m_rows); + numext::swap(m_data, other.m_data); + numext::swap(m_rows, other.m_rows); return *this; } #endif - ~DenseStorage() { internal::conditional_aligned_delete_auto(m_data, _Cols*m_rows); } - void swap(DenseStorage& other) { std::swap(m_data,other.m_data); std::swap(m_rows,other.m_rows); } - DenseIndex rows(void) const {return m_rows;} - static DenseIndex cols(void) {return _Cols;} - void conservativeResize(DenseIndex size, DenseIndex nbRows, DenseIndex) + EIGEN_DEVICE_FUNC ~DenseStorage() { internal::conditional_aligned_delete_auto(m_data, _Cols*m_rows); } + EIGEN_DEVICE_FUNC void swap(DenseStorage& other) { + numext::swap(m_data,other.m_data); + numext::swap(m_rows,other.m_rows); + } + EIGEN_DEVICE_FUNC Index rows(void) const EIGEN_NOEXCEPT {return m_rows;} + EIGEN_DEVICE_FUNC static EIGEN_CONSTEXPR Index cols(void) {return _Cols;} + void conservativeResize(Index size, Index rows, Index) { m_data = internal::conditional_aligned_realloc_new_auto(m_data, size, m_rows*_Cols); - m_rows = nbRows; + m_rows = rows; } - EIGEN_STRONG_INLINE void resize(DenseIndex size, DenseIndex nbRows, DenseIndex) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resize(Index size, Index rows, Index) { if(size != m_rows*_Cols) { internal::conditional_aligned_delete_auto(m_data, _Cols*m_rows); - if (size) + if (size>0) // >0 and not simply !=0 to let the compiler knows that size cannot be negative m_data = internal::conditional_aligned_new_auto(size); else m_data = 0; - EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN + EIGEN_INTERNAL_DENSE_STORAGE_CTOR_PLUGIN({}) } - m_rows = nbRows; + m_rows = rows; } - const T *data() const { return m_data; } - T *data() { return m_data; } - private: - DenseStorage(const DenseStorage&); - DenseStorage& operator=(const DenseStorage&); + EIGEN_DEVICE_FUNC const T *data() const { return m_data; } + EIGEN_DEVICE_FUNC T *data() { return m_data; } }; } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/Diagonal.h b/thirdparty/eigen/Eigen/src/Core/Diagonal.h index 68cf6d4b..ad5bccd7 100644 --- a/thirdparty/eigen/Eigen/src/Core/Diagonal.h +++ b/thirdparty/eigen/Eigen/src/Core/Diagonal.h @@ -11,7 +11,7 @@ #ifndef EIGEN_DIAGONAL_H #define EIGEN_DIAGONAL_H -namespace Eigen { +namespace Eigen { /** \class Diagonal * \ingroup Core_Module @@ -21,7 +21,7 @@ namespace Eigen { * \param MatrixType the type of the object in which we are taking a sub/main/super diagonal * \param DiagIndex the index of the sub/super diagonal. The default is 0 and it means the main diagonal. * A positive value means a superdiagonal, a negative value means a subdiagonal. - * You can also use Dynamic so the index can be set at runtime. + * You can also use DynamicIndex so the index can be set at runtime. * * The matrix is not required to be square. * @@ -37,7 +37,7 @@ template struct traits > : traits { - typedef typename nested::type MatrixTypeNested; + typedef typename ref_selector::type MatrixTypeNested; typedef typename remove_reference::type _MatrixTypeNested; typedef typename MatrixType::StorageKind StorageKind; enum { @@ -52,8 +52,7 @@ struct traits > MatrixType::MaxColsAtCompileTime - EIGEN_PLAIN_ENUM_MAX( DiagIndex, 0))), MaxColsAtCompileTime = 1, MaskLvalueBit = is_lvalue::value ? LvalueBit : 0, - Flags = (unsigned int)_MatrixTypeNested::Flags & (HereditaryBits | LinearAccessBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, - CoeffReadCost = _MatrixTypeNested::CoeffReadCost, + Flags = (unsigned int)_MatrixTypeNested::Flags & (RowMajorBit | MaskLvalueBit | DirectAccessBit) & ~RowMajorBit, // FIXME DirectAccessBit should not be handled by expressions MatrixTypeOuterStride = outer_stride_at_compile_time::ret, InnerStrideAtCompileTime = MatrixTypeOuterStride == Dynamic ? Dynamic : MatrixTypeOuterStride+1, OuterStrideAtCompileTime = 0 @@ -70,24 +69,31 @@ template class Diagonal typedef typename internal::dense_xpr_base::type Base; EIGEN_DENSE_PUBLIC_INTERFACE(Diagonal) - inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index) {} + EIGEN_DEVICE_FUNC + explicit inline Diagonal(MatrixType& matrix, Index a_index = DiagIndex) : m_matrix(matrix), m_index(a_index) + { + eigen_assert( a_index <= m_matrix.cols() && -a_index <= m_matrix.rows() ); + } EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Diagonal) + EIGEN_DEVICE_FUNC inline Index rows() const - { return m_index.value()<0 ? (std::min)(m_matrix.cols(),m_matrix.rows()+m_index.value()) : (std::min)(m_matrix.rows(),m_matrix.cols()-m_index.value()); } + { + return m_index.value()<0 ? numext::mini(m_matrix.cols(),m_matrix.rows()+m_index.value()) + : numext::mini(m_matrix.rows(),m_matrix.cols()-m_index.value()); + } - inline Index cols() const { return 1; } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return 1; } - inline Index innerStride() const - { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const EIGEN_NOEXCEPT { return m_matrix.outerStride() + 1; } - inline Index outerStride() const - { - return 0; - } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const EIGEN_NOEXCEPT { return 0; } typedef typename internal::conditional< internal::is_lvalue::value, @@ -95,62 +101,75 @@ template class Diagonal const Scalar >::type ScalarWithConstIfNotLvalue; - inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); } - inline const Scalar* data() const { return &(m_matrix.const_cast_derived().coeffRef(rowOffset(), colOffset())); } + EIGEN_DEVICE_FUNC + inline ScalarWithConstIfNotLvalue* data() { return &(m_matrix.coeffRef(rowOffset(), colOffset())); } + EIGEN_DEVICE_FUNC + inline const Scalar* data() const { return &(m_matrix.coeffRef(rowOffset(), colOffset())); } + EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index) { EIGEN_STATIC_ASSERT_LVALUE(MatrixType) - return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset()); + return m_matrix.coeffRef(row+rowOffset(), row+colOffset()); } + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index row, Index) const { - return m_matrix.const_cast_derived().coeffRef(row+rowOffset(), row+colOffset()); + return m_matrix.coeffRef(row+rowOffset(), row+colOffset()); } + EIGEN_DEVICE_FUNC inline CoeffReturnType coeff(Index row, Index) const { return m_matrix.coeff(row+rowOffset(), row+colOffset()); } + EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index idx) { EIGEN_STATIC_ASSERT_LVALUE(MatrixType) - return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset()); + return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset()); } + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index idx) const { - return m_matrix.const_cast_derived().coeffRef(idx+rowOffset(), idx+colOffset()); + return m_matrix.coeffRef(idx+rowOffset(), idx+colOffset()); } + EIGEN_DEVICE_FUNC inline CoeffReturnType coeff(Index idx) const { return m_matrix.coeff(idx+rowOffset(), idx+colOffset()); } - const typename internal::remove_all::type& - nestedExpression() const + EIGEN_DEVICE_FUNC + inline const typename internal::remove_all::type& + nestedExpression() const { return m_matrix; } - int index() const + EIGEN_DEVICE_FUNC + inline Index index() const { return m_index.value(); } protected: - typename MatrixType::Nested m_matrix; + typename internal::ref_selector::non_const_type m_matrix; const internal::variable_if_dynamicindex m_index; private: // some compilers may fail to optimize std::max etc in case of compile-time constants... - EIGEN_STRONG_INLINE Index absDiagIndex() const { return m_index.value()>0 ? m_index.value() : -m_index.value(); } - EIGEN_STRONG_INLINE Index rowOffset() const { return m_index.value()>0 ? 0 : -m_index.value(); } - EIGEN_STRONG_INLINE Index colOffset() const { return m_index.value()>0 ? m_index.value() : 0; } - // triger a compile time error is someone try to call packet + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index absDiagIndex() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : -m_index.value(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rowOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? 0 : -m_index.value(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index colOffset() const EIGEN_NOEXCEPT { return m_index.value()>0 ? m_index.value() : 0; } + // trigger a compile-time error if someone try to call packet template typename MatrixType::PacketReturnType packet(Index) const; template typename MatrixType::PacketReturnType packet(Index,Index) const; }; @@ -164,15 +183,16 @@ template class Diagonal * * \sa class Diagonal */ template -inline typename MatrixBase::DiagonalReturnType +EIGEN_DEVICE_FUNC inline typename MatrixBase::DiagonalReturnType MatrixBase::diagonal() { - return derived(); + return DiagonalReturnType(derived()); } /** This is the const version of diagonal(). */ template -inline typename MatrixBase::ConstDiagonalReturnType +EIGEN_DEVICE_FUNC inline +const typename MatrixBase::ConstDiagonalReturnType MatrixBase::diagonal() const { return ConstDiagonalReturnType(derived()); @@ -190,18 +210,18 @@ MatrixBase::diagonal() const * * \sa MatrixBase::diagonal(), class Diagonal */ template -inline typename MatrixBase::DiagonalDynamicIndexReturnType +EIGEN_DEVICE_FUNC inline Diagonal MatrixBase::diagonal(Index index) { - return DiagonalDynamicIndexReturnType(derived(), index); + return Diagonal(derived(), index); } /** This is the const version of diagonal(Index). */ template -inline typename MatrixBase::ConstDiagonalDynamicIndexReturnType +EIGEN_DEVICE_FUNC inline const Diagonal MatrixBase::diagonal(Index index) const { - return ConstDiagonalDynamicIndexReturnType(derived(), index); + return Diagonal(derived(), index); } /** \returns an expression of the \a DiagIndex-th sub or super diagonal of the matrix \c *this @@ -216,20 +236,22 @@ MatrixBase::diagonal(Index index) const * * \sa MatrixBase::diagonal(), class Diagonal */ template -template -inline typename MatrixBase::template DiagonalIndexReturnType::Type +template +EIGEN_DEVICE_FUNC +inline Diagonal MatrixBase::diagonal() { - return derived(); + return Diagonal(derived()); } /** This is the const version of diagonal(). */ template -template -inline typename MatrixBase::template ConstDiagonalIndexReturnType::Type +template +EIGEN_DEVICE_FUNC +inline const Diagonal MatrixBase::diagonal() const { - return derived(); + return Diagonal(derived()); } } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/DiagonalMatrix.h b/thirdparty/eigen/Eigen/src/Core/DiagonalMatrix.h index 53c757be..542685c6 100644 --- a/thirdparty/eigen/Eigen/src/Core/DiagonalMatrix.h +++ b/thirdparty/eigen/Eigen/src/Core/DiagonalMatrix.h @@ -22,7 +22,7 @@ class DiagonalBase : public EigenBase typedef typename DiagonalVectorType::Scalar Scalar; typedef typename DiagonalVectorType::RealScalar RealScalar; typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; + typedef typename internal::traits::StorageIndex StorageIndex; enum { RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime, @@ -30,79 +30,85 @@ class DiagonalBase : public EigenBase MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime, MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime, IsVectorAtCompileTime = 0, - Flags = 0 + Flags = NoPreferredStorageOrderBit }; typedef Matrix DenseMatrixType; typedef DenseMatrixType DenseType; typedef DiagonalMatrix PlainObject; + EIGEN_DEVICE_FUNC inline const Derived& derived() const { return *static_cast(this); } + EIGEN_DEVICE_FUNC inline Derived& derived() { return *static_cast(this); } + EIGEN_DEVICE_FUNC DenseMatrixType toDenseMatrix() const { return derived(); } - template - void evalTo(MatrixBase &other) const; - template - inline void addTo(MatrixBase &other) const - { other.diagonal() += diagonal(); } - template - inline void subTo(MatrixBase &other) const - { other.diagonal() -= diagonal(); } + EIGEN_DEVICE_FUNC inline const DiagonalVectorType& diagonal() const { return derived().diagonal(); } + EIGEN_DEVICE_FUNC inline DiagonalVectorType& diagonal() { return derived().diagonal(); } + EIGEN_DEVICE_FUNC inline Index rows() const { return diagonal().size(); } + EIGEN_DEVICE_FUNC inline Index cols() const { return diagonal().size(); } - /** \returns the diagonal matrix product of \c *this by the matrix \a matrix. - */ template - const DiagonalProduct + EIGEN_DEVICE_FUNC + const Product operator*(const MatrixBase &matrix) const { - return DiagonalProduct(matrix.derived(), derived()); + return Product(derived(),matrix.derived()); } - inline const DiagonalWrapper, const DiagonalVectorType> > + typedef DiagonalWrapper, const DiagonalVectorType> > InverseReturnType; + EIGEN_DEVICE_FUNC + inline const InverseReturnType inverse() const { - return diagonal().cwiseInverse(); + return InverseReturnType(diagonal().cwiseInverse()); } - inline const DiagonalWrapper, const DiagonalVectorType> > + EIGEN_DEVICE_FUNC + inline const DiagonalWrapper operator*(const Scalar& scalar) const { - return diagonal() * scalar; + return DiagonalWrapper(diagonal() * scalar); } - friend inline const DiagonalWrapper, const DiagonalVectorType> > + EIGEN_DEVICE_FUNC + friend inline const DiagonalWrapper operator*(const Scalar& scalar, const DiagonalBase& other) { - return other.diagonal() * scalar; + return DiagonalWrapper(scalar * other.diagonal()); } - - #ifdef EIGEN2_SUPPORT + template - bool isApprox(const DiagonalBase& other, typename NumTraits::Real precision = NumTraits::dummy_precision()) const + EIGEN_DEVICE_FUNC + #ifdef EIGEN_PARSED_BY_DOXYGEN + inline unspecified_expression_type + #else + inline const DiagonalWrapper + #endif + operator+(const DiagonalBase& other) const { - return diagonal().isApprox(other.diagonal(), precision); + return (diagonal() + other.diagonal()).asDiagonal(); } + template - bool isApprox(const MatrixBase& other, typename NumTraits::Real precision = NumTraits::dummy_precision()) const + EIGEN_DEVICE_FUNC + #ifdef EIGEN_PARSED_BY_DOXYGEN + inline unspecified_expression_type + #else + inline const DiagonalWrapper + #endif + operator-(const DiagonalBase& other) const { - return toDenseMatrix().isApprox(other, precision); + return (diagonal() - other.diagonal()).asDiagonal(); } - #endif }; -template -template -inline void DiagonalBase::evalTo(MatrixBase &other) const -{ - other.setZero(); - other.diagonal() = diagonal(); -} #endif /** \class DiagonalMatrix @@ -124,10 +130,9 @@ struct traits > : traits > { typedef Matrix<_Scalar,SizeAtCompileTime,1,0,MaxSizeAtCompileTime,1> DiagonalVectorType; - typedef Dense StorageKind; - typedef DenseIndex Index; + typedef DiagonalShape StorageKind; enum { - Flags = LvalueBit + Flags = LvalueBit | NoPreferredStorageOrderBit }; }; } @@ -141,7 +146,7 @@ class DiagonalMatrix typedef const DiagonalMatrix& Nested; typedef _Scalar Scalar; typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; + typedef typename internal::traits::StorageIndex StorageIndex; #endif protected: @@ -151,24 +156,55 @@ class DiagonalMatrix public: /** const version of diagonal(). */ + EIGEN_DEVICE_FUNC inline const DiagonalVectorType& diagonal() const { return m_diagonal; } /** \returns a reference to the stored vector of diagonal coefficients. */ + EIGEN_DEVICE_FUNC inline DiagonalVectorType& diagonal() { return m_diagonal; } /** Default constructor without initialization */ + EIGEN_DEVICE_FUNC inline DiagonalMatrix() {} /** Constructs a diagonal matrix with given dimension */ - inline DiagonalMatrix(Index dim) : m_diagonal(dim) {} + EIGEN_DEVICE_FUNC + explicit inline DiagonalMatrix(Index dim) : m_diagonal(dim) {} /** 2D constructor. */ + EIGEN_DEVICE_FUNC inline DiagonalMatrix(const Scalar& x, const Scalar& y) : m_diagonal(x,y) {} /** 3D constructor. */ + EIGEN_DEVICE_FUNC inline DiagonalMatrix(const Scalar& x, const Scalar& y, const Scalar& z) : m_diagonal(x,y,z) {} + #if EIGEN_HAS_CXX11 + /** \brief Construct a diagonal matrix with fixed size from an arbitrary number of coefficients. \cpp11 + * + * There exists C++98 anologue constructors for fixed-size diagonal matrices having 2 or 3 coefficients. + * + * \warning To construct a diagonal matrix of fixed size, the number of values passed to this + * constructor must match the fixed dimension of \c *this. + * + * \sa DiagonalMatrix(const Scalar&, const Scalar&) + * \sa DiagonalMatrix(const Scalar&, const Scalar&, const Scalar&) + */ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + DiagonalMatrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const ArgTypes&... args) + : m_diagonal(a0, a1, a2, args...) {} + + /** \brief Constructs a DiagonalMatrix and initializes it by elements given by an initializer list of initializer + * lists \cpp11 + */ + EIGEN_DEVICE_FUNC + explicit EIGEN_STRONG_INLINE DiagonalMatrix(const std::initializer_list>& list) + : m_diagonal(list) {} + #endif // EIGEN_HAS_CXX11 + /** Copy constructor. */ template + EIGEN_DEVICE_FUNC inline DiagonalMatrix(const DiagonalBase& other) : m_diagonal(other.diagonal()) {} #ifndef EIGEN_PARSED_BY_DOXYGEN @@ -178,11 +214,13 @@ class DiagonalMatrix /** generic constructor from expression of the diagonal coefficients */ template + EIGEN_DEVICE_FUNC explicit inline DiagonalMatrix(const MatrixBase& other) : m_diagonal(other) {} /** Copy operator. */ template + EIGEN_DEVICE_FUNC DiagonalMatrix& operator=(const DiagonalBase& other) { m_diagonal = other.diagonal(); @@ -193,6 +231,7 @@ class DiagonalMatrix /** This is a special case of the templated operator=. Its purpose is to * prevent a default operator= from hiding the templated operator=. */ + EIGEN_DEVICE_FUNC DiagonalMatrix& operator=(const DiagonalMatrix& other) { m_diagonal = other.diagonal(); @@ -201,14 +240,19 @@ class DiagonalMatrix #endif /** Resizes to given size. */ + EIGEN_DEVICE_FUNC inline void resize(Index size) { m_diagonal.resize(size); } /** Sets all coefficients to zero. */ + EIGEN_DEVICE_FUNC inline void setZero() { m_diagonal.setZero(); } /** Resizes and sets all coefficients to zero. */ + EIGEN_DEVICE_FUNC inline void setZero(Index size) { m_diagonal.setZero(size); } /** Sets this matrix to be the identity matrix of the current size. */ + EIGEN_DEVICE_FUNC inline void setIdentity() { m_diagonal.setOnes(); } /** Sets this matrix to be the identity matrix of the given size. */ + EIGEN_DEVICE_FUNC inline void setIdentity(Index size) { m_diagonal.setOnes(size); } }; @@ -232,14 +276,15 @@ struct traits > { typedef _DiagonalVectorType DiagonalVectorType; typedef typename DiagonalVectorType::Scalar Scalar; - typedef typename DiagonalVectorType::Index Index; - typedef typename DiagonalVectorType::StorageKind StorageKind; + typedef typename DiagonalVectorType::StorageIndex StorageIndex; + typedef DiagonalShape StorageKind; + typedef typename traits::XprKind XprKind; enum { RowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime, ColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime, - MaxRowsAtCompileTime = DiagonalVectorType::SizeAtCompileTime, - MaxColsAtCompileTime = DiagonalVectorType::SizeAtCompileTime, - Flags = traits::Flags & LvalueBit + MaxRowsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime, + MaxColsAtCompileTime = DiagonalVectorType::MaxSizeAtCompileTime, + Flags = (traits::Flags & LvalueBit) | NoPreferredStorageOrderBit }; }; } @@ -255,9 +300,11 @@ class DiagonalWrapper #endif /** Constructor from expression of diagonal coefficients to wrap. */ - inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {} + EIGEN_DEVICE_FUNC + explicit inline DiagonalWrapper(DiagonalVectorType& a_diagonal) : m_diagonal(a_diagonal) {} /** \returns a const reference to the wrapped expression of diagonal coefficients. */ + EIGEN_DEVICE_FUNC const DiagonalVectorType& diagonal() const { return m_diagonal; } protected: @@ -274,10 +321,10 @@ class DiagonalWrapper * \sa class DiagonalWrapper, class DiagonalMatrix, diagonal(), isDiagonal() **/ template -inline const DiagonalWrapper +EIGEN_DEVICE_FUNC inline const DiagonalWrapper MatrixBase::asDiagonal() const { - return derived(); + return DiagonalWrapper(derived()); } /** \returns true if *this is approximately equal to a diagonal matrix, @@ -291,12 +338,11 @@ MatrixBase::asDiagonal() const template bool MatrixBase::isDiagonal(const RealScalar& prec) const { - using std::abs; if(cols() != rows()) return false; RealScalar maxAbsOnDiagonal = static_cast(-1); for(Index j = 0; j < cols(); ++j) { - RealScalar absOnDiagonal = abs(coeff(j,j)); + RealScalar absOnDiagonal = numext::abs(coeff(j,j)); if(absOnDiagonal > maxAbsOnDiagonal) maxAbsOnDiagonal = absOnDiagonal; } for(Index j = 0; j < cols(); ++j) @@ -308,6 +354,38 @@ bool MatrixBase::isDiagonal(const RealScalar& prec) const return true; } +namespace internal { + +template<> struct storage_kind_to_shape { typedef DiagonalShape Shape; }; + +struct Diagonal2Dense {}; + +template<> struct AssignmentKind { typedef Diagonal2Dense Kind; }; + +// Diagonal matrix to Dense assignment +template< typename DstXprType, typename SrcXprType, typename Functor> +struct Assignment +{ + static void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op &/*func*/) + { + Index dstRows = src.rows(); + Index dstCols = src.cols(); + if((dst.rows()!=dstRows) || (dst.cols()!=dstCols)) + dst.resize(dstRows, dstCols); + + dst.setZero(); + dst.diagonal() = src.diagonal(); + } + + static void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op &/*func*/) + { dst.diagonal() += src.diagonal(); } + + static void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op &/*func*/) + { dst.diagonal() -= src.diagonal(); } +}; + +} // namespace internal + } // end namespace Eigen #endif // EIGEN_DIAGONALMATRIX_H diff --git a/thirdparty/eigen/Eigen/src/Core/DiagonalProduct.h b/thirdparty/eigen/Eigen/src/Core/DiagonalProduct.h index cc6b536e..7911d1cd 100644 --- a/thirdparty/eigen/Eigen/src/Core/DiagonalProduct.h +++ b/thirdparty/eigen/Eigen/src/Core/DiagonalProduct.h @@ -13,117 +13,14 @@ namespace Eigen { -namespace internal { -template -struct traits > - : traits -{ - typedef typename scalar_product_traits::ReturnType Scalar; - enum { - RowsAtCompileTime = MatrixType::RowsAtCompileTime, - ColsAtCompileTime = MatrixType::ColsAtCompileTime, - MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime, - MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime, - - _StorageOrder = MatrixType::Flags & RowMajorBit ? RowMajor : ColMajor, - _ScalarAccessOnDiag = !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft) - ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)), - _SameTypes = is_same::value, - // FIXME currently we need same types, but in the future the next rule should be the one - //_Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))), - _Vectorizable = bool(int(MatrixType::Flags)&PacketAccessBit) && _SameTypes && (_ScalarAccessOnDiag || (bool(int(DiagonalType::DiagonalVectorType::Flags)&PacketAccessBit))), - _LinearAccessMask = (RowsAtCompileTime==1 || ColsAtCompileTime==1) ? LinearAccessBit : 0, - - Flags = ((HereditaryBits|_LinearAccessMask|AlignedBit) & (unsigned int)(MatrixType::Flags)) | (_Vectorizable ? PacketAccessBit : 0),//(int(MatrixType::Flags)&int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit), - Cost0 = EIGEN_ADD_COST(NumTraits::MulCost, MatrixType::CoeffReadCost), - CoeffReadCost = EIGEN_ADD_COST(Cost0,DiagonalType::DiagonalVectorType::CoeffReadCost) - }; -}; -} - -template -class DiagonalProduct : internal::no_assignment_operator, - public MatrixBase > -{ - public: - - typedef MatrixBase Base; - EIGEN_DENSE_PUBLIC_INTERFACE(DiagonalProduct) - - inline DiagonalProduct(const MatrixType& matrix, const DiagonalType& diagonal) - : m_matrix(matrix), m_diagonal(diagonal) - { - eigen_assert(diagonal.diagonal().size() == (ProductOrder == OnTheLeft ? matrix.rows() : matrix.cols())); - } - - EIGEN_STRONG_INLINE Index rows() const { return m_matrix.rows(); } - EIGEN_STRONG_INLINE Index cols() const { return m_matrix.cols(); } - - EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const - { - return m_diagonal.diagonal().coeff(ProductOrder == OnTheLeft ? row : col) * m_matrix.coeff(row, col); - } - - EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const - { - enum { - StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor - }; - return coeff(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index row, Index col) const - { - enum { - StorageOrder = Flags & RowMajorBit ? RowMajor : ColMajor - }; - const Index indexInDiagonalVector = ProductOrder == OnTheLeft ? row : col; - return packet_impl(row,col,indexInDiagonalVector,typename internal::conditional< - ((int(StorageOrder) == RowMajor && int(ProductOrder) == OnTheLeft) - ||(int(StorageOrder) == ColMajor && int(ProductOrder) == OnTheRight)), internal::true_type, internal::false_type>::type()); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet(Index idx) const - { - enum { - StorageOrder = int(MatrixType::Flags) & RowMajorBit ? RowMajor : ColMajor - }; - return packet(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx); - } - - protected: - template - EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::true_type) const - { - return internal::pmul(m_matrix.template packet(row, col), - internal::pset1(m_diagonal.diagonal().coeff(id))); - } - - template - EIGEN_STRONG_INLINE PacketScalar packet_impl(Index row, Index col, Index id, internal::false_type) const - { - enum { - InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime, - DiagonalVectorPacketLoadMode = (LoadMode == Aligned && (((InnerSize%16) == 0) || (int(DiagonalType::DiagonalVectorType::Flags)&AlignedBit)==AlignedBit) ? Aligned : Unaligned) - }; - return internal::pmul(m_matrix.template packet(row, col), - m_diagonal.diagonal().template packet(id)); - } - - typename MatrixType::Nested m_matrix; - typename DiagonalType::Nested m_diagonal; -}; - /** \returns the diagonal matrix product of \c *this by the diagonal matrix \a diagonal. */ template template -inline const DiagonalProduct +EIGEN_DEVICE_FUNC inline const Product MatrixBase::operator*(const DiagonalBase &a_diagonal) const { - return DiagonalProduct(derived(), a_diagonal.derived()); + return Product(derived(),a_diagonal.derived()); } } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/Dot.h b/thirdparty/eigen/Eigen/src/Core/Dot.h index 23aab831..abac7ad4 100644 --- a/thirdparty/eigen/Eigen/src/Core/Dot.h +++ b/thirdparty/eigen/Eigen/src/Core/Dot.h @@ -18,36 +18,38 @@ namespace internal { // with mismatched types, the compiler emits errors about failing to instantiate cwiseProduct BEFORE // looking at the static assertions. Thus this is a trick to get better compile errors. template + bool NeedToTranspose = T::IsVectorAtCompileTime && U::IsVectorAtCompileTime && + ((int(T::RowsAtCompileTime) == 1 && int(U::ColsAtCompileTime) == 1) || + (int(T::ColsAtCompileTime) == 1 && int(U::RowsAtCompileTime) == 1))> struct dot_nocheck { - typedef typename scalar_product_traits::Scalar,typename traits::Scalar>::ReturnType ResScalar; - static inline ResScalar run(const MatrixBase& a, const MatrixBase& b) + typedef scalar_conj_product_op::Scalar,typename traits::Scalar> conj_prod; + typedef typename conj_prod::result_type ResScalar; + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE + static ResScalar run(const MatrixBase& a, const MatrixBase& b) { - return a.template binaryExpr::Scalar,typename traits::Scalar> >(b).sum(); + return a.template binaryExpr(b).sum(); } }; template struct dot_nocheck { - typedef typename scalar_product_traits::Scalar,typename traits::Scalar>::ReturnType ResScalar; - static inline ResScalar run(const MatrixBase& a, const MatrixBase& b) + typedef scalar_conj_product_op::Scalar,typename traits::Scalar> conj_prod; + typedef typename conj_prod::result_type ResScalar; + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE + static ResScalar run(const MatrixBase& a, const MatrixBase& b) { - return a.transpose().template binaryExpr::Scalar,typename traits::Scalar> >(b).sum(); + return a.transpose().template binaryExpr(b).sum(); } }; } // end namespace internal -/** \returns the dot product of *this with other. +/** \fn MatrixBase::dot + * \returns the dot product of *this with other. * * \only_for_vectors * @@ -59,58 +61,34 @@ struct dot_nocheck */ template template -inline typename internal::scalar_product_traits::Scalar,typename internal::traits::Scalar>::ReturnType +EIGEN_DEVICE_FUNC +EIGEN_STRONG_INLINE +typename ScalarBinaryOpTraits::Scalar,typename internal::traits::Scalar>::ReturnType MatrixBase::dot(const MatrixBase& other) const { EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived) EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived) +#if !(defined(EIGEN_NO_STATIC_ASSERT) && defined(EIGEN_NO_DEBUG)) typedef internal::scalar_conj_product_op func; EIGEN_CHECK_BINARY_COMPATIBILIY(func,Scalar,typename OtherDerived::Scalar); - +#endif + eigen_assert(size() == other.size()); return internal::dot_nocheck::run(*this, other); } -#ifdef EIGEN2_SUPPORT -/** \returns the dot product of *this with other, with the Eigen2 convention that the dot product is linear in the first variable - * (conjugating the second variable). Of course this only makes a difference in the complex case. - * - * This method is only available in EIGEN2_SUPPORT mode. - * - * \only_for_vectors - * - * \sa dot() - */ -template -template -typename internal::traits::Scalar -MatrixBase::eigen2_dot(const MatrixBase& other) const -{ - EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) - EIGEN_STATIC_ASSERT_VECTOR_ONLY(OtherDerived) - EIGEN_STATIC_ASSERT_SAME_VECTOR_SIZE(Derived,OtherDerived) - EIGEN_STATIC_ASSERT((internal::is_same::value), - YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) - - eigen_assert(size() == other.size()); - - return internal::dot_nocheck::run(other,*this); -} -#endif - - //---------- implementation of L2 norm and related functions ---------- -/** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the Frobenius norm. +/** \returns, for vectors, the squared \em l2 norm of \c *this, and for matrices the squared Frobenius norm. * In both cases, it consists in the sum of the square of all the matrix entries. * For vectors, this is also equals to the dot product of \c *this with itself. * - * \sa dot(), norm() + * \sa dot(), norm(), lpNorm() */ template -EIGEN_STRONG_INLINE typename NumTraits::Scalar>::Real MatrixBase::squaredNorm() const +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits::Scalar>::Real MatrixBase::squaredNorm() const { return numext::real((*this).cwiseAbs2().sum()); } @@ -119,41 +97,98 @@ EIGEN_STRONG_INLINE typename NumTraits::Scala * In both cases, it consists in the square root of the sum of the square of all the matrix entries. * For vectors, this is also equals to the square root of the dot product of \c *this with itself. * - * \sa dot(), squaredNorm() + * \sa lpNorm(), dot(), squaredNorm() */ template -inline typename NumTraits::Scalar>::Real MatrixBase::norm() const +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename NumTraits::Scalar>::Real MatrixBase::norm() const { - using std::sqrt; - return sqrt(squaredNorm()); + return numext::sqrt(squaredNorm()); } -/** \returns an expression of the quotient of *this by its own norm. +/** \returns an expression of the quotient of \c *this by its own norm. + * + * \warning If the input vector is too small (i.e., this->norm()==0), + * then this function returns a copy of the input. * * \only_for_vectors * * \sa norm(), normalize() */ template -inline const typename MatrixBase::PlainObject +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::PlainObject MatrixBase::normalized() const { - typedef typename internal::nested::type Nested; - typedef typename internal::remove_reference::type _Nested; + typedef typename internal::nested_eval::type _Nested; _Nested n(derived()); - return n / n.norm(); + RealScalar z = n.squaredNorm(); + // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU + if(z>RealScalar(0)) + return n / numext::sqrt(z); + else + return n; } /** Normalizes the vector, i.e. divides it by its own norm. * * \only_for_vectors * + * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged. + * * \sa norm(), normalized() */ template -inline void MatrixBase::normalize() +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase::normalize() { - *this /= norm(); + RealScalar z = squaredNorm(); + // NOTE: after extensive benchmarking, this conditional does not impact performance, at least on recent x86 CPU + if(z>RealScalar(0)) + derived() /= numext::sqrt(z); +} + +/** \returns an expression of the quotient of \c *this by its own norm while avoiding underflow and overflow. + * + * \only_for_vectors + * + * This method is analogue to the normalized() method, but it reduces the risk of + * underflow and overflow when computing the norm. + * + * \warning If the input vector is too small (i.e., this->norm()==0), + * then this function returns a copy of the input. + * + * \sa stableNorm(), stableNormalize(), normalized() + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const typename MatrixBase::PlainObject +MatrixBase::stableNormalized() const +{ + typedef typename internal::nested_eval::type _Nested; + _Nested n(derived()); + RealScalar w = n.cwiseAbs().maxCoeff(); + RealScalar z = (n/w).squaredNorm(); + if(z>RealScalar(0)) + return n / (numext::sqrt(z)*w); + else + return n; +} + +/** Normalizes the vector while avoid underflow and overflow + * + * \only_for_vectors + * + * This method is analogue to the normalize() method, but it reduces the risk of + * underflow and overflow when computing the norm. + * + * \warning If the input vector is too small (i.e., this->norm()==0), then \c *this is left unchanged. + * + * \sa stableNorm(), stableNormalized(), normalize() + */ +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void MatrixBase::stableNormalize() +{ + RealScalar w = cwiseAbs().maxCoeff(); + RealScalar z = (derived()/w).squaredNorm(); + if(z>RealScalar(0)) + derived() /= numext::sqrt(z)*w; } //---------- implementation of other norms ---------- @@ -164,9 +199,10 @@ template struct lpNorm_selector { typedef typename NumTraits::Scalar>::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const MatrixBase& m) { - using std::pow; + EIGEN_USING_STD(pow) return pow(m.cwiseAbs().array().pow(p).sum(), RealScalar(1)/p); } }; @@ -174,6 +210,7 @@ struct lpNorm_selector template struct lpNorm_selector { + EIGEN_DEVICE_FUNC static inline typename NumTraits::Scalar>::Real run(const MatrixBase& m) { return m.cwiseAbs().sum(); @@ -183,6 +220,7 @@ struct lpNorm_selector template struct lpNorm_selector { + EIGEN_DEVICE_FUNC static inline typename NumTraits::Scalar>::Real run(const MatrixBase& m) { return m.norm(); @@ -192,23 +230,35 @@ struct lpNorm_selector template struct lpNorm_selector { - static inline typename NumTraits::Scalar>::Real run(const MatrixBase& m) + typedef typename NumTraits::Scalar>::Real RealScalar; + EIGEN_DEVICE_FUNC + static inline RealScalar run(const MatrixBase& m) { + if(Derived::SizeAtCompileTime==0 || (Derived::SizeAtCompileTime==Dynamic && m.size()==0)) + return RealScalar(0); return m.cwiseAbs().maxCoeff(); } }; } // end namespace internal -/** \returns the \f$ \ell^p \f$ norm of *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values - * of the coefficients of *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$ - * norm, that is the maximum of the absolute values of the coefficients of *this. +/** \returns the \b coefficient-wise \f$ \ell^p \f$ norm of \c *this, that is, returns the p-th root of the sum of the p-th powers of the absolute values + * of the coefficients of \c *this. If \a p is the special value \a Eigen::Infinity, this function returns the \f$ \ell^\infty \f$ + * norm, that is the maximum of the absolute values of the coefficients of \c *this. + * + * In all cases, if \c *this is empty, then the value 0 is returned. + * + * \note For matrices, this function does not compute the operator-norm. That is, if \c *this is a matrix, then its coefficients are interpreted as a 1D vector. Nonetheless, you can easily compute the 1-norm and \f$\infty\f$-norm matrix operator norms using \link TutorialReductionsVisitorsBroadcastingReductionsNorm partial reductions \endlink. * * \sa norm() */ template template -inline typename NumTraits::Scalar>::Real +#ifndef EIGEN_PARSED_BY_DOXYGEN +EIGEN_DEVICE_FUNC inline typename NumTraits::Scalar>::Real +#else +EIGEN_DEVICE_FUNC MatrixBase::RealScalar +#endif MatrixBase::lpNorm() const { return internal::lpNorm_selector::run(*this); @@ -227,8 +277,8 @@ template bool MatrixBase::isOrthogonal (const MatrixBase& other, const RealScalar& prec) const { - typename internal::nested::type nested(derived()); - typename internal::nested::type otherNested(other.derived()); + typename internal::nested_eval::type nested(derived()); + typename internal::nested_eval::type otherNested(other.derived()); return numext::abs2(nested.dot(otherNested)) <= prec * prec * nested.squaredNorm() * otherNested.squaredNorm(); } @@ -246,13 +296,13 @@ bool MatrixBase::isOrthogonal template bool MatrixBase::isUnitary(const RealScalar& prec) const { - typename Derived::Nested nested(derived()); + typename internal::nested_eval::type self(derived()); for(Index i = 0; i < cols(); ++i) { - if(!internal::isApprox(nested.col(i).squaredNorm(), static_cast(1), prec)) + if(!internal::isApprox(self.col(i).squaredNorm(), static_cast(1), prec)) return false; for(Index j = 0; j < i; ++j) - if(!internal::isMuchSmallerThan(nested.col(i).dot(nested.col(j)), static_cast(1), prec)) + if(!internal::isMuchSmallerThan(self.col(i).dot(self.col(j)), static_cast(1), prec)) return false; } return true; diff --git a/thirdparty/eigen/Eigen/src/Core/EigenBase.h b/thirdparty/eigen/Eigen/src/Core/EigenBase.h index fadb4585..6b3c7d37 100644 --- a/thirdparty/eigen/Eigen/src/Core/EigenBase.h +++ b/thirdparty/eigen/Eigen/src/Core/EigenBase.h @@ -13,7 +13,10 @@ namespace Eigen { -/** Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T). +/** \class EigenBase + * \ingroup Core_Module + * + * Common base class for all classes T such that MatrixBase has an operator=(T) and a constructor MatrixBase(T). * * In other words, an EigenBase object is an object that can be copied into a MatrixBase. * @@ -21,39 +24,58 @@ namespace Eigen { * * Notice that this class is trivial, it is only used to disambiguate overloaded functions. * - * \sa \ref TopicClassHierarchy + * \sa \blank \ref TopicClassHierarchy */ template struct EigenBase { // typedef typename internal::plain_matrix_type::type PlainObject; + /** \brief The interface type of indices + * \details To change this, \c \#define the preprocessor symbol \c EIGEN_DEFAULT_DENSE_INDEX_TYPE. + * \sa StorageIndex, \ref TopicPreprocessorDirectives. + * DEPRECATED: Since Eigen 3.3, its usage is deprecated. Use Eigen::Index instead. + * Deprecation is not marked with a doxygen comment because there are too many existing usages to add the deprecation attribute. + */ + typedef Eigen::Index Index; + + // FIXME is it needed? typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; /** \returns a reference to the derived object */ + EIGEN_DEVICE_FUNC Derived& derived() { return *static_cast(this); } /** \returns a const reference to the derived object */ + EIGEN_DEVICE_FUNC const Derived& derived() const { return *static_cast(this); } + EIGEN_DEVICE_FUNC inline Derived& const_cast_derived() const { return *static_cast(const_cast(this)); } + EIGEN_DEVICE_FUNC inline const Derived& const_derived() const { return *static_cast(this); } /** \returns the number of rows. \sa cols(), RowsAtCompileTime */ - inline Index rows() const { return derived().rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rows() const EIGEN_NOEXCEPT { return derived().rows(); } /** \returns the number of columns. \sa rows(), ColsAtCompileTime*/ - inline Index cols() const { return derived().cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return derived().cols(); } /** \returns the number of coefficients, which is rows()*cols(). * \sa rows(), cols(), SizeAtCompileTime. */ - inline Index size() const { return rows() * cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index size() const EIGEN_NOEXCEPT { return rows() * cols(); } /** \internal Don't use it, but do the equivalent: \code dst = *this; \endcode */ - template inline void evalTo(Dest& dst) const + template + EIGEN_DEVICE_FUNC + inline void evalTo(Dest& dst) const { derived().evalTo(dst); } /** \internal Don't use it, but do the equivalent: \code dst += *this; \endcode */ - template inline void addTo(Dest& dst) const + template + EIGEN_DEVICE_FUNC + inline void addTo(Dest& dst) const { // This is the default implementation, // derived class can reimplement it in a more optimized way. @@ -63,7 +85,9 @@ template struct EigenBase } /** \internal Don't use it, but do the equivalent: \code dst -= *this; \endcode */ - template inline void subTo(Dest& dst) const + template + EIGEN_DEVICE_FUNC + inline void subTo(Dest& dst) const { // This is the default implementation, // derived class can reimplement it in a more optimized way. @@ -73,7 +97,8 @@ template struct EigenBase } /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheRight(*this); \endcode */ - template inline void applyThisOnTheRight(Dest& dst) const + template + EIGEN_DEVICE_FUNC inline void applyThisOnTheRight(Dest& dst) const { // This is the default implementation, // derived class can reimplement it in a more optimized way. @@ -81,7 +106,8 @@ template struct EigenBase } /** \internal Don't use it, but do the equivalent: \code dst.applyOnTheLeft(*this); \endcode */ - template inline void applyThisOnTheLeft(Dest& dst) const + template + EIGEN_DEVICE_FUNC inline void applyThisOnTheLeft(Dest& dst) const { // This is the default implementation, // derived class can reimplement it in a more optimized way. @@ -104,25 +130,28 @@ template struct EigenBase */ template template +EIGEN_DEVICE_FUNC Derived& DenseBase::operator=(const EigenBase &other) { - other.derived().evalTo(derived()); + call_assignment(derived(), other.derived()); return derived(); } template template +EIGEN_DEVICE_FUNC Derived& DenseBase::operator+=(const EigenBase &other) { - other.derived().addTo(derived()); + call_assignment(derived(), other.derived(), internal::add_assign_op()); return derived(); } template template +EIGEN_DEVICE_FUNC Derived& DenseBase::operator-=(const EigenBase &other) { - other.derived().subTo(derived()); + call_assignment(derived(), other.derived(), internal::sub_assign_op()); return derived(); } diff --git a/thirdparty/eigen/Eigen/src/Core/Flagged.h b/thirdparty/eigen/Eigen/src/Core/Flagged.h deleted file mode 100644 index 1f2955fc..00000000 --- a/thirdparty/eigen/Eigen/src/Core/Flagged.h +++ /dev/null @@ -1,140 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_FLAGGED_H -#define EIGEN_FLAGGED_H - -namespace Eigen { - -/** \class Flagged - * \ingroup Core_Module - * - * \brief Expression with modified flags - * - * \param ExpressionType the type of the object of which we are modifying the flags - * \param Added the flags added to the expression - * \param Removed the flags removed from the expression (has priority over Added). - * - * This class represents an expression whose flags have been modified. - * It is the return type of MatrixBase::flagged() - * and most of the time this is the only way it is used. - * - * \sa MatrixBase::flagged() - */ - -namespace internal { -template -struct traits > : traits -{ - enum { Flags = (ExpressionType::Flags | Added) & ~Removed }; -}; -} - -template class Flagged - : public MatrixBase > -{ - public: - - typedef MatrixBase Base; - - EIGEN_DENSE_PUBLIC_INTERFACE(Flagged) - typedef typename internal::conditional::ret, - ExpressionType, const ExpressionType&>::type ExpressionTypeNested; - typedef typename ExpressionType::InnerIterator InnerIterator; - - inline Flagged(const ExpressionType& matrix) : m_matrix(matrix) {} - - inline Index rows() const { return m_matrix.rows(); } - inline Index cols() const { return m_matrix.cols(); } - inline Index outerStride() const { return m_matrix.outerStride(); } - inline Index innerStride() const { return m_matrix.innerStride(); } - - inline CoeffReturnType coeff(Index row, Index col) const - { - return m_matrix.coeff(row, col); - } - - inline CoeffReturnType coeff(Index index) const - { - return m_matrix.coeff(index); - } - - inline const Scalar& coeffRef(Index row, Index col) const - { - return m_matrix.const_cast_derived().coeffRef(row, col); - } - - inline const Scalar& coeffRef(Index index) const - { - return m_matrix.const_cast_derived().coeffRef(index); - } - - inline Scalar& coeffRef(Index row, Index col) - { - return m_matrix.const_cast_derived().coeffRef(row, col); - } - - inline Scalar& coeffRef(Index index) - { - return m_matrix.const_cast_derived().coeffRef(index); - } - - template - inline const PacketScalar packet(Index row, Index col) const - { - return m_matrix.template packet(row, col); - } - - template - inline void writePacket(Index row, Index col, const PacketScalar& x) - { - m_matrix.const_cast_derived().template writePacket(row, col, x); - } - - template - inline const PacketScalar packet(Index index) const - { - return m_matrix.template packet(index); - } - - template - inline void writePacket(Index index, const PacketScalar& x) - { - m_matrix.const_cast_derived().template writePacket(index, x); - } - - const ExpressionType& _expression() const { return m_matrix; } - - template - typename ExpressionType::PlainObject solveTriangular(const MatrixBase& other) const; - - template - void solveTriangularInPlace(const MatrixBase& other) const; - - protected: - ExpressionTypeNested m_matrix; -}; - -/** \returns an expression of *this with added and removed flags - * - * This is mostly for internal use. - * - * \sa class Flagged - */ -template -template -inline const Flagged -DenseBase::flagged() const -{ - return derived(); -} - -} // end namespace Eigen - -#endif // EIGEN_FLAGGED_H diff --git a/thirdparty/eigen/Eigen/src/Core/ForceAlignedAccess.h b/thirdparty/eigen/Eigen/src/Core/ForceAlignedAccess.h index 807c7a29..817a43af 100644 --- a/thirdparty/eigen/Eigen/src/Core/ForceAlignedAccess.h +++ b/thirdparty/eigen/Eigen/src/Core/ForceAlignedAccess.h @@ -39,29 +39,33 @@ template class ForceAlignedAccess typedef typename internal::dense_xpr_base::type Base; EIGEN_DENSE_PUBLIC_INTERFACE(ForceAlignedAccess) - inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {} - - inline Index rows() const { return m_expression.rows(); } - inline Index cols() const { return m_expression.cols(); } - inline Index outerStride() const { return m_expression.outerStride(); } - inline Index innerStride() const { return m_expression.innerStride(); } - - inline const CoeffReturnType coeff(Index row, Index col) const + EIGEN_DEVICE_FUNC explicit inline ForceAlignedAccess(const ExpressionType& matrix) : m_expression(matrix) {} + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const EIGEN_NOEXCEPT { return m_expression.outerStride(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const EIGEN_NOEXCEPT { return m_expression.innerStride(); } + + EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index row, Index col) const { return m_expression.coeff(row, col); } - inline Scalar& coeffRef(Index row, Index col) + EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index row, Index col) { return m_expression.const_cast_derived().coeffRef(row, col); } - inline const CoeffReturnType coeff(Index index) const + EIGEN_DEVICE_FUNC inline const CoeffReturnType coeff(Index index) const { return m_expression.coeff(index); } - inline Scalar& coeffRef(Index index) + EIGEN_DEVICE_FUNC inline Scalar& coeffRef(Index index) { return m_expression.const_cast_derived().coeffRef(index); } @@ -90,7 +94,7 @@ template class ForceAlignedAccess m_expression.const_cast_derived().template writePacket(index, x); } - operator const ExpressionType&() const { return m_expression; } + EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; } protected: const ExpressionType& m_expression; @@ -127,7 +131,7 @@ template inline typename internal::add_const_on_value_type,Derived&>::type>::type MatrixBase::forceAlignedAccessIf() const { - return derived(); + return derived(); // FIXME This should not work but apparently is never used } /** \returns an expression of *this with forced aligned access if \a Enable is true. @@ -138,7 +142,7 @@ template inline typename internal::conditional,Derived&>::type MatrixBase::forceAlignedAccessIf() { - return derived(); + return derived(); // FIXME This should not work but apparently is never used } } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/Functors.h b/thirdparty/eigen/Eigen/src/Core/Functors.h deleted file mode 100644 index 5a1b2f28..00000000 --- a/thirdparty/eigen/Eigen/src/Core/Functors.h +++ /dev/null @@ -1,1029 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2008-2010 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_FUNCTORS_H -#define EIGEN_FUNCTORS_H - -namespace Eigen { - -namespace internal { - -// associative functors: - -/** \internal - * \brief Template functor to compute the sum of two scalars - * - * \sa class CwiseBinaryOp, MatrixBase::operator+, class VectorwiseOp, MatrixBase::sum() - */ -template struct scalar_sum_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_sum_op) - EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a + b; } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const - { return internal::padd(a,b); } - template - EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const - { return internal::predux(a); } -}; -template -struct functor_traits > { - enum { - Cost = NumTraits::AddCost, - PacketAccess = packet_traits::HasAdd - }; -}; - -/** \internal - * \brief Template functor to compute the product of two scalars - * - * \sa class CwiseBinaryOp, Cwise::operator*(), class VectorwiseOp, MatrixBase::redux() - */ -template struct scalar_product_op { - enum { - // TODO vectorize mixed product - Vectorizable = is_same::value && packet_traits::HasMul && packet_traits::HasMul - }; - typedef typename scalar_product_traits::ReturnType result_type; - EIGEN_EMPTY_STRUCT_CTOR(scalar_product_op) - EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a * b; } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const - { return internal::pmul(a,b); } - template - EIGEN_STRONG_INLINE const result_type predux(const Packet& a) const - { return internal::predux_mul(a); } -}; -template -struct functor_traits > { - enum { - Cost = (NumTraits::MulCost + NumTraits::MulCost)/2, // rough estimate! - PacketAccess = scalar_product_op::Vectorizable - }; -}; - -/** \internal - * \brief Template functor to compute the conjugate product of two scalars - * - * This is a short cut for conj(x) * y which is needed for optimization purpose; in Eigen2 support mode, this becomes x * conj(y) - */ -template struct scalar_conj_product_op { - - enum { - Conj = NumTraits::IsComplex - }; - - typedef typename scalar_product_traits::ReturnType result_type; - - EIGEN_EMPTY_STRUCT_CTOR(scalar_conj_product_op) - EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const - { return conj_helper().pmul(a,b); } - - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const - { return conj_helper().pmul(a,b); } -}; -template -struct functor_traits > { - enum { - Cost = NumTraits::MulCost, - PacketAccess = internal::is_same::value && packet_traits::HasMul - }; -}; - -/** \internal - * \brief Template functor to compute the min of two scalars - * - * \sa class CwiseBinaryOp, MatrixBase::cwiseMin, class VectorwiseOp, MatrixBase::minCoeff() - */ -template struct scalar_min_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_min_op) - EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { using std::min; return (min)(a, b); } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const - { return internal::pmin(a,b); } - template - EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const - { return internal::predux_min(a); } -}; -template -struct functor_traits > { - enum { - Cost = NumTraits::AddCost, - PacketAccess = packet_traits::HasMin - }; -}; - -/** \internal - * \brief Template functor to compute the max of two scalars - * - * \sa class CwiseBinaryOp, MatrixBase::cwiseMax, class VectorwiseOp, MatrixBase::maxCoeff() - */ -template struct scalar_max_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_max_op) - EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { using std::max; return (max)(a, b); } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const - { return internal::pmax(a,b); } - template - EIGEN_STRONG_INLINE const Scalar predux(const Packet& a) const - { return internal::predux_max(a); } -}; -template -struct functor_traits > { - enum { - Cost = NumTraits::AddCost, - PacketAccess = packet_traits::HasMax - }; -}; - -/** \internal - * \brief Template functor to compute the hypot of two scalars - * - * \sa MatrixBase::stableNorm(), class Redux - */ -template struct scalar_hypot_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_hypot_op) -// typedef typename NumTraits::Real result_type; - EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& _x, const Scalar& _y) const - { - using std::max; - using std::min; - using std::sqrt; - Scalar p = (max)(_x, _y); - Scalar q = (min)(_x, _y); - Scalar qp = q/p; - return p * sqrt(Scalar(1) + qp*qp); - } -}; -template -struct functor_traits > { - enum { Cost = 5 * NumTraits::MulCost, PacketAccess=0 }; -}; - -/** \internal - * \brief Template functor to compute the pow of two scalars - */ -template struct scalar_binary_pow_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_binary_pow_op) - inline Scalar operator() (const Scalar& a, const OtherScalar& b) const { return numext::pow(a, b); } -}; -template -struct functor_traits > { - enum { Cost = 5 * NumTraits::MulCost, PacketAccess = false }; -}; - -// other binary functors: - -/** \internal - * \brief Template functor to compute the difference of two scalars - * - * \sa class CwiseBinaryOp, MatrixBase::operator- - */ -template struct scalar_difference_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_difference_op) - EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a, const Scalar& b) const { return a - b; } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const - { return internal::psub(a,b); } -}; -template -struct functor_traits > { - enum { - Cost = NumTraits::AddCost, - PacketAccess = packet_traits::HasSub - }; -}; - -/** \internal - * \brief Template functor to compute the quotient of two scalars - * - * \sa class CwiseBinaryOp, Cwise::operator/() - */ -template struct scalar_quotient_op { - enum { - // TODO vectorize mixed product - Vectorizable = is_same::value && packet_traits::HasDiv && packet_traits::HasDiv - }; - typedef typename scalar_product_traits::ReturnType result_type; - EIGEN_EMPTY_STRUCT_CTOR(scalar_quotient_op) - EIGEN_STRONG_INLINE const result_type operator() (const LhsScalar& a, const RhsScalar& b) const { return a / b; } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a, const Packet& b) const - { return internal::pdiv(a,b); } -}; -template -struct functor_traits > { - enum { - Cost = (NumTraits::MulCost + NumTraits::MulCost), // rough estimate! - PacketAccess = scalar_quotient_op::Vectorizable - }; -}; - - - -/** \internal - * \brief Template functor to compute the and of two booleans - * - * \sa class CwiseBinaryOp, ArrayBase::operator&& - */ -struct scalar_boolean_and_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_and_op) - EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a && b; } -}; -template<> struct functor_traits { - enum { - Cost = NumTraits::AddCost, - PacketAccess = false - }; -}; - -/** \internal - * \brief Template functor to compute the or of two booleans - * - * \sa class CwiseBinaryOp, ArrayBase::operator|| - */ -struct scalar_boolean_or_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_boolean_or_op) - EIGEN_STRONG_INLINE bool operator() (const bool& a, const bool& b) const { return a || b; } -}; -template<> struct functor_traits { - enum { - Cost = NumTraits::AddCost, - PacketAccess = false - }; -}; - -/** \internal - * \brief Template functors for comparison of two scalars - * \todo Implement packet-comparisons - */ -template struct scalar_cmp_op; - -template -struct functor_traits > { - enum { - Cost = NumTraits::AddCost, - PacketAccess = false - }; -}; - -template -struct result_of(Scalar,Scalar)> { - typedef bool type; -}; - - -template struct scalar_cmp_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op) - EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a==b;} -}; -template struct scalar_cmp_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op) - EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a struct scalar_cmp_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op) - EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a<=b;} -}; -template struct scalar_cmp_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op) - EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return !(a<=b || b<=a);} -}; -template struct scalar_cmp_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cmp_op) - EIGEN_STRONG_INLINE bool operator()(const Scalar& a, const Scalar& b) const {return a!=b;} -}; - -// unary functors: - -/** \internal - * \brief Template functor to compute the opposite of a scalar - * - * \sa class CwiseUnaryOp, MatrixBase::operator- - */ -template struct scalar_opposite_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_opposite_op) - EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { return -a; } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const - { return internal::pnegate(a); } -}; -template -struct functor_traits > -{ enum { - Cost = NumTraits::AddCost, - PacketAccess = packet_traits::HasNegate }; -}; - -/** \internal - * \brief Template functor to compute the absolute value of a scalar - * - * \sa class CwiseUnaryOp, Cwise::abs - */ -template struct scalar_abs_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_abs_op) - typedef typename NumTraits::Real result_type; - EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { using std::abs; return abs(a); } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const - { return internal::pabs(a); } -}; -template -struct functor_traits > -{ - enum { - Cost = NumTraits::AddCost, - PacketAccess = packet_traits::HasAbs - }; -}; - -/** \internal - * \brief Template functor to compute the squared absolute value of a scalar - * - * \sa class CwiseUnaryOp, Cwise::abs2 - */ -template struct scalar_abs2_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_abs2_op) - typedef typename NumTraits::Real result_type; - EIGEN_STRONG_INLINE const result_type operator() (const Scalar& a) const { return numext::abs2(a); } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const - { return internal::pmul(a,a); } -}; -template -struct functor_traits > -{ enum { Cost = NumTraits::MulCost, PacketAccess = packet_traits::HasAbs2 }; }; - -/** \internal - * \brief Template functor to compute the conjugate of a complex value - * - * \sa class CwiseUnaryOp, MatrixBase::conjugate() - */ -template struct scalar_conjugate_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_conjugate_op) - EIGEN_STRONG_INLINE const Scalar operator() (const Scalar& a) const { using numext::conj; return conj(a); } - template - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const { return internal::pconj(a); } -}; -template -struct functor_traits > -{ - enum { - Cost = NumTraits::IsComplex ? NumTraits::AddCost : 0, - PacketAccess = packet_traits::HasConj - }; -}; - -/** \internal - * \brief Template functor to cast a scalar to another type - * - * \sa class CwiseUnaryOp, MatrixBase::cast() - */ -template -struct scalar_cast_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cast_op) - typedef NewType result_type; - EIGEN_STRONG_INLINE const NewType operator() (const Scalar& a) const { return cast(a); } -}; -template -struct functor_traits > -{ enum { Cost = is_same::value ? 0 : NumTraits::AddCost, PacketAccess = false }; }; - -/** \internal - * \brief Template functor to extract the real part of a complex - * - * \sa class CwiseUnaryOp, MatrixBase::real() - */ -template -struct scalar_real_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_real_op) - typedef typename NumTraits::Real result_type; - EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::real(a); } -}; -template -struct functor_traits > -{ enum { Cost = 0, PacketAccess = false }; }; - -/** \internal - * \brief Template functor to extract the imaginary part of a complex - * - * \sa class CwiseUnaryOp, MatrixBase::imag() - */ -template -struct scalar_imag_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_op) - typedef typename NumTraits::Real result_type; - EIGEN_STRONG_INLINE result_type operator() (const Scalar& a) const { return numext::imag(a); } -}; -template -struct functor_traits > -{ enum { Cost = 0, PacketAccess = false }; }; - -/** \internal - * \brief Template functor to extract the real part of a complex as a reference - * - * \sa class CwiseUnaryOp, MatrixBase::real() - */ -template -struct scalar_real_ref_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_real_ref_op) - typedef typename NumTraits::Real result_type; - EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::real_ref(*const_cast(&a)); } -}; -template -struct functor_traits > -{ enum { Cost = 0, PacketAccess = false }; }; - -/** \internal - * \brief Template functor to extract the imaginary part of a complex as a reference - * - * \sa class CwiseUnaryOp, MatrixBase::imag() - */ -template -struct scalar_imag_ref_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_imag_ref_op) - typedef typename NumTraits::Real result_type; - EIGEN_STRONG_INLINE result_type& operator() (const Scalar& a) const { return numext::imag_ref(*const_cast(&a)); } -}; -template -struct functor_traits > -{ enum { Cost = 0, PacketAccess = false }; }; - -/** \internal - * - * \brief Template functor to compute the exponential of a scalar - * - * \sa class CwiseUnaryOp, Cwise::exp() - */ -template struct scalar_exp_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_exp_op) - inline const Scalar operator() (const Scalar& a) const { using std::exp; return exp(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::pexp(a); } -}; -template -struct functor_traits > -{ enum { Cost = 5 * NumTraits::MulCost, PacketAccess = packet_traits::HasExp }; }; - -/** \internal - * - * \brief Template functor to compute the logarithm of a scalar - * - * \sa class CwiseUnaryOp, Cwise::log() - */ -template struct scalar_log_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_log_op) - inline const Scalar operator() (const Scalar& a) const { using std::log; return log(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::plog(a); } -}; -template -struct functor_traits > -{ enum { Cost = 5 * NumTraits::MulCost, PacketAccess = packet_traits::HasLog }; }; - -/** \internal - * \brief Template functor to multiply a scalar by a fixed other one - * - * \sa class CwiseUnaryOp, MatrixBase::operator*, MatrixBase::operator/ - */ -/* NOTE why doing the pset1() in packetOp *is* an optimization ? - * indeed it seems better to declare m_other as a Packet and do the pset1() once - * in the constructor. However, in practice: - * - GCC does not like m_other as a Packet and generate a load every time it needs it - * - on the other hand GCC is able to moves the pset1() outside the loop :) - * - simpler code ;) - * (ICC and gcc 4.4 seems to perform well in both cases, the issue is visible with y = a*x + b*y) - */ -template -struct scalar_multiple_op { - typedef typename packet_traits::type Packet; - // FIXME default copy constructors seems bugged with std::complex<> - EIGEN_STRONG_INLINE scalar_multiple_op(const scalar_multiple_op& other) : m_other(other.m_other) { } - EIGEN_STRONG_INLINE scalar_multiple_op(const Scalar& other) : m_other(other) { } - EIGEN_STRONG_INLINE Scalar operator() (const Scalar& a) const { return a * m_other; } - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const - { return internal::pmul(a, pset1(m_other)); } - typename add_const_on_value_type::Nested>::type m_other; -}; -template -struct functor_traits > -{ enum { Cost = NumTraits::MulCost, PacketAccess = packet_traits::HasMul }; }; - -template -struct scalar_multiple2_op { - typedef typename scalar_product_traits::ReturnType result_type; - EIGEN_STRONG_INLINE scalar_multiple2_op(const scalar_multiple2_op& other) : m_other(other.m_other) { } - EIGEN_STRONG_INLINE scalar_multiple2_op(const Scalar2& other) : m_other(other) { } - EIGEN_STRONG_INLINE result_type operator() (const Scalar1& a) const { return a * m_other; } - typename add_const_on_value_type::Nested>::type m_other; -}; -template -struct functor_traits > -{ enum { Cost = NumTraits::MulCost, PacketAccess = false }; }; - -/** \internal - * \brief Template functor to divide a scalar by a fixed other one - * - * This functor is used to implement the quotient of a matrix by - * a scalar where the scalar type is not necessarily a floating point type. - * - * \sa class CwiseUnaryOp, MatrixBase::operator/ - */ -template -struct scalar_quotient1_op { - typedef typename packet_traits::type Packet; - // FIXME default copy constructors seems bugged with std::complex<> - EIGEN_STRONG_INLINE scalar_quotient1_op(const scalar_quotient1_op& other) : m_other(other.m_other) { } - EIGEN_STRONG_INLINE scalar_quotient1_op(const Scalar& other) : m_other(other) {} - EIGEN_STRONG_INLINE Scalar operator() (const Scalar& a) const { return a / m_other; } - EIGEN_STRONG_INLINE const Packet packetOp(const Packet& a) const - { return internal::pdiv(a, pset1(m_other)); } - typename add_const_on_value_type::Nested>::type m_other; -}; -template -struct functor_traits > -{ enum { Cost = 2 * NumTraits::MulCost, PacketAccess = packet_traits::HasDiv }; }; - -// nullary functors - -template -struct scalar_constant_op { - typedef typename packet_traits::type Packet; - EIGEN_STRONG_INLINE scalar_constant_op(const scalar_constant_op& other) : m_other(other.m_other) { } - EIGEN_STRONG_INLINE scalar_constant_op(const Scalar& other) : m_other(other) { } - template - EIGEN_STRONG_INLINE const Scalar operator() (Index, Index = 0) const { return m_other; } - template - EIGEN_STRONG_INLINE const Packet packetOp(Index, Index = 0) const { return internal::pset1(m_other); } - const Scalar m_other; -}; -template -struct functor_traits > -// FIXME replace this packet test by a safe one -{ enum { Cost = 1, PacketAccess = packet_traits::Vectorizable, IsRepeatable = true }; }; - -template struct scalar_identity_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_identity_op) - template - EIGEN_STRONG_INLINE const Scalar operator() (Index row, Index col) const { return row==col ? Scalar(1) : Scalar(0); } -}; -template -struct functor_traits > -{ enum { Cost = NumTraits::AddCost, PacketAccess = false, IsRepeatable = true }; }; - -template struct linspaced_op_impl; - -// linear access for packet ops: -// 1) initialization -// base = [low, ..., low] + ([step, ..., step] * [-size, ..., 0]) -// 2) each step (where size is 1 for coeff access or PacketSize for packet access) -// base += [size*step, ..., size*step] -// -// TODO: Perhaps it's better to initialize lazily (so not in the constructor but in packetOp) -// in order to avoid the padd() in operator() ? -template -struct linspaced_op_impl -{ - typedef typename packet_traits::type Packet; - - linspaced_op_impl(const Scalar& low, const Scalar& step) : - m_low(low), m_step(step), - m_packetStep(pset1(packet_traits::size*step)), - m_base(padd(pset1(low), pmul(pset1(step),plset(-packet_traits::size)))) {} - - template - EIGEN_STRONG_INLINE const Scalar operator() (Index i) const - { - m_base = padd(m_base, pset1(m_step)); - return m_low+Scalar(i)*m_step; - } - - template - EIGEN_STRONG_INLINE const Packet packetOp(Index) const { return m_base = padd(m_base,m_packetStep); } - - const Scalar m_low; - const Scalar m_step; - const Packet m_packetStep; - mutable Packet m_base; -}; - -// random access for packet ops: -// 1) each step -// [low, ..., low] + ( [step, ..., step] * ( [i, ..., i] + [0, ..., size] ) ) -template -struct linspaced_op_impl -{ - typedef typename packet_traits::type Packet; - - linspaced_op_impl(const Scalar& low, const Scalar& step) : - m_low(low), m_step(step), - m_lowPacket(pset1(m_low)), m_stepPacket(pset1(m_step)), m_interPacket(plset(0)) {} - - template - EIGEN_STRONG_INLINE const Scalar operator() (Index i) const { return m_low+i*m_step; } - - template - EIGEN_STRONG_INLINE const Packet packetOp(Index i) const - { return internal::padd(m_lowPacket, pmul(m_stepPacket, padd(pset1(Scalar(i)),m_interPacket))); } - - const Scalar m_low; - const Scalar m_step; - const Packet m_lowPacket; - const Packet m_stepPacket; - const Packet m_interPacket; -}; - -// ----- Linspace functor ---------------------------------------------------------------- - -// Forward declaration (we default to random access which does not really give -// us a speed gain when using packet access but it allows to use the functor in -// nested expressions). -template struct linspaced_op; -template struct functor_traits< linspaced_op > -{ enum { Cost = 1, PacketAccess = packet_traits::HasSetLinear, IsRepeatable = true }; }; -template struct linspaced_op -{ - typedef typename packet_traits::type Packet; - linspaced_op(const Scalar& low, const Scalar& high, DenseIndex num_steps) : impl((num_steps==1 ? high : low), (num_steps==1 ? Scalar() : (high-low)/Scalar(num_steps-1))) {} - - template - EIGEN_STRONG_INLINE const Scalar operator() (Index i) const { return impl(i); } - - // We need this function when assigning e.g. a RowVectorXd to a MatrixXd since - // there row==0 and col is used for the actual iteration. - template - EIGEN_STRONG_INLINE const Scalar operator() (Index row, Index col) const - { - eigen_assert(col==0 || row==0); - return impl(col + row); - } - - template - EIGEN_STRONG_INLINE const Packet packetOp(Index i) const { return impl.packetOp(i); } - - // We need this function when assigning e.g. a RowVectorXd to a MatrixXd since - // there row==0 and col is used for the actual iteration. - template - EIGEN_STRONG_INLINE const Packet packetOp(Index row, Index col) const - { - eigen_assert(col==0 || row==0); - return impl.packetOp(col + row); - } - - // This proxy object handles the actual required temporaries, the different - // implementations (random vs. sequential access) as well as the - // correct piping to size 2/4 packet operations. - const linspaced_op_impl impl; -}; - -// all functors allow linear access, except scalar_identity_op. So we fix here a quick meta -// to indicate whether a functor allows linear access, just always answering 'yes' except for -// scalar_identity_op. -// FIXME move this to functor_traits adding a functor_default -template struct functor_has_linear_access { enum { ret = 1 }; }; -template struct functor_has_linear_access > { enum { ret = 0 }; }; - -// In Eigen, any binary op (Product, CwiseBinaryOp) require the Lhs and Rhs to have the same scalar type, except for multiplication -// where the mixing of different types is handled by scalar_product_traits -// In particular, real * complex is allowed. -// FIXME move this to functor_traits adding a functor_default -template struct functor_is_product_like { enum { ret = 0 }; }; -template struct functor_is_product_like > { enum { ret = 1 }; }; -template struct functor_is_product_like > { enum { ret = 1 }; }; -template struct functor_is_product_like > { enum { ret = 1 }; }; - - -/** \internal - * \brief Template functor to add a scalar to a fixed other one - * \sa class CwiseUnaryOp, Array::operator+ - */ -/* If you wonder why doing the pset1() in packetOp() is an optimization check scalar_multiple_op */ -template -struct scalar_add_op { - typedef typename packet_traits::type Packet; - // FIXME default copy constructors seems bugged with std::complex<> - inline scalar_add_op(const scalar_add_op& other) : m_other(other.m_other) { } - inline scalar_add_op(const Scalar& other) : m_other(other) { } - inline Scalar operator() (const Scalar& a) const { return a + m_other; } - inline const Packet packetOp(const Packet& a) const - { return internal::padd(a, pset1(m_other)); } - const Scalar m_other; -}; -template -struct functor_traits > -{ enum { Cost = NumTraits::AddCost, PacketAccess = packet_traits::HasAdd }; }; - -/** \internal - * \brief Template functor to compute the square root of a scalar - * \sa class CwiseUnaryOp, Cwise::sqrt() - */ -template struct scalar_sqrt_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_sqrt_op) - inline const Scalar operator() (const Scalar& a) const { using std::sqrt; return sqrt(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::psqrt(a); } -}; -template -struct functor_traits > -{ enum { - Cost = 5 * NumTraits::MulCost, - PacketAccess = packet_traits::HasSqrt - }; -}; - -/** \internal - * \brief Template functor to compute the cosine of a scalar - * \sa class CwiseUnaryOp, ArrayBase::cos() - */ -template struct scalar_cos_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cos_op) - inline Scalar operator() (const Scalar& a) const { using std::cos; return cos(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::pcos(a); } -}; -template -struct functor_traits > -{ - enum { - Cost = 5 * NumTraits::MulCost, - PacketAccess = packet_traits::HasCos - }; -}; - -/** \internal - * \brief Template functor to compute the sine of a scalar - * \sa class CwiseUnaryOp, ArrayBase::sin() - */ -template struct scalar_sin_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_sin_op) - inline const Scalar operator() (const Scalar& a) const { using std::sin; return sin(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::psin(a); } -}; -template -struct functor_traits > -{ - enum { - Cost = 5 * NumTraits::MulCost, - PacketAccess = packet_traits::HasSin - }; -}; - - -/** \internal - * \brief Template functor to compute the tan of a scalar - * \sa class CwiseUnaryOp, ArrayBase::tan() - */ -template struct scalar_tan_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_tan_op) - inline const Scalar operator() (const Scalar& a) const { using std::tan; return tan(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::ptan(a); } -}; -template -struct functor_traits > -{ - enum { - Cost = 5 * NumTraits::MulCost, - PacketAccess = packet_traits::HasTan - }; -}; - -/** \internal - * \brief Template functor to compute the arc cosine of a scalar - * \sa class CwiseUnaryOp, ArrayBase::acos() - */ -template struct scalar_acos_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_acos_op) - inline const Scalar operator() (const Scalar& a) const { using std::acos; return acos(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::pacos(a); } -}; -template -struct functor_traits > -{ - enum { - Cost = 5 * NumTraits::MulCost, - PacketAccess = packet_traits::HasACos - }; -}; - -/** \internal - * \brief Template functor to compute the arc sine of a scalar - * \sa class CwiseUnaryOp, ArrayBase::asin() - */ -template struct scalar_asin_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_asin_op) - inline const Scalar operator() (const Scalar& a) const { using std::asin; return asin(a); } - typedef typename packet_traits::type Packet; - inline Packet packetOp(const Packet& a) const { return internal::pasin(a); } -}; -template -struct functor_traits > -{ - enum { - Cost = 5 * NumTraits::MulCost, - PacketAccess = packet_traits::HasASin - }; -}; - -/** \internal - * \brief Template functor to raise a scalar to a power - * \sa class CwiseUnaryOp, Cwise::pow - */ -template -struct scalar_pow_op { - // FIXME default copy constructors seems bugged with std::complex<> - inline scalar_pow_op(const scalar_pow_op& other) : m_exponent(other.m_exponent) { } - inline scalar_pow_op(const Scalar& exponent) : m_exponent(exponent) {} - inline Scalar operator() (const Scalar& a) const { return numext::pow(a, m_exponent); } - const Scalar m_exponent; -}; -template -struct functor_traits > -{ enum { Cost = 5 * NumTraits::MulCost, PacketAccess = false }; }; - -/** \internal - * \brief Template functor to compute the quotient between a scalar and array entries. - * \sa class CwiseUnaryOp, Cwise::inverse() - */ -template -struct scalar_inverse_mult_op { - scalar_inverse_mult_op(const Scalar& other) : m_other(other) {} - inline Scalar operator() (const Scalar& a) const { return m_other / a; } - template - inline const Packet packetOp(const Packet& a) const - { return internal::pdiv(pset1(m_other),a); } - Scalar m_other; -}; - -/** \internal - * \brief Template functor to compute the inverse of a scalar - * \sa class CwiseUnaryOp, Cwise::inverse() - */ -template -struct scalar_inverse_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_inverse_op) - inline Scalar operator() (const Scalar& a) const { return Scalar(1)/a; } - template - inline const Packet packetOp(const Packet& a) const - { return internal::pdiv(pset1(Scalar(1)),a); } -}; -template -struct functor_traits > -{ enum { Cost = NumTraits::MulCost, PacketAccess = packet_traits::HasDiv }; }; - -/** \internal - * \brief Template functor to compute the square of a scalar - * \sa class CwiseUnaryOp, Cwise::square() - */ -template -struct scalar_square_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_square_op) - inline Scalar operator() (const Scalar& a) const { return a*a; } - template - inline const Packet packetOp(const Packet& a) const - { return internal::pmul(a,a); } -}; -template -struct functor_traits > -{ enum { Cost = NumTraits::MulCost, PacketAccess = packet_traits::HasMul }; }; - -/** \internal - * \brief Template functor to compute the cube of a scalar - * \sa class CwiseUnaryOp, Cwise::cube() - */ -template -struct scalar_cube_op { - EIGEN_EMPTY_STRUCT_CTOR(scalar_cube_op) - inline Scalar operator() (const Scalar& a) const { return a*a*a; } - template - inline const Packet packetOp(const Packet& a) const - { return internal::pmul(a,pmul(a,a)); } -}; -template -struct functor_traits > -{ enum { Cost = 2*NumTraits::MulCost, PacketAccess = packet_traits::HasMul }; }; - -// default functor traits for STL functors: - -template -struct functor_traits > -{ enum { Cost = NumTraits::MulCost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = NumTraits::MulCost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = NumTraits::AddCost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = NumTraits::AddCost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = NumTraits::AddCost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1, PacketAccess = false }; }; - -#if(__cplusplus < 201103L) -// std::binder* are deprecated since c++11 and will be removed in c++17 -template -struct functor_traits > -{ enum { Cost = functor_traits::Cost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = functor_traits::Cost, PacketAccess = false }; }; -#endif - -template -struct functor_traits > -{ enum { Cost = 1 + functor_traits::Cost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 1 + functor_traits::Cost, PacketAccess = false }; }; - -#ifdef EIGEN_STDEXT_SUPPORT - -template -struct functor_traits > -{ enum { Cost = 0, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = 0, PacketAccess = false }; }; - -template -struct functor_traits > > -{ enum { Cost = 0, PacketAccess = false }; }; - -template -struct functor_traits > > -{ enum { Cost = 0, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = functor_traits::Cost + functor_traits::Cost, PacketAccess = false }; }; - -template -struct functor_traits > -{ enum { Cost = functor_traits::Cost + functor_traits::Cost + functor_traits::Cost, PacketAccess = false }; }; - -#endif // EIGEN_STDEXT_SUPPORT - -// allow to add new functors and specializations of functor_traits from outside Eigen. -// this macro is really needed because functor_traits must be specialized after it is declared but before it is used... -#ifdef EIGEN_FUNCTORS_PLUGIN -#include EIGEN_FUNCTORS_PLUGIN -#endif - -} // end namespace internal - -} // end namespace Eigen - -#endif // EIGEN_FUNCTORS_H diff --git a/thirdparty/eigen/Eigen/src/Core/Fuzzy.h b/thirdparty/eigen/Eigen/src/Core/Fuzzy.h index fe63bd29..43aa49b2 100644 --- a/thirdparty/eigen/Eigen/src/Core/Fuzzy.h +++ b/thirdparty/eigen/Eigen/src/Core/Fuzzy.h @@ -19,18 +19,19 @@ namespace internal template::IsInteger> struct isApprox_selector { + EIGEN_DEVICE_FUNC static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec) { - using std::min; - typename internal::nested::type nested(x); - typename internal::nested::type otherNested(y); - return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * (min)(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum()); + typename internal::nested_eval::type nested(x); + typename internal::nested_eval::type otherNested(y); + return (nested - otherNested).cwiseAbs2().sum() <= prec * prec * numext::mini(nested.cwiseAbs2().sum(), otherNested.cwiseAbs2().sum()); } }; template struct isApprox_selector { + EIGEN_DEVICE_FUNC static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar&) { return x.matrix() == y.matrix(); @@ -40,6 +41,7 @@ struct isApprox_selector template::IsInteger> struct isMuchSmallerThan_object_selector { + EIGEN_DEVICE_FUNC static bool run(const Derived& x, const OtherDerived& y, const typename Derived::RealScalar& prec) { return x.cwiseAbs2().sum() <= numext::abs2(prec) * y.cwiseAbs2().sum(); @@ -49,6 +51,7 @@ struct isMuchSmallerThan_object_selector template struct isMuchSmallerThan_object_selector { + EIGEN_DEVICE_FUNC static bool run(const Derived& x, const OtherDerived&, const typename Derived::RealScalar&) { return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix(); @@ -58,6 +61,7 @@ struct isMuchSmallerThan_object_selector template::IsInteger> struct isMuchSmallerThan_scalar_selector { + EIGEN_DEVICE_FUNC static bool run(const Derived& x, const typename Derived::RealScalar& y, const typename Derived::RealScalar& prec) { return x.cwiseAbs2().sum() <= numext::abs2(prec * y); @@ -67,6 +71,7 @@ struct isMuchSmallerThan_scalar_selector template struct isMuchSmallerThan_scalar_selector { + EIGEN_DEVICE_FUNC static bool run(const Derived& x, const typename Derived::RealScalar&, const typename Derived::RealScalar&) { return x.matrix() == Derived::Zero(x.rows(), x.cols()).matrix(); @@ -95,7 +100,7 @@ struct isMuchSmallerThan_scalar_selector */ template template -bool DenseBase::isApprox( +EIGEN_DEVICE_FUNC bool DenseBase::isApprox( const DenseBase& other, const RealScalar& prec ) const @@ -117,7 +122,7 @@ bool DenseBase::isApprox( * \sa isApprox(), isMuchSmallerThan(const DenseBase&, RealScalar) const */ template -bool DenseBase::isMuchSmallerThan( +EIGEN_DEVICE_FUNC bool DenseBase::isMuchSmallerThan( const typename NumTraits::Real& other, const RealScalar& prec ) const @@ -137,7 +142,7 @@ bool DenseBase::isMuchSmallerThan( */ template template -bool DenseBase::isMuchSmallerThan( +EIGEN_DEVICE_FUNC bool DenseBase::isMuchSmallerThan( const DenseBase& other, const RealScalar& prec ) const diff --git a/thirdparty/eigen/Eigen/src/Core/GeneralProduct.h b/thirdparty/eigen/Eigen/src/Core/GeneralProduct.h index 5744eb71..6906aa75 100644 --- a/thirdparty/eigen/Eigen/src/Core/GeneralProduct.h +++ b/thirdparty/eigen/Eigen/src/Core/GeneralProduct.h @@ -11,46 +11,40 @@ #ifndef EIGEN_GENERAL_PRODUCT_H #define EIGEN_GENERAL_PRODUCT_H -namespace Eigen { - -/** \class GeneralProduct - * \ingroup Core_Module - * - * \brief Expression of the product of two general matrices or vectors - * - * \param LhsNested the type used to store the left-hand side - * \param RhsNested the type used to store the right-hand side - * \param ProductMode the type of the product - * - * This class represents an expression of the product of two general matrices. - * We call a general matrix, a dense matrix with full storage. For instance, - * This excludes triangular, selfadjoint, and sparse matrices. - * It is the return type of the operator* between general matrices. Its template - * arguments are determined automatically by ProductReturnType. Therefore, - * GeneralProduct should never be used direclty. To determine the result type of a - * function which involves a matrix product, use ProductReturnType::Type. - * - * \sa ProductReturnType, MatrixBase::operator*(const MatrixBase&) - */ -template::value> -class GeneralProduct; +namespace Eigen { enum { Large = 2, Small = 3 }; +// Define the threshold value to fallback from the generic matrix-matrix product +// implementation (heavy) to the lightweight coeff-based product one. +// See generic_product_impl +// in products/GeneralMatrixMatrix.h for more details. +// TODO This threshold should also be used in the compile-time selector below. +#ifndef EIGEN_GEMM_TO_COEFFBASED_THRESHOLD +// This default value has been obtained on a Haswell architecture. +#define EIGEN_GEMM_TO_COEFFBASED_THRESHOLD 20 +#endif + namespace internal { template struct product_type_selector; template struct product_size_category { - enum { is_large = MaxSize == Dynamic || - Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD, - value = is_large ? Large - : Size == 1 ? 1 - : Small + enum { + #ifndef EIGEN_GPU_COMPILE_PHASE + is_large = MaxSize == Dynamic || + Size >= EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD || + (Size==Dynamic && MaxSize>=EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD), + #else + is_large = 0, + #endif + value = is_large ? Large + : Size == 1 ? 1 + : Small }; }; @@ -59,15 +53,14 @@ template struct product_type typedef typename remove_all::type _Lhs; typedef typename remove_all::type _Rhs; enum { - MaxRows = _Lhs::MaxRowsAtCompileTime, - Rows = _Lhs::RowsAtCompileTime, - MaxCols = _Rhs::MaxColsAtCompileTime, - Cols = _Rhs::ColsAtCompileTime, - MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::MaxColsAtCompileTime, - _Rhs::MaxRowsAtCompileTime), - Depth = EIGEN_SIZE_MIN_PREFER_FIXED(_Lhs::ColsAtCompileTime, - _Rhs::RowsAtCompileTime), - LargeThreshold = EIGEN_CACHEFRIENDLY_PRODUCT_THRESHOLD + MaxRows = traits<_Lhs>::MaxRowsAtCompileTime, + Rows = traits<_Lhs>::RowsAtCompileTime, + MaxCols = traits<_Rhs>::MaxColsAtCompileTime, + Cols = traits<_Rhs>::ColsAtCompileTime, + MaxDepth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::MaxColsAtCompileTime, + traits<_Rhs>::MaxRowsAtCompileTime), + Depth = EIGEN_SIZE_MIN_PREFER_FIXED(traits<_Lhs>::ColsAtCompileTime, + traits<_Rhs>::RowsAtCompileTime) }; // the splitting into different lines of code here, introducing the _select enums and the typedef below, @@ -82,7 +75,8 @@ template struct product_type public: enum { - value = selector::ret + value = selector::ret, + ret = selector::ret }; #ifdef EIGEN_DEBUG_PRODUCT static void debug() @@ -98,12 +92,13 @@ template struct product_type #endif }; - /* The following allows to select the kind of product at compile time * based on the three dimensions of the product. * This is a compile time mapping from {1,Small,Large}^3 -> {product types} */ // FIXME I'm not sure the current mapping is the ideal one. template struct product_type_selector { enum { ret = OuterProduct }; }; +template struct product_type_selector { enum { ret = LazyCoeffBasedProductMode }; }; +template struct product_type_selector<1, N, 1> { enum { ret = LazyCoeffBasedProductMode }; }; template struct product_type_selector<1, 1, Depth> { enum { ret = InnerProduct }; }; template<> struct product_type_selector<1, 1, 1> { enum { ret = InnerProduct }; }; template<> struct product_type_selector { enum { ret = CoeffBasedProductMode }; }; @@ -122,60 +117,12 @@ template<> struct product_type_selector { enum template<> struct product_type_selector { enum { ret = GemmProduct }; }; template<> struct product_type_selector { enum { ret = GemmProduct }; }; template<> struct product_type_selector { enum { ret = GemmProduct }; }; -template<> struct product_type_selector { enum { ret = GemmProduct }; }; -template<> struct product_type_selector { enum { ret = GemmProduct }; }; +template<> struct product_type_selector { enum { ret = CoeffBasedProductMode }; }; +template<> struct product_type_selector { enum { ret = CoeffBasedProductMode }; }; template<> struct product_type_selector { enum { ret = GemmProduct }; }; } // end namespace internal -/** \class ProductReturnType - * \ingroup Core_Module - * - * \brief Helper class to get the correct and optimized returned type of operator* - * - * \param Lhs the type of the left-hand side - * \param Rhs the type of the right-hand side - * \param ProductMode the type of the product (determined automatically by internal::product_mode) - * - * This class defines the typename Type representing the optimized product expression - * between two matrix expressions. In practice, using ProductReturnType::Type - * is the recommended way to define the result type of a function returning an expression - * which involve a matrix product. The class Product should never be - * used directly. - * - * \sa class Product, MatrixBase::operator*(const MatrixBase&) - */ -template -struct ProductReturnType -{ - // TODO use the nested type to reduce instanciations ???? -// typedef typename internal::nested::type LhsNested; -// typedef typename internal::nested::type RhsNested; - - typedef GeneralProduct Type; -}; - -template -struct ProductReturnType -{ - typedef typename internal::nested::type >::type LhsNested; - typedef typename internal::nested::type >::type RhsNested; - typedef CoeffBasedProduct Type; -}; - -template -struct ProductReturnType -{ - typedef typename internal::nested::type >::type LhsNested; - typedef typename internal::nested::type >::type RhsNested; - typedef CoeffBasedProduct Type; -}; - -// this is a workaround for sun CC -template -struct LazyProductReturnType : public ProductReturnType -{}; - /*********************************************************************** * Implementation of Inner Vector Vector Product ***********************************************************************/ @@ -187,114 +134,10 @@ struct LazyProductReturnType : public ProductReturnType with: operator=(Scalar x); -namespace internal { - -template -struct traits > - : traits::ReturnType,1,1> > -{}; - -} - -template -class GeneralProduct - : internal::no_assignment_operator, - public Matrix::ReturnType,1,1> -{ - typedef Matrix::ReturnType,1,1> Base; - public: - GeneralProduct(const Lhs& lhs, const Rhs& rhs) - { - Base::coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum(); - } - - /** Convertion to scalar */ - operator const typename Base::Scalar() const { - return Base::coeff(0,0); - } -}; - /*********************************************************************** * Implementation of Outer Vector Vector Product ***********************************************************************/ -namespace internal { - -// Column major -template -EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const false_type&) -{ - typedef typename Dest::Index Index; - // FIXME make sure lhs is sequentially stored - // FIXME not very good if rhs is real and lhs complex while alpha is real too - const Index cols = dest.cols(); - for (Index j=0; j -EIGEN_DONT_INLINE void outer_product_selector_run(const ProductType& prod, Dest& dest, const Func& func, const true_type&) { - typedef typename Dest::Index Index; - // FIXME make sure rhs is sequentially stored - // FIXME not very good if lhs is real and rhs complex while alpha is real too - const Index rows = dest.rows(); - for (Index i=0; i -struct traits > - : traits, Lhs, Rhs> > -{}; - -} - -template -class GeneralProduct - : public ProductBase, Lhs, Rhs> -{ - template struct is_row_major : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {}; - - public: - EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct) - - GeneralProduct(const Lhs& lhs, const Rhs& rhs) : Base(lhs,rhs) - { - } - - struct set { template void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() = src; } }; - struct add { template void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } }; - struct sub { template void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } }; - struct adds { - Scalar m_scale; - adds(const Scalar& s) : m_scale(s) {} - template void operator()(const Dst& dst, const Src& src) const { - dst.const_cast_derived() += m_scale * src; - } - }; - - template - inline void evalTo(Dest& dest) const { - internal::outer_product_selector_run(*this, dest, set(), is_row_major()); - } - - template - inline void addTo(Dest& dest) const { - internal::outer_product_selector_run(*this, dest, add(), is_row_major()); - } - - template - inline void subTo(Dest& dest) const { - internal::outer_product_selector_run(*this, dest, sub(), is_row_major()); - } - - template void scaleAndAddTo(Dest& dest, const Scalar& alpha) const - { - internal::outer_product_selector_run(*this, dest, adds(alpha), is_row_major()); - } -}; - /*********************************************************************** * Implementation of General Matrix Vector Product ***********************************************************************/ @@ -308,117 +151,84 @@ class GeneralProduct */ namespace internal { -template -struct traits > - : traits, Lhs, Rhs> > -{}; - template -struct gemv_selector; +struct gemv_dense_selector; } // end namespace internal -template -class GeneralProduct - : public ProductBase, Lhs, Rhs> -{ - public: - EIGEN_PRODUCT_PUBLIC_INTERFACE(GeneralProduct) - - typedef typename Lhs::Scalar LhsScalar; - typedef typename Rhs::Scalar RhsScalar; - - GeneralProduct(const Lhs& a_lhs, const Rhs& a_rhs) : Base(a_lhs,a_rhs) - { -// EIGEN_STATIC_ASSERT((internal::is_same::value), -// YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) - } - - enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight }; - typedef typename internal::conditional::type MatrixType; - - template void scaleAndAddTo(Dest& dst, const Scalar& alpha) const - { - eigen_assert(m_lhs.rows() == dst.rows() && m_rhs.cols() == dst.cols()); - internal::gemv_selector::HasUsableDirectAccess)>::run(*this, dst, alpha); - } -}; - namespace internal { -// The vector is on the left => transposition -template -struct gemv_selector -{ - template - static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha) - { - Transpose destT(dest); - enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor }; - gemv_selector - ::run(GeneralProduct,Transpose, GemvProduct> - (prod.rhs().transpose(), prod.lhs().transpose()), destT, alpha); - } -}; - template struct gemv_static_vector_if; template struct gemv_static_vector_if { - EIGEN_STRONG_INLINE Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; } + EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Scalar* data() { eigen_internal_assert(false && "should never be called"); return 0; } }; template struct gemv_static_vector_if { - EIGEN_STRONG_INLINE Scalar* data() { return 0; } + EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Scalar* data() { return 0; } }; template struct gemv_static_vector_if { - #if EIGEN_ALIGN_STATICALLY - internal::plain_array m_data; - EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; } - #else - // Some architectures cannot align on the stack, - // => let's manually enforce alignment by allocating more data and return the address of the first aligned element. enum { ForceAlignment = internal::packet_traits::Vectorizable, PacketSize = internal::packet_traits::size }; - internal::plain_array m_data; + #if EIGEN_MAX_STATIC_ALIGN_BYTES!=0 + internal::plain_array m_data; + EIGEN_STRONG_INLINE Scalar* data() { return m_data.array; } + #else + // Some architectures cannot align on the stack, + // => let's manually enforce alignment by allocating more data and return the address of the first aligned element. + internal::plain_array m_data; EIGEN_STRONG_INLINE Scalar* data() { return ForceAlignment - ? reinterpret_cast((reinterpret_cast(m_data.array) & ~(size_t(15))) + 16) + ? reinterpret_cast((internal::UIntPtr(m_data.array) & ~(std::size_t(EIGEN_MAX_ALIGN_BYTES-1))) + EIGEN_MAX_ALIGN_BYTES) : m_data.array; } #endif }; -template<> struct gemv_selector +// The vector is on the left => transposition +template +struct gemv_dense_selector +{ + template + static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha) + { + Transpose destT(dest); + enum { OtherStorageOrder = StorageOrder == RowMajor ? ColMajor : RowMajor }; + gemv_dense_selector + ::run(rhs.transpose(), lhs.transpose(), destT, alpha); + } +}; + +template<> struct gemv_dense_selector { - template - static inline void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha) + template + static inline void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha) { - typedef typename ProductType::Index Index; - typedef typename ProductType::LhsScalar LhsScalar; - typedef typename ProductType::RhsScalar RhsScalar; - typedef typename ProductType::Scalar ResScalar; - typedef typename ProductType::RealScalar RealScalar; - typedef typename ProductType::ActualLhsType ActualLhsType; - typedef typename ProductType::ActualRhsType ActualRhsType; - typedef typename ProductType::LhsBlasTraits LhsBlasTraits; - typedef typename ProductType::RhsBlasTraits RhsBlasTraits; - typedef Map, Aligned> MappedDest; - - ActualLhsType actualLhs = LhsBlasTraits::extract(prod.lhs()); - ActualRhsType actualRhs = RhsBlasTraits::extract(prod.rhs()); - - ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs()) - * RhsBlasTraits::extractScalarFactor(prod.rhs()); + typedef typename Lhs::Scalar LhsScalar; + typedef typename Rhs::Scalar RhsScalar; + typedef typename Dest::Scalar ResScalar; + typedef typename Dest::RealScalar RealScalar; + + typedef internal::blas_traits LhsBlasTraits; + typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType; + typedef internal::blas_traits RhsBlasTraits; + typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType; + + typedef Map, EIGEN_PLAIN_ENUM_MIN(AlignedMax,internal::packet_traits::size)> MappedDest; + + ActualLhsType actualLhs = LhsBlasTraits::extract(lhs); + ActualRhsType actualRhs = RhsBlasTraits::extract(rhs); + + ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs); // make sure Dest is a compile-time vector type (bug 1166) typedef typename conditional::type ActualDest; @@ -428,80 +238,96 @@ template<> struct gemv_selector // on, the other hand it is good for the cache to pack the vector anyways... EvalToDestAtCompileTime = (ActualDest::InnerStrideAtCompileTime==1), ComplexByReal = (NumTraits::IsComplex) && (!NumTraits::IsComplex), - MightCannotUseDest = (ActualDest::InnerStrideAtCompileTime!=1) || ComplexByReal + MightCannotUseDest = ((!EvalToDestAtCompileTime) || ComplexByReal) && (ActualDest::MaxSizeAtCompileTime!=0) }; - gemv_static_vector_if static_dest; - - bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0)); - bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible; - + typedef const_blas_data_mapper LhsMapper; + typedef const_blas_data_mapper RhsMapper; RhsScalar compatibleAlpha = get_factor::run(actualAlpha); - ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(), - evalToDest ? dest.data() : static_dest.data()); - - if(!evalToDest) + if(!MightCannotUseDest) { - #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN - int size = dest.size(); - EIGEN_DENSE_STORAGE_CTOR_PLUGIN - #endif - if(!alphaIsCompatible) + // shortcut if we are sure to be able to use dest directly, + // this ease the compiler to generate cleaner and more optimzized code for most common cases + general_matrix_vector_product + ::run( + actualLhs.rows(), actualLhs.cols(), + LhsMapper(actualLhs.data(), actualLhs.outerStride()), + RhsMapper(actualRhs.data(), actualRhs.innerStride()), + dest.data(), 1, + compatibleAlpha); + } + else + { + gemv_static_vector_if static_dest; + + const bool alphaIsCompatible = (!ComplexByReal) || (numext::imag(actualAlpha)==RealScalar(0)); + const bool evalToDest = EvalToDestAtCompileTime && alphaIsCompatible; + + ei_declare_aligned_stack_constructed_variable(ResScalar,actualDestPtr,dest.size(), + evalToDest ? dest.data() : static_dest.data()); + + if(!evalToDest) { - MappedDest(actualDestPtr, dest.size()).setZero(); - compatibleAlpha = RhsScalar(1); + #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN + Index size = dest.size(); + EIGEN_DENSE_STORAGE_CTOR_PLUGIN + #endif + if(!alphaIsCompatible) + { + MappedDest(actualDestPtr, dest.size()).setZero(); + compatibleAlpha = RhsScalar(1); + } + else + MappedDest(actualDestPtr, dest.size()) = dest; } - else - MappedDest(actualDestPtr, dest.size()) = dest; - } - general_matrix_vector_product - ::run( - actualLhs.rows(), actualLhs.cols(), - actualLhs.data(), actualLhs.outerStride(), - actualRhs.data(), actualRhs.innerStride(), - actualDestPtr, 1, - compatibleAlpha); + general_matrix_vector_product + ::run( + actualLhs.rows(), actualLhs.cols(), + LhsMapper(actualLhs.data(), actualLhs.outerStride()), + RhsMapper(actualRhs.data(), actualRhs.innerStride()), + actualDestPtr, 1, + compatibleAlpha); - if (!evalToDest) - { - if(!alphaIsCompatible) - dest += actualAlpha * MappedDest(actualDestPtr, dest.size()); - else - dest = MappedDest(actualDestPtr, dest.size()); + if (!evalToDest) + { + if(!alphaIsCompatible) + dest.matrix() += actualAlpha * MappedDest(actualDestPtr, dest.size()); + else + dest = MappedDest(actualDestPtr, dest.size()); + } } } }; -template<> struct gemv_selector +template<> struct gemv_dense_selector { - template - static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha) + template + static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha) { - typedef typename ProductType::LhsScalar LhsScalar; - typedef typename ProductType::RhsScalar RhsScalar; - typedef typename ProductType::Scalar ResScalar; - typedef typename ProductType::Index Index; - typedef typename ProductType::ActualLhsType ActualLhsType; - typedef typename ProductType::ActualRhsType ActualRhsType; - typedef typename ProductType::_ActualRhsType _ActualRhsType; - typedef typename ProductType::LhsBlasTraits LhsBlasTraits; - typedef typename ProductType::RhsBlasTraits RhsBlasTraits; - - typename add_const::type actualLhs = LhsBlasTraits::extract(prod.lhs()); - typename add_const::type actualRhs = RhsBlasTraits::extract(prod.rhs()); - - ResScalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(prod.lhs()) - * RhsBlasTraits::extractScalarFactor(prod.rhs()); + typedef typename Lhs::Scalar LhsScalar; + typedef typename Rhs::Scalar RhsScalar; + typedef typename Dest::Scalar ResScalar; + + typedef internal::blas_traits LhsBlasTraits; + typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType; + typedef internal::blas_traits RhsBlasTraits; + typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType; + typedef typename internal::remove_all::type ActualRhsTypeCleaned; + + typename add_const::type actualLhs = LhsBlasTraits::extract(lhs); + typename add_const::type actualRhs = RhsBlasTraits::extract(rhs); + + ResScalar actualAlpha = combine_scalar_factors(alpha, lhs, rhs); enum { // FIXME find a way to allow an inner stride on the result if packet_traits::size==1 // on, the other hand it is good for the cache to pack the vector anyways... - DirectlyUseRhs = _ActualRhsType::InnerStrideAtCompileTime==1 + DirectlyUseRhs = ActualRhsTypeCleaned::InnerStrideAtCompileTime==1 || ActualRhsTypeCleaned::MaxSizeAtCompileTime==0 }; - gemv_static_vector_if static_rhs; + gemv_static_vector_if static_rhs; ei_declare_aligned_stack_constructed_variable(RhsScalar,actualRhsPtr,actualRhs.size(), DirectlyUseRhs ? const_cast(actualRhs.data()) : static_rhs.data()); @@ -509,45 +335,48 @@ template<> struct gemv_selector if(!DirectlyUseRhs) { #ifdef EIGEN_DENSE_STORAGE_CTOR_PLUGIN - int size = actualRhs.size(); + Index size = actualRhs.size(); EIGEN_DENSE_STORAGE_CTOR_PLUGIN #endif - Map(actualRhsPtr, actualRhs.size()) = actualRhs; + Map(actualRhsPtr, actualRhs.size()) = actualRhs; } + typedef const_blas_data_mapper LhsMapper; + typedef const_blas_data_mapper RhsMapper; general_matrix_vector_product - ::run( + ::run( actualLhs.rows(), actualLhs.cols(), - actualLhs.data(), actualLhs.outerStride(), - actualRhsPtr, 1, + LhsMapper(actualLhs.data(), actualLhs.outerStride()), + RhsMapper(actualRhsPtr, 1), dest.data(), dest.col(0).innerStride(), //NOTE if dest is not a vector at compile-time, then dest.innerStride() might be wrong. (bug 1166) actualAlpha); } }; -template<> struct gemv_selector +template<> struct gemv_dense_selector { - template - static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha) + template + static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha) { - typedef typename Dest::Index Index; - // TODO makes sure dest is sequentially stored in memory, otherwise use a temp - const Index size = prod.rhs().rows(); + EIGEN_STATIC_ASSERT((!nested_eval::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE); + // TODO if rhs is large enough it might be beneficial to make sure that dest is sequentially stored in memory, otherwise use a temp + typename nested_eval::type actual_rhs(rhs); + const Index size = rhs.rows(); for(Index k=0; k struct gemv_selector +template<> struct gemv_dense_selector { - template - static void run(const ProductType& prod, Dest& dest, const typename ProductType::Scalar& alpha) + template + static void run(const Lhs &lhs, const Rhs &rhs, Dest& dest, const typename Dest::Scalar& alpha) { - typedef typename Dest::Index Index; - // TODO makes sure rhs is sequentially stored in memory, otherwise use a temp - const Index rows = prod.rows(); + EIGEN_STATIC_ASSERT((!nested_eval::Evaluate),EIGEN_INTERNAL_COMPILATION_ERROR_OR_YOU_MADE_A_PROGRAMMING_MISTAKE); + typename nested_eval::type actual_rhs(rhs); + const Index rows = dest.rows(); for(Index i=0; i struct gemv_selector */ template template -inline const typename ProductReturnType::Type +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +const Product MatrixBase::operator*(const MatrixBase &other) const { // A note regarding the function declaration: In MSVC, this function will sometimes @@ -590,7 +420,8 @@ MatrixBase::operator*(const MatrixBase &other) const #ifdef EIGEN_DEBUG_PRODUCT internal::product_type::debug(); #endif - return typename ProductReturnType::Type(derived(), other.derived()); + + return Product(derived(), other.derived()); } /** \returns an expression of the matrix product of \c *this and \a other without implicit evaluation. @@ -606,7 +437,8 @@ MatrixBase::operator*(const MatrixBase &other) const */ template template -const typename LazyProductReturnType::Type +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +const Product MatrixBase::lazyProduct(const MatrixBase &other) const { enum { @@ -625,7 +457,7 @@ MatrixBase::lazyProduct(const MatrixBase &other) const INVALID_MATRIX_PRODUCT__IF_YOU_WANTED_A_COEFF_WISE_PRODUCT_YOU_MUST_USE_THE_EXPLICIT_FUNCTION) EIGEN_STATIC_ASSERT(ProductIsValid || SameSizes, INVALID_MATRIX_PRODUCT) - return typename LazyProductReturnType::Type(derived(), other.derived()); + return Product(derived(), other.derived()); } } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/GenericPacketMath.h b/thirdparty/eigen/Eigen/src/Core/GenericPacketMath.h index c6e93bbb..4aeb77de 100644 --- a/thirdparty/eigen/Eigen/src/Core/GenericPacketMath.h +++ b/thirdparty/eigen/Eigen/src/Core/GenericPacketMath.h @@ -42,21 +42,34 @@ namespace internal { struct default_packet_traits { enum { - HasAdd = 1, - HasSub = 1, - HasMul = 1, - HasNegate = 1, - HasAbs = 1, - HasAbs2 = 1, - HasMin = 1, - HasMax = 1, - HasConj = 1, + HasHalfPacket = 0, + + HasAdd = 1, + HasSub = 1, + HasShift = 1, + HasMul = 1, + HasNegate = 1, + HasAbs = 1, + HasArg = 0, + HasAbs2 = 1, + HasAbsDiff = 0, + HasMin = 1, + HasMax = 1, + HasConj = 1, HasSetLinear = 1, + HasBlend = 0, + // This flag is used to indicate whether packet comparison is supported. + // pcmp_eq, pcmp_lt and pcmp_le should be defined for it to be true. + HasCmp = 0, HasDiv = 0, HasSqrt = 0, + HasRsqrt = 0, HasExp = 0, + HasExpm1 = 0, HasLog = 0, + HasLog1p = 0, + HasLog10 = 0, HasPow = 0, HasSin = 0, @@ -64,17 +77,41 @@ struct default_packet_traits HasTan = 0, HasASin = 0, HasACos = 0, - HasATan = 0 + HasATan = 0, + HasSinh = 0, + HasCosh = 0, + HasTanh = 0, + HasLGamma = 0, + HasDiGamma = 0, + HasZeta = 0, + HasPolygamma = 0, + HasErf = 0, + HasErfc = 0, + HasNdtri = 0, + HasBessel = 0, + HasIGamma = 0, + HasIGammaDerA = 0, + HasGammaSampleDerAlpha = 0, + HasIGammac = 0, + HasBetaInc = 0, + + HasRound = 0, + HasRint = 0, + HasFloor = 0, + HasCeil = 0, + HasSign = 0 }; }; template struct packet_traits : default_packet_traits { typedef T type; + typedef T half; enum { Vectorizable = 0, size = 1, - AlignedOnScalar = 0 + AlignedOnScalar = 0, + HasHalfPacket = 0 }; enum { HasAdd = 0, @@ -90,140 +127,624 @@ template struct packet_traits : default_packet_traits }; }; +template struct packet_traits : packet_traits { }; + +template struct unpacket_traits +{ + typedef T type; + typedef T half; + enum + { + size = 1, + alignment = 1, + vectorizable = false, + masked_load_available=false, + masked_store_available=false + }; +}; + +template struct unpacket_traits : unpacket_traits { }; + +template struct type_casting_traits { + enum { + VectorizedCast = 0, + SrcCoeffRatio = 1, + TgtCoeffRatio = 1 + }; +}; + +/** \internal Wrapper to ensure that multiple packet types can map to the same + same underlying vector type. */ +template +struct eigen_packet_wrapper +{ + EIGEN_ALWAYS_INLINE operator T&() { return m_val; } + EIGEN_ALWAYS_INLINE operator const T&() const { return m_val; } +#if EIGEN_HAS_CXX11 + EIGEN_ALWAYS_INLINE eigen_packet_wrapper() = default; +#else + EIGEN_ALWAYS_INLINE eigen_packet_wrapper() {}; +#endif + EIGEN_ALWAYS_INLINE eigen_packet_wrapper(const T &v) : m_val(v) {} + EIGEN_ALWAYS_INLINE eigen_packet_wrapper& operator=(const T &v) { + m_val = v; + return *this; + } + + T m_val; +}; + + +/** \internal A convenience utility for determining if the type is a scalar. + * This is used to enable some generic packet implementations. + */ +template +struct is_scalar { + typedef typename unpacket_traits::type Scalar; + enum { + value = internal::is_same::value + }; +}; + +/** \internal \returns static_cast(a) (coeff-wise) */ +template +EIGEN_DEVICE_FUNC inline TgtPacket +pcast(const SrcPacket& a) { + return static_cast(a); +} +template +EIGEN_DEVICE_FUNC inline TgtPacket +pcast(const SrcPacket& a, const SrcPacket& /*b*/) { + return static_cast(a); +} +template +EIGEN_DEVICE_FUNC inline TgtPacket +pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/) { + return static_cast(a); +} +template +EIGEN_DEVICE_FUNC inline TgtPacket +pcast(const SrcPacket& a, const SrcPacket& /*b*/, const SrcPacket& /*c*/, const SrcPacket& /*d*/, + const SrcPacket& /*e*/, const SrcPacket& /*f*/, const SrcPacket& /*g*/, const SrcPacket& /*h*/) { + return static_cast(a); +} + +/** \internal \returns reinterpret_cast(a) */ +template +EIGEN_DEVICE_FUNC inline Target +preinterpret(const Packet& a); /* { return reinterpret_cast(a); } */ + /** \internal \returns a + b (coeff-wise) */ -template inline Packet -padd(const Packet& a, - const Packet& b) { return a+b; } +template EIGEN_DEVICE_FUNC inline Packet +padd(const Packet& a, const Packet& b) { return a+b; } +// Avoid compiler warning for boolean algebra. +template<> EIGEN_DEVICE_FUNC inline bool +padd(const bool& a, const bool& b) { return a || b; } /** \internal \returns a - b (coeff-wise) */ -template inline Packet -psub(const Packet& a, - const Packet& b) { return a-b; } +template EIGEN_DEVICE_FUNC inline Packet +psub(const Packet& a, const Packet& b) { return a-b; } /** \internal \returns -a (coeff-wise) */ -template inline Packet +template EIGEN_DEVICE_FUNC inline Packet pnegate(const Packet& a) { return -a; } +template<> EIGEN_DEVICE_FUNC inline bool +pnegate(const bool& a) { return !a; } + /** \internal \returns conj(a) (coeff-wise) */ -template inline Packet +template EIGEN_DEVICE_FUNC inline Packet pconj(const Packet& a) { return numext::conj(a); } /** \internal \returns a * b (coeff-wise) */ -template inline Packet -pmul(const Packet& a, - const Packet& b) { return a*b; } +template EIGEN_DEVICE_FUNC inline Packet +pmul(const Packet& a, const Packet& b) { return a*b; } +// Avoid compiler warning for boolean algebra. +template<> EIGEN_DEVICE_FUNC inline bool +pmul(const bool& a, const bool& b) { return a && b; } /** \internal \returns a / b (coeff-wise) */ -template inline Packet -pdiv(const Packet& a, - const Packet& b) { return a/b; } +template EIGEN_DEVICE_FUNC inline Packet +pdiv(const Packet& a, const Packet& b) { return a/b; } + +// In the generic case, memset to all one bits. +template +struct ptrue_impl { + static EIGEN_DEVICE_FUNC inline Packet run(const Packet& /*a*/){ + Packet b; + memset(static_cast(&b), 0xff, sizeof(Packet)); + return b; + } +}; -/** \internal \returns the min of \a a and \a b (coeff-wise) */ -template inline Packet -pmin(const Packet& a, - const Packet& b) { using std::min; return (min)(a, b); } +// For non-trivial scalars, set to Scalar(1) (i.e. a non-zero value). +// Although this is technically not a valid bitmask, the scalar path for pselect +// uses a comparison to zero, so this should still work in most cases. We don't +// have another option, since the scalar type requires initialization. +template +struct ptrue_impl::value && NumTraits::RequireInitialization>::type > { + static EIGEN_DEVICE_FUNC inline T run(const T& /*a*/){ + return T(1); + } +}; + +/** \internal \returns one bits. */ +template EIGEN_DEVICE_FUNC inline Packet +ptrue(const Packet& a) { + return ptrue_impl::run(a); +} -/** \internal \returns the max of \a a and \a b (coeff-wise) */ -template inline Packet -pmax(const Packet& a, - const Packet& b) { using std::max; return (max)(a, b); } +// In the general case, memset to zero. +template +struct pzero_impl { + static EIGEN_DEVICE_FUNC inline Packet run(const Packet& /*a*/) { + Packet b; + memset(static_cast(&b), 0x00, sizeof(Packet)); + return b; + } +}; -/** \internal \returns the absolute value of \a a */ -template inline Packet -pabs(const Packet& a) { using std::abs; return abs(a); } +// For scalars, explicitly set to Scalar(0), since the underlying representation +// for zero may not consist of all-zero bits. +template +struct pzero_impl::value>::type> { + static EIGEN_DEVICE_FUNC inline T run(const T& /*a*/) { + return T(0); + } +}; + +/** \internal \returns packet of zeros */ +template EIGEN_DEVICE_FUNC inline Packet +pzero(const Packet& a) { + return pzero_impl::run(a); +} + +/** \internal \returns a <= b as a bit mask */ +template EIGEN_DEVICE_FUNC inline Packet +pcmp_le(const Packet& a, const Packet& b) { return a<=b ? ptrue(a) : pzero(a); } + +/** \internal \returns a < b as a bit mask */ +template EIGEN_DEVICE_FUNC inline Packet +pcmp_lt(const Packet& a, const Packet& b) { return a EIGEN_DEVICE_FUNC inline Packet +pcmp_eq(const Packet& a, const Packet& b) { return a==b ? ptrue(a) : pzero(a); } + +/** \internal \returns a < b or a==NaN or b==NaN as a bit mask */ +template EIGEN_DEVICE_FUNC inline Packet +pcmp_lt_or_nan(const Packet& a, const Packet& b) { return a>=b ? pzero(a) : ptrue(a); } + +template +struct bit_and { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a, const T& b) const { + return a & b; + } +}; + +template +struct bit_or { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a, const T& b) const { + return a | b; + } +}; + +template +struct bit_xor { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a, const T& b) const { + return a ^ b; + } +}; + +template +struct bit_not { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR EIGEN_ALWAYS_INLINE T operator()(const T& a) const { + return ~a; + } +}; + +// Use operators &, |, ^, ~. +template +struct operator_bitwise_helper { + EIGEN_DEVICE_FUNC static inline T bitwise_and(const T& a, const T& b) { return bit_and()(a, b); } + EIGEN_DEVICE_FUNC static inline T bitwise_or(const T& a, const T& b) { return bit_or()(a, b); } + EIGEN_DEVICE_FUNC static inline T bitwise_xor(const T& a, const T& b) { return bit_xor()(a, b); } + EIGEN_DEVICE_FUNC static inline T bitwise_not(const T& a) { return bit_not()(a); } +}; + +// Apply binary operations byte-by-byte +template +struct bytewise_bitwise_helper { + EIGEN_DEVICE_FUNC static inline T bitwise_and(const T& a, const T& b) { + return binary(a, b, bit_and()); + } + EIGEN_DEVICE_FUNC static inline T bitwise_or(const T& a, const T& b) { + return binary(a, b, bit_or()); + } + EIGEN_DEVICE_FUNC static inline T bitwise_xor(const T& a, const T& b) { + return binary(a, b, bit_xor()); + } + EIGEN_DEVICE_FUNC static inline T bitwise_not(const T& a) { + return unary(a,bit_not()); + } + + private: + template + EIGEN_DEVICE_FUNC static inline T unary(const T& a, Op op) { + const unsigned char* a_ptr = reinterpret_cast(&a); + T c; + unsigned char* c_ptr = reinterpret_cast(&c); + for (size_t i = 0; i < sizeof(T); ++i) { + *c_ptr++ = op(*a_ptr++); + } + return c; + } + + template + EIGEN_DEVICE_FUNC static inline T binary(const T& a, const T& b, Op op) { + const unsigned char* a_ptr = reinterpret_cast(&a); + const unsigned char* b_ptr = reinterpret_cast(&b); + T c; + unsigned char* c_ptr = reinterpret_cast(&c); + for (size_t i = 0; i < sizeof(T); ++i) { + *c_ptr++ = op(*a_ptr++, *b_ptr++); + } + return c; + } +}; + +// In the general case, use byte-by-byte manipulation. +template +struct bitwise_helper : public bytewise_bitwise_helper {}; + +// For integers or non-trivial scalars, use binary operators. +template +struct bitwise_helper::value && (NumTraits::IsInteger || NumTraits::RequireInitialization)>::type + > : public operator_bitwise_helper {}; /** \internal \returns the bitwise and of \a a and \a b */ -template inline Packet -pand(const Packet& a, const Packet& b) { return a & b; } +template EIGEN_DEVICE_FUNC inline Packet +pand(const Packet& a, const Packet& b) { + return bitwise_helper::bitwise_and(a, b); +} /** \internal \returns the bitwise or of \a a and \a b */ -template inline Packet -por(const Packet& a, const Packet& b) { return a | b; } +template EIGEN_DEVICE_FUNC inline Packet +por(const Packet& a, const Packet& b) { + return bitwise_helper::bitwise_or(a, b); +} /** \internal \returns the bitwise xor of \a a and \a b */ -template inline Packet -pxor(const Packet& a, const Packet& b) { return a ^ b; } +template EIGEN_DEVICE_FUNC inline Packet +pxor(const Packet& a, const Packet& b) { + return bitwise_helper::bitwise_xor(a, b); +} + +/** \internal \returns the bitwise not of \a a */ +template EIGEN_DEVICE_FUNC inline Packet +pnot(const Packet& a) { + return bitwise_helper::bitwise_not(a); +} + +/** \internal \returns the bitwise and of \a a and not \a b */ +template EIGEN_DEVICE_FUNC inline Packet +pandnot(const Packet& a, const Packet& b) { return pand(a, pnot(b)); } + +// In the general case, use bitwise select. +template +struct pselect_impl { + static EIGEN_DEVICE_FUNC inline Packet run(const Packet& mask, const Packet& a, const Packet& b) { + return por(pand(a,mask),pandnot(b,mask)); + } +}; + +// For scalars, use ternary select. +template +struct pselect_impl::value>::type > { + static EIGEN_DEVICE_FUNC inline Packet run(const Packet& mask, const Packet& a, const Packet& b) { + return numext::equal_strict(mask, Packet(0)) ? b : a; + } +}; -/** \internal \returns the bitwise andnot of \a a and \a b */ -template inline Packet -pandnot(const Packet& a, const Packet& b) { return a & (!b); } +/** \internal \returns \a or \b for each field in packet according to \mask */ +template EIGEN_DEVICE_FUNC inline Packet +pselect(const Packet& mask, const Packet& a, const Packet& b) { + return pselect_impl::run(mask, a, b); +} + +template<> EIGEN_DEVICE_FUNC inline bool pselect( + const bool& cond, const bool& a, const bool& b) { + return cond ? a : b; +} + +/** \internal \returns the min or of \a a and \a b (coeff-wise) + If either \a a or \a b are NaN, the result is implementation defined. */ +template +struct pminmax_impl { + template + static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a, const Packet& b, Op op) { + return op(a,b); + } +}; + +/** \internal \returns the min or max of \a a and \a b (coeff-wise) + If either \a a or \a b are NaN, NaN is returned. */ +template<> +struct pminmax_impl { + template + static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a, const Packet& b, Op op) { + Packet not_nan_mask_a = pcmp_eq(a, a); + Packet not_nan_mask_b = pcmp_eq(b, b); + return pselect(not_nan_mask_a, + pselect(not_nan_mask_b, op(a, b), b), + a); + } +}; + +/** \internal \returns the min or max of \a a and \a b (coeff-wise) + If both \a a and \a b are NaN, NaN is returned. + Equivalent to std::fmin(a, b). */ +template<> +struct pminmax_impl { + template + static EIGEN_DEVICE_FUNC inline Packet run(const Packet& a, const Packet& b, Op op) { + Packet not_nan_mask_a = pcmp_eq(a, a); + Packet not_nan_mask_b = pcmp_eq(b, b); + return pselect(not_nan_mask_a, + pselect(not_nan_mask_b, op(a, b), a), + b); + } +}; + + +#ifndef SYCL_DEVICE_ONLY +#define EIGEN_BINARY_OP_NAN_PROPAGATION(Type, Func) Func +#else +#define EIGEN_BINARY_OP_NAN_PROPAGATION(Type, Func) \ +[](const Type& a, const Type& b) { \ + return Func(a, b);} +#endif + +/** \internal \returns the min of \a a and \a b (coeff-wise). + If \a a or \b b is NaN, the return value is implementation defined. */ +template EIGEN_DEVICE_FUNC inline Packet +pmin(const Packet& a, const Packet& b) { return numext::mini(a,b); } + +/** \internal \returns the min of \a a and \a b (coeff-wise). + NaNPropagation determines the NaN propagation semantics. */ +template +EIGEN_DEVICE_FUNC inline Packet pmin(const Packet& a, const Packet& b) { + return pminmax_impl::run(a, b, EIGEN_BINARY_OP_NAN_PROPAGATION(Packet, (pmin))); +} + +/** \internal \returns the max of \a a and \a b (coeff-wise) + If \a a or \b b is NaN, the return value is implementation defined. */ +template EIGEN_DEVICE_FUNC inline Packet +pmax(const Packet& a, const Packet& b) { return numext::maxi(a, b); } + +/** \internal \returns the max of \a a and \a b (coeff-wise). + NaNPropagation determines the NaN propagation semantics. */ +template +EIGEN_DEVICE_FUNC inline Packet pmax(const Packet& a, const Packet& b) { + return pminmax_impl::run(a, b, EIGEN_BINARY_OP_NAN_PROPAGATION(Packet,(pmax))); +} + +/** \internal \returns the absolute value of \a a */ +template EIGEN_DEVICE_FUNC inline Packet +pabs(const Packet& a) { return numext::abs(a); } +template<> EIGEN_DEVICE_FUNC inline unsigned int +pabs(const unsigned int& a) { return a; } +template<> EIGEN_DEVICE_FUNC inline unsigned long +pabs(const unsigned long& a) { return a; } +template<> EIGEN_DEVICE_FUNC inline unsigned long long +pabs(const unsigned long long& a) { return a; } + +/** \internal \returns the addsub value of \a a,b */ +template EIGEN_DEVICE_FUNC inline Packet +paddsub(const Packet& a, const Packet& b) { + return pselect(peven_mask(a), padd(a, b), psub(a, b)); + } + +/** \internal \returns the phase angle of \a a */ +template EIGEN_DEVICE_FUNC inline Packet +parg(const Packet& a) { using numext::arg; return arg(a); } + + +/** \internal \returns \a a logically shifted by N bits to the right */ +template EIGEN_DEVICE_FUNC inline int +parithmetic_shift_right(const int& a) { return a >> N; } +template EIGEN_DEVICE_FUNC inline long int +parithmetic_shift_right(const long int& a) { return a >> N; } + +/** \internal \returns \a a arithmetically shifted by N bits to the right */ +template EIGEN_DEVICE_FUNC inline int +plogical_shift_right(const int& a) { return static_cast(static_cast(a) >> N); } +template EIGEN_DEVICE_FUNC inline long int +plogical_shift_right(const long int& a) { return static_cast(static_cast(a) >> N); } + +/** \internal \returns \a a shifted by N bits to the left */ +template EIGEN_DEVICE_FUNC inline int +plogical_shift_left(const int& a) { return a << N; } +template EIGEN_DEVICE_FUNC inline long int +plogical_shift_left(const long int& a) { return a << N; } + +/** \internal \returns the significant and exponent of the underlying floating point numbers + * See https://en.cppreference.com/w/cpp/numeric/math/frexp + */ +template +EIGEN_DEVICE_FUNC inline Packet pfrexp(const Packet& a, Packet& exponent) { + int exp; + EIGEN_USING_STD(frexp); + Packet result = static_cast(frexp(a, &exp)); + exponent = static_cast(exp); + return result; +} + +/** \internal \returns a * 2^((int)exponent) + * See https://en.cppreference.com/w/cpp/numeric/math/ldexp + */ +template EIGEN_DEVICE_FUNC inline Packet +pldexp(const Packet &a, const Packet &exponent) { + EIGEN_USING_STD(ldexp) + return static_cast(ldexp(a, static_cast(exponent))); +} + +/** \internal \returns the min of \a a and \a b (coeff-wise) */ +template EIGEN_DEVICE_FUNC inline Packet +pabsdiff(const Packet& a, const Packet& b) { return pselect(pcmp_lt(a, b), psub(b, a), psub(a, b)); } /** \internal \returns a packet version of \a *from, from must be 16 bytes aligned */ -template inline Packet +template EIGEN_DEVICE_FUNC inline Packet pload(const typename unpacket_traits::type* from) { return *from; } /** \internal \returns a packet version of \a *from, (un-aligned load) */ -template inline Packet +template EIGEN_DEVICE_FUNC inline Packet ploadu(const typename unpacket_traits::type* from) { return *from; } -/** \internal \returns a packet with elements of \a *from duplicated. - * For instance, for a packet of 8 elements, 4 scalar will be read from \a *from and - * duplicated to form: {from[0],from[0],from[1],from[1],,from[2],from[2],,from[3],from[3]} - * Currently, this function is only used for scalar * complex products. +/** \internal \returns a packet version of \a *from, (un-aligned masked load) + * There is no generic implementation. We only have implementations for specialized + * cases. Generic case should not be called. */ -template inline Packet -ploaddup(const typename unpacket_traits::type* from) { return *from; } +template EIGEN_DEVICE_FUNC inline +typename enable_if::masked_load_available, Packet>::type +ploadu(const typename unpacket_traits::type* from, typename unpacket_traits::mask_t umask); /** \internal \returns a packet with constant coefficients \a a, e.g.: (a,a,a,a) */ -template inline Packet +template EIGEN_DEVICE_FUNC inline Packet pset1(const typename unpacket_traits::type& a) { return a; } +/** \internal \returns a packet with constant coefficients set from bits */ +template EIGEN_DEVICE_FUNC inline Packet +pset1frombits(BitsType a); + +/** \internal \returns a packet with constant coefficients \a a[0], e.g.: (a[0],a[0],a[0],a[0]) */ +template EIGEN_DEVICE_FUNC inline Packet +pload1(const typename unpacket_traits::type *a) { return pset1(*a); } + +/** \internal \returns a packet with elements of \a *from duplicated. + * For instance, for a packet of 8 elements, 4 scalars will be read from \a *from and + * duplicated to form: {from[0],from[0],from[1],from[1],from[2],from[2],from[3],from[3]} + * Currently, this function is only used for scalar * complex products. + */ +template EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet +ploaddup(const typename unpacket_traits::type* from) { return *from; } + +/** \internal \returns a packet with elements of \a *from quadrupled. + * For instance, for a packet of 8 elements, 2 scalars will be read from \a *from and + * replicated to form: {from[0],from[0],from[0],from[0],from[1],from[1],from[1],from[1]} + * Currently, this function is only used in matrix products. + * For packet-size smaller or equal to 4, this function is equivalent to pload1 + */ +template EIGEN_DEVICE_FUNC inline Packet +ploadquad(const typename unpacket_traits::type* from) +{ return pload1(from); } + +/** \internal equivalent to + * \code + * a0 = pload1(a+0); + * a1 = pload1(a+1); + * a2 = pload1(a+2); + * a3 = pload1(a+3); + * \endcode + * \sa pset1, pload1, ploaddup, pbroadcast2 + */ +template EIGEN_DEVICE_FUNC +inline void pbroadcast4(const typename unpacket_traits::type *a, + Packet& a0, Packet& a1, Packet& a2, Packet& a3) +{ + a0 = pload1(a+0); + a1 = pload1(a+1); + a2 = pload1(a+2); + a3 = pload1(a+3); +} + +/** \internal equivalent to + * \code + * a0 = pload1(a+0); + * a1 = pload1(a+1); + * \endcode + * \sa pset1, pload1, ploaddup, pbroadcast4 + */ +template EIGEN_DEVICE_FUNC +inline void pbroadcast2(const typename unpacket_traits::type *a, + Packet& a0, Packet& a1) +{ + a0 = pload1(a+0); + a1 = pload1(a+1); +} + /** \internal \brief Returns a packet with coefficients (a,a+1,...,a+packet_size-1). */ -template inline typename packet_traits::type -plset(const Scalar& a) { return a; } +template EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Packet +plset(const typename unpacket_traits::type& a) { return a; } + +/** \internal \returns a packet with constant coefficients \a a, e.g.: (x, 0, x, 0), + where x is the value of all 1-bits. */ +template EIGEN_DEVICE_FUNC inline Packet +peven_mask(const Packet& /*a*/) { + typedef typename unpacket_traits::type Scalar; + const size_t n = unpacket_traits::size; + EIGEN_ALIGN_TO_BOUNDARY(sizeof(Packet)) Scalar elements[n]; + for(size_t i = 0; i < n; ++i) { + memset(elements+i, ((i & 1) == 0 ? 0xff : 0), sizeof(Scalar)); + } + return ploadu(elements); +} + /** \internal copy the packet \a from to \a *to, \a to must be 16 bytes aligned */ -template inline void pstore(Scalar* to, const Packet& from) +template EIGEN_DEVICE_FUNC inline void pstore(Scalar* to, const Packet& from) { (*to) = from; } /** \internal copy the packet \a from to \a *to, (un-aligned store) */ -template inline void pstoreu(Scalar* to, const Packet& from) -{ (*to) = from; } +template EIGEN_DEVICE_FUNC inline void pstoreu(Scalar* to, const Packet& from) +{ (*to) = from; } + +/** \internal copy the packet \a from to \a *to, (un-aligned store with a mask) + * There is no generic implementation. We only have implementations for specialized + * cases. Generic case should not be called. + */ +template +EIGEN_DEVICE_FUNC inline +typename enable_if::masked_store_available, void>::type +pstoreu(Scalar* to, const Packet& from, typename unpacket_traits::mask_t umask); + + template EIGEN_DEVICE_FUNC inline Packet pgather(const Scalar* from, Index /*stride*/) + { return ploadu(from); } + + template EIGEN_DEVICE_FUNC inline void pscatter(Scalar* to, const Packet& from, Index /*stride*/) + { pstore(to, from); } /** \internal tries to do cache prefetching of \a addr */ -template inline void prefetch(const Scalar* addr) +template EIGEN_DEVICE_FUNC inline void prefetch(const Scalar* addr) { -#if (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC) +#if defined(EIGEN_HIP_DEVICE_COMPILE) + // do nothing +#elif defined(EIGEN_CUDA_ARCH) +#if defined(__LP64__) || EIGEN_OS_WIN64 + // 64-bit pointer operand constraint for inlined asm + asm(" prefetch.L1 [ %1 ];" : "=l"(addr) : "l"(addr)); +#else + // 32-bit pointer operand constraint for inlined asm + asm(" prefetch.L1 [ %1 ];" : "=r"(addr) : "r"(addr)); +#endif +#elif (!EIGEN_COMP_MSVC) && (EIGEN_COMP_GNUC || EIGEN_COMP_CLANG || EIGEN_COMP_ICC) __builtin_prefetch(addr); #endif } -/** \internal \returns the first element of a packet */ -template inline typename unpacket_traits::type pfirst(const Packet& a) -{ return a; } - -/** \internal \returns a packet where the element i contains the sum of the packet of \a vec[i] */ -template inline Packet -preduxp(const Packet* vecs) { return vecs[0]; } - -/** \internal \returns the sum of the elements of \a a*/ -template inline typename unpacket_traits::type predux(const Packet& a) -{ return a; } - -/** \internal \returns the product of the elements of \a a*/ -template inline typename unpacket_traits::type predux_mul(const Packet& a) -{ return a; } - -/** \internal \returns the min of the elements of \a a*/ -template inline typename unpacket_traits::type predux_min(const Packet& a) -{ return a; } - -/** \internal \returns the max of the elements of \a a*/ -template inline typename unpacket_traits::type predux_max(const Packet& a) -{ return a; } - /** \internal \returns the reversed elements of \a a*/ -template inline Packet preverse(const Packet& a) +template EIGEN_DEVICE_FUNC inline Packet preverse(const Packet& a) { return a; } - /** \internal \returns \a a with real and imaginary part flipped (for complex type only) */ -template inline Packet pcplxflip(const Packet& a) +template EIGEN_DEVICE_FUNC inline Packet pcplxflip(const Packet& a) { - // FIXME: uncomment the following in case we drop the internal imag and real functions. -// using std::imag; -// using std::real; - return Packet(imag(a),real(a)); + return Packet(numext::imag(a),numext::real(a)); } /************************** @@ -232,41 +753,201 @@ template inline Packet pcplxflip(const Packet& a) /** \internal \returns the sine of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet psin(const Packet& a) { using std::sin; return sin(a); } +Packet psin(const Packet& a) { EIGEN_USING_STD(sin); return sin(a); } /** \internal \returns the cosine of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet pcos(const Packet& a) { using std::cos; return cos(a); } +Packet pcos(const Packet& a) { EIGEN_USING_STD(cos); return cos(a); } /** \internal \returns the tan of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet ptan(const Packet& a) { using std::tan; return tan(a); } +Packet ptan(const Packet& a) { EIGEN_USING_STD(tan); return tan(a); } /** \internal \returns the arc sine of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet pasin(const Packet& a) { using std::asin; return asin(a); } +Packet pasin(const Packet& a) { EIGEN_USING_STD(asin); return asin(a); } /** \internal \returns the arc cosine of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet pacos(const Packet& a) { using std::acos; return acos(a); } +Packet pacos(const Packet& a) { EIGEN_USING_STD(acos); return acos(a); } + +/** \internal \returns the arc tangent of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet patan(const Packet& a) { EIGEN_USING_STD(atan); return atan(a); } + +/** \internal \returns the hyperbolic sine of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet psinh(const Packet& a) { EIGEN_USING_STD(sinh); return sinh(a); } + +/** \internal \returns the hyperbolic cosine of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet pcosh(const Packet& a) { EIGEN_USING_STD(cosh); return cosh(a); } + +/** \internal \returns the hyperbolic tan of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet ptanh(const Packet& a) { EIGEN_USING_STD(tanh); return tanh(a); } /** \internal \returns the exp of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet pexp(const Packet& a) { using std::exp; return exp(a); } +Packet pexp(const Packet& a) { EIGEN_USING_STD(exp); return exp(a); } + +/** \internal \returns the expm1 of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet pexpm1(const Packet& a) { return numext::expm1(a); } /** \internal \returns the log of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet plog(const Packet& a) { using std::log; return log(a); } +Packet plog(const Packet& a) { EIGEN_USING_STD(log); return log(a); } + +/** \internal \returns the log1p of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet plog1p(const Packet& a) { return numext::log1p(a); } + +/** \internal \returns the log10 of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet plog10(const Packet& a) { EIGEN_USING_STD(log10); return log10(a); } + +/** \internal \returns the log10 of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet plog2(const Packet& a) { + typedef typename internal::unpacket_traits::type Scalar; + return pmul(pset1(Scalar(EIGEN_LOG2E)), plog(a)); +} /** \internal \returns the square-root of \a a (coeff-wise) */ template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS -Packet psqrt(const Packet& a) { using std::sqrt; return sqrt(a); } +Packet psqrt(const Packet& a) { return numext::sqrt(a); } + +/** \internal \returns the reciprocal square-root of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet prsqrt(const Packet& a) { + typedef typename internal::unpacket_traits::type Scalar; + return pdiv(pset1(Scalar(1)), psqrt(a)); +} + +/** \internal \returns the rounded value of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet pround(const Packet& a) { using numext::round; return round(a); } + +/** \internal \returns the floor of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet pfloor(const Packet& a) { using numext::floor; return floor(a); } + +/** \internal \returns the rounded value of \a a (coeff-wise) with current + * rounding mode */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet print(const Packet& a) { using numext::rint; return rint(a); } + +/** \internal \returns the ceil of \a a (coeff-wise) */ +template EIGEN_DECLARE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS +Packet pceil(const Packet& a) { using numext::ceil; return ceil(a); } + +/** \internal \returns the first element of a packet */ +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type +pfirst(const Packet& a) +{ return a; } + +/** \internal \returns the sum of the elements of upper and lower half of \a a if \a a is larger than 4. + * For a packet {a0, a1, a2, a3, a4, a5, a6, a7}, it returns a half packet {a0+a4, a1+a5, a2+a6, a3+a7} + * For packet-size smaller or equal to 4, this boils down to a noop. + */ +template +EIGEN_DEVICE_FUNC inline typename conditional<(unpacket_traits::size%8)==0,typename unpacket_traits::half,Packet>::type +predux_half_dowto4(const Packet& a) +{ return a; } + +// Slow generic implementation of Packet reduction. +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type +predux_helper(const Packet& a, Op op) { + typedef typename unpacket_traits::type Scalar; + const size_t n = unpacket_traits::size; + EIGEN_ALIGN_TO_BOUNDARY(sizeof(Packet)) Scalar elements[n]; + pstoreu(elements, a); + for(size_t k = n / 2; k > 0; k /= 2) { + for(size_t i = 0; i < k; ++i) { + elements[i] = op(elements[i], elements[i + k]); + } + } + return elements[0]; +} + +/** \internal \returns the sum of the elements of \a a*/ +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type +predux(const Packet& a) +{ + return a; +} + +/** \internal \returns the product of the elements of \a a */ +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type predux_mul( + const Packet& a) { + typedef typename unpacket_traits::type Scalar; + return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmul))); +} + +/** \internal \returns the min of the elements of \a a */ +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type predux_min( + const Packet &a) { + typedef typename unpacket_traits::type Scalar; + return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmin))); +} + +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type predux_min( + const Packet& a) { + typedef typename unpacket_traits::type Scalar; + return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmin))); +} + +/** \internal \returns the min of the elements of \a a */ +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type predux_max( + const Packet &a) { + typedef typename unpacket_traits::type Scalar; + return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmax))); +} + +template +EIGEN_DEVICE_FUNC inline typename unpacket_traits::type predux_max( + const Packet& a) { + typedef typename unpacket_traits::type Scalar; + return predux_helper(a, EIGEN_BINARY_OP_NAN_PROPAGATION(Scalar, (pmax))); +} + +#undef EIGEN_BINARY_OP_NAN_PROPAGATION + +/** \internal \returns true if all coeffs of \a a means "true" + * It is supposed to be called on values returned by pcmp_*. + */ +// not needed yet +// template EIGEN_DEVICE_FUNC inline bool predux_all(const Packet& a) +// { return bool(a); } + +/** \internal \returns true if any coeffs of \a a means "true" + * It is supposed to be called on values returned by pcmp_*. + */ +template EIGEN_DEVICE_FUNC inline bool predux_any(const Packet& a) +{ + // Dirty but generic implementation where "true" is assumed to be non 0 and all the sames. + // It is expected that "true" is either: + // - Scalar(1) + // - bits full of ones (NaN for floats), + // - or first bit equals to 1 (1 for ints, smallest denormal for floats). + // For all these cases, taking the sum is just fine, and this boils down to a no-op for scalars. + typedef typename unpacket_traits::type Scalar; + return numext::not_equal_strict(predux(a), Scalar(0)); +} /*************************************************************************** * The following functions might not have to be overwritten for vectorized types ***************************************************************************/ -/** \internal copy a packet with constant coeficient \a a (e.g., [a,a,a,a]) to \a *to. \a to must be 16 bytes aligned */ +/** \internal copy a packet with constant coefficient \a a (e.g., [a,a,a,a]) to \a *to. \a to must be 16 bytes aligned */ // NOTE: this function must really be templated on the packet type (think about different packet types for the same scalar type) template inline void pstore1(typename unpacket_traits::type* to, const typename unpacket_traits::type& a) @@ -275,76 +956,89 @@ inline void pstore1(typename unpacket_traits::type* to, const typename u } /** \internal \returns a * b + c (coeff-wise) */ -template inline Packet +template EIGEN_DEVICE_FUNC inline Packet pmadd(const Packet& a, const Packet& b, const Packet& c) { return padd(pmul(a, b),c); } /** \internal \returns a packet version of \a *from. - * If LoadMode equals #Aligned, \a from must be 16 bytes aligned */ -template -inline Packet ploadt(const typename unpacket_traits::type* from) + * The pointer \a from must be aligned on a \a Alignment bytes boundary. */ +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt(const typename unpacket_traits::type* from) { - if(LoadMode == Aligned) + if(Alignment >= unpacket_traits::alignment) return pload(from); else return ploadu(from); } /** \internal copy the packet \a from to \a *to. - * If StoreMode equals #Aligned, \a to must be 16 bytes aligned */ -template -inline void pstoret(Scalar* to, const Packet& from) + * The pointer \a from must be aligned on a \a Alignment bytes boundary. */ +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE void pstoret(Scalar* to, const Packet& from) { - if(LoadMode == Aligned) + if(Alignment >= unpacket_traits::alignment) pstore(to, from); else pstoreu(to, from); } -/** \internal default implementation of palign() allowing partial specialization */ -template -struct palign_impl -{ - // by default data are aligned, so there is nothing to be done :) - static inline void run(PacketType&, const PacketType&) {} -}; - -/** \internal update \a first using the concatenation of the packet_size minus \a Offset last elements - * of \a first and \a Offset first elements of \a second. - * - * This function is currently only used to optimize matrix-vector products on unligned matrices. - * It takes 2 packets that represent a contiguous memory array, and returns a packet starting - * at the position \a Offset. For instance, for packets of 4 elements, we have: - * Input: - * - first = {f0,f1,f2,f3} - * - second = {s0,s1,s2,s3} - * Output: - * - if Offset==0 then {f0,f1,f2,f3} - * - if Offset==1 then {f1,f2,f3,s0} - * - if Offset==2 then {f2,f3,s0,s1} - * - if Offset==3 then {f3,s0,s1,s3} +/** \internal \returns a packet version of \a *from. + * Unlike ploadt, ploadt_ro takes advantage of the read-only memory path on the + * hardware if available to speedup the loading of data that won't be modified + * by the current computation. */ -template -inline void palign(PacketType& first, const PacketType& second) +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE Packet ploadt_ro(const typename unpacket_traits::type* from) { - palign_impl::run(first,second); + return ploadt(from); } /*************************************************************************** * Fast complex products (GCC generates a function call which is very slow) ***************************************************************************/ +// Eigen+CUDA does not support complexes. +#if !defined(EIGEN_GPUCC) + template<> inline std::complex pmul(const std::complex& a, const std::complex& b) -{ return std::complex(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); } +{ return std::complex(a.real()*b.real() - a.imag()*b.imag(), a.imag()*b.real() + a.real()*b.imag()); } template<> inline std::complex pmul(const std::complex& a, const std::complex& b) -{ return std::complex(real(a)*real(b) - imag(a)*imag(b), imag(a)*real(b) + real(a)*imag(b)); } +{ return std::complex(a.real()*b.real() - a.imag()*b.imag(), a.imag()*b.real() + a.real()*b.imag()); } + +#endif + + +/*************************************************************************** + * PacketBlock, that is a collection of N packets where the number of words + * in the packet is a multiple of N. +***************************************************************************/ +template ::size> struct PacketBlock { + Packet packet[N]; +}; + +template EIGEN_DEVICE_FUNC inline void +ptranspose(PacketBlock& /*kernel*/) { + // Nothing to do in the scalar case, i.e. a 1x1 matrix. +} + +/*************************************************************************** + * Selector, i.e. vector of N boolean values used to select (i.e. blend) + * words from 2 packets. +***************************************************************************/ +template struct Selector { + bool select[N]; +}; + +template EIGEN_DEVICE_FUNC inline Packet +pblend(const Selector::size>& ifPacket, const Packet& thenPacket, const Packet& elsePacket) { + return ifPacket.select[0] ? thenPacket : elsePacket; +} } // end namespace internal } // end namespace Eigen #endif // EIGEN_GENERIC_PACKET_MATH_H - diff --git a/thirdparty/eigen/Eigen/src/Core/GlobalFunctions.h b/thirdparty/eigen/Eigen/src/Core/GlobalFunctions.h index 2acf9772..629af94b 100644 --- a/thirdparty/eigen/Eigen/src/Core/GlobalFunctions.h +++ b/thirdparty/eigen/Eigen/src/Core/GlobalFunctions.h @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2010-2012 Gael Guennebaud +// Copyright (C) 2010-2016 Gael Guennebaud // Copyright (C) 2010 Benoit Jacob // // This Source Code Form is subject to the terms of the Mozilla @@ -11,13 +11,30 @@ #ifndef EIGEN_GLOBAL_FUNCTIONS_H #define EIGEN_GLOBAL_FUNCTIONS_H -#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR) \ +#ifdef EIGEN_PARSED_BY_DOXYGEN + +#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \ + /** \returns an expression of the coefficient-wise DOC_OP of \a x + + DOC_DETAILS + + \sa Math functions, class CwiseUnaryOp + */ \ + template \ + inline const Eigen::CwiseUnaryOp, const Derived> \ + NAME(const Eigen::ArrayBase& x); + +#else + +#define EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(NAME,FUNCTOR,DOC_OP,DOC_DETAILS) \ template \ inline const Eigen::CwiseUnaryOp, const Derived> \ - NAME(const Eigen::ArrayBase& x) { \ - return x.derived(); \ + (NAME)(const Eigen::ArrayBase& x) { \ + return Eigen::CwiseUnaryOp, const Derived>(x.derived()); \ } +#endif // EIGEN_PARSED_BY_DOXYGEN + #define EIGEN_ARRAY_DECLARE_GLOBAL_EIGEN_UNARY(NAME,FUNCTOR) \ \ template \ @@ -30,54 +47,139 @@ { \ static inline typename NAME##_retval >::type run(const Eigen::ArrayBase& x) \ { \ - return x.derived(); \ + return typename NAME##_retval >::type(x.derived()); \ } \ }; - namespace Eigen { - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op) - EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op) - - template - inline const Eigen::CwiseUnaryOp, const Derived> - pow(const Eigen::ArrayBase& x, const typename Derived::Scalar& exponent) { - return x.derived().pow(exponent); + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(real,scalar_real_op,real part,\sa ArrayBase::real) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(imag,scalar_imag_op,imaginary part,\sa ArrayBase::imag) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(conj,scalar_conjugate_op,complex conjugate,\sa ArrayBase::conjugate) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(inverse,scalar_inverse_op,inverse,\sa ArrayBase::inverse) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sin,scalar_sin_op,sine,\sa ArrayBase::sin) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cos,scalar_cos_op,cosine,\sa ArrayBase::cos) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tan,scalar_tan_op,tangent,\sa ArrayBase::tan) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atan,scalar_atan_op,arc-tangent,\sa ArrayBase::atan) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asin,scalar_asin_op,arc-sine,\sa ArrayBase::asin) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acos,scalar_acos_op,arc-consine,\sa ArrayBase::acos) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sinh,scalar_sinh_op,hyperbolic sine,\sa ArrayBase::sinh) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cosh,scalar_cosh_op,hyperbolic cosine,\sa ArrayBase::cosh) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(tanh,scalar_tanh_op,hyperbolic tangent,\sa ArrayBase::tanh) +#if EIGEN_HAS_CXX11_MATH + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(asinh,scalar_asinh_op,inverse hyperbolic sine,\sa ArrayBase::asinh) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(acosh,scalar_acosh_op,inverse hyperbolic cosine,\sa ArrayBase::acosh) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(atanh,scalar_atanh_op,inverse hyperbolic tangent,\sa ArrayBase::atanh) +#endif + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(logistic,scalar_logistic_op,logistic function,\sa ArrayBase::logistic) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(lgamma,scalar_lgamma_op,natural logarithm of the gamma function,\sa ArrayBase::lgamma) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(digamma,scalar_digamma_op,derivative of lgamma,\sa ArrayBase::digamma) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erf,scalar_erf_op,error function,\sa ArrayBase::erf) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(erfc,scalar_erfc_op,complement error function,\sa ArrayBase::erfc) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ndtri,scalar_ndtri_op,inverse normal distribution function,\sa ArrayBase::ndtri) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(exp,scalar_exp_op,exponential,\sa ArrayBase::exp) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(expm1,scalar_expm1_op,exponential of a value minus 1,\sa ArrayBase::expm1) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log,scalar_log_op,natural logarithm,\sa Eigen::log10 DOXCOMMA ArrayBase::log) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log1p,scalar_log1p_op,natural logarithm of 1 plus the value,\sa ArrayBase::log1p) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log10,scalar_log10_op,base 10 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log10) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(log2,scalar_log2_op,base 2 logarithm,\sa Eigen::log DOXCOMMA ArrayBase::log2) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs,scalar_abs_op,absolute value,\sa ArrayBase::abs DOXCOMMA MatrixBase::cwiseAbs) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(abs2,scalar_abs2_op,squared absolute value,\sa ArrayBase::abs2 DOXCOMMA MatrixBase::cwiseAbs2) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(arg,scalar_arg_op,complex argument,\sa ArrayBase::arg DOXCOMMA MatrixBase::cwiseArg) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sqrt,scalar_sqrt_op,square root,\sa ArrayBase::sqrt DOXCOMMA MatrixBase::cwiseSqrt) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rsqrt,scalar_rsqrt_op,reciprocal square root,\sa ArrayBase::rsqrt) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(square,scalar_square_op,square (power 2),\sa Eigen::abs2 DOXCOMMA Eigen::pow DOXCOMMA ArrayBase::square) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(cube,scalar_cube_op,cube (power 3),\sa Eigen::pow DOXCOMMA ArrayBase::cube) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(rint,scalar_rint_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(round,scalar_round_op,nearest integer,\sa Eigen::floor DOXCOMMA Eigen::ceil DOXCOMMA ArrayBase::round) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(floor,scalar_floor_op,nearest integer not greater than the giben value,\sa Eigen::ceil DOXCOMMA ArrayBase::floor) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(ceil,scalar_ceil_op,nearest integer not less than the giben value,\sa Eigen::floor DOXCOMMA ArrayBase::ceil) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isnan,scalar_isnan_op,not-a-number test,\sa Eigen::isinf DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isnan) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isinf,scalar_isinf_op,infinite value test,\sa Eigen::isnan DOXCOMMA Eigen::isfinite DOXCOMMA ArrayBase::isinf) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(isfinite,scalar_isfinite_op,finite value test,\sa Eigen::isinf DOXCOMMA Eigen::isnan DOXCOMMA ArrayBase::isfinite) + EIGEN_ARRAY_DECLARE_GLOBAL_UNARY(sign,scalar_sign_op,sign (or 0),\sa ArrayBase::sign) + + /** \returns an expression of the coefficient-wise power of \a x to the given constant \a exponent. + * + * \tparam ScalarExponent is the scalar type of \a exponent. It must be compatible with the scalar type of the given expression (\c Derived::Scalar). + * + * \sa ArrayBase::pow() + * + * \relates ArrayBase + */ +#ifdef EIGEN_PARSED_BY_DOXYGEN + template + inline const CwiseBinaryOp,Derived,Constant > + pow(const Eigen::ArrayBase& x, const ScalarExponent& exponent); +#else + template + EIGEN_DEVICE_FUNC inline + EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE( + const EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,typename internal::promote_scalar_arg::type,pow)) + pow(const Eigen::ArrayBase& x, const ScalarExponent& exponent) + { + typedef typename internal::promote_scalar_arg::type PromotedExponent; + return EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(Derived,PromotedExponent,pow)(x.derived(), + typename internal::plain_constant_type::type(x.derived().rows(), x.derived().cols(), internal::scalar_constant_op(exponent))); } +#endif - template - inline const Eigen::CwiseBinaryOp, const Derived, const Derived> - pow(const Eigen::ArrayBase& x, const Eigen::ArrayBase& exponents) + /** \returns an expression of the coefficient-wise power of \a x to the given array of \a exponents. + * + * This function computes the coefficient-wise power. + * + * Example: \include Cwise_array_power_array.cpp + * Output: \verbinclude Cwise_array_power_array.out + * + * \sa ArrayBase::pow() + * + * \relates ArrayBase + */ + template + inline const Eigen::CwiseBinaryOp, const Derived, const ExponentDerived> + pow(const Eigen::ArrayBase& x, const Eigen::ArrayBase& exponents) { - return Eigen::CwiseBinaryOp, const Derived, const Derived>( + return Eigen::CwiseBinaryOp, const Derived, const ExponentDerived>( x.derived(), exponents.derived() ); } - - /** - * \brief Component-wise division of a scalar by array elements. - **/ - template - inline const Eigen::CwiseUnaryOp, const Derived> - operator/(const typename Derived::Scalar& s, const Eigen::ArrayBase& a) - { - return Eigen::CwiseUnaryOp, const Derived>( - a.derived(), - Eigen::internal::scalar_inverse_mult_op(s) - ); + + /** \returns an expression of the coefficient-wise power of the scalar \a x to the given array of \a exponents. + * + * This function computes the coefficient-wise power between a scalar and an array of exponents. + * + * \tparam Scalar is the scalar type of \a x. It must be compatible with the scalar type of the given array expression (\c Derived::Scalar). + * + * Example: \include Cwise_scalar_power_array.cpp + * Output: \verbinclude Cwise_scalar_power_array.out + * + * \sa ArrayBase::pow() + * + * \relates ArrayBase + */ +#ifdef EIGEN_PARSED_BY_DOXYGEN + template + inline const CwiseBinaryOp,Constant,Derived> + pow(const Scalar& x,const Eigen::ArrayBase& x); +#else + template + EIGEN_DEVICE_FUNC inline + EIGEN_MSVC10_WORKAROUND_BINARYOP_RETURN_TYPE( + const EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(typename internal::promote_scalar_arg::type,Derived,pow)) + pow(const Scalar& x, const Eigen::ArrayBase& exponents) { + typedef typename internal::promote_scalar_arg::type PromotedScalar; + return EIGEN_SCALAR_BINARYOP_EXPR_RETURN_TYPE(PromotedScalar,Derived,pow)( + typename internal::plain_constant_type::type(exponents.derived().rows(), exponents.derived().cols(), internal::scalar_constant_op(x)), exponents.derived()); } +#endif + namespace internal { diff --git a/thirdparty/eigen/Eigen/src/Core/IO.h b/thirdparty/eigen/Eigen/src/Core/IO.h index 8d4bc59e..e81c3152 100644 --- a/thirdparty/eigen/Eigen/src/Core/IO.h +++ b/thirdparty/eigen/Eigen/src/Core/IO.h @@ -41,6 +41,7 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& * - \b rowSuffix string printed at the end of each row * - \b matPrefix string printed at the beginning of the matrix * - \b matSuffix string printed at the end of the matrix + * - \b fill character printed to fill the empty space in aligned columns * * Example: \include IOFormat.cpp * Output: \verbinclude IOFormat.out @@ -49,14 +50,18 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& */ struct IOFormat { - /** Default contructor, see class IOFormat for the meaning of the parameters */ + /** Default constructor, see class IOFormat for the meaning of the parameters */ IOFormat(int _precision = StreamPrecision, int _flags = 0, const std::string& _coeffSeparator = " ", const std::string& _rowSeparator = "\n", const std::string& _rowPrefix="", const std::string& _rowSuffix="", - const std::string& _matPrefix="", const std::string& _matSuffix="") + const std::string& _matPrefix="", const std::string& _matSuffix="", const char _fill=' ') : matPrefix(_matPrefix), matSuffix(_matSuffix), rowPrefix(_rowPrefix), rowSuffix(_rowSuffix), rowSeparator(_rowSeparator), - rowSpacer(""), coeffSeparator(_coeffSeparator), precision(_precision), flags(_flags) + rowSpacer(""), coeffSeparator(_coeffSeparator), fill(_fill), precision(_precision), flags(_flags) { + // TODO check if rowPrefix, rowSuffix or rowSeparator contains a newline + // don't add rowSpacer if columns are not to be aligned + if((flags & DontAlignCols)) + return; int i = int(matSuffix.length())-1; while (i>=0 && matSuffix[i]!='\n') { @@ -67,6 +72,7 @@ struct IOFormat std::string matPrefix, matSuffix; std::string rowPrefix, rowSuffix, rowSeparator, rowSpacer; std::string coeffSeparator; + char fill; int precision; int flags; }; @@ -76,7 +82,7 @@ struct IOFormat * * \brief Pseudo expression providing matrix output with given format * - * \param ExpressionType the type of the object on which IO stream operations are performed + * \tparam ExpressionType the type of the object on which IO stream operations are performed * * This class represents an expression with stream operators controlled by a given IOFormat. * It is the return type of DenseBase::format() @@ -101,57 +107,32 @@ class WithFormat } protected: - const typename ExpressionType::Nested m_matrix; + typename ExpressionType::Nested m_matrix; IOFormat m_format; }; -/** \returns a WithFormat proxy object allowing to print a matrix the with given - * format \a fmt. - * - * See class IOFormat for some examples. - * - * \sa class IOFormat, class WithFormat - */ -template -inline const WithFormat -DenseBase::format(const IOFormat& fmt) const -{ - return WithFormat(derived(), fmt); -} - namespace internal { -template -struct significant_decimals_default_impl -{ - typedef typename NumTraits::Real RealScalar; - static inline int run() - { - using std::ceil; - using std::log; - return cast(ceil(-log(NumTraits::epsilon())/log(RealScalar(10)))); - } -}; - +// NOTE: This helper is kept for backward compatibility with previous code specializing +// this internal::significant_decimals_impl structure. In the future we should directly +// call digits10() which has been introduced in July 2016 in 3.3. template -struct significant_decimals_default_impl +struct significant_decimals_impl { static inline int run() { - return 0; + return NumTraits::digits10(); } }; -template -struct significant_decimals_impl - : significant_decimals_default_impl::IsInteger> -{}; - /** \internal * print the matrix \a _m to the output stream \a s using the output format \a fmt */ template std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& fmt) { + using internal::is_same; + using internal::conditional; + if(_m.size() == 0) { s << fmt.matPrefix << fmt.matSuffix; @@ -160,7 +141,22 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& typename Derived::Nested m = _m; typedef typename Derived::Scalar Scalar; - typedef typename Derived::Index Index; + typedef typename + conditional< + is_same::value || + is_same::value || + is_same::value || + is_same::value, + int, + typename conditional< + is_same >::value || + is_same >::value || + is_same >::value || + is_same >::value, + std::complex, + const Scalar& + >::type + >::type PrintType; Index width = 0; @@ -197,23 +193,31 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& { std::stringstream sstr; sstr.copyfmt(s); - sstr << m.coeff(i,j); + sstr << static_cast(m.coeff(i,j)); width = std::max(width, Index(sstr.str().length())); } } + std::streamsize old_width = s.width(); + char old_fill_character = s.fill(); s << fmt.matPrefix; for(Index i = 0; i < m.rows(); ++i) { if (i) s << fmt.rowSpacer; s << fmt.rowPrefix; - if(width) s.width(width); - s << m.coeff(i, 0); + if(width) { + s.fill(fmt.fill); + s.width(width); + } + s << static_cast(m.coeff(i, 0)); for(Index j = 1; j < m.cols(); ++j) { s << fmt.coeffSeparator; - if (width) s.width(width); - s << m.coeff(i, j); + if(width) { + s.fill(fmt.fill); + s.width(width); + } + s << static_cast(m.coeff(i, j)); } s << fmt.rowSuffix; if( i < m.rows() - 1) @@ -221,6 +225,10 @@ std::ostream & print_matrix(std::ostream & s, const Derived& _m, const IOFormat& } s << fmt.matSuffix; if(explicit_precision) s.precision(old_precision); + if(width) { + s.fill(old_fill_character); + s.width(old_width); + } return s; } diff --git a/thirdparty/eigen/Eigen/src/Core/IndexedView.h b/thirdparty/eigen/Eigen/src/Core/IndexedView.h new file mode 100644 index 00000000..05c2bc9c --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/IndexedView.h @@ -0,0 +1,247 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2017 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_INDEXED_VIEW_H +#define EIGEN_INDEXED_VIEW_H + +namespace Eigen { + +namespace internal { + +template +struct traits > + : traits +{ + enum { + RowsAtCompileTime = int(array_size::value), + ColsAtCompileTime = int(array_size::value), + MaxRowsAtCompileTime = RowsAtCompileTime != Dynamic ? int(RowsAtCompileTime) : Dynamic, + MaxColsAtCompileTime = ColsAtCompileTime != Dynamic ? int(ColsAtCompileTime) : Dynamic, + + XprTypeIsRowMajor = (int(traits::Flags)&RowMajorBit) != 0, + IsRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1 + : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0 + : XprTypeIsRowMajor, + + RowIncr = int(get_compile_time_incr::value), + ColIncr = int(get_compile_time_incr::value), + InnerIncr = IsRowMajor ? ColIncr : RowIncr, + OuterIncr = IsRowMajor ? RowIncr : ColIncr, + + HasSameStorageOrderAsXprType = (IsRowMajor == XprTypeIsRowMajor), + XprInnerStride = HasSameStorageOrderAsXprType ? int(inner_stride_at_compile_time::ret) : int(outer_stride_at_compile_time::ret), + XprOuterstride = HasSameStorageOrderAsXprType ? int(outer_stride_at_compile_time::ret) : int(inner_stride_at_compile_time::ret), + + InnerSize = XprTypeIsRowMajor ? ColsAtCompileTime : RowsAtCompileTime, + IsBlockAlike = InnerIncr==1 && OuterIncr==1, + IsInnerPannel = HasSameStorageOrderAsXprType && is_same,typename conditional::type>::value, + + InnerStrideAtCompileTime = InnerIncr<0 || InnerIncr==DynamicIndex || XprInnerStride==Dynamic ? Dynamic : XprInnerStride * InnerIncr, + OuterStrideAtCompileTime = OuterIncr<0 || OuterIncr==DynamicIndex || XprOuterstride==Dynamic ? Dynamic : XprOuterstride * OuterIncr, + + ReturnAsScalar = is_same::value && is_same::value, + ReturnAsBlock = (!ReturnAsScalar) && IsBlockAlike, + ReturnAsIndexedView = (!ReturnAsScalar) && (!ReturnAsBlock), + + // FIXME we deal with compile-time strides if and only if we have DirectAccessBit flag, + // but this is too strict regarding negative strides... + DirectAccessMask = (int(InnerIncr)!=UndefinedIncr && int(OuterIncr)!=UndefinedIncr && InnerIncr>=0 && OuterIncr>=0) ? DirectAccessBit : 0, + FlagsRowMajorBit = IsRowMajor ? RowMajorBit : 0, + FlagsLvalueBit = is_lvalue::value ? LvalueBit : 0, + FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0, + Flags = (traits::Flags & (HereditaryBits | DirectAccessMask )) | FlagsLvalueBit | FlagsRowMajorBit | FlagsLinearAccessBit + }; + + typedef Block BlockType; +}; + +} + +template +class IndexedViewImpl; + + +/** \class IndexedView + * \ingroup Core_Module + * + * \brief Expression of a non-sequential sub-matrix defined by arbitrary sequences of row and column indices + * + * \tparam XprType the type of the expression in which we are taking the intersections of sub-rows and sub-columns + * \tparam RowIndices the type of the object defining the sequence of row indices + * \tparam ColIndices the type of the object defining the sequence of column indices + * + * This class represents an expression of a sub-matrix (or sub-vector) defined as the intersection + * of sub-sets of rows and columns, that are themself defined by generic sequences of row indices \f$ \{r_0,r_1,..r_{m-1}\} \f$ + * and column indices \f$ \{c_0,c_1,..c_{n-1} \}\f$. Let \f$ A \f$ be the nested matrix, then the resulting matrix \f$ B \f$ has \c m + * rows and \c n columns, and its entries are given by: \f$ B(i,j) = A(r_i,c_j) \f$. + * + * The \c RowIndices and \c ColIndices types must be compatible with the following API: + * \code + * operator[](Index) const; + * Index size() const; + * \endcode + * + * Typical supported types thus include: + * - std::vector + * - std::valarray + * - std::array + * - Plain C arrays: int[N] + * - Eigen::ArrayXi + * - decltype(ArrayXi::LinSpaced(...)) + * - Any view/expressions of the previous types + * - Eigen::ArithmeticSequence + * - Eigen::internal::AllRange (helper for Eigen::all) + * - Eigen::internal::SingleRange (helper for single index) + * - etc. + * + * In typical usages of %Eigen, this class should never be used directly. It is the return type of + * DenseBase::operator()(const RowIndices&, const ColIndices&). + * + * \sa class Block + */ +template +class IndexedView : public IndexedViewImpl::StorageKind> +{ +public: + typedef typename IndexedViewImpl::StorageKind>::Base Base; + EIGEN_GENERIC_PUBLIC_INTERFACE(IndexedView) + EIGEN_INHERIT_ASSIGNMENT_OPERATORS(IndexedView) + + typedef typename internal::ref_selector::non_const_type MatrixTypeNested; + typedef typename internal::remove_all::type NestedExpression; + + template + IndexedView(XprType& xpr, const T0& rowIndices, const T1& colIndices) + : m_xpr(xpr), m_rowIndices(rowIndices), m_colIndices(colIndices) + {} + + /** \returns number of rows */ + Index rows() const { return internal::index_list_size(m_rowIndices); } + + /** \returns number of columns */ + Index cols() const { return internal::index_list_size(m_colIndices); } + + /** \returns the nested expression */ + const typename internal::remove_all::type& + nestedExpression() const { return m_xpr; } + + /** \returns the nested expression */ + typename internal::remove_reference::type& + nestedExpression() { return m_xpr; } + + /** \returns a const reference to the object storing/generating the row indices */ + const RowIndices& rowIndices() const { return m_rowIndices; } + + /** \returns a const reference to the object storing/generating the column indices */ + const ColIndices& colIndices() const { return m_colIndices; } + +protected: + MatrixTypeNested m_xpr; + RowIndices m_rowIndices; + ColIndices m_colIndices; +}; + + +// Generic API dispatcher +template +class IndexedViewImpl + : public internal::generic_xpr_base >::type +{ +public: + typedef typename internal::generic_xpr_base >::type Base; +}; + +namespace internal { + + +template +struct unary_evaluator, IndexBased> + : evaluator_base > +{ + typedef IndexedView XprType; + + enum { + CoeffReadCost = evaluator::CoeffReadCost /* TODO + cost of row/col index */, + + FlagsLinearAccessBit = (traits::RowsAtCompileTime == 1 || traits::ColsAtCompileTime == 1) ? LinearAccessBit : 0, + + FlagsRowMajorBit = traits::FlagsRowMajorBit, + + Flags = (evaluator::Flags & (HereditaryBits & ~RowMajorBit /*| LinearAccessBit | DirectAccessBit*/)) | FlagsLinearAccessBit | FlagsRowMajorBit, + + Alignment = 0 + }; + + EIGEN_DEVICE_FUNC explicit unary_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_xpr(xpr) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeff(Index row, Index col) const + { + eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows() + && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols()); + return m_argImpl.coeff(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index row, Index col) + { + eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows() + && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols()); + return m_argImpl.coeffRef(m_xpr.rowIndices()[row], m_xpr.colIndices()[col]); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Scalar& coeffRef(Index index) + { + EIGEN_STATIC_ASSERT_LVALUE(XprType) + Index row = XprType::RowsAtCompileTime == 1 ? 0 : index; + Index col = XprType::RowsAtCompileTime == 1 ? index : 0; + eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows() + && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols()); + return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const Scalar& coeffRef(Index index) const + { + Index row = XprType::RowsAtCompileTime == 1 ? 0 : index; + Index col = XprType::RowsAtCompileTime == 1 ? index : 0; + eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows() + && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols()); + return m_argImpl.coeffRef( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const CoeffReturnType coeff(Index index) const + { + Index row = XprType::RowsAtCompileTime == 1 ? 0 : index; + Index col = XprType::RowsAtCompileTime == 1 ? index : 0; + eigen_assert(m_xpr.rowIndices()[row] >= 0 && m_xpr.rowIndices()[row] < m_xpr.nestedExpression().rows() + && m_xpr.colIndices()[col] >= 0 && m_xpr.colIndices()[col] < m_xpr.nestedExpression().cols()); + return m_argImpl.coeff( m_xpr.rowIndices()[row], m_xpr.colIndices()[col]); + } + +protected: + + evaluator m_argImpl; + const XprType& m_xpr; + +}; + +} // end namespace internal + +} // end namespace Eigen + +#endif // EIGEN_INDEXED_VIEW_H diff --git a/thirdparty/eigen/Eigen/src/Core/Inverse.h b/thirdparty/eigen/Eigen/src/Core/Inverse.h new file mode 100644 index 00000000..c514438c --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/Inverse.h @@ -0,0 +1,117 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2014-2019 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_INVERSE_H +#define EIGEN_INVERSE_H + +namespace Eigen { + +template class InverseImpl; + +namespace internal { + +template +struct traits > + : traits +{ + typedef typename XprType::PlainObject PlainObject; + typedef traits BaseTraits; + enum { + Flags = BaseTraits::Flags & RowMajorBit + }; +}; + +} // end namespace internal + +/** \class Inverse + * + * \brief Expression of the inverse of another expression + * + * \tparam XprType the type of the expression we are taking the inverse + * + * This class represents an abstract expression of A.inverse() + * and most of the time this is the only way it is used. + * + */ +template +class Inverse : public InverseImpl::StorageKind> +{ +public: + typedef typename XprType::StorageIndex StorageIndex; + typedef typename XprType::Scalar Scalar; + typedef typename internal::ref_selector::type XprTypeNested; + typedef typename internal::remove_all::type XprTypeNestedCleaned; + typedef typename internal::ref_selector::type Nested; + typedef typename internal::remove_all::type NestedExpression; + + explicit EIGEN_DEVICE_FUNC Inverse(const XprType &xpr) + : m_xpr(xpr) + {} + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index rows() const EIGEN_NOEXCEPT { return m_xpr.cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index cols() const EIGEN_NOEXCEPT { return m_xpr.rows(); } + + EIGEN_DEVICE_FUNC const XprTypeNestedCleaned& nestedExpression() const { return m_xpr; } + +protected: + XprTypeNested m_xpr; +}; + +// Generic API dispatcher +template +class InverseImpl + : public internal::generic_xpr_base >::type +{ +public: + typedef typename internal::generic_xpr_base >::type Base; + typedef typename XprType::Scalar Scalar; +private: + + Scalar coeff(Index row, Index col) const; + Scalar coeff(Index i) const; +}; + +namespace internal { + +/** \internal + * \brief Default evaluator for Inverse expression. + * + * This default evaluator for Inverse expression simply evaluate the inverse into a temporary + * by a call to internal::call_assignment_no_alias. + * Therefore, inverse implementers only have to specialize Assignment, ...> for + * there own nested expression. + * + * \sa class Inverse + */ +template +struct unary_evaluator > + : public evaluator::PlainObject> +{ + typedef Inverse InverseType; + typedef typename InverseType::PlainObject PlainObject; + typedef evaluator Base; + + enum { Flags = Base::Flags | EvalBeforeNestingBit }; + + unary_evaluator(const InverseType& inv_xpr) + : m_result(inv_xpr.rows(), inv_xpr.cols()) + { + ::new (static_cast(this)) Base(m_result); + internal::call_assignment_no_alias(m_result, inv_xpr); + } + +protected: + PlainObject m_result; +}; + +} // end namespace internal + +} // end namespace Eigen + +#endif // EIGEN_INVERSE_H diff --git a/thirdparty/eigen/Eigen/src/Core/Map.h b/thirdparty/eigen/Eigen/src/Core/Map.h index f804c89d..218cc157 100644 --- a/thirdparty/eigen/Eigen/src/Core/Map.h +++ b/thirdparty/eigen/Eigen/src/Core/Map.h @@ -11,7 +11,35 @@ #ifndef EIGEN_MAP_H #define EIGEN_MAP_H -namespace Eigen { +namespace Eigen { + +namespace internal { +template +struct traits > + : public traits +{ + typedef traits TraitsBase; + enum { + PlainObjectTypeInnerSize = ((traits::Flags&RowMajorBit)==RowMajorBit) + ? PlainObjectType::ColsAtCompileTime + : PlainObjectType::RowsAtCompileTime, + + InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0 + ? int(PlainObjectType::InnerStrideAtCompileTime) + : int(StrideType::InnerStrideAtCompileTime), + OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0 + ? (InnerStrideAtCompileTime==Dynamic || PlainObjectTypeInnerSize==Dynamic + ? Dynamic + : int(InnerStrideAtCompileTime) * int(PlainObjectTypeInnerSize)) + : int(StrideType::OuterStrideAtCompileTime), + Alignment = int(MapOptions)&int(AlignedMask), + Flags0 = TraitsBase::Flags & (~NestByRefBit), + Flags = is_lvalue::value ? int(Flags0) : (int(Flags0) & ~LvalueBit) + }; +private: + enum { Options }; // Expressions don't have Options +}; +} /** \class Map * \ingroup Core_Module @@ -19,7 +47,7 @@ namespace Eigen { * \brief A matrix or vector expression mapping an existing array of data. * * \tparam PlainObjectType the equivalent matrix type of the mapped data - * \tparam MapOptions specifies whether the pointer is \c #Aligned, or \c #Unaligned. + * \tparam MapOptions specifies the pointer alignment in bytes. It can be: \c #Aligned128, \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned. * The default is \c #Unaligned. * \tparam StrideType optionally specifies strides. By default, Map assumes the memory layout * of an ordinary, contiguous array. This can be overridden by specifying strides. @@ -63,44 +91,6 @@ namespace Eigen { * * \sa PlainObjectBase::Map(), \ref TopicStorageOrders */ - -namespace internal { -template -struct traits > - : public traits -{ - typedef traits TraitsBase; - typedef typename PlainObjectType::Index Index; - typedef typename PlainObjectType::Scalar Scalar; - enum { - InnerStrideAtCompileTime = StrideType::InnerStrideAtCompileTime == 0 - ? int(PlainObjectType::InnerStrideAtCompileTime) - : int(StrideType::InnerStrideAtCompileTime), - OuterStrideAtCompileTime = StrideType::OuterStrideAtCompileTime == 0 - ? int(PlainObjectType::OuterStrideAtCompileTime) - : int(StrideType::OuterStrideAtCompileTime), - HasNoInnerStride = InnerStrideAtCompileTime == 1, - HasNoOuterStride = StrideType::OuterStrideAtCompileTime == 0, - HasNoStride = HasNoInnerStride && HasNoOuterStride, - IsAligned = bool(EIGEN_ALIGN) && ((int(MapOptions)&Aligned)==Aligned), - IsDynamicSize = PlainObjectType::SizeAtCompileTime==Dynamic, - KeepsPacketAccess = bool(HasNoInnerStride) - && ( bool(IsDynamicSize) - || HasNoOuterStride - || ( OuterStrideAtCompileTime!=Dynamic - && ((static_cast(sizeof(Scalar))*OuterStrideAtCompileTime)%16)==0 ) ), - Flags0 = TraitsBase::Flags & (~NestByRefBit), - Flags1 = IsAligned ? (int(Flags0) | AlignedBit) : (int(Flags0) & ~AlignedBit), - Flags2 = (bool(HasNoStride) || bool(PlainObjectType::IsVectorAtCompileTime)) - ? int(Flags1) : int(Flags1 & ~LinearAccessBit), - Flags3 = is_lvalue::value ? int(Flags2) : (int(Flags2) & ~LvalueBit), - Flags = KeepsPacketAccess ? int(Flags3) : (int(Flags3) & ~PacketAccessBit) - }; -private: - enum { Options }; // Expressions don't have Options -}; -} - template class Map : public MapBase > { @@ -110,34 +100,34 @@ template class Ma EIGEN_DENSE_PUBLIC_INTERFACE(Map) typedef typename Base::PointerType PointerType; -#if EIGEN2_SUPPORT_STAGE <= STAGE30_FULL_EIGEN3_API - typedef const Scalar* PointerArgType; - inline PointerType cast_to_pointer_type(PointerArgType ptr) { return const_cast(ptr); } -#else typedef PointerType PointerArgType; + EIGEN_DEVICE_FUNC inline PointerType cast_to_pointer_type(PointerArgType ptr) { return ptr; } -#endif + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const { return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1; } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const { return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer() - : IsVectorAtCompileTime ? this->size() - : int(Flags)&RowMajorBit ? this->cols() - : this->rows(); + : internal::traits::OuterStrideAtCompileTime != Dynamic ? Index(internal::traits::OuterStrideAtCompileTime) + : IsVectorAtCompileTime ? (this->size() * innerStride()) + : int(Flags)&RowMajorBit ? (this->cols() * innerStride()) + : (this->rows() * innerStride()); } /** Constructor in the fixed-size case. * * \param dataPtr pointer to the array to map - * \param a_stride optional Stride object, passing the strides. + * \param stride optional Stride object, passing the strides. */ - inline Map(PointerArgType dataPtr, const StrideType& a_stride = StrideType()) - : Base(cast_to_pointer_type(dataPtr)), m_stride(a_stride) + EIGEN_DEVICE_FUNC + explicit inline Map(PointerArgType dataPtr, const StrideType& stride = StrideType()) + : Base(cast_to_pointer_type(dataPtr)), m_stride(stride) { PlainObjectType::Base::_check_template_params(); } @@ -145,11 +135,12 @@ template class Ma /** Constructor in the dynamic-size vector case. * * \param dataPtr pointer to the array to map - * \param a_size the size of the vector expression - * \param a_stride optional Stride object, passing the strides. + * \param size the size of the vector expression + * \param stride optional Stride object, passing the strides. */ - inline Map(PointerArgType dataPtr, Index a_size, const StrideType& a_stride = StrideType()) - : Base(cast_to_pointer_type(dataPtr), a_size), m_stride(a_stride) + EIGEN_DEVICE_FUNC + inline Map(PointerArgType dataPtr, Index size, const StrideType& stride = StrideType()) + : Base(cast_to_pointer_type(dataPtr), size), m_stride(stride) { PlainObjectType::Base::_check_template_params(); } @@ -157,12 +148,13 @@ template class Ma /** Constructor in the dynamic-size matrix case. * * \param dataPtr pointer to the array to map - * \param nbRows the number of rows of the matrix expression - * \param nbCols the number of columns of the matrix expression - * \param a_stride optional Stride object, passing the strides. + * \param rows the number of rows of the matrix expression + * \param cols the number of columns of the matrix expression + * \param stride optional Stride object, passing the strides. */ - inline Map(PointerArgType dataPtr, Index nbRows, Index nbCols, const StrideType& a_stride = StrideType()) - : Base(cast_to_pointer_type(dataPtr), nbRows, nbCols), m_stride(a_stride) + EIGEN_DEVICE_FUNC + inline Map(PointerArgType dataPtr, Index rows, Index cols, const StrideType& stride = StrideType()) + : Base(cast_to_pointer_type(dataPtr), rows, cols), m_stride(stride) { PlainObjectType::Base::_check_template_params(); } @@ -173,19 +165,6 @@ template class Ma StrideType m_stride; }; -template -inline Array<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> - ::Array(const Scalar *data) -{ - this->_set_noalias(Eigen::Map(data)); -} - -template -inline Matrix<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols> - ::Matrix(const Scalar *data) -{ - this->_set_noalias(Eigen::Map(data)); -} } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/MapBase.h b/thirdparty/eigen/Eigen/src/Core/MapBase.h index 81efc4a6..d856447f 100644 --- a/thirdparty/eigen/Eigen/src/Core/MapBase.h +++ b/thirdparty/eigen/Eigen/src/Core/MapBase.h @@ -12,15 +12,25 @@ #define EIGEN_MAPBASE_H #define EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) \ - EIGEN_STATIC_ASSERT((int(internal::traits::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \ + EIGEN_STATIC_ASSERT((int(internal::evaluator::Flags) & LinearAccessBit) || Derived::IsVectorAtCompileTime, \ YOU_ARE_TRYING_TO_USE_AN_INDEX_BASED_ACCESSOR_ON_AN_EXPRESSION_THAT_DOES_NOT_SUPPORT_THAT) -namespace Eigen { +namespace Eigen { -/** \class MapBase - * \ingroup Core_Module +/** \ingroup Core_Module * - * \brief Base class for Map and Block expression with direct access + * \brief Base class for dense Map and Block expression with direct access + * + * This base class provides the const low-level accessors (e.g. coeff, coeffRef) of dense + * Map and Block objects with direct access. + * Typical users do not have to directly deal with this class. + * + * This class can be extended by through the macro plugin \c EIGEN_MAPBASE_PLUGIN. + * See \link TopicCustomizing_Plugins customizing Eigen \endlink for details. + * + * The \c Derived class has to provide the following two methods describing the memory layout: + * \code Index innerStride() const; \endcode + * \code Index outerStride() const; \endcode * * \sa class Map, class Block */ @@ -33,11 +43,11 @@ template class MapBase enum { RowsAtCompileTime = internal::traits::RowsAtCompileTime, ColsAtCompileTime = internal::traits::ColsAtCompileTime, + InnerStrideAtCompileTime = internal::traits::InnerStrideAtCompileTime, SizeAtCompileTime = Base::SizeAtCompileTime }; typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; typedef typename internal::packet_traits::type PacketScalar; typedef typename NumTraits::Real RealScalar; @@ -76,8 +86,12 @@ template class MapBase typedef typename Base::CoeffReturnType CoeffReturnType; - inline Index rows() const { return m_rows.value(); } - inline Index cols() const { return m_cols.value(); } + /** \copydoc DenseBase::rows() */ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rows() const EIGEN_NOEXCEPT { return m_rows.value(); } + /** \copydoc DenseBase::cols() */ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return m_cols.value(); } /** Returns a pointer to the first coefficient of the matrix or vector. * @@ -85,30 +99,39 @@ template class MapBase * * \sa innerStride(), outerStride() */ - inline const Scalar* data() const { return m_data; } + EIGEN_DEVICE_FUNC inline const Scalar* data() const { return m_data; } + /** \copydoc PlainObjectBase::coeff(Index,Index) const */ + EIGEN_DEVICE_FUNC inline const Scalar& coeff(Index rowId, Index colId) const { return m_data[colId * colStride() + rowId * rowStride()]; } + /** \copydoc PlainObjectBase::coeff(Index) const */ + EIGEN_DEVICE_FUNC inline const Scalar& coeff(Index index) const { EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) return m_data[index * innerStride()]; } + /** \copydoc PlainObjectBase::coeffRef(Index,Index) const */ + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index rowId, Index colId) const { return this->m_data[colId * colStride() + rowId * rowStride()]; } + /** \copydoc PlainObjectBase::coeffRef(Index) const */ + EIGEN_DEVICE_FUNC inline const Scalar& coeffRef(Index index) const { EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) return this->m_data[index * innerStride()]; } + /** \internal */ template inline PacketScalar packet(Index rowId, Index colId) const { @@ -116,6 +139,7 @@ template class MapBase (m_data + (colId * colStride() + rowId * rowStride())); } + /** \internal */ template inline PacketScalar packet(Index index) const { @@ -123,12 +147,16 @@ template class MapBase return internal::ploadt(m_data + index * innerStride()); } + /** \internal Constructor for fixed size matrices or vectors */ + EIGEN_DEVICE_FUNC explicit inline MapBase(PointerType dataPtr) : m_data(dataPtr), m_rows(RowsAtCompileTime), m_cols(ColsAtCompileTime) { EIGEN_STATIC_ASSERT_FIXED_SIZE(Derived) - checkSanity(); + checkSanity(); } + /** \internal Constructor for dynamically sized vectors */ + EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index vecSize) : m_data(dataPtr), m_rows(RowsAtCompileTime == Dynamic ? vecSize : Index(RowsAtCompileTime)), @@ -137,16 +165,18 @@ template class MapBase EIGEN_STATIC_ASSERT_VECTOR_ONLY(Derived) eigen_assert(vecSize >= 0); eigen_assert(dataPtr == 0 || SizeAtCompileTime == Dynamic || SizeAtCompileTime == vecSize); - checkSanity(); + checkSanity(); } - inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols) - : m_data(dataPtr), m_rows(nbRows), m_cols(nbCols) + /** \internal Constructor for dynamically sized matrices */ + EIGEN_DEVICE_FUNC + inline MapBase(PointerType dataPtr, Index rows, Index cols) + : m_data(dataPtr), m_rows(rows), m_cols(cols) { eigen_assert( (dataPtr == 0) - || ( nbRows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == nbRows) - && nbCols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == nbCols))); - checkSanity(); + || ( rows >= 0 && (RowsAtCompileTime == Dynamic || RowsAtCompileTime == rows) + && cols >= 0 && (ColsAtCompileTime == Dynamic || ColsAtCompileTime == cols))); + checkSanity(); } #ifdef EIGEN_MAPBASE_PLUGIN @@ -154,21 +184,42 @@ template class MapBase #endif protected: + EIGEN_DEFAULT_COPY_CONSTRUCTOR(MapBase) + EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MapBase) - void checkSanity() const + template + EIGEN_DEVICE_FUNC + void checkSanity(typename internal::enable_if<(internal::traits::Alignment>0),void*>::type = 0) const { - EIGEN_STATIC_ASSERT(EIGEN_IMPLIES(internal::traits::Flags&PacketAccessBit, - internal::inner_stride_at_compile_time::ret==1), - PACKET_ACCESS_REQUIRES_TO_HAVE_INNER_STRIDE_FIXED_TO_1); - eigen_assert(EIGEN_IMPLIES(internal::traits::Flags&AlignedBit, (size_t(m_data) % 16) == 0) - && "input pointer is not aligned on a 16 byte boundary"); +#if EIGEN_MAX_ALIGN_BYTES>0 + // innerStride() is not set yet when this function is called, so we optimistically assume the lowest plausible value: + const Index minInnerStride = InnerStrideAtCompileTime == Dynamic ? 1 : Index(InnerStrideAtCompileTime); + EIGEN_ONLY_USED_FOR_DEBUG(minInnerStride); + eigen_assert(( ((internal::UIntPtr(m_data) % internal::traits::Alignment) == 0) + || (cols() * rows() * minInnerStride * sizeof(Scalar)) < internal::traits::Alignment ) && "data is not aligned"); +#endif } + template + EIGEN_DEVICE_FUNC + void checkSanity(typename internal::enable_if::Alignment==0,void*>::type = 0) const + {} + PointerType m_data; const internal::variable_if_dynamic m_rows; const internal::variable_if_dynamic m_cols; }; +/** \ingroup Core_Module + * + * \brief Base class for non-const dense Map and Block expression with direct access + * + * This base class provides the non-const low-level accessors (e.g. coeff and coeffRef) of + * dense Map and Block objects with direct access. + * It inherits MapBase which defines the const variant for reading specific entries. + * + * \sa class Map, class Block + */ template class MapBase : public MapBase { @@ -179,7 +230,7 @@ template class MapBase typedef typename Base::Scalar Scalar; typedef typename Base::PacketScalar PacketScalar; - typedef typename Base::Index Index; + typedef typename Base::StorageIndex StorageIndex; typedef typename Base::PointerType PointerType; using Base::derived; @@ -200,14 +251,18 @@ template class MapBase const Scalar >::type ScalarWithConstIfNotLvalue; + EIGEN_DEVICE_FUNC inline const Scalar* data() const { return this->m_data; } + EIGEN_DEVICE_FUNC inline ScalarWithConstIfNotLvalue* data() { return this->m_data; } // no const-cast here so non-const-correct code will give a compile error + EIGEN_DEVICE_FUNC inline ScalarWithConstIfNotLvalue& coeffRef(Index row, Index col) { return this->m_data[col * colStride() + row * rowStride()]; } + EIGEN_DEVICE_FUNC inline ScalarWithConstIfNotLvalue& coeffRef(Index index) { EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS(Derived) @@ -229,10 +284,11 @@ template class MapBase (this->m_data + index * innerStride(), val); } - explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {} - inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {} - inline MapBase(PointerType dataPtr, Index nbRows, Index nbCols) : Base(dataPtr, nbRows, nbCols) {} + EIGEN_DEVICE_FUNC explicit inline MapBase(PointerType dataPtr) : Base(dataPtr) {} + EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index vecSize) : Base(dataPtr, vecSize) {} + EIGEN_DEVICE_FUNC inline MapBase(PointerType dataPtr, Index rows, Index cols) : Base(dataPtr, rows, cols) {} + EIGEN_DEVICE_FUNC Derived& operator=(const MapBase& other) { ReadOnlyMapBase::Base::operator=(other); @@ -242,6 +298,9 @@ template class MapBase // In theory we could simply refer to Base:Base::operator=, but MSVC does not like Base::Base, // see bugs 821 and 920. using ReadOnlyMapBase::Base::operator=; + protected: + EIGEN_DEFAULT_COPY_CONSTRUCTOR(MapBase) + EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MapBase) }; #undef EIGEN_STATIC_ASSERT_INDEX_BASED_ACCESS diff --git a/thirdparty/eigen/Eigen/src/Core/MathFunctions.h b/thirdparty/eigen/Eigen/src/Core/MathFunctions.h index f707aa41..764c41c9 100644 --- a/thirdparty/eigen/Eigen/src/Core/MathFunctions.h +++ b/thirdparty/eigen/Eigen/src/Core/MathFunctions.h @@ -2,6 +2,7 @@ // for linear algebra. // // Copyright (C) 2006-2010 Benoit Jacob +// Copyright (c) 2021, NVIDIA CORPORATION. All rights reserved. // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -10,11 +11,26 @@ #ifndef EIGEN_MATHFUNCTIONS_H #define EIGEN_MATHFUNCTIONS_H +// TODO this should better be moved to NumTraits +// Source: WolframAlpha +#define EIGEN_PI 3.141592653589793238462643383279502884197169399375105820974944592307816406L +#define EIGEN_LOG2E 1.442695040888963407359924681001892137426645954152985934135449406931109219L +#define EIGEN_LN2 0.693147180559945309417232121458176568075500134360255254120680009493393621L + namespace Eigen { +// On WINCE, std::abs is defined for int only, so let's defined our own overloads: +// This issue has been confirmed with MSVC 2008 only, but the issue might exist for more recent versions too. +#if EIGEN_OS_WINCE && EIGEN_COMP_MSVC && EIGEN_COMP_MSVC<=1500 +long abs(long x) { return (labs(x)); } +double abs(double x) { return (fabs(x)); } +float abs(float x) { return (fabsf(x)); } +long double abs(long double x) { return (fabsl(x)); } +#endif + namespace internal { -/** \internal \struct global_math_functions_filtering_base +/** \internal \class global_math_functions_filtering_base * * What it does: * Defines a typedef 'type' as follows: @@ -62,6 +78,7 @@ template::IsComplex> struct real_default_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar& x) { return x; @@ -72,6 +89,7 @@ template struct real_default_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar& x) { using std::real; @@ -81,13 +99,25 @@ struct real_default_impl template struct real_impl : real_default_impl {}; +#if defined(EIGEN_GPU_COMPILE_PHASE) +template +struct real_impl > +{ + typedef T RealScalar; + EIGEN_DEVICE_FUNC + static inline T run(const std::complex& x) + { + return x.real(); + } +}; +#endif + template struct real_retval { typedef typename NumTraits::Real type; }; - /**************************************************************************** * Implementation of imag * ****************************************************************************/ @@ -96,6 +126,7 @@ template::IsComplex> struct imag_default_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar&) { return RealScalar(0); @@ -106,6 +137,7 @@ template struct imag_default_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar& x) { using std::imag; @@ -115,6 +147,19 @@ struct imag_default_impl template struct imag_impl : imag_default_impl {}; +#if defined(EIGEN_GPU_COMPILE_PHASE) +template +struct imag_impl > +{ + typedef T RealScalar; + EIGEN_DEVICE_FUNC + static inline T run(const std::complex& x) + { + return x.imag(); + } +}; +#endif + template struct imag_retval { @@ -129,10 +174,12 @@ template struct real_ref_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar& run(Scalar& x) { return reinterpret_cast(&x)[0]; } + EIGEN_DEVICE_FUNC static inline const RealScalar& run(const Scalar& x) { return reinterpret_cast(&x)[0]; @@ -153,10 +200,12 @@ template struct imag_ref_default_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar& run(Scalar& x) { return reinterpret_cast(&x)[1]; } + EIGEN_DEVICE_FUNC static inline const RealScalar& run(const Scalar& x) { return reinterpret_cast(&x)[1]; @@ -166,10 +215,12 @@ struct imag_ref_default_impl template struct imag_ref_default_impl { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline Scalar run(Scalar&) { return Scalar(0); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline const Scalar run(const Scalar&) { return Scalar(0); @@ -190,8 +241,9 @@ struct imag_ref_retval ****************************************************************************/ template::IsComplex> -struct conj_impl +struct conj_default_impl { + EIGEN_DEVICE_FUNC static inline Scalar run(const Scalar& x) { return x; @@ -199,8 +251,9 @@ struct conj_impl }; template -struct conj_impl +struct conj_default_impl { + EIGEN_DEVICE_FUNC static inline Scalar run(const Scalar& x) { using std::conj; @@ -208,6 +261,9 @@ struct conj_impl } }; +template::IsComplex> +struct conj_impl : conj_default_impl {}; + template struct conj_retval { @@ -222,6 +278,7 @@ template struct abs2_impl_default { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar& x) { return x*x; @@ -232,9 +289,10 @@ template struct abs2_impl_default // IsComplex { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar& x) { - return real(x)*real(x) + imag(x)*imag(x); + return x.real()*x.real() + x.imag()*x.imag(); } }; @@ -242,6 +300,7 @@ template struct abs2_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar& x) { return abs2_impl_default::IsComplex>::run(x); @@ -254,27 +313,91 @@ struct abs2_retval typedef typename NumTraits::Real type; }; +/**************************************************************************** +* Implementation of sqrt/rsqrt * +****************************************************************************/ + +template +struct sqrt_impl +{ + EIGEN_DEVICE_FUNC + static EIGEN_ALWAYS_INLINE Scalar run(const Scalar& x) + { + EIGEN_USING_STD(sqrt); + return sqrt(x); + } +}; + +// Complex sqrt defined in MathFunctionsImpl.h. +template EIGEN_DEVICE_FUNC std::complex complex_sqrt(const std::complex& a_x); + +// Custom implementation is faster than `std::sqrt`, works on +// GPU, and correctly handles special cases (unlike MSVC). +template +struct sqrt_impl > +{ + EIGEN_DEVICE_FUNC + static EIGEN_ALWAYS_INLINE std::complex run(const std::complex& x) + { + return complex_sqrt(x); + } +}; + +template +struct sqrt_retval +{ + typedef Scalar type; +}; + +// Default implementation relies on numext::sqrt, at bottom of file. +template +struct rsqrt_impl; + +// Complex rsqrt defined in MathFunctionsImpl.h. +template EIGEN_DEVICE_FUNC std::complex complex_rsqrt(const std::complex& a_x); + +template +struct rsqrt_impl > +{ + EIGEN_DEVICE_FUNC + static EIGEN_ALWAYS_INLINE std::complex run(const std::complex& x) + { + return complex_rsqrt(x); + } +}; + +template +struct rsqrt_retval +{ + typedef Scalar type; +}; + /**************************************************************************** * Implementation of norm1 * ****************************************************************************/ template -struct norm1_default_impl +struct norm1_default_impl; + +template +struct norm1_default_impl { typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC static inline RealScalar run(const Scalar& x) { - using std::abs; - return abs(real(x)) + abs(imag(x)); + EIGEN_USING_STD(abs); + return abs(x.real()) + abs(x.imag()); } }; template struct norm1_default_impl { + EIGEN_DEVICE_FUNC static inline Scalar run(const Scalar& x) { - using std::abs; + EIGEN_USING_STD(abs); return abs(x); } }; @@ -292,25 +415,7 @@ struct norm1_retval * Implementation of hypot * ****************************************************************************/ -template -struct hypot_impl -{ - typedef typename NumTraits::Real RealScalar; - static inline RealScalar run(const Scalar& x, const Scalar& y) - { - using std::max; - using std::min; - using std::abs; - using std::sqrt; - RealScalar _x = abs(x); - RealScalar _y = abs(y); - RealScalar p = (max)(_x, _y); - if(p==RealScalar(0)) return RealScalar(0); - RealScalar q = (min)(_x, _y); - RealScalar qp = q/p; - return p * sqrt(RealScalar(1) + qp*qp); - } -}; +template struct hypot_impl; template struct hypot_retval @@ -322,60 +427,335 @@ struct hypot_retval * Implementation of cast * ****************************************************************************/ -template +template struct cast_impl { + EIGEN_DEVICE_FUNC static inline NewType run(const OldType& x) { return static_cast(x); } }; +// Casting from S -> Complex leads to an implicit conversion from S to T, +// generating warnings on clang. Here we explicitly cast the real component. +template +struct cast_impl::IsComplex && NumTraits::IsComplex + >::type> +{ + EIGEN_DEVICE_FUNC + static inline NewType run(const OldType& x) + { + typedef typename NumTraits::Real NewReal; + return static_cast(static_cast(x)); + } +}; + // here, for once, we're plainly returning NewType: we don't want cast to do weird things. template +EIGEN_DEVICE_FUNC inline NewType cast(const OldType& x) { return cast_impl::run(x); } /**************************************************************************** -* Implementation of atanh2 * +* Implementation of round * +****************************************************************************/ + +template +struct round_impl +{ + EIGEN_DEVICE_FUNC + static inline Scalar run(const Scalar& x) + { + EIGEN_STATIC_ASSERT((!NumTraits::IsComplex), NUMERIC_TYPE_MUST_BE_REAL) +#if EIGEN_HAS_CXX11_MATH + EIGEN_USING_STD(round); +#endif + return Scalar(round(x)); + } +}; + +#if !EIGEN_HAS_CXX11_MATH +#if EIGEN_HAS_C99_MATH +// Use ::roundf for float. +template<> +struct round_impl { + EIGEN_DEVICE_FUNC + static inline float run(const float& x) + { + return ::roundf(x); + } +}; +#else +template +struct round_using_floor_ceil_impl +{ + EIGEN_DEVICE_FUNC + static inline Scalar run(const Scalar& x) + { + EIGEN_STATIC_ASSERT((!NumTraits::IsComplex), NUMERIC_TYPE_MUST_BE_REAL) + // Without C99 round/roundf, resort to floor/ceil. + EIGEN_USING_STD(floor); + EIGEN_USING_STD(ceil); + // If not enough precision to resolve a decimal at all, return the input. + // Otherwise, adding 0.5 can trigger an increment by 1. + const Scalar limit = Scalar(1ull << (NumTraits::digits() - 1)); + if (x >= limit || x <= -limit) { + return x; + } + return (x > Scalar(0)) ? Scalar(floor(x + Scalar(0.5))) : Scalar(ceil(x - Scalar(0.5))); + } +}; + +template<> +struct round_impl : round_using_floor_ceil_impl {}; + +template<> +struct round_impl : round_using_floor_ceil_impl {}; +#endif // EIGEN_HAS_C99_MATH +#endif // !EIGEN_HAS_CXX11_MATH + +template +struct round_retval +{ + typedef Scalar type; +}; + +/**************************************************************************** +* Implementation of rint * ****************************************************************************/ -template -struct atanh2_default_impl +template +struct rint_impl { + EIGEN_DEVICE_FUNC + static inline Scalar run(const Scalar& x) + { + EIGEN_STATIC_ASSERT((!NumTraits::IsComplex), NUMERIC_TYPE_MUST_BE_REAL) +#if EIGEN_HAS_CXX11_MATH + EIGEN_USING_STD(rint); +#endif + return rint(x); + } +}; + +#if !EIGEN_HAS_CXX11_MATH +template<> +struct rint_impl { + EIGEN_DEVICE_FUNC + static inline double run(const double& x) + { + return ::rint(x); + } +}; +template<> +struct rint_impl { + EIGEN_DEVICE_FUNC + static inline float run(const float& x) + { + return ::rintf(x); + } +}; +#endif + +template +struct rint_retval { - typedef Scalar retval; + typedef Scalar type; +}; + +/**************************************************************************** +* Implementation of arg * +****************************************************************************/ + +// Visual Studio 2017 has a bug where arg(float) returns 0 for negative inputs. +// This seems to be fixed in VS 2019. +#if EIGEN_HAS_CXX11_MATH && (!EIGEN_COMP_MSVC || EIGEN_COMP_MSVC >= 1920) +// std::arg is only defined for types of std::complex, or integer types or float/double/long double +template::IsComplex || is_integral::value + || is_same::value || is_same::value + || is_same::value > +struct arg_default_impl; + +template +struct arg_default_impl { typedef typename NumTraits::Real RealScalar; - static inline Scalar run(const Scalar& x, const Scalar& y) + EIGEN_DEVICE_FUNC + static inline RealScalar run(const Scalar& x) + { + // There is no official ::arg on device in CUDA/HIP, so we always need to use std::arg. + using std::arg; + return static_cast(arg(x)); + } +}; + +// Must be non-complex floating-point type (e.g. half/bfloat16). +template +struct arg_default_impl { + typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC + static inline RealScalar run(const Scalar& x) + { + return (x < Scalar(0)) ? RealScalar(EIGEN_PI) : RealScalar(0); + } +}; +#else +template::IsComplex> +struct arg_default_impl +{ + typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC + static inline RealScalar run(const Scalar& x) { - using std::abs; - using std::log; - using std::sqrt; - Scalar z = x / y; - if (y == Scalar(0) || abs(z) > sqrt(NumTraits::epsilon())) - return RealScalar(0.5) * log((y + x) / (y - x)); - else - return z + z*z*z / RealScalar(3); + return (x < RealScalar(0)) ? RealScalar(EIGEN_PI) : RealScalar(0); } }; template -struct atanh2_default_impl +struct arg_default_impl { - static inline Scalar run(const Scalar&, const Scalar&) + typedef typename NumTraits::Real RealScalar; + EIGEN_DEVICE_FUNC + static inline RealScalar run(const Scalar& x) { + EIGEN_USING_STD(arg); + return arg(x); + } +}; +#endif +template struct arg_impl : arg_default_impl {}; + +template +struct arg_retval +{ + typedef typename NumTraits::Real type; +}; + +/**************************************************************************** +* Implementation of expm1 * +****************************************************************************/ + +// This implementation is based on GSL Math's expm1. +namespace std_fallback { + // fallback expm1 implementation in case there is no expm1(Scalar) function in namespace of Scalar, + // or that there is no suitable std::expm1 function available. Implementation + // attributed to Kahan. See: http://www.plunk.org/~hatch/rightway.php. + template + EIGEN_DEVICE_FUNC inline Scalar expm1(const Scalar& x) { EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar) - return Scalar(0); + typedef typename NumTraits::Real RealScalar; + + EIGEN_USING_STD(exp); + Scalar u = exp(x); + if (numext::equal_strict(u, Scalar(1))) { + return x; + } + Scalar um1 = u - RealScalar(1); + if (numext::equal_strict(um1, Scalar(-1))) { + return RealScalar(-1); + } + + EIGEN_USING_STD(log); + Scalar logu = log(u); + return numext::equal_strict(u, logu) ? u : (u - RealScalar(1)) * x / logu; + } +} + +template +struct expm1_impl { + EIGEN_DEVICE_FUNC static inline Scalar run(const Scalar& x) + { + EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar) + #if EIGEN_HAS_CXX11_MATH + using std::expm1; + #else + using std_fallback::expm1; + #endif + return expm1(x); + } +}; + +template +struct expm1_retval +{ + typedef Scalar type; +}; + +/**************************************************************************** +* Implementation of log * +****************************************************************************/ + +// Complex log defined in MathFunctionsImpl.h. +template EIGEN_DEVICE_FUNC std::complex complex_log(const std::complex& z); + +template +struct log_impl { + EIGEN_DEVICE_FUNC static inline Scalar run(const Scalar& x) + { + EIGEN_USING_STD(log); + return static_cast(log(x)); + } +}; + +template +struct log_impl > { + EIGEN_DEVICE_FUNC static inline std::complex run(const std::complex& z) + { + return complex_log(z); } }; +/**************************************************************************** +* Implementation of log1p * +****************************************************************************/ + +namespace std_fallback { + // fallback log1p implementation in case there is no log1p(Scalar) function in namespace of Scalar, + // or that there is no suitable std::log1p function available + template + EIGEN_DEVICE_FUNC inline Scalar log1p(const Scalar& x) { + EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar) + typedef typename NumTraits::Real RealScalar; + EIGEN_USING_STD(log); + Scalar x1p = RealScalar(1) + x; + Scalar log_1p = log_impl::run(x1p); + const bool is_small = numext::equal_strict(x1p, Scalar(1)); + const bool is_inf = numext::equal_strict(x1p, log_1p); + return (is_small || is_inf) ? x : x * (log_1p / (x1p - RealScalar(1))); + } +} + template -struct atanh2_impl : atanh2_default_impl::IsInteger> {}; +struct log1p_impl { + EIGEN_DEVICE_FUNC static inline Scalar run(const Scalar& x) + { + EIGEN_STATIC_ASSERT_NON_INTEGER(Scalar) + #if EIGEN_HAS_CXX11_MATH + using std::log1p; + #else + using std_fallback::log1p; + #endif + return log1p(x); + } +}; + +// Specialization for complex types that are not supported by std::log1p. +template +struct log1p_impl > { + EIGEN_DEVICE_FUNC static inline std::complex run( + const std::complex& x) { + EIGEN_STATIC_ASSERT_NON_INTEGER(RealScalar) + return std_fallback::log1p(x); + } +}; template -struct atanh2_retval +struct log1p_retval { typedef Scalar type; }; @@ -384,24 +764,26 @@ struct atanh2_retval * Implementation of pow * ****************************************************************************/ -template -struct pow_default_impl +template::IsInteger&&NumTraits::IsInteger> +struct pow_impl { - typedef Scalar retval; - static inline Scalar run(const Scalar& x, const Scalar& y) + //typedef Scalar retval; + typedef typename ScalarBinaryOpTraits >::ReturnType result_type; + static EIGEN_DEVICE_FUNC inline result_type run(const ScalarX& x, const ScalarY& y) { - using std::pow; + EIGEN_USING_STD(pow); return pow(x, y); } }; -template -struct pow_default_impl +template +struct pow_impl { - static inline Scalar run(Scalar x, Scalar y) + typedef ScalarX result_type; + static EIGEN_DEVICE_FUNC inline ScalarX run(ScalarX x, ScalarY y) { - Scalar res(1); - eigen_assert(!NumTraits::IsSigned || y >= 0); + ScalarX res(1); + eigen_assert(!NumTraits::IsSigned || y >= 0); if(y & 1) res *= x; y >>= 1; while(y) @@ -414,15 +796,6 @@ struct pow_default_impl } }; -template -struct pow_impl : pow_default_impl::IsInteger> {}; - -template -struct pow_retval -{ - typedef Scalar type; -}; - /**************************************************************************** * Implementation of random * ****************************************************************************/ @@ -458,109 +831,262 @@ struct random_default_impl }; enum { - floor_log2_terminate, - floor_log2_move_up, - floor_log2_move_down, - floor_log2_bogus + meta_floor_log2_terminate, + meta_floor_log2_move_up, + meta_floor_log2_move_down, + meta_floor_log2_bogus }; -template struct floor_log2_selector +template struct meta_floor_log2_selector { enum { middle = (lower + upper) / 2, - value = (upper <= lower + 1) ? int(floor_log2_terminate) - : (n < (1 << middle)) ? int(floor_log2_move_down) - : (n==0) ? int(floor_log2_bogus) - : int(floor_log2_move_up) + value = (upper <= lower + 1) ? int(meta_floor_log2_terminate) + : (n < (1 << middle)) ? int(meta_floor_log2_move_down) + : (n==0) ? int(meta_floor_log2_bogus) + : int(meta_floor_log2_move_up) }; }; template::value> -struct floor_log2 {}; + int selector = meta_floor_log2_selector::value> +struct meta_floor_log2 {}; template -struct floor_log2 +struct meta_floor_log2 { - enum { value = floor_log2::middle>::value }; + enum { value = meta_floor_log2::middle>::value }; }; template -struct floor_log2 +struct meta_floor_log2 { - enum { value = floor_log2::middle, upper>::value }; + enum { value = meta_floor_log2::middle, upper>::value }; }; template -struct floor_log2 +struct meta_floor_log2 { enum { value = (n >= ((unsigned int)(1) << (lower+1))) ? lower+1 : lower }; }; template -struct floor_log2 +struct meta_floor_log2 { // no value, error at compile time }; -template -struct random_default_impl -{ - static inline Scalar run(const Scalar& x, const Scalar& y) - { - typedef typename conditional::IsSigned,std::ptrdiff_t,std::size_t>::type ScalarX; - if(y=x the result converted to an unsigned long is still correct. - std::size_t range = ScalarX(y)-ScalarX(x); - std::size_t offset = 0; - // rejection sampling - std::size_t divisor = 1; - std::size_t multiplier = 1; - if(range range); - return Scalar(ScalarX(x) + offset); +template +struct count_bits_impl { + static EIGEN_DEVICE_FUNC inline int clz(BitsType bits) { + EIGEN_STATIC_ASSERT( + is_integral::value && !NumTraits::IsSigned, + THIS_TYPE_IS_NOT_SUPPORTED); + int n = CHAR_BIT * sizeof(BitsType); + int shift = n / 2; + while (bits > 0 && shift > 0) { + BitsType y = bits >> shift; + if (y > 0) { + n -= shift; + bits = y; + } + shift /= 2; + } + if (shift == 0) { + --n; + } + return n; } - static inline Scalar run() - { -#ifdef EIGEN_MAKING_DOCS - return run(Scalar(NumTraits::IsSigned ? -10 : 0), Scalar(10)); -#else - enum { rand_bits = floor_log2<(unsigned int)(RAND_MAX)+1>::value, - scalar_bits = sizeof(Scalar) * CHAR_BIT, - shift = EIGEN_PLAIN_ENUM_MAX(0, int(rand_bits) - int(scalar_bits)), - offset = NumTraits::IsSigned ? (1 << (EIGEN_PLAIN_ENUM_MIN(rand_bits,scalar_bits)-1)) : 0 - }; - return Scalar((std::rand() >> shift) - offset); -#endif + static EIGEN_DEVICE_FUNC inline int ctz(BitsType bits) { + EIGEN_STATIC_ASSERT( + is_integral::value && !NumTraits::IsSigned, + THIS_TYPE_IS_NOT_SUPPORTED); + int n = CHAR_BIT * sizeof(BitsType); + int shift = n / 2; + while (bits > 0 && shift > 0) { + BitsType y = bits << shift; + if (y > 0) { + n -= shift; + bits = y; + } + shift /= 2; + } + if (shift == 0) { + --n; + } + return n; } }; -template -struct random_default_impl -{ - static inline Scalar run(const Scalar& x, const Scalar& y) - { - return Scalar(random(real(x), real(y)), - random(imag(x), imag(y))); +// Count leading zeros. +template +EIGEN_DEVICE_FUNC inline int clz(BitsType bits) { + return count_bits_impl::clz(bits); +} + +// Count trailing zeros. +template +EIGEN_DEVICE_FUNC inline int ctz(BitsType bits) { + return count_bits_impl::ctz(bits); +} + +#if EIGEN_COMP_GNUC || EIGEN_COMP_CLANG + +template +struct count_bits_impl::type> { + static const int kNumBits = static_cast(sizeof(BitsType) * CHAR_BIT); + static EIGEN_DEVICE_FUNC inline int clz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + static const int kLeadingBitsOffset = (sizeof(unsigned int) - sizeof(BitsType)) * CHAR_BIT; + return bits == 0 ? kNumBits : __builtin_clz(static_cast(bits)) - kLeadingBitsOffset; } - static inline Scalar run() - { - typedef typename NumTraits::Real RealScalar; - return Scalar(random(), random()); + + static EIGEN_DEVICE_FUNC inline int ctz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + return bits == 0 ? kNumBits : __builtin_ctz(static_cast(bits)); } }; -template -inline EIGEN_MATHFUNC_RETVAL(random, Scalar) random(const Scalar& x, const Scalar& y) -{ - return EIGEN_MATHFUNC_IMPL(random, Scalar)::run(x, y); +template +struct count_bits_impl< + BitsType, typename enable_if::type> { + static const int kNumBits = static_cast(sizeof(BitsType) * CHAR_BIT); + static EIGEN_DEVICE_FUNC inline int clz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + static const int kLeadingBitsOffset = (sizeof(unsigned long) - sizeof(BitsType)) * CHAR_BIT; + return bits == 0 ? kNumBits : __builtin_clzl(static_cast(bits)) - kLeadingBitsOffset; + } + + static EIGEN_DEVICE_FUNC inline int ctz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + return bits == 0 ? kNumBits : __builtin_ctzl(static_cast(bits)); + } +}; + +template +struct count_bits_impl::type> { + static const int kNumBits = static_cast(sizeof(BitsType) * CHAR_BIT); + static EIGEN_DEVICE_FUNC inline int clz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + static const int kLeadingBitsOffset = (sizeof(unsigned long long) - sizeof(BitsType)) * CHAR_BIT; + return bits == 0 ? kNumBits : __builtin_clzll(static_cast(bits)) - kLeadingBitsOffset; + } + + static EIGEN_DEVICE_FUNC inline int ctz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + return bits == 0 ? kNumBits : __builtin_ctzll(static_cast(bits)); + } +}; + +#elif EIGEN_COMP_MSVC + +template +struct count_bits_impl::type> { + static const int kNumBits = static_cast(sizeof(BitsType) * CHAR_BIT); + static EIGEN_DEVICE_FUNC inline int clz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + unsigned long out; + _BitScanReverse(&out, static_cast(bits)); + return bits == 0 ? kNumBits : (kNumBits - 1) - static_cast(out); + } + + static EIGEN_DEVICE_FUNC inline int ctz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + unsigned long out; + _BitScanForward(&out, static_cast(bits)); + return bits == 0 ? kNumBits : static_cast(out); + } +}; + +#ifdef _WIN64 + +template +struct count_bits_impl< + BitsType, typename enable_if::type> { + static const int kNumBits = static_cast(sizeof(BitsType) * CHAR_BIT); + static EIGEN_DEVICE_FUNC inline int clz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + unsigned long out; + _BitScanReverse64(&out, static_cast(bits)); + return bits == 0 ? kNumBits : (kNumBits - 1) - static_cast(out); + } + + static EIGEN_DEVICE_FUNC inline int ctz(BitsType bits) { + EIGEN_STATIC_ASSERT(is_integral::value, THIS_TYPE_IS_NOT_SUPPORTED); + unsigned long out; + _BitScanForward64(&out, static_cast(bits)); + return bits == 0 ? kNumBits : static_cast(out); + } +}; + +#endif // _WIN64 + +#endif // EIGEN_COMP_GNUC || EIGEN_COMP_CLANG + +template +struct random_default_impl { + static inline Scalar run(const Scalar& x, const Scalar& y) { + if (y <= x) return x; + // ScalarU is the unsigned counterpart of Scalar, possibly Scalar itself. + typedef typename make_unsigned::type ScalarU; + // ScalarX is the widest of ScalarU and unsigned int. + // We'll deal only with ScalarX and unsigned int below thus avoiding signed + // types and arithmetic and signed overflows (which are undefined behavior). + typedef typename conditional<(ScalarU(-1) > unsigned(-1)), ScalarU, unsigned>::type ScalarX; + // The following difference doesn't overflow, provided our integer types are two's + // complement and have the same number of padding bits in signed and unsigned variants. + // This is the case in most modern implementations of C++. + ScalarX range = ScalarX(y) - ScalarX(x); + ScalarX offset = 0; + ScalarX divisor = 1; + ScalarX multiplier = 1; + const unsigned rand_max = RAND_MAX; + if (range <= rand_max) divisor = (rand_max + 1) / (range + 1); + else multiplier = 1 + range / (rand_max + 1); + // Rejection sampling. + do { + offset = (unsigned(std::rand()) * multiplier) / divisor; + } while (offset > range); + return Scalar(ScalarX(x) + offset); + } + + static inline Scalar run() + { +#ifdef EIGEN_MAKING_DOCS + return run(Scalar(NumTraits::IsSigned ? -10 : 0), Scalar(10)); +#else + enum { rand_bits = meta_floor_log2<(unsigned int)(RAND_MAX)+1>::value, + scalar_bits = sizeof(Scalar) * CHAR_BIT, + shift = EIGEN_PLAIN_ENUM_MAX(0, int(rand_bits) - int(scalar_bits)), + offset = NumTraits::IsSigned ? (1 << (EIGEN_PLAIN_ENUM_MIN(rand_bits,scalar_bits)-1)) : 0 + }; + return Scalar((std::rand() >> shift) - offset); +#endif + } +}; + +template +struct random_default_impl +{ + static inline Scalar run(const Scalar& x, const Scalar& y) + { + return Scalar(random(x.real(), y.real()), + random(x.imag(), y.imag())); + } + static inline Scalar run() + { + typedef typename NumTraits::Real RealScalar; + return Scalar(random(), random()); + } +}; + +template +inline EIGEN_MATHFUNC_RETVAL(random, Scalar) random(const Scalar& x, const Scalar& y) +{ + return EIGEN_MATHFUNC_IMPL(random, Scalar)::run(x, y); } template @@ -569,97 +1095,932 @@ inline EIGEN_MATHFUNC_RETVAL(random, Scalar) random() return EIGEN_MATHFUNC_IMPL(random, Scalar)::run(); } +// Implementation of is* functions + +// std::is* do not work with fast-math and gcc, std::is* are available on MSVC 2013 and newer, as well as in clang. +#if (EIGEN_HAS_CXX11_MATH && !(EIGEN_COMP_GNUC_STRICT && __FINITE_MATH_ONLY__)) || (EIGEN_COMP_MSVC>=1800) || (EIGEN_COMP_CLANG) +#define EIGEN_USE_STD_FPCLASSIFY 1 +#else +#define EIGEN_USE_STD_FPCLASSIFY 0 +#endif + +template +EIGEN_DEVICE_FUNC +typename internal::enable_if::value,bool>::type +isnan_impl(const T&) { return false; } + +template +EIGEN_DEVICE_FUNC +typename internal::enable_if::value,bool>::type +isinf_impl(const T&) { return false; } + +template +EIGEN_DEVICE_FUNC +typename internal::enable_if::value,bool>::type +isfinite_impl(const T&) { return true; } + +template +EIGEN_DEVICE_FUNC +typename internal::enable_if<(!internal::is_integral::value)&&(!NumTraits::IsComplex),bool>::type +isfinite_impl(const T& x) +{ + #if defined(EIGEN_GPU_COMPILE_PHASE) + return (::isfinite)(x); + #elif EIGEN_USE_STD_FPCLASSIFY + using std::isfinite; + return isfinite EIGEN_NOT_A_MACRO (x); + #else + return x<=NumTraits::highest() && x>=NumTraits::lowest(); + #endif +} + +template +EIGEN_DEVICE_FUNC +typename internal::enable_if<(!internal::is_integral::value)&&(!NumTraits::IsComplex),bool>::type +isinf_impl(const T& x) +{ + #if defined(EIGEN_GPU_COMPILE_PHASE) + return (::isinf)(x); + #elif EIGEN_USE_STD_FPCLASSIFY + using std::isinf; + return isinf EIGEN_NOT_A_MACRO (x); + #else + return x>NumTraits::highest() || x::lowest(); + #endif +} + +template +EIGEN_DEVICE_FUNC +typename internal::enable_if<(!internal::is_integral::value)&&(!NumTraits::IsComplex),bool>::type +isnan_impl(const T& x) +{ + #if defined(EIGEN_GPU_COMPILE_PHASE) + return (::isnan)(x); + #elif EIGEN_USE_STD_FPCLASSIFY + using std::isnan; + return isnan EIGEN_NOT_A_MACRO (x); + #else + return x != x; + #endif +} + +#if (!EIGEN_USE_STD_FPCLASSIFY) + +#if EIGEN_COMP_MSVC + +template EIGEN_DEVICE_FUNC bool isinf_msvc_helper(T x) +{ + return _fpclass(x)==_FPCLASS_NINF || _fpclass(x)==_FPCLASS_PINF; +} + +//MSVC defines a _isnan builtin function, but for double only +#ifndef EIGEN_GPU_COMPILE_PHASE +EIGEN_DEVICE_FUNC inline bool isnan_impl(const long double& x) { return _isnan(x)!=0; } +#endif +EIGEN_DEVICE_FUNC inline bool isnan_impl(const double& x) { return _isnan(x)!=0; } +EIGEN_DEVICE_FUNC inline bool isnan_impl(const float& x) { return _isnan(x)!=0; } + +#ifndef EIGEN_GPU_COMPILE_PHASE +EIGEN_DEVICE_FUNC inline bool isinf_impl(const long double& x) { return isinf_msvc_helper(x); } +#endif +EIGEN_DEVICE_FUNC inline bool isinf_impl(const double& x) { return isinf_msvc_helper(x); } +EIGEN_DEVICE_FUNC inline bool isinf_impl(const float& x) { return isinf_msvc_helper(x); } + +#elif (defined __FINITE_MATH_ONLY__ && __FINITE_MATH_ONLY__ && EIGEN_COMP_GNUC) + +#if EIGEN_GNUC_AT_LEAST(5,0) + #define EIGEN_TMP_NOOPT_ATTRIB EIGEN_DEVICE_FUNC inline __attribute__((optimize("no-finite-math-only"))) +#else + // NOTE the inline qualifier and noinline attribute are both needed: the former is to avoid linking issue (duplicate symbol), + // while the second prevent too aggressive optimizations in fast-math mode: + #define EIGEN_TMP_NOOPT_ATTRIB EIGEN_DEVICE_FUNC inline __attribute__((noinline,optimize("no-finite-math-only"))) +#endif + +#ifndef EIGEN_GPU_COMPILE_PHASE +template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const long double& x) { return __builtin_isnan(x); } +#endif +template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const double& x) { return __builtin_isnan(x); } +template<> EIGEN_TMP_NOOPT_ATTRIB bool isnan_impl(const float& x) { return __builtin_isnan(x); } +template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const double& x) { return __builtin_isinf(x); } +template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const float& x) { return __builtin_isinf(x); } +#ifndef EIGEN_GPU_COMPILE_PHASE +template<> EIGEN_TMP_NOOPT_ATTRIB bool isinf_impl(const long double& x) { return __builtin_isinf(x); } +#endif + +#undef EIGEN_TMP_NOOPT_ATTRIB + +#endif + +#endif + +// The following overload are defined at the end of this file +template EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex& x); +template EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex& x); +template EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex& x); + +template T generic_fast_tanh_float(const T& a_x); } // end namespace internal /**************************************************************************** -* Generic math function * +* Generic math functions * ****************************************************************************/ namespace numext { +#if (!defined(EIGEN_GPUCC) || defined(EIGEN_CONSTEXPR_ARE_DEVICE_FUNC)) +template +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE T mini(const T& x, const T& y) +{ + EIGEN_USING_STD(min) + return min EIGEN_NOT_A_MACRO (x,y); +} + +template +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE T maxi(const T& x, const T& y) +{ + EIGEN_USING_STD(max) + return max EIGEN_NOT_A_MACRO (x,y); +} +#else +template +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE T mini(const T& x, const T& y) +{ + return y < x ? y : x; +} +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE float mini(const float& x, const float& y) +{ + return fminf(x, y); +} +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE double mini(const double& x, const double& y) +{ + return fmin(x, y); +} + +#ifndef EIGEN_GPU_COMPILE_PHASE +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE long double mini(const long double& x, const long double& y) +{ +#if defined(EIGEN_HIPCC) + // no "fminl" on HIP yet + return (x < y) ? x : y; +#else + return fminl(x, y); +#endif +} +#endif + +template +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE T maxi(const T& x, const T& y) +{ + return x < y ? y : x; +} +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE float maxi(const float& x, const float& y) +{ + return fmaxf(x, y); +} +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE double maxi(const double& x, const double& y) +{ + return fmax(x, y); +} +#ifndef EIGEN_GPU_COMPILE_PHASE +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE long double maxi(const long double& x, const long double& y) +{ +#if defined(EIGEN_HIPCC) + // no "fmaxl" on HIP yet + return (x > y) ? x : y; +#else + return fmaxl(x, y); +#endif +} +#endif +#endif + +#if defined(SYCL_DEVICE_ONLY) + + +#define SYCL_SPECIALIZE_SIGNED_INTEGER_TYPES_BINARY(NAME, FUNC) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_char) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_short) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_int) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_long) +#define SYCL_SPECIALIZE_SIGNED_INTEGER_TYPES_UNARY(NAME, FUNC) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_char) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_short) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_int) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_long) +#define SYCL_SPECIALIZE_UNSIGNED_INTEGER_TYPES_BINARY(NAME, FUNC) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_uchar) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_ushort) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_uint) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_ulong) +#define SYCL_SPECIALIZE_UNSIGNED_INTEGER_TYPES_UNARY(NAME, FUNC) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_uchar) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_ushort) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_uint) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_ulong) +#define SYCL_SPECIALIZE_INTEGER_TYPES_BINARY(NAME, FUNC) \ + SYCL_SPECIALIZE_SIGNED_INTEGER_TYPES_BINARY(NAME, FUNC) \ + SYCL_SPECIALIZE_UNSIGNED_INTEGER_TYPES_BINARY(NAME, FUNC) +#define SYCL_SPECIALIZE_INTEGER_TYPES_UNARY(NAME, FUNC) \ + SYCL_SPECIALIZE_SIGNED_INTEGER_TYPES_UNARY(NAME, FUNC) \ + SYCL_SPECIALIZE_UNSIGNED_INTEGER_TYPES_UNARY(NAME, FUNC) +#define SYCL_SPECIALIZE_FLOATING_TYPES_BINARY(NAME, FUNC) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, cl::sycl::cl_float) \ + SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC,cl::sycl::cl_double) +#define SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(NAME, FUNC) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, cl::sycl::cl_float) \ + SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC,cl::sycl::cl_double) +#define SYCL_SPECIALIZE_FLOATING_TYPES_UNARY_FUNC_RET_TYPE(NAME, FUNC, RET_TYPE) \ + SYCL_SPECIALIZE_GEN_UNARY_FUNC(NAME, FUNC, RET_TYPE, cl::sycl::cl_float) \ + SYCL_SPECIALIZE_GEN_UNARY_FUNC(NAME, FUNC, RET_TYPE, cl::sycl::cl_double) + +#define SYCL_SPECIALIZE_GEN_UNARY_FUNC(NAME, FUNC, RET_TYPE, ARG_TYPE) \ +template<> \ + EIGEN_DEVICE_FUNC \ + EIGEN_ALWAYS_INLINE RET_TYPE NAME(const ARG_TYPE& x) { \ + return cl::sycl::FUNC(x); \ + } + +#define SYCL_SPECIALIZE_UNARY_FUNC(NAME, FUNC, TYPE) \ + SYCL_SPECIALIZE_GEN_UNARY_FUNC(NAME, FUNC, TYPE, TYPE) + +#define SYCL_SPECIALIZE_GEN1_BINARY_FUNC(NAME, FUNC, RET_TYPE, ARG_TYPE1, ARG_TYPE2) \ + template<> \ + EIGEN_DEVICE_FUNC \ + EIGEN_ALWAYS_INLINE RET_TYPE NAME(const ARG_TYPE1& x, const ARG_TYPE2& y) { \ + return cl::sycl::FUNC(x, y); \ + } + +#define SYCL_SPECIALIZE_GEN2_BINARY_FUNC(NAME, FUNC, RET_TYPE, ARG_TYPE) \ + SYCL_SPECIALIZE_GEN1_BINARY_FUNC(NAME, FUNC, RET_TYPE, ARG_TYPE, ARG_TYPE) + +#define SYCL_SPECIALIZE_BINARY_FUNC(NAME, FUNC, TYPE) \ + SYCL_SPECIALIZE_GEN2_BINARY_FUNC(NAME, FUNC, TYPE, TYPE) + +SYCL_SPECIALIZE_INTEGER_TYPES_BINARY(mini, min) +SYCL_SPECIALIZE_FLOATING_TYPES_BINARY(mini, fmin) +SYCL_SPECIALIZE_INTEGER_TYPES_BINARY(maxi, max) +SYCL_SPECIALIZE_FLOATING_TYPES_BINARY(maxi, fmax) + +#endif + + template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(real, Scalar) real(const Scalar& x) { return EIGEN_MATHFUNC_IMPL(real, Scalar)::run(x); -} +} template +EIGEN_DEVICE_FUNC inline typename internal::add_const_on_value_type< EIGEN_MATHFUNC_RETVAL(real_ref, Scalar) >::type real_ref(const Scalar& x) { return internal::real_ref_impl::run(x); } template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(real_ref, Scalar) real_ref(Scalar& x) { return EIGEN_MATHFUNC_IMPL(real_ref, Scalar)::run(x); } template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(imag, Scalar) imag(const Scalar& x) { return EIGEN_MATHFUNC_IMPL(imag, Scalar)::run(x); } template +EIGEN_DEVICE_FUNC +inline EIGEN_MATHFUNC_RETVAL(arg, Scalar) arg(const Scalar& x) +{ + return EIGEN_MATHFUNC_IMPL(arg, Scalar)::run(x); +} + +template +EIGEN_DEVICE_FUNC inline typename internal::add_const_on_value_type< EIGEN_MATHFUNC_RETVAL(imag_ref, Scalar) >::type imag_ref(const Scalar& x) { return internal::imag_ref_impl::run(x); } template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(imag_ref, Scalar) imag_ref(Scalar& x) { return EIGEN_MATHFUNC_IMPL(imag_ref, Scalar)::run(x); } template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(conj, Scalar) conj(const Scalar& x) { return EIGEN_MATHFUNC_IMPL(conj, Scalar)::run(x); } template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(abs2, Scalar) abs2(const Scalar& x) { return EIGEN_MATHFUNC_IMPL(abs2, Scalar)::run(x); } +EIGEN_DEVICE_FUNC +inline bool abs2(bool x) { return x; } + +template +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE T absdiff(const T& x, const T& y) +{ + return x > y ? x - y : y - x; +} +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE float absdiff(const float& x, const float& y) +{ + return fabsf(x - y); +} +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE double absdiff(const double& x, const double& y) +{ + return fabs(x - y); +} + +// HIP and CUDA do not support long double. +#ifndef EIGEN_GPU_COMPILE_PHASE +template<> +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE long double absdiff(const long double& x, const long double& y) { + return fabsl(x - y); +} +#endif + template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(norm1, Scalar) norm1(const Scalar& x) { return EIGEN_MATHFUNC_IMPL(norm1, Scalar)::run(x); } template +EIGEN_DEVICE_FUNC inline EIGEN_MATHFUNC_RETVAL(hypot, Scalar) hypot(const Scalar& x, const Scalar& y) { return EIGEN_MATHFUNC_IMPL(hypot, Scalar)::run(x, y); } +#if defined(SYCL_DEVICE_ONLY) + SYCL_SPECIALIZE_FLOATING_TYPES_BINARY(hypot, hypot) +#endif + template -inline EIGEN_MATHFUNC_RETVAL(atanh2, Scalar) atanh2(const Scalar& x, const Scalar& y) +EIGEN_DEVICE_FUNC +inline EIGEN_MATHFUNC_RETVAL(log1p, Scalar) log1p(const Scalar& x) +{ + return EIGEN_MATHFUNC_IMPL(log1p, Scalar)::run(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(log1p, log1p) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float log1p(const float &x) { return ::log1pf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double log1p(const double &x) { return ::log1p(x); } +#endif + +template +EIGEN_DEVICE_FUNC +inline typename internal::pow_impl::result_type pow(const ScalarX& x, const ScalarY& y) +{ + return internal::pow_impl::run(x, y); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_BINARY(pow, pow) +#endif + +template EIGEN_DEVICE_FUNC bool (isnan) (const T &x) { return internal::isnan_impl(x); } +template EIGEN_DEVICE_FUNC bool (isinf) (const T &x) { return internal::isinf_impl(x); } +template EIGEN_DEVICE_FUNC bool (isfinite)(const T &x) { return internal::isfinite_impl(x); } + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY_FUNC_RET_TYPE(isnan, isnan, bool) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY_FUNC_RET_TYPE(isinf, isinf, bool) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY_FUNC_RET_TYPE(isfinite, isfinite, bool) +#endif + +template +EIGEN_DEVICE_FUNC +inline EIGEN_MATHFUNC_RETVAL(rint, Scalar) rint(const Scalar& x) +{ + return EIGEN_MATHFUNC_IMPL(rint, Scalar)::run(x); +} + +template +EIGEN_DEVICE_FUNC +inline EIGEN_MATHFUNC_RETVAL(round, Scalar) round(const Scalar& x) +{ + return EIGEN_MATHFUNC_IMPL(round, Scalar)::run(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(round, round) +#endif + +template +EIGEN_DEVICE_FUNC +T (floor)(const T& x) +{ + EIGEN_USING_STD(floor) + return floor(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(floor, floor) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float floor(const float &x) { return ::floorf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double floor(const double &x) { return ::floor(x); } +#endif + +template +EIGEN_DEVICE_FUNC +T (ceil)(const T& x) { - return EIGEN_MATHFUNC_IMPL(atanh2, Scalar)::run(x, y); + EIGEN_USING_STD(ceil); + return ceil(x); } +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(ceil, ceil) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float ceil(const float &x) { return ::ceilf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double ceil(const double &x) { return ::ceil(x); } +#endif + + +/** Log base 2 for 32 bits positive integers. + * Conveniently returns 0 for x==0. */ +inline int log2(int x) +{ + eigen_assert(x>=0); + unsigned int v(x); + static const int table[32] = { 0, 9, 1, 10, 13, 21, 2, 29, 11, 14, 16, 18, 22, 25, 3, 30, 8, 12, 20, 28, 15, 17, 24, 7, 19, 27, 23, 6, 26, 5, 4, 31 }; + v |= v >> 1; + v |= v >> 2; + v |= v >> 4; + v |= v >> 8; + v |= v >> 16; + return table[(v * 0x07C4ACDDU) >> 27]; +} + +/** \returns the square root of \a x. + * + * It is essentially equivalent to + * \code using std::sqrt; return sqrt(x); \endcode + * but slightly faster for float/double and some compilers (e.g., gcc), thanks to + * specializations when SSE is enabled. + * + * It's usage is justified in performance critical functions, like norm/normalize. + */ template -inline EIGEN_MATHFUNC_RETVAL(pow, Scalar) pow(const Scalar& x, const Scalar& y) +EIGEN_DEVICE_FUNC +EIGEN_ALWAYS_INLINE EIGEN_MATHFUNC_RETVAL(sqrt, Scalar) sqrt(const Scalar& x) { - return EIGEN_MATHFUNC_IMPL(pow, Scalar)::run(x, y); + return EIGEN_MATHFUNC_IMPL(sqrt, Scalar)::run(x); } -// std::isfinite is non standard, so let's define our own version, -// even though it is not very efficient. -template bool (isfinite)(const T& x) +// Boolean specialization, avoids implicit float to bool conversion (-Wimplicit-conversion-floating-point-to-bool). +template<> +EIGEN_DEFINE_FUNCTION_ALLOWING_MULTIPLE_DEFINITIONS EIGEN_DEVICE_FUNC +bool sqrt(const bool &x) { return x; } + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(sqrt, sqrt) +#endif + +/** \returns the reciprocal square root of \a x. **/ +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T rsqrt(const T& x) { - return x::highest() && x>NumTraits::lowest(); + return internal::rsqrt_impl::run(x); +} + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T log(const T &x) { + return internal::log_impl::run(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(log, log) +#endif + + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float log(const float &x) { return ::logf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double log(const double &x) { return ::log(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +typename internal::enable_if::IsSigned || NumTraits::IsComplex,typename NumTraits::Real>::type +abs(const T &x) { + EIGEN_USING_STD(abs); + return abs(x); +} + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +typename internal::enable_if::IsSigned || NumTraits::IsComplex),typename NumTraits::Real>::type +abs(const T &x) { + return x; +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_INTEGER_TYPES_UNARY(abs, abs) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(abs, fabs) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float abs(const float &x) { return ::fabsf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double abs(const double &x) { return ::fabs(x); } + +template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float abs(const std::complex& x) { + return ::hypotf(x.real(), x.imag()); +} + +template <> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double abs(const std::complex& x) { + return ::hypot(x.real(), x.imag()); +} +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T exp(const T &x) { + EIGEN_USING_STD(exp); + return exp(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(exp, exp) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float exp(const float &x) { return ::expf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double exp(const double &x) { return ::exp(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +std::complex exp(const std::complex& x) { + float com = ::expf(x.real()); + float res_real = com * ::cosf(x.imag()); + float res_imag = com * ::sinf(x.imag()); + return std::complex(res_real, res_imag); +} + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +std::complex exp(const std::complex& x) { + double com = ::exp(x.real()); + double res_real = com * ::cos(x.imag()); + double res_imag = com * ::sin(x.imag()); + return std::complex(res_real, res_imag); +} +#endif + +template +EIGEN_DEVICE_FUNC +inline EIGEN_MATHFUNC_RETVAL(expm1, Scalar) expm1(const Scalar& x) +{ + return EIGEN_MATHFUNC_IMPL(expm1, Scalar)::run(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(expm1, expm1) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float expm1(const float &x) { return ::expm1f(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double expm1(const double &x) { return ::expm1(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T cos(const T &x) { + EIGEN_USING_STD(cos); + return cos(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(cos,cos) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float cos(const float &x) { return ::cosf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double cos(const double &x) { return ::cos(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T sin(const T &x) { + EIGEN_USING_STD(sin); + return sin(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(sin, sin) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float sin(const float &x) { return ::sinf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double sin(const double &x) { return ::sin(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T tan(const T &x) { + EIGEN_USING_STD(tan); + return tan(x); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(tan, tan) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float tan(const float &x) { return ::tanf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double tan(const double &x) { return ::tan(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T acos(const T &x) { + EIGEN_USING_STD(acos); + return acos(x); +} + +#if EIGEN_HAS_CXX11_MATH +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T acosh(const T &x) { + EIGEN_USING_STD(acosh); + return static_cast(acosh(x)); +} +#endif + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(acos, acos) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(acosh, acosh) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float acos(const float &x) { return ::acosf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double acos(const double &x) { return ::acos(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T asin(const T &x) { + EIGEN_USING_STD(asin); + return asin(x); +} + +#if EIGEN_HAS_CXX11_MATH +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T asinh(const T &x) { + EIGEN_USING_STD(asinh); + return static_cast(asinh(x)); +} +#endif + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(asin, asin) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(asinh, asinh) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float asin(const float &x) { return ::asinf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double asin(const double &x) { return ::asin(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T atan(const T &x) { + EIGEN_USING_STD(atan); + return static_cast(atan(x)); +} + +#if EIGEN_HAS_CXX11_MATH +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T atanh(const T &x) { + EIGEN_USING_STD(atanh); + return static_cast(atanh(x)); +} +#endif + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(atan, atan) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(atanh, atanh) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float atan(const float &x) { return ::atanf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double atan(const double &x) { return ::atan(x); } +#endif + + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T cosh(const T &x) { + EIGEN_USING_STD(cosh); + return static_cast(cosh(x)); } +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(cosh, cosh) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float cosh(const float &x) { return ::coshf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double cosh(const double &x) { return ::cosh(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T sinh(const T &x) { + EIGEN_USING_STD(sinh); + return static_cast(sinh(x)); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(sinh, sinh) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float sinh(const float &x) { return ::sinhf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double sinh(const double &x) { return ::sinh(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T tanh(const T &x) { + EIGEN_USING_STD(tanh); + return tanh(x); +} + +#if (!defined(EIGEN_GPUCC)) && EIGEN_FAST_MATH && !defined(SYCL_DEVICE_ONLY) +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float tanh(float x) { return internal::generic_fast_tanh_float(x); } +#endif + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_UNARY(tanh, tanh) +#endif + +#if defined(EIGEN_GPUCC) +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float tanh(const float &x) { return ::tanhf(x); } + +template<> EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double tanh(const double &x) { return ::tanh(x); } +#endif + +template +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +T fmod(const T& a, const T& b) { + EIGEN_USING_STD(fmod); + return fmod(a, b); +} + +#if defined(SYCL_DEVICE_ONLY) +SYCL_SPECIALIZE_FLOATING_TYPES_BINARY(fmod, fmod) +#endif + +#if defined(EIGEN_GPUCC) +template <> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +float fmod(const float& a, const float& b) { + return ::fmodf(a, b); +} + +template <> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +double fmod(const double& a, const double& b) { + return ::fmod(a, b); +} +#endif + +#if defined(SYCL_DEVICE_ONLY) +#undef SYCL_SPECIALIZE_SIGNED_INTEGER_TYPES_BINARY +#undef SYCL_SPECIALIZE_SIGNED_INTEGER_TYPES_UNARY +#undef SYCL_SPECIALIZE_UNSIGNED_INTEGER_TYPES_BINARY +#undef SYCL_SPECIALIZE_UNSIGNED_INTEGER_TYPES_UNARY +#undef SYCL_SPECIALIZE_INTEGER_TYPES_BINARY +#undef SYCL_SPECIALIZE_UNSIGNED_INTEGER_TYPES_UNARY +#undef SYCL_SPECIALIZE_FLOATING_TYPES_BINARY +#undef SYCL_SPECIALIZE_FLOATING_TYPES_UNARY +#undef SYCL_SPECIALIZE_FLOATING_TYPES_UNARY_FUNC_RET_TYPE +#undef SYCL_SPECIALIZE_GEN_UNARY_FUNC +#undef SYCL_SPECIALIZE_UNARY_FUNC +#undef SYCL_SPECIALIZE_GEN1_BINARY_FUNC +#undef SYCL_SPECIALIZE_GEN2_BINARY_FUNC +#undef SYCL_SPECIALIZE_BINARY_FUNC +#endif + } // end namespace numext namespace internal { +template +EIGEN_DEVICE_FUNC bool isfinite_impl(const std::complex& x) +{ + return (numext::isfinite)(numext::real(x)) && (numext::isfinite)(numext::imag(x)); +} + +template +EIGEN_DEVICE_FUNC bool isnan_impl(const std::complex& x) +{ + return (numext::isnan)(numext::real(x)) || (numext::isnan)(numext::imag(x)); +} + +template +EIGEN_DEVICE_FUNC bool isinf_impl(const std::complex& x) +{ + return ((numext::isinf)(numext::real(x)) || (numext::isinf)(numext::imag(x))) && (!(numext::isnan)(x)); +} + /**************************************************************************** * Implementation of fuzzy comparisons * ****************************************************************************/ @@ -673,18 +2034,17 @@ template struct scalar_fuzzy_default_impl { typedef typename NumTraits::Real RealScalar; - template + template EIGEN_DEVICE_FUNC static inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const RealScalar& prec) { - using std::abs; - return abs(x) <= abs(y) * prec; + return numext::abs(x) <= numext::abs(y) * prec; } + EIGEN_DEVICE_FUNC static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar& prec) { - using std::min; - using std::abs; - return abs(x - y) <= (min)(abs(x), abs(y)) * prec; + return numext::abs(x - y) <= numext::mini(numext::abs(x), numext::abs(y)) * prec; } + EIGEN_DEVICE_FUNC static inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const RealScalar& prec) { return x <= y || isApprox(x, y, prec); @@ -695,15 +2055,17 @@ template struct scalar_fuzzy_default_impl { typedef typename NumTraits::Real RealScalar; - template + template EIGEN_DEVICE_FUNC static inline bool isMuchSmallerThan(const Scalar& x, const Scalar&, const RealScalar&) { return x == Scalar(0); } + EIGEN_DEVICE_FUNC static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar&) { return x == y; } + EIGEN_DEVICE_FUNC static inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const RealScalar&) { return x <= y; @@ -714,36 +2076,36 @@ template struct scalar_fuzzy_default_impl { typedef typename NumTraits::Real RealScalar; - template + template EIGEN_DEVICE_FUNC static inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const RealScalar& prec) { return numext::abs2(x) <= numext::abs2(y) * prec * prec; } + EIGEN_DEVICE_FUNC static inline bool isApprox(const Scalar& x, const Scalar& y, const RealScalar& prec) { - using std::min; - return numext::abs2(x - y) <= (min)(numext::abs2(x), numext::abs2(y)) * prec * prec; + return numext::abs2(x - y) <= numext::mini(numext::abs2(x), numext::abs2(y)) * prec * prec; } }; template struct scalar_fuzzy_impl : scalar_fuzzy_default_impl::IsComplex, NumTraits::IsInteger> {}; -template +template EIGEN_DEVICE_FUNC inline bool isMuchSmallerThan(const Scalar& x, const OtherScalar& y, const typename NumTraits::Real &precision = NumTraits::dummy_precision()) { return scalar_fuzzy_impl::template isMuchSmallerThan(x, y, precision); } -template +template EIGEN_DEVICE_FUNC inline bool isApprox(const Scalar& x, const Scalar& y, const typename NumTraits::Real &precision = NumTraits::dummy_precision()) { return scalar_fuzzy_impl::isApprox(x, y, precision); } -template +template EIGEN_DEVICE_FUNC inline bool isApproxOrLessThan(const Scalar& x, const Scalar& y, const typename NumTraits::Real &precision = NumTraits::dummy_precision()) { @@ -760,31 +2122,89 @@ template<> struct random_impl { return random(0,1)==0 ? false : true; } + + static inline bool run(const bool& a, const bool& b) + { + return random(a, b)==0 ? false : true; + } }; template<> struct scalar_fuzzy_impl { typedef bool RealScalar; - - template + + template EIGEN_DEVICE_FUNC static inline bool isMuchSmallerThan(const bool& x, const bool&, const bool&) { return !x; } - + + EIGEN_DEVICE_FUNC static inline bool isApprox(bool x, bool y, bool) { return x == y; } + EIGEN_DEVICE_FUNC static inline bool isApproxOrLessThan(const bool& x, const bool& y, const bool&) { return (!x) || y; } - + +}; + +} // end namespace internal + +// Default implementations that rely on other numext implementations +namespace internal { + +// Specialization for complex types that are not supported by std::expm1. +template +struct expm1_impl > { + EIGEN_DEVICE_FUNC static inline std::complex run( + const std::complex& x) { + EIGEN_STATIC_ASSERT_NON_INTEGER(RealScalar) + RealScalar xr = x.real(); + RealScalar xi = x.imag(); + // expm1(z) = exp(z) - 1 + // = exp(x + i * y) - 1 + // = exp(x) * (cos(y) + i * sin(y)) - 1 + // = exp(x) * cos(y) - 1 + i * exp(x) * sin(y) + // Imag(expm1(z)) = exp(x) * sin(y) + // Real(expm1(z)) = exp(x) * cos(y) - 1 + // = exp(x) * cos(y) - 1. + // = expm1(x) + exp(x) * (cos(y) - 1) + // = expm1(x) + exp(x) * (2 * sin(y / 2) ** 2) + RealScalar erm1 = numext::expm1(xr); + RealScalar er = erm1 + RealScalar(1.); + RealScalar sin2 = numext::sin(xi / RealScalar(2.)); + sin2 = sin2 * sin2; + RealScalar s = numext::sin(xi); + RealScalar real_part = erm1 - RealScalar(2.) * er * sin2; + return std::complex(real_part, er * s); + } +}; + +template +struct rsqrt_impl { + EIGEN_DEVICE_FUNC + static EIGEN_ALWAYS_INLINE T run(const T& x) { + return T(1)/numext::sqrt(x); + } +}; + +#if defined(EIGEN_GPU_COMPILE_PHASE) +template +struct conj_impl, true> +{ + EIGEN_DEVICE_FUNC + static inline std::complex run(const std::complex& x) + { + return std::complex(numext::real(x), -numext::imag(x)); + } }; +#endif - } // end namespace internal } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/MathFunctionsImpl.h b/thirdparty/eigen/Eigen/src/Core/MathFunctionsImpl.h new file mode 100644 index 00000000..4eaaaa78 --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/MathFunctionsImpl.h @@ -0,0 +1,200 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2014 Pedro Gonnet (pedro.gonnet@gmail.com) +// Copyright (C) 2016 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_MATHFUNCTIONSIMPL_H +#define EIGEN_MATHFUNCTIONSIMPL_H + +namespace Eigen { + +namespace internal { + +/** \internal \returns the hyperbolic tan of \a a (coeff-wise) + Doesn't do anything fancy, just a 13/6-degree rational interpolant which + is accurate up to a couple of ulps in the (approximate) range [-8, 8], + outside of which tanh(x) = +/-1 in single precision. The input is clamped + to the range [-c, c]. The value c is chosen as the smallest value where + the approximation evaluates to exactly 1. In the reange [-0.0004, 0.0004] + the approxmation tanh(x) ~= x is used for better accuracy as x tends to zero. + + This implementation works on both scalars and packets. +*/ +template +T generic_fast_tanh_float(const T& a_x) +{ + // Clamp the inputs to the range [-c, c] +#ifdef EIGEN_VECTORIZE_FMA + const T plus_clamp = pset1(7.99881172180175781f); + const T minus_clamp = pset1(-7.99881172180175781f); +#else + const T plus_clamp = pset1(7.90531110763549805f); + const T minus_clamp = pset1(-7.90531110763549805f); +#endif + const T tiny = pset1(0.0004f); + const T x = pmax(pmin(a_x, plus_clamp), minus_clamp); + const T tiny_mask = pcmp_lt(pabs(a_x), tiny); + // The monomial coefficients of the numerator polynomial (odd). + const T alpha_1 = pset1(4.89352455891786e-03f); + const T alpha_3 = pset1(6.37261928875436e-04f); + const T alpha_5 = pset1(1.48572235717979e-05f); + const T alpha_7 = pset1(5.12229709037114e-08f); + const T alpha_9 = pset1(-8.60467152213735e-11f); + const T alpha_11 = pset1(2.00018790482477e-13f); + const T alpha_13 = pset1(-2.76076847742355e-16f); + + // The monomial coefficients of the denominator polynomial (even). + const T beta_0 = pset1(4.89352518554385e-03f); + const T beta_2 = pset1(2.26843463243900e-03f); + const T beta_4 = pset1(1.18534705686654e-04f); + const T beta_6 = pset1(1.19825839466702e-06f); + + // Since the polynomials are odd/even, we need x^2. + const T x2 = pmul(x, x); + + // Evaluate the numerator polynomial p. + T p = pmadd(x2, alpha_13, alpha_11); + p = pmadd(x2, p, alpha_9); + p = pmadd(x2, p, alpha_7); + p = pmadd(x2, p, alpha_5); + p = pmadd(x2, p, alpha_3); + p = pmadd(x2, p, alpha_1); + p = pmul(x, p); + + // Evaluate the denominator polynomial q. + T q = pmadd(x2, beta_6, beta_4); + q = pmadd(x2, q, beta_2); + q = pmadd(x2, q, beta_0); + + // Divide the numerator by the denominator. + return pselect(tiny_mask, x, pdiv(p, q)); +} + +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE +RealScalar positive_real_hypot(const RealScalar& x, const RealScalar& y) +{ + // IEEE IEC 6059 special cases. + if ((numext::isinf)(x) || (numext::isinf)(y)) + return NumTraits::infinity(); + if ((numext::isnan)(x) || (numext::isnan)(y)) + return NumTraits::quiet_NaN(); + + EIGEN_USING_STD(sqrt); + RealScalar p, qp; + p = numext::maxi(x,y); + if(p==RealScalar(0)) return RealScalar(0); + qp = numext::mini(y,x) / p; + return p * sqrt(RealScalar(1) + qp*qp); +} + +template +struct hypot_impl +{ + typedef typename NumTraits::Real RealScalar; + static EIGEN_DEVICE_FUNC + inline RealScalar run(const Scalar& x, const Scalar& y) + { + EIGEN_USING_STD(abs); + return positive_real_hypot(abs(x), abs(y)); + } +}; + +// Generic complex sqrt implementation that correctly handles corner cases +// according to https://en.cppreference.com/w/cpp/numeric/complex/sqrt +template +EIGEN_DEVICE_FUNC std::complex complex_sqrt(const std::complex& z) { + // Computes the principal sqrt of the input. + // + // For a complex square root of the number x + i*y. We want to find real + // numbers u and v such that + // (u + i*v)^2 = x + i*y <=> + // u^2 - v^2 + i*2*u*v = x + i*v. + // By equating the real and imaginary parts we get: + // u^2 - v^2 = x + // 2*u*v = y. + // + // For x >= 0, this has the numerically stable solution + // u = sqrt(0.5 * (x + sqrt(x^2 + y^2))) + // v = y / (2 * u) + // and for x < 0, + // v = sign(y) * sqrt(0.5 * (-x + sqrt(x^2 + y^2))) + // u = y / (2 * v) + // + // Letting w = sqrt(0.5 * (|x| + |z|)), + // if x == 0: u = w, v = sign(y) * w + // if x > 0: u = w, v = y / (2 * w) + // if x < 0: u = |y| / (2 * w), v = sign(y) * w + + const T x = numext::real(z); + const T y = numext::imag(z); + const T zero = T(0); + const T w = numext::sqrt(T(0.5) * (numext::abs(x) + numext::hypot(x, y))); + + return + (numext::isinf)(y) ? std::complex(NumTraits::infinity(), y) + : x == zero ? std::complex(w, y < zero ? -w : w) + : x > zero ? std::complex(w, y / (2 * w)) + : std::complex(numext::abs(y) / (2 * w), y < zero ? -w : w ); +} + +// Generic complex rsqrt implementation. +template +EIGEN_DEVICE_FUNC std::complex complex_rsqrt(const std::complex& z) { + // Computes the principal reciprocal sqrt of the input. + // + // For a complex reciprocal square root of the number z = x + i*y. We want to + // find real numbers u and v such that + // (u + i*v)^2 = 1 / (x + i*y) <=> + // u^2 - v^2 + i*2*u*v = x/|z|^2 - i*v/|z|^2. + // By equating the real and imaginary parts we get: + // u^2 - v^2 = x/|z|^2 + // 2*u*v = y/|z|^2. + // + // For x >= 0, this has the numerically stable solution + // u = sqrt(0.5 * (x + |z|)) / |z| + // v = -y / (2 * u * |z|) + // and for x < 0, + // v = -sign(y) * sqrt(0.5 * (-x + |z|)) / |z| + // u = -y / (2 * v * |z|) + // + // Letting w = sqrt(0.5 * (|x| + |z|)), + // if x == 0: u = w / |z|, v = -sign(y) * w / |z| + // if x > 0: u = w / |z|, v = -y / (2 * w * |z|) + // if x < 0: u = |y| / (2 * w * |z|), v = -sign(y) * w / |z| + + const T x = numext::real(z); + const T y = numext::imag(z); + const T zero = T(0); + + const T abs_z = numext::hypot(x, y); + const T w = numext::sqrt(T(0.5) * (numext::abs(x) + abs_z)); + const T woz = w / abs_z; + // Corner cases consistent with 1/sqrt(z) on gcc/clang. + return + abs_z == zero ? std::complex(NumTraits::infinity(), NumTraits::quiet_NaN()) + : ((numext::isinf)(x) || (numext::isinf)(y)) ? std::complex(zero, zero) + : x == zero ? std::complex(woz, y < zero ? woz : -woz) + : x > zero ? std::complex(woz, -y / (2 * w * abs_z)) + : std::complex(numext::abs(y) / (2 * w * abs_z), y < zero ? woz : -woz ); +} + +template +EIGEN_DEVICE_FUNC std::complex complex_log(const std::complex& z) { + // Computes complex log. + T a = numext::abs(z); + EIGEN_USING_STD(atan2); + T b = atan2(z.imag(), z.real()); + return std::complex(numext::log(a), b); +} + +} // end namespace internal + +} // end namespace Eigen + +#endif // EIGEN_MATHFUNCTIONSIMPL_H diff --git a/thirdparty/eigen/Eigen/src/Core/Matrix.h b/thirdparty/eigen/Eigen/src/Core/Matrix.h index 02be142d..29c3b5c6 100644 --- a/thirdparty/eigen/Eigen/src/Core/Matrix.h +++ b/thirdparty/eigen/Eigen/src/Core/Matrix.h @@ -13,6 +13,45 @@ namespace Eigen { +namespace internal { +template +struct traits > +{ +private: + enum { size = internal::size_at_compile_time<_Rows,_Cols>::ret }; + typedef typename find_best_packet<_Scalar,size>::type PacketScalar; + enum { + row_major_bit = _Options&RowMajor ? RowMajorBit : 0, + is_dynamic_size_storage = _MaxRows==Dynamic || _MaxCols==Dynamic, + max_size = is_dynamic_size_storage ? Dynamic : _MaxRows*_MaxCols, + default_alignment = compute_default_alignment<_Scalar,max_size>::value, + actual_alignment = ((_Options&DontAlign)==0) ? default_alignment : 0, + required_alignment = unpacket_traits::alignment, + packet_access_bit = (packet_traits<_Scalar>::Vectorizable && (EIGEN_UNALIGNED_VECTORIZE || (actual_alignment>=required_alignment))) ? PacketAccessBit : 0 + }; + +public: + typedef _Scalar Scalar; + typedef Dense StorageKind; + typedef Eigen::Index StorageIndex; + typedef MatrixXpr XprKind; + enum { + RowsAtCompileTime = _Rows, + ColsAtCompileTime = _Cols, + MaxRowsAtCompileTime = _MaxRows, + MaxColsAtCompileTime = _MaxCols, + Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret, + Options = _Options, + InnerStrideAtCompileTime = 1, + OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime, + + // FIXME, the following flag in only used to define NeedsToAlign in PlainObjectBase + EvaluatorFlags = LinearAccessBit | DirectAccessBit | packet_access_bit | row_major_bit, + Alignment = actual_alignment + }; +}; +} + /** \class Matrix * \ingroup Core_Module * @@ -24,13 +63,13 @@ namespace Eigen { * The %Matrix class encompasses \em both fixed-size and dynamic-size objects (\ref fixedsize "note"). * * The first three template parameters are required: - * \tparam _Scalar \anchor matrix_tparam_scalar Numeric type, e.g. float, double, int or std::complex. - * User defined sclar types are supported as well (see \ref user_defined_scalars "here"). + * \tparam _Scalar Numeric type, e.g. float, double, int or std::complex. + * User defined scalar types are supported as well (see \ref user_defined_scalars "here"). * \tparam _Rows Number of rows, or \b Dynamic * \tparam _Cols Number of columns, or \b Dynamic * * The remaining template parameters are optional -- in most cases you don't have to worry about them. - * \tparam _Options \anchor matrix_tparam_options A combination of either \b #RowMajor or \b #ColMajor, and of either + * \tparam _Options A combination of either \b #RowMajor or \b #ColMajor, and of either * \b #AutoAlign or \b #DontAlign. * The former controls \ref TopicStorageOrders "storage order", and defaults to column-major. The latter controls alignment, which is required * for vectorization. It defaults to aligning matrices except for fixed sizes that aren't a multiple of the packet size. @@ -67,7 +106,7 @@ namespace Eigen { * \endcode * * This class can be extended with the help of the plugin mechanism described on the page - * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN. + * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIX_PLUGIN. * * Some notes: * @@ -97,32 +136,44 @@ namespace Eigen { * are the dimensions of the original matrix, while _Rows and _Cols are Dynamic. * * - * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy, - * \ref TopicStorageOrders + * ABI and storage layout + * + * The table below summarizes the ABI of some possible Matrix instances which is fixed thorough the lifetime of Eigen 3. + * + * + * + * + * + * + *
Matrix typeEquivalent C structure
\code Matrix \endcode\code + * struct { + * T *data; // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0 + * Eigen::Index rows, cols; + * }; + * \endcode
\code + * Matrix + * Matrix \endcode\code + * struct { + * T *data; // with (size_t(data)%EIGEN_MAX_ALIGN_BYTES)==0 + * Eigen::Index size; + * }; + * \endcode
\code Matrix \endcode\code + * struct { + * T data[Rows*Cols]; // with (size_t(data)%A(Rows*Cols*sizeof(T)))==0 + * }; + * \endcode
\code Matrix \endcode\code + * struct { + * T data[MaxRows*MaxCols]; // with (size_t(data)%A(MaxRows*MaxCols*sizeof(T)))==0 + * Eigen::Index rows, cols; + * }; + * \endcode
+ * Note that in this table Rows, Cols, MaxRows and MaxCols are all positive integers. A(S) is defined to the largest possible power-of-two + * smaller to EIGEN_MAX_STATIC_ALIGN_BYTES. + * + * \see MatrixBase for the majority of the API methods for matrices, \ref TopicClassHierarchy, + * \ref TopicStorageOrders */ -namespace internal { -template -struct traits > -{ - typedef _Scalar Scalar; - typedef Dense StorageKind; - typedef DenseIndex Index; - typedef MatrixXpr XprKind; - enum { - RowsAtCompileTime = _Rows, - ColsAtCompileTime = _Cols, - MaxRowsAtCompileTime = _MaxRows, - MaxColsAtCompileTime = _MaxCols, - Flags = compute_matrix_flags<_Scalar, _Rows, _Cols, _Options, _MaxRows, _MaxCols>::ret, - CoeffReadCost = NumTraits::ReadCost, - Options = _Options, - InnerStrideAtCompileTime = 1, - OuterStrideAtCompileTime = (Options&RowMajor) ? ColsAtCompileTime : RowsAtCompileTime - }; -}; -} - template class Matrix : public PlainObjectBase > @@ -151,6 +202,7 @@ class Matrix * * \callgraph */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Matrix& operator=(const Matrix& other) { return Base::_set(other); @@ -167,24 +219,25 @@ class Matrix * remain row-vectors and vectors remain vectors. */ template - EIGEN_STRONG_INLINE Matrix& operator=(const MatrixBase& other) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Matrix& operator=(const DenseBase& other) { return Base::_set(other); } - /* Here, doxygen failed to copy the brief information when using \copydoc */ - /** * \brief Copies the generic expression \a other into *this. * \copydetails DenseBase::operator=(const EigenBase &other) */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Matrix& operator=(const EigenBase &other) { return Base::operator=(other); } template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Matrix& operator=(const ReturnByValue& func) { return Base::operator=(func); @@ -200,67 +253,147 @@ class Matrix * * \sa resize(Index,Index) */ - EIGEN_STRONG_INLINE Matrix() : Base() + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Matrix() : Base() { Base::_check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED } // FIXME is it still needed - Matrix(internal::constructor_without_unaligned_array_assert) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit Matrix(internal::constructor_without_unaligned_array_assert) : Base(internal::constructor_without_unaligned_array_assert()) { Base::_check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED } -#ifdef EIGEN_HAVE_RVALUE_REFERENCES - Matrix(Matrix&& other) +#if EIGEN_HAS_RVALUE_REFERENCES + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Matrix(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_constructible::value) : Base(std::move(other)) { Base::_check_template_params(); - if (RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic) - Base::_set_noalias(other); } - Matrix& operator=(Matrix&& other) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Matrix& operator=(Matrix&& other) EIGEN_NOEXCEPT_IF(std::is_nothrow_move_assignable::value) { - other.swap(*this); + Base::operator=(std::move(other)); return *this; } #endif - /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors +#if EIGEN_HAS_CXX11 + /** \brief Construct a row of column vector with fixed size from an arbitrary number of coefficients. \cpp11 + * + * \only_for_vectors + * + * This constructor is for 1D array or vectors with more than 4 coefficients. + * There exists C++98 analogue constructors for fixed-size array/vector having 1, 2, 3, or 4 coefficients. + * + * \warning To construct a column (resp. row) vector of fixed length, the number of values passed to this + * constructor must match the the fixed number of rows (resp. columns) of \c *this. + * + * Example: \include Matrix_variadic_ctor_cxx11.cpp + * Output: \verbinclude Matrix_variadic_ctor_cxx11.out + * + * \sa Matrix(const std::initializer_list>&) + */ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) + : Base(a0, a1, a2, a3, args...) {} + + /** \brief Constructs a Matrix and initializes it from the coefficients given as initializer-lists grouped by row. \cpp11 + * + * \anchor matrix_constructor_initializer_list + * + * In the general case, the constructor takes a list of rows, each row being represented as a list of coefficients: + * + * Example: \include Matrix_initializer_list_23_cxx11.cpp + * Output: \verbinclude Matrix_initializer_list_23_cxx11.out + * + * Each of the inner initializer lists must contain the exact same number of elements, otherwise an assertion is triggered. * - * Note that this is only useful for dynamic-size vectors. For fixed-size vectors, - * it is redundant to pass the dimension here, so it makes more sense to use the default - * constructor Matrix() instead. + * In the case of a compile-time column vector, implicit transposition from a single row is allowed. + * Therefore VectorXd{{1,2,3,4,5}} is legal and the more verbose syntax + * RowVectorXd{{1},{2},{3},{4},{5}} can be avoided: + * + * Example: \include Matrix_initializer_list_vector_cxx11.cpp + * Output: \verbinclude Matrix_initializer_list_vector_cxx11.out + * + * In the case of fixed-sized matrices, the initializer list sizes must exactly match the matrix sizes, + * and implicit transposition is allowed for compile-time vectors only. + * + * \sa Matrix(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) */ - EIGEN_STRONG_INLINE explicit Matrix(Index dim) - : Base(dim, RowsAtCompileTime == 1 ? 1 : dim, ColsAtCompileTime == 1 ? 1 : dim) + EIGEN_DEVICE_FUNC + explicit EIGEN_STRONG_INLINE Matrix(const std::initializer_list>& list) : Base(list) {} +#endif // end EIGEN_HAS_CXX11 + +#ifndef EIGEN_PARSED_BY_DOXYGEN + + // This constructor is for both 1x1 matrices and dynamic vectors + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit Matrix(const T& x) { Base::_check_template_params(); - EIGEN_STATIC_ASSERT_VECTOR_ONLY(Matrix) - eigen_assert(dim >= 0); - eigen_assert(SizeAtCompileTime == Dynamic || SizeAtCompileTime == dim); - EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED + Base::template _init1(x); } - #ifndef EIGEN_PARSED_BY_DOXYGEN template - EIGEN_STRONG_INLINE Matrix(const T0& x, const T1& y) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Matrix(const T0& x, const T1& y) { Base::_check_template_params(); Base::template _init2(x, y); } - #else + + +#else + /** \brief Constructs a fixed-sized matrix initialized with coefficients starting at \a data */ + EIGEN_DEVICE_FUNC + explicit Matrix(const Scalar *data); + + /** \brief Constructs a vector or row-vector with given dimension. \only_for_vectors + * + * This is useful for dynamic-size vectors. For fixed-size vectors, + * it is redundant to pass these parameters, so one should use the default constructor + * Matrix() instead. + * + * \warning This constructor is disabled for fixed-size \c 1x1 matrices. For instance, + * calling Matrix(1) will call the initialization constructor: Matrix(const Scalar&). + * For fixed-size \c 1x1 matrices it is therefore recommended to use the default + * constructor Matrix() instead, especially when using one of the non standard + * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives). + */ + EIGEN_STRONG_INLINE explicit Matrix(Index dim); + /** \brief Constructs an initialized 1x1 matrix with the given coefficient + * \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...) */ + Matrix(const Scalar& x); /** \brief Constructs an uninitialized matrix with \a rows rows and \a cols columns. * * This is useful for dynamic-size matrices. For fixed-size matrices, * it is redundant to pass these parameters, so one should use the default constructor - * Matrix() instead. */ + * Matrix() instead. + * + * \warning This constructor is disabled for fixed-size \c 1x2 and \c 2x1 vectors. For instance, + * calling Matrix2f(2,1) will call the initialization constructor: Matrix(const Scalar& x, const Scalar& y). + * For fixed-size \c 1x2 or \c 2x1 vectors it is therefore recommended to use the default + * constructor Matrix() instead, especially when using one of the non standard + * \c EIGEN_INITIALIZE_MATRICES_BY_{ZERO,\c NAN} macros (see \ref TopicPreprocessorDirectives). + */ + EIGEN_DEVICE_FUNC Matrix(Index rows, Index cols); - /** \brief Constructs an initialized 2D vector with given coefficients */ + + /** \brief Constructs an initialized 2D vector with given coefficients + * \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...) */ Matrix(const Scalar& x, const Scalar& y); - #endif + #endif // end EIGEN_PARSED_BY_DOXYGEN - /** \brief Constructs an initialized 3D vector with given coefficients */ + /** \brief Constructs an initialized 3D vector with given coefficients + * \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...) + */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z) { Base::_check_template_params(); @@ -269,7 +402,10 @@ class Matrix m_storage.data()[1] = y; m_storage.data()[2] = z; } - /** \brief Constructs an initialized 4D vector with given coefficients */ + /** \brief Constructs an initialized 4D vector with given coefficients + * \sa Matrix(const Scalar&, const Scalar&, const Scalar&, const Scalar&, const ArgTypes&...) + */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Matrix(const Scalar& x, const Scalar& y, const Scalar& z, const Scalar& w) { Base::_check_template_params(); @@ -280,76 +416,35 @@ class Matrix m_storage.data()[3] = w; } - explicit Matrix(const Scalar *data); - /** \brief Constructor copying the value of the expression \a other */ - template - EIGEN_STRONG_INLINE Matrix(const MatrixBase& other) - : Base(other.rows() * other.cols(), other.rows(), other.cols()) - { - // This test resides here, to bring the error messages closer to the user. Normally, these checks - // are performed deeply within the library, thus causing long and scary error traces. - EIGEN_STATIC_ASSERT((internal::is_same::value), - YOU_MIXED_DIFFERENT_NUMERIC_TYPES__YOU_NEED_TO_USE_THE_CAST_METHOD_OF_MATRIXBASE_TO_CAST_NUMERIC_TYPES_EXPLICITLY) - - Base::_check_template_params(); - Base::_set_noalias(other); - } /** \brief Copy constructor */ - EIGEN_STRONG_INLINE Matrix(const Matrix& other) - : Base(other.rows() * other.cols(), other.rows(), other.cols()) - { - Base::_check_template_params(); - Base::_set_noalias(other); - } - /** \brief Copy constructor with in-place evaluation */ - template - EIGEN_STRONG_INLINE Matrix(const ReturnByValue& other) - { - Base::_check_template_params(); - Base::resize(other.rows(), other.cols()); - other.evalTo(*this); - } + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE Matrix(const Matrix& other) : Base(other) + { } /** \brief Copy constructor for generic expressions. * \sa MatrixBase::operator=(const EigenBase&) */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Matrix(const EigenBase &other) - : Base(other.derived().rows() * other.derived().cols(), other.derived().rows(), other.derived().cols()) - { - Base::_check_template_params(); - Base::_resize_to_match(other); - // FIXME/CHECK: isn't *this = other.derived() more efficient. it allows to - // go for pure _set() implementations, right? - *this = other; - } + : Base(other.derived()) + { } - /** \internal - * \brief Override MatrixBase::swap() since for dynamic-sized matrices - * of same type it is enough to swap the data pointers. - */ - template - void swap(MatrixBase const & other) - { this->_swap(other.derived()); } - - inline Index innerStride() const { return 1; } - inline Index outerStride() const { return this->innerSize(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const EIGEN_NOEXCEPT { return 1; } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const EIGEN_NOEXCEPT { return this->innerSize(); } /////////// Geometry module /////////// template + EIGEN_DEVICE_FUNC explicit Matrix(const RotationBase& r); template + EIGEN_DEVICE_FUNC Matrix& operator=(const RotationBase& r); - #ifdef EIGEN2_SUPPORT - template - explicit Matrix(const eigen2_RotationBase& r); - template - Matrix& operator=(const eigen2_RotationBase& r); - #endif - // allow to extend Matrix outside Eigen #ifdef EIGEN_MATRIX_PLUGIN #include EIGEN_MATRIX_PLUGIN @@ -366,7 +461,7 @@ class Matrix * * \ingroup Core_Module * - * Eigen defines several typedef shortcuts for most common matrix and vector types. + * %Eigen defines several typedef shortcuts for most common matrix and vector types. * * The general patterns are the following: * @@ -379,21 +474,35 @@ class Matrix * There are also \c VectorSizeType and \c RowVectorSizeType which are self-explanatory. For example, \c Vector4cf is * a fixed-size vector of 4 complex floats. * + * With \cpp11, template alias are also defined for common sizes. + * They follow the same pattern as above except that the scalar type suffix is replaced by a + * template parameter, i.e.: + * - `MatrixSize` where `Size` can be \c 2,\c 3,\c 4 for fixed size square matrices or \c X for dynamic size. + * - `MatrixXSize` and `MatrixSizeX` where `Size` can be \c 2,\c 3,\c 4 for hybrid dynamic/fixed matrices. + * - `VectorSize` and `RowVectorSize` for column and row vectors. + * + * With \cpp11, you can also use fully generic column and row vector types: `Vector` and `RowVector`. + * * \sa class Matrix */ #define EIGEN_MAKE_TYPEDEFS(Type, TypeSuffix, Size, SizeSuffix) \ /** \ingroup matrixtypedefs */ \ +/** \brief \noop */ \ typedef Matrix Matrix##SizeSuffix##TypeSuffix; \ /** \ingroup matrixtypedefs */ \ +/** \brief \noop */ \ typedef Matrix Vector##SizeSuffix##TypeSuffix; \ /** \ingroup matrixtypedefs */ \ +/** \brief \noop */ \ typedef Matrix RowVector##SizeSuffix##TypeSuffix; #define EIGEN_MAKE_FIXED_TYPEDEFS(Type, TypeSuffix, Size) \ /** \ingroup matrixtypedefs */ \ +/** \brief \noop */ \ typedef Matrix Matrix##Size##X##TypeSuffix; \ /** \ingroup matrixtypedefs */ \ +/** \brief \noop */ \ typedef Matrix Matrix##X##Size##TypeSuffix; #define EIGEN_MAKE_TYPEDEFS_ALL_SIZES(Type, TypeSuffix) \ @@ -415,6 +524,55 @@ EIGEN_MAKE_TYPEDEFS_ALL_SIZES(std::complex, cd) #undef EIGEN_MAKE_TYPEDEFS #undef EIGEN_MAKE_FIXED_TYPEDEFS +#if EIGEN_HAS_CXX11 + +#define EIGEN_MAKE_TYPEDEFS(Size, SizeSuffix) \ +/** \ingroup matrixtypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Matrix##SizeSuffix = Matrix; \ +/** \ingroup matrixtypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Vector##SizeSuffix = Matrix; \ +/** \ingroup matrixtypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using RowVector##SizeSuffix = Matrix; + +#define EIGEN_MAKE_FIXED_TYPEDEFS(Size) \ +/** \ingroup matrixtypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Matrix##Size##X = Matrix; \ +/** \ingroup matrixtypedefs */ \ +/** \brief \cpp11 */ \ +template \ +using Matrix##X##Size = Matrix; + +EIGEN_MAKE_TYPEDEFS(2, 2) +EIGEN_MAKE_TYPEDEFS(3, 3) +EIGEN_MAKE_TYPEDEFS(4, 4) +EIGEN_MAKE_TYPEDEFS(Dynamic, X) +EIGEN_MAKE_FIXED_TYPEDEFS(2) +EIGEN_MAKE_FIXED_TYPEDEFS(3) +EIGEN_MAKE_FIXED_TYPEDEFS(4) + +/** \ingroup matrixtypedefs + * \brief \cpp11 */ +template +using Vector = Matrix; + +/** \ingroup matrixtypedefs + * \brief \cpp11 */ +template +using RowVector = Matrix; + +#undef EIGEN_MAKE_TYPEDEFS +#undef EIGEN_MAKE_FIXED_TYPEDEFS + +#endif // EIGEN_HAS_CXX11 + } // end namespace Eigen #endif // EIGEN_MATRIX_H diff --git a/thirdparty/eigen/Eigen/src/Core/MatrixBase.h b/thirdparty/eigen/Eigen/src/Core/MatrixBase.h index e83ef4dc..d93a7e37 100644 --- a/thirdparty/eigen/Eigen/src/Core/MatrixBase.h +++ b/thirdparty/eigen/Eigen/src/Core/MatrixBase.h @@ -41,9 +41,9 @@ namespace Eigen { * \endcode * * This class can be extended with the help of the plugin mechanism described on the page - * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN. + * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_MATRIXBASE_PLUGIN. * - * \sa \ref TopicClassHierarchy + * \sa \blank \ref TopicClassHierarchy */ template class MatrixBase : public DenseBase @@ -52,7 +52,7 @@ template class MatrixBase #ifndef EIGEN_PARSED_BY_DOXYGEN typedef MatrixBase StorageBaseType; typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; + typedef typename internal::traits::StorageIndex StorageIndex; typedef typename internal::traits::Scalar Scalar; typedef typename internal::packet_traits::type PacketScalar; typedef typename NumTraits::Real RealScalar; @@ -66,7 +66,6 @@ template class MatrixBase using Base::MaxSizeAtCompileTime; using Base::IsVectorAtCompileTime; using Base::Flags; - using Base::CoeffReadCost; using Base::derived; using Base::const_cast_derived; @@ -77,6 +76,7 @@ template class MatrixBase using Base::coeffRef; using Base::lazyAssign; using Base::eval; + using Base::operator-; using Base::operator+=; using Base::operator-=; using Base::operator*=; @@ -98,25 +98,14 @@ template class MatrixBase /** \returns the size of the main diagonal, which is min(rows(),cols()). * \sa rows(), cols(), SizeAtCompileTime. */ - inline Index diagonalSize() const { return (std::min)(rows(),cols()); } + EIGEN_DEVICE_FUNC + inline Index diagonalSize() const { return (numext::mini)(rows(),cols()); } - /** \brief The plain matrix type corresponding to this expression. - * - * This is not necessarily exactly the return type of eval(). In the case of plain matrices, - * the return type of eval() is a const reference to a matrix, not a matrix! It is however guaranteed - * that the return type of eval() is either PlainObject or const PlainObject&. - */ - typedef Matrix::Scalar, - internal::traits::RowsAtCompileTime, - internal::traits::ColsAtCompileTime, - AutoAlign | (internal::traits::Flags&RowMajorBit ? RowMajor : ColMajor), - internal::traits::MaxRowsAtCompileTime, - internal::traits::MaxColsAtCompileTime - > PlainObject; + typedef typename Base::PlainObject PlainObject; #ifndef EIGEN_PARSED_BY_DOXYGEN /** \internal Represents a matrix with all coefficients equal to one another*/ - typedef CwiseNullaryOp,Derived> ConstantReturnType; + typedef CwiseNullaryOp,PlainObject> ConstantReturnType; /** \internal the return type of MatrixBase::adjoint() */ typedef typename internal::conditional::IsComplex, CwiseUnaryOp, ConstTransposeReturnType>, @@ -125,7 +114,7 @@ template class MatrixBase /** \internal Return type of eigenvalues() */ typedef Matrix, internal::traits::ColsAtCompileTime, 1, ColMajor> EigenvaluesReturnType; /** \internal the return type of identity */ - typedef CwiseNullaryOp,Derived> IdentityReturnType; + typedef CwiseNullaryOp,PlainObject> IdentityReturnType; /** \internal the return type of unit vectors */ typedef Block, SquareMatrixType>, internal::traits::RowsAtCompileTime, @@ -133,7 +122,7 @@ template class MatrixBase #endif // not EIGEN_PARSED_BY_DOXYGEN #define EIGEN_CURRENT_STORAGE_BASE_CLASS Eigen::MatrixBase -# include "../plugins/CommonCwiseUnaryOps.h" +#define EIGEN_DOC_UNARY_ADDONS(X,Y) # include "../plugins/CommonCwiseBinaryOps.h" # include "../plugins/MatrixCwiseUnaryOps.h" # include "../plugins/MatrixCwiseBinaryOps.h" @@ -141,41 +130,44 @@ template class MatrixBase # include EIGEN_MATRIXBASE_PLUGIN # endif #undef EIGEN_CURRENT_STORAGE_BASE_CLASS +#undef EIGEN_DOC_UNARY_ADDONS /** Special case of the template operator=, in order to prevent the compiler * from generating a default operator= (issue hit with g++ 4.1) */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const MatrixBase& other); // We cannot inherit here via Base::operator= since it is causing // trouble with MSVC. template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const DenseBase& other); template + EIGEN_DEVICE_FUNC Derived& operator=(const EigenBase& other); template + EIGEN_DEVICE_FUNC Derived& operator=(const ReturnByValue& other); - template - Derived& lazyAssign(const ProductBase& other); - - template - Derived& lazyAssign(const MatrixPowerProduct& other); - template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator+=(const MatrixBase& other); template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator-=(const MatrixBase& other); template - const typename ProductReturnType::Type + EIGEN_DEVICE_FUNC + const Product operator*(const MatrixBase &other) const; template - const typename LazyProductReturnType::Type + EIGEN_DEVICE_FUNC + const Product lazyProduct(const MatrixBase &other) const; template @@ -188,85 +180,90 @@ template class MatrixBase void applyOnTheRight(const EigenBase& other); template - const DiagonalProduct + EIGEN_DEVICE_FUNC + const Product operator*(const DiagonalBase &diagonal) const; template - typename internal::scalar_product_traits::Scalar,typename internal::traits::Scalar>::ReturnType + EIGEN_DEVICE_FUNC + typename ScalarBinaryOpTraits::Scalar,typename internal::traits::Scalar>::ReturnType dot(const MatrixBase& other) const; - #ifdef EIGEN2_SUPPORT - template - Scalar eigen2_dot(const MatrixBase& other) const; - #endif - - RealScalar squaredNorm() const; - RealScalar norm() const; + EIGEN_DEVICE_FUNC RealScalar squaredNorm() const; + EIGEN_DEVICE_FUNC RealScalar norm() const; RealScalar stableNorm() const; RealScalar blueNorm() const; RealScalar hypotNorm() const; - const PlainObject normalized() const; - void normalize(); + EIGEN_DEVICE_FUNC const PlainObject normalized() const; + EIGEN_DEVICE_FUNC const PlainObject stableNormalized() const; + EIGEN_DEVICE_FUNC void normalize(); + EIGEN_DEVICE_FUNC void stableNormalize(); - const AdjointReturnType adjoint() const; - void adjointInPlace(); + EIGEN_DEVICE_FUNC const AdjointReturnType adjoint() const; + EIGEN_DEVICE_FUNC void adjointInPlace(); typedef Diagonal DiagonalReturnType; + EIGEN_DEVICE_FUNC DiagonalReturnType diagonal(); - typedef typename internal::add_const >::type ConstDiagonalReturnType; - ConstDiagonalReturnType diagonal() const; - - template struct DiagonalIndexReturnType { typedef Diagonal Type; }; - template struct ConstDiagonalIndexReturnType { typedef const Diagonal Type; }; - - template typename DiagonalIndexReturnType::Type diagonal(); - template typename ConstDiagonalIndexReturnType::Type diagonal() const; - - typedef Diagonal DiagonalDynamicIndexReturnType; - typedef typename internal::add_const >::type ConstDiagonalDynamicIndexReturnType; - - DiagonalDynamicIndexReturnType diagonal(Index index); - ConstDiagonalDynamicIndexReturnType diagonal(Index index) const; - - #ifdef EIGEN2_SUPPORT - template typename internal::eigen2_part_return_type::type part(); - template const typename internal::eigen2_part_return_type::type part() const; - - // huuuge hack. make Eigen2's matrix.part() work in eigen3. Problem: Diagonal is now a class template instead - // of an integer constant. Solution: overload the part() method template wrt template parameters list. - template class U> - const DiagonalWrapper part() const - { return diagonal().asDiagonal(); } - #endif // EIGEN2_SUPPORT + + typedef Diagonal ConstDiagonalReturnType; + EIGEN_DEVICE_FUNC + const ConstDiagonalReturnType diagonal() const; + + template + EIGEN_DEVICE_FUNC + Diagonal diagonal(); + + template + EIGEN_DEVICE_FUNC + const Diagonal diagonal() const; + + EIGEN_DEVICE_FUNC + Diagonal diagonal(Index index); + EIGEN_DEVICE_FUNC + const Diagonal diagonal(Index index) const; template struct TriangularViewReturnType { typedef TriangularView Type; }; template struct ConstTriangularViewReturnType { typedef const TriangularView Type; }; - template typename TriangularViewReturnType::Type triangularView(); - template typename ConstTriangularViewReturnType::Type triangularView() const; + template + EIGEN_DEVICE_FUNC + typename TriangularViewReturnType::Type triangularView(); + template + EIGEN_DEVICE_FUNC + typename ConstTriangularViewReturnType::Type triangularView() const; template struct SelfAdjointViewReturnType { typedef SelfAdjointView Type; }; template struct ConstSelfAdjointViewReturnType { typedef const SelfAdjointView Type; }; - template typename SelfAdjointViewReturnType::Type selfadjointView(); - template typename ConstSelfAdjointViewReturnType::Type selfadjointView() const; + template + EIGEN_DEVICE_FUNC + typename SelfAdjointViewReturnType::Type selfadjointView(); + template + EIGEN_DEVICE_FUNC + typename ConstSelfAdjointViewReturnType::Type selfadjointView() const; const SparseView sparseView(const Scalar& m_reference = Scalar(0), const typename NumTraits::Real& m_epsilon = NumTraits::dummy_precision()) const; - static const IdentityReturnType Identity(); - static const IdentityReturnType Identity(Index rows, Index cols); - static const BasisReturnType Unit(Index size, Index i); - static const BasisReturnType Unit(Index i); - static const BasisReturnType UnitX(); - static const BasisReturnType UnitY(); - static const BasisReturnType UnitZ(); - static const BasisReturnType UnitW(); - + EIGEN_DEVICE_FUNC static const IdentityReturnType Identity(); + EIGEN_DEVICE_FUNC static const IdentityReturnType Identity(Index rows, Index cols); + EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index size, Index i); + EIGEN_DEVICE_FUNC static const BasisReturnType Unit(Index i); + EIGEN_DEVICE_FUNC static const BasisReturnType UnitX(); + EIGEN_DEVICE_FUNC static const BasisReturnType UnitY(); + EIGEN_DEVICE_FUNC static const BasisReturnType UnitZ(); + EIGEN_DEVICE_FUNC static const BasisReturnType UnitW(); + + EIGEN_DEVICE_FUNC const DiagonalWrapper asDiagonal() const; const PermutationWrapper asPermutation() const; + EIGEN_DEVICE_FUNC Derived& setIdentity(); + EIGEN_DEVICE_FUNC Derived& setIdentity(Index rows, Index cols); + EIGEN_DEVICE_FUNC Derived& setUnit(Index i); + EIGEN_DEVICE_FUNC Derived& setUnit(Index newSize, Index i); bool isIdentity(const RealScalar& prec = NumTraits::dummy_precision()) const; bool isDiagonal(const RealScalar& prec = NumTraits::dummy_precision()) const; @@ -284,7 +281,7 @@ template class MatrixBase * fuzzy comparison such as isApprox() * \sa isApprox(), operator!= */ template - inline bool operator==(const MatrixBase& other) const + EIGEN_DEVICE_FUNC inline bool operator==(const MatrixBase& other) const { return cwiseEqual(other).all(); } /** \returns true if at least one pair of coefficients of \c *this and \a other are not exactly equal to each other. @@ -292,143 +289,142 @@ template class MatrixBase * fuzzy comparison such as isApprox() * \sa isApprox(), operator== */ template - inline bool operator!=(const MatrixBase& other) const + EIGEN_DEVICE_FUNC inline bool operator!=(const MatrixBase& other) const { return cwiseNotEqual(other).any(); } - NoAlias noalias(); + NoAlias EIGEN_DEVICE_FUNC noalias(); - inline const ForceAlignedAccess forceAlignedAccess() const; - inline ForceAlignedAccess forceAlignedAccess(); - template inline typename internal::add_const_on_value_type,Derived&>::type>::type forceAlignedAccessIf() const; - template inline typename internal::conditional,Derived&>::type forceAlignedAccessIf(); + // TODO forceAlignedAccess is temporarily disabled + // Need to find a nicer workaround. + inline const Derived& forceAlignedAccess() const { return derived(); } + inline Derived& forceAlignedAccess() { return derived(); } + template inline const Derived& forceAlignedAccessIf() const { return derived(); } + template inline Derived& forceAlignedAccessIf() { return derived(); } - Scalar trace() const; + EIGEN_DEVICE_FUNC Scalar trace() const; -/////////// Array module /////////// + template EIGEN_DEVICE_FUNC RealScalar lpNorm() const; - template RealScalar lpNorm() const; - - MatrixBase& matrix() { return *this; } - const MatrixBase& matrix() const { return *this; } + EIGEN_DEVICE_FUNC MatrixBase& matrix() { return *this; } + EIGEN_DEVICE_FUNC const MatrixBase& matrix() const { return *this; } /** \returns an \link Eigen::ArrayBase Array \endlink expression of this matrix * \sa ArrayBase::matrix() */ - ArrayWrapper array() { return derived(); } - const ArrayWrapper array() const { return derived(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ArrayWrapper array() { return ArrayWrapper(derived()); } + /** \returns a const \link Eigen::ArrayBase Array \endlink expression of this matrix + * \sa ArrayBase::matrix() */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const ArrayWrapper array() const { return ArrayWrapper(derived()); } /////////// LU module /////////// - const FullPivLU fullPivLu() const; - const PartialPivLU partialPivLu() const; + inline const FullPivLU fullPivLu() const; + inline const PartialPivLU partialPivLu() const; - #if EIGEN2_SUPPORT_STAGE < STAGE20_RESOLVE_API_CONFLICTS - const LU lu() const; - #endif + inline const PartialPivLU lu() const; - #ifdef EIGEN2_SUPPORT - const LU eigen2_lu() const; - #endif - - #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS - const PartialPivLU lu() const; - #endif - - #ifdef EIGEN2_SUPPORT - template - void computeInverse(MatrixBase *result) const { - *result = this->inverse(); - } - #endif + EIGEN_DEVICE_FUNC + inline const Inverse inverse() const; - const internal::inverse_impl inverse() const; template - void computeInverseAndDetWithCheck( + inline void computeInverseAndDetWithCheck( ResultType& inverse, typename ResultType::Scalar& determinant, bool& invertible, const RealScalar& absDeterminantThreshold = NumTraits::dummy_precision() ) const; + template - void computeInverseWithCheck( + inline void computeInverseWithCheck( ResultType& inverse, bool& invertible, const RealScalar& absDeterminantThreshold = NumTraits::dummy_precision() ) const; + + EIGEN_DEVICE_FUNC Scalar determinant() const; /////////// Cholesky module /////////// - const LLT llt() const; - const LDLT ldlt() const; + inline const LLT llt() const; + inline const LDLT ldlt() const; /////////// QR module /////////// - const HouseholderQR householderQr() const; - const ColPivHouseholderQR colPivHouseholderQr() const; - const FullPivHouseholderQR fullPivHouseholderQr() const; - - #ifdef EIGEN2_SUPPORT - const QR qr() const; - #endif + inline const HouseholderQR householderQr() const; + inline const ColPivHouseholderQR colPivHouseholderQr() const; + inline const FullPivHouseholderQR fullPivHouseholderQr() const; + inline const CompleteOrthogonalDecomposition completeOrthogonalDecomposition() const; - EigenvaluesReturnType eigenvalues() const; - RealScalar operatorNorm() const; +/////////// Eigenvalues module /////////// -/////////// SVD module /////////// + inline EigenvaluesReturnType eigenvalues() const; + inline RealScalar operatorNorm() const; - JacobiSVD jacobiSvd(unsigned int computationOptions = 0) const; +/////////// SVD module /////////// - #ifdef EIGEN2_SUPPORT - SVD svd() const; - #endif + inline JacobiSVD jacobiSvd(unsigned int computationOptions = 0) const; + inline BDCSVD bdcSvd(unsigned int computationOptions = 0) const; /////////// Geometry module /////////// #ifndef EIGEN_PARSED_BY_DOXYGEN /// \internal helper struct to form the return type of the cross product template struct cross_product_return_type { - typedef typename internal::scalar_product_traits::Scalar,typename internal::traits::Scalar>::ReturnType Scalar; + typedef typename ScalarBinaryOpTraits::Scalar,typename internal::traits::Scalar>::ReturnType Scalar; typedef Matrix type; }; #endif // EIGEN_PARSED_BY_DOXYGEN template - typename cross_product_return_type::type + EIGEN_DEVICE_FUNC +#ifndef EIGEN_PARSED_BY_DOXYGEN + inline typename cross_product_return_type::type +#else + inline PlainObject +#endif cross(const MatrixBase& other) const; + template - PlainObject cross3(const MatrixBase& other) const; - PlainObject unitOrthogonal(void) const; - Matrix eulerAngles(Index a0, Index a1, Index a2) const; - - #if EIGEN2_SUPPORT_STAGE > STAGE20_RESOLVE_API_CONFLICTS - ScalarMultipleReturnType operator*(const UniformScaling& s) const; + EIGEN_DEVICE_FUNC + inline PlainObject cross3(const MatrixBase& other) const; + + EIGEN_DEVICE_FUNC + inline PlainObject unitOrthogonal(void) const; + + EIGEN_DEVICE_FUNC + inline Matrix eulerAngles(Index a0, Index a1, Index a2) const; + // put this as separate enum value to work around possible GCC 4.3 bug (?) - enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1?Vertical:Horizontal }; + enum { HomogeneousReturnTypeDirection = ColsAtCompileTime==1&&RowsAtCompileTime==1 ? ((internal::traits::Flags&RowMajorBit)==RowMajorBit ? Horizontal : Vertical) + : ColsAtCompileTime==1 ? Vertical : Horizontal }; typedef Homogeneous HomogeneousReturnType; - HomogeneousReturnType homogeneous() const; - #endif - + EIGEN_DEVICE_FUNC + inline HomogeneousReturnType homogeneous() const; + enum { SizeMinusOne = SizeAtCompileTime==Dynamic ? Dynamic : SizeAtCompileTime-1 }; typedef Block::ColsAtCompileTime==1 ? SizeMinusOne : 1, internal::traits::ColsAtCompileTime==1 ? 1 : SizeMinusOne> ConstStartMinusOne; - typedef CwiseUnaryOp::Scalar>, - const ConstStartMinusOne > HNormalizedReturnType; - - const HNormalizedReturnType hnormalized() const; + typedef EIGEN_EXPR_BINARYOP_SCALAR_RETURN_TYPE(ConstStartMinusOne,Scalar,quotient) HNormalizedReturnType; + EIGEN_DEVICE_FUNC + inline const HNormalizedReturnType hnormalized() const; ////////// Householder module /////////// + EIGEN_DEVICE_FUNC void makeHouseholderInPlace(Scalar& tau, RealScalar& beta); template + EIGEN_DEVICE_FUNC void makeHouseholder(EssentialPart& essential, Scalar& tau, RealScalar& beta) const; template + EIGEN_DEVICE_FUNC void applyHouseholderOnTheLeft(const EssentialPart& essential, const Scalar& tau, Scalar* workspace); template + EIGEN_DEVICE_FUNC void applyHouseholderOnTheRight(const EssentialPart& essential, const Scalar& tau, Scalar* workspace); @@ -436,8 +432,10 @@ template class MatrixBase ///////// Jacobi module ///////// template + EIGEN_DEVICE_FUNC void applyOnTheLeft(Index p, Index q, const JacobiRotation& j); template + EIGEN_DEVICE_FUNC void applyOnTheRight(Index p, Index q, const JacobiRotation& j); ///////// SparseCore module ///////// @@ -452,58 +450,38 @@ template class MatrixBase ///////// MatrixFunctions module ///////// typedef typename internal::stem_function::type StemFunction; - const MatrixExponentialReturnValue exp() const; +#define EIGEN_MATRIX_FUNCTION(ReturnType, Name, Description) \ + /** \returns an expression of the matrix Description of \c *this. \brief This function requires the unsupported MatrixFunctions module. To compute the coefficient-wise Description use ArrayBase::##Name . */ \ + const ReturnType Name() const; +#define EIGEN_MATRIX_FUNCTION_1(ReturnType, Name, Description, Argument) \ + /** \returns an expression of the matrix Description of \c *this. \brief This function requires the unsupported MatrixFunctions module. To compute the coefficient-wise Description use ArrayBase::##Name . */ \ + const ReturnType Name(Argument) const; + + EIGEN_MATRIX_FUNCTION(MatrixExponentialReturnValue, exp, exponential) + /** \brief Helper function for the unsupported MatrixFunctions module.*/ const MatrixFunctionReturnValue matrixFunction(StemFunction f) const; - const MatrixFunctionReturnValue cosh() const; - const MatrixFunctionReturnValue sinh() const; - const MatrixFunctionReturnValue cos() const; - const MatrixFunctionReturnValue sin() const; - const MatrixSquareRootReturnValue sqrt() const; - const MatrixLogarithmReturnValue log() const; - const MatrixPowerReturnValue pow(const RealScalar& p) const; - -#ifdef EIGEN2_SUPPORT - template - Derived& operator+=(const Flagged, 0, - EvalBeforeAssigningBit>& other); - - template - Derived& operator-=(const Flagged, 0, - EvalBeforeAssigningBit>& other); - - /** \deprecated because .lazy() is deprecated - * Overloaded for cache friendly product evaluation */ - template - Derived& lazyAssign(const Flagged& other) - { return lazyAssign(other._expression()); } - - template - const Flagged marked() const; - const Flagged lazy() const; - - inline const Cwise cwise() const; - inline Cwise cwise(); - - VectorBlock start(Index size); - const VectorBlock start(Index size) const; - VectorBlock end(Index size); - const VectorBlock end(Index size) const; - template VectorBlock start(); - template const VectorBlock start() const; - template VectorBlock end(); - template const VectorBlock end() const; - - Minor minor(Index row, Index col); - const Minor minor(Index row, Index col) const; + EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cosh, hyperbolic cosine) + EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sinh, hyperbolic sine) +#if EIGEN_HAS_CXX11_MATH + EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, atanh, inverse hyperbolic cosine) + EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, acosh, inverse hyperbolic cosine) + EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, asinh, inverse hyperbolic sine) #endif + EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, cos, cosine) + EIGEN_MATRIX_FUNCTION(MatrixFunctionReturnValue, sin, sine) + EIGEN_MATRIX_FUNCTION(MatrixSquareRootReturnValue, sqrt, square root) + EIGEN_MATRIX_FUNCTION(MatrixLogarithmReturnValue, log, logarithm) + EIGEN_MATRIX_FUNCTION_1(MatrixPowerReturnValue, pow, power to \c p, const RealScalar& p) + EIGEN_MATRIX_FUNCTION_1(MatrixComplexPowerReturnValue, pow, power to \c p, const std::complex& p) protected: - MatrixBase() : Base() {} + EIGEN_DEFAULT_COPY_CONSTRUCTOR(MatrixBase) + EIGEN_DEFAULT_EMPTY_CONSTRUCTOR_AND_DESTRUCTOR(MatrixBase) private: - explicit MatrixBase(int); - MatrixBase(int,int); - template explicit MatrixBase(const MatrixBase&); + EIGEN_DEVICE_FUNC explicit MatrixBase(int); + EIGEN_DEVICE_FUNC MatrixBase(int,int); + template EIGEN_DEVICE_FUNC explicit MatrixBase(const MatrixBase&); protected: // mixing arrays and matrices is not legal template Derived& operator+=(const ArrayBase& ) diff --git a/thirdparty/eigen/Eigen/src/Core/NestByValue.h b/thirdparty/eigen/Eigen/src/Core/NestByValue.h index a893b176..b4275768 100644 --- a/thirdparty/eigen/Eigen/src/Core/NestByValue.h +++ b/thirdparty/eigen/Eigen/src/Core/NestByValue.h @@ -13,25 +13,28 @@ namespace Eigen { +namespace internal { +template +struct traits > : public traits +{ + enum { + Flags = traits::Flags & ~NestByRefBit + }; +}; +} + /** \class NestByValue * \ingroup Core_Module * * \brief Expression which must be nested by value * - * \param ExpressionType the type of the object of which we are requiring nesting-by-value + * \tparam ExpressionType the type of the object of which we are requiring nesting-by-value * * This class is the return type of MatrixBase::nestByValue() * and most of the time this is the only way it is used. * * \sa MatrixBase::nestByValue() */ - -namespace internal { -template -struct traits > : public traits -{}; -} - template class NestByValue : public internal::dense_xpr_base< NestByValue >::type { @@ -40,58 +43,14 @@ template class NestByValue typedef typename internal::dense_xpr_base::type Base; EIGEN_DENSE_PUBLIC_INTERFACE(NestByValue) - inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {} - - inline Index rows() const { return m_expression.rows(); } - inline Index cols() const { return m_expression.cols(); } - inline Index outerStride() const { return m_expression.outerStride(); } - inline Index innerStride() const { return m_expression.innerStride(); } - - inline const CoeffReturnType coeff(Index row, Index col) const - { - return m_expression.coeff(row, col); - } - - inline Scalar& coeffRef(Index row, Index col) - { - return m_expression.const_cast_derived().coeffRef(row, col); - } - - inline const CoeffReturnType coeff(Index index) const - { - return m_expression.coeff(index); - } - - inline Scalar& coeffRef(Index index) - { - return m_expression.const_cast_derived().coeffRef(index); - } - - template - inline const PacketScalar packet(Index row, Index col) const - { - return m_expression.template packet(row, col); - } - - template - inline void writePacket(Index row, Index col, const PacketScalar& x) - { - m_expression.const_cast_derived().template writePacket(row, col, x); - } - - template - inline const PacketScalar packet(Index index) const - { - return m_expression.template packet(index); - } - - template - inline void writePacket(Index index, const PacketScalar& x) - { - m_expression.const_cast_derived().template writePacket(index, x); - } - - operator const ExpressionType&() const { return m_expression; } + EIGEN_DEVICE_FUNC explicit inline NestByValue(const ExpressionType& matrix) : m_expression(matrix) {} + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index rows() const EIGEN_NOEXCEPT { return m_expression.rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index cols() const EIGEN_NOEXCEPT { return m_expression.cols(); } + + EIGEN_DEVICE_FUNC operator const ExpressionType&() const { return m_expression; } + + EIGEN_DEVICE_FUNC const ExpressionType& nestedExpression() const { return m_expression; } protected: const ExpressionType m_expression; @@ -100,12 +59,27 @@ template class NestByValue /** \returns an expression of the temporary version of *this. */ template -inline const NestByValue +EIGEN_DEVICE_FUNC inline const NestByValue DenseBase::nestByValue() const { return NestByValue(derived()); } +namespace internal { + +// Evaluator of Solve -> eval into a temporary +template +struct evaluator > + : public evaluator +{ + typedef evaluator Base; + + EIGEN_DEVICE_FUNC explicit evaluator(const NestByValue& xpr) + : Base(xpr.nestedExpression()) + {} +}; +} + } // end namespace Eigen #endif // EIGEN_NESTBYVALUE_H diff --git a/thirdparty/eigen/Eigen/src/Core/NoAlias.h b/thirdparty/eigen/Eigen/src/Core/NoAlias.h index 768bfb18..570283d9 100644 --- a/thirdparty/eigen/Eigen/src/Core/NoAlias.h +++ b/thirdparty/eigen/Eigen/src/Core/NoAlias.h @@ -17,7 +17,7 @@ namespace Eigen { * * \brief Pseudo expression providing an operator = assuming no aliasing * - * \param ExpressionType the type of the object on which to do the lazy assignment + * \tparam ExpressionType the type of the object on which to do the lazy assignment * * This class represents an expression with special assignment operators * assuming no aliasing between the target expression and the source expression. @@ -30,62 +30,37 @@ namespace Eigen { template class StorageBase> class NoAlias { - typedef typename ExpressionType::Scalar Scalar; public: - NoAlias(ExpressionType& expression) : m_expression(expression) {} - - /** Behaves like MatrixBase::lazyAssign(other) - * \sa MatrixBase::lazyAssign() */ + typedef typename ExpressionType::Scalar Scalar; + + EIGEN_DEVICE_FUNC + explicit NoAlias(ExpressionType& expression) : m_expression(expression) {} + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ExpressionType& operator=(const StorageBase& other) - { return internal::assign_selector::run(m_expression,other.derived()); } - - /** \sa MatrixBase::operator+= */ + { + call_assignment_no_alias(m_expression, other.derived(), internal::assign_op()); + return m_expression; + } + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ExpressionType& operator+=(const StorageBase& other) { - typedef SelfCwiseBinaryOp, ExpressionType, OtherDerived> SelfAdder; - SelfAdder tmp(m_expression); - typedef typename internal::nested::type OtherDerivedNested; - typedef typename internal::remove_all::type _OtherDerivedNested; - internal::assign_selector::run(tmp,OtherDerivedNested(other.derived())); + call_assignment_no_alias(m_expression, other.derived(), internal::add_assign_op()); return m_expression; } - - /** \sa MatrixBase::operator-= */ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE ExpressionType& operator-=(const StorageBase& other) { - typedef SelfCwiseBinaryOp, ExpressionType, OtherDerived> SelfAdder; - SelfAdder tmp(m_expression); - typedef typename internal::nested::type OtherDerivedNested; - typedef typename internal::remove_all::type _OtherDerivedNested; - internal::assign_selector::run(tmp,OtherDerivedNested(other.derived())); + call_assignment_no_alias(m_expression, other.derived(), internal::sub_assign_op()); return m_expression; } -#ifndef EIGEN_PARSED_BY_DOXYGEN - template - EIGEN_STRONG_INLINE ExpressionType& operator+=(const ProductBase& other) - { other.derived().addTo(m_expression); return m_expression; } - - template - EIGEN_STRONG_INLINE ExpressionType& operator-=(const ProductBase& other) - { other.derived().subTo(m_expression); return m_expression; } - - template - EIGEN_STRONG_INLINE ExpressionType& operator+=(const CoeffBasedProduct& other) - { return m_expression.derived() += CoeffBasedProduct(other.lhs(), other.rhs()); } - - template - EIGEN_STRONG_INLINE ExpressionType& operator-=(const CoeffBasedProduct& other) - { return m_expression.derived() -= CoeffBasedProduct(other.lhs(), other.rhs()); } - - template - ExpressionType& operator=(const ReturnByValue& func) - { return m_expression = func; } -#endif - + EIGEN_DEVICE_FUNC ExpressionType& expression() const { return m_expression; @@ -100,10 +75,10 @@ class NoAlias * * More precisely, noalias() allows to bypass the EvalBeforeAssignBit flag. * Currently, even though several expressions may alias, only product - * expressions have this flag. Therefore, noalias() is only usefull when + * expressions have this flag. Therefore, noalias() is only useful when * the source expression contains a matrix product. * - * Here are some examples where noalias is usefull: + * Here are some examples where noalias is useful: * \code * D.noalias() = A * B; * D.noalias() += A.transpose() * B; @@ -124,9 +99,9 @@ class NoAlias * \sa class NoAlias */ template -NoAlias MatrixBase::noalias() +NoAlias EIGEN_DEVICE_FUNC MatrixBase::noalias() { - return derived(); + return NoAlias(derived()); } } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/NumTraits.h b/thirdparty/eigen/Eigen/src/Core/NumTraits.h index bac9e50b..7c2c50b8 100644 --- a/thirdparty/eigen/Eigen/src/Core/NumTraits.h +++ b/thirdparty/eigen/Eigen/src/Core/NumTraits.h @@ -12,41 +12,145 @@ namespace Eigen { +namespace internal { + +// default implementation of digits10(), based on numeric_limits if specialized, +// 0 for integer types, and log10(epsilon()) otherwise. +template< typename T, + bool use_numeric_limits = std::numeric_limits::is_specialized, + bool is_integer = NumTraits::IsInteger> +struct default_digits10_impl +{ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static int run() { return std::numeric_limits::digits10; } +}; + +template +struct default_digits10_impl // Floating point +{ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static int run() { + using std::log10; + using std::ceil; + typedef typename NumTraits::Real Real; + return int(ceil(-log10(NumTraits::epsilon()))); + } +}; + +template +struct default_digits10_impl // Integer +{ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static int run() { return 0; } +}; + + +// default implementation of digits(), based on numeric_limits if specialized, +// 0 for integer types, and log2(epsilon()) otherwise. +template< typename T, + bool use_numeric_limits = std::numeric_limits::is_specialized, + bool is_integer = NumTraits::IsInteger> +struct default_digits_impl +{ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static int run() { return std::numeric_limits::digits; } +}; + +template +struct default_digits_impl // Floating point +{ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static int run() { + using std::log; + using std::ceil; + typedef typename NumTraits::Real Real; + return int(ceil(-log(NumTraits::epsilon())/log(static_cast(2)))); + } +}; + +template +struct default_digits_impl // Integer +{ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static int run() { return 0; } +}; + +} // end namespace internal + +namespace numext { +/** \internal bit-wise cast without changing the underlying bit representation. */ + +// TODO: Replace by std::bit_cast (available in C++20) +template +EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC Tgt bit_cast(const Src& src) { +#if EIGEN_HAS_TYPE_TRAITS + // The behaviour of memcpy is not specified for non-trivially copyable types + EIGEN_STATIC_ASSERT(std::is_trivially_copyable::value, THIS_TYPE_IS_NOT_SUPPORTED); + EIGEN_STATIC_ASSERT(std::is_trivially_copyable::value && std::is_default_constructible::value, + THIS_TYPE_IS_NOT_SUPPORTED); +#endif + + EIGEN_STATIC_ASSERT(sizeof(Src) == sizeof(Tgt), THIS_TYPE_IS_NOT_SUPPORTED); + Tgt tgt; + EIGEN_USING_STD(memcpy) + memcpy(&tgt, &src, sizeof(Tgt)); + return tgt; +} +} // namespace numext + +// clang-format off /** \class NumTraits * \ingroup Core_Module * * \brief Holds information about the various numeric (i.e. scalar) types allowed by Eigen. * - * \param T the numeric type at hand + * \tparam T the numeric type at hand * * This class stores enums, typedefs and static methods giving information about a numeric type. * * The provided data consists of: - * \li A typedef \a Real, giving the "real part" type of \a T. If \a T is already real, - * then \a Real is just a typedef to \a T. If \a T is \c std::complex then \a Real + * \li A typedef \c Real, giving the "real part" type of \a T. If \a T is already real, + * then \c Real is just a typedef to \a T. If \a T is `std::complex` then \c Real * is a typedef to \a U. - * \li A typedef \a NonInteger, giving the type that should be used for operations producing non-integral values, + * \li A typedef \c NonInteger, giving the type that should be used for operations producing non-integral values, * such as quotients, square roots, etc. If \a T is a floating-point type, then this typedef just gives - * \a T again. Note however that many Eigen functions such as internal::sqrt simply refuse to + * \a T again. Note however that many Eigen functions such as `internal::sqrt` simply refuse to * take integers. Outside of a few cases, Eigen doesn't do automatic type promotion. Thus, this typedef is * only intended as a helper for code that needs to explicitly promote types. - * \li A typedef \a Nested giving the type to use to nest a value inside of the expression tree. If you don't know what + * \li A typedef \c Literal giving the type to use for numeric literals such as "2" or "0.5". For instance, for `std::complex`, + * Literal is defined as \c U. + * Of course, this type must be fully compatible with \a T. In doubt, just use \a T here. + * \li A typedef \c Nested giving the type to use to nest a value inside of the expression tree. If you don't know what * this means, just use \a T here. - * \li An enum value \a IsComplex. It is equal to 1 if \a T is a \c std::complex + * \li An enum value \c IsComplex. It is equal to 1 if \a T is a \c std::complex * type, and to 0 otherwise. - * \li An enum value \a IsInteger. It is equal to \c 1 if \a T is an integer type such as \c int, + * \li An enum value \c IsInteger. It is equal to \c 1 if \a T is an integer type such as \c int, * and to \c 0 otherwise. - * \li Enum values ReadCost, AddCost and MulCost representing a rough estimate of the number of CPU cycles needed + * \li Enum values \c ReadCost, \c AddCost and \c MulCost representing a rough estimate of the number of CPU cycles needed * to by move / add / mul instructions respectively, assuming the data is already stored in CPU registers. - * Stay vague here. No need to do architecture-specific stuff. - * \li An enum value \a IsSigned. It is equal to \c 1 if \a T is a signed type and to 0 if \a T is unsigned. - * \li An enum value \a RequireInitialization. It is equal to \c 1 if the constructor of the numeric type \a T must + * Stay vague here. No need to do architecture-specific stuff. If you don't know what this means, just use \c Eigen::HugeCost. + * \li An enum value \c IsSigned. It is equal to \c 1 if \a T is a signed type and to 0 if \a T is unsigned. + * \li An enum value \c RequireInitialization. It is equal to \c 1 if the constructor of the numeric type \a T must * be called, and to 0 if it is safe not to call it. Default is 0 if \a T is an arithmetic type, and 1 otherwise. - * \li An epsilon() function which, unlike std::numeric_limits::epsilon(), returns a \a Real instead of a \a T. - * \li A dummy_precision() function returning a weak epsilon value. It is mainly used as a default + * \li An `epsilon()` function which, unlike `std::numeric_limits::epsilon()`, + * it returns a \c Real instead of a \a T. + * \li A `dummy_precision()` function returning a weak epsilon value. It is mainly used as a default * value by the fuzzy comparison operators. - * \li highest() and lowest() functions returning the highest and lowest possible values respectively. + * \li `highest()` and `lowest()` functions returning the highest and lowest possible values respectively. + * \li `digits()` function returning the number of radix digits (non-sign digits for integers, mantissa for floating-point). This is + * the analogue of std::numeric_limits::digits + * which is used as the default implementation if specialized. + * \li `digits10()` function returning the number of decimal digits that can be represented without change. This is + * the analogue of std::numeric_limits::digits10 + * which is used as the default implementation if specialized. + * \li `min_exponent()` and `max_exponent()` functions returning the highest and lowest possible values, respectively, + * such that the radix raised to the power exponent-1 is a normalized floating-point number. These are equivalent to + * `std::numeric_limits::min_exponent`/ + * `std::numeric_limits::max_exponent`. + * \li `infinity()` function returning a representation of positive infinity, if available. + * \li `quiet_NaN` function returning a non-signaling "not-a-number", if available. */ + // clang-format on template struct GenericNumTraits { @@ -67,22 +171,65 @@ template struct GenericNumTraits T >::type NonInteger; typedef T Nested; + typedef T Literal; + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline Real epsilon() + { + return numext::numeric_limits::epsilon(); + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline int digits10() + { + return internal::default_digits10_impl::run(); + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline int digits() + { + return internal::default_digits_impl::run(); + } - static inline Real epsilon() { return std::numeric_limits::epsilon(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline int min_exponent() + { + return numext::numeric_limits::min_exponent; + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline int max_exponent() + { + return numext::numeric_limits::max_exponent; + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline Real dummy_precision() { // make sure to override this for floating-point types return Real(0); } - static inline T highest() { return (std::numeric_limits::max)(); } - static inline T lowest() { return IsInteger ? (std::numeric_limits::min)() : (-(std::numeric_limits::max)()); } - -#ifdef EIGEN2_SUPPORT - enum { - HasFloatingPoint = !IsInteger - }; - typedef NonInteger FloatingPoint; -#endif + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline T highest() { + return (numext::numeric_limits::max)(); + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline T lowest() { + return IsInteger ? (numext::numeric_limits::min)() + : static_cast(-(numext::numeric_limits::max)()); + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline T infinity() { + return numext::numeric_limits::infinity(); + } + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline T quiet_NaN() { + return numext::numeric_limits::quiet_NaN(); + } }; template struct NumTraits : GenericNumTraits @@ -91,24 +238,41 @@ template struct NumTraits : GenericNumTraits template<> struct NumTraits : GenericNumTraits { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline float dummy_precision() { return 1e-5f; } }; template<> struct NumTraits : GenericNumTraits { + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline double dummy_precision() { return 1e-12; } }; +// GPU devices treat `long double` as `double`. +#ifndef EIGEN_GPU_COMPILE_PHASE template<> struct NumTraits : GenericNumTraits { - static inline long double dummy_precision() { return 1e-15l; } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline long double dummy_precision() { return static_cast(1e-15l); } + +#if defined(EIGEN_ARCH_PPC) && (__LDBL_MANT_DIG__ == 106) + // PowerPC double double causes issues with some values + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline long double epsilon() + { + // 2^(-(__LDBL_MANT_DIG__)+1) + return static_cast(2.4651903288156618919116517665087e-32l); + } +#endif }; +#endif template struct NumTraits > : GenericNumTraits > { typedef _Real Real; + typedef typename NumTraits<_Real>::Literal Literal; enum { IsComplex = 1, RequireInitialization = NumTraits<_Real>::RequireInitialization, @@ -117,8 +281,12 @@ template struct NumTraits > MulCost = 4 * NumTraits::MulCost + 2 * NumTraits::AddCost }; + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline Real epsilon() { return NumTraits::epsilon(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline Real dummy_precision() { return NumTraits::dummy_precision(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + static inline int digits10() { return NumTraits::digits10(); } }; template @@ -130,21 +298,54 @@ struct NumTraits > typedef typename NumTraits::NonInteger NonIntegerScalar; typedef Array NonInteger; typedef ArrayType & Nested; - + typedef typename NumTraits::Literal Literal; + enum { IsComplex = NumTraits::IsComplex, IsInteger = NumTraits::IsInteger, IsSigned = NumTraits::IsSigned, RequireInitialization = 1, - ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits::ReadCost, - AddCost = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits::AddCost, - MulCost = ArrayType::SizeAtCompileTime==Dynamic ? Dynamic : ArrayType::SizeAtCompileTime * NumTraits::MulCost + ReadCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits::ReadCost), + AddCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits::AddCost), + MulCost = ArrayType::SizeAtCompileTime==Dynamic ? HugeCost : ArrayType::SizeAtCompileTime * int(NumTraits::MulCost) }; - + + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline RealScalar epsilon() { return NumTraits::epsilon(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR static inline RealScalar dummy_precision() { return NumTraits::dummy_precision(); } + + EIGEN_CONSTEXPR + static inline int digits10() { return NumTraits::digits10(); } }; +template<> struct NumTraits + : GenericNumTraits +{ + enum { + RequireInitialization = 1, + ReadCost = HugeCost, + AddCost = HugeCost, + MulCost = HugeCost + }; + + EIGEN_CONSTEXPR + static inline int digits10() { return 0; } + +private: + static inline std::string epsilon(); + static inline std::string dummy_precision(); + static inline std::string lowest(); + static inline std::string highest(); + static inline std::string infinity(); + static inline std::string quiet_NaN(); +}; + +// Empty specialization for void to allow template specialization based on NumTraits::Real with T==void and SFINAE. +template<> struct NumTraits {}; + +template<> struct NumTraits : GenericNumTraits {}; + } // end namespace Eigen #endif // EIGEN_NUMTRAITS_H diff --git a/thirdparty/eigen/Eigen/src/Core/PartialReduxEvaluator.h b/thirdparty/eigen/Eigen/src/Core/PartialReduxEvaluator.h new file mode 100644 index 00000000..17c06f07 --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/PartialReduxEvaluator.h @@ -0,0 +1,237 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2011-2018 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_PARTIALREDUX_H +#define EIGEN_PARTIALREDUX_H + +namespace Eigen { + +namespace internal { + + +/*************************************************************************** +* +* This file provides evaluators for partial reductions. +* There are two modes: +* +* - scalar path: simply calls the respective function on the column or row. +* -> nothing special here, all the tricky part is handled by the return +* types of VectorwiseOp's members. They embed the functor calling the +* respective DenseBase's member function. +* +* - vectorized path: implements a packet-wise reductions followed by +* some (optional) processing of the outcome, e.g., division by n for mean. +* +* For the vectorized path let's observe that the packet-size and outer-unrolling +* are both decided by the assignement logic. So all we have to do is to decide +* on the inner unrolling. +* +* For the unrolling, we can reuse "internal::redux_vec_unroller" from Redux.h, +* but be need to be careful to specify correct increment. +* +***************************************************************************/ + + +/* logic deciding a strategy for unrolling of vectorized paths */ +template +struct packetwise_redux_traits +{ + enum { + OuterSize = int(Evaluator::IsRowMajor) ? Evaluator::RowsAtCompileTime : Evaluator::ColsAtCompileTime, + Cost = OuterSize == Dynamic ? HugeCost + : OuterSize * Evaluator::CoeffReadCost + (OuterSize-1) * functor_traits::Cost, + Unrolling = Cost <= EIGEN_UNROLLING_LIMIT ? CompleteUnrolling : NoUnrolling + }; + +}; + +/* Value to be returned when size==0 , by default let's return 0 */ +template +EIGEN_DEVICE_FUNC +PacketType packetwise_redux_empty_value(const Func& ) { + const typename unpacket_traits::type zero(0); + return pset1(zero); +} + +/* For products the default is 1 */ +template +EIGEN_DEVICE_FUNC +PacketType packetwise_redux_empty_value(const scalar_product_op& ) { + return pset1(Scalar(1)); +} + +/* Perform the actual reduction */ +template::Unrolling +> +struct packetwise_redux_impl; + +/* Perform the actual reduction with unrolling */ +template +struct packetwise_redux_impl +{ + typedef redux_novec_unroller Base; + typedef typename Evaluator::Scalar Scalar; + + template + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE + PacketType run(const Evaluator &eval, const Func& func, Index /*size*/) + { + return redux_vec_unroller::OuterSize>::template run(eval,func); + } +}; + +/* Add a specialization of redux_vec_unroller for size==0 at compiletime. + * This specialization is not required for general reductions, which is + * why it is defined here. + */ +template +struct redux_vec_unroller +{ + template + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE PacketType run(const Evaluator &, const Func& f) + { + return packetwise_redux_empty_value(f); + } +}; + +/* Perform the actual reduction for dynamic sizes */ +template +struct packetwise_redux_impl +{ + typedef typename Evaluator::Scalar Scalar; + typedef typename redux_traits::PacketType PacketScalar; + + template + EIGEN_DEVICE_FUNC + static PacketType run(const Evaluator &eval, const Func& func, Index size) + { + if(size==0) + return packetwise_redux_empty_value(func); + + const Index size4 = (size-1)&(~3); + PacketType p = eval.template packetByOuterInner(0,0); + Index i = 1; + // This loop is optimized for instruction pipelining: + // - each iteration generates two independent instructions + // - thanks to branch prediction and out-of-order execution we have independent instructions across loops + for(; i(i+0,0),eval.template packetByOuterInner(i+1,0)), + func.packetOp(eval.template packetByOuterInner(i+2,0),eval.template packetByOuterInner(i+3,0)))); + for(; i(i,0)); + return p; + } +}; + +template< typename ArgType, typename MemberOp, int Direction> +struct evaluator > + : evaluator_base > +{ + typedef PartialReduxExpr XprType; + typedef typename internal::nested_eval::type ArgTypeNested; + typedef typename internal::add_const_on_value_type::type ConstArgTypeNested; + typedef typename internal::remove_all::type ArgTypeNestedCleaned; + typedef typename ArgType::Scalar InputScalar; + typedef typename XprType::Scalar Scalar; + enum { + TraversalSize = Direction==int(Vertical) ? int(ArgType::RowsAtCompileTime) : int(ArgType::ColsAtCompileTime) + }; + typedef typename MemberOp::template Cost CostOpType; + enum { + CoeffReadCost = TraversalSize==Dynamic ? HugeCost + : TraversalSize==0 ? 1 + : int(TraversalSize) * int(evaluator::CoeffReadCost) + int(CostOpType::value), + + _ArgFlags = evaluator::Flags, + + _Vectorizable = bool(int(_ArgFlags)&PacketAccessBit) + && bool(MemberOp::Vectorizable) + && (Direction==int(Vertical) ? bool(_ArgFlags&RowMajorBit) : (_ArgFlags&RowMajorBit)==0) + && (TraversalSize!=0), + + Flags = (traits::Flags&RowMajorBit) + | (evaluator::Flags&(HereditaryBits&(~RowMajorBit))) + | (_Vectorizable ? PacketAccessBit : 0) + | LinearAccessBit, + + Alignment = 0 // FIXME this will need to be improved once PartialReduxExpr is vectorized + }; + + EIGEN_DEVICE_FUNC explicit evaluator(const XprType xpr) + : m_arg(xpr.nestedExpression()), m_functor(xpr.functor()) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(TraversalSize==Dynamic ? HugeCost : (TraversalSize==0 ? 1 : int(CostOpType::value))); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const Scalar coeff(Index i, Index j) const + { + return coeff(Direction==Vertical ? j : i); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const Scalar coeff(Index index) const + { + return m_functor(m_arg.template subVector(index)); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + PacketType packet(Index i, Index j) const + { + return packet(Direction==Vertical ? j : i); + } + + template + EIGEN_STRONG_INLINE EIGEN_DEVICE_FUNC + PacketType packet(Index idx) const + { + enum { PacketSize = internal::unpacket_traits::size }; + typedef Block PanelType; + + PanelType panel(m_arg, + Direction==Vertical ? 0 : idx, + Direction==Vertical ? idx : 0, + Direction==Vertical ? m_arg.rows() : Index(PacketSize), + Direction==Vertical ? Index(PacketSize) : m_arg.cols()); + + // FIXME + // See bug 1612, currently if PacketSize==1 (i.e. complex with 128bits registers) then the storage-order of panel get reversed + // and methods like packetByOuterInner do not make sense anymore in this context. + // So let's just by pass "vectorization" in this case: + if(PacketSize==1) + return internal::pset1(coeff(idx)); + + typedef typename internal::redux_evaluator PanelEvaluator; + PanelEvaluator panel_eval(panel); + typedef typename MemberOp::BinaryOp BinaryOp; + PacketType p = internal::packetwise_redux_impl::template run(panel_eval,m_functor.binaryFunc(),m_arg.outerSize()); + return p; + } + +protected: + ConstArgTypeNested m_arg; + const MemberOp m_functor; +}; + +} // end namespace internal + +} // end namespace Eigen + +#endif // EIGEN_PARTIALREDUX_H diff --git a/thirdparty/eigen/Eigen/src/Core/PermutationMatrix.h b/thirdparty/eigen/Eigen/src/Core/PermutationMatrix.h index bda79fa0..69401bf4 100644 --- a/thirdparty/eigen/Eigen/src/Core/PermutationMatrix.h +++ b/thirdparty/eigen/Eigen/src/Core/PermutationMatrix.h @@ -2,7 +2,7 @@ // for linear algebra. // // Copyright (C) 2009 Benoit Jacob -// Copyright (C) 2009-2011 Gael Guennebaud +// Copyright (C) 2009-2015 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -13,14 +13,18 @@ namespace Eigen { -template class PermutedImpl; +namespace internal { + +enum PermPermProduct_t {PermPermProduct}; + +} // end namespace internal /** \class PermutationBase * \ingroup Core_Module * * \brief Base class for permutations * - * \param Derived the derived class + * \tparam Derived the derived class * * This class is the base class for all expressions representing a permutation matrix, * internally stored as a vector of integers. @@ -38,17 +42,6 @@ template -struct permut_matrix_product_retval; -template -struct permut_sparsematrix_product_retval; -enum PermPermProduct_t {PermPermProduct}; - -} // end namespace internal - template class PermutationBase : public EigenBase { @@ -60,19 +53,20 @@ class PermutationBase : public EigenBase typedef typename Traits::IndicesType IndicesType; enum { Flags = Traits::Flags, - CoeffReadCost = Traits::CoeffReadCost, RowsAtCompileTime = Traits::RowsAtCompileTime, ColsAtCompileTime = Traits::ColsAtCompileTime, MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime, MaxColsAtCompileTime = Traits::MaxColsAtCompileTime }; - typedef typename Traits::Scalar Scalar; - typedef typename Traits::Index Index; - typedef Matrix + typedef typename Traits::StorageIndex StorageIndex; + typedef Matrix DenseMatrixType; - typedef PermutationMatrix + typedef PermutationMatrix PlainPermutationType; + typedef PlainPermutationType PlainObject; using Base::derived; + typedef Inverse InverseReturnType; + typedef void Scalar; #endif /** Copies the other permutation into *this */ @@ -93,32 +87,21 @@ class PermutationBase : public EigenBase return derived(); } - #ifndef EIGEN_PARSED_BY_DOXYGEN - /** This is a special case of the templated operator=. Its purpose is to - * prevent a default operator= from hiding the templated operator=. - */ - Derived& operator=(const PermutationBase& other) - { - indices() = other.indices(); - return derived(); - } - #endif - /** \returns the number of rows */ - inline Index rows() const { return Index(indices().size()); } + inline EIGEN_DEVICE_FUNC Index rows() const { return Index(indices().size()); } /** \returns the number of columns */ - inline Index cols() const { return Index(indices().size()); } + inline EIGEN_DEVICE_FUNC Index cols() const { return Index(indices().size()); } /** \returns the size of a side of the respective square matrix, i.e., the number of indices */ - inline Index size() const { return Index(indices().size()); } + inline EIGEN_DEVICE_FUNC Index size() const { return Index(indices().size()); } #ifndef EIGEN_PARSED_BY_DOXYGEN template void evalTo(MatrixBase& other) const { other.setZero(); - for (int i=0; i /** Sets *this to be the identity permutation matrix */ void setIdentity() { - for(Index i = 0; i < size(); ++i) + StorageIndex n = StorageIndex(size()); + for(StorageIndex i = 0; i < n; ++i) indices().coeffRef(i) = i; } @@ -163,18 +147,18 @@ class PermutationBase : public EigenBase * * \returns a reference to *this. * - * \warning This is much slower than applyTranspositionOnTheRight(int,int): + * \warning This is much slower than applyTranspositionOnTheRight(Index,Index): * this has linear complexity and requires a lot of branching. * - * \sa applyTranspositionOnTheRight(int,int) + * \sa applyTranspositionOnTheRight(Index,Index) */ Derived& applyTranspositionOnTheLeft(Index i, Index j) { eigen_assert(i>=0 && j>=0 && i * * This is a fast operation, it only consists in swapping two indices. * - * \sa applyTranspositionOnTheLeft(int,int) + * \sa applyTranspositionOnTheLeft(Index,Index) */ Derived& applyTranspositionOnTheRight(Index i, Index j) { @@ -196,16 +180,16 @@ class PermutationBase : public EigenBase /** \returns the inverse permutation matrix. * - * \note \note_try_to_help_rvo + * \note \blank \note_try_to_help_rvo */ - inline Transpose inverse() const - { return derived(); } + inline InverseReturnType inverse() const + { return InverseReturnType(derived()); } /** \returns the tranpose permutation matrix. * - * \note \note_try_to_help_rvo + * \note \blank \note_try_to_help_rvo */ - inline Transpose transpose() const - { return derived(); } + inline InverseReturnType transpose() const + { return InverseReturnType(derived()); } /**** multiplication helpers to hopefully get RVO ****/ @@ -215,13 +199,13 @@ class PermutationBase : public EigenBase template void assignTranspose(const PermutationBase& other) { - for (int i=0; i void assignProduct(const Lhs& lhs, const Rhs& rhs) { eigen_assert(lhs.cols() == rhs.rows()); - for (int i=0; i /** \returns the product permutation matrix. * - * \note \note_try_to_help_rvo + * \note \blank \note_try_to_help_rvo */ template inline PlainPermutationType operator*(const PermutationBase& other) const @@ -237,18 +221,18 @@ class PermutationBase : public EigenBase /** \returns the product of a permutation with another inverse permutation. * - * \note \note_try_to_help_rvo + * \note \blank \note_try_to_help_rvo */ template - inline PlainPermutationType operator*(const Transpose >& other) const + inline PlainPermutationType operator*(const InverseImpl& other) const { return PlainPermutationType(internal::PermPermProduct, *this, other.eval()); } /** \returns the product of an inverse permutation with another permutation. * - * \note \note_try_to_help_rvo + * \note \blank \note_try_to_help_rvo */ template friend - inline PlainPermutationType operator*(const Transpose >& other, const PermutationBase& perm) + inline PlainPermutationType operator*(const InverseImpl& other, const PermutationBase& perm) { return PlainPermutationType(internal::PermPermProduct, other.eval(), perm); } /** \returns the determinant of the permutation matrix, which is either 1 or -1 depending on the parity of the permutation. @@ -284,39 +268,43 @@ class PermutationBase : public EigenBase }; +namespace internal { +template +struct traits > + : traits > +{ + typedef PermutationStorage StorageKind; + typedef Matrix<_StorageIndex, SizeAtCompileTime, 1, 0, MaxSizeAtCompileTime, 1> IndicesType; + typedef _StorageIndex StorageIndex; + typedef void Scalar; +}; +} + /** \class PermutationMatrix * \ingroup Core_Module * * \brief Permutation matrix * - * \param SizeAtCompileTime the number of rows/cols, or Dynamic - * \param MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it. - * \param IndexType the interger type of the indices + * \tparam SizeAtCompileTime the number of rows/cols, or Dynamic + * \tparam MaxSizeAtCompileTime the maximum number of rows/cols, or Dynamic. This optional parameter defaults to SizeAtCompileTime. Most of the time, you should not have to specify it. + * \tparam _StorageIndex the integer type of the indices * * This class represents a permutation matrix, internally stored as a vector of integers. * * \sa class PermutationBase, class PermutationWrapper, class DiagonalMatrix */ - -namespace internal { -template -struct traits > - : traits > -{ - typedef IndexType Index; - typedef Matrix IndicesType; -}; -} - -template -class PermutationMatrix : public PermutationBase > +template +class PermutationMatrix : public PermutationBase > { typedef PermutationBase Base; typedef internal::traits Traits; public: + typedef const PermutationMatrix& Nested; + #ifndef EIGEN_PARSED_BY_DOXYGEN typedef typename Traits::IndicesType IndicesType; + typedef typename Traits::StorageIndex StorageIndex; #endif inline PermutationMatrix() @@ -324,20 +312,16 @@ class PermutationMatrix : public PermutationBase::highest()); + } /** Copy constructor. */ template inline PermutationMatrix(const PermutationBase& other) : m_indices(other.indices()) {} - #ifndef EIGEN_PARSED_BY_DOXYGEN - /** Standard copy constructor. Defined only to prevent a default copy constructor - * from hiding the other templated constructor */ - inline PermutationMatrix(const PermutationMatrix& other) : m_indices(other.indices()) {} - #endif - /** Generic constructor from expression of the indices. The indices * array has the meaning that the permutations sends each integer i to indices[i]. * @@ -346,7 +330,7 @@ class PermutationMatrix : public PermutationBase - explicit inline PermutationMatrix(const MatrixBase& a_indices) : m_indices(a_indices) + explicit inline PermutationMatrix(const MatrixBase& indices) : m_indices(indices) {} /** Convert the Transpositions \a tr to a permutation matrix */ @@ -372,17 +356,6 @@ class PermutationMatrix : public PermutationBase - PermutationMatrix(const Transpose >& other) - : m_indices(other.nestedPermutation().size()) + PermutationMatrix(const InverseImpl& other) + : m_indices(other.derived().nestedExpression().size()) { - for (int i=0; i::highest()); + StorageIndex end = StorageIndex(m_indices.size()); + for (StorageIndex i=0; i PermutationMatrix(internal::PermPermProduct_t, const Lhs& lhs, const Rhs& rhs) @@ -413,18 +389,20 @@ class PermutationMatrix : public PermutationBase -struct traits,_PacketAccess> > - : traits > +template +struct traits,_PacketAccess> > + : traits > { - typedef IndexType Index; - typedef Map, _PacketAccess> IndicesType; + typedef PermutationStorage StorageKind; + typedef Map, _PacketAccess> IndicesType; + typedef _StorageIndex StorageIndex; + typedef void Scalar; }; } -template -class Map,_PacketAccess> - : public PermutationBase,_PacketAccess> > +template +class Map,_PacketAccess> + : public PermutationBase,_PacketAccess> > { typedef PermutationBase Base; typedef internal::traits Traits; @@ -432,14 +410,14 @@ class Map, #ifndef EIGEN_PARSED_BY_DOXYGEN typedef typename Traits::IndicesType IndicesType; - typedef typename IndicesType::Scalar Index; + typedef typename IndicesType::Scalar StorageIndex; #endif - inline Map(const Index* indicesPtr) + inline Map(const StorageIndex* indicesPtr) : m_indices(indicesPtr) {} - inline Map(const Index* indicesPtr, Index size) + inline Map(const StorageIndex* indicesPtr, Index size) : m_indices(indicesPtr,size) {} @@ -474,40 +452,36 @@ class Map, IndicesType m_indices; }; -/** \class PermutationWrapper - * \ingroup Core_Module - * - * \brief Class to view a vector of integers as a permutation matrix - * - * \param _IndicesType the type of the vector of integer (can be any compatible expression) - * - * This class allows to view any vector expression of integers as a permutation matrix. - * - * \sa class PermutationBase, class PermutationMatrix - */ - -struct PermutationStorage {}; - template class TranspositionsWrapper; namespace internal { template struct traits > { typedef PermutationStorage StorageKind; - typedef typename _IndicesType::Scalar Scalar; - typedef typename _IndicesType::Scalar Index; + typedef void Scalar; + typedef typename _IndicesType::Scalar StorageIndex; typedef _IndicesType IndicesType; enum { RowsAtCompileTime = _IndicesType::SizeAtCompileTime, ColsAtCompileTime = _IndicesType::SizeAtCompileTime, - MaxRowsAtCompileTime = IndicesType::MaxRowsAtCompileTime, - MaxColsAtCompileTime = IndicesType::MaxColsAtCompileTime, - Flags = 0, - CoeffReadCost = _IndicesType::CoeffReadCost + MaxRowsAtCompileTime = IndicesType::MaxSizeAtCompileTime, + MaxColsAtCompileTime = IndicesType::MaxSizeAtCompileTime, + Flags = 0 }; }; } +/** \class PermutationWrapper + * \ingroup Core_Module + * + * \brief Class to view a vector of integers as a permutation matrix + * + * \tparam _IndicesType the type of the vector of integer (can be any compatible expression) + * + * This class allows to view any vector expression of integers as a permutation matrix. + * + * \sa class PermutationBase, class PermutationMatrix + */ template class PermutationWrapper : public PermutationBase > { @@ -519,8 +493,8 @@ class PermutationWrapper : public PermutationBase -inline const internal::permut_matrix_product_retval -operator*(const MatrixBase& matrix, - const PermutationBase &permutation) +template +EIGEN_DEVICE_FUNC +const Product +operator*(const MatrixBase &matrix, + const PermutationBase& permutation) { - return internal::permut_matrix_product_retval - - (permutation.derived(), matrix.derived()); + return Product + (matrix.derived(), permutation.derived()); } /** \returns the matrix with the permutation applied to the rows. */ -template -inline const internal::permut_matrix_product_retval - +template +EIGEN_DEVICE_FUNC +const Product operator*(const PermutationBase &permutation, - const MatrixBase& matrix) + const MatrixBase& matrix) { - return internal::permut_matrix_product_retval - - (permutation.derived(), matrix.derived()); + return Product + (permutation.derived(), matrix.derived()); } -namespace internal { -template -struct traits > +template +class InverseImpl + : public EigenBase > { - typedef typename MatrixType::PlainObject ReturnType; -}; - -template -struct permut_matrix_product_retval - : public ReturnByValue > -{ - typedef typename remove_all::type MatrixTypeNestedCleaned; - typedef typename MatrixType::Index Index; - - permut_matrix_product_retval(const PermutationType& perm, const MatrixType& matrix) - : m_permutation(perm), m_matrix(matrix) - {} - - inline Index rows() const { return m_matrix.rows(); } - inline Index cols() const { return m_matrix.cols(); } - - template inline void evalTo(Dest& dst) const - { - const Index n = Side==OnTheLeft ? rows() : cols(); - // FIXME we need an is_same for expression that is not sensitive to constness. For instance - // is_same_xpr, Block >::value should be true. - const typename Dest::Scalar *dst_data = internal::extract_data(dst); - if( is_same::value - && blas_traits::HasUsableDirectAccess - && blas_traits::HasUsableDirectAccess - && dst_data!=0 && dst_data == extract_data(m_matrix)) - { - // apply the permutation inplace - Matrix mask(m_permutation.size()); - mask.fill(false); - Index r = 0; - while(r < m_permutation.size()) - { - // search for the next seed - while(r=m_permutation.size()) - break; - // we got one, let's follow it until we are back to the seed - Index k0 = r++; - Index kPrev = k0; - mask.coeffRef(k0) = true; - for(Index k=m_permutation.indices().coeff(k0); k!=k0; k=m_permutation.indices().coeff(k)) - { - Block(dst, k) - .swap(Block - (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev)); - - mask.coeffRef(k) = true; - kPrev = k; - } - } - } - else - { - for(int i = 0; i < n; ++i) - { - Block - (dst, ((Side==OnTheLeft) ^ Transposed) ? m_permutation.indices().coeff(i) : i) - - = - - Block - (m_matrix, ((Side==OnTheRight) ^ Transposed) ? m_permutation.indices().coeff(i) : i); - } - } - } - - protected: - const PermutationType& m_permutation; - typename MatrixType::Nested m_matrix; -}; - -/* Template partial specialization for transposed/inverse permutations */ - -template -struct traits > > - : traits -{}; - -} // end namespace internal - -template -class Transpose > - : public EigenBase > > -{ - typedef Derived PermutationType; - typedef typename PermutationType::IndicesType IndicesType; typedef typename PermutationType::PlainPermutationType PlainPermutationType; + typedef internal::traits PermTraits; + protected: + InverseImpl() {} public: + typedef Inverse InverseType; + using EigenBase >::derived; #ifndef EIGEN_PARSED_BY_DOXYGEN - typedef internal::traits Traits; - typedef typename Derived::DenseMatrixType DenseMatrixType; + typedef typename PermutationType::DenseMatrixType DenseMatrixType; enum { - Flags = Traits::Flags, - CoeffReadCost = Traits::CoeffReadCost, - RowsAtCompileTime = Traits::RowsAtCompileTime, - ColsAtCompileTime = Traits::ColsAtCompileTime, - MaxRowsAtCompileTime = Traits::MaxRowsAtCompileTime, - MaxColsAtCompileTime = Traits::MaxColsAtCompileTime + RowsAtCompileTime = PermTraits::RowsAtCompileTime, + ColsAtCompileTime = PermTraits::ColsAtCompileTime, + MaxRowsAtCompileTime = PermTraits::MaxRowsAtCompileTime, + MaxColsAtCompileTime = PermTraits::MaxColsAtCompileTime }; - typedef typename Traits::Scalar Scalar; #endif - Transpose(const PermutationType& p) : m_permutation(p) {} - - inline int rows() const { return m_permutation.rows(); } - inline int cols() const { return m_permutation.cols(); } - #ifndef EIGEN_PARSED_BY_DOXYGEN template void evalTo(MatrixBase& other) const { other.setZero(); - for (int i=0; i friend - inline const internal::permut_matrix_product_retval - operator*(const MatrixBase& matrix, const Transpose& trPerm) + const Product + operator*(const MatrixBase& matrix, const InverseType& trPerm) { - return internal::permut_matrix_product_retval(trPerm.m_permutation, matrix.derived()); + return Product(matrix.derived(), trPerm.derived()); } /** \returns the matrix with the inverse permutation applied to the rows. */ template - inline const internal::permut_matrix_product_retval + const Product operator*(const MatrixBase& matrix) const { - return internal::permut_matrix_product_retval(m_permutation, matrix.derived()); + return Product(derived(), matrix.derived()); } - - const PermutationType& nestedPermutation() const { return m_permutation; } - - protected: - const PermutationType& m_permutation; }; template @@ -717,6 +594,12 @@ const PermutationWrapper MatrixBase::asPermutation() con return derived(); } +namespace internal { + +template<> struct AssignmentKind { typedef EigenBase2EigenBase Kind; }; + +} // end namespace internal + } // end namespace Eigen #endif // EIGEN_PERMUTATIONMATRIX_H diff --git a/thirdparty/eigen/Eigen/src/Core/PlainObjectBase.h b/thirdparty/eigen/Eigen/src/Core/PlainObjectBase.h index 9f71956f..e2ddbd1d 100644 --- a/thirdparty/eigen/Eigen/src/Core/PlainObjectBase.h +++ b/thirdparty/eigen/Eigen/src/Core/PlainObjectBase.h @@ -13,10 +13,10 @@ #if defined(EIGEN_INITIALIZE_MATRICES_BY_ZERO) # define EIGEN_INITIALIZE_COEFFS -# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(int i=0;i::quiet_NaN(); +# define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED for(Index i=0;i::quiet_NaN(); #else # undef EIGEN_INITIALIZE_COEFFS # define EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED @@ -28,6 +28,7 @@ namespace internal { template struct check_rows_cols_for_overflow { template + EIGEN_DEVICE_FUNC static EIGEN_ALWAYS_INLINE void run(Index, Index) { } @@ -35,11 +36,12 @@ template struct check_rows_cols_for_overflow { template<> struct check_rows_cols_for_overflow { template + EIGEN_DEVICE_FUNC static EIGEN_ALWAYS_INLINE void run(Index rows, Index cols) { // http://hg.mozilla.org/mozilla-central/file/6c8a909977d3/xpcom/ds/CheckedInt.h#l242 // we assume Index is signed - Index max_index = (size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed + Index max_index = (std::size_t(1) << (8 * sizeof(Index) - 1)) - 1; // assume Index is signed bool error = (rows == 0 || cols == 0) ? false : (rows > max_index / cols); if (error) @@ -56,33 +58,41 @@ template struct m } // end namespace internal -/** \class PlainObjectBase - * \brief %Dense storage base class for matrices and arrays. - * - * This class can be extended with the help of the plugin mechanism described on the page - * \ref TopicCustomizingEigen by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN. - * - * \sa \ref TopicClassHierarchy - */ #ifdef EIGEN_PARSED_BY_DOXYGEN -namespace internal { +namespace doxygen { -// this is a warkaround to doxygen not being able to understand the inheritence logic +// This is a workaround to doxygen not being able to understand the inheritance logic // when it is hidden by the dense_xpr_base helper struct. -template struct dense_xpr_base_dispatcher_for_doxygen;// : public MatrixBase {}; +// Moreover, doxygen fails to include members that are not documented in the declaration body of +// MatrixBase if we inherits MatrixBase >, +// this is why we simply inherits MatrixBase, though this does not make sense. + +/** This class is just a workaround for Doxygen and it does not not actually exist. */ +template struct dense_xpr_base_dispatcher; /** This class is just a workaround for Doxygen and it does not not actually exist. */ template -struct dense_xpr_base_dispatcher_for_doxygen > - : public MatrixBase > {}; +struct dense_xpr_base_dispatcher > + : public MatrixBase {}; /** This class is just a workaround for Doxygen and it does not not actually exist. */ template -struct dense_xpr_base_dispatcher_for_doxygen > - : public ArrayBase > {}; +struct dense_xpr_base_dispatcher > + : public ArrayBase {}; -} // namespace internal +} // namespace doxygen +/** \class PlainObjectBase + * \ingroup Core_Module + * \brief %Dense storage base class for matrices and arrays. + * + * This class can be extended with the help of the plugin mechanism described on the page + * \ref TopicCustomizing_Plugins by defining the preprocessor symbol \c EIGEN_PLAINOBJECTBASE_PLUGIN. + * + * \tparam Derived is the derived type, e.g., a Matrix or Array + * + * \sa \ref TopicClassHierarchy + */ template -class PlainObjectBase : public internal::dense_xpr_base_dispatcher_for_doxygen +class PlainObjectBase : public doxygen::dense_xpr_base_dispatcher #else template class PlainObjectBase : public internal::dense_xpr_base::type @@ -93,8 +103,8 @@ class PlainObjectBase : public internal::dense_xpr_base::type typedef typename internal::dense_xpr_base::type Base; typedef typename internal::traits::StorageKind StorageKind; - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; + typedef typename internal::packet_traits::type PacketScalar; typedef typename NumTraits::Real RealScalar; typedef Derived DenseType; @@ -108,33 +118,37 @@ class PlainObjectBase : public internal::dense_xpr_base::type using Base::IsVectorAtCompileTime; using Base::Flags; - template friend class Eigen::Map; - friend class Eigen::Map; typedef Eigen::Map MapType; - friend class Eigen::Map; typedef const Eigen::Map ConstMapType; - friend class Eigen::Map; - typedef Eigen::Map AlignedMapType; - friend class Eigen::Map; - typedef const Eigen::Map ConstAlignedMapType; + typedef Eigen::Map AlignedMapType; + typedef const Eigen::Map ConstAlignedMapType; template struct StridedMapType { typedef Eigen::Map type; }; template struct StridedConstMapType { typedef Eigen::Map type; }; - template struct StridedAlignedMapType { typedef Eigen::Map type; }; - template struct StridedConstAlignedMapType { typedef Eigen::Map type; }; + template struct StridedAlignedMapType { typedef Eigen::Map type; }; + template struct StridedConstAlignedMapType { typedef Eigen::Map type; }; protected: DenseStorage m_storage; public: - enum { NeedsToAlign = SizeAtCompileTime != Dynamic && (internal::traits::Flags & AlignedBit) != 0 }; + enum { NeedsToAlign = (SizeAtCompileTime != Dynamic) && (internal::traits::Alignment>0) }; EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(NeedsToAlign) + EIGEN_DEVICE_FUNC Base& base() { return *static_cast(this); } + EIGEN_DEVICE_FUNC const Base& base() const { return *static_cast(this); } - EIGEN_STRONG_INLINE Index rows() const { return m_storage.rows(); } - EIGEN_STRONG_INLINE Index cols() const { return m_storage.cols(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rows() const EIGEN_NOEXCEPT { return m_storage.rows(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index cols() const EIGEN_NOEXCEPT { return m_storage.cols(); } + /** This is an overloaded version of DenseCoeffsBase::coeff(Index,Index) const + * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts. + * + * See DenseCoeffsBase::coeff(Index) const for details. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar& coeff(Index rowId, Index colId) const { if(Flags & RowMajorBit) @@ -143,11 +157,21 @@ class PlainObjectBase : public internal::dense_xpr_base::type return m_storage.data()[rowId + colId * m_storage.rows()]; } + /** This is an overloaded version of DenseCoeffsBase::coeff(Index) const + * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts. + * + * See DenseCoeffsBase::coeff(Index) const for details. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar& coeff(Index index) const { return m_storage.data()[index]; } + /** This is an overloaded version of DenseCoeffsBase::coeffRef(Index,Index) const + * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts. + * + * See DenseCoeffsBase::coeffRef(Index,Index) const for details. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar& coeffRef(Index rowId, Index colId) { if(Flags & RowMajorBit) @@ -156,11 +180,19 @@ class PlainObjectBase : public internal::dense_xpr_base::type return m_storage.data()[rowId + colId * m_storage.rows()]; } + /** This is an overloaded version of DenseCoeffsBase::coeffRef(Index) const + * provided to by-pass the creation of an evaluator of the expression, thus saving compilation efforts. + * + * See DenseCoeffsBase::coeffRef(Index) const for details. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar& coeffRef(Index index) { return m_storage.data()[index]; } + /** This is the const version of coeffRef(Index,Index) which is thus synonym of coeff(Index,Index). + * It is provided for convenience. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar& coeffRef(Index rowId, Index colId) const { if(Flags & RowMajorBit) @@ -169,6 +201,9 @@ class PlainObjectBase : public internal::dense_xpr_base::type return m_storage.data()[rowId + colId * m_storage.rows()]; } + /** This is the const version of coeffRef(Index) which is thus synonym of coeff(Index). + * It is provided for convenience. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar& coeffRef(Index index) const { return m_storage.data()[index]; @@ -209,11 +244,11 @@ class PlainObjectBase : public internal::dense_xpr_base::type } /** \returns a const pointer to the data array of this matrix */ - EIGEN_STRONG_INLINE const Scalar *data() const + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar *data() const { return m_storage.data(); } /** \returns a pointer to the data array of this matrix */ - EIGEN_STRONG_INLINE Scalar *data() + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar *data() { return m_storage.data(); } /** Resizes \c *this to a \a rows x \a cols matrix. @@ -232,22 +267,22 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * \sa resize(Index) for vectors, resize(NoChange_t, Index), resize(Index, NoChange_t) */ - EIGEN_STRONG_INLINE void resize(Index nbRows, Index nbCols) - { - eigen_assert( EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,nbRows==RowsAtCompileTime) - && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,nbCols==ColsAtCompileTime) - && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,nbRows<=MaxRowsAtCompileTime) - && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,nbCols<=MaxColsAtCompileTime) - && nbRows>=0 && nbCols>=0 && "Invalid sizes when resizing a matrix or array."); - internal::check_rows_cols_for_overflow::run(nbRows, nbCols); + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void resize(Index rows, Index cols) + { + eigen_assert( EIGEN_IMPLIES(RowsAtCompileTime!=Dynamic,rows==RowsAtCompileTime) + && EIGEN_IMPLIES(ColsAtCompileTime!=Dynamic,cols==ColsAtCompileTime) + && EIGEN_IMPLIES(RowsAtCompileTime==Dynamic && MaxRowsAtCompileTime!=Dynamic,rows<=MaxRowsAtCompileTime) + && EIGEN_IMPLIES(ColsAtCompileTime==Dynamic && MaxColsAtCompileTime!=Dynamic,cols<=MaxColsAtCompileTime) + && rows>=0 && cols>=0 && "Invalid sizes when resizing a matrix or array."); + internal::check_rows_cols_for_overflow::run(rows, cols); #ifdef EIGEN_INITIALIZE_COEFFS - Index size = nbRows*nbCols; + Index size = rows*cols; bool size_changed = size != this->size(); - m_storage.resize(size, nbRows, nbCols); + m_storage.resize(size, rows, cols); if(size_changed) EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED #else - internal::check_rows_cols_for_overflow::run(nbRows, nbCols); - m_storage.resize(nbRows*nbCols, nbRows, nbCols); + m_storage.resize(rows*cols, rows, cols); #endif } @@ -262,6 +297,7 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * \sa resize(Index,Index), resize(NoChange_t, Index), resize(Index, NoChange_t) */ + EIGEN_DEVICE_FUNC inline void resize(Index size) { EIGEN_STATIC_ASSERT_VECTOR_ONLY(PlainObjectBase) @@ -286,9 +322,10 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * \sa resize(Index,Index) */ - inline void resize(NoChange_t, Index nbCols) + EIGEN_DEVICE_FUNC + inline void resize(NoChange_t, Index cols) { - resize(rows(), nbCols); + resize(rows(), cols); } /** Resizes the matrix, changing only the number of rows. For the parameter of type NoChange_t, just pass the special value \c NoChange @@ -299,9 +336,10 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * \sa resize(Index,Index) */ - inline void resize(Index nbRows, NoChange_t) + EIGEN_DEVICE_FUNC + inline void resize(Index rows, NoChange_t) { - resize(nbRows, cols()); + resize(rows, cols()); } /** Resizes \c *this to have the same dimensions as \a other. @@ -312,11 +350,12 @@ class PlainObjectBase : public internal::dense_xpr_base::type * remain row-vectors and vectors remain vectors. */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void resizeLike(const EigenBase& _other) { const OtherDerived& other = _other.derived(); - internal::check_rows_cols_for_overflow::run(Index(other.rows()), Index(other.cols())); - const Index othersize = Index(other.rows())*Index(other.cols()); + internal::check_rows_cols_for_overflow::run(other.rows(), other.cols()); + const Index othersize = other.rows()*other.cols(); if(RowsAtCompileTime == 1) { eigen_assert(other.rows() == 1 || other.cols() == 1); @@ -336,12 +375,13 @@ class PlainObjectBase : public internal::dense_xpr_base::type * of rows and/or of columns, you can use conservativeResize(NoChange_t, Index) or * conservativeResize(Index, NoChange_t). * - * Matrices are resized relative to the top-left element. In case values need to be + * Matrices are resized relative to the top-left element. In case values need to be * appended to the matrix they will be uninitialized. */ - EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, Index nbCols) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void conservativeResize(Index rows, Index cols) { - internal::conservative_resize_like_impl::run(*this, nbRows, nbCols); + internal::conservative_resize_like_impl::run(*this, rows, cols); } /** Resizes the matrix to \a rows x \a cols while leaving old values untouched. @@ -351,10 +391,11 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * In case the matrix is growing, new rows will be uninitialized. */ - EIGEN_STRONG_INLINE void conservativeResize(Index nbRows, NoChange_t) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void conservativeResize(Index rows, NoChange_t) { // Note: see the comment in conservativeResize(Index,Index) - conservativeResize(nbRows, cols()); + conservativeResize(rows, cols()); } /** Resizes the matrix to \a rows x \a cols while leaving old values untouched. @@ -364,10 +405,11 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * In case the matrix is growing, new columns will be uninitialized. */ - EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index nbCols) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void conservativeResize(NoChange_t, Index cols) { // Note: see the comment in conservativeResize(Index,Index) - conservativeResize(rows(), nbCols); + conservativeResize(rows(), cols); } /** Resizes the vector to \a size while retaining old values. @@ -378,6 +420,7 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * When values are appended, they will be uninitialized. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void conservativeResize(Index size) { internal::conservative_resize_like_impl::run(*this, size); @@ -389,10 +432,11 @@ class PlainObjectBase : public internal::dense_xpr_base::type * of rows and/or of columns, you can use conservativeResize(NoChange_t, Index) or * conservativeResize(Index, NoChange_t). * - * Matrices are resized relative to the top-left element. In case values need to be + * Matrices are resized relative to the top-left element. In case values need to be * appended to the matrix they will copied from \c other. */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void conservativeResizeLike(const DenseBase& other) { internal::conservative_resize_like_impl::run(*this, other); @@ -401,6 +445,7 @@ class PlainObjectBase : public internal::dense_xpr_base::type /** This is a special case of the templated operator=. Its purpose is to * prevent a default operator= from hiding the templated operator=. */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const PlainObjectBase& other) { return _set(other); @@ -408,6 +453,7 @@ class PlainObjectBase : public internal::dense_xpr_base::type /** \sa MatrixBase::lazyAssign() */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& lazyAssign(const DenseBase& other) { _resize_to_match(other); @@ -415,12 +461,18 @@ class PlainObjectBase : public internal::dense_xpr_base::type } template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const ReturnByValue& func) { resize(func.rows(), func.cols()); return Base::operator=(func); } + // Prevent user from trying to instantiate PlainObjectBase objects + // by making all its constructor protected. See bug 1074. + protected: + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PlainObjectBase() : m_storage() { // _check_template_params(); @@ -430,53 +482,146 @@ class PlainObjectBase : public internal::dense_xpr_base::type #ifndef EIGEN_PARSED_BY_DOXYGEN // FIXME is it still needed ? /** \internal */ - PlainObjectBase(internal::constructor_without_unaligned_array_assert) + EIGEN_DEVICE_FUNC + explicit PlainObjectBase(internal::constructor_without_unaligned_array_assert) : m_storage(internal::constructor_without_unaligned_array_assert()) { // _check_template_params(); EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED } #endif -#ifdef EIGEN_HAVE_RVALUE_REFERENCES - PlainObjectBase(PlainObjectBase&& other) +#if EIGEN_HAS_RVALUE_REFERENCES + EIGEN_DEVICE_FUNC + PlainObjectBase(PlainObjectBase&& other) EIGEN_NOEXCEPT : m_storage( std::move(other.m_storage) ) { } - PlainObjectBase& operator=(PlainObjectBase&& other) + EIGEN_DEVICE_FUNC + PlainObjectBase& operator=(PlainObjectBase&& other) EIGEN_NOEXCEPT { - using std::swap; - swap(m_storage, other.m_storage); + _check_template_params(); + m_storage = std::move(other.m_storage); return *this; } #endif /** Copy constructor */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PlainObjectBase(const PlainObjectBase& other) + : Base(), m_storage(other.m_storage) { } + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE PlainObjectBase(Index size, Index rows, Index cols) + : m_storage(size, rows, cols) + { +// _check_template_params(); +// EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED + } + + #if EIGEN_HAS_CXX11 + /** \brief Construct a row of column vector with fixed size from an arbitrary number of coefficients. \cpp11 + * + * \only_for_vectors + * + * This constructor is for 1D array or vectors with more than 4 coefficients. + * There exists C++98 analogue constructors for fixed-size array/vector having 1, 2, 3, or 4 coefficients. + * + * \warning To construct a column (resp. row) vector of fixed length, the number of values passed to this + * constructor must match the the fixed number of rows (resp. columns) of \c *this. + */ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + PlainObjectBase(const Scalar& a0, const Scalar& a1, const Scalar& a2, const Scalar& a3, const ArgTypes&... args) + : m_storage() + { + _check_template_params(); + EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, sizeof...(args) + 4); + m_storage.data()[0] = a0; + m_storage.data()[1] = a1; + m_storage.data()[2] = a2; + m_storage.data()[3] = a3; + Index i = 4; + auto x = {(m_storage.data()[i++] = args, 0)...}; + static_cast(x); + } + + /** \brief Constructs a Matrix or Array and initializes it by elements given by an initializer list of initializer + * lists \cpp11 + */ + EIGEN_DEVICE_FUNC + explicit EIGEN_STRONG_INLINE PlainObjectBase(const std::initializer_list>& list) : m_storage() { _check_template_params(); - lazyAssign(other); + + size_t list_size = 0; + if (list.begin() != list.end()) { + list_size = list.begin()->size(); + } + + // This is to allow syntax like VectorXi {{1, 2, 3, 4}} + if (ColsAtCompileTime == 1 && list.size() == 1) { + eigen_assert(list_size == static_cast(RowsAtCompileTime) || RowsAtCompileTime == Dynamic); + resize(list_size, ColsAtCompileTime); + std::copy(list.begin()->begin(), list.begin()->end(), m_storage.data()); + } else { + eigen_assert(list.size() == static_cast(RowsAtCompileTime) || RowsAtCompileTime == Dynamic); + eigen_assert(list_size == static_cast(ColsAtCompileTime) || ColsAtCompileTime == Dynamic); + resize(list.size(), list_size); + + Index row_index = 0; + for (const std::initializer_list& row : list) { + eigen_assert(list_size == row.size()); + Index col_index = 0; + for (const Scalar& e : row) { + coeffRef(row_index, col_index) = e; + ++col_index; + } + ++row_index; + } + } } + #endif // end EIGEN_HAS_CXX11 + /** \sa PlainObjectBase::operator=(const EigenBase&) */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE PlainObjectBase(const DenseBase &other) : m_storage() { _check_template_params(); - lazyAssign(other); + resizeLike(other); + _set_noalias(other); } - EIGEN_STRONG_INLINE PlainObjectBase(Index a_size, Index nbRows, Index nbCols) - : m_storage(a_size, nbRows, nbCols) + /** \sa PlainObjectBase::operator=(const EigenBase&) */ + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase &other) + : m_storage() { -// _check_template_params(); -// EIGEN_INITIALIZE_COEFFS_IF_THAT_OPTION_IS_ENABLED + _check_template_params(); + resizeLike(other); + *this = other.derived(); + } + /** \brief Copy constructor with in-place evaluation */ + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE PlainObjectBase(const ReturnByValue& other) + { + _check_template_params(); + // FIXME this does not automatically transpose vectors if necessary + resize(other.rows(), other.cols()); + other.evalTo(this->derived()); } - /** \copydoc MatrixBase::operator=(const EigenBase&) + public: + + /** \brief Copies the generic expression \a other into *this. + * \copydetails DenseBase::operator=(const EigenBase &other) */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& operator=(const EigenBase &other) { _resize_to_match(other); @@ -484,21 +629,15 @@ class PlainObjectBase : public internal::dense_xpr_base::type return this->derived(); } - /** \sa MatrixBase::operator=(const EigenBase&) */ - template - EIGEN_STRONG_INLINE PlainObjectBase(const EigenBase &other) - : m_storage(Index(other.derived().rows()) * Index(other.derived().cols()), other.derived().rows(), other.derived().cols()) - { - _check_template_params(); - internal::check_rows_cols_for_overflow::run(other.derived().rows(), other.derived().cols()); - Base::operator=(other.derived()); - } - /** \name Map * These are convenience functions returning Map objects. The Map() static functions return unaligned Map objects, * while the AlignedMap() functions return aligned Map objects and thus should be called only with 16-byte-aligned * \a data pointers. * + * Here is an example using strides: + * \include Matrix_Map_stride.cpp + * Output: \verbinclude Matrix_Map_stride.out + * * \see class Map */ //@{ @@ -568,20 +707,28 @@ class PlainObjectBase : public internal::dense_xpr_base::type //@} using Base::setConstant; - Derived& setConstant(Index size, const Scalar& value); - Derived& setConstant(Index rows, Index cols, const Scalar& value); + EIGEN_DEVICE_FUNC Derived& setConstant(Index size, const Scalar& val); + EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, Index cols, const Scalar& val); + EIGEN_DEVICE_FUNC Derived& setConstant(NoChange_t, Index cols, const Scalar& val); + EIGEN_DEVICE_FUNC Derived& setConstant(Index rows, NoChange_t, const Scalar& val); using Base::setZero; - Derived& setZero(Index size); - Derived& setZero(Index rows, Index cols); + EIGEN_DEVICE_FUNC Derived& setZero(Index size); + EIGEN_DEVICE_FUNC Derived& setZero(Index rows, Index cols); + EIGEN_DEVICE_FUNC Derived& setZero(NoChange_t, Index cols); + EIGEN_DEVICE_FUNC Derived& setZero(Index rows, NoChange_t); using Base::setOnes; - Derived& setOnes(Index size); - Derived& setOnes(Index rows, Index cols); + EIGEN_DEVICE_FUNC Derived& setOnes(Index size); + EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, Index cols); + EIGEN_DEVICE_FUNC Derived& setOnes(NoChange_t, Index cols); + EIGEN_DEVICE_FUNC Derived& setOnes(Index rows, NoChange_t); using Base::setRandom; Derived& setRandom(Index size); Derived& setRandom(Index rows, Index cols); + Derived& setRandom(NoChange_t, Index cols); + Derived& setRandom(Index rows, NoChange_t); #ifdef EIGEN_PLAINOBJECTBASE_PLUGIN #include EIGEN_PLAINOBJECTBASE_PLUGIN @@ -596,6 +743,7 @@ class PlainObjectBase : public internal::dense_xpr_base::type * remain row-vectors and vectors remain vectors. */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void _resize_to_match(const EigenBase& other) { #ifdef EIGEN_NO_AUTOMATIC_RESIZING @@ -603,8 +751,6 @@ class PlainObjectBase : public internal::dense_xpr_base::type : (rows() == other.rows() && cols() == other.cols()))) && "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined"); EIGEN_ONLY_USED_FOR_DEBUG(other); - if(this->size()==0) - resizeLike(other); #else resizeLike(other); #endif @@ -624,25 +770,23 @@ class PlainObjectBase : public internal::dense_xpr_base::type * * \internal */ + // aliasing is dealt once in internal::call_assignment + // so at this stage we have to assume aliasing... and resising has to be done later. template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& _set(const DenseBase& other) { - _set_selector(other.derived(), typename internal::conditional(int(OtherDerived::Flags) & EvalBeforeAssigningBit), internal::true_type, internal::false_type>::type()); + internal::call_assignment(this->derived(), other.derived()); return this->derived(); } - template - EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::true_type&) { _set_noalias(other.eval()); } - - template - EIGEN_STRONG_INLINE void _set_selector(const OtherDerived& other, const internal::false_type&) { _set_noalias(other); } - /** \internal Like _set() but additionally makes the assumption that no aliasing effect can happen (which * is the case when creating a new matrix) so one can enforce lazy evaluation. * * \sa operator=(const MatrixBase&), _set() */ template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Derived& _set_noalias(const DenseBase& other) { // I don't think we need this resize call since the lazyAssign will anyways resize @@ -650,44 +794,181 @@ class PlainObjectBase : public internal::dense_xpr_base::type //_resize_to_match(other); // the 'false' below means to enforce lazy evaluation. We don't use lazyAssign() because // it wouldn't allow to copy a row-vector into a column-vector. - return internal::assign_selector::run(this->derived(), other.derived()); + internal::call_assignment_no_alias(this->derived(), other.derived(), internal::assign_op()); + return this->derived(); } template - EIGEN_STRONG_INLINE void _init2(Index nbRows, Index nbCols, typename internal::enable_if::type* = 0) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init2(Index rows, Index cols, typename internal::enable_if::type* = 0) { - EIGEN_STATIC_ASSERT(bool(NumTraits::IsInteger) && - bool(NumTraits::IsInteger), + const bool t0_is_integer_alike = internal::is_valid_index_type::value; + const bool t1_is_integer_alike = internal::is_valid_index_type::value; + EIGEN_STATIC_ASSERT(t0_is_integer_alike && + t1_is_integer_alike, FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED) - resize(nbRows,nbCols); + resize(rows,cols); + } + + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init2(const T0& val0, const T1& val1, typename internal::enable_if::type* = 0) + { + EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2) + m_storage.data()[0] = Scalar(val0); + m_storage.data()[1] = Scalar(val1); } + template - EIGEN_STRONG_INLINE void _init2(const Scalar& val0, const Scalar& val1, typename internal::enable_if::type* = 0) + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init2(const Index& val0, const Index& val1, + typename internal::enable_if< (!internal::is_same::value) + && (internal::is_same::value) + && (internal::is_same::value) + && Base::SizeAtCompileTime==2,T1>::type* = 0) { EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 2) + m_storage.data()[0] = Scalar(val0); + m_storage.data()[1] = Scalar(val1); + } + + // The argument is convertible to the Index type and we either have a non 1x1 Matrix, or a dynamic-sized Array, + // then the argument is meant to be the size of the object. + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(Index size, typename internal::enable_if< (Base::SizeAtCompileTime!=1 || !internal::is_convertible::value) + && ((!internal::is_same::XprKind,ArrayXpr>::value || Base::SizeAtCompileTime==Dynamic)),T>::type* = 0) + { + // NOTE MSVC 2008 complains if we directly put bool(NumTraits::IsInteger) as the EIGEN_STATIC_ASSERT argument. + const bool is_integer_alike = internal::is_valid_index_type::value; + EIGEN_UNUSED_VARIABLE(is_integer_alike); + EIGEN_STATIC_ASSERT(is_integer_alike, + FLOATING_POINT_ARGUMENT_PASSED__INTEGER_WAS_EXPECTED) + resize(size); + } + + // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type can be implicitly converted) + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const Scalar& val0, typename internal::enable_if::value,T>::type* = 0) + { + EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1) m_storage.data()[0] = val0; - m_storage.data()[1] = val1; + } + + // We have a 1x1 matrix/array => the argument is interpreted as the value of the unique coefficient (case where scalar type match the index type) + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const Index& val0, + typename internal::enable_if< (!internal::is_same::value) + && (internal::is_same::value) + && Base::SizeAtCompileTime==1 + && internal::is_convertible::value,T*>::type* = 0) + { + EIGEN_STATIC_ASSERT_VECTOR_SPECIFIC_SIZE(PlainObjectBase, 1) + m_storage.data()[0] = Scalar(val0); + } + + // Initialize a fixed size matrix from a pointer to raw data + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const Scalar* data){ + this->_set_noalias(ConstMapType(data)); + } + + // Initialize an arbitrary matrix from a dense expression + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const DenseBase& other){ + this->_set_noalias(other); + } + + // Initialize an arbitrary matrix from an object convertible to the Derived type. + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const Derived& other){ + this->_set_noalias(other); + } + + // Initialize an arbitrary matrix from a generic Eigen expression + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const EigenBase& other){ + this->derived() = other; + } + + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const ReturnByValue& other) + { + resize(other.rows(), other.cols()); + other.evalTo(this->derived()); + } + + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const RotationBase& r) + { + this->derived() = r; + } + + // For fixed-size Array + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const Scalar& val0, + typename internal::enable_if< Base::SizeAtCompileTime!=Dynamic + && Base::SizeAtCompileTime!=1 + && internal::is_convertible::value + && internal::is_same::XprKind,ArrayXpr>::value,T>::type* = 0) + { + Base::setConstant(val0); + } + + // For fixed-size Array + template + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE void _init1(const Index& val0, + typename internal::enable_if< (!internal::is_same::value) + && (internal::is_same::value) + && Base::SizeAtCompileTime!=Dynamic + && Base::SizeAtCompileTime!=1 + && internal::is_convertible::value + && internal::is_same::XprKind,ArrayXpr>::value,T*>::type* = 0) + { + Base::setConstant(val0); } template friend struct internal::matrix_swap_impl; - /** \internal generic implementation of swap for dense storage since for dynamic-sized matrices of same type it is enough to swap the - * data pointers. + public: + +#ifndef EIGEN_PARSED_BY_DOXYGEN + /** \internal + * \brief Override DenseBase::swap() since for dynamic-sized matrices + * of same type it is enough to swap the data pointers. */ template - void _swap(DenseBase const & other) + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void swap(DenseBase & other) { enum { SwapPointers = internal::is_same::value && Base::SizeAtCompileTime==Dynamic }; - internal::matrix_swap_impl::run(this->derived(), other.const_cast_derived()); + internal::matrix_swap_impl::run(this->derived(), other.derived()); } - public: -#ifndef EIGEN_PARSED_BY_DOXYGEN + /** \internal + * \brief const version forwarded to DenseBase::swap + */ + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void swap(DenseBase const & other) + { Base::swap(other.derived()); } + + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE void _check_template_params() { - EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (Options&RowMajor)==RowMajor) - && EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, (Options&RowMajor)==0) + EIGEN_STATIC_ASSERT((EIGEN_IMPLIES(MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1, (int(Options)&RowMajor)==RowMajor) + && EIGEN_IMPLIES(MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1, (int(Options)&RowMajor)==0) && ((RowsAtCompileTime == Dynamic) || (RowsAtCompileTime >= 0)) && ((ColsAtCompileTime == Dynamic) || (ColsAtCompileTime >= 0)) && ((MaxRowsAtCompileTime == Dynamic) || (MaxRowsAtCompileTime >= 0)) @@ -697,10 +978,20 @@ class PlainObjectBase : public internal::dense_xpr_base::type && (Options & (DontAlign|RowMajor)) == Options), INVALID_MATRIX_TEMPLATE_PARAMETERS) } -#endif -private: - enum { ThisConstantIsPrivateInPlainObjectBase }; + enum { IsPlainObjectBase = 1 }; +#endif + public: + // These apparently need to be down here for nvcc+icc to prevent duplicate + // Map symbol. + template friend class Eigen::Map; + friend class Eigen::Map; + friend class Eigen::Map; +#if EIGEN_MAX_ALIGN_BYTES>0 + // for EIGEN_MAX_ALIGN_BYTES==0, AlignedMax==Unaligned, and many compilers generate warnings for friend-ing a class twice. + friend class Eigen::Map; + friend class Eigen::Map; +#endif }; namespace internal { @@ -708,14 +999,19 @@ namespace internal { template struct conservative_resize_like_impl { - typedef typename Derived::Index Index; + #if EIGEN_HAS_TYPE_TRAITS + static const bool IsRelocatable = std::is_trivially_copyable::value; + #else + static const bool IsRelocatable = !NumTraits::RequireInitialization; + #endif static void run(DenseBase& _this, Index rows, Index cols) { if (_this.rows() == rows && _this.cols() == cols) return; EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived) - if ( ( Derived::IsRowMajor && _this.cols() == cols) || // row-major and we change only the number of rows - (!Derived::IsRowMajor && _this.rows() == rows) ) // column-major and we change only the number of columns + if ( IsRelocatable + && (( Derived::IsRowMajor && _this.cols() == cols) || // row-major and we change only the number of rows + (!Derived::IsRowMajor && _this.rows() == rows) )) // column-major and we change only the number of columns { internal::check_rows_cols_for_overflow::run(rows, cols); _this.derived().m_storage.conservativeResize(rows*cols,rows,cols); @@ -723,9 +1019,9 @@ struct conservative_resize_like_impl else { // The storage order does not allow us to use reallocation. - typename Derived::PlainObject tmp(rows,cols); - const Index common_rows = (std::min)(rows, _this.rows()); - const Index common_cols = (std::min)(cols, _this.cols()); + Derived tmp(rows,cols); + const Index common_rows = numext::mini(rows, _this.rows()); + const Index common_cols = numext::mini(cols, _this.cols()); tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols); _this.derived().swap(tmp); } @@ -743,8 +1039,9 @@ struct conservative_resize_like_impl EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(Derived) EIGEN_STATIC_ASSERT_DYNAMIC_SIZE(OtherDerived) - if ( ( Derived::IsRowMajor && _this.cols() == other.cols()) || // row-major and we change only the number of rows - (!Derived::IsRowMajor && _this.rows() == other.rows()) ) // column-major and we change only the number of columns + if ( IsRelocatable && + (( Derived::IsRowMajor && _this.cols() == other.cols()) || // row-major and we change only the number of rows + (!Derived::IsRowMajor && _this.rows() == other.rows()) )) // column-major and we change only the number of columns { const Index new_rows = other.rows() - _this.rows(); const Index new_cols = other.cols() - _this.cols(); @@ -757,9 +1054,9 @@ struct conservative_resize_like_impl else { // The storage order does not allow us to use reallocation. - typename Derived::PlainObject tmp(other); - const Index common_rows = (std::min)(tmp.rows(), _this.rows()); - const Index common_cols = (std::min)(tmp.cols(), _this.cols()); + Derived tmp(other); + const Index common_rows = numext::mini(tmp.rows(), _this.rows()); + const Index common_cols = numext::mini(tmp.cols(), _this.cols()); tmp.block(0,0,common_rows,common_cols) = _this.block(0,0,common_rows,common_cols); _this.derived().swap(tmp); } @@ -772,14 +1069,18 @@ template struct conservative_resize_like_impl : conservative_resize_like_impl { - using conservative_resize_like_impl::run; - - typedef typename Derived::Index Index; + typedef conservative_resize_like_impl Base; + using Base::run; + using Base::IsRelocatable; + static void run(DenseBase& _this, Index size) { const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : size; const Index new_cols = Derived::RowsAtCompileTime==1 ? size : 1; - _this.derived().m_storage.conservativeResize(size,new_rows,new_cols); + if(IsRelocatable) + _this.derived().m_storage.conservativeResize(size,new_rows,new_cols); + else + Base::run(_this.derived(), new_rows, new_cols); } static void run(DenseBase& _this, const DenseBase& other) @@ -790,7 +1091,10 @@ struct conservative_resize_like_impl const Index new_rows = Derived::RowsAtCompileTime==1 ? 1 : other.rows(); const Index new_cols = Derived::RowsAtCompileTime==1 ? other.cols() : 1; - _this.derived().m_storage.conservativeResize(other.size(),new_rows,new_cols); + if(IsRelocatable) + _this.derived().m_storage.conservativeResize(other.size(),new_rows,new_cols); + else + Base::run(_this.derived(), new_rows, new_cols); if (num_new_elements > 0) _this.tail(num_new_elements) = other.tail(num_new_elements); @@ -800,7 +1104,8 @@ struct conservative_resize_like_impl template struct matrix_swap_impl { - static inline void run(MatrixTypeA& a, MatrixTypeB& b) + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE void run(MatrixTypeA& a, MatrixTypeB& b) { a.base().swap(b); } @@ -809,6 +1114,7 @@ struct matrix_swap_impl template struct matrix_swap_impl { + EIGEN_DEVICE_FUNC static inline void run(MatrixTypeA& a, MatrixTypeB& b) { static_cast(a).m_storage.swap(static_cast(b).m_storage); diff --git a/thirdparty/eigen/Eigen/src/Core/Product.h b/thirdparty/eigen/Eigen/src/Core/Product.h new file mode 100644 index 00000000..70a6c106 --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/Product.h @@ -0,0 +1,191 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2008-2011 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_PRODUCT_H +#define EIGEN_PRODUCT_H + +namespace Eigen { + +template class ProductImpl; + +namespace internal { + +template +struct traits > +{ + typedef typename remove_all::type LhsCleaned; + typedef typename remove_all::type RhsCleaned; + typedef traits LhsTraits; + typedef traits RhsTraits; + + typedef MatrixXpr XprKind; + + typedef typename ScalarBinaryOpTraits::Scalar, typename traits::Scalar>::ReturnType Scalar; + typedef typename product_promote_storage_type::ret>::ret StorageKind; + typedef typename promote_index_type::type StorageIndex; + + enum { + RowsAtCompileTime = LhsTraits::RowsAtCompileTime, + ColsAtCompileTime = RhsTraits::ColsAtCompileTime, + MaxRowsAtCompileTime = LhsTraits::MaxRowsAtCompileTime, + MaxColsAtCompileTime = RhsTraits::MaxColsAtCompileTime, + + // FIXME: only needed by GeneralMatrixMatrixTriangular + InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsTraits::ColsAtCompileTime, RhsTraits::RowsAtCompileTime), + + // The storage order is somewhat arbitrary here. The correct one will be determined through the evaluator. + Flags = (MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1) ? RowMajorBit + : (MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1) ? 0 + : ( ((LhsTraits::Flags&NoPreferredStorageOrderBit) && (RhsTraits::Flags&RowMajorBit)) + || ((RhsTraits::Flags&NoPreferredStorageOrderBit) && (LhsTraits::Flags&RowMajorBit)) ) ? RowMajorBit + : NoPreferredStorageOrderBit + }; +}; + +} // end namespace internal + +/** \class Product + * \ingroup Core_Module + * + * \brief Expression of the product of two arbitrary matrices or vectors + * + * \tparam _Lhs the type of the left-hand side expression + * \tparam _Rhs the type of the right-hand side expression + * + * This class represents an expression of the product of two arbitrary matrices. + * + * The other template parameters are: + * \tparam Option can be DefaultProduct, AliasFreeProduct, or LazyProduct + * + */ +template +class Product : public ProductImpl<_Lhs,_Rhs,Option, + typename internal::product_promote_storage_type::StorageKind, + typename internal::traits<_Rhs>::StorageKind, + internal::product_type<_Lhs,_Rhs>::ret>::ret> +{ + public: + + typedef _Lhs Lhs; + typedef _Rhs Rhs; + + typedef typename ProductImpl< + Lhs, Rhs, Option, + typename internal::product_promote_storage_type::StorageKind, + typename internal::traits::StorageKind, + internal::product_type::ret>::ret>::Base Base; + EIGEN_GENERIC_PUBLIC_INTERFACE(Product) + + typedef typename internal::ref_selector::type LhsNested; + typedef typename internal::ref_selector::type RhsNested; + typedef typename internal::remove_all::type LhsNestedCleaned; + typedef typename internal::remove_all::type RhsNestedCleaned; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + Product(const Lhs& lhs, const Rhs& rhs) : m_lhs(lhs), m_rhs(rhs) + { + eigen_assert(lhs.cols() == rhs.rows() + && "invalid matrix product" + && "if you wanted a coeff-wise or a dot product use the respective explicit functions"); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index rows() const EIGEN_NOEXCEPT { return m_lhs.rows(); } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE EIGEN_CONSTEXPR + Index cols() const EIGEN_NOEXCEPT { return m_rhs.cols(); } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const LhsNestedCleaned& lhs() const { return m_lhs; } + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const RhsNestedCleaned& rhs() const { return m_rhs; } + + protected: + + LhsNested m_lhs; + RhsNested m_rhs; +}; + +namespace internal { + +template::ret> +class dense_product_base + : public internal::dense_xpr_base >::type +{}; + +/** Conversion to scalar for inner-products */ +template +class dense_product_base + : public internal::dense_xpr_base >::type +{ + typedef Product ProductXpr; + typedef typename internal::dense_xpr_base::type Base; +public: + using Base::derived; + typedef typename Base::Scalar Scalar; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE operator const Scalar() const + { + return internal::evaluator(derived()).coeff(0,0); + } +}; + +} // namespace internal + +// Generic API dispatcher +template +class ProductImpl : public internal::generic_xpr_base, MatrixXpr, StorageKind>::type +{ + public: + typedef typename internal::generic_xpr_base, MatrixXpr, StorageKind>::type Base; +}; + +template +class ProductImpl + : public internal::dense_product_base +{ + typedef Product Derived; + + public: + + typedef typename internal::dense_product_base Base; + EIGEN_DENSE_PUBLIC_INTERFACE(Derived) + protected: + enum { + IsOneByOne = (RowsAtCompileTime == 1 || RowsAtCompileTime == Dynamic) && + (ColsAtCompileTime == 1 || ColsAtCompileTime == Dynamic), + EnableCoeff = IsOneByOne || Option==LazyProduct + }; + + public: + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index row, Index col) const + { + EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS); + eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) ); + + return internal::evaluator(derived()).coeff(row,col); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE Scalar coeff(Index i) const + { + EIGEN_STATIC_ASSERT(EnableCoeff, THIS_METHOD_IS_ONLY_FOR_INNER_OR_LAZY_PRODUCTS); + eigen_assert( (Option==LazyProduct) || (this->rows() == 1 && this->cols() == 1) ); + + return internal::evaluator(derived()).coeff(i); + } + + +}; + +} // end namespace Eigen + +#endif // EIGEN_PRODUCT_H diff --git a/thirdparty/eigen/Eigen/src/Core/ProductBase.h b/thirdparty/eigen/Eigen/src/Core/ProductBase.h deleted file mode 100644 index cf74470a..00000000 --- a/thirdparty/eigen/Eigen/src/Core/ProductBase.h +++ /dev/null @@ -1,290 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2009-2010 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_PRODUCTBASE_H -#define EIGEN_PRODUCTBASE_H - -namespace Eigen { - -/** \class ProductBase - * \ingroup Core_Module - * - */ - -namespace internal { -template -struct traits > -{ - typedef MatrixXpr XprKind; - typedef typename remove_all<_Lhs>::type Lhs; - typedef typename remove_all<_Rhs>::type Rhs; - typedef typename scalar_product_traits::ReturnType Scalar; - typedef typename promote_storage_type::StorageKind, - typename traits::StorageKind>::ret StorageKind; - typedef typename promote_index_type::Index, - typename traits::Index>::type Index; - enum { - RowsAtCompileTime = traits::RowsAtCompileTime, - ColsAtCompileTime = traits::ColsAtCompileTime, - MaxRowsAtCompileTime = traits::MaxRowsAtCompileTime, - MaxColsAtCompileTime = traits::MaxColsAtCompileTime, - Flags = (MaxRowsAtCompileTime==1 ? RowMajorBit : 0) - | EvalBeforeNestingBit | EvalBeforeAssigningBit | NestByRefBit, - // Note that EvalBeforeNestingBit and NestByRefBit - // are not used in practice because nested is overloaded for products - CoeffReadCost = 0 // FIXME why is it needed ? - }; -}; -} - -#define EIGEN_PRODUCT_PUBLIC_INTERFACE(Derived) \ - typedef ProductBase Base; \ - EIGEN_DENSE_PUBLIC_INTERFACE(Derived) \ - typedef typename Base::LhsNested LhsNested; \ - typedef typename Base::_LhsNested _LhsNested; \ - typedef typename Base::LhsBlasTraits LhsBlasTraits; \ - typedef typename Base::ActualLhsType ActualLhsType; \ - typedef typename Base::_ActualLhsType _ActualLhsType; \ - typedef typename Base::RhsNested RhsNested; \ - typedef typename Base::_RhsNested _RhsNested; \ - typedef typename Base::RhsBlasTraits RhsBlasTraits; \ - typedef typename Base::ActualRhsType ActualRhsType; \ - typedef typename Base::_ActualRhsType _ActualRhsType; \ - using Base::m_lhs; \ - using Base::m_rhs; - -template -class ProductBase : public MatrixBase -{ - public: - typedef MatrixBase Base; - EIGEN_DENSE_PUBLIC_INTERFACE(ProductBase) - - typedef typename Lhs::Nested LhsNested; - typedef typename internal::remove_all::type _LhsNested; - typedef internal::blas_traits<_LhsNested> LhsBlasTraits; - typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType; - typedef typename internal::remove_all::type _ActualLhsType; - typedef typename internal::traits::Scalar LhsScalar; - - typedef typename Rhs::Nested RhsNested; - typedef typename internal::remove_all::type _RhsNested; - typedef internal::blas_traits<_RhsNested> RhsBlasTraits; - typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType; - typedef typename internal::remove_all::type _ActualRhsType; - typedef typename internal::traits::Scalar RhsScalar; - - // Diagonal of a product: no need to evaluate the arguments because they are going to be evaluated only once - typedef CoeffBasedProduct FullyLazyCoeffBaseProductType; - - public: - -#ifndef EIGEN_NO_MALLOC - typedef typename Base::PlainObject BasePlainObject; - typedef Matrix DynPlainObject; - typedef typename internal::conditional<(BasePlainObject::SizeAtCompileTime==Dynamic) || (BasePlainObject::SizeAtCompileTime*int(sizeof(Scalar)) < int(EIGEN_STACK_ALLOCATION_LIMIT)), - BasePlainObject, DynPlainObject>::type PlainObject; -#else - typedef typename Base::PlainObject PlainObject; -#endif - - ProductBase(const Lhs& a_lhs, const Rhs& a_rhs) - : m_lhs(a_lhs), m_rhs(a_rhs) - { - eigen_assert(a_lhs.cols() == a_rhs.rows() - && "invalid matrix product" - && "if you wanted a coeff-wise or a dot product use the respective explicit functions"); - } - - inline Index rows() const { return m_lhs.rows(); } - inline Index cols() const { return m_rhs.cols(); } - - template - inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst,Scalar(1)); } - - template - inline void addTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(1)); } - - template - inline void subTo(Dest& dst) const { scaleAndAddTo(dst,Scalar(-1)); } - - template - inline void scaleAndAddTo(Dest& dst, const Scalar& alpha) const { derived().scaleAndAddTo(dst,alpha); } - - const _LhsNested& lhs() const { return m_lhs; } - const _RhsNested& rhs() const { return m_rhs; } - - // Implicit conversion to the nested type (trigger the evaluation of the product) - operator const PlainObject& () const - { - m_result.resize(m_lhs.rows(), m_rhs.cols()); - derived().evalTo(m_result); - return m_result; - } - - const Diagonal diagonal() const - { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); } - - template - const Diagonal diagonal() const - { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs); } - - const Diagonal diagonal(Index index) const - { return FullyLazyCoeffBaseProductType(m_lhs, m_rhs).diagonal(index); } - - // restrict coeff accessors to 1x1 expressions. No need to care about mutators here since this isnt a Lvalue expression - typename Base::CoeffReturnType coeff(Index row, Index col) const - { -#ifdef EIGEN2_SUPPORT - return lhs().row(row).cwiseProduct(rhs().col(col).transpose()).sum(); -#else - EIGEN_STATIC_ASSERT_SIZE_1x1(Derived) - eigen_assert(this->rows() == 1 && this->cols() == 1); - Matrix result = *this; - return result.coeff(row,col); -#endif - } - - typename Base::CoeffReturnType coeff(Index i) const - { - EIGEN_STATIC_ASSERT_SIZE_1x1(Derived) - eigen_assert(this->rows() == 1 && this->cols() == 1); - Matrix result = *this; - return result.coeff(i); - } - - const Scalar& coeffRef(Index row, Index col) const - { - EIGEN_STATIC_ASSERT_SIZE_1x1(Derived) - eigen_assert(this->rows() == 1 && this->cols() == 1); - return derived().coeffRef(row,col); - } - - const Scalar& coeffRef(Index i) const - { - EIGEN_STATIC_ASSERT_SIZE_1x1(Derived) - eigen_assert(this->rows() == 1 && this->cols() == 1); - return derived().coeffRef(i); - } - - protected: - - LhsNested m_lhs; - RhsNested m_rhs; - - mutable PlainObject m_result; -}; - -// here we need to overload the nested rule for products -// such that the nested type is a const reference to a plain matrix -namespace internal { -template -struct nested, N, PlainObject> -{ - typedef typename GeneralProduct::PlainObject const& type; -}; -template -struct nested, N, PlainObject> -{ - typedef typename GeneralProduct::PlainObject const& type; -}; -} - -template -class ScaledProduct; - -// Note that these two operator* functions are not defined as member -// functions of ProductBase, because, otherwise we would have to -// define all overloads defined in MatrixBase. Furthermore, Using -// "using Base::operator*" would not work with MSVC. -// -// Also note that here we accept any compatible scalar types -template -const ScaledProduct -operator*(const ProductBase& prod, const typename Derived::Scalar& x) -{ return ScaledProduct(prod.derived(), x); } - -template -typename internal::enable_if::value, - const ScaledProduct >::type -operator*(const ProductBase& prod, const typename Derived::RealScalar& x) -{ return ScaledProduct(prod.derived(), x); } - - -template -const ScaledProduct -operator*(const typename Derived::Scalar& x,const ProductBase& prod) -{ return ScaledProduct(prod.derived(), x); } - -template -typename internal::enable_if::value, - const ScaledProduct >::type -operator*(const typename Derived::RealScalar& x,const ProductBase& prod) -{ return ScaledProduct(prod.derived(), x); } - -namespace internal { -template -struct traits > - : traits, - typename NestedProduct::_LhsNested, - typename NestedProduct::_RhsNested> > -{ - typedef typename traits::StorageKind StorageKind; -}; -} - -template -class ScaledProduct - : public ProductBase, - typename NestedProduct::_LhsNested, - typename NestedProduct::_RhsNested> -{ - public: - typedef ProductBase, - typename NestedProduct::_LhsNested, - typename NestedProduct::_RhsNested> Base; - typedef typename Base::Scalar Scalar; - typedef typename Base::PlainObject PlainObject; -// EIGEN_PRODUCT_PUBLIC_INTERFACE(ScaledProduct) - - ScaledProduct(const NestedProduct& prod, const Scalar& x) - : Base(prod.lhs(),prod.rhs()), m_prod(prod), m_alpha(x) {} - - template - inline void evalTo(Dest& dst) const { dst.setZero(); scaleAndAddTo(dst, Scalar(1)); } - - template - inline void addTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(1)); } - - template - inline void subTo(Dest& dst) const { scaleAndAddTo(dst, Scalar(-1)); } - - template - inline void scaleAndAddTo(Dest& dst, const Scalar& a_alpha) const { m_prod.derived().scaleAndAddTo(dst,a_alpha * m_alpha); } - - const Scalar& alpha() const { return m_alpha; } - - protected: - const NestedProduct& m_prod; - Scalar m_alpha; -}; - -/** \internal - * Overloaded to perform an efficient C = (A*B).lazy() */ -template -template -Derived& MatrixBase::lazyAssign(const ProductBase& other) -{ - other.derived().evalTo(derived()); - return derived(); -} - -} // end namespace Eigen - -#endif // EIGEN_PRODUCTBASE_H diff --git a/thirdparty/eigen/Eigen/src/Core/ProductEvaluators.h b/thirdparty/eigen/Eigen/src/Core/ProductEvaluators.h new file mode 100644 index 00000000..8cf294b2 --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/ProductEvaluators.h @@ -0,0 +1,1179 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2006-2008 Benoit Jacob +// Copyright (C) 2008-2010 Gael Guennebaud +// Copyright (C) 2011 Jitse Niesen +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + + +#ifndef EIGEN_PRODUCTEVALUATORS_H +#define EIGEN_PRODUCTEVALUATORS_H + +namespace Eigen { + +namespace internal { + +/** \internal + * Evaluator of a product expression. + * Since products require special treatments to handle all possible cases, + * we simply defer the evaluation logic to a product_evaluator class + * which offers more partial specialization possibilities. + * + * \sa class product_evaluator + */ +template +struct evaluator > + : public product_evaluator > +{ + typedef Product XprType; + typedef product_evaluator Base; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr) : Base(xpr) {} +}; + +// Catch "scalar * ( A * B )" and transform it to "(A*scalar) * B" +// TODO we should apply that rule only if that's really helpful +template +struct evaluator_assume_aliasing, + const CwiseNullaryOp, Plain1>, + const Product > > +{ + static const bool value = true; +}; +template +struct evaluator, + const CwiseNullaryOp, Plain1>, + const Product > > + : public evaluator > +{ + typedef CwiseBinaryOp, + const CwiseNullaryOp, Plain1>, + const Product > XprType; + typedef evaluator > Base; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr) + : Base(xpr.lhs().functor().m_other * xpr.rhs().lhs() * xpr.rhs().rhs()) + {} +}; + + +template +struct evaluator, DiagIndex> > + : public evaluator, DiagIndex> > +{ + typedef Diagonal, DiagIndex> XprType; + typedef evaluator, DiagIndex> > Base; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE explicit evaluator(const XprType& xpr) + : Base(Diagonal, DiagIndex>( + Product(xpr.nestedExpression().lhs(), xpr.nestedExpression().rhs()), + xpr.index() )) + {} +}; + + +// Helper class to perform a matrix product with the destination at hand. +// Depending on the sizes of the factors, there are different evaluation strategies +// as controlled by internal::product_type. +template< typename Lhs, typename Rhs, + typename LhsShape = typename evaluator_traits::Shape, + typename RhsShape = typename evaluator_traits::Shape, + int ProductType = internal::product_type::value> +struct generic_product_impl; + +template +struct evaluator_assume_aliasing > { + static const bool value = true; +}; + +// This is the default evaluator implementation for products: +// It creates a temporary and call generic_product_impl +template +struct product_evaluator, ProductTag, LhsShape, RhsShape> + : public evaluator::PlainObject> +{ + typedef Product XprType; + typedef typename XprType::PlainObject PlainObject; + typedef evaluator Base; + enum { + Flags = Base::Flags | EvalBeforeNestingBit + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit product_evaluator(const XprType& xpr) + : m_result(xpr.rows(), xpr.cols()) + { + ::new (static_cast(this)) Base(m_result); + +// FIXME shall we handle nested_eval here?, +// if so, then we must take care at removing the call to nested_eval in the specializations (e.g., in permutation_matrix_product, transposition_matrix_product, etc.) +// typedef typename internal::nested_eval::type LhsNested; +// typedef typename internal::nested_eval::type RhsNested; +// typedef typename internal::remove_all::type LhsNestedCleaned; +// typedef typename internal::remove_all::type RhsNestedCleaned; +// +// const LhsNested lhs(xpr.lhs()); +// const RhsNested rhs(xpr.rhs()); +// +// generic_product_impl::evalTo(m_result, lhs, rhs); + + generic_product_impl::evalTo(m_result, xpr.lhs(), xpr.rhs()); + } + +protected: + PlainObject m_result; +}; + +// The following three shortcuts are enabled only if the scalar types match exactly. +// TODO: we could enable them for different scalar types when the product is not vectorized. + +// Dense = Product +template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar> +struct Assignment, internal::assign_op, Dense2Dense, + typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type> +{ + typedef Product SrcXprType; + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void run(DstXprType &dst, const SrcXprType &src, const internal::assign_op &) + { + Index dstRows = src.rows(); + Index dstCols = src.cols(); + if((dst.rows()!=dstRows) || (dst.cols()!=dstCols)) + dst.resize(dstRows, dstCols); + // FIXME shall we handle nested_eval here? + generic_product_impl::evalTo(dst, src.lhs(), src.rhs()); + } +}; + +// Dense += Product +template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar> +struct Assignment, internal::add_assign_op, Dense2Dense, + typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type> +{ + typedef Product SrcXprType; + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void run(DstXprType &dst, const SrcXprType &src, const internal::add_assign_op &) + { + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); + // FIXME shall we handle nested_eval here? + generic_product_impl::addTo(dst, src.lhs(), src.rhs()); + } +}; + +// Dense -= Product +template< typename DstXprType, typename Lhs, typename Rhs, int Options, typename Scalar> +struct Assignment, internal::sub_assign_op, Dense2Dense, + typename enable_if<(Options==DefaultProduct || Options==AliasFreeProduct)>::type> +{ + typedef Product SrcXprType; + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void run(DstXprType &dst, const SrcXprType &src, const internal::sub_assign_op &) + { + eigen_assert(dst.rows() == src.rows() && dst.cols() == src.cols()); + // FIXME shall we handle nested_eval here? + generic_product_impl::subTo(dst, src.lhs(), src.rhs()); + } +}; + + +// Dense ?= scalar * Product +// TODO we should apply that rule if that's really helpful +// for instance, this is not good for inner products +template< typename DstXprType, typename Lhs, typename Rhs, typename AssignFunc, typename Scalar, typename ScalarBis, typename Plain> +struct Assignment, const CwiseNullaryOp,Plain>, + const Product >, AssignFunc, Dense2Dense> +{ + typedef CwiseBinaryOp, + const CwiseNullaryOp,Plain>, + const Product > SrcXprType; + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void run(DstXprType &dst, const SrcXprType &src, const AssignFunc& func) + { + call_assignment_no_alias(dst, (src.lhs().functor().m_other * src.rhs().lhs())*src.rhs().rhs(), func); + } +}; + +//---------------------------------------- +// Catch "Dense ?= xpr + Product<>" expression to save one temporary +// FIXME we could probably enable these rules for any product, i.e., not only Dense and DefaultProduct + +template +struct evaluator_assume_aliasing::Scalar>, const OtherXpr, + const Product >, DenseShape > { + static const bool value = true; +}; + +template +struct evaluator_assume_aliasing::Scalar>, const OtherXpr, + const Product >, DenseShape > { + static const bool value = true; +}; + +template +struct assignment_from_xpr_op_product +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void run(DstXprType &dst, const SrcXprType &src, const InitialFunc& /*func*/) + { + call_assignment_no_alias(dst, src.lhs(), Func1()); + call_assignment_no_alias(dst, src.rhs(), Func2()); + } +}; + +#define EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(ASSIGN_OP,BINOP,ASSIGN_OP2) \ + template< typename DstXprType, typename OtherXpr, typename Lhs, typename Rhs, typename DstScalar, typename SrcScalar, typename OtherScalar,typename ProdScalar> \ + struct Assignment, const OtherXpr, \ + const Product >, internal::ASSIGN_OP, Dense2Dense> \ + : assignment_from_xpr_op_product, internal::ASSIGN_OP, internal::ASSIGN_OP2 > \ + {} + +EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(assign_op, scalar_sum_op,add_assign_op); +EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(add_assign_op,scalar_sum_op,add_assign_op); +EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(sub_assign_op,scalar_sum_op,sub_assign_op); + +EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(assign_op, scalar_difference_op,sub_assign_op); +EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(add_assign_op,scalar_difference_op,sub_assign_op); +EIGEN_CATCH_ASSIGN_XPR_OP_PRODUCT(sub_assign_op,scalar_difference_op,add_assign_op); + +//---------------------------------------- + +template +struct generic_product_impl +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + dst.coeffRef(0,0) = (lhs.transpose().cwiseProduct(rhs)).sum(); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + dst.coeffRef(0,0) += (lhs.transpose().cwiseProduct(rhs)).sum(); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { dst.coeffRef(0,0) -= (lhs.transpose().cwiseProduct(rhs)).sum(); } +}; + + +/*********************************************************************** +* Implementation of outer dense * dense vector product +***********************************************************************/ + +// Column major result +template +void EIGEN_DEVICE_FUNC outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const false_type&) +{ + evaluator rhsEval(rhs); + ei_declare_local_nested_eval(Lhs,lhs,Rhs::SizeAtCompileTime,actual_lhs); + // FIXME if cols is large enough, then it might be useful to make sure that lhs is sequentially stored + // FIXME not very good if rhs is real and lhs complex while alpha is real too + const Index cols = dst.cols(); + for (Index j=0; j +void EIGEN_DEVICE_FUNC outer_product_selector_run(Dst& dst, const Lhs &lhs, const Rhs &rhs, const Func& func, const true_type&) +{ + evaluator lhsEval(lhs); + ei_declare_local_nested_eval(Rhs,rhs,Lhs::SizeAtCompileTime,actual_rhs); + // FIXME if rows is large enough, then it might be useful to make sure that rhs is sequentially stored + // FIXME not very good if lhs is real and rhs complex while alpha is real too + const Index rows = dst.rows(); + for (Index i=0; i +struct generic_product_impl +{ + template struct is_row_major : internal::conditional<(int(T::Flags)&RowMajorBit), internal::true_type, internal::false_type>::type {}; + typedef typename Product::Scalar Scalar; + + // TODO it would be nice to be able to exploit our *_assign_op functors for that purpose + struct set { template EIGEN_DEVICE_FUNC void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() = src; } }; + struct add { template EIGEN_DEVICE_FUNC void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() += src; } }; + struct sub { template EIGEN_DEVICE_FUNC void operator()(const Dst& dst, const Src& src) const { dst.const_cast_derived() -= src; } }; + struct adds { + Scalar m_scale; + explicit adds(const Scalar& s) : m_scale(s) {} + template void EIGEN_DEVICE_FUNC operator()(const Dst& dst, const Src& src) const { + dst.const_cast_derived() += m_scale * src; + } + }; + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + internal::outer_product_selector_run(dst, lhs, rhs, set(), is_row_major()); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + internal::outer_product_selector_run(dst, lhs, rhs, add(), is_row_major()); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + internal::outer_product_selector_run(dst, lhs, rhs, sub(), is_row_major()); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha) + { + internal::outer_product_selector_run(dst, lhs, rhs, adds(alpha), is_row_major()); + } + +}; + + +// This base class provides default implementations for evalTo, addTo, subTo, in terms of scaleAndAddTo +template +struct generic_product_impl_base +{ + typedef typename Product::Scalar Scalar; + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { dst.setZero(); scaleAndAddTo(dst, lhs, rhs, Scalar(1)); } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { scaleAndAddTo(dst,lhs, rhs, Scalar(1)); } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { scaleAndAddTo(dst, lhs, rhs, Scalar(-1)); } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void scaleAndAddTo(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha) + { Derived::scaleAndAddTo(dst,lhs,rhs,alpha); } + +}; + +template +struct generic_product_impl + : generic_product_impl_base > +{ + typedef typename nested_eval::type LhsNested; + typedef typename nested_eval::type RhsNested; + typedef typename Product::Scalar Scalar; + enum { Side = Lhs::IsVectorAtCompileTime ? OnTheLeft : OnTheRight }; + typedef typename internal::remove_all::type>::type MatrixType; + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha) + { + // Fallback to inner product if both the lhs and rhs is a runtime vector. + if (lhs.rows() == 1 && rhs.cols() == 1) { + dst.coeffRef(0,0) += alpha * lhs.row(0).conjugate().dot(rhs.col(0)); + return; + } + LhsNested actual_lhs(lhs); + RhsNested actual_rhs(rhs); + internal::gemv_dense_selector::HasUsableDirectAccess) + >::run(actual_lhs, actual_rhs, dst, alpha); + } +}; + +template +struct generic_product_impl +{ + typedef typename Product::Scalar Scalar; + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + // Same as: dst.noalias() = lhs.lazyProduct(rhs); + // but easier on the compiler side + call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::assign_op()); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + // dst.noalias() += lhs.lazyProduct(rhs); + call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::add_assign_op()); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) + { + // dst.noalias() -= lhs.lazyProduct(rhs); + call_assignment_no_alias(dst, lhs.lazyProduct(rhs), internal::sub_assign_op()); + } + + // This is a special evaluation path called from generic_product_impl<...,GemmProduct> in file GeneralMatrixMatrix.h + // This variant tries to extract scalar multiples from both the LHS and RHS and factor them out. For instance: + // dst {,+,-}= (s1*A)*(B*s2) + // will be rewritten as: + // dst {,+,-}= (s1*s2) * (A.lazyProduct(B)) + // There are at least four benefits of doing so: + // 1 - huge performance gain for heap-allocated matrix types as it save costly allocations. + // 2 - it is faster than simply by-passing the heap allocation through stack allocation. + // 3 - it makes this fallback consistent with the heavy GEMM routine. + // 4 - it fully by-passes huge stack allocation attempts when multiplying huge fixed-size matrices. + // (see https://stackoverflow.com/questions/54738495) + // For small fixed sizes matrices, howver, the gains are less obvious, it is sometimes x2 faster, but sometimes x3 slower, + // and the behavior depends also a lot on the compiler... This is why this re-writting strategy is currently + // enabled only when falling back from the main GEMM. + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void eval_dynamic(Dst& dst, const Lhs& lhs, const Rhs& rhs, const Func &func) + { + enum { + HasScalarFactor = blas_traits::HasScalarFactor || blas_traits::HasScalarFactor, + ConjLhs = blas_traits::NeedToConjugate, + ConjRhs = blas_traits::NeedToConjugate + }; + // FIXME: in c++11 this should be auto, and extractScalarFactor should also return auto + // this is important for real*complex_mat + Scalar actualAlpha = combine_scalar_factors(lhs, rhs); + + eval_dynamic_impl(dst, + blas_traits::extract(lhs).template conjugateIf(), + blas_traits::extract(rhs).template conjugateIf(), + func, + actualAlpha, + typename conditional::type()); + } + +protected: + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void eval_dynamic_impl(Dst& dst, const LhsT& lhs, const RhsT& rhs, const Func &func, const Scalar& s /* == 1 */, false_type) + { + EIGEN_UNUSED_VARIABLE(s); + eigen_internal_assert(s==Scalar(1)); + call_restricted_packet_assignment_no_alias(dst, lhs.lazyProduct(rhs), func); + } + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + void eval_dynamic_impl(Dst& dst, const LhsT& lhs, const RhsT& rhs, const Func &func, const Scalar& s, true_type) + { + call_restricted_packet_assignment_no_alias(dst, s * lhs.lazyProduct(rhs), func); + } +}; + +// This specialization enforces the use of a coefficient-based evaluation strategy +template +struct generic_product_impl + : generic_product_impl {}; + +// Case 2: Evaluate coeff by coeff +// +// This is mostly taken from CoeffBasedProduct.h +// The main difference is that we add an extra argument to the etor_product_*_impl::run() function +// for the inner dimension of the product, because evaluator object do not know their size. + +template +struct etor_product_coeff_impl; + +template +struct etor_product_packet_impl; + +template +struct product_evaluator, ProductTag, DenseShape, DenseShape> + : evaluator_base > +{ + typedef Product XprType; + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit product_evaluator(const XprType& xpr) + : m_lhs(xpr.lhs()), + m_rhs(xpr.rhs()), + m_lhsImpl(m_lhs), // FIXME the creation of the evaluator objects should result in a no-op, but check that! + m_rhsImpl(m_rhs), // Moreover, they are only useful for the packet path, so we could completely disable them when not needed, + // or perhaps declare them on the fly on the packet method... We have experiment to check what's best. + m_innerDim(xpr.lhs().cols()) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits::MulCost); + EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits::AddCost); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); +#if 0 + std::cerr << "LhsOuterStrideBytes= " << LhsOuterStrideBytes << "\n"; + std::cerr << "RhsOuterStrideBytes= " << RhsOuterStrideBytes << "\n"; + std::cerr << "LhsAlignment= " << LhsAlignment << "\n"; + std::cerr << "RhsAlignment= " << RhsAlignment << "\n"; + std::cerr << "CanVectorizeLhs= " << CanVectorizeLhs << "\n"; + std::cerr << "CanVectorizeRhs= " << CanVectorizeRhs << "\n"; + std::cerr << "CanVectorizeInner= " << CanVectorizeInner << "\n"; + std::cerr << "EvalToRowMajor= " << EvalToRowMajor << "\n"; + std::cerr << "Alignment= " << Alignment << "\n"; + std::cerr << "Flags= " << Flags << "\n"; +#endif + } + + // Everything below here is taken from CoeffBasedProduct.h + + typedef typename internal::nested_eval::type LhsNested; + typedef typename internal::nested_eval::type RhsNested; + + typedef typename internal::remove_all::type LhsNestedCleaned; + typedef typename internal::remove_all::type RhsNestedCleaned; + + typedef evaluator LhsEtorType; + typedef evaluator RhsEtorType; + + enum { + RowsAtCompileTime = LhsNestedCleaned::RowsAtCompileTime, + ColsAtCompileTime = RhsNestedCleaned::ColsAtCompileTime, + InnerSize = EIGEN_SIZE_MIN_PREFER_FIXED(LhsNestedCleaned::ColsAtCompileTime, RhsNestedCleaned::RowsAtCompileTime), + MaxRowsAtCompileTime = LhsNestedCleaned::MaxRowsAtCompileTime, + MaxColsAtCompileTime = RhsNestedCleaned::MaxColsAtCompileTime + }; + + typedef typename find_best_packet::type LhsVecPacketType; + typedef typename find_best_packet::type RhsVecPacketType; + + enum { + + LhsCoeffReadCost = LhsEtorType::CoeffReadCost, + RhsCoeffReadCost = RhsEtorType::CoeffReadCost, + CoeffReadCost = InnerSize==0 ? NumTraits::ReadCost + : InnerSize == Dynamic ? HugeCost + : InnerSize * (NumTraits::MulCost + int(LhsCoeffReadCost) + int(RhsCoeffReadCost)) + + (InnerSize - 1) * NumTraits::AddCost, + + Unroll = CoeffReadCost <= EIGEN_UNROLLING_LIMIT, + + LhsFlags = LhsEtorType::Flags, + RhsFlags = RhsEtorType::Flags, + + LhsRowMajor = LhsFlags & RowMajorBit, + RhsRowMajor = RhsFlags & RowMajorBit, + + LhsVecPacketSize = unpacket_traits::size, + RhsVecPacketSize = unpacket_traits::size, + + // Here, we don't care about alignment larger than the usable packet size. + LhsAlignment = EIGEN_PLAIN_ENUM_MIN(LhsEtorType::Alignment,LhsVecPacketSize*int(sizeof(typename LhsNestedCleaned::Scalar))), + RhsAlignment = EIGEN_PLAIN_ENUM_MIN(RhsEtorType::Alignment,RhsVecPacketSize*int(sizeof(typename RhsNestedCleaned::Scalar))), + + SameType = is_same::value, + + CanVectorizeRhs = bool(RhsRowMajor) && (RhsFlags & PacketAccessBit) && (ColsAtCompileTime!=1), + CanVectorizeLhs = (!LhsRowMajor) && (LhsFlags & PacketAccessBit) && (RowsAtCompileTime!=1), + + EvalToRowMajor = (MaxRowsAtCompileTime==1&&MaxColsAtCompileTime!=1) ? 1 + : (MaxColsAtCompileTime==1&&MaxRowsAtCompileTime!=1) ? 0 + : (bool(RhsRowMajor) && !CanVectorizeLhs), + + Flags = ((int(LhsFlags) | int(RhsFlags)) & HereditaryBits & ~RowMajorBit) + | (EvalToRowMajor ? RowMajorBit : 0) + // TODO enable vectorization for mixed types + | (SameType && (CanVectorizeLhs || CanVectorizeRhs) ? PacketAccessBit : 0) + | (XprType::IsVectorAtCompileTime ? LinearAccessBit : 0), + + LhsOuterStrideBytes = int(LhsNestedCleaned::OuterStrideAtCompileTime) * int(sizeof(typename LhsNestedCleaned::Scalar)), + RhsOuterStrideBytes = int(RhsNestedCleaned::OuterStrideAtCompileTime) * int(sizeof(typename RhsNestedCleaned::Scalar)), + + Alignment = bool(CanVectorizeLhs) ? (LhsOuterStrideBytes<=0 || (int(LhsOuterStrideBytes) % EIGEN_PLAIN_ENUM_MAX(1,LhsAlignment))!=0 ? 0 : LhsAlignment) + : bool(CanVectorizeRhs) ? (RhsOuterStrideBytes<=0 || (int(RhsOuterStrideBytes) % EIGEN_PLAIN_ENUM_MAX(1,RhsAlignment))!=0 ? 0 : RhsAlignment) + : 0, + + /* CanVectorizeInner deserves special explanation. It does not affect the product flags. It is not used outside + * of Product. If the Product itself is not a packet-access expression, there is still a chance that the inner + * loop of the product might be vectorized. This is the meaning of CanVectorizeInner. Since it doesn't affect + * the Flags, it is safe to make this value depend on ActualPacketAccessBit, that doesn't affect the ABI. + */ + CanVectorizeInner = SameType + && LhsRowMajor + && (!RhsRowMajor) + && (int(LhsFlags) & int(RhsFlags) & ActualPacketAccessBit) + && (int(InnerSize) % packet_traits::size == 0) + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index row, Index col) const + { + return (m_lhs.row(row).transpose().cwiseProduct( m_rhs.col(col) )).sum(); + } + + /* Allow index-based non-packet access. It is impossible though to allow index-based packed access, + * which is why we don't set the LinearAccessBit. + * TODO: this seems possible when the result is a vector + */ + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const CoeffReturnType coeff(Index index) const + { + const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index; + const Index col = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? index : 0; + return (m_lhs.row(row).transpose().cwiseProduct( m_rhs.col(col) )).sum(); + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const PacketType packet(Index row, Index col) const + { + PacketType res; + typedef etor_product_packet_impl PacketImpl; + PacketImpl::run(row, col, m_lhsImpl, m_rhsImpl, m_innerDim, res); + return res; + } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + const PacketType packet(Index index) const + { + const Index row = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? 0 : index; + const Index col = (RowsAtCompileTime == 1 || MaxRowsAtCompileTime==1) ? index : 0; + return packet(row,col); + } + +protected: + typename internal::add_const_on_value_type::type m_lhs; + typename internal::add_const_on_value_type::type m_rhs; + + LhsEtorType m_lhsImpl; + RhsEtorType m_rhsImpl; + + // TODO: Get rid of m_innerDim if known at compile time + Index m_innerDim; +}; + +template +struct product_evaluator, LazyCoeffBasedProductMode, DenseShape, DenseShape> + : product_evaluator, CoeffBasedProductMode, DenseShape, DenseShape> +{ + typedef Product XprType; + typedef Product BaseProduct; + typedef product_evaluator Base; + enum { + Flags = Base::Flags | EvalBeforeNestingBit + }; + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit product_evaluator(const XprType& xpr) + : Base(BaseProduct(xpr.lhs(),xpr.rhs())) + {} +}; + +/**************************************** +*** Coeff based product, Packet path *** +****************************************/ + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res) + { + etor_product_packet_impl::run(row, col, lhs, rhs, innerDim, res); + res = pmadd(pset1(lhs.coeff(row, Index(UnrollingIndex-1))), rhs.template packet(Index(UnrollingIndex-1), col), res); + } +}; + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet &res) + { + etor_product_packet_impl::run(row, col, lhs, rhs, innerDim, res); + res = pmadd(lhs.template packet(row, Index(UnrollingIndex-1)), pset1(rhs.coeff(Index(UnrollingIndex-1), col)), res); + } +}; + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res) + { + res = pmul(pset1(lhs.coeff(row, Index(0))),rhs.template packet(Index(0), col)); + } +}; + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index /*innerDim*/, Packet &res) + { + res = pmul(lhs.template packet(row, Index(0)), pset1(rhs.coeff(Index(0), col))); + } +}; + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res) + { + res = pset1(typename unpacket_traits::type(0)); + } +}; + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index /*row*/, Index /*col*/, const Lhs& /*lhs*/, const Rhs& /*rhs*/, Index /*innerDim*/, Packet &res) + { + res = pset1(typename unpacket_traits::type(0)); + } +}; + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res) + { + res = pset1(typename unpacket_traits::type(0)); + for(Index i = 0; i < innerDim; ++i) + res = pmadd(pset1(lhs.coeff(row, i)), rhs.template packet(i, col), res); + } +}; + +template +struct etor_product_packet_impl +{ + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Index row, Index col, const Lhs& lhs, const Rhs& rhs, Index innerDim, Packet& res) + { + res = pset1(typename unpacket_traits::type(0)); + for(Index i = 0; i < innerDim; ++i) + res = pmadd(lhs.template packet(row, i), pset1(rhs.coeff(i, col)), res); + } +}; + + +/*************************************************************************** +* Triangular products +***************************************************************************/ +template +struct triangular_product_impl; + +template +struct generic_product_impl + : generic_product_impl_base > +{ + typedef typename Product::Scalar Scalar; + + template + static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha) + { + triangular_product_impl + ::run(dst, lhs.nestedExpression(), rhs, alpha); + } +}; + +template +struct generic_product_impl +: generic_product_impl_base > +{ + typedef typename Product::Scalar Scalar; + + template + static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha) + { + triangular_product_impl::run(dst, lhs, rhs.nestedExpression(), alpha); + } +}; + + +/*************************************************************************** +* SelfAdjoint products +***************************************************************************/ +template +struct selfadjoint_product_impl; + +template +struct generic_product_impl + : generic_product_impl_base > +{ + typedef typename Product::Scalar Scalar; + + template + static EIGEN_DEVICE_FUNC + void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha) + { + selfadjoint_product_impl::run(dst, lhs.nestedExpression(), rhs, alpha); + } +}; + +template +struct generic_product_impl +: generic_product_impl_base > +{ + typedef typename Product::Scalar Scalar; + + template + static void scaleAndAddTo(Dest& dst, const Lhs& lhs, const Rhs& rhs, const Scalar& alpha) + { + selfadjoint_product_impl::run(dst, lhs, rhs.nestedExpression(), alpha); + } +}; + + +/*************************************************************************** +* Diagonal products +***************************************************************************/ + +template +struct diagonal_product_evaluator_base + : evaluator_base +{ + typedef typename ScalarBinaryOpTraits::ReturnType Scalar; +public: + enum { + CoeffReadCost = int(NumTraits::MulCost) + int(evaluator::CoeffReadCost) + int(evaluator::CoeffReadCost), + + MatrixFlags = evaluator::Flags, + DiagFlags = evaluator::Flags, + + _StorageOrder = (Derived::MaxRowsAtCompileTime==1 && Derived::MaxColsAtCompileTime!=1) ? RowMajor + : (Derived::MaxColsAtCompileTime==1 && Derived::MaxRowsAtCompileTime!=1) ? ColMajor + : MatrixFlags & RowMajorBit ? RowMajor : ColMajor, + _SameStorageOrder = _StorageOrder == (MatrixFlags & RowMajorBit ? RowMajor : ColMajor), + + _ScalarAccessOnDiag = !((int(_StorageOrder) == ColMajor && int(ProductOrder) == OnTheLeft) + ||(int(_StorageOrder) == RowMajor && int(ProductOrder) == OnTheRight)), + _SameTypes = is_same::value, + // FIXME currently we need same types, but in the future the next rule should be the one + //_Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) && ((!_PacketOnDiag) || (_SameTypes && bool(int(DiagFlags)&PacketAccessBit))), + _Vectorizable = bool(int(MatrixFlags)&PacketAccessBit) + && _SameTypes + && (_SameStorageOrder || (MatrixFlags&LinearAccessBit)==LinearAccessBit) + && (_ScalarAccessOnDiag || (bool(int(DiagFlags)&PacketAccessBit))), + _LinearAccessMask = (MatrixType::RowsAtCompileTime==1 || MatrixType::ColsAtCompileTime==1) ? LinearAccessBit : 0, + Flags = ((HereditaryBits|_LinearAccessMask) & (unsigned int)(MatrixFlags)) | (_Vectorizable ? PacketAccessBit : 0), + Alignment = evaluator::Alignment, + + AsScalarProduct = (DiagonalType::SizeAtCompileTime==1) + || (DiagonalType::SizeAtCompileTime==Dynamic && MatrixType::RowsAtCompileTime==1 && ProductOrder==OnTheLeft) + || (DiagonalType::SizeAtCompileTime==Dynamic && MatrixType::ColsAtCompileTime==1 && ProductOrder==OnTheRight) + }; + + EIGEN_DEVICE_FUNC diagonal_product_evaluator_base(const MatrixType &mat, const DiagonalType &diag) + : m_diagImpl(diag), m_matImpl(mat) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(NumTraits::MulCost); + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index idx) const + { + if(AsScalarProduct) + return m_diagImpl.coeff(0) * m_matImpl.coeff(idx); + else + return m_diagImpl.coeff(idx) * m_matImpl.coeff(idx); + } + +protected: + template + EIGEN_STRONG_INLINE PacketType packet_impl(Index row, Index col, Index id, internal::true_type) const + { + return internal::pmul(m_matImpl.template packet(row, col), + internal::pset1(m_diagImpl.coeff(id))); + } + + template + EIGEN_STRONG_INLINE PacketType packet_impl(Index row, Index col, Index id, internal::false_type) const + { + enum { + InnerSize = (MatrixType::Flags & RowMajorBit) ? MatrixType::ColsAtCompileTime : MatrixType::RowsAtCompileTime, + DiagonalPacketLoadMode = EIGEN_PLAIN_ENUM_MIN(LoadMode,((InnerSize%16) == 0) ? int(Aligned16) : int(evaluator::Alignment)) // FIXME hardcoded 16!! + }; + return internal::pmul(m_matImpl.template packet(row, col), + m_diagImpl.template packet(id)); + } + + evaluator m_diagImpl; + evaluator m_matImpl; +}; + +// diagonal * dense +template +struct product_evaluator, ProductTag, DiagonalShape, DenseShape> + : diagonal_product_evaluator_base, OnTheLeft> +{ + typedef diagonal_product_evaluator_base, OnTheLeft> Base; + using Base::m_diagImpl; + using Base::m_matImpl; + using Base::coeff; + typedef typename Base::Scalar Scalar; + + typedef Product XprType; + typedef typename XprType::PlainObject PlainObject; + typedef typename Lhs::DiagonalVectorType DiagonalType; + + + enum { StorageOrder = Base::_StorageOrder }; + + EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr) + : Base(xpr.rhs(), xpr.lhs().diagonal()) + { + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const + { + return m_diagImpl.coeff(row) * m_matImpl.coeff(row, col); + } + +#ifndef EIGEN_GPUCC + template + EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const + { + // FIXME: NVCC used to complain about the template keyword, but we have to check whether this is still the case. + // See also similar calls below. + return this->template packet_impl(row,col, row, + typename internal::conditional::type()); + } + + template + EIGEN_STRONG_INLINE PacketType packet(Index idx) const + { + return packet(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx); + } +#endif +}; + +// dense * diagonal +template +struct product_evaluator, ProductTag, DenseShape, DiagonalShape> + : diagonal_product_evaluator_base, OnTheRight> +{ + typedef diagonal_product_evaluator_base, OnTheRight> Base; + using Base::m_diagImpl; + using Base::m_matImpl; + using Base::coeff; + typedef typename Base::Scalar Scalar; + + typedef Product XprType; + typedef typename XprType::PlainObject PlainObject; + + enum { StorageOrder = Base::_StorageOrder }; + + EIGEN_DEVICE_FUNC explicit product_evaluator(const XprType& xpr) + : Base(xpr.lhs(), xpr.rhs().diagonal()) + { + } + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE const Scalar coeff(Index row, Index col) const + { + return m_matImpl.coeff(row, col) * m_diagImpl.coeff(col); + } + +#ifndef EIGEN_GPUCC + template + EIGEN_STRONG_INLINE PacketType packet(Index row, Index col) const + { + return this->template packet_impl(row,col, col, + typename internal::conditional::type()); + } + + template + EIGEN_STRONG_INLINE PacketType packet(Index idx) const + { + return packet(int(StorageOrder)==ColMajor?idx:0,int(StorageOrder)==ColMajor?0:idx); + } +#endif +}; + +/*************************************************************************** +* Products with permutation matrices +***************************************************************************/ + +/** \internal + * \class permutation_matrix_product + * Internal helper class implementing the product between a permutation matrix and a matrix. + * This class is specialized for DenseShape below and for SparseShape in SparseCore/SparsePermutation.h + */ +template +struct permutation_matrix_product; + +template +struct permutation_matrix_product +{ + typedef typename nested_eval::type MatrixType; + typedef typename remove_all::type MatrixTypeCleaned; + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Dest& dst, const PermutationType& perm, const ExpressionType& xpr) + { + MatrixType mat(xpr); + const Index n = Side==OnTheLeft ? mat.rows() : mat.cols(); + // FIXME we need an is_same for expression that is not sensitive to constness. For instance + // is_same_xpr, Block >::value should be true. + //if(is_same::value && extract_data(dst) == extract_data(mat)) + if(is_same_dense(dst, mat)) + { + // apply the permutation inplace + Matrix mask(perm.size()); + mask.fill(false); + Index r = 0; + while(r < perm.size()) + { + // search for the next seed + while(r=perm.size()) + break; + // we got one, let's follow it until we are back to the seed + Index k0 = r++; + Index kPrev = k0; + mask.coeffRef(k0) = true; + for(Index k=perm.indices().coeff(k0); k!=k0; k=perm.indices().coeff(k)) + { + Block(dst, k) + .swap(Block + (dst,((Side==OnTheLeft) ^ Transposed) ? k0 : kPrev)); + + mask.coeffRef(k) = true; + kPrev = k; + } + } + } + else + { + for(Index i = 0; i < n; ++i) + { + Block + (dst, ((Side==OnTheLeft) ^ Transposed) ? perm.indices().coeff(i) : i) + + = + + Block + (mat, ((Side==OnTheRight) ^ Transposed) ? perm.indices().coeff(i) : i); + } + } + } +}; + +template +struct generic_product_impl +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs) + { + permutation_matrix_product::run(dst, lhs, rhs); + } +}; + +template +struct generic_product_impl +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs) + { + permutation_matrix_product::run(dst, rhs, lhs); + } +}; + +template +struct generic_product_impl, Rhs, PermutationShape, MatrixShape, ProductTag> +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Inverse& lhs, const Rhs& rhs) + { + permutation_matrix_product::run(dst, lhs.nestedExpression(), rhs); + } +}; + +template +struct generic_product_impl, MatrixShape, PermutationShape, ProductTag> +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Inverse& rhs) + { + permutation_matrix_product::run(dst, rhs.nestedExpression(), lhs); + } +}; + + +/*************************************************************************** +* Products with transpositions matrices +***************************************************************************/ + +// FIXME could we unify Transpositions and Permutation into a single "shape"?? + +/** \internal + * \class transposition_matrix_product + * Internal helper class implementing the product between a permutation matrix and a matrix. + */ +template +struct transposition_matrix_product +{ + typedef typename nested_eval::type MatrixType; + typedef typename remove_all::type MatrixTypeCleaned; + + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void run(Dest& dst, const TranspositionType& tr, const ExpressionType& xpr) + { + MatrixType mat(xpr); + typedef typename TranspositionType::StorageIndex StorageIndex; + const Index size = tr.size(); + StorageIndex j = 0; + + if(!is_same_dense(dst,mat)) + dst = mat; + + for(Index k=(Transposed?size-1:0) ; Transposed?k>=0:k +struct generic_product_impl +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs) + { + transposition_matrix_product::run(dst, lhs, rhs); + } +}; + +template +struct generic_product_impl +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Rhs& rhs) + { + transposition_matrix_product::run(dst, rhs, lhs); + } +}; + + +template +struct generic_product_impl, Rhs, TranspositionsShape, MatrixShape, ProductTag> +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Transpose& lhs, const Rhs& rhs) + { + transposition_matrix_product::run(dst, lhs.nestedExpression(), rhs); + } +}; + +template +struct generic_product_impl, MatrixShape, TranspositionsShape, ProductTag> +{ + template + static EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE void evalTo(Dest& dst, const Lhs& lhs, const Transpose& rhs) + { + transposition_matrix_product::run(dst, rhs.nestedExpression(), lhs); + } +}; + +} // end namespace internal + +} // end namespace Eigen + +#endif // EIGEN_PRODUCT_EVALUATORS_H diff --git a/thirdparty/eigen/Eigen/src/Core/Random.h b/thirdparty/eigen/Eigen/src/Core/Random.h index 480fea40..dab2ac8e 100644 --- a/thirdparty/eigen/Eigen/src/Core/Random.h +++ b/thirdparty/eigen/Eigen/src/Core/Random.h @@ -16,8 +16,7 @@ namespace internal { template struct scalar_random_op { EIGEN_EMPTY_STRUCT_CTOR(scalar_random_op) - template - inline const Scalar operator() (Index, Index = 0) const { return random(); } + inline const Scalar operator() () const { return random(); } }; template @@ -28,12 +27,18 @@ struct functor_traits > /** \returns a random matrix expression * + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. + * * The parameters \a rows and \a cols are the number of rows and of columns of * the returned matrix. Must be compatible with this MatrixBase type. * + * \not_reentrant + * * This variant is meant to be used for dynamic-size matrix types. For fixed-size types, * it is redundant to pass \a rows and \a cols as arguments, so Random() should be used * instead. + * * * Example: \include MatrixBase_random_int_int.cpp * Output: \verbinclude MatrixBase_random_int_int.out @@ -41,22 +46,28 @@ struct functor_traits > * This expression has the "evaluate before nesting" flag so that it will be evaluated into * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected * behavior with expressions involving random matrices. + * + * See DenseBase::NullaryExpr(Index, const CustomNullaryOp&) for an example using C++11 random generators. * - * \sa MatrixBase::setRandom(), MatrixBase::Random(Index), MatrixBase::Random() + * \sa DenseBase::setRandom(), DenseBase::Random(Index), DenseBase::Random() */ template -inline const CwiseNullaryOp::Scalar>, Derived> +inline const typename DenseBase::RandomReturnType DenseBase::Random(Index rows, Index cols) { return NullaryExpr(rows, cols, internal::scalar_random_op()); } /** \returns a random vector expression + * + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. * * The parameter \a size is the size of the returned vector. * Must be compatible with this MatrixBase type. * * \only_for_vectors + * \not_reentrant * * This variant is meant to be used for dynamic-size vector types. For fixed-size types, * it is redundant to pass \a size as argument, so Random() should be used @@ -69,10 +80,10 @@ DenseBase::Random(Index rows, Index cols) * a temporary vector whenever it is nested in a larger expression. This prevents unexpected * behavior with expressions involving random matrices. * - * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random() + * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random() */ template -inline const CwiseNullaryOp::Scalar>, Derived> +inline const typename DenseBase::RandomReturnType DenseBase::Random(Index size) { return NullaryExpr(size, internal::scalar_random_op()); @@ -80,6 +91,9 @@ DenseBase::Random(Index size) /** \returns a fixed-size random matrix or vector expression * + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. + * * This variant is only for fixed-size MatrixBase types. For dynamic-size types, you * need to use the variants taking size arguments. * @@ -89,11 +103,13 @@ DenseBase::Random(Index size) * This expression has the "evaluate before nesting" flag so that it will be evaluated into * a temporary matrix whenever it is nested in a larger expression. This prevents unexpected * behavior with expressions involving random matrices. + * + * \not_reentrant * - * \sa MatrixBase::setRandom(), MatrixBase::Random(Index,Index), MatrixBase::Random(Index) + * \sa DenseBase::setRandom(), DenseBase::Random(Index,Index), DenseBase::Random(Index) */ template -inline const CwiseNullaryOp::Scalar>, Derived> +inline const typename DenseBase::RandomReturnType DenseBase::Random() { return NullaryExpr(RowsAtCompileTime, ColsAtCompileTime, internal::scalar_random_op()); @@ -101,25 +117,34 @@ DenseBase::Random() /** Sets all coefficients in this expression to random values. * + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. + * + * \not_reentrant + * * Example: \include MatrixBase_setRandom.cpp * Output: \verbinclude MatrixBase_setRandom.out * * \sa class CwiseNullaryOp, setRandom(Index), setRandom(Index,Index) */ template -inline Derived& DenseBase::setRandom() +EIGEN_DEVICE_FUNC inline Derived& DenseBase::setRandom() { return *this = Random(rows(), cols()); } /** Resizes to the given \a newSize, and sets all coefficients in this expression to random values. * + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. + * * \only_for_vectors + * \not_reentrant * * Example: \include Matrix_setRandom_int.cpp * Output: \verbinclude Matrix_setRandom_int.out * - * \sa MatrixBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, MatrixBase::Random() + * \sa DenseBase::setRandom(), setRandom(Index,Index), class CwiseNullaryOp, DenseBase::Random() */ template EIGEN_STRONG_INLINE Derived& @@ -131,22 +156,63 @@ PlainObjectBase::setRandom(Index newSize) /** Resizes to the given size, and sets all coefficients in this expression to random values. * - * \param nbRows the new number of rows - * \param nbCols the new number of columns + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. + * + * \not_reentrant + * + * \param rows the new number of rows + * \param cols the new number of columns * * Example: \include Matrix_setRandom_int_int.cpp * Output: \verbinclude Matrix_setRandom_int_int.out * - * \sa MatrixBase::setRandom(), setRandom(Index), class CwiseNullaryOp, MatrixBase::Random() + * \sa DenseBase::setRandom(), setRandom(Index), class CwiseNullaryOp, DenseBase::Random() */ template EIGEN_STRONG_INLINE Derived& -PlainObjectBase::setRandom(Index nbRows, Index nbCols) +PlainObjectBase::setRandom(Index rows, Index cols) { - resize(nbRows, nbCols); + resize(rows, cols); return setRandom(); } +/** Resizes to the given size, changing only the number of columns, and sets all + * coefficients in this expression to random values. For the parameter of type + * NoChange_t, just pass the special value \c NoChange. + * + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. + * + * \not_reentrant + * + * \sa DenseBase::setRandom(), setRandom(Index), setRandom(Index, NoChange_t), class CwiseNullaryOp, DenseBase::Random() + */ +template +EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setRandom(NoChange_t, Index cols) +{ + return setRandom(rows(), cols); +} + +/** Resizes to the given size, changing only the number of rows, and sets all + * coefficients in this expression to random values. For the parameter of type + * NoChange_t, just pass the special value \c NoChange. + * + * Numbers are uniformly spread through their whole definition range for integer types, + * and in the [-1:1] range for floating point scalar types. + * + * \not_reentrant + * + * \sa DenseBase::setRandom(), setRandom(Index), setRandom(NoChange_t, Index), class CwiseNullaryOp, DenseBase::Random() + */ +template +EIGEN_STRONG_INLINE Derived& +PlainObjectBase::setRandom(Index rows, NoChange_t) +{ + return setRandom(rows, cols()); +} + } // end namespace Eigen #endif // EIGEN_RANDOM_H diff --git a/thirdparty/eigen/Eigen/src/Core/Redux.h b/thirdparty/eigen/Eigen/src/Core/Redux.h index 9b8662a6..b6790d11 100644 --- a/thirdparty/eigen/Eigen/src/Core/Redux.h +++ b/thirdparty/eigen/Eigen/src/Core/Redux.h @@ -23,22 +23,29 @@ namespace internal { * Part 1 : the logic deciding a strategy for vectorization and unrolling ***************************************************************************/ -template +template struct redux_traits { public: + typedef typename find_best_packet::type PacketType; enum { - PacketSize = packet_traits::size, - InnerMaxSize = int(Derived::IsRowMajor) - ? Derived::MaxColsAtCompileTime - : Derived::MaxRowsAtCompileTime + PacketSize = unpacket_traits::size, + InnerMaxSize = int(Evaluator::IsRowMajor) + ? Evaluator::MaxColsAtCompileTime + : Evaluator::MaxRowsAtCompileTime, + OuterMaxSize = int(Evaluator::IsRowMajor) + ? Evaluator::MaxRowsAtCompileTime + : Evaluator::MaxColsAtCompileTime, + SliceVectorizedWork = int(InnerMaxSize)==Dynamic ? Dynamic + : int(OuterMaxSize)==Dynamic ? (int(InnerMaxSize)>=int(PacketSize) ? Dynamic : 0) + : (int(InnerMaxSize)/int(PacketSize)) * int(OuterMaxSize) }; enum { - MightVectorize = (int(Derived::Flags)&ActualPacketAccessBit) + MightVectorize = (int(Evaluator::Flags)&ActualPacketAccessBit) && (functor_traits::PacketAccess), - MayLinearVectorize = MightVectorize && (int(Derived::Flags)&LinearAccessBit), - MaySliceVectorize = MightVectorize && int(InnerMaxSize)>=3*PacketSize + MayLinearVectorize = bool(MightVectorize) && (int(Evaluator::Flags)&LinearAccessBit), + MaySliceVectorize = bool(MightVectorize) && (int(SliceVectorizedWork)==Dynamic || int(SliceVectorizedWork)>=3) }; public: @@ -50,21 +57,36 @@ struct redux_traits public: enum { - Cost = ( Derived::SizeAtCompileTime == Dynamic - || Derived::CoeffReadCost == Dynamic - || (Derived::SizeAtCompileTime!=1 && functor_traits::Cost == Dynamic) - ) ? Dynamic - : Derived::SizeAtCompileTime * Derived::CoeffReadCost - + (Derived::SizeAtCompileTime-1) * functor_traits::Cost, + Cost = Evaluator::SizeAtCompileTime == Dynamic ? HugeCost + : int(Evaluator::SizeAtCompileTime) * int(Evaluator::CoeffReadCost) + (Evaluator::SizeAtCompileTime-1) * functor_traits::Cost, UnrollingLimit = EIGEN_UNROLLING_LIMIT * (int(Traversal) == int(DefaultTraversal) ? 1 : int(PacketSize)) }; public: enum { - Unrolling = Cost != Dynamic && Cost <= UnrollingLimit - ? CompleteUnrolling - : NoUnrolling + Unrolling = Cost <= UnrollingLimit ? CompleteUnrolling : NoUnrolling }; + +#ifdef EIGEN_DEBUG_ASSIGN + static void debug() + { + std::cerr << "Xpr: " << typeid(typename Evaluator::XprType).name() << std::endl; + std::cerr.setf(std::ios::hex, std::ios::basefield); + EIGEN_DEBUG_VAR(Evaluator::Flags) + std::cerr.unsetf(std::ios::hex); + EIGEN_DEBUG_VAR(InnerMaxSize) + EIGEN_DEBUG_VAR(OuterMaxSize) + EIGEN_DEBUG_VAR(SliceVectorizedWork) + EIGEN_DEBUG_VAR(PacketSize) + EIGEN_DEBUG_VAR(MightVectorize) + EIGEN_DEBUG_VAR(MayLinearVectorize) + EIGEN_DEBUG_VAR(MaySliceVectorize) + std::cerr << "Traversal" << " = " << Traversal << " (" << demangle_traversal(Traversal) << ")" << std::endl; + EIGEN_DEBUG_VAR(UnrollingLimit) + std::cerr << "Unrolling" << " = " << Unrolling << " (" << demangle_unrolling(Unrolling) << ")" << std::endl; + std::cerr << std::endl; + } +#endif }; /*************************************************************************** @@ -73,85 +95,86 @@ struct redux_traits /*** no vectorization ***/ -template +template struct redux_novec_unroller { enum { HalfLength = Length/2 }; - typedef typename Derived::Scalar Scalar; + typedef typename Evaluator::Scalar Scalar; - static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func& func) + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE Scalar run(const Evaluator &eval, const Func& func) { - return func(redux_novec_unroller::run(mat,func), - redux_novec_unroller::run(mat,func)); + return func(redux_novec_unroller::run(eval,func), + redux_novec_unroller::run(eval,func)); } }; -template -struct redux_novec_unroller +template +struct redux_novec_unroller { enum { - outer = Start / Derived::InnerSizeAtCompileTime, - inner = Start % Derived::InnerSizeAtCompileTime + outer = Start / Evaluator::InnerSizeAtCompileTime, + inner = Start % Evaluator::InnerSizeAtCompileTime }; - typedef typename Derived::Scalar Scalar; + typedef typename Evaluator::Scalar Scalar; - static EIGEN_STRONG_INLINE Scalar run(const Derived &mat, const Func&) + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE Scalar run(const Evaluator &eval, const Func&) { - return mat.coeffByOuterInner(outer, inner); + return eval.coeffByOuterInner(outer, inner); } }; // This is actually dead code and will never be called. It is required // to prevent false warnings regarding failed inlining though // for 0 length run() will never be called at all. -template -struct redux_novec_unroller +template +struct redux_novec_unroller { - typedef typename Derived::Scalar Scalar; - static EIGEN_STRONG_INLINE Scalar run(const Derived&, const Func&) { return Scalar(); } + typedef typename Evaluator::Scalar Scalar; + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE Scalar run(const Evaluator&, const Func&) { return Scalar(); } }; /*** vectorization ***/ -template +template struct redux_vec_unroller { - enum { - PacketSize = packet_traits::size, - HalfLength = Length/2 - }; - - typedef typename Derived::Scalar Scalar; - typedef typename packet_traits::type PacketScalar; - - static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func& func) + template + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE PacketType run(const Evaluator &eval, const Func& func) { + enum { + PacketSize = unpacket_traits::size, + HalfLength = Length/2 + }; + return func.packetOp( - redux_vec_unroller::run(mat,func), - redux_vec_unroller::run(mat,func) ); + redux_vec_unroller::template run(eval,func), + redux_vec_unroller::template run(eval,func) ); } }; -template -struct redux_vec_unroller +template +struct redux_vec_unroller { - enum { - index = Start * packet_traits::size, - outer = index / int(Derived::InnerSizeAtCompileTime), - inner = index % int(Derived::InnerSizeAtCompileTime), - alignment = (Derived::Flags & AlignedBit) ? Aligned : Unaligned - }; - - typedef typename Derived::Scalar Scalar; - typedef typename packet_traits::type PacketScalar; - - static EIGEN_STRONG_INLINE PacketScalar run(const Derived &mat, const Func&) + template + EIGEN_DEVICE_FUNC + static EIGEN_STRONG_INLINE PacketType run(const Evaluator &eval, const Func&) { - return mat.template packetByOuterInner(outer, inner); + enum { + PacketSize = unpacket_traits::size, + index = Start * PacketSize, + outer = index / int(Evaluator::InnerSizeAtCompileTime), + inner = index % int(Evaluator::InnerSizeAtCompileTime), + alignment = Evaluator::Alignment + }; + return eval.template packetByOuterInner(outer, inner); } }; @@ -159,53 +182,65 @@ struct redux_vec_unroller * Part 3 : implementation of all cases ***************************************************************************/ -template::Traversal, - int Unrolling = redux_traits::Unrolling +template::Traversal, + int Unrolling = redux_traits::Unrolling > struct redux_impl; -template -struct redux_impl +template +struct redux_impl { - typedef typename Derived::Scalar Scalar; - typedef typename Derived::Index Index; - static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func) + typedef typename Evaluator::Scalar Scalar; + + template + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE + Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr) { - eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix"); + eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix"); Scalar res; - res = mat.coeffByOuterInner(0, 0); - for(Index i = 1; i < mat.innerSize(); ++i) - res = func(res, mat.coeffByOuterInner(0, i)); - for(Index i = 1; i < mat.outerSize(); ++i) - for(Index j = 0; j < mat.innerSize(); ++j) - res = func(res, mat.coeffByOuterInner(i, j)); + res = eval.coeffByOuterInner(0, 0); + for(Index i = 1; i < xpr.innerSize(); ++i) + res = func(res, eval.coeffByOuterInner(0, i)); + for(Index i = 1; i < xpr.outerSize(); ++i) + for(Index j = 0; j < xpr.innerSize(); ++j) + res = func(res, eval.coeffByOuterInner(i, j)); return res; } }; -template -struct redux_impl - : public redux_novec_unroller -{}; +template +struct redux_impl + : redux_novec_unroller +{ + typedef redux_novec_unroller Base; + typedef typename Evaluator::Scalar Scalar; + template + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE + Scalar run(const Evaluator &eval, const Func& func, const XprType& /*xpr*/) + { + return Base::run(eval,func); + } +}; -template -struct redux_impl +template +struct redux_impl { - typedef typename Derived::Scalar Scalar; - typedef typename packet_traits::type PacketScalar; - typedef typename Derived::Index Index; + typedef typename Evaluator::Scalar Scalar; + typedef typename redux_traits::PacketType PacketScalar; - static Scalar run(const Derived& mat, const Func& func) + template + static Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr) { - const Index size = mat.size(); - eigen_assert(size && "you are using an empty matrix"); - const Index packetSize = packet_traits::size; - const Index alignedStart = internal::first_aligned(mat); + const Index size = xpr.size(); + + const Index packetSize = redux_traits::PacketSize; + const int packetAlignment = unpacket_traits::alignment; enum { - alignment = bool(Derived::Flags & DirectAccessBit) || bool(Derived::Flags & AlignedBit) - ? Aligned : Unaligned + alignment0 = (bool(Evaluator::Flags & DirectAccessBit) && bool(packet_traits::AlignedOnScalar)) ? int(packetAlignment) : int(Unaligned), + alignment = EIGEN_PLAIN_ENUM_MAX(alignment0, Evaluator::Alignment) }; + const Index alignedStart = internal::first_default_aligned(xpr); const Index alignedSize2 = ((size-alignedStart)/(2*packetSize))*(2*packetSize); const Index alignedSize = ((size-alignedStart)/(packetSize))*(packetSize); const Index alignedEnd2 = alignedStart + alignedSize2; @@ -213,34 +248,34 @@ struct redux_impl Scalar res; if(alignedSize) { - PacketScalar packet_res0 = mat.template packet(alignedStart); + PacketScalar packet_res0 = eval.template packet(alignedStart); if(alignedSize>packetSize) // we have at least two packets to partly unroll the loop { - PacketScalar packet_res1 = mat.template packet(alignedStart+packetSize); + PacketScalar packet_res1 = eval.template packet(alignedStart+packetSize); for(Index index = alignedStart + 2*packetSize; index < alignedEnd2; index += 2*packetSize) { - packet_res0 = func.packetOp(packet_res0, mat.template packet(index)); - packet_res1 = func.packetOp(packet_res1, mat.template packet(index+packetSize)); + packet_res0 = func.packetOp(packet_res0, eval.template packet(index)); + packet_res1 = func.packetOp(packet_res1, eval.template packet(index+packetSize)); } packet_res0 = func.packetOp(packet_res0,packet_res1); if(alignedEnd>alignedEnd2) - packet_res0 = func.packetOp(packet_res0, mat.template packet(alignedEnd2)); + packet_res0 = func.packetOp(packet_res0, eval.template packet(alignedEnd2)); } res = func.predux(packet_res0); for(Index index = 0; index < alignedStart; ++index) - res = func(res,mat.coeff(index)); + res = func(res,eval.coeff(index)); for(Index index = alignedEnd; index < size; ++index) - res = func(res,mat.coeff(index)); + res = func(res,eval.coeff(index)); } else // too small to vectorize anything. // since this is dynamic-size hence inefficient anyway for such small sizes, don't try to optimize. { - res = mat.coeff(0); + res = eval.coeff(0); for(Index index = 1; index < size; ++index) - res = func(res,mat.coeff(index)); + res = func(res,eval.coeff(index)); } return res; @@ -248,65 +283,110 @@ struct redux_impl }; // NOTE: for SliceVectorizedTraversal we simply bypass unrolling -template -struct redux_impl +template +struct redux_impl { - typedef typename Derived::Scalar Scalar; - typedef typename packet_traits::type PacketScalar; - typedef typename Derived::Index Index; + typedef typename Evaluator::Scalar Scalar; + typedef typename redux_traits::PacketType PacketType; - static Scalar run(const Derived& mat, const Func& func) + template + EIGEN_DEVICE_FUNC static Scalar run(const Evaluator &eval, const Func& func, const XprType& xpr) { - eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix"); - const Index innerSize = mat.innerSize(); - const Index outerSize = mat.outerSize(); + eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix"); + const Index innerSize = xpr.innerSize(); + const Index outerSize = xpr.outerSize(); enum { - packetSize = packet_traits::size + packetSize = redux_traits::PacketSize }; const Index packetedInnerSize = ((innerSize)/packetSize)*packetSize; Scalar res; if(packetedInnerSize) { - PacketScalar packet_res = mat.template packet(0,0); + PacketType packet_res = eval.template packet(0,0); for(Index j=0; j(j,i)); + packet_res = func.packetOp(packet_res, eval.template packetByOuterInner(j,i)); res = func.predux(packet_res); for(Index j=0; j::run(mat, func); + res = redux_impl::run(eval, func, xpr); } return res; } }; -template -struct redux_impl +template +struct redux_impl { - typedef typename Derived::Scalar Scalar; - typedef typename packet_traits::type PacketScalar; + typedef typename Evaluator::Scalar Scalar; + + typedef typename redux_traits::PacketType PacketType; enum { - PacketSize = packet_traits::size, - Size = Derived::SizeAtCompileTime, - VectorizedSize = (Size / PacketSize) * PacketSize + PacketSize = redux_traits::PacketSize, + Size = Evaluator::SizeAtCompileTime, + VectorizedSize = (int(Size) / int(PacketSize)) * int(PacketSize) }; - static EIGEN_STRONG_INLINE Scalar run(const Derived& mat, const Func& func) + + template + EIGEN_DEVICE_FUNC static EIGEN_STRONG_INLINE + Scalar run(const Evaluator &eval, const Func& func, const XprType &xpr) { - eigen_assert(mat.rows()>0 && mat.cols()>0 && "you are using an empty matrix"); - Scalar res = func.predux(redux_vec_unroller::run(mat,func)); - if (VectorizedSize != Size) - res = func(res,redux_novec_unroller::run(mat,func)); - return res; + EIGEN_ONLY_USED_FOR_DEBUG(xpr) + eigen_assert(xpr.rows()>0 && xpr.cols()>0 && "you are using an empty matrix"); + if (VectorizedSize > 0) { + Scalar res = func.predux(redux_vec_unroller::template run(eval,func)); + if (VectorizedSize != Size) + res = func(res,redux_novec_unroller::run(eval,func)); + return res; + } + else { + return redux_novec_unroller::run(eval,func); + } } }; +// evaluator adaptor +template +class redux_evaluator : public internal::evaluator<_XprType> +{ + typedef internal::evaluator<_XprType> Base; +public: + typedef _XprType XprType; + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + explicit redux_evaluator(const XprType &xpr) : Base(xpr) {} + + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + typedef typename XprType::PacketScalar PacketScalar; + + enum { + MaxRowsAtCompileTime = XprType::MaxRowsAtCompileTime, + MaxColsAtCompileTime = XprType::MaxColsAtCompileTime, + // TODO we should not remove DirectAccessBit and rather find an elegant way to query the alignment offset at runtime from the evaluator + Flags = Base::Flags & ~DirectAccessBit, + IsRowMajor = XprType::IsRowMajor, + SizeAtCompileTime = XprType::SizeAtCompileTime, + InnerSizeAtCompileTime = XprType::InnerSizeAtCompileTime + }; + + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + CoeffReturnType coeffByOuterInner(Index outer, Index inner) const + { return Base::coeff(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); } + + template + EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE + PacketType packetByOuterInner(Index outer, Index inner) const + { return Base::template packet(IsRowMajor ? outer : inner, IsRowMajor ? inner : outer); } + +}; + } // end namespace internal /*************************************************************************** @@ -317,51 +397,70 @@ struct redux_impl /** \returns the result of a full redux operation on the whole matrix or vector using \a func * * The template parameter \a BinaryOp is the type of the functor \a func which must be - * an associative operator. Both current STL and TR1 functor styles are handled. + * an associative operator. Both current C++98 and C++11 functor styles are handled. + * + * \warning the matrix must be not empty, otherwise an assertion is triggered. * * \sa DenseBase::sum(), DenseBase::minCoeff(), DenseBase::maxCoeff(), MatrixBase::colwise(), MatrixBase::rowwise() */ template template -EIGEN_STRONG_INLINE typename internal::result_of::Scalar)>::type +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits::Scalar DenseBase::redux(const Func& func) const { - typedef typename internal::remove_all::type ThisNested; - return internal::redux_impl - ::run(derived(), func); + eigen_assert(this->rows()>0 && this->cols()>0 && "you are using an empty matrix"); + + typedef typename internal::redux_evaluator ThisEvaluator; + ThisEvaluator thisEval(derived()); + + // The initial expression is passed to the reducer as an additional argument instead of + // passing it as a member of redux_evaluator to help + return internal::redux_impl::run(thisEval, func, derived()); } /** \returns the minimum of all coefficients of \c *this. - * \warning the result is undefined if \c *this contains NaN. + * In case \c *this contains NaN, NaNPropagation determines the behavior: + * NaNPropagation == PropagateFast : undefined + * NaNPropagation == PropagateNaN : result is NaN + * NaNPropagation == PropagateNumbers : result is minimum of elements that are not NaN + * \warning the matrix must be not empty, otherwise an assertion is triggered. */ template -EIGEN_STRONG_INLINE typename internal::traits::Scalar +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits::Scalar DenseBase::minCoeff() const { - return this->redux(Eigen::internal::scalar_min_op()); + return derived().redux(Eigen::internal::scalar_min_op()); } -/** \returns the maximum of all coefficients of \c *this. - * \warning the result is undefined if \c *this contains NaN. +/** \returns the maximum of all coefficients of \c *this. + * In case \c *this contains NaN, NaNPropagation determines the behavior: + * NaNPropagation == PropagateFast : undefined + * NaNPropagation == PropagateNaN : result is NaN + * NaNPropagation == PropagateNumbers : result is maximum of elements that are not NaN + * \warning the matrix must be not empty, otherwise an assertion is triggered. */ template -EIGEN_STRONG_INLINE typename internal::traits::Scalar +template +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits::Scalar DenseBase::maxCoeff() const { - return this->redux(Eigen::internal::scalar_max_op()); + return derived().redux(Eigen::internal::scalar_max_op()); } -/** \returns the sum of all coefficients of *this +/** \returns the sum of all coefficients of \c *this + * + * If \c *this is empty, then the value 0 is returned. * * \sa trace(), prod(), mean() */ template -EIGEN_STRONG_INLINE typename internal::traits::Scalar +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits::Scalar DenseBase::sum() const { if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0)) return Scalar(0); - return this->redux(Eigen::internal::scalar_sum_op()); + return derived().redux(Eigen::internal::scalar_sum_op()); } /** \returns the mean of all coefficients of *this @@ -369,10 +468,17 @@ DenseBase::sum() const * \sa trace(), prod(), sum() */ template -EIGEN_STRONG_INLINE typename internal::traits::Scalar +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits::Scalar DenseBase::mean() const { - return Scalar(this->redux(Eigen::internal::scalar_sum_op())) / Scalar(this->size()); +#ifdef __INTEL_COMPILER + #pragma warning push + #pragma warning ( disable : 2259 ) +#endif + return Scalar(derived().redux(Eigen::internal::scalar_sum_op())) / Scalar(this->size()); +#ifdef __INTEL_COMPILER + #pragma warning pop +#endif } /** \returns the product of all coefficients of *this @@ -383,12 +489,12 @@ DenseBase::mean() const * \sa sum(), mean(), trace() */ template -EIGEN_STRONG_INLINE typename internal::traits::Scalar +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits::Scalar DenseBase::prod() const { if(SizeAtCompileTime==0 || (SizeAtCompileTime==Dynamic && size()==0)) return Scalar(1); - return this->redux(Eigen::internal::scalar_product_op()); + return derived().redux(Eigen::internal::scalar_product_op()); } /** \returns the trace of \c *this, i.e. the sum of the coefficients on the main diagonal. @@ -398,7 +504,7 @@ DenseBase::prod() const * \sa diagonal(), sum() */ template -EIGEN_STRONG_INLINE typename internal::traits::Scalar +EIGEN_DEVICE_FUNC EIGEN_STRONG_INLINE typename internal::traits::Scalar MatrixBase::trace() const { return derived().diagonal().sum(); diff --git a/thirdparty/eigen/Eigen/src/Core/Ref.h b/thirdparty/eigen/Eigen/src/Core/Ref.h index 7a3becaf..07da1555 100644 --- a/thirdparty/eigen/Eigen/src/Core/Ref.h +++ b/thirdparty/eigen/Eigen/src/Core/Ref.h @@ -10,80 +10,7 @@ #ifndef EIGEN_REF_H #define EIGEN_REF_H -namespace Eigen { - -template class RefBase; -template,OuterStride<> >::type > class Ref; - -/** \class Ref - * \ingroup Core_Module - * - * \brief A matrix or vector expression mapping an existing expressions - * - * \tparam PlainObjectType the equivalent matrix type of the mapped data - * \tparam Options specifies whether the pointer is \c #Aligned, or \c #Unaligned. - * The default is \c #Unaligned. - * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1), - * but accept a variable outer stride (leading dimension). - * This can be overridden by specifying strides. - * The type passed here must be a specialization of the Stride template, see examples below. - * - * This class permits to write non template functions taking Eigen's object as parameters while limiting the number of copies. - * A Ref<> object can represent either a const expression or a l-value: - * \code - * // in-out argument: - * void foo1(Ref x); - * - * // read-only const argument: - * void foo2(const Ref& x); - * \endcode - * - * In the in-out case, the input argument must satisfies the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered. - * By default, a Ref can reference any dense vector expression of float having a contiguous memory layout. - * Likewise, a Ref can reference any column major dense matrix expression of float whose column's elements are contiguously stored with - * the possibility to have a constant space inbetween each column, i.e.: the inner stride mmust be equal to 1, but the outer-stride (or leading dimension), - * can be greater than the number of rows. - * - * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function. - * Here are some examples: - * \code - * MatrixXf A; - * VectorXf a; - * foo1(a.head()); // OK - * foo1(A.col()); // OK - * foo1(A.row()); // compilation error because here innerstride!=1 - * foo2(A.row()); // The row is copied into a contiguous temporary - * foo2(2*a); // The expression is evaluated into a temporary - * foo2(A.col().segment(2,4)); // No temporary - * \endcode - * - * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameter. - * Here is an example accepting an innerstride!=1: - * \code - * // in-out argument: - * void foo3(Ref > x); - * foo3(A.row()); // OK - * \endcode - * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involved more - * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overloads internally calling a - * template function, e.g.: - * \code - * // in the .h: - * void foo(const Ref& A); - * void foo(const Ref >& A); - * - * // in the .cpp: - * template void foo_impl(const TypeOfA& A) { - * ... // crazy code goes here - * } - * void foo(const Ref& A) { foo_impl(A); } - * void foo(const Ref >& A) { foo_impl(A); } - * \endcode - * - * - * \sa PlainObjectBase::Map(), \ref TopicStorageOrders - */ +namespace Eigen { namespace internal { @@ -95,25 +22,33 @@ struct traits > typedef _StrideType StrideType; enum { Options = _Options, - Flags = traits >::Flags | NestByRefBit + Flags = traits >::Flags | NestByRefBit, + Alignment = traits >::Alignment }; template struct match { enum { + IsVectorAtCompileTime = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime, HasDirectAccess = internal::has_direct_access::ret, - StorageOrderMatch = PlainObjectType::IsVectorAtCompileTime || Derived::IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)), + StorageOrderMatch = IsVectorAtCompileTime || ((PlainObjectType::Flags&RowMajorBit)==(Derived::Flags&RowMajorBit)), InnerStrideMatch = int(StrideType::InnerStrideAtCompileTime)==int(Dynamic) || int(StrideType::InnerStrideAtCompileTime)==int(Derived::InnerStrideAtCompileTime) || (int(StrideType::InnerStrideAtCompileTime)==0 && int(Derived::InnerStrideAtCompileTime)==1), - OuterStrideMatch = Derived::IsVectorAtCompileTime + OuterStrideMatch = IsVectorAtCompileTime || int(StrideType::OuterStrideAtCompileTime)==int(Dynamic) || int(StrideType::OuterStrideAtCompileTime)==int(Derived::OuterStrideAtCompileTime), - AlignmentMatch = (_Options!=Aligned) || ((PlainObjectType::Flags&AlignedBit)==0) || ((traits::Flags&AlignedBit)==AlignedBit), + // NOTE, this indirection of evaluator::Alignment is needed + // to workaround a very strange bug in MSVC related to the instantiation + // of has_*ary_operator in evaluator. + // This line is surprisingly very sensitive. For instance, simply adding parenthesis + // as "DerivedAlignment = (int(evaluator::Alignment))," will make MSVC fail... + DerivedAlignment = int(evaluator::Alignment), + AlignmentMatch = (int(traits::Alignment)==int(Unaligned)) || (DerivedAlignment >= int(Alignment)), // FIXME the first condition is not very clear, it should be replaced by the required alignment ScalarTypeMatch = internal::is_same::value, MatchAtCompileTime = HasDirectAccess && StorageOrderMatch && InnerStrideMatch && OuterStrideMatch && AlignmentMatch && ScalarTypeMatch }; typedef typename internal::conditional::type type; }; - + }; template @@ -132,12 +67,12 @@ template class RefBase typedef MapBase Base; EIGEN_DENSE_PUBLIC_INTERFACE(RefBase) - inline Index innerStride() const + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index innerStride() const { return StrideType::InnerStrideAtCompileTime != 0 ? m_stride.inner() : 1; } - inline Index outerStride() const + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index outerStride() const { return StrideType::OuterStrideAtCompileTime != 0 ? m_stride.outer() : IsVectorAtCompileTime ? this->size() @@ -145,54 +80,212 @@ template class RefBase : this->rows(); } - RefBase() + EIGEN_DEVICE_FUNC RefBase() : Base(0,RowsAtCompileTime==Dynamic?0:RowsAtCompileTime,ColsAtCompileTime==Dynamic?0:ColsAtCompileTime), // Stride<> does not allow default ctor for Dynamic strides, so let' initialize it with dummy values: m_stride(StrideType::OuterStrideAtCompileTime==Dynamic?0:StrideType::OuterStrideAtCompileTime, StrideType::InnerStrideAtCompileTime==Dynamic?0:StrideType::InnerStrideAtCompileTime) {} - + EIGEN_INHERIT_ASSIGNMENT_OPERATORS(RefBase) protected: typedef Stride StrideBase; + // Resolves inner stride if default 0. + static EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index resolveInnerStride(Index inner) { + return inner == 0 ? 1 : inner; + } + + // Resolves outer stride if default 0. + static EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR Index resolveOuterStride(Index inner, Index outer, Index rows, Index cols, bool isVectorAtCompileTime, bool isRowMajor) { + return outer == 0 ? isVectorAtCompileTime ? inner * rows * cols : isRowMajor ? inner * cols : inner * rows : outer; + } + + // Returns true if construction is valid, false if there is a stride mismatch, + // and fails if there is a size mismatch. template - void construct(Expression& expr) + EIGEN_DEVICE_FUNC bool construct(Expression& expr) { + // Check matrix sizes. If this is a compile-time vector, we do allow + // implicitly transposing. + EIGEN_STATIC_ASSERT( + EIGEN_PREDICATE_SAME_MATRIX_SIZE(PlainObjectType, Expression) + // If it is a vector, the transpose sizes might match. + || ( PlainObjectType::IsVectorAtCompileTime + && ((int(PlainObjectType::RowsAtCompileTime)==Eigen::Dynamic + || int(Expression::ColsAtCompileTime)==Eigen::Dynamic + || int(PlainObjectType::RowsAtCompileTime)==int(Expression::ColsAtCompileTime)) + && (int(PlainObjectType::ColsAtCompileTime)==Eigen::Dynamic + || int(Expression::RowsAtCompileTime)==Eigen::Dynamic + || int(PlainObjectType::ColsAtCompileTime)==int(Expression::RowsAtCompileTime)))), + YOU_MIXED_MATRICES_OF_DIFFERENT_SIZES + ) + + // Determine runtime rows and columns. + Index rows = expr.rows(); + Index cols = expr.cols(); if(PlainObjectType::RowsAtCompileTime==1) { eigen_assert(expr.rows()==1 || expr.cols()==1); - ::new (static_cast(this)) Base(expr.data(), 1, expr.size()); + rows = 1; + cols = expr.size(); } else if(PlainObjectType::ColsAtCompileTime==1) { eigen_assert(expr.rows()==1 || expr.cols()==1); - ::new (static_cast(this)) Base(expr.data(), expr.size(), 1); + rows = expr.size(); + cols = 1; + } + // Verify that the sizes are valid. + eigen_assert( + (PlainObjectType::RowsAtCompileTime == Dynamic) || (PlainObjectType::RowsAtCompileTime == rows)); + eigen_assert( + (PlainObjectType::ColsAtCompileTime == Dynamic) || (PlainObjectType::ColsAtCompileTime == cols)); + + + // If this is a vector, we might be transposing, which means that stride should swap. + const bool transpose = PlainObjectType::IsVectorAtCompileTime && (rows != expr.rows()); + // If the storage format differs, we also need to swap the stride. + const bool row_major = ((PlainObjectType::Flags)&RowMajorBit) != 0; + const bool expr_row_major = (Expression::Flags&RowMajorBit) != 0; + const bool storage_differs = (row_major != expr_row_major); + + const bool swap_stride = (transpose != storage_differs); + + // Determine expr's actual strides, resolving any defaults if zero. + const Index expr_inner_actual = resolveInnerStride(expr.innerStride()); + const Index expr_outer_actual = resolveOuterStride(expr_inner_actual, + expr.outerStride(), + expr.rows(), + expr.cols(), + Expression::IsVectorAtCompileTime != 0, + expr_row_major); + + // If this is a column-major row vector or row-major column vector, the inner-stride + // is arbitrary, so set it to either the compile-time inner stride or 1. + const bool row_vector = (rows == 1); + const bool col_vector = (cols == 1); + const Index inner_stride = + ( (!row_major && row_vector) || (row_major && col_vector) ) ? + ( StrideType::InnerStrideAtCompileTime > 0 ? Index(StrideType::InnerStrideAtCompileTime) : 1) + : swap_stride ? expr_outer_actual : expr_inner_actual; + + // If this is a column-major column vector or row-major row vector, the outer-stride + // is arbitrary, so set it to either the compile-time outer stride or vector size. + const Index outer_stride = + ( (!row_major && col_vector) || (row_major && row_vector) ) ? + ( StrideType::OuterStrideAtCompileTime > 0 ? Index(StrideType::OuterStrideAtCompileTime) : rows * cols * inner_stride) + : swap_stride ? expr_inner_actual : expr_outer_actual; + + // Check if given inner/outer strides are compatible with compile-time strides. + const bool inner_valid = (StrideType::InnerStrideAtCompileTime == Dynamic) + || (resolveInnerStride(Index(StrideType::InnerStrideAtCompileTime)) == inner_stride); + if (!inner_valid) { + return false; + } + + const bool outer_valid = (StrideType::OuterStrideAtCompileTime == Dynamic) + || (resolveOuterStride( + inner_stride, + Index(StrideType::OuterStrideAtCompileTime), + rows, cols, PlainObjectType::IsVectorAtCompileTime != 0, + row_major) + == outer_stride); + if (!outer_valid) { + return false; } - else - ::new (static_cast(this)) Base(expr.data(), expr.rows(), expr.cols()); - - if(Expression::IsVectorAtCompileTime && (!PlainObjectType::IsVectorAtCompileTime) && ((Expression::Flags&RowMajorBit)!=(PlainObjectType::Flags&RowMajorBit))) - ::new (&m_stride) StrideBase(expr.innerStride(), StrideType::InnerStrideAtCompileTime==0?0:1); - else - ::new (&m_stride) StrideBase(StrideType::OuterStrideAtCompileTime==0?0:expr.outerStride(), - StrideType::InnerStrideAtCompileTime==0?0:expr.innerStride()); + + ::new (static_cast(this)) Base(expr.data(), rows, cols); + ::new (&m_stride) StrideBase( + (StrideType::OuterStrideAtCompileTime == 0) ? 0 : outer_stride, + (StrideType::InnerStrideAtCompileTime == 0) ? 0 : inner_stride ); + return true; } StrideBase m_stride; }; - +/** \class Ref + * \ingroup Core_Module + * + * \brief A matrix or vector expression mapping an existing expression + * + * \tparam PlainObjectType the equivalent matrix type of the mapped data + * \tparam Options specifies the pointer alignment in bytes. It can be: \c #Aligned128, , \c #Aligned64, \c #Aligned32, \c #Aligned16, \c #Aligned8 or \c #Unaligned. + * The default is \c #Unaligned. + * \tparam StrideType optionally specifies strides. By default, Ref implies a contiguous storage along the inner dimension (inner stride==1), + * but accepts a variable outer stride (leading dimension). + * This can be overridden by specifying strides. + * The type passed here must be a specialization of the Stride template, see examples below. + * + * This class provides a way to write non-template functions taking Eigen objects as parameters while limiting the number of copies. + * A Ref<> object can represent either a const expression or a l-value: + * \code + * // in-out argument: + * void foo1(Ref x); + * + * // read-only const argument: + * void foo2(const Ref& x); + * \endcode + * + * In the in-out case, the input argument must satisfy the constraints of the actual Ref<> type, otherwise a compilation issue will be triggered. + * By default, a Ref can reference any dense vector expression of float having a contiguous memory layout. + * Likewise, a Ref can reference any column-major dense matrix expression of float whose column's elements are contiguously stored with + * the possibility to have a constant space in-between each column, i.e. the inner stride must be equal to 1, but the outer stride (or leading dimension) + * can be greater than the number of rows. + * + * In the const case, if the input expression does not match the above requirement, then it is evaluated into a temporary before being passed to the function. + * Here are some examples: + * \code + * MatrixXf A; + * VectorXf a; + * foo1(a.head()); // OK + * foo1(A.col()); // OK + * foo1(A.row()); // Compilation error because here innerstride!=1 + * foo2(A.row()); // Compilation error because A.row() is a 1xN object while foo2 is expecting a Nx1 object + * foo2(A.row().transpose()); // The row is copied into a contiguous temporary + * foo2(2*a); // The expression is evaluated into a temporary + * foo2(A.col().segment(2,4)); // No temporary + * \endcode + * + * The range of inputs that can be referenced without temporary can be enlarged using the last two template parameters. + * Here is an example accepting an innerstride!=1: + * \code + * // in-out argument: + * void foo3(Ref > x); + * foo3(A.row()); // OK + * \endcode + * The downside here is that the function foo3 might be significantly slower than foo1 because it won't be able to exploit vectorization, and will involve more + * expensive address computations even if the input is contiguously stored in memory. To overcome this issue, one might propose to overload internally calling a + * template function, e.g.: + * \code + * // in the .h: + * void foo(const Ref& A); + * void foo(const Ref >& A); + * + * // in the .cpp: + * template void foo_impl(const TypeOfA& A) { + * ... // crazy code goes here + * } + * void foo(const Ref& A) { foo_impl(A); } + * void foo(const Ref >& A) { foo_impl(A); } + * \endcode + * + * See also the following stackoverflow questions for further references: + * - Correct usage of the Eigen::Ref<> class + * + * \sa PlainObjectBase::Map(), \ref TopicStorageOrders + */ template class Ref : public RefBase > { private: typedef internal::traits Traits; template - inline Ref(const PlainObjectBase& expr, - typename internal::enable_if::MatchAtCompileTime),Derived>::type* = 0); + EIGEN_DEVICE_FUNC inline Ref(const PlainObjectBase& expr, + typename internal::enable_if::MatchAtCompileTime),Derived>::type* = 0); public: typedef RefBase Base; @@ -201,24 +294,31 @@ template class Ref #ifndef EIGEN_PARSED_BY_DOXYGEN template - inline Ref(PlainObjectBase& expr, - typename internal::enable_if::MatchAtCompileTime),Derived>::type* = 0) + EIGEN_DEVICE_FUNC inline Ref(PlainObjectBase& expr, + typename internal::enable_if::MatchAtCompileTime),Derived>::type* = 0) { - EIGEN_STATIC_ASSERT(static_cast(Traits::template match::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH); - Base::construct(expr.derived()); + EIGEN_STATIC_ASSERT(bool(Traits::template match::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH); + // Construction must pass since we will not create temprary storage in the non-const case. + const bool success = Base::construct(expr.derived()); + EIGEN_UNUSED_VARIABLE(success) + eigen_assert(success); } template - inline Ref(const DenseBase& expr, - typename internal::enable_if::MatchAtCompileTime),Derived>::type* = 0) + EIGEN_DEVICE_FUNC inline Ref(const DenseBase& expr, + typename internal::enable_if::MatchAtCompileTime),Derived>::type* = 0) #else + /** Implicit constructor from any dense expression */ template inline Ref(DenseBase& expr) #endif { - EIGEN_STATIC_ASSERT(static_cast(internal::is_lvalue::value), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY); - EIGEN_STATIC_ASSERT(static_cast(Traits::template match::MatchAtCompileTime), STORAGE_LAYOUT_DOES_NOT_MATCH); - enum { THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY = Derived::ThisConstantIsPrivateInPlainObjectBase}; - Base::construct(expr.const_cast_derived()); + EIGEN_STATIC_ASSERT((static_cast(internal::is_lvalue::value)), THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY); + EIGEN_STATIC_ASSERT((static_cast(Traits::template match::MatchAtCompileTime)), STORAGE_LAYOUT_DOES_NOT_MATCH); + EIGEN_STATIC_ASSERT(!Derived::IsPlainObjectBase,THIS_EXPRESSION_IS_NOT_A_LVALUE__IT_IS_READ_ONLY); + // Construction must pass since we will not create temporary storage in the non-const case. + const bool success = Base::construct(expr.const_cast_derived()); + EIGEN_UNUSED_VARIABLE(success) + eigen_assert(success); } EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Ref) @@ -236,36 +336,39 @@ template class Ref< EIGEN_DENSE_PUBLIC_INTERFACE(Ref) template - inline Ref(const DenseBase& expr, - typename internal::enable_if::ScalarTypeMatch),Derived>::type* = 0) + EIGEN_DEVICE_FUNC inline Ref(const DenseBase& expr, + typename internal::enable_if::ScalarTypeMatch),Derived>::type* = 0) { // std::cout << match_helper::HasDirectAccess << "," << match_helper::OuterStrideMatch << "," << match_helper::InnerStrideMatch << "\n"; // std::cout << int(StrideType::OuterStrideAtCompileTime) << " - " << int(Derived::OuterStrideAtCompileTime) << "\n"; // std::cout << int(StrideType::InnerStrideAtCompileTime) << " - " << int(Derived::InnerStrideAtCompileTime) << "\n"; construct(expr.derived(), typename Traits::template match::type()); } - - inline Ref(const Ref& other) : Base(other) { + + EIGEN_DEVICE_FUNC inline Ref(const Ref& other) : Base(other) { // copy constructor shall not copy the m_object, to avoid unnecessary malloc and copy } template - inline Ref(const RefBase& other) { + EIGEN_DEVICE_FUNC inline Ref(const RefBase& other) { construct(other.derived(), typename Traits::template match::type()); } protected: template - void construct(const Expression& expr,internal::true_type) + EIGEN_DEVICE_FUNC void construct(const Expression& expr,internal::true_type) { - Base::construct(expr); + // Check if we can use the underlying expr's storage directly, otherwise call the copy version. + if (!Base::construct(expr)) { + construct(expr, internal::false_type()); + } } template - void construct(const Expression& expr, internal::false_type) + EIGEN_DEVICE_FUNC void construct(const Expression& expr, internal::false_type) { - m_object.lazyAssign(expr); + internal::call_assignment_no_alias(m_object,expr,internal::assign_op()); Base::construct(m_object); } diff --git a/thirdparty/eigen/Eigen/src/Core/Replicate.h b/thirdparty/eigen/Eigen/src/Core/Replicate.h index ac4537c1..ab5be7e6 100644 --- a/thirdparty/eigen/Eigen/src/Core/Replicate.h +++ b/thirdparty/eigen/Eigen/src/Core/Replicate.h @@ -10,22 +10,7 @@ #ifndef EIGEN_REPLICATE_H #define EIGEN_REPLICATE_H -namespace Eigen { - -/** - * \class Replicate - * \ingroup Core_Module - * - * \brief Expression of the multiple replication of a matrix or vector - * - * \param MatrixType the type of the object we are replicating - * - * This class represents an expression of the multiple replication of a matrix or vector. - * It is the return type of DenseBase::replicate() and most of the time - * this is the only way it is used. - * - * \sa DenseBase::replicate() - */ +namespace Eigen { namespace internal { template @@ -35,10 +20,7 @@ struct traits > typedef typename MatrixType::Scalar Scalar; typedef typename traits::StorageKind StorageKind; typedef typename traits::XprKind XprKind; - enum { - Factor = (RowFactor==Dynamic || ColFactor==Dynamic) ? Dynamic : RowFactor*ColFactor - }; - typedef typename nested::type MatrixTypeNested; + typedef typename ref_selector::type MatrixTypeNested; typedef typename remove_reference::type _MatrixTypeNested; enum { RowsAtCompileTime = RowFactor==Dynamic || int(MatrixType::RowsAtCompileTime)==Dynamic @@ -53,12 +35,29 @@ struct traits > IsRowMajor = MaxRowsAtCompileTime==1 && MaxColsAtCompileTime!=1 ? 1 : MaxColsAtCompileTime==1 && MaxRowsAtCompileTime!=1 ? 0 : (MatrixType::Flags & RowMajorBit) ? 1 : 0, - Flags = (_MatrixTypeNested::Flags & HereditaryBits & ~RowMajorBit) | (IsRowMajor ? RowMajorBit : 0), - CoeffReadCost = _MatrixTypeNested::CoeffReadCost + + // FIXME enable DirectAccess with negative strides? + Flags = IsRowMajor ? RowMajorBit : 0 }; }; } +/** + * \class Replicate + * \ingroup Core_Module + * + * \brief Expression of the multiple replication of a matrix or vector + * + * \tparam MatrixType the type of the object we are replicating + * \tparam RowFactor number of repetitions at compile time along the vertical direction, can be Dynamic. + * \tparam ColFactor number of repetitions at compile time along the horizontal direction, can be Dynamic. + * + * This class represents an expression of the multiple replication of a matrix or vector. + * It is the return type of DenseBase::replicate() and most of the time + * this is the only way it is used. + * + * \sa DenseBase::replicate() + */ template class Replicate : public internal::dense_xpr_base< Replicate >::type { @@ -68,10 +67,12 @@ template class Replicate typedef typename internal::dense_xpr_base::type Base; EIGEN_DENSE_PUBLIC_INTERFACE(Replicate) + typedef typename internal::remove_all::type NestedExpression; template - inline explicit Replicate(const OriginalMatrixType& a_matrix) - : m_matrix(a_matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor) + EIGEN_DEVICE_FUNC + inline explicit Replicate(const OriginalMatrixType& matrix) + : m_matrix(matrix), m_rowFactor(RowFactor), m_colFactor(ColFactor) { EIGEN_STATIC_ASSERT((internal::is_same::type,OriginalMatrixType>::value), THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE) @@ -79,44 +80,23 @@ template class Replicate } template - inline Replicate(const OriginalMatrixType& a_matrix, Index rowFactor, Index colFactor) - : m_matrix(a_matrix), m_rowFactor(rowFactor), m_colFactor(colFactor) + EIGEN_DEVICE_FUNC + inline Replicate(const OriginalMatrixType& matrix, Index rowFactor, Index colFactor) + : m_matrix(matrix), m_rowFactor(rowFactor), m_colFactor(colFactor) { EIGEN_STATIC_ASSERT((internal::is_same::type,OriginalMatrixType>::value), THE_MATRIX_OR_EXPRESSION_THAT_YOU_PASSED_DOES_NOT_HAVE_THE_EXPECTED_TYPE) } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index rows() const { return m_matrix.rows() * m_rowFactor.value(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR inline Index cols() const { return m_matrix.cols() * m_colFactor.value(); } - inline Scalar coeff(Index rowId, Index colId) const - { - // try to avoid using modulo; this is a pure optimization strategy - const Index actual_row = internal::traits::RowsAtCompileTime==1 ? 0 - : RowFactor==1 ? rowId - : rowId%m_matrix.rows(); - const Index actual_col = internal::traits::ColsAtCompileTime==1 ? 0 - : ColFactor==1 ? colId - : colId%m_matrix.cols(); - - return m_matrix.coeff(actual_row, actual_col); - } - template - inline PacketScalar packet(Index rowId, Index colId) const - { - const Index actual_row = internal::traits::RowsAtCompileTime==1 ? 0 - : RowFactor==1 ? rowId - : rowId%m_matrix.rows(); - const Index actual_col = internal::traits::ColsAtCompileTime==1 ? 0 - : ColFactor==1 ? colId - : colId%m_matrix.cols(); - - return m_matrix.template packet(actual_row, actual_col); - } - + EIGEN_DEVICE_FUNC const _MatrixTypeNested& nestedExpression() const - { - return m_matrix; + { + return m_matrix; } protected: @@ -135,27 +115,12 @@ template class Replicate */ template template -const Replicate +EIGEN_DEVICE_FUNC const Replicate DenseBase::replicate() const { return Replicate(derived()); } -/** - * \return an expression of the replication of \c *this - * - * Example: \include MatrixBase_replicate_int_int.cpp - * Output: \verbinclude MatrixBase_replicate_int_int.out - * - * \sa VectorwiseOp::replicate(), DenseBase::replicate(), class Replicate - */ -template -const typename DenseBase::ReplicateReturnType -DenseBase::replicate(Index rowFactor,Index colFactor) const -{ - return Replicate(derived(),rowFactor,colFactor); -} - /** * \return an expression of the replication of each column (or row) of \c *this * @@ -165,7 +130,7 @@ DenseBase::replicate(Index rowFactor,Index colFactor) const * \sa VectorwiseOp::replicate(), DenseBase::replicate(), class Replicate */ template -const typename VectorwiseOp::ReplicateReturnType +EIGEN_DEVICE_FUNC const typename VectorwiseOp::ReplicateReturnType VectorwiseOp::replicate(Index factor) const { return typename VectorwiseOp::ReplicateReturnType diff --git a/thirdparty/eigen/Eigen/src/Core/Reshaped.h b/thirdparty/eigen/Eigen/src/Core/Reshaped.h new file mode 100644 index 00000000..882314cf --- /dev/null +++ b/thirdparty/eigen/Eigen/src/Core/Reshaped.h @@ -0,0 +1,454 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2008-2017 Gael Guennebaud +// Copyright (C) 2014 yoco +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_RESHAPED_H +#define EIGEN_RESHAPED_H + +namespace Eigen { + +/** \class Reshaped + * \ingroup Core_Module + * + * \brief Expression of a fixed-size or dynamic-size reshape + * + * \tparam XprType the type of the expression in which we are taking a reshape + * \tparam Rows the number of rows of the reshape we are taking at compile time (optional) + * \tparam Cols the number of columns of the reshape we are taking at compile time (optional) + * \tparam Order can be ColMajor or RowMajor, default is ColMajor. + * + * This class represents an expression of either a fixed-size or dynamic-size reshape. + * It is the return type of DenseBase::reshaped(NRowsType,NColsType) and + * most of the time this is the only way it is used. + * + * However, in C++98, if you want to directly maniputate reshaped expressions, + * for instance if you want to write a function returning such an expression, you + * will need to use this class. In C++11, it is advised to use the \em auto + * keyword for such use cases. + * + * Here is an example illustrating the dynamic case: + * \include class_Reshaped.cpp + * Output: \verbinclude class_Reshaped.out + * + * Here is an example illustrating the fixed-size case: + * \include class_FixedReshaped.cpp + * Output: \verbinclude class_FixedReshaped.out + * + * \sa DenseBase::reshaped(NRowsType,NColsType) + */ + +namespace internal { + +template +struct traits > : traits +{ + typedef typename traits::Scalar Scalar; + typedef typename traits::StorageKind StorageKind; + typedef typename traits::XprKind XprKind; + enum{ + MatrixRows = traits::RowsAtCompileTime, + MatrixCols = traits::ColsAtCompileTime, + RowsAtCompileTime = Rows, + ColsAtCompileTime = Cols, + MaxRowsAtCompileTime = Rows, + MaxColsAtCompileTime = Cols, + XpxStorageOrder = ((int(traits::Flags) & RowMajorBit) == RowMajorBit) ? RowMajor : ColMajor, + ReshapedStorageOrder = (RowsAtCompileTime == 1 && ColsAtCompileTime != 1) ? RowMajor + : (ColsAtCompileTime == 1 && RowsAtCompileTime != 1) ? ColMajor + : XpxStorageOrder, + HasSameStorageOrderAsXprType = (ReshapedStorageOrder == XpxStorageOrder), + InnerSize = (ReshapedStorageOrder==int(RowMajor)) ? int(ColsAtCompileTime) : int(RowsAtCompileTime), + InnerStrideAtCompileTime = HasSameStorageOrderAsXprType + ? int(inner_stride_at_compile_time::ret) + : Dynamic, + OuterStrideAtCompileTime = Dynamic, + + HasDirectAccess = internal::has_direct_access::ret + && (Order==int(XpxStorageOrder)) + && ((evaluator::Flags&LinearAccessBit)==LinearAccessBit), + + MaskPacketAccessBit = (InnerSize == Dynamic || (InnerSize % packet_traits::size) == 0) + && (InnerStrideAtCompileTime == 1) + ? PacketAccessBit : 0, + //MaskAlignedBit = ((OuterStrideAtCompileTime!=Dynamic) && (((OuterStrideAtCompileTime * int(sizeof(Scalar))) % 16) == 0)) ? AlignedBit : 0, + FlagsLinearAccessBit = (RowsAtCompileTime == 1 || ColsAtCompileTime == 1) ? LinearAccessBit : 0, + FlagsLvalueBit = is_lvalue::value ? LvalueBit : 0, + FlagsRowMajorBit = (ReshapedStorageOrder==int(RowMajor)) ? RowMajorBit : 0, + FlagsDirectAccessBit = HasDirectAccess ? DirectAccessBit : 0, + Flags0 = traits::Flags & ( (HereditaryBits & ~RowMajorBit) | MaskPacketAccessBit), + + Flags = (Flags0 | FlagsLinearAccessBit | FlagsLvalueBit | FlagsRowMajorBit | FlagsDirectAccessBit) + }; +}; + +template class ReshapedImpl_dense; + +} // end namespace internal + +template class ReshapedImpl; + +template class Reshaped + : public ReshapedImpl::StorageKind> +{ + typedef ReshapedImpl::StorageKind> Impl; + public: + //typedef typename Impl::Base Base; + typedef Impl Base; + EIGEN_GENERIC_PUBLIC_INTERFACE(Reshaped) + EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reshaped) + + /** Fixed-size constructor + */ + EIGEN_DEVICE_FUNC + inline Reshaped(XprType& xpr) + : Impl(xpr) + { + EIGEN_STATIC_ASSERT(RowsAtCompileTime!=Dynamic && ColsAtCompileTime!=Dynamic,THIS_METHOD_IS_ONLY_FOR_FIXED_SIZE) + eigen_assert(Rows * Cols == xpr.rows() * xpr.cols()); + } + + /** Dynamic-size constructor + */ + EIGEN_DEVICE_FUNC + inline Reshaped(XprType& xpr, + Index reshapeRows, Index reshapeCols) + : Impl(xpr, reshapeRows, reshapeCols) + { + eigen_assert((RowsAtCompileTime==Dynamic || RowsAtCompileTime==reshapeRows) + && (ColsAtCompileTime==Dynamic || ColsAtCompileTime==reshapeCols)); + eigen_assert(reshapeRows * reshapeCols == xpr.rows() * xpr.cols()); + } +}; + +// The generic default implementation for dense reshape simply forward to the internal::ReshapedImpl_dense +// that must be specialized for direct and non-direct access... +template +class ReshapedImpl + : public internal::ReshapedImpl_dense >::HasDirectAccess> +{ + typedef internal::ReshapedImpl_dense >::HasDirectAccess> Impl; + public: + typedef Impl Base; + EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl) + EIGEN_DEVICE_FUNC inline ReshapedImpl(XprType& xpr) : Impl(xpr) {} + EIGEN_DEVICE_FUNC inline ReshapedImpl(XprType& xpr, Index reshapeRows, Index reshapeCols) + : Impl(xpr, reshapeRows, reshapeCols) {} +}; + +namespace internal { + +/** \internal Internal implementation of dense Reshaped in the general case. */ +template +class ReshapedImpl_dense + : public internal::dense_xpr_base >::type +{ + typedef Reshaped ReshapedType; + public: + + typedef typename internal::dense_xpr_base::type Base; + EIGEN_DENSE_PUBLIC_INTERFACE(ReshapedType) + EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl_dense) + + typedef typename internal::ref_selector::non_const_type MatrixTypeNested; + typedef typename internal::remove_all::type NestedExpression; + + class InnerIterator; + + /** Fixed-size constructor + */ + EIGEN_DEVICE_FUNC + inline ReshapedImpl_dense(XprType& xpr) + : m_xpr(xpr), m_rows(Rows), m_cols(Cols) + {} + + /** Dynamic-size constructor + */ + EIGEN_DEVICE_FUNC + inline ReshapedImpl_dense(XprType& xpr, Index nRows, Index nCols) + : m_xpr(xpr), m_rows(nRows), m_cols(nCols) + {} + + EIGEN_DEVICE_FUNC Index rows() const { return m_rows; } + EIGEN_DEVICE_FUNC Index cols() const { return m_cols; } + + #ifdef EIGEN_PARSED_BY_DOXYGEN + /** \sa MapBase::data() */ + EIGEN_DEVICE_FUNC inline const Scalar* data() const; + EIGEN_DEVICE_FUNC inline Index innerStride() const; + EIGEN_DEVICE_FUNC inline Index outerStride() const; + #endif + + /** \returns the nested expression */ + EIGEN_DEVICE_FUNC + const typename internal::remove_all::type& + nestedExpression() const { return m_xpr; } + + /** \returns the nested expression */ + EIGEN_DEVICE_FUNC + typename internal::remove_reference::type& + nestedExpression() { return m_xpr; } + + protected: + + MatrixTypeNested m_xpr; + const internal::variable_if_dynamic m_rows; + const internal::variable_if_dynamic m_cols; +}; + + +/** \internal Internal implementation of dense Reshaped in the direct access case. */ +template +class ReshapedImpl_dense + : public MapBase > +{ + typedef Reshaped ReshapedType; + typedef typename internal::ref_selector::non_const_type XprTypeNested; + public: + + typedef MapBase Base; + EIGEN_DENSE_PUBLIC_INTERFACE(ReshapedType) + EIGEN_INHERIT_ASSIGNMENT_OPERATORS(ReshapedImpl_dense) + + /** Fixed-size constructor + */ + EIGEN_DEVICE_FUNC + inline ReshapedImpl_dense(XprType& xpr) + : Base(xpr.data()), m_xpr(xpr) + {} + + /** Dynamic-size constructor + */ + EIGEN_DEVICE_FUNC + inline ReshapedImpl_dense(XprType& xpr, Index nRows, Index nCols) + : Base(xpr.data(), nRows, nCols), + m_xpr(xpr) + {} + + EIGEN_DEVICE_FUNC + const typename internal::remove_all::type& nestedExpression() const + { + return m_xpr; + } + + EIGEN_DEVICE_FUNC + XprType& nestedExpression() { return m_xpr; } + + /** \sa MapBase::innerStride() */ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index innerStride() const + { + return m_xpr.innerStride(); + } + + /** \sa MapBase::outerStride() */ + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index outerStride() const + { + return (((Flags&RowMajorBit)==RowMajorBit) ? this->cols() : this->rows()) * m_xpr.innerStride(); + } + + protected: + + XprTypeNested m_xpr; +}; + +// Evaluators +template struct reshaped_evaluator; + +template +struct evaluator > + : reshaped_evaluator >::HasDirectAccess> +{ + typedef Reshaped XprType; + typedef typename XprType::Scalar Scalar; + // TODO: should check for smaller packet types + typedef typename packet_traits::type PacketScalar; + + enum { + CoeffReadCost = evaluator::CoeffReadCost, + HasDirectAccess = traits::HasDirectAccess, + +// RowsAtCompileTime = traits::RowsAtCompileTime, +// ColsAtCompileTime = traits::ColsAtCompileTime, +// MaxRowsAtCompileTime = traits::MaxRowsAtCompileTime, +// MaxColsAtCompileTime = traits::MaxColsAtCompileTime, +// +// InnerStrideAtCompileTime = traits::HasSameStorageOrderAsXprType +// ? int(inner_stride_at_compile_time::ret) +// : Dynamic, +// OuterStrideAtCompileTime = Dynamic, + + FlagsLinearAccessBit = (traits::RowsAtCompileTime == 1 || traits::ColsAtCompileTime == 1 || HasDirectAccess) ? LinearAccessBit : 0, + FlagsRowMajorBit = (traits::ReshapedStorageOrder==int(RowMajor)) ? RowMajorBit : 0, + FlagsDirectAccessBit = HasDirectAccess ? DirectAccessBit : 0, + Flags0 = evaluator::Flags & (HereditaryBits & ~RowMajorBit), + Flags = Flags0 | FlagsLinearAccessBit | FlagsRowMajorBit | FlagsDirectAccessBit, + + PacketAlignment = unpacket_traits::alignment, + Alignment = evaluator::Alignment + }; + typedef reshaped_evaluator reshaped_evaluator_type; + EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) : reshaped_evaluator_type(xpr) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } +}; + +template +struct reshaped_evaluator + : evaluator_base > +{ + typedef Reshaped XprType; + + enum { + CoeffReadCost = evaluator::CoeffReadCost /* TODO + cost of index computations */, + + Flags = (evaluator::Flags & (HereditaryBits /*| LinearAccessBit | DirectAccessBit*/)), + + Alignment = 0 + }; + + EIGEN_DEVICE_FUNC explicit reshaped_evaluator(const XprType& xpr) : m_argImpl(xpr.nestedExpression()), m_xpr(xpr) + { + EIGEN_INTERNAL_CHECK_COST_VALUE(CoeffReadCost); + } + + typedef typename XprType::Scalar Scalar; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + typedef std::pair RowCol; + + inline RowCol index_remap(Index rowId, Index colId) const + { + if(Order==ColMajor) + { + const Index nth_elem_idx = colId * m_xpr.rows() + rowId; + return RowCol(nth_elem_idx % m_xpr.nestedExpression().rows(), + nth_elem_idx / m_xpr.nestedExpression().rows()); + } + else + { + const Index nth_elem_idx = colId + rowId * m_xpr.cols(); + return RowCol(nth_elem_idx / m_xpr.nestedExpression().cols(), + nth_elem_idx % m_xpr.nestedExpression().cols()); + } + } + + EIGEN_DEVICE_FUNC + inline Scalar& coeffRef(Index rowId, Index colId) + { + EIGEN_STATIC_ASSERT_LVALUE(XprType) + const RowCol row_col = index_remap(rowId, colId); + return m_argImpl.coeffRef(row_col.first, row_col.second); + } + + EIGEN_DEVICE_FUNC + inline const Scalar& coeffRef(Index rowId, Index colId) const + { + const RowCol row_col = index_remap(rowId, colId); + return m_argImpl.coeffRef(row_col.first, row_col.second); + } + + EIGEN_DEVICE_FUNC + EIGEN_STRONG_INLINE const CoeffReturnType coeff(Index rowId, Index colId) const + { + const RowCol row_col = index_remap(rowId, colId); + return m_argImpl.coeff(row_col.first, row_col.second); + } + + EIGEN_DEVICE_FUNC + inline Scalar& coeffRef(Index index) + { + EIGEN_STATIC_ASSERT_LVALUE(XprType) + const RowCol row_col = index_remap(Rows == 1 ? 0 : index, + Rows == 1 ? index : 0); + return m_argImpl.coeffRef(row_col.first, row_col.second); + + } + + EIGEN_DEVICE_FUNC + inline const Scalar& coeffRef(Index index) const + { + const RowCol row_col = index_remap(Rows == 1 ? 0 : index, + Rows == 1 ? index : 0); + return m_argImpl.coeffRef(row_col.first, row_col.second); + } + + EIGEN_DEVICE_FUNC + inline const CoeffReturnType coeff(Index index) const + { + const RowCol row_col = index_remap(Rows == 1 ? 0 : index, + Rows == 1 ? index : 0); + return m_argImpl.coeff(row_col.first, row_col.second); + } +#if 0 + EIGEN_DEVICE_FUNC + template + inline PacketScalar packet(Index rowId, Index colId) const + { + const RowCol row_col = index_remap(rowId, colId); + return m_argImpl.template packet(row_col.first, row_col.second); + + } + + template + EIGEN_DEVICE_FUNC + inline void writePacket(Index rowId, Index colId, const PacketScalar& val) + { + const RowCol row_col = index_remap(rowId, colId); + m_argImpl.const_cast_derived().template writePacket + (row_col.first, row_col.second, val); + } + + template + EIGEN_DEVICE_FUNC + inline PacketScalar packet(Index index) const + { + const RowCol row_col = index_remap(RowsAtCompileTime == 1 ? 0 : index, + RowsAtCompileTime == 1 ? index : 0); + return m_argImpl.template packet(row_col.first, row_col.second); + } + + template + EIGEN_DEVICE_FUNC + inline void writePacket(Index index, const PacketScalar& val) + { + const RowCol row_col = index_remap(RowsAtCompileTime == 1 ? 0 : index, + RowsAtCompileTime == 1 ? index : 0); + return m_argImpl.template packet(row_col.first, row_col.second, val); + } +#endif +protected: + + evaluator m_argImpl; + const XprType& m_xpr; + +}; + +template +struct reshaped_evaluator +: mapbase_evaluator, + typename Reshaped::PlainObject> +{ + typedef Reshaped XprType; + typedef typename XprType::Scalar Scalar; + + EIGEN_DEVICE_FUNC explicit reshaped_evaluator(const XprType& xpr) + : mapbase_evaluator(xpr) + { + // TODO: for the 3.4 release, this should be turned to an internal assertion, but let's keep it as is for the beta lifetime + eigen_assert(((internal::UIntPtr(xpr.data()) % EIGEN_PLAIN_ENUM_MAX(1,evaluator::Alignment)) == 0) && "data is not aligned"); + } +}; + +} // end namespace internal + +} // end namespace Eigen + +#endif // EIGEN_RESHAPED_H diff --git a/thirdparty/eigen/Eigen/src/Core/ReturnByValue.h b/thirdparty/eigen/Eigen/src/Core/ReturnByValue.h index f635598d..4dad13ea 100644 --- a/thirdparty/eigen/Eigen/src/Core/ReturnByValue.h +++ b/thirdparty/eigen/Eigen/src/Core/ReturnByValue.h @@ -13,11 +13,6 @@ namespace Eigen { -/** \class ReturnByValue - * \ingroup Core_Module - * - */ - namespace internal { template @@ -38,17 +33,22 @@ struct traits > * So internal::nested always gives the plain return matrix type. * * FIXME: I don't understand why we need this specialization: isn't this taken care of by the EvalBeforeNestingBit ?? + * Answer: EvalBeforeNestingBit should be deprecated since we have the evaluators */ template -struct nested, n, PlainObject> +struct nested_eval, n, PlainObject> { typedef typename traits::ReturnType type; }; } // end namespace internal +/** \class ReturnByValue + * \ingroup Core_Module + * + */ template class ReturnByValue - : internal::no_assignment_operator, public internal::dense_xpr_base< ReturnByValue >::type + : public internal::dense_xpr_base< ReturnByValue >::type, internal::no_assignment_operator { public: typedef typename internal::traits::ReturnType ReturnType; @@ -57,10 +57,13 @@ template class ReturnByValue EIGEN_DENSE_PUBLIC_INTERFACE(ReturnByValue) template + EIGEN_DEVICE_FUNC inline void evalTo(Dest& dst) const { static_cast(this)->evalTo(dst); } - inline Index rows() const { return static_cast(this)->rows(); } - inline Index cols() const { return static_cast(this)->cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rows() const EIGEN_NOEXCEPT { return static_cast(this)->rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return static_cast(this)->cols(); } #ifndef EIGEN_PARSED_BY_DOXYGEN #define Unusable YOU_ARE_TRYING_TO_ACCESS_A_SINGLE_COEFFICIENT_IN_A_SPECIAL_EXPRESSION_WHERE_THAT_IS_NOT_ALLOWED_BECAUSE_THAT_WOULD_BE_INEFFICIENT @@ -72,27 +75,44 @@ template class ReturnByValue const Unusable& coeff(Index,Index) const { return *reinterpret_cast(this); } Unusable& coeffRef(Index) { return *reinterpret_cast(this); } Unusable& coeffRef(Index,Index) { return *reinterpret_cast(this); } - template Unusable& packet(Index) const; - template Unusable& packet(Index, Index) const; +#undef Unusable #endif }; template template -Derived& DenseBase::operator=(const ReturnByValue& other) +EIGEN_DEVICE_FUNC Derived& DenseBase::operator=(const ReturnByValue& other) { other.evalTo(derived()); return derived(); } +namespace internal { + +// Expression is evaluated in a temporary; default implementation of Assignment is bypassed so that +// when a ReturnByValue expression is assigned, the evaluator is not constructed. +// TODO: Finalize port to new regime; ReturnByValue should not exist in the expression world + template -template -Derived& DenseBase::lazyAssign(const ReturnByValue& other) +struct evaluator > + : public evaluator::ReturnType> { - other.evalTo(derived()); - return derived(); -} + typedef ReturnByValue XprType; + typedef typename internal::traits::ReturnType PlainObject; + typedef evaluator Base; + + EIGEN_DEVICE_FUNC explicit evaluator(const XprType& xpr) + : m_result(xpr.rows(), xpr.cols()) + { + ::new (static_cast(this)) Base(m_result); + xpr.evalTo(m_result); + } + +protected: + PlainObject m_result; +}; +} // end namespace internal } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/Reverse.h b/thirdparty/eigen/Eigen/src/Core/Reverse.h index 041f8222..28cdd76a 100644 --- a/thirdparty/eigen/Eigen/src/Core/Reverse.h +++ b/thirdparty/eigen/Eigen/src/Core/Reverse.h @@ -12,21 +12,7 @@ #ifndef EIGEN_REVERSE_H #define EIGEN_REVERSE_H -namespace Eigen { - -/** \class Reverse - * \ingroup Core_Module - * - * \brief Expression of the reverse of a vector or matrix - * - * \param MatrixType the type of the object of which we are taking the reverse - * - * This class represents an expression of the reverse of a vector. - * It is the return type of MatrixBase::reverse() and VectorwiseOp::reverse() - * and most of the time this is the only way it is used. - * - * \sa MatrixBase::reverse(), VectorwiseOp::reverse() - */ +namespace Eigen { namespace internal { @@ -37,36 +23,43 @@ struct traits > typedef typename MatrixType::Scalar Scalar; typedef typename traits::StorageKind StorageKind; typedef typename traits::XprKind XprKind; - typedef typename nested::type MatrixTypeNested; + typedef typename ref_selector::type MatrixTypeNested; typedef typename remove_reference::type _MatrixTypeNested; enum { RowsAtCompileTime = MatrixType::RowsAtCompileTime, ColsAtCompileTime = MatrixType::ColsAtCompileTime, MaxRowsAtCompileTime = MatrixType::MaxRowsAtCompileTime, MaxColsAtCompileTime = MatrixType::MaxColsAtCompileTime, - - // let's enable LinearAccess only with vectorization because of the product overhead - LinearAccess = ( (Direction==BothDirections) && (int(_MatrixTypeNested::Flags)&PacketAccessBit) ) - ? LinearAccessBit : 0, - - Flags = int(_MatrixTypeNested::Flags) & (HereditaryBits | LvalueBit | PacketAccessBit | LinearAccess), - - CoeffReadCost = _MatrixTypeNested::CoeffReadCost + Flags = _MatrixTypeNested::Flags & (RowMajorBit | LvalueBit) }; }; -template struct reverse_packet_cond +template struct reverse_packet_cond { - static inline PacketScalar run(const PacketScalar& x) { return preverse(x); } + static inline PacketType run(const PacketType& x) { return preverse(x); } }; -template struct reverse_packet_cond +template struct reverse_packet_cond { - static inline PacketScalar run(const PacketScalar& x) { return x; } + static inline PacketType run(const PacketType& x) { return x; } }; -} // end namespace internal +} // end namespace internal +/** \class Reverse + * \ingroup Core_Module + * + * \brief Expression of the reverse of a vector or matrix + * + * \tparam MatrixType the type of the object of which we are taking the reverse + * \tparam Direction defines the direction of the reverse operation, can be Vertical, Horizontal, or BothDirections + * + * This class represents an expression of the reverse of a vector. + * It is the return type of MatrixBase::reverse() and VectorwiseOp::reverse() + * and most of the time this is the only way it is used. + * + * \sa MatrixBase::reverse(), VectorwiseOp::reverse() + */ template class Reverse : public internal::dense_xpr_base< Reverse >::type { @@ -74,26 +67,9 @@ template class Reverse typedef typename internal::dense_xpr_base::type Base; EIGEN_DENSE_PUBLIC_INTERFACE(Reverse) + typedef typename internal::remove_all::type NestedExpression; using Base::IsRowMajor; - // The following two operators are provided to worarkound - // a MSVC 2013 issue. In theory, we could simply do: - // using Base::operator(); - // to make const version of operator() visible. - // Otheriwse, they would be hidden by the non-const versions defined in this file - - inline CoeffReturnType operator()(Index row, Index col) const - { - eigen_assert(row >= 0 && row < rows() && col >= 0 && col < cols()); - return coeff(row, col); - } - - inline CoeffReturnType operator()(Index index) const - { - eigen_assert(index >= 0 && index < m_matrix.size()); - return coeff(index); - } - protected: enum { PacketSize = internal::packet_traits::size, @@ -109,83 +85,22 @@ template class Reverse typedef internal::reverse_packet_cond reverse_packet; public: - inline Reverse(const MatrixType& matrix) : m_matrix(matrix) { } + EIGEN_DEVICE_FUNC explicit inline Reverse(const MatrixType& matrix) : m_matrix(matrix) { } EIGEN_INHERIT_ASSIGNMENT_OPERATORS(Reverse) - inline Index rows() const { return m_matrix.rows(); } - inline Index cols() const { return m_matrix.cols(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index rows() const EIGEN_NOEXCEPT { return m_matrix.rows(); } + EIGEN_DEVICE_FUNC EIGEN_CONSTEXPR + inline Index cols() const EIGEN_NOEXCEPT { return m_matrix.cols(); } - inline Index innerStride() const + EIGEN_DEVICE_FUNC inline Index innerStride() const { return -m_matrix.innerStride(); } - inline Scalar& operator()(Index row, Index col) - { - eigen_assert(row >= 0 && row < rows() && col >= 0 && col < cols()); - return coeffRef(row, col); - } - - inline Scalar& coeffRef(Index row, Index col) - { - return m_matrix.const_cast_derived().coeffRef(ReverseRow ? m_matrix.rows() - row - 1 : row, - ReverseCol ? m_matrix.cols() - col - 1 : col); - } - - inline CoeffReturnType coeff(Index row, Index col) const - { - return m_matrix.coeff(ReverseRow ? m_matrix.rows() - row - 1 : row, - ReverseCol ? m_matrix.cols() - col - 1 : col); - } - - inline CoeffReturnType coeff(Index index) const - { - return m_matrix.coeff(m_matrix.size() - index - 1); - } - - inline Scalar& coeffRef(Index index) - { - return m_matrix.const_cast_derived().coeffRef(m_matrix.size() - index - 1); - } - - inline Scalar& operator()(Index index) - { - eigen_assert(index >= 0 && index < m_matrix.size()); - return coeffRef(index); - } - - template - inline const PacketScalar packet(Index row, Index col) const - { - return reverse_packet::run(m_matrix.template packet( - ReverseRow ? m_matrix.rows() - row - OffsetRow : row, - ReverseCol ? m_matrix.cols() - col - OffsetCol : col)); - } - - template - inline void writePacket(Index row, Index col, const PacketScalar& x) - { - m_matrix.const_cast_derived().template writePacket( - ReverseRow ? m_matrix.rows() - row - OffsetRow : row, - ReverseCol ? m_matrix.cols() - col - OffsetCol : col, - reverse_packet::run(x)); - } - - template - inline const PacketScalar packet(Index index) const - { - return internal::preverse(m_matrix.template packet( m_matrix.size() - index - PacketSize )); - } - - template - inline void writePacket(Index index, const PacketScalar& x) - { - m_matrix.const_cast_derived().template writePacket(m_matrix.size() - index - PacketSize, internal::preverse(x)); - } - - const typename internal::remove_all::type& - nestedExpression() const + EIGEN_DEVICE_FUNC const typename internal::remove_all::type& + nestedExpression() const { return m_matrix; } @@ -201,36 +116,100 @@ template class Reverse * */ template -inline typename DenseBase::ReverseReturnType +EIGEN_DEVICE_FUNC inline typename DenseBase::ReverseReturnType DenseBase::reverse() { - return derived(); + return ReverseReturnType(derived()); } -/** This is the const version of reverse(). */ -template -inline const typename DenseBase::ConstReverseReturnType -DenseBase::reverse() const -{ - return derived(); -} + +//reverse const overload moved DenseBase.h due to a CUDA compiler bug /** This is the "in place" version of reverse: it reverses \c *this. * * In most cases it is probably better to simply use the reversed expression * of a matrix. However, when reversing the matrix data itself is really needed, * then this "in-place" version is probably the right choice because it provides - * the following additional features: + * the following additional benefits: * - less error prone: doing the same operation with .reverse() requires special care: * \code m = m.reverse().eval(); \endcode - * - this API allows to avoid creating a temporary (the current implementation creates a temporary, but that could be avoided using swap) + * - this API enables reverse operations without the need for a temporary * - it allows future optimizations (cache friendliness, etc.) * - * \sa reverse() */ + * \sa VectorwiseOp::reverseInPlace(), reverse() */ template -inline void DenseBase::reverseInPlace() +EIGEN_DEVICE_FUNC inline void DenseBase::reverseInPlace() +{ + if(cols()>rows()) + { + Index half = cols()/2; + leftCols(half).swap(rightCols(half).reverse()); + if((cols()%2)==1) + { + Index half2 = rows()/2; + col(half).head(half2).swap(col(half).tail(half2).reverse()); + } + } + else + { + Index half = rows()/2; + topRows(half).swap(bottomRows(half).reverse()); + if((rows()%2)==1) + { + Index half2 = cols()/2; + row(half).head(half2).swap(row(half).tail(half2).reverse()); + } + } +} + +namespace internal { + +template +struct vectorwise_reverse_inplace_impl; + +template<> +struct vectorwise_reverse_inplace_impl +{ + template + static void run(ExpressionType &xpr) + { + const int HalfAtCompileTime = ExpressionType::RowsAtCompileTime==Dynamic?Dynamic:ExpressionType::RowsAtCompileTime/2; + Index half = xpr.rows()/2; + xpr.topRows(fix(half)) + .swap(xpr.bottomRows(fix(half)).colwise().reverse()); + } +}; + +template<> +struct vectorwise_reverse_inplace_impl +{ + template + static void run(ExpressionType &xpr) + { + const int HalfAtCompileTime = ExpressionType::ColsAtCompileTime==Dynamic?Dynamic:ExpressionType::ColsAtCompileTime/2; + Index half = xpr.cols()/2; + xpr.leftCols(fix(half)) + .swap(xpr.rightCols(fix(half)).rowwise().reverse()); + } +}; + +} // end namespace internal + +/** This is the "in place" version of VectorwiseOp::reverse: it reverses each column or row of \c *this. + * + * In most cases it is probably better to simply use the reversed expression + * of a matrix. However, when reversing the matrix data itself is really needed, + * then this "in-place" version is probably the right choice because it provides + * the following additional benefits: + * - less error prone: doing the same operation with .reverse() requires special care: + * \code m = m.reverse().eval(); \endcode + * - this API enables reverse operations without the need for a temporary + * + * \sa DenseBase::reverseInPlace(), reverse() */ +template +EIGEN_DEVICE_FUNC void VectorwiseOp::reverseInPlace() { - derived() = derived().reverse().eval(); + internal::vectorwise_reverse_inplace_impl::run(m_matrix); } } // end namespace Eigen diff --git a/thirdparty/eigen/Eigen/src/Core/Select.h b/thirdparty/eigen/Eigen/src/Core/Select.h index 87993bbb..7c86bf87 100644 --- a/thirdparty/eigen/Eigen/src/Core/Select.h +++ b/thirdparty/eigen/Eigen/src/Core/Select.h @@ -10,7 +10,7 @@ #ifndef EIGEN_SELECT_H #define EIGEN_SELECT_H -namespace Eigen { +namespace Eigen { /** \class Select * \ingroup Core_Module @@ -43,23 +43,21 @@ struct traits > ColsAtCompileTime = ConditionMatrixType::ColsAtCompileTime, MaxRowsAtCompileTime = ConditionMatrixType::MaxRowsAtCompileTime, MaxColsAtCompileTime = ConditionMatrixType::MaxColsAtCompileTime, - Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & HereditaryBits, - CoeffReadCost = traits::type>::CoeffReadCost - + EIGEN_SIZE_MAX(traits::type>::CoeffReadCost, - traits::type>::CoeffReadCost) + Flags = (unsigned int)ThenMatrixType::Flags & ElseMatrixType::Flags & RowMajorBit }; }; } template -class Select : internal::no_assignment_operator, - public internal::dense_xpr_base< Select >::type +class Select : public internal::dense_xpr_base< Select >::type, + internal::no_assignment_operator { public: typedef typename internal::dense_xpr_base" << endl; + cerr << "available actions:" << endl; + for (auto it = available_actions.begin(); it != available_actions.end(); ++it) { + cerr << " " << (*it)->invokation_name() << endl; + } + cerr << "the input files should each contain an output of benchmark-blocking-sizes" << endl; + exit(1); +} + +int main(int argc, char* argv[]) +{ + cout.precision(default_precision); + cerr.precision(default_precision); + + vector> available_actions; + available_actions.emplace_back(new partition_action_t); + available_actions.emplace_back(new evaluate_defaults_action_t); + + vector input_filenames; + + action_t* action = nullptr; + + if (argc < 2) { + show_usage_and_exit(argc, argv, available_actions); + } + for (int i = 1; i < argc; i++) { + bool arg_handled = false; + // Step 1. Try to match action invocation names. + for (auto it = available_actions.begin(); it != available_actions.end(); ++it) { + if (!strcmp(argv[i], (*it)->invokation_name())) { + if (!action) { + action = it->get(); + arg_handled = true; + break; + } else { + cerr << "can't specify more than one action!" << endl; + show_usage_and_exit(argc, argv, available_actions); + } + } + } + if (arg_handled) { + continue; + } + // Step 2. Try to match option names. + if (argv[i][0] == '-') { + if (!strcmp(argv[i], "--only-cubic-sizes")) { + only_cubic_sizes = true; + arg_handled = true; + } + if (!strcmp(argv[i], "--dump-tables")) { + dump_tables = true; + arg_handled = true; + } + if (!arg_handled) { + cerr << "Unrecognized option: " << argv[i] << endl; + show_usage_and_exit(argc, argv, available_actions); + } + } + if (arg_handled) { + continue; + } + // Step 3. Default to interpreting args as input filenames. + input_filenames.emplace_back(argv[i]); + } + + if (dump_tables && only_cubic_sizes) { + cerr << "Incompatible options: --only-cubic-sizes and --dump-tables." << endl; + show_usage_and_exit(argc, argv, available_actions); + } + + if (!action) { + show_usage_and_exit(argc, argv, available_actions); + } + + action->run(input_filenames); +} diff --git a/thirdparty/eigen/bench/basicbenchmark.h b/thirdparty/eigen/bench/basicbenchmark.h index 3fdc3573..8059375b 100644 --- a/thirdparty/eigen/bench/basicbenchmark.h +++ b/thirdparty/eigen/bench/basicbenchmark.h @@ -16,13 +16,13 @@ void benchBasic_loop(const MatrixType& I, MatrixType& m, int iterations) { asm("#begin_bench_loop LazyEval"); if (MatrixType::SizeAtCompileTime!=Eigen::Dynamic) asm("#fixedsize"); - m = (I + 0.00005 * (m + m.lazy() * m)).eval(); + m = (I + 0.00005 * (m + m.lazyProduct(m))).eval(); } else if (Mode==OmpEval) { asm("#begin_bench_loop OmpEval"); if (MatrixType::SizeAtCompileTime!=Eigen::Dynamic) asm("#fixedsize"); - m = (I + 0.00005 * (m + m.lazy() * m)).evalOMP(); + m = (I + 0.00005 * (m + m.lazyProduct(m))).eval(); } else { diff --git a/thirdparty/eigen/bench/benchCholesky.cpp b/thirdparty/eigen/bench/benchCholesky.cpp index 42b3e128..0dc94e5b 100644 --- a/thirdparty/eigen/bench/benchCholesky.cpp +++ b/thirdparty/eigen/bench/benchCholesky.cpp @@ -1,5 +1,4 @@ - -// g++ -DNDEBUG -O3 -I.. benchLLT.cpp -o benchLLT && ./benchLLT +// g++ -DNDEBUG -O3 -I.. benchCholesky.cpp -o benchCholesky && ./benchCholesky // options: // -DBENCH_GSL -lgsl /usr/lib/libcblas.so.3 // -DEIGEN_DONT_VECTORIZE @@ -31,7 +30,7 @@ __attribute__ ((noinline)) void benchLLT(const MatrixType& m) int rows = m.rows(); int cols = m.cols(); - int cost = 0; + double cost = 0; for (int j=0; j0; ++i) + for (int i=0; dynsizes[i]>0; ++i) benchLLT(Matrix(dynsizes[i],dynsizes[i])); benchLLT(Matrix()); diff --git a/thirdparty/eigen/bench/bench_gemm.cpp b/thirdparty/eigen/bench/bench_gemm.cpp index 41ca8b3b..78ca1cd1 100644 --- a/thirdparty/eigen/bench/bench_gemm.cpp +++ b/thirdparty/eigen/bench/bench_gemm.cpp @@ -2,9 +2,18 @@ // g++-4.4 bench_gemm.cpp -I .. -O2 -DNDEBUG -lrt -fopenmp && OMP_NUM_THREADS=2 ./a.out // icpc bench_gemm.cpp -I .. -O3 -DNDEBUG -lrt -openmp && OMP_NUM_THREADS=2 ./a.out +// Compilation options: +// +// -DSCALAR=std::complex +// -DSCALARA=double or -DSCALARB=double +// -DHAVE_BLAS +// -DDECOUPLED +// + #include -#include #include +#include + using namespace std; using namespace Eigen; @@ -14,10 +23,30 @@ using namespace Eigen; #define SCALAR float #endif +#ifndef SCALARA +#define SCALARA SCALAR +#endif + +#ifndef SCALARB +#define SCALARB SCALAR +#endif + +#ifdef ROWMAJ_A +const int opt_A = RowMajor; +#else +const int opt_A = ColMajor; +#endif + +#ifdef ROWMAJ_B +const int opt_B = RowMajor; +#else +const int opt_B = ColMajor; +#endif + typedef SCALAR Scalar; typedef NumTraits::Real RealScalar; -typedef Matrix A; -typedef Matrix B; +typedef Matrix A; +typedef Matrix B; typedef Matrix C; typedef Matrix M; @@ -42,45 +71,61 @@ static char lower = 'L'; static char right = 'R'; static int intone = 1; -void blas_gemm(const MatrixXf& a, const MatrixXf& b, MatrixXf& c) +#ifdef ROWMAJ_A +const char transA = trans; +#else +const char transA = notrans; +#endif + +#ifdef ROWMAJ_B +const char transB = trans; +#else +const char transB = notrans; +#endif + +template +void blas_gemm(const A& a, const B& b, MatrixXf& c) { int M = c.rows(); int N = c.cols(); int K = a.cols(); - int lda = a.rows(); int ldb = b.rows(); int ldc = c.rows(); + int lda = a.outerStride(); int ldb = b.outerStride(); int ldc = c.rows(); - sgemm_(¬rans,¬rans,&M,&N,&K,&fone, + sgemm_(&transA,&transB,&M,&N,&K,&fone, const_cast(a.data()),&lda, const_cast(b.data()),&ldb,&fone, c.data(),&ldc); } -EIGEN_DONT_INLINE void blas_gemm(const MatrixXd& a, const MatrixXd& b, MatrixXd& c) +template +void blas_gemm(const A& a, const B& b, MatrixXd& c) { int M = c.rows(); int N = c.cols(); int K = a.cols(); - int lda = a.rows(); int ldb = b.rows(); int ldc = c.rows(); + int lda = a.outerStride(); int ldb = b.outerStride(); int ldc = c.rows(); - dgemm_(¬rans,¬rans,&M,&N,&K,&done, + dgemm_(&transA,&transB,&M,&N,&K,&done, const_cast(a.data()),&lda, const_cast(b.data()),&ldb,&done, c.data(),&ldc); } -void blas_gemm(const MatrixXcf& a, const MatrixXcf& b, MatrixXcf& c) +template +void blas_gemm(const A& a, const B& b, MatrixXcf& c) { int M = c.rows(); int N = c.cols(); int K = a.cols(); - int lda = a.rows(); int ldb = b.rows(); int ldc = c.rows(); + int lda = a.outerStride(); int ldb = b.outerStride(); int ldc = c.rows(); - cgemm_(¬rans,¬rans,&M,&N,&K,(float*)&cfone, + cgemm_(&transA,&transB,&M,&N,&K,(float*)&cfone, const_cast((const float*)a.data()),&lda, const_cast((const float*)b.data()),&ldb,(float*)&cfone, (float*)c.data(),&ldc); } -void blas_gemm(const MatrixXcd& a, const MatrixXcd& b, MatrixXcd& c) +template +void blas_gemm(const A& a, const B& b, MatrixXcd& c) { int M = c.rows(); int N = c.cols(); int K = a.cols(); - int lda = a.rows(); int ldb = b.rows(); int ldc = c.rows(); + int lda = a.outerStride(); int ldb = b.outerStride(); int ldc = c.rows(); - zgemm_(¬rans,¬rans,&M,&N,&K,(double*)&cdone, + zgemm_(&transA,&transB,&M,&N,&K,(double*)&cdone, const_cast((const double*)a.data()),&lda, const_cast((const double*)b.data()),&ldb,(double*)&cdone, (double*)c.data(),&ldc); @@ -96,6 +141,7 @@ void matlab_cplx_cplx(const M& ar, const M& ai, const M& br, const M& bi, M& cr, cr.noalias() -= ai * bi; ci.noalias() += ar * bi; ci.noalias() += ai * br; + // [cr ci] += [ar ai] * br + [-ai ar] * bi } void matlab_real_cplx(const M& a, const M& br, const M& bi, M& cr, M& ci) @@ -110,10 +156,12 @@ void matlab_cplx_real(const M& ar, const M& ai, const M& b, M& cr, M& ci) ci.noalias() += ai * b; } + + template EIGEN_DONT_INLINE void gemm(const A& a, const B& b, C& c) { - c.noalias() += a * b; + c.noalias() += a * b; } int main(int argc, char ** argv) @@ -129,35 +177,69 @@ int main(int argc, char ** argv) int tries = 2; // number of tries, we keep the best int s = 2048; - int cache_size = -1; + int m = s; + int n = s; + int p = s; + int cache_size1=-1, cache_size2=l2, cache_size3 = 0; bool need_help = false; - for (int i=1; i c t p\n"; + std::cout << argv[0] << " -s -c -t -p \n"; + std::cout << " : size\n"; + std::cout << " : rows columns depth\n"; return 1; } - if(cache_size>0) - setCpuCacheSizes(cache_size,96*cache_size); - - int m = s; - int n = s; - int p = s; +#if EIGEN_VERSION_AT_LEAST(3,2,90) + if(cache_size1>0) + setCpuCacheSizes(cache_size1,cache_size2,cache_size3); +#endif + A a(m,p); a.setRandom(); B b(p,n); b.setRandom(); C c(m,n); c.setOnes(); @@ -166,12 +248,13 @@ int main(int argc, char ** argv) std::cout << "Matrix sizes = " << m << "x" << p << " * " << p << "x" << n << "\n"; std::ptrdiff_t mc(m), nc(n), kc(p); internal::computeProductBlockingSizes(kc, mc, nc); - std::cout << "blocking size (mc x kc) = " << mc << " x " << kc << "\n"; + std::cout << "blocking size (mc x kc) = " << mc << " x " << kc << " x " << nc << "\n"; C r = c; // check the parallel product is correct #if defined EIGEN_HAS_OPENMP + Eigen::initParallel(); int procs = omp_get_max_threads(); if(procs>1) { @@ -188,11 +271,20 @@ int main(int argc, char ** argv) #elif defined HAVE_BLAS blas_gemm(a,b,r); c.noalias() += a * b; - if(!r.isApprox(c)) std::cerr << "Warning, your product is crap!\n\n"; + if(!r.isApprox(c)) { + std::cout << (r - c).norm()/r.norm() << "\n"; + std::cerr << "Warning, your product is crap!\n\n"; + } #else - gemm(a,b,c); - r.noalias() += a.cast() * b.cast(); - if(!r.isApprox(c)) std::cerr << "Warning, your product is crap!\n\n"; + if(1.*m*n*p<2000.*2000*2000) + { + gemm(a,b,c); + r.noalias() += a.cast() .lazyProduct( b.cast() ); + if(!r.isApprox(c)) { + std::cout << (r - c).norm()/r.norm() << "\n"; + std::cerr << "Warning, your product is crap!\n\n"; + } + } #endif #ifdef HAVE_BLAS @@ -203,6 +295,9 @@ int main(int argc, char ** argv) std::cout << "blas real " << tblas.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tblas.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tblas.total(REAL_TIMER) << "s)\n"; #endif + // warm start + if(b.norm()+a.norm()==123.554) std::cout << "\n"; + BenchTimer tmt; c = rc; BENCH(tmt, tries, rep, gemm(a,b,c)); @@ -214,7 +309,7 @@ int main(int argc, char ** argv) { BenchTimer tmono; omp_set_num_threads(1); - Eigen::internal::setNbThreads(1); + Eigen::setNbThreads(1); c = rc; BENCH(tmono, tries, rep, gemm(a,b,c)); std::cout << "eigen mono cpu " << tmono.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmono.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tmono.total(CPU_TIMER) << "s)\n"; @@ -223,6 +318,15 @@ int main(int argc, char ** argv) } #endif + if(1.*m*n*p<30*30*30) + { + BenchTimer tmt; + c = rc; + BENCH(tmt, tries, rep, c.noalias()+=a.lazyProduct(b)); + std::cout << "lazy cpu " << tmt.best(CPU_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmt.best(CPU_TIMER))*1e-9 << " GFLOPS \t(" << tmt.total(CPU_TIMER) << "s)\n"; + std::cout << "lazy real " << tmt.best(REAL_TIMER)/rep << "s \t" << (double(m)*n*p*rep*2/tmt.best(REAL_TIMER))*1e-9 << " GFLOPS \t(" << tmt.total(REAL_TIMER) << "s)\n"; + } + #ifdef DECOUPLED if((NumTraits::IsComplex) && (NumTraits::IsComplex)) { diff --git a/thirdparty/eigen/bench/bench_move_semantics.cpp b/thirdparty/eigen/bench/bench_move_semantics.cpp new file mode 100644 index 00000000..323d8041 --- /dev/null +++ b/thirdparty/eigen/bench/bench_move_semantics.cpp @@ -0,0 +1,57 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2020 Sebastien Boisvert +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "BenchTimer.h" +#include "../test/MovableScalar.h" + +#include + +#include +#include + +template +void copy_matrix(MatrixType& m) +{ + MatrixType tmp(m); + m = tmp; +} + +template +void move_matrix(MatrixType&& m) +{ + MatrixType tmp(std::move(m)); + m = std::move(tmp); +} + +template +void bench(const std::string& label) +{ + using MatrixType = Eigen::Matrix,1,10>; + Eigen::BenchTimer t; + + int tries = 10; + int rep = 1000000; + + MatrixType data = MatrixType::Random().eval(); + MatrixType dest; + + BENCH(t, tries, rep, copy_matrix(data)); + std::cout << label << " copy semantics: " << 1e3*t.best(Eigen::CPU_TIMER) << " ms" << std::endl; + + BENCH(t, tries, rep, move_matrix(std::move(data))); + std::cout << label << " move semantics: " << 1e3*t.best(Eigen::CPU_TIMER) << " ms" << std::endl; +} + +int main() +{ + bench("float"); + bench("double"); + return 0; +} + diff --git a/thirdparty/eigen/bench/bench_norm.cpp b/thirdparty/eigen/bench/bench_norm.cpp index 806db292..592f25d6 100644 --- a/thirdparty/eigen/bench/bench_norm.cpp +++ b/thirdparty/eigen/bench/bench_norm.cpp @@ -6,19 +6,25 @@ using namespace Eigen; using namespace std; template -EIGEN_DONT_INLINE typename T::Scalar sqsumNorm(const T& v) +EIGEN_DONT_INLINE typename T::Scalar sqsumNorm(T& v) { return v.norm(); } template -EIGEN_DONT_INLINE typename T::Scalar hypotNorm(const T& v) +EIGEN_DONT_INLINE typename T::Scalar stableNorm(T& v) +{ + return v.stableNorm(); +} + +template +EIGEN_DONT_INLINE typename T::Scalar hypotNorm(T& v) { return v.hypotNorm(); } template -EIGEN_DONT_INLINE typename T::Scalar blueNorm(const T& v) +EIGEN_DONT_INLINE typename T::Scalar blueNorm(T& v) { return v.blueNorm(); } @@ -32,25 +38,25 @@ EIGEN_DONT_INLINE typename T::Scalar lapackNorm(T& v) Scalar ssq = 1; for (int i=0;i= ax) { - ssq += internal::abs2(ax/scale); + ssq += numext::abs2(ax/scale); } else { - ssq = Scalar(1) + ssq * internal::abs2(scale/ax); + ssq = Scalar(1) + ssq * numext::abs2(scale/ax); scale = ax; } } - return scale * internal::sqrt(ssq); + return scale * std::sqrt(ssq); } template EIGEN_DONT_INLINE typename T::Scalar twopassNorm(T& v) { typedef typename T::Scalar Scalar; - Scalar s = v.cwise().abs().maxCoeff(); + Scalar s = v.array().abs().maxCoeff(); return s*(v/s).norm(); } @@ -73,16 +79,20 @@ EIGEN_DONT_INLINE typename T::Scalar divacNorm(T& v) v(i) = v(2*i) + v(2*i+1); n = n/2; } - return internal::sqrt(v(0)); + return std::sqrt(v(0)); } +namespace Eigen { +namespace internal { #ifdef EIGEN_VECTORIZE -Packet4f internal::plt(const Packet4f& a, Packet4f& b) { return _mm_cmplt_ps(a,b); } -Packet2d internal::plt(const Packet2d& a, Packet2d& b) { return _mm_cmplt_pd(a,b); } +Packet4f plt(const Packet4f& a, Packet4f& b) { return _mm_cmplt_ps(a,b); } +Packet2d plt(const Packet2d& a, Packet2d& b) { return _mm_cmplt_pd(a,b); } -Packet4f internal::pandnot(const Packet4f& a, Packet4f& b) { return _mm_andnot_ps(a,b); } -Packet2d internal::pandnot(const Packet2d& a, Packet2d& b) { return _mm_andnot_pd(a,b); } +Packet4f pandnot(const Packet4f& a, Packet4f& b) { return _mm_andnot_ps(a,b); } +Packet2d pandnot(const Packet2d& a, Packet2d& b) { return _mm_andnot_pd(a,b); } #endif +} +} template EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v) @@ -101,12 +111,12 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v) int nbig, ibeta, it, iemin, iemax, iexp; Scalar abig, eps; - nbig = std::numeric_limits::max(); // largest integer - ibeta = std::numeric_limits::radix; //NumTraits::Base; // base for floating-point numbers - it = std::numeric_limits::digits; //NumTraits::Mantissa; // number of base-beta digits in mantissa - iemin = std::numeric_limits::min_exponent; // minimum exponent - iemax = std::numeric_limits::max_exponent; // maximum exponent - rbig = std::numeric_limits::max(); // largest floating-point number + nbig = NumTraits::highest(); // largest integer + ibeta = std::numeric_limits::radix; // NumTraits::Base; // base for floating-point numbers + it = NumTraits::digits(); // NumTraits::Mantissa; // number of base-beta digits in mantissa + iemin = NumTraits::min_exponent(); // minimum exponent + iemax = NumTraits::max_exponent(); // maximum exponent + rbig = NumTraits::highest(); // largest floating-point number // Check the basic machine-dependent constants. if(iemin > 1 - 2*it || 1+it>iemax || (it==2 && ibeta<5) @@ -124,9 +134,9 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v) iexp = - ((iemax+it)/2); s2m = std::pow(ibeta,iexp); // scaling factor for upper range - overfl = rbig*s2m; // overfow boundary for abig + overfl = rbig*s2m; // overflow boundary for abig eps = std::pow(ibeta, 1-it); - relerr = internal::sqrt(eps); // tolerance for neglecting asml + relerr = std::sqrt(eps); // tolerance for neglecting asml abig = 1.0/eps - 1.0; if (Scalar(nbig)>abig) nmax = abig; // largest safe n else nmax = nbig; @@ -134,13 +144,13 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v) typedef typename internal::packet_traits::type Packet; const int ps = internal::packet_traits::size; - Packet pasml = internal::pset1(Scalar(0)); - Packet pamed = internal::pset1(Scalar(0)); - Packet pabig = internal::pset1(Scalar(0)); - Packet ps2m = internal::pset1(s2m); - Packet ps1m = internal::pset1(s1m); - Packet pb2 = internal::pset1(b2); - Packet pb1 = internal::pset1(b1); + Packet pasml = internal::pset1(Scalar(0)); + Packet pamed = internal::pset1(Scalar(0)); + Packet pabig = internal::pset1(Scalar(0)); + Packet ps2m = internal::pset1(s2m); + Packet ps1m = internal::pset1(s1m); + Packet pb2 = internal::pset1(b2); + Packet pb1 = internal::pset1(b1); for(int j=0; j(j)); @@ -170,7 +180,7 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v) Scalar amed = internal::predux(pamed); if(abig > Scalar(0)) { - abig = internal::sqrt(abig); + abig = std::sqrt(abig); if(abig > overfl) { eigen_assert(false && "overflow"); @@ -179,7 +189,7 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v) if(amed > Scalar(0)) { abig = abig/s2m; - amed = internal::sqrt(amed); + amed = std::sqrt(amed); } else { @@ -191,55 +201,56 @@ EIGEN_DONT_INLINE typename T::Scalar pblueNorm(const T& v) { if (amed > Scalar(0)) { - abig = internal::sqrt(amed); - amed = internal::sqrt(asml) / s1m; + abig = std::sqrt(amed); + amed = std::sqrt(asml) / s1m; } else { - return internal::sqrt(asml)/s1m; + return std::sqrt(asml)/s1m; } } else { - return internal::sqrt(amed); + return std::sqrt(amed); } asml = std::min(abig, amed); abig = std::max(abig, amed); if(asml <= abig*relerr) return abig; else - return abig * internal::sqrt(Scalar(1) + internal::abs2(asml/abig)); + return abig * std::sqrt(Scalar(1) + numext::abs2(asml/abig)); #endif } #define BENCH_PERF(NRM) { \ + float af = 0; double ad = 0; std::complex ac = 0; \ Eigen::BenchTimer tf, td, tcf; tf.reset(); td.reset(); tcf.reset();\ for (int k=0; k()); - double yd = based * internal::abs(internal::random()); + double yf = basef * std::abs(internal::random()); + double yd = based * std::abs(internal::random()); VectorXf vf = VectorXf::Ones(s) * yf; VectorXd vd = VectorXd::Ones(s) * yd; - std::cout << "reference\t" << internal::sqrt(double(s))*yf << "\t" << internal::sqrt(double(s))*yd << "\n"; + std::cout << "reference\t" << std::sqrt(double(s))*yf << "\t" << std::sqrt(double(s))*yd << "\n"; std::cout << "sqsumNorm\t" << sqsumNorm(vf) << "\t" << sqsumNorm(vd) << "\n"; std::cout << "hypotNorm\t" << hypotNorm(vf) << "\t" << hypotNorm(vd) << "\n"; std::cout << "blueNorm\t" << blueNorm(vf) << "\t" << blueNorm(vd) << "\n"; @@ -255,8 +266,8 @@ void check_accuracy_var(int ef0, int ef1, int ed0, int ed1, int s) VectorXd vd(s); for (int i=0; i()) * std::pow(double(10), internal::random(ef0,ef1)); - vd[i] = internal::abs(internal::random()) * std::pow(double(10), internal::random(ed0,ed1)); + vf[i] = std::abs(internal::random()) * std::pow(double(10), internal::random(ef0,ef1)); + vd[i] = std::abs(internal::random()) * std::pow(double(10), internal::random(ed0,ed1)); } //std::cout << "reference\t" << internal::sqrt(double(s))*yf << "\t" << internal::sqrt(double(s))*yd << "\n"; @@ -312,34 +323,38 @@ int main(int argc, char** argv) std::cout << "\n"; } + y = 1; std::cout.precision(4); - std::cerr << "Performance (out of cache):\n"; + int s1 = 1024*1024*32; + std::cerr << "Performance (out of cache, " << s1 << "):\n"; { int iters = 1; - VectorXf vf = VectorXf::Random(1024*1024*32) * y; - VectorXd vd = VectorXd::Random(1024*1024*32) * y; - VectorXcf vcf = VectorXcf::Random(1024*1024*32) * y; + VectorXf vf = VectorXf::Random(s1) * y; + VectorXd vd = VectorXd::Random(s1) * y; + VectorXcf vcf = VectorXcf::Random(s1) * y; BENCH_PERF(sqsumNorm); + BENCH_PERF(stableNorm); BENCH_PERF(blueNorm); -// BENCH_PERF(pblueNorm); -// BENCH_PERF(lapackNorm); -// BENCH_PERF(hypotNorm); -// BENCH_PERF(twopassNorm); + BENCH_PERF(pblueNorm); + BENCH_PERF(lapackNorm); + BENCH_PERF(hypotNorm); + BENCH_PERF(twopassNorm); BENCH_PERF(bl2passNorm); } - std::cerr << "\nPerformance (in cache):\n"; + std::cerr << "\nPerformance (in cache, " << 512 << "):\n"; { int iters = 100000; VectorXf vf = VectorXf::Random(512) * y; VectorXd vd = VectorXd::Random(512) * y; VectorXcf vcf = VectorXcf::Random(512) * y; BENCH_PERF(sqsumNorm); + BENCH_PERF(stableNorm); BENCH_PERF(blueNorm); -// BENCH_PERF(pblueNorm); -// BENCH_PERF(lapackNorm); -// BENCH_PERF(hypotNorm); -// BENCH_PERF(twopassNorm); + BENCH_PERF(pblueNorm); + BENCH_PERF(lapackNorm); + BENCH_PERF(hypotNorm); + BENCH_PERF(twopassNorm); BENCH_PERF(bl2passNorm); } } diff --git a/thirdparty/eigen/bench/benchmark-blocking-sizes.cpp b/thirdparty/eigen/bench/benchmark-blocking-sizes.cpp new file mode 100644 index 00000000..827be288 --- /dev/null +++ b/thirdparty/eigen/bench/benchmark-blocking-sizes.cpp @@ -0,0 +1,677 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2015 Benoit Jacob +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include +#include +#include +#include +#include +#include +#include + +bool eigen_use_specific_block_size; +int eigen_block_size_k, eigen_block_size_m, eigen_block_size_n; +#define EIGEN_TEST_SPECIFIC_BLOCKING_SIZES eigen_use_specific_block_size +#define EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_K eigen_block_size_k +#define EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_M eigen_block_size_m +#define EIGEN_TEST_SPECIFIC_BLOCKING_SIZE_N eigen_block_size_n +#include + +#include + +using namespace Eigen; +using namespace std; + +static BenchTimer timer; + +// how many times we repeat each measurement. +// measurements are randomly shuffled - we're not doing +// all N identical measurements in a row. +const int measurement_repetitions = 3; + +// Timings below this value are too short to be accurate, +// we'll repeat measurements with more iterations until +// we get a timing above that threshold. +const float min_accurate_time = 1e-2f; + +// See --min-working-set-size command line parameter. +size_t min_working_set_size = 0; + +float max_clock_speed = 0.0f; + +// range of sizes that we will benchmark (in all 3 K,M,N dimensions) +const size_t maxsize = 2048; +const size_t minsize = 16; + +typedef MatrixXf MatrixType; +typedef MatrixType::Scalar Scalar; +typedef internal::packet_traits::type Packet; + +static_assert((maxsize & (maxsize - 1)) == 0, "maxsize must be a power of two"); +static_assert((minsize & (minsize - 1)) == 0, "minsize must be a power of two"); +static_assert(maxsize > minsize, "maxsize must be larger than minsize"); +static_assert(maxsize < (minsize << 16), "maxsize must be less than (minsize<<16)"); + +// just a helper to store a triple of K,M,N sizes for matrix product +struct size_triple_t +{ + size_t k, m, n; + size_triple_t() : k(0), m(0), n(0) {} + size_triple_t(size_t _k, size_t _m, size_t _n) : k(_k), m(_m), n(_n) {} + size_triple_t(const size_triple_t& o) : k(o.k), m(o.m), n(o.n) {} + size_triple_t(uint16_t compact) + { + k = 1 << ((compact & 0xf00) >> 8); + m = 1 << ((compact & 0x0f0) >> 4); + n = 1 << ((compact & 0x00f) >> 0); + } +}; + +uint8_t log2_pot(size_t x) { + size_t l = 0; + while (x >>= 1) l++; + return l; +} + +// Convert between size tripes and a compact form fitting in 12 bits +// where each size, which must be a POT, is encoded as its log2, on 4 bits +// so the largest representable size is 2^15 == 32k ... big enough. +uint16_t compact_size_triple(size_t k, size_t m, size_t n) +{ + return (log2_pot(k) << 8) | (log2_pot(m) << 4) | log2_pot(n); +} + +uint16_t compact_size_triple(const size_triple_t& t) +{ + return compact_size_triple(t.k, t.m, t.n); +} + +// A single benchmark. Initially only contains benchmark params. +// Then call run(), which stores the result in the gflops field. +struct benchmark_t +{ + uint16_t compact_product_size; + uint16_t compact_block_size; + bool use_default_block_size; + float gflops; + benchmark_t() + : compact_product_size(0) + , compact_block_size(0) + , use_default_block_size(false) + , gflops(0) + { + } + benchmark_t(size_t pk, size_t pm, size_t pn, + size_t bk, size_t bm, size_t bn) + : compact_product_size(compact_size_triple(pk, pm, pn)) + , compact_block_size(compact_size_triple(bk, bm, bn)) + , use_default_block_size(false) + , gflops(0) + {} + benchmark_t(size_t pk, size_t pm, size_t pn) + : compact_product_size(compact_size_triple(pk, pm, pn)) + , compact_block_size(0) + , use_default_block_size(true) + , gflops(0) + {} + + void run(); +}; + +ostream& operator<<(ostream& s, const benchmark_t& b) +{ + s << hex << b.compact_product_size << dec; + if (b.use_default_block_size) { + size_triple_t t(b.compact_product_size); + Index k = t.k, m = t.m, n = t.n; + internal::computeProductBlockingSizes(k, m, n); + s << " default(" << k << ", " << m << ", " << n << ")"; + } else { + s << " " << hex << b.compact_block_size << dec; + } + s << " " << b.gflops; + return s; +} + +// We sort first by increasing benchmark parameters, +// then by decreasing performance. +bool operator<(const benchmark_t& b1, const benchmark_t& b2) +{ + return b1.compact_product_size < b2.compact_product_size || + (b1.compact_product_size == b2.compact_product_size && ( + (b1.compact_block_size < b2.compact_block_size || ( + b1.compact_block_size == b2.compact_block_size && + b1.gflops > b2.gflops)))); +} + +void benchmark_t::run() +{ + size_triple_t productsizes(compact_product_size); + + if (use_default_block_size) { + eigen_use_specific_block_size = false; + } else { + // feed eigen with our custom blocking params + eigen_use_specific_block_size = true; + size_triple_t blocksizes(compact_block_size); + eigen_block_size_k = blocksizes.k; + eigen_block_size_m = blocksizes.m; + eigen_block_size_n = blocksizes.n; + } + + // set up the matrix pool + + const size_t combined_three_matrices_sizes = + sizeof(Scalar) * + (productsizes.k * productsizes.m + + productsizes.k * productsizes.n + + productsizes.m * productsizes.n); + + // 64 M is large enough that nobody has a cache bigger than that, + // while still being small enough that everybody has this much RAM, + // so conveniently we don't need to special-case platforms here. + const size_t unlikely_large_cache_size = 64 << 20; + + const size_t working_set_size = + min_working_set_size ? min_working_set_size : unlikely_large_cache_size; + + const size_t matrix_pool_size = + 1 + working_set_size / combined_three_matrices_sizes; + + MatrixType *lhs = new MatrixType[matrix_pool_size]; + MatrixType *rhs = new MatrixType[matrix_pool_size]; + MatrixType *dst = new MatrixType[matrix_pool_size]; + + for (size_t i = 0; i < matrix_pool_size; i++) { + lhs[i] = MatrixType::Zero(productsizes.m, productsizes.k); + rhs[i] = MatrixType::Zero(productsizes.k, productsizes.n); + dst[i] = MatrixType::Zero(productsizes.m, productsizes.n); + } + + // main benchmark loop + + int iters_at_a_time = 1; + float time_per_iter = 0.0f; + size_t matrix_index = 0; + while (true) { + + double starttime = timer.getCpuTime(); + for (int i = 0; i < iters_at_a_time; i++) { + dst[matrix_index].noalias() = lhs[matrix_index] * rhs[matrix_index]; + matrix_index++; + if (matrix_index == matrix_pool_size) { + matrix_index = 0; + } + } + double endtime = timer.getCpuTime(); + + const float timing = float(endtime - starttime); + + if (timing >= min_accurate_time) { + time_per_iter = timing / iters_at_a_time; + break; + } + + iters_at_a_time *= 2; + } + + delete[] lhs; + delete[] rhs; + delete[] dst; + + gflops = 2e-9 * productsizes.k * productsizes.m * productsizes.n / time_per_iter; +} + +void print_cpuinfo() +{ +#ifdef __linux__ + cout << "contents of /proc/cpuinfo:" << endl; + string line; + ifstream cpuinfo("/proc/cpuinfo"); + if (cpuinfo.is_open()) { + while (getline(cpuinfo, line)) { + cout << line << endl; + } + cpuinfo.close(); + } + cout << endl; +#elif defined __APPLE__ + cout << "output of sysctl hw:" << endl; + system("sysctl hw"); + cout << endl; +#endif +} + +template +string type_name() +{ + return "unknown"; +} + +template<> +string type_name() +{ + return "float"; +} + +template<> +string type_name() +{ + return "double"; +} + +struct action_t +{ + virtual const char* invokation_name() const { abort(); return nullptr; } + virtual void run() const { abort(); } + virtual ~action_t() {} +}; + +void show_usage_and_exit(int /*argc*/, char* argv[], + const vector>& available_actions) +{ + cerr << "usage: " << argv[0] << " [options...]" << endl << endl; + cerr << "available actions:" << endl << endl; + for (auto it = available_actions.begin(); it != available_actions.end(); ++it) { + cerr << " " << (*it)->invokation_name() << endl; + } + cerr << endl; + cerr << "options:" << endl << endl; + cerr << " --min-working-set-size=N:" << endl; + cerr << " Set the minimum working set size to N bytes." << endl; + cerr << " This is rounded up as needed to a multiple of matrix size." << endl; + cerr << " A larger working set lowers the chance of a warm cache." << endl; + cerr << " The default value 0 means use a large enough working" << endl; + cerr << " set to likely outsize caches." << endl; + cerr << " A value of 1 (that is, 1 byte) would mean don't do anything to" << endl; + cerr << " avoid warm caches." << endl; + exit(1); +} + +float measure_clock_speed() +{ + cerr << "Measuring clock speed... \r" << flush; + + vector all_gflops; + for (int i = 0; i < 8; i++) { + benchmark_t b(1024, 1024, 1024); + b.run(); + all_gflops.push_back(b.gflops); + } + + sort(all_gflops.begin(), all_gflops.end()); + float stable_estimate = all_gflops[2] + all_gflops[3] + all_gflops[4] + all_gflops[5]; + + // multiply by an arbitrary constant to discourage trying doing anything with the + // returned values besides just comparing them with each other. + float result = stable_estimate * 123.456f; + + return result; +} + +struct human_duration_t +{ + int seconds; + human_duration_t(int s) : seconds(s) {} +}; + +ostream& operator<<(ostream& s, const human_duration_t& d) +{ + int remainder = d.seconds; + if (remainder > 3600) { + int hours = remainder / 3600; + s << hours << " h "; + remainder -= hours * 3600; + } + if (remainder > 60) { + int minutes = remainder / 60; + s << minutes << " min "; + remainder -= minutes * 60; + } + if (d.seconds < 600) { + s << remainder << " s"; + } + return s; +} + +const char session_filename[] = "/data/local/tmp/benchmark-blocking-sizes-session.data"; + +void serialize_benchmarks(const char* filename, const vector& benchmarks, size_t first_benchmark_to_run) +{ + FILE* file = fopen(filename, "w"); + if (!file) { + cerr << "Could not open file " << filename << " for writing." << endl; + cerr << "Do you have write permissions on the current working directory?" << endl; + exit(1); + } + size_t benchmarks_vector_size = benchmarks.size(); + fwrite(&max_clock_speed, sizeof(max_clock_speed), 1, file); + fwrite(&benchmarks_vector_size, sizeof(benchmarks_vector_size), 1, file); + fwrite(&first_benchmark_to_run, sizeof(first_benchmark_to_run), 1, file); + fwrite(benchmarks.data(), sizeof(benchmark_t), benchmarks.size(), file); + fclose(file); +} + +bool deserialize_benchmarks(const char* filename, vector& benchmarks, size_t& first_benchmark_to_run) +{ + FILE* file = fopen(filename, "r"); + if (!file) { + return false; + } + if (1 != fread(&max_clock_speed, sizeof(max_clock_speed), 1, file)) { + return false; + } + size_t benchmarks_vector_size = 0; + if (1 != fread(&benchmarks_vector_size, sizeof(benchmarks_vector_size), 1, file)) { + return false; + } + if (1 != fread(&first_benchmark_to_run, sizeof(first_benchmark_to_run), 1, file)) { + return false; + } + benchmarks.resize(benchmarks_vector_size); + if (benchmarks.size() != fread(benchmarks.data(), sizeof(benchmark_t), benchmarks.size(), file)) { + return false; + } + unlink(filename); + return true; +} + +void try_run_some_benchmarks( + vector& benchmarks, + double time_start, + size_t& first_benchmark_to_run) +{ + if (first_benchmark_to_run == benchmarks.size()) { + return; + } + + double time_last_progress_update = 0; + double time_last_clock_speed_measurement = 0; + double time_now = 0; + + size_t benchmark_index = first_benchmark_to_run; + + while (true) { + float ratio_done = float(benchmark_index) / benchmarks.size(); + time_now = timer.getRealTime(); + + // We check clock speed every minute and at the end. + if (benchmark_index == benchmarks.size() || + time_now > time_last_clock_speed_measurement + 60.0f) + { + time_last_clock_speed_measurement = time_now; + + // Ensure that clock speed is as expected + float current_clock_speed = measure_clock_speed(); + + // The tolerance needs to be smaller than the relative difference between + // clock speeds that a device could operate under. + // It seems unlikely that a device would be throttling clock speeds by + // amounts smaller than 2%. + // With a value of 1%, I was getting within noise on a Sandy Bridge. + const float clock_speed_tolerance = 0.02f; + + if (current_clock_speed > (1 + clock_speed_tolerance) * max_clock_speed) { + // Clock speed is now higher than we previously measured. + // Either our initial measurement was inaccurate, which won't happen + // too many times as we are keeping the best clock speed value and + // and allowing some tolerance; or something really weird happened, + // which invalidates all benchmark results collected so far. + // Either way, we better restart all over again now. + if (benchmark_index) { + cerr << "Restarting at " << 100.0f * ratio_done + << " % because clock speed increased. " << endl; + } + max_clock_speed = current_clock_speed; + first_benchmark_to_run = 0; + return; + } + + bool rerun_last_tests = false; + + if (current_clock_speed < (1 - clock_speed_tolerance) * max_clock_speed) { + cerr << "Measurements completed so far: " + << 100.0f * ratio_done + << " % " << endl; + cerr << "Clock speed seems to be only " + << current_clock_speed/max_clock_speed + << " times what it used to be." << endl; + + unsigned int seconds_to_sleep_if_lower_clock_speed = 1; + + while (current_clock_speed < (1 - clock_speed_tolerance) * max_clock_speed) { + if (seconds_to_sleep_if_lower_clock_speed > 32) { + cerr << "Sleeping longer probably won't make a difference." << endl; + cerr << "Serializing benchmarks to " << session_filename << endl; + serialize_benchmarks(session_filename, benchmarks, first_benchmark_to_run); + cerr << "Now restart this benchmark, and it should pick up where we left." << endl; + exit(2); + } + rerun_last_tests = true; + cerr << "Sleeping " + << seconds_to_sleep_if_lower_clock_speed + << " s... \r" << endl; + sleep(seconds_to_sleep_if_lower_clock_speed); + current_clock_speed = measure_clock_speed(); + seconds_to_sleep_if_lower_clock_speed *= 2; + } + } + + if (rerun_last_tests) { + cerr << "Redoing the last " + << 100.0f * float(benchmark_index - first_benchmark_to_run) / benchmarks.size() + << " % because clock speed had been low. " << endl; + return; + } + + // nothing wrong with the clock speed so far, so there won't be a need to rerun + // benchmarks run so far in case we later encounter a lower clock speed. + first_benchmark_to_run = benchmark_index; + } + + if (benchmark_index == benchmarks.size()) { + // We're done! + first_benchmark_to_run = benchmarks.size(); + // Erase progress info + cerr << " " << endl; + return; + } + + // Display progress info on stderr + if (time_now > time_last_progress_update + 1.0f) { + time_last_progress_update = time_now; + cerr << "Measurements... " << 100.0f * ratio_done + << " %, ETA " + << human_duration_t(float(time_now - time_start) * (1.0f - ratio_done) / ratio_done) + << " \r" << flush; + } + + // This is where we actually run a benchmark! + benchmarks[benchmark_index].run(); + benchmark_index++; + } +} + +void run_benchmarks(vector& benchmarks) +{ + size_t first_benchmark_to_run; + vector deserialized_benchmarks; + bool use_deserialized_benchmarks = false; + if (deserialize_benchmarks(session_filename, deserialized_benchmarks, first_benchmark_to_run)) { + cerr << "Found serialized session with " + << 100.0f * first_benchmark_to_run / deserialized_benchmarks.size() + << " % already done" << endl; + if (deserialized_benchmarks.size() == benchmarks.size() && + first_benchmark_to_run > 0 && + first_benchmark_to_run < benchmarks.size()) + { + use_deserialized_benchmarks = true; + } + } + + if (use_deserialized_benchmarks) { + benchmarks = deserialized_benchmarks; + } else { + // not using deserialized benchmarks, starting from scratch + first_benchmark_to_run = 0; + + // Randomly shuffling benchmarks allows us to get accurate enough progress info, + // as now the cheap/expensive benchmarks are randomly mixed so they average out. + // It also means that if data is corrupted for some time span, the odds are that + // not all repetitions of a given benchmark will be corrupted. + random_shuffle(benchmarks.begin(), benchmarks.end()); + } + + for (int i = 0; i < 4; i++) { + max_clock_speed = max(max_clock_speed, measure_clock_speed()); + } + + double time_start = 0.0; + while (first_benchmark_to_run < benchmarks.size()) { + if (first_benchmark_to_run == 0) { + time_start = timer.getRealTime(); + } + try_run_some_benchmarks(benchmarks, + time_start, + first_benchmark_to_run); + } + + // Sort timings by increasing benchmark parameters, and decreasing gflops. + // The latter is very important. It means that we can ignore all but the first + // benchmark with given parameters. + sort(benchmarks.begin(), benchmarks.end()); + + // Collect best (i.e. now first) results for each parameter values. + vector best_benchmarks; + for (auto it = benchmarks.begin(); it != benchmarks.end(); ++it) { + if (best_benchmarks.empty() || + best_benchmarks.back().compact_product_size != it->compact_product_size || + best_benchmarks.back().compact_block_size != it->compact_block_size) + { + best_benchmarks.push_back(*it); + } + } + + // keep and return only the best benchmarks + benchmarks = best_benchmarks; +} + +struct measure_all_pot_sizes_action_t : action_t +{ + virtual const char* invokation_name() const { return "all-pot-sizes"; } + virtual void run() const + { + vector benchmarks; + for (int repetition = 0; repetition < measurement_repetitions; repetition++) { + for (size_t ksize = minsize; ksize <= maxsize; ksize *= 2) { + for (size_t msize = minsize; msize <= maxsize; msize *= 2) { + for (size_t nsize = minsize; nsize <= maxsize; nsize *= 2) { + for (size_t kblock = minsize; kblock <= ksize; kblock *= 2) { + for (size_t mblock = minsize; mblock <= msize; mblock *= 2) { + for (size_t nblock = minsize; nblock <= nsize; nblock *= 2) { + benchmarks.emplace_back(ksize, msize, nsize, kblock, mblock, nblock); + } + } + } + } + } + } + } + + run_benchmarks(benchmarks); + + cout << "BEGIN MEASUREMENTS ALL POT SIZES" << endl; + for (auto it = benchmarks.begin(); it != benchmarks.end(); ++it) { + cout << *it << endl; + } + } +}; + +struct measure_default_sizes_action_t : action_t +{ + virtual const char* invokation_name() const { return "default-sizes"; } + virtual void run() const + { + vector benchmarks; + for (int repetition = 0; repetition < measurement_repetitions; repetition++) { + for (size_t ksize = minsize; ksize <= maxsize; ksize *= 2) { + for (size_t msize = minsize; msize <= maxsize; msize *= 2) { + for (size_t nsize = minsize; nsize <= maxsize; nsize *= 2) { + benchmarks.emplace_back(ksize, msize, nsize); + } + } + } + } + + run_benchmarks(benchmarks); + + cout << "BEGIN MEASUREMENTS DEFAULT SIZES" << endl; + for (auto it = benchmarks.begin(); it != benchmarks.end(); ++it) { + cout << *it << endl; + } + } +}; + +int main(int argc, char* argv[]) +{ + double time_start = timer.getRealTime(); + cout.precision(4); + cerr.precision(4); + + vector> available_actions; + available_actions.emplace_back(new measure_all_pot_sizes_action_t); + available_actions.emplace_back(new measure_default_sizes_action_t); + + auto action = available_actions.end(); + + if (argc <= 1) { + show_usage_and_exit(argc, argv, available_actions); + } + for (auto it = available_actions.begin(); it != available_actions.end(); ++it) { + if (!strcmp(argv[1], (*it)->invokation_name())) { + action = it; + break; + } + } + + if (action == available_actions.end()) { + show_usage_and_exit(argc, argv, available_actions); + } + + for (int i = 2; i < argc; i++) { + if (argv[i] == strstr(argv[i], "--min-working-set-size=")) { + const char* equals_sign = strchr(argv[i], '='); + min_working_set_size = strtoul(equals_sign+1, nullptr, 10); + } else { + cerr << "unrecognized option: " << argv[i] << endl << endl; + show_usage_and_exit(argc, argv, available_actions); + } + } + + print_cpuinfo(); + + cout << "benchmark parameters:" << endl; + cout << "pointer size: " << 8*sizeof(void*) << " bits" << endl; + cout << "scalar type: " << type_name() << endl; + cout << "packet size: " << internal::packet_traits::size << endl; + cout << "minsize = " << minsize << endl; + cout << "maxsize = " << maxsize << endl; + cout << "measurement_repetitions = " << measurement_repetitions << endl; + cout << "min_accurate_time = " << min_accurate_time << endl; + cout << "min_working_set_size = " << min_working_set_size; + if (min_working_set_size == 0) { + cout << " (try to outsize caches)"; + } + cout << endl << endl; + + (*action)->run(); + + double time_end = timer.getRealTime(); + cerr << "Finished in " << human_duration_t(time_end - time_start) << endl; +} diff --git a/thirdparty/eigen/bench/btl/CMakeLists.txt b/thirdparty/eigen/bench/btl/CMakeLists.txt index 119b470d..42094e86 100644 --- a/thirdparty/eigen/bench/btl/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/CMakeLists.txt @@ -1,40 +1,35 @@ -PROJECT(BTL) +project(BTL) -CMAKE_MINIMUM_REQUIRED(VERSION 2.6.2) +cmake_minimum_required(VERSION 2.6.2) set(CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake ${Eigen_SOURCE_DIR}/cmake) include(MacroOptionalAddSubdirectory) -OPTION(BTL_NOVEC "Disable SSE/Altivec optimizations when possible" OFF) +option(BTL_NOVEC "Disable SSE/Altivec optimizations when possible" OFF) -SET(CMAKE_INCLUDE_CURRENT_DIR ON) +set(CMAKE_INCLUDE_CURRENT_DIR ON) string(REGEX MATCH icpc IS_ICPC ${CMAKE_CXX_COMPILER}) -IF(CMAKE_COMPILER_IS_GNUCXX OR IS_ICPC) - SET(CMAKE_CXX_FLAGS "-g0 -O3 -DNDEBUG") - SET(CMAKE_Fortran_FLAGS "-g0 -O3 -DNDEBUG") - IF(NOT BTL_NOVEC) - SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -msse2") - SET(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -msse2") - ELSE(NOT BTL_NOVEC) - SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DEIGEN_DONT_VECTORIZE") - ENDIF(NOT BTL_NOVEC) -ENDIF(CMAKE_COMPILER_IS_GNUCXX OR IS_ICPC) - -IF(MSVC) - SET(CMAKE_CXX_FLAGS " /O2 /Ot /GL /fp:fast -DNDEBUG") -# SET(CMAKE_Fortran_FLAGS "-g0 -O3 -DNDEBUG") - IF(NOT BTL_NOVEC) - SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /arch:SSE2") - ELSE(NOT BTL_NOVEC) - SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DEIGEN_DONT_VECTORIZE") - ENDIF(NOT BTL_NOVEC) -ENDIF(MSVC) +if(CMAKE_COMPILER_IS_GNUCXX OR IS_ICPC) + set(CMAKE_CXX_FLAGS "-g0 -O3 -DNDEBUG ${CMAKE_CXX_FLAGS}") + set(CMAKE_Fortran_FLAGS "-g0 -O3 -DNDEBUG ${CMAKE_Fortran_FLAGS}") + if(BTL_NOVEC) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DEIGEN_DONT_VECTORIZE") + endif(BTL_NOVEC) +endif(CMAKE_COMPILER_IS_GNUCXX OR IS_ICPC) + +if(MSVC) + set(CMAKE_CXX_FLAGS " /O2 /Ot /GL /fp:fast -DNDEBUG") +# set(CMAKE_Fortran_FLAGS "-g0 -O3 -DNDEBUG") + if(BTL_NOVEC) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -DEIGEN_DONT_VECTORIZE") + endif(BTL_NOVEC) +endif(MSVC) if(IS_ICPC) - set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fast") - set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fast") -endif(IS_ICPC) + set(CMAKE_CXX_FLAGS "-fast ${CMAKE_CXX_FLAGS}") + set(CMAKE_Fortran_FLAGS "-fast ${CMAKE_Fortran_FLAGS}") +endif() include_directories( ${PROJECT_SOURCE_DIR}/actions @@ -46,13 +41,19 @@ include_directories( # if (MKL_FOUND) # add_definitions(-DHAVE_MKL) # set(DEFAULT_LIBRARIES ${MKL_LIBRARIES}) -# endif (MKL_FOUND) +# endif () -MACRO(BTL_ADD_BENCH targetname) +find_library(EIGEN_BTL_RT_LIBRARY rt) +# if we cannot find it easily, then we don't need it! +if(NOT EIGEN_BTL_RT_LIBRARY) + set(EIGEN_BTL_RT_LIBRARY "") +endif() + +macro(BTL_ADD_BENCH targetname) foreach(_current_var ${ARGN}) set(_last_var ${_current_var}) - endforeach(_current_var) + endforeach() set(_sources ${ARGN}) list(LENGTH _sources _argn_length) @@ -63,17 +64,17 @@ MACRO(BTL_ADD_BENCH targetname) if (${_argn_length} EQUAL ${_src_length}) set(_last_var ON) - endif (${_argn_length} EQUAL ${_src_length}) + endif () - OPTION(BUILD_${targetname} "Build benchmark ${targetname}" ${_last_var}) + option(BUILD_${targetname} "Build benchmark ${targetname}" ${_last_var}) - IF(BUILD_${targetname}) - ADD_EXECUTABLE(${targetname} ${_sources}) - ADD_TEST(${targetname} "${targetname}") - target_link_libraries(${targetname} ${DEFAULT_LIBRARIES} rt) - ENDIF(BUILD_${targetname}) + if(BUILD_${targetname}) + add_executable(${targetname} ${_sources}) + add_test(${targetname} "${targetname}") + target_link_libraries(${targetname} ${DEFAULT_LIBRARIES} ${EIGEN_BTL_RT_LIBRARY}) + endif(BUILD_${targetname}) -ENDMACRO(BTL_ADD_BENCH) +endmacro(BTL_ADD_BENCH) macro(btl_add_target_property target prop value) @@ -85,12 +86,13 @@ macro(btl_add_target_property target prop value) set_target_properties(${target} PROPERTIES ${prop} "${previous} ${value}") endif() -endmacro(btl_add_target_property) +endmacro() -ENABLE_TESTING() +enable_testing() add_subdirectory(libs/eigen3) add_subdirectory(libs/eigen2) +add_subdirectory(libs/tensors) add_subdirectory(libs/BLAS) add_subdirectory(libs/ublas) add_subdirectory(libs/gmm) @@ -98,6 +100,7 @@ add_subdirectory(libs/mtl4) add_subdirectory(libs/blitz) add_subdirectory(libs/tvmet) add_subdirectory(libs/STL) +add_subdirectory(libs/blaze) add_subdirectory(data) diff --git a/thirdparty/eigen/bench/btl/README b/thirdparty/eigen/bench/btl/README index f3f5fb36..ebed8896 100644 --- a/thirdparty/eigen/bench/btl/README +++ b/thirdparty/eigen/bench/btl/README @@ -36,7 +36,7 @@ For instance: You can also select a given set of actions defining the environment variable BTL_CONFIG this way: BTL_CONFIG="-a action1{:action2}*" ctest -V -An exemple: +An example: BTL_CONFIG="-a axpy:vector_matrix:trisolve:ata" ctest -V -R eigen2 Finally, if bench results already exist (the bench*.dat files) then they merges by keeping the best for each matrix size. If you want to overwrite the previous ones you can simply add the "--overwrite" option: diff --git a/thirdparty/eigen/bench/btl/actions/action_axpby.hh b/thirdparty/eigen/bench/btl/actions/action_axpby.hh index 98511ab6..dadd0ccf 100644 --- a/thirdparty/eigen/bench/btl/actions/action_axpby.hh +++ b/thirdparty/eigen/bench/btl/actions/action_axpby.hh @@ -33,7 +33,7 @@ class Action_axpby { public : // Ctor - Action_axpby( int size ):_size(size),_alpha(0.5),_beta(0.95) + Action_axpby( int size ):_alpha(0.5),_beta(0.95),_size(size) { MESSAGE("Action_axpby Ctor"); diff --git a/thirdparty/eigen/bench/btl/actions/action_axpy.hh b/thirdparty/eigen/bench/btl/actions/action_axpy.hh index e4cb3a5b..261be4cb 100644 --- a/thirdparty/eigen/bench/btl/actions/action_axpy.hh +++ b/thirdparty/eigen/bench/btl/actions/action_axpy.hh @@ -35,7 +35,7 @@ public : // Ctor - Action_axpy( int size ):_size(size),_coef(1.0) + Action_axpy( int size ):_coef(1.0),_size(size) { MESSAGE("Action_axpy Ctor"); diff --git a/thirdparty/eigen/bench/btl/actions/basic_actions.hh b/thirdparty/eigen/bench/btl/actions/basic_actions.hh index a3333ea2..62442f01 100644 --- a/thirdparty/eigen/bench/btl/actions/basic_actions.hh +++ b/thirdparty/eigen/bench/btl/actions/basic_actions.hh @@ -6,7 +6,7 @@ #include "action_atv_product.hh" #include "action_matrix_matrix_product.hh" -// #include "action_ata_product.hh" +#include "action_ata_product.hh" #include "action_aat_product.hh" #include "action_trisolve.hh" diff --git a/thirdparty/eigen/bench/btl/cmake/FindACML.cmake b/thirdparty/eigen/bench/btl/cmake/FindACML.cmake index f45ae1b0..daeeb535 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindACML.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindACML.cmake @@ -1,7 +1,7 @@ if (ACML_LIBRARIES) set(ACML_FIND_QUIETLY TRUE) -endif (ACML_LIBRARIES) +endif () find_library(ACML_LIBRARIES NAMES @@ -17,6 +17,7 @@ find_file(ACML_LIBRARIES libacml_mp.so PATHS /usr/lib + /usr/lib64 $ENV{ACMLDIR}/lib ${LIB_INSTALL_DIR} ) @@ -35,6 +36,7 @@ if(NOT ACML_LIBRARIES) libacml.so libacml_mv.so PATHS /usr/lib + /usr/lib64 $ENV{ACMLDIR}/lib ${LIB_INSTALL_DIR} ) diff --git a/thirdparty/eigen/bench/btl/cmake/FindATLAS.cmake b/thirdparty/eigen/bench/btl/cmake/FindATLAS.cmake index 6b906520..572a4c0b 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindATLAS.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindATLAS.cmake @@ -1,37 +1,29 @@ if (ATLAS_LIBRARIES) set(ATLAS_FIND_QUIETLY TRUE) -endif (ATLAS_LIBRARIES) +endif () -find_file(ATLAS_LIB libatlas.so.3 PATHS /usr/lib $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) -find_library(ATLAS_LIB atlas PATHS $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) +find_file(ATLAS_LIB libatlas.so.3 PATHS /usr/lib /usr/lib/atlas /usr/lib64 /usr/lib64/atlas $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) +find_library(ATLAS_LIB satlas PATHS $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) -find_file(ATLAS_CBLAS libcblas.so.3 PATHS /usr/lib $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) -find_library(ATLAS_CBLAS cblas PATHS $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) +find_file(ATLAS_LAPACK NAMES liblapack_atlas.so.3 liblapack.so.3 PATHS /usr/lib /usr/lib/atlas /usr/lib64 /usr/lib64/atlas $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) +find_library(ATLAS_LAPACK NAMES lapack_atlas lapack PATHS $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) -find_file(ATLAS_LAPACK liblapack_atlas.so.3 PATHS /usr/lib $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) -find_library(ATLAS_LAPACK lapack_atlas PATHS $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) - -if(NOT ATLAS_LAPACK) - find_file(ATLAS_LAPACK liblapack.so.3 PATHS /usr/lib/atlas $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) - find_library(ATLAS_LAPACK lapack PATHS $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) -endif(NOT ATLAS_LAPACK) - -find_file(ATLAS_F77BLAS libf77blas.so.3 PATHS /usr/lib $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) +find_file(ATLAS_F77BLAS libf77blas.so.3 PATHS /usr/lib /usr/lib/atlas /usr/lib64 /usr/lib64/atlas $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) find_library(ATLAS_F77BLAS f77blas PATHS $ENV{ATLASDIR} ${LIB_INSTALL_DIR}) if(ATLAS_LIB AND ATLAS_CBLAS AND ATLAS_LAPACK AND ATLAS_F77BLAS) - set(ATLAS_LIBRARIES ${ATLAS_LAPACK} ${ATLAS_CBLAS} ${ATLAS_F77BLAS} ${ATLAS_LIB}) + set(ATLAS_LIBRARIES ${ATLAS_LAPACK} ${ATLAS_LIB}) # search the default lapack lib link to it find_file(ATLAS_REFERENCE_LAPACK liblapack.so.3 PATHS /usr/lib /usr/lib64) find_library(ATLAS_REFERENCE_LAPACK NAMES lapack) - if(ATLAS_REFERENCE_LAPACK) - set(ATLAS_LIBRARIES ${ATLAS_LIBRARIES} ${ATLAS_REFERENCE_LAPACK}) - endif() +# if(ATLAS_REFERENCE_LAPACK) +# set(ATLAS_LIBRARIES ${ATLAS_LIBRARIES} ${ATLAS_REFERENCE_LAPACK}) +# endif() -endif(ATLAS_LIB AND ATLAS_CBLAS AND ATLAS_LAPACK AND ATLAS_F77BLAS) +endif() include(FindPackageHandleStandardArgs) find_package_handle_standard_args(ATLAS DEFAULT_MSG ATLAS_LIBRARIES) diff --git a/thirdparty/eigen/bench/btl/cmake/FindBLAZE.cmake b/thirdparty/eigen/bench/btl/cmake/FindBLAZE.cmake new file mode 100644 index 00000000..18a878ff --- /dev/null +++ b/thirdparty/eigen/bench/btl/cmake/FindBLAZE.cmake @@ -0,0 +1,31 @@ +# - Try to find eigen2 headers +# Once done this will define +# +# BLAZE_FOUND - system has blaze lib +# BLAZE_INCLUDE_DIR - the blaze include directory +# +# Copyright (C) 2008 Gael Guennebaud +# Adapted from FindEigen.cmake: +# Copyright (c) 2006, 2007 Montel Laurent, +# Redistribution and use is allowed according to the terms of the BSD license. +# For details see the accompanying COPYING-CMAKE-SCRIPTS file. + +if (BLAZE_INCLUDE_DIR) + + # in cache already + set(BLAZE_FOUND TRUE) + +else () + +find_path(BLAZE_INCLUDE_DIR NAMES blaze/Blaze.h + PATHS + ${INCLUDE_INSTALL_DIR} + ) + +include(FindPackageHandleStandardArgs) +find_package_handle_standard_args(BLAZE DEFAULT_MSG BLAZE_INCLUDE_DIR) + +mark_as_advanced(BLAZE_INCLUDE_DIR) + +endif() + diff --git a/thirdparty/eigen/bench/btl/cmake/FindBlitz.cmake b/thirdparty/eigen/bench/btl/cmake/FindBlitz.cmake index 92880bbe..7ab375fd 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindBlitz.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindBlitz.cmake @@ -15,7 +15,7 @@ if (BLITZ_INCLUDES AND BLITZ_LIBRARIES) set(Blitz_FIND_QUIETLY TRUE) -endif (BLITZ_INCLUDES AND BLITZ_LIBRARIES) +endif () find_path(BLITZ_INCLUDES NAMES diff --git a/thirdparty/eigen/bench/btl/cmake/FindCBLAS.cmake b/thirdparty/eigen/bench/btl/cmake/FindCBLAS.cmake index 554f0291..43a90f7f 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindCBLAS.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindCBLAS.cmake @@ -2,7 +2,7 @@ if (CBLAS_INCLUDES AND CBLAS_LIBRARIES) set(CBLAS_FIND_QUIETLY TRUE) -endif (CBLAS_INCLUDES AND CBLAS_LIBRARIES) +endif () find_path(CBLAS_INCLUDES NAMES @@ -23,6 +23,7 @@ find_file(CBLAS_LIBRARIES libcblas.so.3 PATHS /usr/lib + /usr/lib64 $ENV{CBLASDIR}/lib ${LIB_INSTALL_DIR} ) diff --git a/thirdparty/eigen/bench/btl/cmake/FindGMM.cmake b/thirdparty/eigen/bench/btl/cmake/FindGMM.cmake index 5049c64e..ff45e6a0 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindGMM.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindGMM.cmake @@ -1,7 +1,7 @@ if (GMM_INCLUDE_DIR) # in cache already set(GMM_FOUND TRUE) -else (GMM_INCLUDE_DIR) +else () find_path(GMM_INCLUDE_DIR NAMES gmm/gmm.h PATHS @@ -14,4 +14,4 @@ FIND_PACKAGE_HANDLE_STANDARD_ARGS(GMM DEFAULT_MSG GMM_INCLUDE_DIR ) mark_as_advanced(GMM_INCLUDE_DIR) -endif(GMM_INCLUDE_DIR) +endif() diff --git a/thirdparty/eigen/bench/btl/cmake/FindGOTO.cmake b/thirdparty/eigen/bench/btl/cmake/FindGOTO.cmake deleted file mode 100644 index 67ea0934..00000000 --- a/thirdparty/eigen/bench/btl/cmake/FindGOTO.cmake +++ /dev/null @@ -1,15 +0,0 @@ - -if (GOTO_LIBRARIES) - set(GOTO_FIND_QUIETLY TRUE) -endif (GOTO_LIBRARIES) - -find_library(GOTO_LIBRARIES goto PATHS $ENV{GOTODIR} ${LIB_INSTALL_DIR}) - -if(GOTO_LIBRARIES AND CMAKE_COMPILER_IS_GNUCXX) - set(GOTO_LIBRARIES ${GOTO_LIBRARIES} "-lpthread -lgfortran") -endif() - -include(FindPackageHandleStandardArgs) -find_package_handle_standard_args(GOTO DEFAULT_MSG GOTO_LIBRARIES) - -mark_as_advanced(GOTO_LIBRARIES) diff --git a/thirdparty/eigen/bench/btl/cmake/FindGOTO2.cmake b/thirdparty/eigen/bench/btl/cmake/FindGOTO2.cmake deleted file mode 100644 index baa68d21..00000000 --- a/thirdparty/eigen/bench/btl/cmake/FindGOTO2.cmake +++ /dev/null @@ -1,25 +0,0 @@ - -if (GOTO2_LIBRARIES) - set(GOTO2_FIND_QUIETLY TRUE) -endif (GOTO2_LIBRARIES) -# -# find_path(GOTO_INCLUDES -# NAMES -# cblas.h -# PATHS -# $ENV{GOTODIR}/include -# ${INCLUDE_INSTALL_DIR} -# ) - -find_file(GOTO2_LIBRARIES libgoto2.so PATHS /usr/lib $ENV{GOTO2DIR} ${LIB_INSTALL_DIR}) -find_library(GOTO2_LIBRARIES goto2 PATHS $ENV{GOTO2DIR} ${LIB_INSTALL_DIR}) - -if(GOTO2_LIBRARIES AND CMAKE_COMPILER_IS_GNUCXX) - set(GOTO2_LIBRARIES ${GOTO2_LIBRARIES} "-lpthread -lgfortran") -endif() - -include(FindPackageHandleStandardArgs) -find_package_handle_standard_args(GOTO2 DEFAULT_MSG - GOTO2_LIBRARIES) - -mark_as_advanced(GOTO2_LIBRARIES) diff --git a/thirdparty/eigen/bench/btl/cmake/FindMKL.cmake b/thirdparty/eigen/bench/btl/cmake/FindMKL.cmake index f4d7c6eb..23e77279 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindMKL.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindMKL.cmake @@ -1,7 +1,7 @@ if (MKL_LIBRARIES) set(MKL_FIND_QUIETLY TRUE) -endif (MKL_LIBRARIES) +endif () if(CMAKE_MINOR_VERSION GREATER 4) @@ -30,7 +30,7 @@ if(MKL_LIBRARIES AND MKL_GUIDE) set(MKL_LIBRARIES ${MKL_LIBRARIES} mkl_intel_lp64 mkl_sequential ${MKL_GUIDE} pthread) endif() -else(${CMAKE_HOST_SYSTEM_PROCESSOR} STREQUAL "x86_64") +else() find_library(MKL_LIBRARIES mkl_core @@ -55,9 +55,9 @@ if(MKL_LIBRARIES AND MKL_GUIDE) set(MKL_LIBRARIES ${MKL_LIBRARIES} mkl_intel mkl_sequential ${MKL_GUIDE} pthread) endif() -endif(${CMAKE_HOST_SYSTEM_PROCESSOR} STREQUAL "x86_64") +endif() -endif(CMAKE_MINOR_VERSION GREATER 4) +endif() include(FindPackageHandleStandardArgs) find_package_handle_standard_args(MKL DEFAULT_MSG MKL_LIBRARIES) diff --git a/thirdparty/eigen/bench/btl/cmake/FindMTL4.cmake b/thirdparty/eigen/bench/btl/cmake/FindMTL4.cmake index 3de49098..1bafc93a 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindMTL4.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindMTL4.cmake @@ -15,7 +15,7 @@ if (MTL4_INCLUDE_DIR) # in cache already set(MTL4_FOUND TRUE) -else (MTL4_INCLUDE_DIR) +else () find_path(MTL4_INCLUDE_DIR NAMES boost/numeric/mtl/mtl.hpp PATHS @@ -27,5 +27,5 @@ find_package_handle_standard_args(MTL4 DEFAULT_MSG MTL4_INCLUDE_DIR) mark_as_advanced(MTL4_INCLUDE_DIR) -endif(MTL4_INCLUDE_DIR) +endif() diff --git a/thirdparty/eigen/bench/btl/cmake/FindOPENBLAS.cmake b/thirdparty/eigen/bench/btl/cmake/FindOPENBLAS.cmake new file mode 100644 index 00000000..5c076230 --- /dev/null +++ b/thirdparty/eigen/bench/btl/cmake/FindOPENBLAS.cmake @@ -0,0 +1,17 @@ + +if (OPENBLAS_LIBRARIES) + set(OPENBLAS_FIND_QUIETLY TRUE) +endif () + +find_file(OPENBLAS_LIBRARIES NAMES libopenblas.so libopenblas.so.0 PATHS /usr/lib /usr/lib64 $ENV{OPENBLASDIR} ${LIB_INSTALL_DIR}) +find_library(OPENBLAS_LIBRARIES openblas PATHS $ENV{OPENBLASDIR} ${LIB_INSTALL_DIR}) + +if(OPENBLAS_LIBRARIES AND CMAKE_COMPILER_IS_GNUCXX) + set(OPENBLAS_LIBRARIES ${OPENBLAS_LIBRARIES} "-lpthread -lgfortran") +endif() + +include(FindPackageHandleStandardArgs) +find_package_handle_standard_args(OPENBLAS DEFAULT_MSG + OPENBLAS_LIBRARIES) + +mark_as_advanced(OPENBLAS_LIBRARIES) diff --git a/thirdparty/eigen/bench/btl/cmake/FindPackageHandleStandardArgs.cmake b/thirdparty/eigen/bench/btl/cmake/FindPackageHandleStandardArgs.cmake index 7f122edc..05d7e65b 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindPackageHandleStandardArgs.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindPackageHandleStandardArgs.cmake @@ -1,7 +1,7 @@ # FIND_PACKAGE_HANDLE_STANDARD_ARGS(NAME (DEFAULT_MSG|"Custom failure message") VAR1 ... ) # # This macro is intended to be used in FindXXX.cmake modules files. -# It handles the REQUIRED and QUIET argument to FIND_PACKAGE() and +# It handles the REQUIRED and QUIET argument to find_package() and # it also sets the _FOUND variable. # The package is found if all variables listed are TRUE. # Example: @@ -19,42 +19,42 @@ # be "Could NOT find LibXml2", if you don't like this message you can specify # your own custom failure message there. -MACRO(FIND_PACKAGE_HANDLE_STANDARD_ARGS _NAME _FAIL_MSG _VAR1 ) +macro(FIND_PACKAGE_HANDLE_STANDARD_ARGS _NAME _FAIL_MSG _VAR1 ) - IF("${_FAIL_MSG}" STREQUAL "DEFAULT_MSG") - IF (${_NAME}_FIND_REQUIRED) - SET(_FAIL_MESSAGE "Could not find REQUIRED package ${_NAME}") - ELSE (${_NAME}_FIND_REQUIRED) - SET(_FAIL_MESSAGE "Could not find OPTIONAL package ${_NAME}") - ENDIF (${_NAME}_FIND_REQUIRED) - ELSE("${_FAIL_MSG}" STREQUAL "DEFAULT_MSG") - SET(_FAIL_MESSAGE "${_FAIL_MSG}") - ENDIF("${_FAIL_MSG}" STREQUAL "DEFAULT_MSG") + if("${_FAIL_MSG}" STREQUAL "DEFAULT_MSG") + if (${_NAME}_FIND_REQUIRED) + set(_FAIL_MESSAGE "Could not find REQUIRED package ${_NAME}") + else (${_NAME}_FIND_REQUIRED) + set(_FAIL_MESSAGE "Could not find OPTIONAL package ${_NAME}") + endif (${_NAME}_FIND_REQUIRED) + else("${_FAIL_MSG}" STREQUAL "DEFAULT_MSG") + set(_FAIL_MESSAGE "${_FAIL_MSG}") + endif("${_FAIL_MSG}" STREQUAL "DEFAULT_MSG") - STRING(TOUPPER ${_NAME} _NAME_UPPER) + string(TOUPPER ${_NAME} _NAME_UPPER) - SET(${_NAME_UPPER}_FOUND TRUE) - IF(NOT ${_VAR1}) - SET(${_NAME_UPPER}_FOUND FALSE) - ENDIF(NOT ${_VAR1}) + set(${_NAME_UPPER}_FOUND TRUE) + if(NOT ${_VAR1}) + set(${_NAME_UPPER}_FOUND FALSE) + endif(NOT ${_VAR1}) - FOREACH(_CURRENT_VAR ${ARGN}) - IF(NOT ${_CURRENT_VAR}) - SET(${_NAME_UPPER}_FOUND FALSE) - ENDIF(NOT ${_CURRENT_VAR}) - ENDFOREACH(_CURRENT_VAR) + foreach(_CURRENT_VAR ${ARGN}) + if(NOT ${_CURRENT_VAR}) + set(${_NAME_UPPER}_FOUND FALSE) + endif(NOT ${_CURRENT_VAR}) + endforeach(_CURRENT_VAR) - IF (${_NAME_UPPER}_FOUND) - IF (NOT ${_NAME}_FIND_QUIETLY) - MESSAGE(STATUS "Found ${_NAME}: ${${_VAR1}}") - ENDIF (NOT ${_NAME}_FIND_QUIETLY) - ELSE (${_NAME_UPPER}_FOUND) - IF (${_NAME}_FIND_REQUIRED) - MESSAGE(FATAL_ERROR "${_FAIL_MESSAGE}") - ELSE (${_NAME}_FIND_REQUIRED) - IF (NOT ${_NAME}_FIND_QUIETLY) - MESSAGE(STATUS "${_FAIL_MESSAGE}") - ENDIF (NOT ${_NAME}_FIND_QUIETLY) - ENDIF (${_NAME}_FIND_REQUIRED) - ENDIF (${_NAME_UPPER}_FOUND) -ENDMACRO(FIND_PACKAGE_HANDLE_STANDARD_ARGS) + if (${_NAME_UPPER}_FOUND) + if (NOT ${_NAME}_FIND_QUIETLY) + message(STATUS "Found ${_NAME}: ${${_VAR1}}") + endif (NOT ${_NAME}_FIND_QUIETLY) + else (${_NAME_UPPER}_FOUND) + if (${_NAME}_FIND_REQUIRED) + message(FATAL_ERROR "${_FAIL_MESSAGE}") + else (${_NAME}_FIND_REQUIRED) + if (NOT ${_NAME}_FIND_QUIETLY) + message(STATUS "${_FAIL_MESSAGE}") + endif (NOT ${_NAME}_FIND_QUIETLY) + endif (${_NAME}_FIND_REQUIRED) + endif (${_NAME_UPPER}_FOUND) +endmacro(FIND_PACKAGE_HANDLE_STANDARD_ARGS) diff --git a/thirdparty/eigen/bench/btl/cmake/FindTvmet.cmake b/thirdparty/eigen/bench/btl/cmake/FindTvmet.cmake index 26a29d96..8ccae271 100644 --- a/thirdparty/eigen/bench/btl/cmake/FindTvmet.cmake +++ b/thirdparty/eigen/bench/btl/cmake/FindTvmet.cmake @@ -15,7 +15,7 @@ if (TVMET_INCLUDE_DIR) # in cache already set(TVMET_FOUND TRUE) -else (TVMET_INCLUDE_DIR) +else () find_path(TVMET_INCLUDE_DIR NAMES tvmet/tvmet.h PATHS @@ -28,5 +28,5 @@ find_package_handle_standard_args(Tvmet DEFAULT_MSG TVMET_INCLUDE_DIR) mark_as_advanced(TVMET_INCLUDE_DIR) -endif(TVMET_INCLUDE_DIR) +endif() diff --git a/thirdparty/eigen/bench/btl/cmake/MacroOptionalAddSubdirectory.cmake b/thirdparty/eigen/bench/btl/cmake/MacroOptionalAddSubdirectory.cmake index 545048b6..8d46fcea 100644 --- a/thirdparty/eigen/bench/btl/cmake/MacroOptionalAddSubdirectory.cmake +++ b/thirdparty/eigen/bench/btl/cmake/MacroOptionalAddSubdirectory.cmake @@ -1,6 +1,6 @@ -# - MACRO_OPTIONAL_ADD_SUBDIRECTORY() combines ADD_SUBDIRECTORY() with an OPTION() +# - MACRO_OPTIONAL_ADD_SUBDIRECTORY() combines add_subdirectory() with an option() # MACRO_OPTIONAL_ADD_SUBDIRECTORY( ) -# If you use MACRO_OPTIONAL_ADD_SUBDIRECTORY() instead of ADD_SUBDIRECTORY(), +# If you use MACRO_OPTIONAL_ADD_SUBDIRECTORY() instead of add_subdirectory(), # this will have two effects # 1 - CMake will not complain if the directory doesn't exist # This makes sense if you want to distribute just one of the subdirs @@ -16,16 +16,16 @@ # For details see the accompanying COPYING-CMAKE-SCRIPTS file. -MACRO (MACRO_OPTIONAL_ADD_SUBDIRECTORY _dir ) - GET_FILENAME_COMPONENT(_fullPath ${_dir} ABSOLUTE) - IF(EXISTS ${_fullPath}) - IF(${ARGC} EQUAL 2) - OPTION(BUILD_${_dir} "Build directory ${_dir}" ${ARGV1}) - ELSE(${ARGC} EQUAL 2) - OPTION(BUILD_${_dir} "Build directory ${_dir}" TRUE) - ENDIF(${ARGC} EQUAL 2) - IF(BUILD_${_dir}) - ADD_SUBDIRECTORY(${_dir}) - ENDIF(BUILD_${_dir}) - ENDIF(EXISTS ${_fullPath}) -ENDMACRO (MACRO_OPTIONAL_ADD_SUBDIRECTORY) +macro (MACRO_OPTIONAL_ADD_SUBDIRECTORY _dir ) + get_filename_component(_fullPath ${_dir} ABSOLUTE) + if(EXISTS ${_fullPath}) + if(${ARGC} EQUAL 2) + option(BUILD_${_dir} "Build directory ${_dir}" ${ARGV1}) + else(${ARGC} EQUAL 2) + option(BUILD_${_dir} "Build directory ${_dir}" TRUE) + endif(${ARGC} EQUAL 2) + if(BUILD_${_dir}) + add_subdirectory(${_dir}) + endif(BUILD_${_dir}) + endif(EXISTS ${_fullPath}) +endmacro (MACRO_OPTIONAL_ADD_SUBDIRECTORY) diff --git a/thirdparty/eigen/bench/btl/data/CMakeLists.txt b/thirdparty/eigen/bench/btl/data/CMakeLists.txt index 6af2a366..580c1ced 100644 --- a/thirdparty/eigen/bench/btl/data/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/data/CMakeLists.txt @@ -1,25 +1,25 @@ -ADD_CUSTOM_TARGET(copy_scripts) +add_custom_target(copy_scripts) -SET(script_files go_mean mk_mean_script.sh mk_new_gnuplot.sh +set(script_files go_mean mk_mean_script.sh mk_new_gnuplot.sh perlib_plot_settings.txt action_settings.txt gnuplot_common_settings.hh ) -FOREACH(script_file ${script_files}) -ADD_CUSTOM_COMMAND( +foreach(script_file ${script_files}) +add_custom_command( TARGET copy_scripts POST_BUILD COMMAND ${CMAKE_COMMAND} -E copy ${CMAKE_CURRENT_SOURCE_DIR}/${script_file} ${CMAKE_CURRENT_BINARY_DIR}/ ARGS ) -ENDFOREACH(script_file) +endforeach(script_file) -ADD_CUSTOM_COMMAND( +add_custom_command( TARGET copy_scripts POST_BUILD COMMAND ${CMAKE_CXX_COMPILER} --version | head -n 1 > ${CMAKE_CURRENT_BINARY_DIR}/compiler_version.txt ARGS ) -ADD_CUSTOM_COMMAND( +add_custom_command( TARGET copy_scripts POST_BUILD COMMAND echo "${Eigen_SOURCE_DIR}" > ${CMAKE_CURRENT_BINARY_DIR}/eigen_root_dir.txt diff --git a/thirdparty/eigen/bench/btl/data/action_settings.txt b/thirdparty/eigen/bench/btl/data/action_settings.txt index e32213e2..39d2b5dc 100644 --- a/thirdparty/eigen/bench/btl/data/action_settings.txt +++ b/thirdparty/eigen/bench/btl/data/action_settings.txt @@ -1,19 +1,19 @@ -aat ; "{/*1.5 A x A^T}" ; "matrix size" ; 4:3000 -ata ; "{/*1.5 A^T x A}" ; "matrix size" ; 4:3000 -atv ; "{/*1.5 matrix^T x vector}" ; "matrix size" ; 4:3000 +aat ; "{/*1.5 A x A^T}" ; "matrix size" ; 4:5000 +ata ; "{/*1.5 A^T x A}" ; "matrix size" ; 4:5000 +atv ; "{/*1.5 matrix^T x vector}" ; "matrix size" ; 4:5000 axpby ; "{/*1.5 Y = alpha X + beta Y}" ; "vector size" ; 5:1000000 axpy ; "{/*1.5 Y += alpha X}" ; "vector size" ; 5:1000000 -matrix_matrix ; "{/*1.5 matrix matrix product}" ; "matrix size" ; 4:3000 -matrix_vector ; "{/*1.5 matrix vector product}" ; "matrix size" ; 4:3000 -trmm ; "{/*1.5 triangular matrix matrix product}" ; "matrix size" ; 4:3000 -trisolve_vector ; "{/*1.5 triangular solver - vector (X = inv(L) X)}" ; "size" ; 4:3000 -trisolve_matrix ; "{/*1.5 triangular solver - matrix (M = inv(L) M)}" ; "size" ; 4:3000 -cholesky ; "{/*1.5 Cholesky decomposition}" ; "matrix size" ; 4:3000 -complete_lu_decomp ; "{/*1.5 Complete LU decomposition}" ; "matrix size" ; 4:3000 -partial_lu_decomp ; "{/*1.5 Partial LU decomposition}" ; "matrix size" ; 4:3000 -tridiagonalization ; "{/*1.5 Tridiagonalization}" ; "matrix size" ; 4:3000 -hessenberg ; "{/*1.5 Hessenberg decomposition}" ; "matrix size" ; 4:3000 -symv ; "{/*1.5 symmetric matrix vector product}" ; "matrix size" ; 4:3000 -syr2 ; "{/*1.5 symmretric rank-2 update (A += u^T v + u v^T)}" ; "matrix size" ; 4:3000 -ger ; "{/*1.5 general rank-1 update (A += u v^T)}" ; "matrix size" ; 4:3000 -rot ; "{/*1.5 apply rotation in the plane}" ; "vector size" ; 4:1000000 \ No newline at end of file +matrix_matrix ; "{/*1.5 matrix matrix product}" ; "matrix size" ; 4:5000 +matrix_vector ; "{/*1.5 matrix vector product}" ; "matrix size" ; 4:5000 +trmm ; "{/*1.5 triangular matrix matrix product}" ; "matrix size" ; 4:5000 +trisolve_vector ; "{/*1.5 triangular solver - vector (X = inv(L) X)}" ; "size" ; 4:5000 +trisolve_matrix ; "{/*1.5 triangular solver - matrix (M = inv(L) M)}" ; "size" ; 4:5000 +cholesky ; "{/*1.5 Cholesky decomposition}" ; "matrix size" ; 4:5000 +complete_lu_decomp ; "{/*1.5 Complete LU decomposition}" ; "matrix size" ; 4:5000 +partial_lu_decomp ; "{/*1.5 Partial LU decomposition}" ; "matrix size" ; 4:5000 +tridiagonalization ; "{/*1.5 Tridiagonalization}" ; "matrix size" ; 4:5000 +hessenberg ; "{/*1.5 Hessenberg decomposition}" ; "matrix size" ; 4:5000 +symv ; "{/*1.5 symmetric matrix vector product}" ; "matrix size" ; 4:5000 +syr2 ; "{/*1.5 symmretric rank-2 update (A += u^T v + u v^T)}" ; "matrix size" ; 4:5000 +ger ; "{/*1.5 general rank-1 update (A += u v^T)}" ; "matrix size" ; 4:5000 +rot ; "{/*1.5 apply rotation in the plane}" ; "vector size" ; 4:1000000 diff --git a/thirdparty/eigen/bench/btl/data/go_mean b/thirdparty/eigen/bench/btl/data/go_mean index 42338ca2..d0142690 100755 --- a/thirdparty/eigen/bench/btl/data/go_mean +++ b/thirdparty/eigen/bench/btl/data/go_mean @@ -27,7 +27,7 @@ echo '
    '\ '
  • ' `cat /proc/cpuinfo | grep "model name" | head -n 1`\ ' (' `uname -m` ')
    • '\ '
  • compiler: ' `cat compiler_version.txt` '
    • '\ - '
  • eigen3: ' `hg identify -i $EIGENDIR` '
    • '\ + '
  • eigen3: ' `git ls-remote --refs -q $EIGENDIR HEAD | cut -f 1` '
    • '\ '
' \ '

' >> $webpagefilename diff --git a/thirdparty/eigen/bench/btl/data/perlib_plot_settings.txt b/thirdparty/eigen/bench/btl/data/perlib_plot_settings.txt index 6844bab2..f023cfe0 100644 --- a/thirdparty/eigen/bench/btl/data/perlib_plot_settings.txt +++ b/thirdparty/eigen/bench/btl/data/perlib_plot_settings.txt @@ -10,7 +10,7 @@ ublas ; with lines lw 3 lt 1 lc rgbcolor "#00b7ff" mtl4 ; with lines lw 3 lt 1 lc rgbcolor "#d18847" blitz ; with lines lw 3 lt 1 lc rgbcolor "#ff00ff" F77 ; with lines lw 3 lt 3 lc rgbcolor "#e6e64c" -GOTO ; with lines lw 3 lt 3 lc rgbcolor "#C05600" -GOTO2 ; with lines lw 3 lt 1 lc rgbcolor "#C05600" +OPENBLAS ; with lines lw 3 lt 1 lc rgbcolor "#C05600" C ; with lines lw 3 lt 3 lc rgbcolor "#e6bd96" ACML ; with lines lw 2 lt 3 lc rgbcolor "#e6e64c" +blaze ; with lines lw 3 lt 1 lc rgbcolor "#ff00ff" diff --git a/thirdparty/eigen/bench/btl/generic_bench/bench.hh b/thirdparty/eigen/bench/btl/generic_bench/bench.hh index 005c3639..0732940d 100644 --- a/thirdparty/eigen/bench/btl/generic_bench/bench.hh +++ b/thirdparty/eigen/bench/btl/generic_bench/bench.hh @@ -102,8 +102,8 @@ BTL_DONT_INLINE void bench( int size_min, int size_max, int nb_point ) // merge the two data std::vector newSizes; std::vector newFlops; - int i=0; - int j=0; + unsigned int i=0; + unsigned int j=0; while (i(size_min,size_max,nb_point); - // Only for small problem size. Otherwize it will be too long + // Only for small problem size. Otherwise it will be too long // bench(size_min,size_max,nb_point); // bench(size_min,size_max,nb_point); diff --git a/thirdparty/eigen/bench/btl/generic_bench/bench_parameter.hh b/thirdparty/eigen/bench/btl/generic_bench/bench_parameter.hh index 4c355cd6..2b01149f 100644 --- a/thirdparty/eigen/bench/btl/generic_bench/bench_parameter.hh +++ b/thirdparty/eigen/bench/btl/generic_bench/bench_parameter.hh @@ -29,11 +29,11 @@ // min vector size for axpy bench #define MIN_AXPY 5 // max vector size for axpy bench -#define MAX_AXPY 1000000 +#define MAX_AXPY 3000000 // min matrix size for matrix vector product bench #define MIN_MV 5 // max matrix size for matrix vector product bench -#define MAX_MV 3000 +#define MAX_MV 5000 // min matrix size for matrix matrix product bench #define MIN_MM 5 // max matrix size for matrix matrix product bench diff --git a/thirdparty/eigen/bench/btl/generic_bench/btl.hh b/thirdparty/eigen/bench/btl/generic_bench/btl.hh index 86a8438c..706b00fb 100644 --- a/thirdparty/eigen/bench/btl/generic_bench/btl.hh +++ b/thirdparty/eigen/bench/btl/generic_bench/btl.hh @@ -171,7 +171,7 @@ public: if (_config!=NULL) { std::vector config = BtlString(_config).split(" \t\n"); - for (int i = 0; i BTL_DONT_INLINE void init_matrix(Vector & A, int size){ A.resize(size); - for (int row=0; row(A[row],size,row); } } @@ -50,11 +50,11 @@ BTL_DONT_INLINE void init_matrix(Vector & A, int size){ template BTL_DONT_INLINE void init_matrix_symm(Matrix& A, int size){ A.resize(size); - for (int row=0; row @@ -87,6 +87,48 @@ }; // Portable_Timer +#elif defined(__APPLE__) +#include +#include + + +class Portable_Timer +{ + public: + + Portable_Timer() + { + } + + void start() + { + m_start_time = double(mach_absolute_time())*1e-9;; + + } + + void stop() + { + m_stop_time = double(mach_absolute_time())*1e-9;; + + } + + double elapsed() + { + return user_time(); + } + + double user_time() + { + return m_stop_time - m_start_time; + } + + +private: + + double m_stop_time, m_start_time; + +}; // Portable_Timer (Apple) + #else #include @@ -138,7 +180,7 @@ private: int m_clkid; double m_stop_time, m_start_time; -}; // Portable_Timer +}; // Portable_Timer (Linux) #endif diff --git a/thirdparty/eigen/bench/btl/generic_bench/utils/size_lin_log.hh b/thirdparty/eigen/bench/btl/generic_bench/utils/size_lin_log.hh index bca3932a..bbc9f543 100644 --- a/thirdparty/eigen/bench/btl/generic_bench/utils/size_lin_log.hh +++ b/thirdparty/eigen/bench/btl/generic_bench/utils/size_lin_log.hh @@ -23,7 +23,7 @@ #include "size_log.hh" template -void size_lin_log(const int nb_point, const int size_min, const int size_max, Vector & X) +void size_lin_log(const int nb_point, const int /*size_min*/, const int size_max, Vector & X) { int ten=10; int nine=9; diff --git a/thirdparty/eigen/bench/btl/generic_bench/utils/size_log.hh b/thirdparty/eigen/bench/btl/generic_bench/utils/size_log.hh index 13a3da7a..68945e7c 100644 --- a/thirdparty/eigen/bench/btl/generic_bench/utils/size_log.hh +++ b/thirdparty/eigen/bench/btl/generic_bench/utils/size_log.hh @@ -23,7 +23,7 @@ #include "math.h" // The Vector class must satisfy the following part of STL vector concept : // resize() method -// [] operator for seting element +// [] operator for setting element // the vector element are int compatible. template void size_log(const int nb_point, const int size_min, const int size_max, Vector & X) diff --git a/thirdparty/eigen/bench/btl/generic_bench/utils/xy_file.hh b/thirdparty/eigen/bench/btl/generic_bench/utils/xy_file.hh index 4571bed8..0492faf0 100644 --- a/thirdparty/eigen/bench/btl/generic_bench/utils/xy_file.hh +++ b/thirdparty/eigen/bench/btl/generic_bench/utils/xy_file.hh @@ -55,7 +55,7 @@ bool read_xy_file(const std::string & filename, std::vector & tab_sizes, // The Vector class must satisfy the following part of STL vector concept : // resize() method -// [] operator for seting element +// [] operator for setting element // the vector element must have the << operator define using namespace std; diff --git a/thirdparty/eigen/bench/btl/libs/BLAS/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/BLAS/CMakeLists.txt index de42fe04..f2738f16 100644 --- a/thirdparty/eigen/bench/btl/libs/BLAS/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/libs/BLAS/CMakeLists.txt @@ -5,8 +5,8 @@ if (ATLAS_FOUND) if(BUILD_btl_atlas) target_link_libraries(btl_atlas ${ATLAS_LIBRARIES}) set_target_properties(btl_atlas PROPERTIES COMPILE_FLAGS "-DCBLASNAME=ATLAS -DHAS_LAPACK=1") - endif(BUILD_btl_atlas) -endif (ATLAS_FOUND) + endif() +endif () find_package(MKL) if (MKL_FOUND) @@ -14,31 +14,18 @@ if (MKL_FOUND) if(BUILD_btl_mkl) target_link_libraries(btl_mkl ${MKL_LIBRARIES}) set_target_properties(btl_mkl PROPERTIES COMPILE_FLAGS "-DCBLASNAME=INTEL_MKL -DHAS_LAPACK=1") - endif(BUILD_btl_mkl) -endif (MKL_FOUND) - + endif() +endif () -find_package(GOTO2) -if (GOTO2_FOUND) - btl_add_bench(btl_goto2 main.cpp) - if(BUILD_btl_goto2) - target_link_libraries(btl_goto2 ${GOTO_LIBRARIES} ) - set_target_properties(btl_goto2 PROPERTIES COMPILE_FLAGS "-DCBLASNAME=GOTO2") - endif(BUILD_btl_goto2) -endif (GOTO2_FOUND) -find_package(GOTO) -if (GOTO_FOUND) - if(GOTO2_FOUND) - btl_add_bench(btl_goto main.cpp OFF) - else() - btl_add_bench(btl_goto main.cpp) +find_package(OPENBLAS) +if (OPENBLAS_FOUND) + btl_add_bench(btl_openblas main.cpp) + if(BUILD_btl_openblas) + target_link_libraries(btl_openblas ${OPENBLAS_LIBRARIES} ) + set_target_properties(btl_openblas PROPERTIES COMPILE_FLAGS "-DCBLASNAME=OPENBLAS") endif() - if(BUILD_btl_goto) - target_link_libraries(btl_goto ${GOTO_LIBRARIES} ) - set_target_properties(btl_goto PROPERTIES COMPILE_FLAGS "-DCBLASNAME=GOTO") - endif(BUILD_btl_goto) -endif (GOTO_FOUND) +endif () find_package(ACML) if (ACML_FOUND) @@ -46,8 +33,8 @@ if (ACML_FOUND) if(BUILD_btl_acml) target_link_libraries(btl_acml ${ACML_LIBRARIES} ) set_target_properties(btl_acml PROPERTIES COMPILE_FLAGS "-DCBLASNAME=ACML -DHAS_LAPACK=1") - endif(BUILD_btl_acml) -endif (ACML_FOUND) + endif() +endif () if(Eigen_SOURCE_DIR AND CMAKE_Fortran_COMPILER_WORKS) # we are inside Eigen and blas/lapack interface is compilable diff --git a/thirdparty/eigen/bench/btl/libs/BLAS/blas_interface_impl.hh b/thirdparty/eigen/bench/btl/libs/BLAS/blas_interface_impl.hh index 0e84df03..9e0a6490 100644 --- a/thirdparty/eigen/bench/btl/libs/BLAS/blas_interface_impl.hh +++ b/thirdparty/eigen/bench/btl/libs/BLAS/blas_interface_impl.hh @@ -46,9 +46,9 @@ public : BLAS_FUNC(gemm)(¬rans,¬rans,&N,&N,&N,&fone,A,&N,B,&N,&fzero,X,&N); } -// static inline void ata_product(gene_matrix & A, gene_matrix & X, int N){ -// ssyrk_(&lower,&trans,&N,&N,&fone,A,&N,&fzero,X,&N); -// } + static inline void ata_product(gene_matrix & A, gene_matrix & X, int N){ + BLAS_FUNC(syrk)(&lower,&trans,&N,&N,&fone,A,&N,&fzero,X,&N); + } static inline void aat_product(gene_matrix & A, gene_matrix & X, int N){ BLAS_FUNC(syrk)(&lower,¬rans,&N,&N,&fone,A,&N,&fzero,X,&N); @@ -75,7 +75,6 @@ public : static inline void partial_lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ int N2 = N*N; BLAS_FUNC(copy)(&N2, X, &intone, C, &intone); - char uplo = 'L'; int info = 0; int * ipiv = (int*)alloca(sizeof(int)*N); BLAS_FUNC(getrf)(&N, &N, C, &N, ipiv, &info); @@ -92,7 +91,7 @@ public : BLAS_FUNC(trsm)(&right, &lower, ¬rans, &nonunit, &N, &N, &fone, L, &N, X, &N); } - static inline void trmm(gene_matrix & A, gene_matrix & B, gene_matrix & X, int N){ + static inline void trmm(gene_matrix & A, gene_matrix & B, gene_matrix & /*X*/, int N){ BLAS_FUNC(trmm)(&left, &lower, ¬rans,&nonunit, &N,&N,&fone,A,&N,B,&N); } @@ -101,7 +100,6 @@ public : static inline void lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ int N2 = N*N; BLAS_FUNC(copy)(&N2, X, &intone, C, &intone); - char uplo = 'L'; int info = 0; int * ipiv = (int*)alloca(sizeof(int)*N); int * jpiv = (int*)alloca(sizeof(int)*N); @@ -134,8 +132,6 @@ public : } char uplo = 'U'; int info = 0; - int ilo = 1; - int ihi = N; int bsize = 64; int worksize = N*bsize; SCALAR* d = new SCALAR[3*N+worksize]; diff --git a/thirdparty/eigen/bench/btl/libs/BLAS/c_interface_base.h b/thirdparty/eigen/bench/btl/libs/BLAS/c_interface_base.h index 515d8dcf..de613803 100644 --- a/thirdparty/eigen/bench/btl/libs/BLAS/c_interface_base.h +++ b/thirdparty/eigen/bench/btl/libs/BLAS/c_interface_base.h @@ -17,12 +17,12 @@ template class c_interface_base typedef real* gene_matrix; typedef real* gene_vector; - static void free_matrix(gene_matrix & A, int N){ - delete A; + static void free_matrix(gene_matrix & A, int /*N*/){ + delete[] A; } static void free_vector(gene_vector & B){ - delete B; + delete[] B; } static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ diff --git a/thirdparty/eigen/bench/btl/libs/BLAS/main.cpp b/thirdparty/eigen/bench/btl/libs/BLAS/main.cpp index 8347c9f0..fd991490 100644 --- a/thirdparty/eigen/bench/btl/libs/BLAS/main.cpp +++ b/thirdparty/eigen/bench/btl/libs/BLAS/main.cpp @@ -48,7 +48,7 @@ int main() bench > >(MIN_AXPY,MAX_AXPY,NB_POINT); bench > >(MIN_MM,MAX_MM,NB_POINT); -// bench > >(MIN_MM,MAX_MM,NB_POINT); + bench > >(MIN_MM,MAX_MM,NB_POINT); bench > >(MIN_MM,MAX_MM,NB_POINT); bench > >(MIN_MM,MAX_MM,NB_POINT); @@ -56,13 +56,13 @@ int main() bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); #ifdef HAS_LAPACK - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); +// bench > >(MIN_LU,MAX_LU,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); #endif //bench > >(MIN_LU,MAX_LU,NB_POINT); diff --git a/thirdparty/eigen/bench/btl/libs/STL/STL_interface.hh b/thirdparty/eigen/bench/btl/libs/STL/STL_interface.hh index 93e76bd5..5b391c6e 100644 --- a/thirdparty/eigen/bench/btl/libs/STL/STL_interface.hh +++ b/thirdparty/eigen/bench/btl/libs/STL/STL_interface.hh @@ -44,9 +44,9 @@ public : return "STL"; } - static void free_matrix(gene_matrix & A, int N){} + static void free_matrix(gene_matrix & /*A*/, int /*N*/){} - static void free_vector(gene_vector & B){} + static void free_vector(gene_vector & /*B*/){} static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ A = A_stl; @@ -78,18 +78,21 @@ public : cible[i][j]=source[i][j]; } -// static inline void ata_product(const gene_matrix & A, gene_matrix & X, int N) -// { -// real somme; -// for (int j=0;j=j) + { + for (int k=0;k +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#ifndef BLAZE_INTERFACE_HH +#define BLAZE_INTERFACE_HH + +#include +#include +#include +// using namespace blaze; + +#include + +template +class blaze_interface { + +public : + + typedef real real_type ; + + typedef std::vector stl_vector; + typedef std::vector stl_matrix; + + typedef blaze::DynamicMatrix gene_matrix; + typedef blaze::DynamicVector gene_vector; + + static inline std::string name() { return "blaze"; } + + static void free_matrix(gene_matrix & A, int N){ + return ; + } + + static void free_vector(gene_vector & B){ + return ; + } + + static inline void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.resize(A_stl[0].size(), A_stl.size()); + + for (int j=0; j ipvt(N); +// lu_factor(R, ipvt); +// } + +// static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector & X, int N){ +// X = lower_trisolve(L, B); +// } + + static inline void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + cible = source; + } + + static inline void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + cible = source; + } + +}; + +#endif diff --git a/thirdparty/eigen/bench/btl/libs/blaze/main.cpp b/thirdparty/eigen/bench/btl/libs/blaze/main.cpp new file mode 100644 index 00000000..ccae0cbd --- /dev/null +++ b/thirdparty/eigen/bench/btl/libs/blaze/main.cpp @@ -0,0 +1,40 @@ +//===================================================== +// Copyright (C) 2008 Gael Guennebaud +//===================================================== +// +// This program is free software; you can redistribute it and/or +// modify it under the terms of the GNU General Public License +// as published by the Free Software Foundation; either version 2 +// of the License, or (at your option) any later version. +// +// This program is distributed in the hope that it will be useful, +// but WITHOUT ANY WARRANTY; without even the implied warranty of +// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the +// GNU General Public License for more details. +// You should have received a copy of the GNU General Public License +// along with this program; if not, write to the Free Software +// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. +// +#include "utilities.h" +#include "blaze_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + + bench > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + bench > >(MIN_MV,MAX_MV,NB_POINT); + bench > >(MIN_MV,MAX_MV,NB_POINT); + bench > >(MIN_MM,MAX_MM,NB_POINT); + bench > >(MIN_MM,MAX_MM,NB_POINT); + bench > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} + + diff --git a/thirdparty/eigen/bench/btl/libs/blitz/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/blitz/CMakeLists.txt index 880ab733..e203c815 100644 --- a/thirdparty/eigen/bench/btl/libs/blitz/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/libs/blitz/CMakeLists.txt @@ -7,11 +7,11 @@ if (BLITZ_FOUND) btl_add_bench(btl_blitz btl_blitz.cpp) if (BUILD_btl_blitz) target_link_libraries(btl_blitz ${BLITZ_LIBRARIES}) - endif (BUILD_btl_blitz) + endif () btl_add_bench(btl_tiny_blitz btl_tiny_blitz.cpp OFF) if (BUILD_btl_tiny_blitz) target_link_libraries(btl_tiny_blitz ${BLITZ_LIBRARIES}) - endif (BUILD_btl_tiny_blitz) + endif () -endif (BLITZ_FOUND) +endif () diff --git a/thirdparty/eigen/bench/btl/libs/eigen2/eigen2_interface.hh b/thirdparty/eigen/bench/btl/libs/eigen2/eigen2_interface.hh index 47fe5813..1deabdae 100644 --- a/thirdparty/eigen/bench/btl/libs/eigen2/eigen2_interface.hh +++ b/thirdparty/eigen/bench/btl/libs/eigen2/eigen2_interface.hh @@ -47,7 +47,7 @@ public : { #if defined(EIGEN_VECTORIZE_SSE) if (SIZE==Dynamic) return "eigen2"; else return "tiny_eigen2"; - #elif defined(EIGEN_VECTORIZE_ALTIVEC) + #elif defined(EIGEN_VECTORIZE_ALTIVEC) || defined(EIGEN_VECTORIZE_VSX) if (SIZE==Dynamic) return "eigen2"; else return "tiny_eigen2"; #else if (SIZE==Dynamic) return "eigen2_novec"; else return "tiny_eigen2_novec"; diff --git a/thirdparty/eigen/bench/btl/libs/eigen3/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/eigen3/CMakeLists.txt index 00cae23d..06a72b4f 100644 --- a/thirdparty/eigen/bench/btl/libs/eigen3/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/libs/eigen3/CMakeLists.txt @@ -47,9 +47,9 @@ if (EIGEN3_FOUND) # if(BUILD_btl_eigen3_adv) # target_link_libraries(btl_eigen3_adv ${MKL_LIBRARIES}) -# endif(BUILD_btl_eigen3_adv) +# endif() - endif(NOT BTL_NOVEC) + endif() btl_add_bench(btl_tiny_eigen3 btl_tiny_eigen3.cpp OFF) @@ -59,7 +59,7 @@ if (EIGEN3_FOUND) if(BUILD_btl_tiny_eigen3_novec) btl_add_target_property(btl_tiny_eigen3_novec COMPILE_FLAGS "-DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=eigen3_tiny_novec") - endif(BUILD_btl_tiny_eigen3_novec) - endif(NOT BTL_NOVEC) + endif() + endif() -endif (EIGEN3_FOUND) +endif () diff --git a/thirdparty/eigen/bench/btl/libs/eigen3/eigen3_interface.hh b/thirdparty/eigen/bench/btl/libs/eigen3/eigen3_interface.hh index 31bcc1f9..2e302d07 100644 --- a/thirdparty/eigen/bench/btl/libs/eigen3/eigen3_interface.hh +++ b/thirdparty/eigen/bench/btl/libs/eigen3/eigen3_interface.hh @@ -45,15 +45,15 @@ public : return EIGEN_MAKESTRING(BTL_PREFIX); } - static void free_matrix(gene_matrix & A, int N) {} + static void free_matrix(gene_matrix & /*A*/, int /*N*/) {} - static void free_vector(gene_vector & B) {} + static void free_vector(gene_vector & /*B*/) {} static BTL_DONT_INLINE void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ A.resize(A_stl[0].size(), A_stl.size()); - for (int j=0; j().setZero(); + X.template selfadjointView().rankUpdate(A.transpose()); + } - static inline void aat_product(const gene_matrix & A, gene_matrix & X, int N){ + static inline void aat_product(const gene_matrix & A, gene_matrix & X, int /*N*/){ X.template triangularView().setZero(); X.template selfadjointView().rankUpdate(A); } - static inline void matrix_vector_product(const gene_matrix & A, const gene_vector & B, gene_vector & X, int N){ + static inline void matrix_vector_product(const gene_matrix & A, const gene_vector & B, gene_vector & X, int /*N*/){ X.noalias() = A*B; } - static inline void symv(const gene_matrix & A, const gene_vector & B, gene_vector & X, int N){ + static inline void symv(const gene_matrix & A, const gene_vector & B, gene_vector & X, int /*N*/){ X.noalias() = (A.template selfadjointView() * B); // internal::product_selfadjoint_vector(N,A.data(),N, B.data(), 1, X.data(), 1); } @@ -155,54 +157,54 @@ public : } } - static EIGEN_DONT_INLINE void syr2(gene_matrix & A, gene_vector & X, gene_vector & Y, int N){ + static EIGEN_DONT_INLINE void syr2(gene_matrix & A, gene_vector & X, gene_vector & Y, int N){ // internal::product_selfadjoint_rank2_update(N,A.data(),N, X.data(), 1, Y.data(), 1, -1); for(int j=0; j(c,s)); } - static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ + static inline void atv_product(gene_matrix & A, gene_vector & B, gene_vector & X, int /*N*/){ X.noalias() = (A.transpose()*B); } - static inline void axpy(real coef, const gene_vector & X, gene_vector & Y, int N){ + static inline void axpy(real coef, const gene_vector & X, gene_vector & Y, int /*N*/){ Y += coef * X; } - static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int N){ + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int /*N*/){ Y = a*X + b*Y; } - static EIGEN_DONT_INLINE void copy_matrix(const gene_matrix & source, gene_matrix & cible, int N){ + static EIGEN_DONT_INLINE void copy_matrix(const gene_matrix & source, gene_matrix & cible, int /*N*/){ cible = source; } - static EIGEN_DONT_INLINE void copy_vector(const gene_vector & source, gene_vector & cible, int N){ + static EIGEN_DONT_INLINE void copy_vector(const gene_vector & source, gene_vector & cible, int /*N*/){ cible = source; } - static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector& X, int N){ + static inline void trisolve_lower(const gene_matrix & L, const gene_vector& B, gene_vector& X, int /*N*/){ X = L.template triangularView().solve(B); } - static inline void trisolve_lower_matrix(const gene_matrix & L, const gene_matrix& B, gene_matrix& X, int N){ + static inline void trisolve_lower_matrix(const gene_matrix & L, const gene_matrix& B, gene_matrix& X, int /*N*/){ X = L.template triangularView().solve(B); } - static inline void trmm(const gene_matrix & L, const gene_matrix& B, gene_matrix& X, int N){ + static inline void trmm(const gene_matrix & L, const gene_matrix& B, gene_matrix& X, int /*N*/){ X.noalias() = L.template triangularView() * B; } - static inline void cholesky(const gene_matrix & X, gene_matrix & C, int N){ + static inline void cholesky(const gene_matrix & X, gene_matrix & C, int /*N*/){ C = X; internal::llt_inplace::blocked(C); //C = X.llt().matrixL(); @@ -211,11 +213,11 @@ public : // Cholesky::computeInPlaceBlock(C); } - static inline void lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ + static inline void lu_decomp(const gene_matrix & X, gene_matrix & C, int /*N*/){ C = X.fullPivLu().matrixLU(); } - static inline void partial_lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ + static inline void partial_lu_decomp(const gene_matrix & X, gene_matrix & C, int N){ Matrix piv(N); DenseIndex nb; C = X; @@ -223,13 +225,13 @@ public : // C = X.partialPivLu().matrixLU(); } - static inline void tridiagonalization(const gene_matrix & X, gene_matrix & C, int N){ + static inline void tridiagonalization(const gene_matrix & X, gene_matrix & C, int N){ typename Tridiagonalization::CoeffVectorType aux(N-1); C = X; internal::tridiagonalization_inplace(C, aux); } - static inline void hessenberg(const gene_matrix & X, gene_matrix & C, int N){ + static inline void hessenberg(const gene_matrix & X, gene_matrix & C, int /*N*/){ C = HessenbergDecomposition(X).packedMatrix(); } diff --git a/thirdparty/eigen/bench/btl/libs/eigen3/main_adv.cpp b/thirdparty/eigen/bench/btl/libs/eigen3/main_adv.cpp index efe5857e..95865357 100644 --- a/thirdparty/eigen/bench/btl/libs/eigen3/main_adv.cpp +++ b/thirdparty/eigen/bench/btl/libs/eigen3/main_adv.cpp @@ -29,14 +29,14 @@ BTL_MAIN; int main() { - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); +// bench > >(MIN_LU,MAX_LU,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); - bench > >(MIN_MM,MAX_MM,NB_POINT); +// bench > >(MIN_LU,MAX_LU,NB_POINT); + bench > >(MIN_LU,MAX_LU,NB_POINT); return 0; } diff --git a/thirdparty/eigen/bench/btl/libs/eigen3/main_matmat.cpp b/thirdparty/eigen/bench/btl/libs/eigen3/main_matmat.cpp index 926fa2b0..052810a1 100644 --- a/thirdparty/eigen/bench/btl/libs/eigen3/main_matmat.cpp +++ b/thirdparty/eigen/bench/btl/libs/eigen3/main_matmat.cpp @@ -25,7 +25,7 @@ BTL_MAIN; int main() { bench > >(MIN_MM,MAX_MM,NB_POINT); -// bench > >(MIN_MM,MAX_MM,NB_POINT); + bench > >(MIN_MM,MAX_MM,NB_POINT); bench > >(MIN_MM,MAX_MM,NB_POINT); bench > >(MIN_MM,MAX_MM,NB_POINT); diff --git a/thirdparty/eigen/bench/btl/libs/gmm/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/gmm/CMakeLists.txt index bc258624..0bcb0465 100644 --- a/thirdparty/eigen/bench/btl/libs/gmm/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/libs/gmm/CMakeLists.txt @@ -3,4 +3,4 @@ find_package(GMM) if (GMM_FOUND) include_directories(${GMM_INCLUDES}) btl_add_bench(btl_gmm main.cpp) -endif (GMM_FOUND) +endif () diff --git a/thirdparty/eigen/bench/btl/libs/mtl4/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/mtl4/CMakeLists.txt index 14b47a80..132a501b 100644 --- a/thirdparty/eigen/bench/btl/libs/mtl4/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/libs/mtl4/CMakeLists.txt @@ -3,4 +3,4 @@ find_package(MTL4) if (MTL4_FOUND) include_directories(${MTL4_INCLUDE_DIR}) btl_add_bench(btl_mtl4 main.cpp) -endif (MTL4_FOUND) +endif () diff --git a/thirdparty/eigen/bench/btl/libs/tensors/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/tensors/CMakeLists.txt new file mode 100644 index 00000000..e10a736f --- /dev/null +++ b/thirdparty/eigen/bench/btl/libs/tensors/CMakeLists.txt @@ -0,0 +1,44 @@ + + +if((NOT TENSOR_INCLUDE_DIR) AND Eigen_SOURCE_DIR) + # unless TENSOR_INCLUDE_DIR is defined, let's use current Eigen version + set(TENSOR_INCLUDE_DIR ${Eigen_SOURCE_DIR}) + set(TENSOR_FOUND TRUE) +else() + find_package(Tensor) +endif() + +if (TENSOR_FOUND) + + include_directories(${TENSOR_INCLUDE_DIR}) + btl_add_bench(btl_tensor_linear main_linear.cpp) + btl_add_bench(btl_tensor_vecmat main_vecmat.cpp) + btl_add_bench(btl_tensor_matmat main_matmat.cpp) + + btl_add_target_property(btl_tensor_linear COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=tensor") + btl_add_target_property(btl_tensor_vecmat COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=tensor") + btl_add_target_property(btl_tensor_matmat COMPILE_FLAGS "-fno-exceptions -DBTL_PREFIX=tensor") + + option(BTL_BENCH_NOGCCVEC "also bench Eigen explicit vec without GCC's auto vec" OFF) + if(CMAKE_COMPILER_IS_GNUCXX AND BTL_BENCH_NOGCCVEC) + btl_add_bench(btl_tensor_nogccvec_linear main_linear.cpp) + btl_add_bench(btl_tensor_nogccvec_vecmat main_vecmat.cpp) + btl_add_bench(btl_tensor_nogccvec_matmat main_matmat.cpp) + + btl_add_target_property(btl_tensor_nogccvec_linear COMPILE_FLAGS "-fno-exceptions -fno-tree-vectorize -DBTL_PREFIX=tensor_nogccvec") + btl_add_target_property(btl_tensor_nogccvec_vecmat COMPILE_FLAGS "-fno-exceptions -fno-tree-vectorize -DBTL_PREFIX=tensor_nogccvec") + btl_add_target_property(btl_tensor_nogccvec_matmat COMPILE_FLAGS "-fno-exceptions -fno-tree-vectorize -DBTL_PREFIX=tensor_nogccvec") + endif() + + + if(NOT BTL_NOVEC) + btl_add_bench(btl_tensor_novec_linear main_linear.cpp OFF) + btl_add_bench(btl_tensor_novec_vecmat main_vecmat.cpp OFF) + btl_add_bench(btl_tensor_novec_matmat main_matmat.cpp OFF) + btl_add_target_property(btl_tensor_novec_linear COMPILE_FLAGS "-fno-exceptions -DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=tensor_novec") + btl_add_target_property(btl_tensor_novec_vecmat COMPILE_FLAGS "-fno-exceptions -DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=tensor_novec") + btl_add_target_property(btl_tensor_novec_matmat COMPILE_FLAGS "-fno-exceptions -DEIGEN_DONT_VECTORIZE -DBTL_PREFIX=tensor_novec") + + endif() + +endif () diff --git a/thirdparty/eigen/bench/btl/libs/tensors/main_linear.cpp b/thirdparty/eigen/bench/btl/libs/tensors/main_linear.cpp new file mode 100644 index 00000000..e257f1e7 --- /dev/null +++ b/thirdparty/eigen/bench/btl/libs/tensors/main_linear.cpp @@ -0,0 +1,23 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2014 Benoit Steiner +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "utilities.h" +#include "tensor_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench > >(MIN_AXPY,MAX_AXPY,NB_POINT); + bench > >(MIN_AXPY,MAX_AXPY,NB_POINT); + + return 0; +} diff --git a/thirdparty/eigen/bench/btl/libs/tensors/main_matmat.cpp b/thirdparty/eigen/bench/btl/libs/tensors/main_matmat.cpp new file mode 100644 index 00000000..675fcfc6 --- /dev/null +++ b/thirdparty/eigen/bench/btl/libs/tensors/main_matmat.cpp @@ -0,0 +1,21 @@ +//===================================================== +// Copyright (C) 2014 Benoit Steiner +//===================================================== +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +// +#include "utilities.h" +#include "tensor_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench > >(MIN_MM,MAX_MM,NB_POINT); + + return 0; +} diff --git a/thirdparty/eigen/bench/btl/libs/tensors/main_vecmat.cpp b/thirdparty/eigen/bench/btl/libs/tensors/main_vecmat.cpp new file mode 100644 index 00000000..1af00c81 --- /dev/null +++ b/thirdparty/eigen/bench/btl/libs/tensors/main_vecmat.cpp @@ -0,0 +1,21 @@ +//===================================================== +// Copyright (C) 2014 Benoit Steiner +//===================================================== +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +// +#include "utilities.h" +#include "tensor_interface.hh" +#include "bench.hh" +#include "basic_actions.hh" + +BTL_MAIN; + +int main() +{ + bench > >(MIN_MV,MAX_MV,NB_POINT); + + return 0; +} diff --git a/thirdparty/eigen/bench/btl/libs/tensors/tensor_interface.hh b/thirdparty/eigen/bench/btl/libs/tensors/tensor_interface.hh new file mode 100644 index 00000000..97b8e0f0 --- /dev/null +++ b/thirdparty/eigen/bench/btl/libs/tensors/tensor_interface.hh @@ -0,0 +1,105 @@ +//===================================================== +// Copyright (C) 2014 Benoit Steiner +//===================================================== +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +// +#ifndef TENSOR_INTERFACE_HH +#define TENSOR_INTERFACE_HH + +#include +#include +#include "btl.hh" + +using namespace Eigen; + +template +class tensor_interface +{ +public : + typedef real real_type; + typedef typename Eigen::Tensor::Index Index; + + typedef std::vector stl_vector; + typedef std::vector stl_matrix; + + typedef Eigen::Tensor gene_matrix; + typedef Eigen::Tensor gene_vector; + + + static inline std::string name( void ) + { + return EIGEN_MAKESTRING(BTL_PREFIX); + } + + static void free_matrix(gene_matrix & /*A*/, int /*N*/) {} + + static void free_vector(gene_vector & /*B*/) {} + + static BTL_DONT_INLINE void matrix_from_stl(gene_matrix & A, stl_matrix & A_stl){ + A.resize(Eigen::array(A_stl[0].size(), A_stl.size())); + + for (unsigned int j=0; j(i,j)) = A_stl[j][i]; + } + } + } + + static BTL_DONT_INLINE void vector_from_stl(gene_vector & B, stl_vector & B_stl){ + B.resize(B_stl.size()); + + for (unsigned int i=0; i(i,j)); + } + } + } + + static inline void matrix_matrix_product(const gene_matrix & A, const gene_matrix & B, gene_matrix & X, int /*N*/){ + typedef typename Eigen::Tensor::DimensionPair DimPair; + const Eigen::array dims(DimPair(1, 0)); + X/*.noalias()*/ = A.contract(B, dims); + } + + static inline void matrix_vector_product(const gene_matrix & A, const gene_vector & B, gene_vector & X, int /*N*/){ + typedef typename Eigen::Tensor::DimensionPair DimPair; + const Eigen::array dims(DimPair(1, 0)); + X/*.noalias()*/ = A.contract(B, dims); + } + + static inline void axpy(real coef, const gene_vector & X, gene_vector & Y, int /*N*/){ + Y += X.constant(coef) * X; + } + + static inline void axpby(real a, const gene_vector & X, real b, gene_vector & Y, int /*N*/){ + Y = X.constant(a)*X + Y.constant(b)*Y; + } + + static EIGEN_DONT_INLINE void copy_matrix(const gene_matrix & source, gene_matrix & cible, int /*N*/){ + cible = source; + } + + static EIGEN_DONT_INLINE void copy_vector(const gene_vector & source, gene_vector & cible, int /*N*/){ + cible = source; + } +}; + +#endif diff --git a/thirdparty/eigen/bench/btl/libs/tvmet/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/tvmet/CMakeLists.txt index 25b565b9..e7376972 100644 --- a/thirdparty/eigen/bench/btl/libs/tvmet/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/libs/tvmet/CMakeLists.txt @@ -3,4 +3,4 @@ find_package(Tvmet) if (TVMET_FOUND) include_directories(${TVMET_INCLUDE_DIR}) btl_add_bench(btl_tvmet main.cpp OFF) -endif (TVMET_FOUND) +endif () diff --git a/thirdparty/eigen/bench/btl/libs/ublas/CMakeLists.txt b/thirdparty/eigen/bench/btl/libs/ublas/CMakeLists.txt index bdb58bea..5accf5b8 100644 --- a/thirdparty/eigen/bench/btl/libs/ublas/CMakeLists.txt +++ b/thirdparty/eigen/bench/btl/libs/ublas/CMakeLists.txt @@ -4,4 +4,4 @@ if (Boost_FOUND) include_directories(${Boost_INCLUDE_DIRS}) include_directories(${Boost_INCLUDES}) btl_add_bench(btl_ublas main.cpp) -endif (Boost_FOUND) +endif () diff --git a/thirdparty/eigen/bench/btl/libs/ublas/ublas_interface.hh b/thirdparty/eigen/bench/btl/libs/ublas/ublas_interface.hh index 95cad519..f59b7cf2 100644 --- a/thirdparty/eigen/bench/btl/libs/ublas/ublas_interface.hh +++ b/thirdparty/eigen/bench/btl/libs/ublas/ublas_interface.hh @@ -100,7 +100,7 @@ public : Y+=coef*X; } - // alias free assignements + // alias free assignments static inline void matrix_vector_product(gene_matrix & A, gene_vector & B, gene_vector & X, int N){ X.assign(prod(A,B)); diff --git a/thirdparty/eigen/bench/dense_solvers.cpp b/thirdparty/eigen/bench/dense_solvers.cpp new file mode 100644 index 00000000..24343dcd --- /dev/null +++ b/thirdparty/eigen/bench/dense_solvers.cpp @@ -0,0 +1,186 @@ +#include +#include "BenchTimer.h" +#include +#include +#include +#include +#include +using namespace Eigen; + +std::map > results; +std::vector labels; +std::vector sizes; + +template +EIGEN_DONT_INLINE +void compute_norm_equation(Solver &solver, const MatrixType &A) { + if(A.rows()!=A.cols()) + solver.compute(A.transpose()*A); + else + solver.compute(A); +} + +template +EIGEN_DONT_INLINE +void compute(Solver &solver, const MatrixType &A) { + solver.compute(A); +} + +template +void bench(int id, int rows, int size = Size) +{ + typedef Matrix Mat; + typedef Matrix MatDyn; + typedef Matrix MatSquare; + Mat A(rows,size); + A.setRandom(); + if(rows==size) + A = A*A.adjoint(); + BenchTimer t_llt, t_ldlt, t_lu, t_fplu, t_qr, t_cpqr, t_cod, t_fpqr, t_jsvd, t_bdcsvd; + + int svd_opt = ComputeThinU|ComputeThinV; + + int tries = 5; + int rep = 1000/size; + if(rep==0) rep = 1; +// rep = rep*rep; + + LLT llt(size); + LDLT ldlt(size); + PartialPivLU lu(size); + FullPivLU fplu(size,size); + HouseholderQR qr(A.rows(),A.cols()); + ColPivHouseholderQR cpqr(A.rows(),A.cols()); + CompleteOrthogonalDecomposition cod(A.rows(),A.cols()); + FullPivHouseholderQR fpqr(A.rows(),A.cols()); + JacobiSVD jsvd(A.rows(),A.cols()); + BDCSVD bdcsvd(A.rows(),A.cols()); + + BENCH(t_llt, tries, rep, compute_norm_equation(llt,A)); + BENCH(t_ldlt, tries, rep, compute_norm_equation(ldlt,A)); + BENCH(t_lu, tries, rep, compute_norm_equation(lu,A)); + if(size<=1000) + BENCH(t_fplu, tries, rep, compute_norm_equation(fplu,A)); + BENCH(t_qr, tries, rep, compute(qr,A)); + BENCH(t_cpqr, tries, rep, compute(cpqr,A)); + BENCH(t_cod, tries, rep, compute(cod,A)); + if(size*rows<=10000000) + BENCH(t_fpqr, tries, rep, compute(fpqr,A)); + if(size<500) // JacobiSVD is really too slow for too large matrices + BENCH(t_jsvd, tries, rep, jsvd.compute(A,svd_opt)); +// if(size*rows<=20000000) + BENCH(t_bdcsvd, tries, rep, bdcsvd.compute(A,svd_opt)); + + results["LLT"][id] = t_llt.best(); + results["LDLT"][id] = t_ldlt.best(); + results["PartialPivLU"][id] = t_lu.best(); + results["FullPivLU"][id] = t_fplu.best(); + results["HouseholderQR"][id] = t_qr.best(); + results["ColPivHouseholderQR"][id] = t_cpqr.best(); + results["CompleteOrthogonalDecomposition"][id] = t_cod.best(); + results["FullPivHouseholderQR"][id] = t_fpqr.best(); + results["JacobiSVD"][id] = t_jsvd.best(); + results["BDCSVD"][id] = t_bdcsvd.best(); +} + + +int main() +{ + labels.push_back("LLT"); + labels.push_back("LDLT"); + labels.push_back("PartialPivLU"); + labels.push_back("FullPivLU"); + labels.push_back("HouseholderQR"); + labels.push_back("ColPivHouseholderQR"); + labels.push_back("CompleteOrthogonalDecomposition"); + labels.push_back("FullPivHouseholderQR"); + labels.push_back("JacobiSVD"); + labels.push_back("BDCSVD"); + + for(int i=0; i(k,sizes[k](0),sizes[k](1)); + } + + cout.width(32); + cout << "solver/size"; + cout << " "; + for(int k=0; k=1e6) cout << "-"; + else cout << r(k); + cout << " "; + } + cout << endl; + } + + // HTML output + cout << "" << endl; + cout << "" << endl; + for(int k=0; k" << sizes[k](0) << "x" << sizes[k](1) << ""; + cout << "" << endl; + for(int i=0; i"; + ArrayXf r = (results[labels[i]]*100000.f).floor()/100.f; + for(int k=0; k=1e6) cout << ""; + else + { + cout << ""; + } + } + cout << "" << endl; + } + cout << "
solver/size
" << labels[i] << "-" << r(k); + if(i>0) + cout << " (x" << numext::round(10.f*results[labels[i]](k)/results["LLT"](k))/10.f << ")"; + if(i<4 && sizes[k](0)!=sizes[k](1)) + cout << " *"; + cout << "
" << endl; + +// cout << "LLT (ms) " << (results["LLT"]*1000.).format(fmt) << "\n"; +// cout << "LDLT (%) " << (results["LDLT"]/results["LLT"]).format(fmt) << "\n"; +// cout << "PartialPivLU (%) " << (results["PartialPivLU"]/results["LLT"]).format(fmt) << "\n"; +// cout << "FullPivLU (%) " << (results["FullPivLU"]/results["LLT"]).format(fmt) << "\n"; +// cout << "HouseholderQR (%) " << (results["HouseholderQR"]/results["LLT"]).format(fmt) << "\n"; +// cout << "ColPivHouseholderQR (%) " << (results["ColPivHouseholderQR"]/results["LLT"]).format(fmt) << "\n"; +// cout << "CompleteOrthogonalDecomposition (%) " << (results["CompleteOrthogonalDecomposition"]/results["LLT"]).format(fmt) << "\n"; +// cout << "FullPivHouseholderQR (%) " << (results["FullPivHouseholderQR"]/results["LLT"]).format(fmt) << "\n"; +// cout << "JacobiSVD (%) " << (results["JacobiSVD"]/results["LLT"]).format(fmt) << "\n"; +// cout << "BDCSVD (%) " << (results["BDCSVD"]/results["LLT"]).format(fmt) << "\n"; +} diff --git a/thirdparty/eigen/bench/eig33.cpp b/thirdparty/eigen/bench/eig33.cpp index 1608b999..f003d8a5 100644 --- a/thirdparty/eigen/bench/eig33.cpp +++ b/thirdparty/eigen/bench/eig33.cpp @@ -50,7 +50,7 @@ inline void computeRoots(const Matrix& m, Roots& roots) { typedef typename Matrix::Scalar Scalar; const Scalar s_inv3 = 1.0/3.0; - const Scalar s_sqrt3 = internal::sqrt(Scalar(3.0)); + const Scalar s_sqrt3 = std::sqrt(Scalar(3.0)); // The characteristic equation is x^3 - c2*x^2 + c1*x - c0 = 0. The // eigenvalues are the roots to this equation, all guaranteed to be @@ -73,23 +73,13 @@ inline void computeRoots(const Matrix& m, Roots& roots) q = Scalar(0); // Compute the eigenvalues by solving for the roots of the polynomial. - Scalar rho = internal::sqrt(-a_over_3); - Scalar theta = std::atan2(internal::sqrt(-q),half_b)*s_inv3; - Scalar cos_theta = internal::cos(theta); - Scalar sin_theta = internal::sin(theta); - roots(0) = c2_over_3 + Scalar(2)*rho*cos_theta; - roots(1) = c2_over_3 - rho*(cos_theta + s_sqrt3*sin_theta); - roots(2) = c2_over_3 - rho*(cos_theta - s_sqrt3*sin_theta); - - // Sort in increasing order. - if (roots(0) >= roots(1)) - std::swap(roots(0),roots(1)); - if (roots(1) >= roots(2)) - { - std::swap(roots(1),roots(2)); - if (roots(0) >= roots(1)) - std::swap(roots(0),roots(1)); - } + Scalar rho = std::sqrt(-a_over_3); + Scalar theta = std::atan2(std::sqrt(-q),half_b)*s_inv3; + Scalar cos_theta = std::cos(theta); + Scalar sin_theta = std::sin(theta); + roots(2) = c2_over_3 + Scalar(2)*rho*cos_theta; + roots(0) = c2_over_3 - rho*(cos_theta + s_sqrt3*sin_theta); + roots(1) = c2_over_3 - rho*(cos_theta - s_sqrt3*sin_theta); } template @@ -99,16 +89,19 @@ void eigen33(const Matrix& mat, Matrix& evecs, Vector& evals) // Scale the matrix so its entries are in [-1,1]. The scaling is applied // only when at least one matrix entry has magnitude larger than 1. - Scalar scale = mat.cwiseAbs()/*.template triangularView()*/.maxCoeff(); + Scalar shift = mat.trace()/3; + Matrix scaledMat = mat; + scaledMat.diagonal().array() -= shift; + Scalar scale = scaledMat.cwiseAbs()/*.template triangularView()*/.maxCoeff(); scale = std::max(scale,Scalar(1)); - Matrix scaledMat = mat / scale; + scaledMat/=scale; // Compute the eigenvalues // scaledMat.setZero(); computeRoots(scaledMat,evals); // compute the eigen vectors - // **here we assume 3 differents eigenvalues** + // **here we assume 3 different eigenvalues** // "optimized version" which appears to be slower with gcc! // Vector base; @@ -166,6 +159,7 @@ void eigen33(const Matrix& mat, Matrix& evecs, Vector& evals) // Rescale back to the original size. evals *= scale; + evals.array()+=shift; } int main() @@ -173,24 +167,29 @@ int main() BenchTimer t; int tries = 10; int rep = 400000; - typedef Matrix3f Mat; - typedef Vector3f Vec; + typedef Matrix3d Mat; + typedef Vector3d Vec; Mat A = Mat::Random(3,3); A = A.adjoint() * A; +// Mat Q = A.householderQr().householderQ(); +// A = Q * Vec(2.2424567,2.2424566,7.454353).asDiagonal() * Q.transpose(); SelfAdjointEigenSolver eig(A); BENCH(t, tries, rep, eig.compute(A)); - std::cout << "Eigen: " << t.best() << "s\n"; + std::cout << "Eigen iterative: " << t.best() << "s\n"; + + BENCH(t, tries, rep, eig.computeDirect(A)); + std::cout << "Eigen direct : " << t.best() << "s\n"; Mat evecs; Vec evals; BENCH(t, tries, rep, eigen33(A,evecs,evals)); std::cout << "Direct: " << t.best() << "s\n\n"; - std::cerr << "Eigenvalue/eigenvector diffs:\n"; - std::cerr << (evals - eig.eigenvalues()).transpose() << "\n"; - for(int k=0;k<3;++k) - if(evecs.col(k).dot(eig.eigenvectors().col(k))<0) - evecs.col(k) = -evecs.col(k); - std::cerr << evecs - eig.eigenvectors() << "\n\n"; +// std::cerr << "Eigenvalue/eigenvector diffs:\n"; +// std::cerr << (evals - eig.eigenvalues()).transpose() << "\n"; +// for(int k=0;k<3;++k) +// if(evecs.col(k).dot(eig.eigenvectors().col(k))<0) +// evecs.col(k) = -evecs.col(k); +// std::cerr << evecs - eig.eigenvectors() << "\n\n"; } diff --git a/thirdparty/eigen/bench/perf_monitoring/changesets.txt b/thirdparty/eigen/bench/perf_monitoring/changesets.txt new file mode 100644 index 00000000..efdd9a0f --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/changesets.txt @@ -0,0 +1,95 @@ +Load hg-to-git hash maps from ./eigen_git/.git/ +#3.0.1 +#3.1.1 +#3.2.0 +3.2.4 +#574a7621809fe +58964a85800bd # introduce AVX +#589cbd7e98174 # merge +589db7d49efbb # introduce FMA +#590a078f442a3 # complex and AVX +590a419cea4a0 # improve packing with ptranspose +#59251e85c936d # merge +#592e497a27ddc +593d5a795f673 # New gebp kernel: up to 3 packets x 4 register-level blocks +#5942c3c95990d # merge +#596c9788d55b9 # Disable 3pX4 kernel on Altivec +#5999aa3dc4e21 # merge +6209452eb38f8 # before-evaluators +#6333eba5e1101 # Implement evaluator for sparse outer products +#663b9d314ae19 +#6655ef95fabee # Properly detect FMA support on ARM +#667fe25f3b8e3 # FMA has been wrongly disabled +#668409547a0c8 +#6694304c73542 # merge default to tensors +#67216047c8d4a # merge default to tensors +#67410a79ca3a3 # merge default to tensors +#674b7271dffb5 # Generalized the gebp apis +676bfdd9f3ac9 # Made the blocking computation aware of the l3 cache;
Also optimized the blocking parameters to take
into account the number of threads used for a computation. +6782dde63499c # generalized gemv +6799f98650d0a # ensured that contractions that can be reduced to a matrix vector product +#6840918c51e60 # merge tensor +684e972b55ec4 # change prefetching in gebp +#68598604576d1 # merge index conversion +68963eb0f6fe6 # clean blocking size computation +689db05f2d01e # rotating kernel for ARM only +#6901b7e12847d # result_of +69226275b250a # fix prefetching change for ARM +692692136350b # prefetching +693a8ad8887bf # blocking size strategy +693bcf9bb5c1f # avoid redundant pack_rhs +6987550107028 # dynamic loop swapping +69858740ce4c6 # rm dynamic loop swapping,
adjust lhs's micro panel height to fully exploit L1 cache +698cd3bbffa73 # blocking heuristic:
block on the rhs in L1 if the lhs fit in L1. +701488c15615a # organize a little our default cache sizes,
and use a saner default L1 outside of x86 (10% faster on Nexus 5) +701e56aabf205 # Refactor computeProductBlockingSizes to make room
for the possibility of using lookup tables +701ca5c12587b # Polish lookup tables generation +7013589a9c115 # actual_panel_rows computation should always be resilient
to parameters not consistent with the known L1 cache size, see comment +70102babb9c0f # Provide a empirical lookup table for blocking sizes measured on a Nexus 5.
Only for float, only for Android on ARM 32bit for now. +7088481dc21ea # Bug 986: add support for coefficient-based
product with 0 depth. +709d7f51feb07 # Bug 992: don't select a 3p GEMM path with non-SIMD scalar types. +759f9303cc7c5 # 3.3-alpha1 +765aba1eda71e # help clang inlining +770fe630c9873 # Improve numerical accuracy in LLT and triangular solve
by using true scalar divisions (instead of x * (1/y)) +#8741d23430628 # Improved the matrix multiplication blocking in the case
where mr is not a power of 2 (e.g on Haswell CPUs) +878f629fe95c8 # Made the index type a template parameter to evaluateProductBlockingSizes.
Use numext::mini and numext::maxi instead of
std::min/std::max to compute blocking sizes. +8975d51a7f12c # Don't optimize the processing of the last rows of
a matrix matrix product in cases that violate
the assumptions made by the optimized code path. +8986136f4fdd4 # Remove the rotating kernel. +898e68e165a23 # Bug 256: enable vectorization with unaligned loads/stores. +91466e99ab6a1 # Relax mixing-type constraints for binary coeff-wise operators +91776236cdea4 # merge +917101ea26f5e # Include the cost of stores in unrolling +921672076db5d # Fix perf regression introduced in changeset e56aabf205 +9210fa9e4a15c # Fix perf regression in dgemm introduced by changeset 5d51a7f12c +936f6b3cf8de9 # 3.3-beta2 +944504a4404f1 # Optimize expression matching 'd?=a-b*c' as 'd?=a; d?=b*c;' +95877e27fbeee # 3.3-rc1 +959779774f98c # Bug 1311: fix alignment logic in some cases
of (scalar*small).lazyProduct(small) +9729f9d8d2f62 # Disabled part of the matrix matrix peeling code
that's incompatible with 512 bit registers +979eeac81b8c0 # 3.3.0 +989c927af60ed # Fix a performance regression in (mat*mat)*vec
for which mat*mat was evaluated multiple times. +994fe696022ec # Operators += and -= do not resize! +99466f65ccc36 # Ease compiler generating clean and efficient code in mat*vec +9946a5fe86098 # Complete rewrite of column-major-matrix * vector product
to deliver higher performance of modern CPU. +99591003f3b86 # Improve performance of row-major-dense-matrix * vector products
for recent CPUs. +997eb621413c1 # Revert vec/y to vec*(1/y) in row-major TRSM +10444bbc320468 # Bug 1435: fix aliasing issue in exressions like: A = C - B*A; +1073624df50945 # Adds missing EIGEN_STRONG_INLINE to support MSVC
properly inlining small vector calculations +1094d428a199ab # Bug 1562: optimize evaluation of small products
of the form s*A*B by rewriting them as: s*(A.lazyProduct(B))
to save a costly temporary.
Measured speedup from 2x to 5x. +1096de9e31a06d # Introduce the macro ei_declare_local_nested_eval to
help allocating on the stack local temporaries via alloca,
and let outer-products makes a good use of it. +11087b91c11207 # Bug 1578: Improve prefetching in matrix multiplication on MIPS. +1153aa110e681b # PR 526: Speed up multiplication of small, dynamically sized matrices +11544ad359237a # Vectorize row-by-row gebp loop iterations on 16 packets as well +1157a476054879 # Bug 1624: improve matrix-matrix product on ARM 64, 20% speedup +1160a4159dba08 # do not read buffers out of bounds +1163c53eececb0 # Implement AVX512 vectorization of std::complex +11644e7746fe22 # Bug 1636: fix gemm performance issue with gcc>=6 and no FMA +1164956678a4ef # Bug 1515: disable gebp's 3pX4 micro kernel
for MSVC<=19.14 because of register spilling. +1165426bce7529 # fix EIGEN_GEBP_2PX4_SPILLING_WORKAROUND
for non vectorized type, and non x86/64 target +11660d90637838 # enable spilling workaround on architectures with SSE/AVX +1166f159cf3d75 # Artificially increase l1-blocking size for AVX512.
+10% speedup with current kernels. +11686dd93f7e3b # Make code compile again for older compilers. +1175dbfcceabf5 # Bug: 1633: refactor gebp kernel and optimize for neon +117670e133333d # Bug 1661: fix regression in GEBP and AVX512 +11760f028f61cb # GEBP: cleanup logic to choose between
a 4 packets of 1 packet (=e118ce86fd+fix) +1180de77bf5d6c # gebp: Add new ½ and ¼ packet rows per (peeling) round on the lhs diff --git a/thirdparty/eigen/bench/perf_monitoring/gemm.cpp b/thirdparty/eigen/bench/perf_monitoring/gemm.cpp new file mode 100644 index 00000000..804139db --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemm.cpp @@ -0,0 +1,12 @@ +#include "gemm_common.h" + +EIGEN_DONT_INLINE +void gemm(const Mat &A, const Mat &B, Mat &C) +{ + C.noalias() += A * B; +} + +int main(int argc, char **argv) +{ + return main_gemm(argc, argv, gemm); +} diff --git a/thirdparty/eigen/bench/perf_monitoring/gemm_common.h b/thirdparty/eigen/bench/perf_monitoring/gemm_common.h new file mode 100644 index 00000000..30dbc0df --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemm_common.h @@ -0,0 +1,67 @@ +#include +#include +#include +#include +#include "eigen_src/Eigen/Core" +#include "../BenchTimer.h" +using namespace Eigen; + +#ifndef SCALAR +#error SCALAR must be defined +#endif + +typedef SCALAR Scalar; + +typedef Matrix Mat; + +template +EIGEN_DONT_INLINE +double bench(long m, long n, long k, const Func& f) +{ + Mat A(m,k); + Mat B(k,n); + Mat C(m,n); + A.setRandom(); + B.setRandom(); + C.setZero(); + + BenchTimer t; + + double up = 1e8*4/sizeof(Scalar); + double tm0 = 4, tm1 = 10; + if(NumTraits::IsComplex) + { + up /= 4; + tm0 = 2; + tm1 = 4; + } + + double flops = 2. * m * n * k; + long rep = std::max(1., std::min(100., up/flops) ); + long tries = std::max(tm0, std::min(tm1, up/flops) ); + + BENCH(t, tries, rep, f(A,B,C)); + + return 1e-9 * rep * flops / t.best(); +} + +template +int main_gemm(int argc, char **argv, const Func& f) +{ + std::vector results; + + std::string filename = std::string("gemm_settings.txt"); + if(argc>1) + filename = std::string(argv[1]); + std::ifstream settings(filename); + long m, n, k; + while(settings >> m >> n >> k) + { + //std::cerr << " Testing " << m << " " << n << " " << k << std::endl; + results.push_back( bench(m, n, k, f) ); + } + + std::cout << RowVectorXd::Map(results.data(), results.size()); + + return 0; +} diff --git a/thirdparty/eigen/bench/perf_monitoring/gemm_settings.txt b/thirdparty/eigen/bench/perf_monitoring/gemm_settings.txt new file mode 100644 index 00000000..5c43e1c7 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemm_settings.txt @@ -0,0 +1,15 @@ +8 8 8 +9 9 9 +24 24 24 +239 239 239 +240 240 240 +2400 24 24 +24 2400 24 +24 24 2400 +24 2400 2400 +2400 24 2400 +2400 2400 24 +2400 2400 64 +4800 23 160 +23 4800 160 +2400 2400 2400 diff --git a/thirdparty/eigen/bench/perf_monitoring/gemm_square_settings.txt b/thirdparty/eigen/bench/perf_monitoring/gemm_square_settings.txt new file mode 100644 index 00000000..98474d17 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemm_square_settings.txt @@ -0,0 +1,11 @@ +8 8 8 +9 9 9 +12 12 12 +15 15 15 +16 16 16 +24 24 24 +102 102 102 +239 239 239 +240 240 240 +2400 2400 2400 +2463 2463 2463 diff --git a/thirdparty/eigen/bench/perf_monitoring/gemv.cpp b/thirdparty/eigen/bench/perf_monitoring/gemv.cpp new file mode 100644 index 00000000..82e5ab96 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemv.cpp @@ -0,0 +1,12 @@ +#include "gemv_common.h" + +EIGEN_DONT_INLINE +void gemv(const Mat &A, const Vec &B, Vec &C) +{ + C.noalias() += A * B; +} + +int main(int argc, char **argv) +{ + return main_gemv(argc, argv, gemv); +} diff --git a/thirdparty/eigen/bench/perf_monitoring/gemv_common.h b/thirdparty/eigen/bench/perf_monitoring/gemv_common.h new file mode 100644 index 00000000..cc325772 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemv_common.h @@ -0,0 +1,69 @@ +#include +#include +#include +#include +#include +#include "eigen_src/Eigen/Core" +#include "../BenchTimer.h" +using namespace Eigen; + +#ifndef SCALAR +#error SCALAR must be defined +#endif + +typedef SCALAR Scalar; + +typedef Matrix Mat; +typedef Matrix Vec; + +template +EIGEN_DONT_INLINE +double bench(long m, long n, Func &f) +{ + Mat A(m,n); + Vec B(n); + Vec C(m); + A.setRandom(); + B.setRandom(); + C.setRandom(); + + BenchTimer t; + + double up = 1e8/sizeof(Scalar); + double tm0 = 4, tm1 = 10; + if(NumTraits::IsComplex) + { + up /= 4; + tm0 = 2; + tm1 = 4; + } + + double flops = 2. * m * n; + long rep = std::max(1., std::min(100., up/flops) ); + long tries = std::max(tm0, std::min(tm1, up/flops) ); + + BENCH(t, tries, rep, f(A,B,C)); + + return 1e-9 * rep * flops / t.best(); +} + +template +int main_gemv(int argc, char **argv, Func& f) +{ + std::vector results; + + std::string filename = std::string("gemv_settings.txt"); + if(argc>1) + filename = std::string(argv[1]); + std::ifstream settings(filename); + long m, n; + while(settings >> m >> n) + { + //std::cerr << " Testing " << m << " " << n << std::endl; + results.push_back( bench(m, n, f) ); + } + + std::cout << RowVectorXd::Map(results.data(), results.size()); + + return 0; +} diff --git a/thirdparty/eigen/bench/perf_monitoring/gemv_settings.txt b/thirdparty/eigen/bench/perf_monitoring/gemv_settings.txt new file mode 100644 index 00000000..21a5ee05 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemv_settings.txt @@ -0,0 +1,11 @@ +8 8 +9 9 +24 24 +239 239 +240 240 +2400 24 +24 2400 +24 240 +2400 2400 +4800 23 +23 4800 diff --git a/thirdparty/eigen/bench/perf_monitoring/gemv_square_settings.txt b/thirdparty/eigen/bench/perf_monitoring/gemv_square_settings.txt new file mode 100644 index 00000000..5165759f --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemv_square_settings.txt @@ -0,0 +1,13 @@ +8 8 +9 9 +12 12 +15 15 +16 16 +24 24 +53 53 +74 74 +102 102 +239 239 +240 240 +2400 2400 +2463 2463 diff --git a/thirdparty/eigen/bench/perf_monitoring/gemvt.cpp b/thirdparty/eigen/bench/perf_monitoring/gemvt.cpp new file mode 100644 index 00000000..fe945767 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/gemvt.cpp @@ -0,0 +1,12 @@ +#include "gemv_common.h" + +EIGEN_DONT_INLINE +void gemv(const Mat &A, Vec &B, const Vec &C) +{ + B.noalias() += A.transpose() * C; +} + +int main(int argc, char **argv) +{ + return main_gemv(argc, argv, gemv); +} diff --git a/thirdparty/eigen/bench/perf_monitoring/lazy_gemm.cpp b/thirdparty/eigen/bench/perf_monitoring/lazy_gemm.cpp new file mode 100644 index 00000000..77330604 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/lazy_gemm.cpp @@ -0,0 +1,101 @@ +#include +#include +#include +#include +#include "../../BenchTimer.h" +using namespace Eigen; + +#ifndef SCALAR +#error SCALAR must be defined +#endif + +typedef SCALAR Scalar; + +template +EIGEN_DONT_INLINE +void lazy_gemm(const MatA &A, const MatB &B, MatC &C) +{ +// escape((void*)A.data()); +// escape((void*)B.data()); + C.noalias() += A.lazyProduct(B); +// escape((void*)C.data()); +} + +template +EIGEN_DONT_INLINE +double bench() +{ + typedef Matrix MatA; + typedef Matrix MatB; + typedef Matrix MatC; + + MatA A(m,k); + MatB B(k,n); + MatC C(m,n); + A.setRandom(); + B.setRandom(); + C.setZero(); + + BenchTimer t; + + double up = 1e7*4/sizeof(Scalar); + double tm0 = 10, tm1 = 20; + + double flops = 2. * m * n * k; + long rep = std::max(10., std::min(10000., up/flops) ); + long tries = std::max(tm0, std::min(tm1, up/flops) ); + + BENCH(t, tries, rep, lazy_gemm(A,B,C)); + + return 1e-9 * rep * flops / t.best(); +} + +template +double bench_t(int t) +{ + if(t) + return bench(); + else + return bench(); +} + +EIGEN_DONT_INLINE +double bench_mnk(int m, int n, int k, int t) +{ + int id = m*10000 + n*100 + k; + switch(id) { + case 10101 : return bench_t< 1, 1, 1>(t); break; + case 20202 : return bench_t< 2, 2, 2>(t); break; + case 30303 : return bench_t< 3, 3, 3>(t); break; + case 40404 : return bench_t< 4, 4, 4>(t); break; + case 50505 : return bench_t< 5, 5, 5>(t); break; + case 60606 : return bench_t< 6, 6, 6>(t); break; + case 70707 : return bench_t< 7, 7, 7>(t); break; + case 80808 : return bench_t< 8, 8, 8>(t); break; + case 90909 : return bench_t< 9, 9, 9>(t); break; + case 101010 : return bench_t<10,10,10>(t); break; + case 111111 : return bench_t<11,11,11>(t); break; + case 121212 : return bench_t<12,12,12>(t); break; + } + return 0; +} + +int main(int argc, char **argv) +{ + std::vector results; + + std::string filename = std::string("lazy_gemm_settings.txt"); + if(argc>1) + filename = std::string(argv[1]); + std::ifstream settings(filename); + long m, n, k, t; + while(settings >> m >> n >> k >> t) + { + //std::cerr << " Testing " << m << " " << n << " " << k << std::endl; + results.push_back( bench_mnk(m, n, k, t) ); + } + + std::cout << RowVectorXd::Map(results.data(), results.size()); + + return 0; +} diff --git a/thirdparty/eigen/bench/perf_monitoring/lazy_gemm_settings.txt b/thirdparty/eigen/bench/perf_monitoring/lazy_gemm_settings.txt new file mode 100644 index 00000000..407d5d4f --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/lazy_gemm_settings.txt @@ -0,0 +1,15 @@ +1 1 1 0 +2 2 2 0 +3 3 3 0 +4 4 4 0 +4 4 4 1 +5 5 5 0 +6 6 6 0 +7 7 7 0 +7 7 7 1 +8 8 8 0 +9 9 9 0 +10 10 10 0 +11 11 11 0 +12 12 12 0 +12 12 12 1 diff --git a/thirdparty/eigen/bench/perf_monitoring/llt.cpp b/thirdparty/eigen/bench/perf_monitoring/llt.cpp new file mode 100644 index 00000000..d55b7d80 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/llt.cpp @@ -0,0 +1,15 @@ +#include "gemm_common.h" +#include + +EIGEN_DONT_INLINE +void llt(const Mat &A, const Mat &B, Mat &C) +{ + C = A; + C.diagonal().array() += 1000; + Eigen::internal::llt_inplace::blocked(C); +} + +int main(int argc, char **argv) +{ + return main_gemm(argc, argv, llt); +} diff --git a/thirdparty/eigen/bench/perf_monitoring/make_plot.sh b/thirdparty/eigen/bench/perf_monitoring/make_plot.sh new file mode 100755 index 00000000..65aaf66f --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/make_plot.sh @@ -0,0 +1,112 @@ +#!/bin/bash + +# base name of the bench +# it reads $1.out +# and generates $1.pdf +WHAT=$1 +bench=$2 +settings_file=$3 + +header="rev " +while read line +do + if [ ! -z '$line' ]; then + header="$header \"$line\"" + fi +done < $settings_file + +echo $header > $WHAT.out.header +cat $WHAT.out >> $WHAT.out.header + + +echo "set title '$WHAT'" > $WHAT.gnuplot +echo "set key autotitle columnhead outside " >> $WHAT.gnuplot +echo "set xtics rotate 1" >> $WHAT.gnuplot + +echo "set term pdf color rounded enhanced fontscale 0.35 size 7in,5in" >> $WHAT.gnuplot +echo set output "'"$WHAT.pdf"'" >> $WHAT.gnuplot + +col=`cat $settings_file | wc -l` +echo "plot for [col=2:$col+1] '$WHAT.out.header' using 0:col:xticlabels(1) with lines" >> $WHAT.gnuplot +echo " " >> $WHAT.gnuplot + +gnuplot -persist < $WHAT.gnuplot + +# generate a png file (thumbnail) +convert -colors 256 -background white -density 300 -resize 300 -quality 0 $WHAT.pdf -background white -flatten $WHAT.png + +# clean +rm $WHAT.out.header $WHAT.gnuplot + + +# generate html/svg graph + +echo " " > $WHAT.html +cat resources/chart_header.html > $WHAT.html +echo 'var customSettings = {"TITLE":"","SUBTITLE":"","XLABEL":"","YLABEL":""};' >> $WHAT.html +# 'data' is an array of datasets (i.e. curves), each of which is an object of the form +# { +# key: , +# color: , +# values: [{ +# r: , +# v: +# }] +# } +echo 'var data = [' >> $WHAT.html + +col=2 +while read line +do + if [ ! -z '$line' ]; then + header="$header \"$line\"" + echo '{"key":"'$line'","values":[' >> $WHAT.html + i=0 + while read line2 + do + if [ ! -z "$line2" ]; then + val=`echo $line2 | cut -s -f $col -d ' '` + if [ -n "$val" ]; then # skip build failures + echo '{"r":'$i',"v":'$val'},' >> $WHAT.html + fi + fi + ((i++)) + done < $WHAT.out + echo ']},' >> $WHAT.html + fi + ((col++)) +done < $settings_file +echo '];' >> $WHAT.html + +echo 'var changesets = [' >> $WHAT.html +while read line2 +do + if [ ! -z '$line2' ]; then + echo '"'`echo $line2 | cut -f 1 -d ' '`'",' >> $WHAT.html + fi +done < $WHAT.out +echo '];' >> $WHAT.html + +echo 'var changesets_details = [' >> $WHAT.html +while read line2 +do + if [ ! -z '$line2' ]; then + num=`echo "$line2" | cut -f 1 -d ' '` + comment=`grep ":$num" changesets.txt | cut -f 2 -d '#'` + echo '"'"$comment"'",' >> $WHAT.html + fi +done < $WHAT.out +echo '];' >> $WHAT.html + +echo 'var changesets_count = [' >> $WHAT.html +i=0 +while read line2 +do + if [ ! -z '$line2' ]; then + echo $i ',' >> $WHAT.html + fi + ((i++)) +done < $WHAT.out +echo '];' >> $WHAT.html + +cat resources/chart_footer.html >> $WHAT.html diff --git a/thirdparty/eigen/bench/perf_monitoring/resources/chart_footer.html b/thirdparty/eigen/bench/perf_monitoring/resources/chart_footer.html new file mode 100644 index 00000000..a96cdb89 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/resources/chart_footer.html @@ -0,0 +1,41 @@ + /* setup the chart and its options */ + var chart = nv.models.lineChart() + .color(d3.scale.category10().range()) + .margin({left: 75, bottom: 100}) + .forceX([0]).forceY([0]); + + chart.x(function(datum){ return datum.r; }) + .xAxis.options({ + axisLabel: customSettings.XLABEL || 'Changeset', + tickFormat: d3.format('.0f') + }); + chart.xAxis + .tickValues(changesets_count) + .tickFormat(function(d){return changesets[d]}) + .rotateLabels(-90); + + chart.y(function(datum){ return datum.v; }) + .yAxis.options({ + axisLabel: customSettings.YLABEL || 'GFlops'/*, + tickFormat: function(val){ return d3.format('.0f')(val) + ' GFlops'; }*/ + }); + + chart.tooltip.headerFormatter(function(d) { return changesets[d] + + '

' + + changesets_details[d] + "

"; }); + + //chart.useInteractiveGuideline(true); + d3.select('#chart').datum(data).call(chart); + var plot = d3.select('#chart > g'); + + /* setup the title */ + plot.append('text') + .style('font-size', '24px') + .attr('text-anchor', 'middle').attr('x', '50%').attr('y', '20px') + .text(customSettings.TITLE || ''); + + /* ensure the chart is responsive */ + nv.utils.windowResize(chart.update); + + + diff --git a/thirdparty/eigen/bench/perf_monitoring/resources/chart_header.html b/thirdparty/eigen/bench/perf_monitoring/resources/chart_header.html new file mode 100644 index 00000000..27eb02e5 --- /dev/null +++ b/thirdparty/eigen/bench/perf_monitoring/resources/chart_header.html @@ -0,0 +1,45 @@ + + + + + + + + + + + + + - - - - diff --git a/thirdparty/eigen/doc/eigendoxy_header.html.in b/thirdparty/eigen/doc/eigendoxy_header.html.in index 0f3859f4..e377b26f 100644 --- a/thirdparty/eigen/doc/eigendoxy_header.html.in +++ b/thirdparty/eigen/doc/eigendoxy_header.html.in @@ -1,39 +1,54 @@ - - + + + - + + $projectname: $title $title - - - + + + + + + + + + + + + + + $treeview $search $mathjax - +$darkmode + - - - - - +$extrastylesheet + + +
+ + +
-
- + - + - - + + + + + + + +
+
$projectname  $projectnumber
@@ -42,20 +57,26 @@ $mathjax - 		+
$projectbrief
 $searchbox$searchbox
$searchbox
- diff --git a/thirdparty/eigen/doc/eigendoxy_layout.xml.in b/thirdparty/eigen/doc/eigendoxy_layout.xml.in index c14b621e..9e3021af 100644 --- a/thirdparty/eigen/doc/eigendoxy_layout.xml.in +++ b/thirdparty/eigen/doc/eigendoxy_layout.xml.in @@ -1,67 +1,103 @@ - + + - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + - - + + - - - - - - - - - - - - - - - - - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + - - - - - - - - - - + + + + + + + + + + + + @@ -71,24 +107,44 @@ + + - - - - - + + + + + + + + + + + - + - - - - - + + + + + + + + + + + + + + + + + @@ -97,23 +153,34 @@ + + + - - - - - - + + + + + + + + + + + - + - - - - - - + + + + + + + + + @@ -121,50 +188,74 @@ - + + + - - - - - - - - - - - - - + + + + + + + + + + + + + + + - - - - - - - - - - - - - - - + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + + @@ -173,6 +264,6 @@ - + diff --git a/thirdparty/eigen/doc/examples/CMakeLists.txt b/thirdparty/eigen/doc/examples/CMakeLists.txt index 08cf8efd..9a1949e2 100644 --- a/thirdparty/eigen/doc/examples/CMakeLists.txt +++ b/thirdparty/eigen/doc/examples/CMakeLists.txt @@ -6,6 +6,7 @@ foreach(example_src ${examples_SRCS}) if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) target_link_libraries(${example} ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}) endif() + target_link_libraries(${example} Eigen3::Eigen) add_custom_command( TARGET ${example} POST_BUILD @@ -13,4 +14,4 @@ foreach(example_src ${examples_SRCS}) ARGS >${CMAKE_CURRENT_BINARY_DIR}/${example}.out ) add_dependencies(all_examples ${example}) -endforeach(example_src) +endforeach() \ No newline at end of file diff --git a/thirdparty/eigen/doc/examples/CustomizingEigen_Inheritance.cpp b/thirdparty/eigen/doc/examples/CustomizingEigen_Inheritance.cpp new file mode 100644 index 00000000..48df64ee --- /dev/null +++ b/thirdparty/eigen/doc/examples/CustomizingEigen_Inheritance.cpp @@ -0,0 +1,30 @@ +#include +#include + +class MyVectorType : public Eigen::VectorXd +{ +public: + MyVectorType(void):Eigen::VectorXd() {} + + // This constructor allows you to construct MyVectorType from Eigen expressions + template + MyVectorType(const Eigen::MatrixBase& other) + : Eigen::VectorXd(other) + { } + + // This method allows you to assign Eigen expressions to MyVectorType + template + MyVectorType& operator=(const Eigen::MatrixBase & other) + { + this->Eigen::VectorXd::operator=(other); + return *this; + } +}; + +int main() +{ + MyVectorType v = MyVectorType::Ones(4); + v(2) += 10; + v = 2 * v; + std::cout << v.transpose() << std::endl; +} diff --git a/thirdparty/eigen/doc/examples/Cwise_erf.cpp b/thirdparty/eigen/doc/examples/Cwise_erf.cpp new file mode 100644 index 00000000..e7cd2c1c --- /dev/null +++ b/thirdparty/eigen/doc/examples/Cwise_erf.cpp @@ -0,0 +1,9 @@ +#include +#include +#include +using namespace Eigen; +int main() +{ + Array4d v(-0.5,2,0,-7); + std::cout << v.erf() << std::endl; +} diff --git a/thirdparty/eigen/doc/examples/Cwise_erfc.cpp b/thirdparty/eigen/doc/examples/Cwise_erfc.cpp new file mode 100644 index 00000000..d8bb04c3 --- /dev/null +++ b/thirdparty/eigen/doc/examples/Cwise_erfc.cpp @@ -0,0 +1,9 @@ +#include +#include +#include +using namespace Eigen; +int main() +{ + Array4d v(-0.5,2,0,-7); + std::cout << v.erfc() << std::endl; +} diff --git a/thirdparty/eigen/doc/examples/Cwise_lgamma.cpp b/thirdparty/eigen/doc/examples/Cwise_lgamma.cpp new file mode 100644 index 00000000..6bfaccbc --- /dev/null +++ b/thirdparty/eigen/doc/examples/Cwise_lgamma.cpp @@ -0,0 +1,9 @@ +#include +#include +#include +using namespace Eigen; +int main() +{ + Array4d v(0.5,10,0,-1); + std::cout << v.lgamma() << std::endl; +} diff --git a/thirdparty/eigen/doc/examples/MatrixBase_cwise_const.cpp b/thirdparty/eigen/doc/examples/MatrixBase_cwise_const.cpp deleted file mode 100644 index 23700e0b..00000000 --- a/thirdparty/eigen/doc/examples/MatrixBase_cwise_const.cpp +++ /dev/null @@ -1,18 +0,0 @@ -#define EIGEN2_SUPPORT -#include -#include - -using namespace Eigen; -using namespace std; - -int main() -{ - Matrix3i m = Matrix3i::Random(); - cout << "Here is the matrix m:" << endl << m << endl; - Matrix3i n = Matrix3i::Random(); - cout << "And here is the matrix n:" << endl << n << endl; - cout << "The coefficient-wise product of m and n is:" << endl; - cout << m.cwise() * n << endl; - cout << "Taking the cube of the coefficients of m yields:" << endl; - cout << m.cwise().pow(3) << endl; -} diff --git a/thirdparty/eigen/doc/examples/TutorialInplaceLU.cpp b/thirdparty/eigen/doc/examples/TutorialInplaceLU.cpp new file mode 100644 index 00000000..cb9c59b6 --- /dev/null +++ b/thirdparty/eigen/doc/examples/TutorialInplaceLU.cpp @@ -0,0 +1,61 @@ +#include +struct init { + init() { std::cout << "[" << "init" << "]" << std::endl; } +}; +init init_obj; +// [init] +#include +#include + +using namespace std; +using namespace Eigen; + +int main() +{ + MatrixXd A(2,2); + A << 2, -1, 1, 3; + cout << "Here is the input matrix A before decomposition:\n" << A << endl; +cout << "[init]" << endl; + +cout << "[declaration]" << endl; + PartialPivLU > lu(A); + cout << "Here is the input matrix A after decomposition:\n" << A << endl; +cout << "[declaration]" << endl; + +cout << "[matrixLU]" << endl; + cout << "Here is the matrix storing the L and U factors:\n" << lu.matrixLU() << endl; +cout << "[matrixLU]" << endl; + +cout << "[solve]" << endl; + MatrixXd A0(2,2); A0 << 2, -1, 1, 3; + VectorXd b(2); b << 1, 2; + VectorXd x = lu.solve(b); + cout << "Residual: " << (A0 * x - b).norm() << endl; +cout << "[solve]" << endl; + +cout << "[modifyA]" << endl; + A << 3, 4, -2, 1; + x = lu.solve(b); + cout << "Residual: " << (A0 * x - b).norm() << endl; +cout << "[modifyA]" << endl; + +cout << "[recompute]" << endl; + A0 = A; // save A + lu.compute(A); + x = lu.solve(b); + cout << "Residual: " << (A0 * x - b).norm() << endl; +cout << "[recompute]" << endl; + +cout << "[recompute_bis0]" << endl; + MatrixXd A1(2,2); + A1 << 5,-2,3,4; + lu.compute(A1); + cout << "Here is the input matrix A1 after decomposition:\n" << A1 << endl; +cout << "[recompute_bis0]" << endl; + +cout << "[recompute_bis1]" << endl; + x = lu.solve(b); + cout << "Residual: " << (A1 * x - b).norm() << endl; +cout << "[recompute_bis1]" << endl; + +} diff --git a/thirdparty/eigen/doc/examples/TutorialLinAlgInverseDeterminant.cpp b/thirdparty/eigen/doc/examples/TutorialLinAlgInverseDeterminant.cpp index 43970ff0..14dde5b3 100644 --- a/thirdparty/eigen/doc/examples/TutorialLinAlgInverseDeterminant.cpp +++ b/thirdparty/eigen/doc/examples/TutorialLinAlgInverseDeterminant.cpp @@ -13,4 +13,4 @@ int main() cout << "Here is the matrix A:\n" << A << endl; cout << "The determinant of A is " << A.determinant() << endl; cout << "The inverse of A is:\n" << A.inverse() << endl; -} \ No newline at end of file +} diff --git a/thirdparty/eigen/doc/examples/TutorialLinAlgSVDSolve.cpp b/thirdparty/eigen/doc/examples/TutorialLinAlgSVDSolve.cpp index 9fbc031d..f109f04e 100644 --- a/thirdparty/eigen/doc/examples/TutorialLinAlgSVDSolve.cpp +++ b/thirdparty/eigen/doc/examples/TutorialLinAlgSVDSolve.cpp @@ -11,5 +11,5 @@ int main() VectorXf b = VectorXf::Random(3); cout << "Here is the right hand side b:\n" << b << endl; cout << "The least-squares solution is:\n" - << A.jacobiSvd(ComputeThinU | ComputeThinV).solve(b) << endl; + << A.bdcSvd(ComputeThinU | ComputeThinV).solve(b) << endl; } diff --git a/thirdparty/eigen/doc/examples/Tutorial_BlockOperations_block_assignment.cpp b/thirdparty/eigen/doc/examples/Tutorial_BlockOperations_block_assignment.cpp index 76f49f2f..0b87313a 100644 --- a/thirdparty/eigen/doc/examples/Tutorial_BlockOperations_block_assignment.cpp +++ b/thirdparty/eigen/doc/examples/Tutorial_BlockOperations_block_assignment.cpp @@ -14,5 +14,5 @@ int main() a.block<2,2>(1,1) = m; cout << "Here is now a with m copied into its central 2x2 block:" << endl << a << endl << endl; a.block(0,0,2,3) = a.block(2,1,2,3); - cout << "Here is now a with bottom-right 2x3 block copied into top-left 2x2 block:" << endl << a << endl << endl; + cout << "Here is now a with bottom-right 2x3 block copied into top-left 2x3 block:" << endl << a << endl << endl; } diff --git a/thirdparty/eigen/doc/examples/Tutorial_ReductionsVisitorsBroadcasting_reductions_operatornorm.cpp b/thirdparty/eigen/doc/examples/Tutorial_ReductionsVisitorsBroadcasting_reductions_operatornorm.cpp new file mode 100644 index 00000000..62e28fc3 --- /dev/null +++ b/thirdparty/eigen/doc/examples/Tutorial_ReductionsVisitorsBroadcasting_reductions_operatornorm.cpp @@ -0,0 +1,18 @@ +#include +#include + +using namespace Eigen; +using namespace std; + +int main() +{ + MatrixXf m(2,2); + m << 1,-2, + -3,4; + + cout << "1-norm(m) = " << m.cwiseAbs().colwise().sum().maxCoeff() + << " == " << m.colwise().lpNorm<1>().maxCoeff() << endl; + + cout << "infty-norm(m) = " << m.cwiseAbs().rowwise().sum().maxCoeff() + << " == " << m.rowwise().lpNorm<1>().maxCoeff() << endl; +} diff --git a/thirdparty/eigen/doc/examples/Tutorial_simple_example_dynamic_size.cpp b/thirdparty/eigen/doc/examples/Tutorial_simple_example_dynamic_size.cpp index 0f0280e0..defcb1ee 100644 --- a/thirdparty/eigen/doc/examples/Tutorial_simple_example_dynamic_size.cpp +++ b/thirdparty/eigen/doc/examples/Tutorial_simple_example_dynamic_size.cpp @@ -10,7 +10,7 @@ int main() MatrixXi m(size,size+1); // a (size)x(size+1)-matrix of int's for (int j=0; j +#include +using namespace Eigen; +using namespace std; + +template +Eigen::Reshaped +reshape_helper(MatrixBase& m) +{ + return Eigen::Reshaped(m.derived()); +} + +int main(int, char**) +{ + MatrixXd m(2, 4); + m << 1, 2, 3, 4, + 5, 6, 7, 8; + MatrixXd n = reshape_helper(m); + cout << "matrix m is:" << endl << m << endl; + cout << "matrix n is:" << endl << n << endl; + return 0; +} diff --git a/thirdparty/eigen/doc/examples/class_Reshaped.cpp b/thirdparty/eigen/doc/examples/class_Reshaped.cpp new file mode 100644 index 00000000..18fb4545 --- /dev/null +++ b/thirdparty/eigen/doc/examples/class_Reshaped.cpp @@ -0,0 +1,23 @@ +#include +#include +using namespace std; +using namespace Eigen; + +template +const Reshaped +reshape_helper(const MatrixBase& m, int rows, int cols) +{ + return Reshaped(m.derived(), rows, cols); +} + +int main(int, char**) +{ + MatrixXd m(3, 4); + m << 1, 4, 7, 10, + 2, 5, 8, 11, + 3, 6, 9, 12; + cout << m << endl; + Ref n = reshape_helper(m, 2, 6); + cout << "Matrix m is:" << endl << m << endl; + cout << "Matrix n is:" << endl << n << endl; +} diff --git a/thirdparty/eigen/doc/examples/make_circulant.cpp b/thirdparty/eigen/doc/examples/make_circulant.cpp new file mode 100644 index 00000000..92e6aaa2 --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant.cpp @@ -0,0 +1,11 @@ +/* +This program is presented in several fragments in the doc page. +Every fragment is in its own file; this file simply combines them. +*/ + +#include "make_circulant.cpp.preamble" +#include "make_circulant.cpp.traits" +#include "make_circulant.cpp.expression" +#include "make_circulant.cpp.evaluator" +#include "make_circulant.cpp.entry" +#include "make_circulant.cpp.main" diff --git a/thirdparty/eigen/doc/examples/make_circulant.cpp.entry b/thirdparty/eigen/doc/examples/make_circulant.cpp.entry new file mode 100644 index 00000000..f9d2eb8a --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant.cpp.entry @@ -0,0 +1,5 @@ +template +Circulant makeCirculant(const Eigen::MatrixBase& arg) +{ + return Circulant(arg.derived()); +} diff --git a/thirdparty/eigen/doc/examples/make_circulant.cpp.evaluator b/thirdparty/eigen/doc/examples/make_circulant.cpp.evaluator new file mode 100644 index 00000000..2ba79e78 --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant.cpp.evaluator @@ -0,0 +1,32 @@ +namespace Eigen { + namespace internal { + template + struct evaluator > + : evaluator_base > + { + typedef Circulant XprType; + typedef typename nested_eval::type ArgTypeNested; + typedef typename remove_all::type ArgTypeNestedCleaned; + typedef typename XprType::CoeffReturnType CoeffReturnType; + + enum { + CoeffReadCost = evaluator::CoeffReadCost, + Flags = Eigen::ColMajor + }; + + evaluator(const XprType& xpr) + : m_argImpl(xpr.m_arg), m_rows(xpr.rows()) + { } + + CoeffReturnType coeff(Index row, Index col) const + { + Index index = row - col; + if (index < 0) index += m_rows; + return m_argImpl.coeff(index); + } + + evaluator m_argImpl; + const Index m_rows; + }; + } +} diff --git a/thirdparty/eigen/doc/examples/make_circulant.cpp.expression b/thirdparty/eigen/doc/examples/make_circulant.cpp.expression new file mode 100644 index 00000000..380cd445 --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant.cpp.expression @@ -0,0 +1,20 @@ +template +class Circulant : public Eigen::MatrixBase > +{ +public: + Circulant(const ArgType& arg) + : m_arg(arg) + { + EIGEN_STATIC_ASSERT(ArgType::ColsAtCompileTime == 1, + YOU_TRIED_CALLING_A_VECTOR_METHOD_ON_A_MATRIX); + } + + typedef typename Eigen::internal::ref_selector::type Nested; + + typedef Eigen::Index Index; + Index rows() const { return m_arg.rows(); } + Index cols() const { return m_arg.rows(); } + + typedef typename Eigen::internal::ref_selector::type ArgTypeNested; + ArgTypeNested m_arg; +}; diff --git a/thirdparty/eigen/doc/examples/make_circulant.cpp.main b/thirdparty/eigen/doc/examples/make_circulant.cpp.main new file mode 100644 index 00000000..877f97f6 --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant.cpp.main @@ -0,0 +1,8 @@ +int main() +{ + Eigen::VectorXd vec(4); + vec << 1, 2, 4, 8; + Eigen::MatrixXd mat; + mat = makeCirculant(vec); + std::cout << mat << std::endl; +} diff --git a/thirdparty/eigen/doc/examples/make_circulant.cpp.preamble b/thirdparty/eigen/doc/examples/make_circulant.cpp.preamble new file mode 100644 index 00000000..e575cce1 --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant.cpp.preamble @@ -0,0 +1,4 @@ +#include +#include + +template class Circulant; diff --git a/thirdparty/eigen/doc/examples/make_circulant.cpp.traits b/thirdparty/eigen/doc/examples/make_circulant.cpp.traits new file mode 100644 index 00000000..4e04535d --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant.cpp.traits @@ -0,0 +1,19 @@ +namespace Eigen { + namespace internal { + template + struct traits > + { + typedef Eigen::Dense StorageKind; + typedef Eigen::MatrixXpr XprKind; + typedef typename ArgType::StorageIndex StorageIndex; + typedef typename ArgType::Scalar Scalar; + enum { + Flags = Eigen::ColMajor, + RowsAtCompileTime = ArgType::RowsAtCompileTime, + ColsAtCompileTime = ArgType::RowsAtCompileTime, + MaxRowsAtCompileTime = ArgType::MaxRowsAtCompileTime, + MaxColsAtCompileTime = ArgType::MaxRowsAtCompileTime + }; + }; + } +} diff --git a/thirdparty/eigen/doc/examples/make_circulant2.cpp b/thirdparty/eigen/doc/examples/make_circulant2.cpp new file mode 100644 index 00000000..95d3dd31 --- /dev/null +++ b/thirdparty/eigen/doc/examples/make_circulant2.cpp @@ -0,0 +1,52 @@ +#include +#include + +using namespace Eigen; + +// [circulant_func] +template +class circulant_functor { + const ArgType &m_vec; +public: + circulant_functor(const ArgType& arg) : m_vec(arg) {} + + const typename ArgType::Scalar& operator() (Index row, Index col) const { + Index index = row - col; + if (index < 0) index += m_vec.size(); + return m_vec(index); + } +}; +// [circulant_func] + +// [square] +template +struct circulant_helper { + typedef Matrix MatrixType; +}; +// [square] + +// [makeCirculant] +template +CwiseNullaryOp, typename circulant_helper::MatrixType> +makeCirculant(const Eigen::MatrixBase& arg) +{ + typedef typename circulant_helper::MatrixType MatrixType; + return MatrixType::NullaryExpr(arg.size(), arg.size(), circulant_functor(arg.derived())); +} +// [makeCirculant] + +// [main] +int main() +{ + Eigen::VectorXd vec(4); + vec << 1, 2, 4, 8; + Eigen::MatrixXd mat; + mat = makeCirculant(vec); + std::cout << mat << std::endl; +} +// [main] diff --git a/thirdparty/eigen/doc/examples/matrixfree_cg.cpp b/thirdparty/eigen/doc/examples/matrixfree_cg.cpp index f0631c3a..cc0eead1 100644 --- a/thirdparty/eigen/doc/examples/matrixfree_cg.cpp +++ b/thirdparty/eigen/doc/examples/matrixfree_cg.cpp @@ -2,179 +2,128 @@ #include #include #include +#include class MatrixReplacement; -template class MatrixReplacement_ProductReturnType; +using Eigen::SparseMatrix; namespace Eigen { namespace internal { + // MatrixReplacement looks-like a SparseMatrix, so let's inherit its traits: template<> - struct traits : Eigen::internal::traits > + struct traits : public Eigen::internal::traits > {}; - - template - struct traits > { - // The equivalent plain objet type of the product. This type is used if the product needs to be evaluated into a temporary. - typedef Eigen::Matrix ReturnType; - }; } } -// Inheriting EigenBase should not be needed in the future. +// Example of a matrix-free wrapper from a user type to Eigen's compatible type +// For the sake of simplicity, this example simply wrap a Eigen::SparseMatrix. class MatrixReplacement : public Eigen::EigenBase { public: - // Expose some compile-time information to Eigen: + // Required typedefs, constants, and method: typedef double Scalar; typedef double RealScalar; + typedef int StorageIndex; enum { ColsAtCompileTime = Eigen::Dynamic, - RowsAtCompileTime = Eigen::Dynamic, MaxColsAtCompileTime = Eigen::Dynamic, - MaxRowsAtCompileTime = Eigen::Dynamic + IsRowMajor = false }; - Index rows() const { return 4; } - Index cols() const { return 4; } + Index rows() const { return mp_mat->rows(); } + Index cols() const { return mp_mat->cols(); } - void resize(Index a_rows, Index a_cols) - { - // This method should not be needed in the future. - assert(a_rows==0 && a_cols==0 || a_rows==rows() && a_cols==cols()); - } - - // In the future, the return type should be Eigen::Product template - MatrixReplacement_ProductReturnType operator*(const Eigen::MatrixBase& x) const { - return MatrixReplacement_ProductReturnType(*this, x.derived()); + Eigen::Product operator*(const Eigen::MatrixBase& x) const { + return Eigen::Product(*this, x.derived()); } -}; + // Custom API: + MatrixReplacement() : mp_mat(0) {} -// The proxy class representing the product of a MatrixReplacement with a MatrixBase<> -template -class MatrixReplacement_ProductReturnType : public Eigen::ReturnByValue > { -public: - typedef MatrixReplacement::Index Index; - - // The ctor store references to the matrix and right-hand-side object (usually a vector). - MatrixReplacement_ProductReturnType(const MatrixReplacement& matrix, const Rhs& rhs) - : m_matrix(matrix), m_rhs(rhs) - {} - - Index rows() const { return m_matrix.rows(); } - Index cols() const { return m_rhs.cols(); } - - // This function is automatically called by Eigen. It must evaluate the product of matrix * rhs into y. - template - void evalTo(Dest& y) const - { - y.setZero(4); - - y(0) += 2 * m_rhs(0); y(1) += 1 * m_rhs(0); - y(0) += 1 * m_rhs(1); y(1) += 2 * m_rhs(1); y(2) += 1 * m_rhs(1); - y(1) += 1 * m_rhs(2); y(2) += 2 * m_rhs(2); y(3) += 1 * m_rhs(2); - y(2) += 1 * m_rhs(3); y(3) += 2 * m_rhs(3); + void attachMyMatrix(const SparseMatrix &mat) { + mp_mat = &mat; } + const SparseMatrix my_matrix() const { return *mp_mat; } -protected: - const MatrixReplacement& m_matrix; - typename Rhs::Nested m_rhs; +private: + const SparseMatrix *mp_mat; }; -/*****/ - -// This class simply warp a diagonal matrix as a Jacobi preconditioner. -// In the future such simple and generic wrapper should be shipped within Eigen itsel. -template -class MyJacobiPreconditioner -{ - typedef _Scalar Scalar; - typedef Eigen::Matrix Vector; - typedef typename Vector::Index Index; - - public: - // this typedef is only to export the scalar type and compile-time dimensions to solve_retval - typedef Eigen::Matrix MatrixType; - - MyJacobiPreconditioner() : m_isInitialized(false) {} - - void setInvDiag(const Eigen::VectorXd &invdiag) { - m_invdiag=invdiag; - m_isInitialized=true; - } - - Index rows() const { return m_invdiag.size(); } - Index cols() const { return m_invdiag.size(); } - - template - MyJacobiPreconditioner& analyzePattern(const MatType& ) { return *this; } - - template - MyJacobiPreconditioner& factorize(const MatType& mat) { return *this; } - - template - MyJacobiPreconditioner& compute(const MatType& mat) { return *this; } - - template - void _solve(const Rhs& b, Dest& x) const - { - x = m_invdiag.array() * b.array() ; - } - - template inline const Eigen::internal::solve_retval - solve(const Eigen::MatrixBase& b) const - { - eigen_assert(m_isInitialized && "MyJacobiPreconditioner is not initialized."); - eigen_assert(m_invdiag.size()==b.rows() - && "MyJacobiPreconditioner::solve(): invalid number of rows of the right hand side matrix b"); - return Eigen::internal::solve_retval(*this, b.derived()); - } - - protected: - Vector m_invdiag; - bool m_isInitialized; -}; - +// Implementation of MatrixReplacement * Eigen::DenseVector though a specialization of internal::generic_product_impl: namespace Eigen { namespace internal { -template -struct solve_retval, Rhs> - : solve_retval_base, Rhs> -{ - typedef MyJacobiPreconditioner<_MatrixType> Dec; - EIGEN_MAKE_SOLVE_HELPERS(Dec,Rhs) - - template void evalTo(Dest& dst) const + template + struct generic_product_impl // GEMV stands for matrix-vector + : generic_product_impl_base > { - dec()._solve(rhs(),dst); - } -}; + typedef typename Product::Scalar Scalar; + + template + static void scaleAndAddTo(Dest& dst, const MatrixReplacement& lhs, const Rhs& rhs, const Scalar& alpha) + { + // This method should implement "dst += alpha * lhs * rhs" inplace, + // however, for iterative solvers, alpha is always equal to 1, so let's not bother about it. + eigen_assert(alpha==Scalar(1) && "scaling is not implemented"); + EIGEN_ONLY_USED_FOR_DEBUG(alpha); + + // Here we could simply call dst.noalias() += lhs.my_matrix() * rhs, + // but let's do something fancier (and less efficient): + for(Index i=0; i S = Eigen::MatrixXd::Random(n,n).sparseView(0.5,1); + S = S.transpose()*S; + MatrixReplacement A; - Eigen::VectorXd b(4), x; - b << 1, 1, 1, 1; + A.attachMyMatrix(S); - // solve Ax = b using CG with matrix-free version: - Eigen::ConjugateGradient < MatrixReplacement, Eigen::Lower|Eigen::Upper, MyJacobiPreconditioner > cg; + Eigen::VectorXd b(n), x; + b.setRandom(); - Eigen::VectorXd invdiag(4); - invdiag << 1./3., 1./4., 1./4., 1./3.; + // Solve Ax = b using various iterative solver with matrix-free version: + { + Eigen::ConjugateGradient cg; + cg.compute(A); + x = cg.solve(b); + std::cout << "CG: #iterations: " << cg.iterations() << ", estimated error: " << cg.error() << std::endl; + } + + { + Eigen::BiCGSTAB bicg; + bicg.compute(A); + x = bicg.solve(b); + std::cout << "BiCGSTAB: #iterations: " << bicg.iterations() << ", estimated error: " << bicg.error() << std::endl; + } + + { + Eigen::GMRES gmres; + gmres.compute(A); + x = gmres.solve(b); + std::cout << "GMRES: #iterations: " << gmres.iterations() << ", estimated error: " << gmres.error() << std::endl; + } - cg.preconditioner().setInvDiag(invdiag); - cg.compute(A); - x = cg.solve(b); + { + Eigen::DGMRES gmres; + gmres.compute(A); + x = gmres.solve(b); + std::cout << "DGMRES: #iterations: " << gmres.iterations() << ", estimated error: " << gmres.error() << std::endl; + } - std::cout << "#iterations: " << cg.iterations() << std::endl; - std::cout << "estimated error: " << cg.error() << std::endl; + { + Eigen::MINRES minres; + minres.compute(A); + x = minres.solve(b); + std::cout << "MINRES: #iterations: " << minres.iterations() << ", estimated error: " << minres.error() << std::endl; + } } diff --git a/thirdparty/eigen/doc/examples/nullary_indexing.cpp b/thirdparty/eigen/doc/examples/nullary_indexing.cpp new file mode 100644 index 00000000..b74db5fd --- /dev/null +++ b/thirdparty/eigen/doc/examples/nullary_indexing.cpp @@ -0,0 +1,66 @@ +#include +#include + +using namespace Eigen; + +// [functor] +template +class indexing_functor { + const ArgType &m_arg; + const RowIndexType &m_rowIndices; + const ColIndexType &m_colIndices; +public: + typedef Matrix MatrixType; + + indexing_functor(const ArgType& arg, const RowIndexType& row_indices, const ColIndexType& col_indices) + : m_arg(arg), m_rowIndices(row_indices), m_colIndices(col_indices) + {} + + const typename ArgType::Scalar& operator() (Index row, Index col) const { + return m_arg(m_rowIndices[row], m_colIndices[col]); + } +}; +// [functor] + +// [function] +template +CwiseNullaryOp, typename indexing_functor::MatrixType> +mat_indexing(const Eigen::MatrixBase& arg, const RowIndexType& row_indices, const ColIndexType& col_indices) +{ + typedef indexing_functor Func; + typedef typename Func::MatrixType MatrixType; + return MatrixType::NullaryExpr(row_indices.size(), col_indices.size(), Func(arg.derived(), row_indices, col_indices)); +} +// [function] + + +int main() +{ + std::cout << "[main1]\n"; + Eigen::MatrixXi A = Eigen::MatrixXi::Random(4,4); + Array3i ri(1,2,1); + ArrayXi ci(6); ci << 3,2,1,0,0,2; + Eigen::MatrixXi B = mat_indexing(A, ri, ci); + std::cout << "A =" << std::endl; + std::cout << A << std::endl << std::endl; + std::cout << "A([" << ri.transpose() << "], [" << ci.transpose() << "]) =" << std::endl; + std::cout << B << std::endl; + std::cout << "[main1]\n"; + + std::cout << "[main2]\n"; + B = mat_indexing(A, ri+1, ci); + std::cout << "A(ri+1,ci) =" << std::endl; + std::cout << B << std::endl << std::endl; +#if EIGEN_COMP_CXXVER >= 11 + B = mat_indexing(A, ArrayXi::LinSpaced(13,0,12).unaryExpr([](int x){return x%4;}), ArrayXi::LinSpaced(4,0,3)); + std::cout << "A(ArrayXi::LinSpaced(13,0,12).unaryExpr([](int x){return x%4;}), ArrayXi::LinSpaced(4,0,3)) =" << std::endl; + std::cout << B << std::endl << std::endl; +#endif + std::cout << "[main2]\n"; +} + diff --git a/thirdparty/eigen/doc/ftv2node.png b/thirdparty/eigen/doc/ftv2node.png new file mode 100644 index 00000000..63c605bb Binary files /dev/null and b/thirdparty/eigen/doc/ftv2node.png differ diff --git a/thirdparty/eigen/doc/ftv2pnode.png b/thirdparty/eigen/doc/ftv2pnode.png new file mode 100644 index 00000000..c6ee22f9 Binary files /dev/null and b/thirdparty/eigen/doc/ftv2pnode.png differ diff --git a/thirdparty/eigen/doc/snippets/Array_initializer_list_23_cxx11.cpp b/thirdparty/eigen/doc/snippets/Array_initializer_list_23_cxx11.cpp new file mode 100644 index 00000000..2c2166ea --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Array_initializer_list_23_cxx11.cpp @@ -0,0 +1,5 @@ +ArrayXXi a { + {1, 2, 3}, + {3, 4, 5} +}; +cout << a << endl; diff --git a/thirdparty/eigen/doc/snippets/Array_initializer_list_vector_cxx11.cpp b/thirdparty/eigen/doc/snippets/Array_initializer_list_vector_cxx11.cpp new file mode 100644 index 00000000..a668d84a --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Array_initializer_list_vector_cxx11.cpp @@ -0,0 +1,2 @@ +Array v {{1, 2, 3, 4, 5}}; +cout << v << endl; diff --git a/thirdparty/eigen/doc/snippets/Array_variadic_ctor_cxx11.cpp b/thirdparty/eigen/doc/snippets/Array_variadic_ctor_cxx11.cpp new file mode 100644 index 00000000..0e4ec446 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Array_variadic_ctor_cxx11.cpp @@ -0,0 +1,3 @@ +Array a(1, 2, 3, 4, 5, 6); +Array b {1, 2, 3}; +cout << a << "\n\n" << b << endl; diff --git a/thirdparty/eigen/doc/snippets/BiCGSTAB_simple.cpp b/thirdparty/eigen/doc/snippets/BiCGSTAB_simple.cpp new file mode 100644 index 00000000..8c8829fd --- /dev/null +++ b/thirdparty/eigen/doc/snippets/BiCGSTAB_simple.cpp @@ -0,0 +1,11 @@ + int n = 10000; + VectorXd x(n), b(n); + SparseMatrix A(n,n); + /* ... fill A and b ... */ + BiCGSTAB > solver; + solver.compute(A); + x = solver.solve(b); + std::cout << "#iterations: " << solver.iterations() << std::endl; + std::cout << "estimated error: " << solver.error() << std::endl; + /* ... update b ... */ + x = solver.solve(b); // solve again diff --git a/thirdparty/eigen/doc/snippets/BiCGSTAB_step_by_step.cpp b/thirdparty/eigen/doc/snippets/BiCGSTAB_step_by_step.cpp new file mode 100644 index 00000000..6c95d5a9 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/BiCGSTAB_step_by_step.cpp @@ -0,0 +1,14 @@ + int n = 10000; + VectorXd x(n), b(n); + SparseMatrix A(n,n); + /* ... fill A and b ... */ + BiCGSTAB > solver(A); + // start from a random solution + x = VectorXd::Random(n); + solver.setMaxIterations(1); + int i = 0; + do { + x = solver.solveWithGuess(b,x); + std::cout << i << " : " << solver.error() << std::endl; + ++i; + } while (solver.info()!=Success && i<100); diff --git a/thirdparty/eigen/doc/snippets/CMakeLists.txt b/thirdparty/eigen/doc/snippets/CMakeLists.txt index 1135900c..4cf1c6a0 100644 --- a/thirdparty/eigen/doc/snippets/CMakeLists.txt +++ b/thirdparty/eigen/doc/snippets/CMakeLists.txt @@ -6,23 +6,32 @@ foreach(snippet_src ${snippets_SRCS}) get_filename_component(snippet ${snippet_src} NAME_WE) set(compile_snippet_target compile_${snippet}) set(compile_snippet_src ${compile_snippet_target}.cpp) - file(READ ${snippet_src} snippet_source_code) - configure_file(${CMAKE_CURRENT_SOURCE_DIR}/compile_snippet.cpp.in - ${CMAKE_CURRENT_BINARY_DIR}/${compile_snippet_src}) - add_executable(${compile_snippet_target} - ${CMAKE_CURRENT_BINARY_DIR}/${compile_snippet_src}) - if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) - target_link_libraries(${compile_snippet_target} ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}) + if((NOT ${snippet_src} MATCHES "cxx11") OR EIGEN_COMPILER_SUPPORT_CPP11) + file(READ ${snippet_src} snippet_source_code) + configure_file(${CMAKE_CURRENT_SOURCE_DIR}/compile_snippet.cpp.in + ${CMAKE_CURRENT_BINARY_DIR}/${compile_snippet_src}) + add_executable(${compile_snippet_target} + ${CMAKE_CURRENT_BINARY_DIR}/${compile_snippet_src}) + if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) + target_link_libraries(${compile_snippet_target} ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}) + endif() + target_link_libraries(${compile_snippet_target} Eigen3::Eigen) + if(${snippet_src} MATCHES "cxx11") + set_target_properties(${compile_snippet_target} PROPERTIES COMPILE_FLAGS "-std=c++11") + endif() + if(${snippet_src} MATCHES "deprecated") + set_target_properties(${compile_snippet_target} PROPERTIES COMPILE_FLAGS "-DEIGEN_NO_DEPRECATED_WARNING") + endif() + add_custom_command( + TARGET ${compile_snippet_target} + POST_BUILD + COMMAND ${compile_snippet_target} + ARGS >${CMAKE_CURRENT_BINARY_DIR}/${snippet}.out + ) + add_dependencies(all_snippets ${compile_snippet_target}) + set_source_files_properties(${CMAKE_CURRENT_BINARY_DIR}/${compile_snippet_src} + PROPERTIES OBJECT_DEPENDS ${snippet_src}) + else() + message("skip snippet ${snippet_src} because compiler does not support C++11") endif() - add_custom_command( - TARGET ${compile_snippet_target} - POST_BUILD - COMMAND ${compile_snippet_target} - ARGS >${CMAKE_CURRENT_BINARY_DIR}/${snippet}.out - ) - add_dependencies(all_snippets ${compile_snippet_target}) - set_source_files_properties(${CMAKE_CURRENT_BINARY_DIR}/${compile_snippet_src} - PROPERTIES OBJECT_DEPENDS ${snippet_src}) -endforeach(snippet_src) - -ei_add_target_property(compile_tut_arithmetic_transpose_aliasing COMPILE_FLAGS -DEIGEN_NO_DEBUG) +endforeach() diff --git a/thirdparty/eigen/doc/snippets/ComplexEigenSolver_eigenvectors.cpp b/thirdparty/eigen/doc/snippets/ComplexEigenSolver_eigenvectors.cpp index bb1c2ccf..adeed9af 100644 --- a/thirdparty/eigen/doc/snippets/ComplexEigenSolver_eigenvectors.cpp +++ b/thirdparty/eigen/doc/snippets/ComplexEigenSolver_eigenvectors.cpp @@ -1,4 +1,4 @@ MatrixXcf ones = MatrixXcf::Ones(3,3); ComplexEigenSolver ces(ones); cout << "The first eigenvector of the 3x3 matrix of ones is:" - << endl << ces.eigenvectors().col(1) << endl; + << endl << ces.eigenvectors().col(0) << endl; diff --git a/thirdparty/eigen/doc/snippets/Cwise_arg.cpp b/thirdparty/eigen/doc/snippets/Cwise_arg.cpp new file mode 100644 index 00000000..3f45133b --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_arg.cpp @@ -0,0 +1,3 @@ +ArrayXcf v = ArrayXcf::Random(3); +cout << v << endl << endl; +cout << arg(v) << endl; diff --git a/thirdparty/eigen/doc/snippets/Cwise_array_power_array.cpp b/thirdparty/eigen/doc/snippets/Cwise_array_power_array.cpp new file mode 100644 index 00000000..432a76ee --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_array_power_array.cpp @@ -0,0 +1,4 @@ +Array x(8,25,3), + e(1./3.,0.5,2.); +cout << "[" << x << "]^[" << e << "] = " << x.pow(e) << endl; // using ArrayBase::pow +cout << "[" << x << "]^[" << e << "] = " << pow(x,e) << endl; // using Eigen::pow diff --git a/thirdparty/eigen/doc/snippets/Cwise_atan.cpp b/thirdparty/eigen/doc/snippets/Cwise_atan.cpp new file mode 100644 index 00000000..44684472 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_atan.cpp @@ -0,0 +1,2 @@ +ArrayXd v = ArrayXd::LinSpaced(5,0,1); +cout << v.atan() << endl; diff --git a/thirdparty/eigen/doc/snippets/Cwise_boolean_not.cpp b/thirdparty/eigen/doc/snippets/Cwise_boolean_not.cpp new file mode 100644 index 00000000..40009f15 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_boolean_not.cpp @@ -0,0 +1,5 @@ +Array3d v(1,2,3); +v(1) *= 0.0/0.0; +v(2) /= 0.0; +cout << v << endl << endl; +cout << !isfinite(v) << endl; diff --git a/thirdparty/eigen/doc/snippets/Cwise_boolean_xor.cpp b/thirdparty/eigen/doc/snippets/Cwise_boolean_xor.cpp new file mode 100644 index 00000000..fafbec80 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_boolean_xor.cpp @@ -0,0 +1,2 @@ +Array3d v(-1,2,1), w(-3,2,3); +cout << ((v e(2,-3,1./3.); +cout << "10^[" << e << "] = " << pow(10,e) << endl; diff --git a/thirdparty/eigen/doc/snippets/Cwise_sign.cpp b/thirdparty/eigen/doc/snippets/Cwise_sign.cpp new file mode 100644 index 00000000..49920e4f --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_sign.cpp @@ -0,0 +1,2 @@ +Array3d v(-3,5,0); +cout << v.sign() << endl; diff --git a/thirdparty/eigen/doc/snippets/Cwise_sinh.cpp b/thirdparty/eigen/doc/snippets/Cwise_sinh.cpp new file mode 100644 index 00000000..fac9b19a --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_sinh.cpp @@ -0,0 +1,2 @@ +ArrayXd v = ArrayXd::LinSpaced(5,0,1); +cout << sinh(v) << endl; diff --git a/thirdparty/eigen/doc/snippets/Cwise_tanh.cpp b/thirdparty/eigen/doc/snippets/Cwise_tanh.cpp new file mode 100644 index 00000000..30cd0450 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Cwise_tanh.cpp @@ -0,0 +1,2 @@ +ArrayXd v = ArrayXd::LinSpaced(5,0,1); +cout << tanh(v) << endl; diff --git a/thirdparty/eigen/doc/snippets/DenseBase_LinSpacedInt.cpp b/thirdparty/eigen/doc/snippets/DenseBase_LinSpacedInt.cpp new file mode 100644 index 00000000..0d7ae068 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/DenseBase_LinSpacedInt.cpp @@ -0,0 +1,8 @@ +cout << "Even spacing inputs:" << endl; +cout << VectorXi::LinSpaced(8,1,4).transpose() << endl; +cout << VectorXi::LinSpaced(8,1,8).transpose() << endl; +cout << VectorXi::LinSpaced(8,1,15).transpose() << endl; +cout << "Uneven spacing inputs:" << endl; +cout << VectorXi::LinSpaced(8,1,7).transpose() << endl; +cout << VectorXi::LinSpaced(8,1,9).transpose() << endl; +cout << VectorXi::LinSpaced(8,1,16).transpose() << endl; diff --git a/thirdparty/eigen/doc/snippets/DenseBase_LinSpaced_seq.cpp b/thirdparty/eigen/doc/snippets/DenseBase_LinSpaced_seq_deprecated.cpp similarity index 100% rename from thirdparty/eigen/doc/snippets/DenseBase_LinSpaced_seq.cpp rename to thirdparty/eigen/doc/snippets/DenseBase_LinSpaced_seq_deprecated.cpp diff --git a/thirdparty/eigen/doc/snippets/DirectionWise_hnormalized.cpp b/thirdparty/eigen/doc/snippets/DirectionWise_hnormalized.cpp new file mode 100644 index 00000000..2451f6e7 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/DirectionWise_hnormalized.cpp @@ -0,0 +1,6 @@ +Matrix4Xd M = Matrix4Xd::Random(4,5); +Projective3d P(Matrix4d::Random()); +cout << "The matrix M is:" << endl << M << endl << endl; +cout << "M.colwise().hnormalized():" << endl << M.colwise().hnormalized() << endl << endl; +cout << "P*M:" << endl << P*M << endl << endl; +cout << "(P*M).colwise().hnormalized():" << endl << (P*M).colwise().hnormalized() << endl << endl; diff --git a/thirdparty/eigen/doc/snippets/EigenSolver_eigenvectors.cpp b/thirdparty/eigen/doc/snippets/EigenSolver_eigenvectors.cpp index 0fad4dad..8355f76c 100644 --- a/thirdparty/eigen/doc/snippets/EigenSolver_eigenvectors.cpp +++ b/thirdparty/eigen/doc/snippets/EigenSolver_eigenvectors.cpp @@ -1,4 +1,4 @@ MatrixXd ones = MatrixXd::Ones(3,3); EigenSolver es(ones); -cout << "The first eigenvector of the 3x3 matrix of ones is:" - << endl << es.eigenvectors().col(1) << endl; +cout << "The first eigenvector of the 3x3 matrix of ones is:" + << endl << es.eigenvectors().col(0) << endl; diff --git a/thirdparty/eigen/doc/snippets/Jacobi_makeGivens.cpp b/thirdparty/eigen/doc/snippets/Jacobi_makeGivens.cpp index 4b733c30..6f8ec054 100644 --- a/thirdparty/eigen/doc/snippets/Jacobi_makeGivens.cpp +++ b/thirdparty/eigen/doc/snippets/Jacobi_makeGivens.cpp @@ -3,4 +3,4 @@ JacobiRotation G; G.makeGivens(v.x(), v.y()); cout << "Here is the vector v:" << endl << v << endl; v.applyOnTheLeft(0, 1, G.adjoint()); -cout << "Here is the vector J' * v:" << endl << v << endl; \ No newline at end of file +cout << "Here is the vector J' * v:" << endl << v << endl; diff --git a/thirdparty/eigen/doc/snippets/Jacobi_makeJacobi.cpp b/thirdparty/eigen/doc/snippets/Jacobi_makeJacobi.cpp index 0cc331d9..a86e80a6 100644 --- a/thirdparty/eigen/doc/snippets/Jacobi_makeJacobi.cpp +++ b/thirdparty/eigen/doc/snippets/Jacobi_makeJacobi.cpp @@ -5,4 +5,4 @@ J.makeJacobi(m, 0, 1); cout << "Here is the matrix m:" << endl << m << endl; m.applyOnTheLeft(0, 1, J.adjoint()); m.applyOnTheRight(0, 1, J); -cout << "Here is the matrix J' * m * J:" << endl << m << endl; \ No newline at end of file +cout << "Here is the matrix J' * m * J:" << endl << m << endl; diff --git a/thirdparty/eigen/doc/snippets/LeastSquaresNormalEquations.cpp b/thirdparty/eigen/doc/snippets/LeastSquaresNormalEquations.cpp new file mode 100644 index 00000000..997cf171 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/LeastSquaresNormalEquations.cpp @@ -0,0 +1,4 @@ +MatrixXf A = MatrixXf::Random(3, 2); +VectorXf b = VectorXf::Random(3); +cout << "The solution using normal equations is:\n" + << (A.transpose() * A).ldlt().solve(A.transpose() * b) << endl; diff --git a/thirdparty/eigen/doc/snippets/LeastSquaresQR.cpp b/thirdparty/eigen/doc/snippets/LeastSquaresQR.cpp new file mode 100644 index 00000000..6c970454 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/LeastSquaresQR.cpp @@ -0,0 +1,4 @@ +MatrixXf A = MatrixXf::Random(3, 2); +VectorXf b = VectorXf::Random(3); +cout << "The solution using the QR decomposition is:\n" + << A.colPivHouseholderQr().solve(b) << endl; diff --git a/thirdparty/eigen/doc/snippets/Map_placement_new.cpp b/thirdparty/eigen/doc/snippets/Map_placement_new.cpp index 2e40eca3..83b83a89 100644 --- a/thirdparty/eigen/doc/snippets/Map_placement_new.cpp +++ b/thirdparty/eigen/doc/snippets/Map_placement_new.cpp @@ -2,4 +2,4 @@ int data[] = {1,2,3,4,5,6,7,8,9}; Map v(data,4); cout << "The mapped vector v is: " << v << "\n"; new (&v) Map(data+4,5); -cout << "Now v is: " << v << "\n"; \ No newline at end of file +cout << "Now v is: " << v << "\n"; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_colwise_iterator_cxx11.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_colwise_iterator_cxx11.cpp new file mode 100644 index 00000000..116063fb --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_colwise_iterator_cxx11.cpp @@ -0,0 +1,12 @@ +Matrix3i m = Matrix3i::Random(); +cout << "Here is the initial matrix m:" << endl << m << endl; +int i = -1; +for(auto c: m.colwise()) { + c *= i; + ++i; +} +cout << "Here is the matrix m after the for-range-loop:" << endl << m << endl; +auto cols = m.colwise(); +auto it = std::find_if(cols.cbegin(), cols.cend(), + [](Matrix3i::ConstColXpr x) { return x.squaredNorm() == 0; }); +cout << "The first empty column is: " << distance(cols.cbegin(),it) << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_cwiseArg.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseArg.cpp new file mode 100644 index 00000000..e0857cf9 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseArg.cpp @@ -0,0 +1,3 @@ +MatrixXcf v = MatrixXcf::Random(2, 3); +cout << v << endl << endl; +cout << v.cwiseArg() << endl; \ No newline at end of file diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_cwiseEqual.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseEqual.cpp index eb3656f4..469af642 100644 --- a/thirdparty/eigen/doc/snippets/MatrixBase_cwiseEqual.cpp +++ b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseEqual.cpp @@ -3,5 +3,5 @@ m << 1, 0, 1, 1; cout << "Comparing m with identity matrix:" << endl; cout << m.cwiseEqual(MatrixXi::Identity(2,2)) << endl; -int count = m.cwiseEqual(MatrixXi::Identity(2,2)).count(); +Index count = m.cwiseEqual(MatrixXi::Identity(2,2)).count(); cout << "Number of coefficients that are equal: " << count << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_cwiseNotEqual.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseNotEqual.cpp index 6a2e4fb6..7f0a105d 100644 --- a/thirdparty/eigen/doc/snippets/MatrixBase_cwiseNotEqual.cpp +++ b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseNotEqual.cpp @@ -3,5 +3,5 @@ m << 1, 0, 1, 1; cout << "Comparing m with identity matrix:" << endl; cout << m.cwiseNotEqual(MatrixXi::Identity(2,2)) << endl; -int count = m.cwiseNotEqual(MatrixXi::Identity(2,2)).count(); +Index count = m.cwiseNotEqual(MatrixXi::Identity(2,2)).count(); cout << "Number of coefficients that are not equal: " << count << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_cwiseSign.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseSign.cpp new file mode 100644 index 00000000..efd71795 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_cwiseSign.cpp @@ -0,0 +1,4 @@ +MatrixXd m(2,3); +m << 2, -4, 6, + -5, 1, 0; +cout << m.cwiseSign() << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_hnormalized.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_hnormalized.cpp new file mode 100644 index 00000000..b714adcc --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_hnormalized.cpp @@ -0,0 +1,6 @@ +Vector4d v = Vector4d::Random(); +Projective3d P(Matrix4d::Random()); +cout << "v = " << v.transpose() << "]^T" << endl; +cout << "v.hnormalized() = " << v.hnormalized().transpose() << "]^T" << endl; +cout << "P*v = " << (P*v).transpose() << "]^T" << endl; +cout << "(P*v).hnormalized() = " << (P*v).hnormalized().transpose() << "]^T" << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_homogeneous.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_homogeneous.cpp new file mode 100644 index 00000000..26319609 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_homogeneous.cpp @@ -0,0 +1,6 @@ +Vector3d v = Vector3d::Random(), w; +Projective3d P(Matrix4d::Random()); +cout << "v = [" << v.transpose() << "]^T" << endl; +cout << "h.homogeneous() = [" << v.homogeneous().transpose() << "]^T" << endl; +cout << "(P * v.homogeneous()) = [" << (P * v.homogeneous()).transpose() << "]^T" << endl; +cout << "(P * v.homogeneous()).hnormalized() = [" << (P * v.homogeneous()).eval().hnormalized().transpose() << "]^T" << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_marked.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_marked.cpp deleted file mode 100644 index f6071217..00000000 --- a/thirdparty/eigen/doc/snippets/MatrixBase_marked.cpp +++ /dev/null @@ -1,14 +0,0 @@ -#ifndef _MSC_VER - #warning deprecated -#endif -/* -Matrix3d m = Matrix3d::Zero(); -m.part().setOnes(); -cout << "Here is the matrix m:" << endl << m << endl; -Matrix3d n = Matrix3d::Ones(); -n.part() *= 2; -cout << "Here is the matrix n:" << endl << n << endl; -cout << "And now here is m.inverse()*n, taking advantage of the fact that" - " m is upper-triangular:" << endl - << m.marked().solveTriangular(n); -*/ \ No newline at end of file diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_part.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_part.cpp deleted file mode 100644 index d3e7f482..00000000 --- a/thirdparty/eigen/doc/snippets/MatrixBase_part.cpp +++ /dev/null @@ -1,13 +0,0 @@ -#ifndef _MSC_VER - #warning deprecated -#endif -/* -Matrix3d m = Matrix3d::Zero(); -m.part().setOnes(); -cout << "Here is the matrix m:" << endl << m << endl; -cout << "And let us now compute m*m.adjoint() in a very optimized way" << endl - << "taking advantage of the symmetry." << endl; -Matrix3d n; -n.part() = (m*m.adjoint()).lazy(); -cout << "The result is:" << endl << n << endl; -*/ \ No newline at end of file diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_auto.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_auto.cpp new file mode 100644 index 00000000..59f9d3f6 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_auto.cpp @@ -0,0 +1,4 @@ +Matrix4i m = Matrix4i::Random(); +cout << "Here is the matrix m:" << endl << m << endl; +cout << "Here is m.reshaped(2, AutoSize):" << endl << m.reshaped(2, AutoSize) << endl; +cout << "Here is m.reshaped(AutoSize, fix<8>):" << endl << m.reshaped(AutoSize, fix<8>) << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_fixed.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_fixed.cpp new file mode 100644 index 00000000..3e9e2cfb --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_fixed.cpp @@ -0,0 +1,3 @@ +Matrix4i m = Matrix4i::Random(); +cout << "Here is the matrix m:" << endl << m << endl; +cout << "Here is m.reshaped(fix<2>,fix<8>):" << endl << m.reshaped(fix<2>,fix<8>) << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_int_int.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_int_int.cpp new file mode 100644 index 00000000..af4ca592 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_int_int.cpp @@ -0,0 +1,3 @@ +Matrix4i m = Matrix4i::Random(); +cout << "Here is the matrix m:" << endl << m << endl; +cout << "Here is m.reshaped(2, 8):" << endl << m.reshaped(2, 8) << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_to_vector.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_to_vector.cpp new file mode 100644 index 00000000..37f65f7c --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_reshaped_to_vector.cpp @@ -0,0 +1,4 @@ +Matrix4i m = Matrix4i::Random(); +cout << "Here is the matrix m:" << endl << m << endl; +cout << "Here is m.reshaped().transpose():" << endl << m.reshaped().transpose() << endl; +cout << "Here is m.reshaped().transpose(): " << endl << m.reshaped().transpose() << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_selfadjointView.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_selfadjointView.cpp new file mode 100644 index 00000000..4bd3c7ee --- /dev/null +++ b/thirdparty/eigen/doc/snippets/MatrixBase_selfadjointView.cpp @@ -0,0 +1,6 @@ +Matrix3i m = Matrix3i::Random(); +cout << "Here is the matrix m:" << endl << m << endl; +cout << "Here is the symmetric matrix extracted from the upper part of m:" << endl + << Matrix3i(m.selfadjointView()) << endl; +cout << "Here is the symmetric matrix extracted from the lower part of m:" << endl + << Matrix3i(m.selfadjointView()) << endl; diff --git a/thirdparty/eigen/doc/snippets/MatrixBase_extract.cpp b/thirdparty/eigen/doc/snippets/MatrixBase_triangularView.cpp similarity index 55% rename from thirdparty/eigen/doc/snippets/MatrixBase_extract.cpp rename to thirdparty/eigen/doc/snippets/MatrixBase_triangularView.cpp index c96220f7..03aa303f 100644 --- a/thirdparty/eigen/doc/snippets/MatrixBase_extract.cpp +++ b/thirdparty/eigen/doc/snippets/MatrixBase_triangularView.cpp @@ -1,13 +1,9 @@ -#ifndef _MSC_VER - #warning deprecated -#endif -/* deprecated Matrix3i m = Matrix3i::Random(); cout << "Here is the matrix m:" << endl << m << endl; cout << "Here is the upper-triangular matrix extracted from m:" << endl - << m.part() << endl; + << Matrix3i(m.triangularView()) << endl; cout << "Here is the strictly-upper-triangular matrix extracted from m:" << endl - << m.part() << endl; + << Matrix3i(m.triangularView()) << endl; cout << "Here is the unit-lower-triangular matrix extracted from m:" << endl - << m.part() << endl; -*/ \ No newline at end of file + << Matrix3i(m.triangularView()) << endl; +// FIXME need to implement output for triangularViews (Bug 885) diff --git a/thirdparty/eigen/doc/snippets/Matrix_Map_stride.cpp b/thirdparty/eigen/doc/snippets/Matrix_Map_stride.cpp new file mode 100644 index 00000000..ae42a127 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Matrix_Map_stride.cpp @@ -0,0 +1,7 @@ +Matrix4i A; +A << 1, 2, 3, 4, + 5, 6, 7, 8, + 9, 10, 11, 12, + 13, 14, 15, 16; + +std::cout << Matrix2i::Map(&A(1,1),Stride<8,2>()) << std::endl; diff --git a/thirdparty/eigen/doc/snippets/Matrix_initializer_list_23_cxx11.cpp b/thirdparty/eigen/doc/snippets/Matrix_initializer_list_23_cxx11.cpp new file mode 100644 index 00000000..60280ab5 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Matrix_initializer_list_23_cxx11.cpp @@ -0,0 +1,5 @@ +MatrixXd m { + {1, 2, 3}, + {4, 5, 6} +}; +cout << m << endl; diff --git a/thirdparty/eigen/doc/snippets/Matrix_initializer_list_vector_cxx11.cpp b/thirdparty/eigen/doc/snippets/Matrix_initializer_list_vector_cxx11.cpp new file mode 100644 index 00000000..325257cb --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Matrix_initializer_list_vector_cxx11.cpp @@ -0,0 +1,2 @@ +VectorXi v {{1, 2}}; +cout << v << endl; diff --git a/thirdparty/eigen/doc/snippets/Matrix_variadic_ctor_cxx11.cpp b/thirdparty/eigen/doc/snippets/Matrix_variadic_ctor_cxx11.cpp new file mode 100644 index 00000000..06d33f57 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Matrix_variadic_ctor_cxx11.cpp @@ -0,0 +1,3 @@ +Matrix a(1, 2, 3, 4, 5, 6); +Matrix b {1, 2, 3}; +cout << a << "\n\n" << b << endl; diff --git a/thirdparty/eigen/doc/snippets/PartialRedux_count.cpp b/thirdparty/eigen/doc/snippets/PartialRedux_count.cpp index c7b3097e..1c3b3a28 100644 --- a/thirdparty/eigen/doc/snippets/PartialRedux_count.cpp +++ b/thirdparty/eigen/doc/snippets/PartialRedux_count.cpp @@ -1,3 +1,5 @@ Matrix3d m = Matrix3d::Random(); cout << "Here is the matrix m:" << endl << m << endl; -cout << "Here is the count of elements larger or equal than 0.5 of each row:" << endl << (m.array() >= 0.5).rowwise().count() << endl; +Matrix res = (m.array() >= 0.5).rowwise().count(); +cout << "Here is the count of elements larger or equal than 0.5 of each row:" << endl; +cout << res << endl; diff --git a/thirdparty/eigen/doc/snippets/SelfAdjointEigenSolver_eigenvectors.cpp b/thirdparty/eigen/doc/snippets/SelfAdjointEigenSolver_eigenvectors.cpp index cfc8b0d5..94b0d6eb 100644 --- a/thirdparty/eigen/doc/snippets/SelfAdjointEigenSolver_eigenvectors.cpp +++ b/thirdparty/eigen/doc/snippets/SelfAdjointEigenSolver_eigenvectors.cpp @@ -1,4 +1,4 @@ MatrixXd ones = MatrixXd::Ones(3,3); SelfAdjointEigenSolver es(ones); cout << "The first eigenvector of the 3x3 matrix of ones is:" - << endl << es.eigenvectors().col(1) << endl; + << endl << es.eigenvectors().col(0) << endl; diff --git a/thirdparty/eigen/doc/snippets/Slicing_arrayexpr.cpp b/thirdparty/eigen/doc/snippets/Slicing_arrayexpr.cpp new file mode 100644 index 00000000..2df81809 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Slicing_arrayexpr.cpp @@ -0,0 +1,4 @@ +ArrayXi ind(5); ind<<4,2,5,5,3; +MatrixXi A = MatrixXi::Random(4,6); +cout << "Initial matrix A:\n" << A << "\n\n"; +cout << "A(all,ind-1):\n" << A(all,ind-1) << "\n\n"; diff --git a/thirdparty/eigen/doc/snippets/Slicing_custom_padding_cxx11.cpp b/thirdparty/eigen/doc/snippets/Slicing_custom_padding_cxx11.cpp new file mode 100644 index 00000000..24db98b7 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Slicing_custom_padding_cxx11.cpp @@ -0,0 +1,12 @@ +struct pad { + Index size() const { return out_size; } + Index operator[] (Index i) const { return std::max(0,i-(out_size-in_size)); } + Index in_size, out_size; +}; + +Matrix3i A; +A.reshaped() = VectorXi::LinSpaced(9,1,9); +cout << "Initial matrix A:\n" << A << "\n\n"; +MatrixXi B(5,5); +B = A(pad{3,5}, pad{3,5}); +cout << "A(pad{3,N}, pad{3,N}):\n" << B << "\n\n"; diff --git a/thirdparty/eigen/doc/snippets/Slicing_rawarray_cxx11.cpp b/thirdparty/eigen/doc/snippets/Slicing_rawarray_cxx11.cpp new file mode 100644 index 00000000..1087131a --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Slicing_rawarray_cxx11.cpp @@ -0,0 +1,5 @@ +#if EIGEN_HAS_STATIC_ARRAY_TEMPLATE +MatrixXi A = MatrixXi::Random(4,6); +cout << "Initial matrix A:\n" << A << "\n\n"; +cout << "A(all,{4,2,5,5,3}):\n" << A(all,{4,2,5,5,3}) << "\n\n"; +#endif diff --git a/thirdparty/eigen/doc/snippets/Slicing_stdvector_cxx11.cpp b/thirdparty/eigen/doc/snippets/Slicing_stdvector_cxx11.cpp new file mode 100644 index 00000000..555f6625 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Slicing_stdvector_cxx11.cpp @@ -0,0 +1,4 @@ +std::vector ind{4,2,5,5,3}; +MatrixXi A = MatrixXi::Random(4,6); +cout << "Initial matrix A:\n" << A << "\n\n"; +cout << "A(all,ind):\n" << A(all,ind) << "\n\n"; diff --git a/thirdparty/eigen/doc/snippets/SparseMatrix_coeffs.cpp b/thirdparty/eigen/doc/snippets/SparseMatrix_coeffs.cpp new file mode 100644 index 00000000..f71a69b0 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/SparseMatrix_coeffs.cpp @@ -0,0 +1,9 @@ +SparseMatrix A(3,3); +A.insert(1,2) = 0; +A.insert(0,1) = 1; +A.insert(2,0) = 2; +A.makeCompressed(); +cout << "The matrix A is:" << endl << MatrixXd(A) << endl; +cout << "it has " << A.nonZeros() << " stored non zero coefficients that are: " << A.coeffs().transpose() << endl; +A.coeffs() += 10; +cout << "After adding 10 to every stored non zero coefficient, the matrix A is:" << endl << MatrixXd(A) << endl; diff --git a/thirdparty/eigen/doc/snippets/TopicAliasing_mult4.cpp b/thirdparty/eigen/doc/snippets/TopicAliasing_mult4.cpp new file mode 100644 index 00000000..01c1c6d7 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/TopicAliasing_mult4.cpp @@ -0,0 +1,5 @@ +MatrixXf A(2,2), B(3,2); +B << 2, 0, 0, 3, 1, 1; +A << 2, 0, 0, -2; +A = (B * A).cwiseAbs(); +cout << A; diff --git a/thirdparty/eigen/doc/snippets/TopicAliasing_mult5.cpp b/thirdparty/eigen/doc/snippets/TopicAliasing_mult5.cpp new file mode 100644 index 00000000..1a36defd --- /dev/null +++ b/thirdparty/eigen/doc/snippets/TopicAliasing_mult5.cpp @@ -0,0 +1,5 @@ +MatrixXf A(2,2), B(3,2); +B << 2, 0, 0, 3, 1, 1; +A << 2, 0, 0, -2; +A = (B * A).eval().cwiseAbs(); +cout << A; diff --git a/thirdparty/eigen/doc/snippets/Triangular_solve.cpp b/thirdparty/eigen/doc/snippets/Triangular_solve.cpp new file mode 100644 index 00000000..54844246 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Triangular_solve.cpp @@ -0,0 +1,11 @@ +Matrix3d m = Matrix3d::Zero(); +m.triangularView().setOnes(); +cout << "Here is the matrix m:\n" << m << endl; +Matrix3d n = Matrix3d::Ones(); +n.triangularView() *= 2; +cout << "Here is the matrix n:\n" << n << endl; +cout << "And now here is m.inverse()*n, taking advantage of the fact that" + " m is upper-triangular:\n" + << m.triangularView().solve(n) << endl; +cout << "And this is n*m.inverse():\n" + << m.triangularView().solve(n); diff --git a/thirdparty/eigen/doc/snippets/Tridiagonalization_decomposeInPlace.cpp b/thirdparty/eigen/doc/snippets/Tridiagonalization_decomposeInPlace.cpp index 93dcfca1..3cdce679 100644 --- a/thirdparty/eigen/doc/snippets/Tridiagonalization_decomposeInPlace.cpp +++ b/thirdparty/eigen/doc/snippets/Tridiagonalization_decomposeInPlace.cpp @@ -4,7 +4,8 @@ cout << "Here is a random symmetric 5x5 matrix:" << endl << A << endl << endl; VectorXd diag(5); VectorXd subdiag(4); -internal::tridiagonalization_inplace(A, diag, subdiag, true); +VectorXd hcoeffs(4); // Scratch space for householder reflector. +internal::tridiagonalization_inplace(A, diag, subdiag, hcoeffs, true); cout << "The orthogonal matrix Q is:" << endl << A << endl; cout << "The diagonal of the tridiagonal matrix T is:" << endl << diag << endl; cout << "The subdiagonal of the tridiagonal matrix T is:" << endl << subdiag << endl; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_AdvancedInitialization_Join.cpp b/thirdparty/eigen/doc/snippets/Tutorial_AdvancedInitialization_Join.cpp index 84e8715c..55a21539 100644 --- a/thirdparty/eigen/doc/snippets/Tutorial_AdvancedInitialization_Join.cpp +++ b/thirdparty/eigen/doc/snippets/Tutorial_AdvancedInitialization_Join.cpp @@ -3,7 +3,7 @@ vec1 << 1, 2, 3; std::cout << "vec1 = " << vec1 << std::endl; RowVectorXd vec2(4); -vec2 << 1, 4, 9, 16;; +vec2 << 1, 4, 9, 16; std::cout << "vec2 = " << vec2 << std::endl; RowVectorXd joined(7); diff --git a/thirdparty/eigen/doc/snippets/Tutorial_ReshapeMat2Mat.cpp b/thirdparty/eigen/doc/snippets/Tutorial_ReshapeMat2Mat.cpp new file mode 100644 index 00000000..737afecb --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_ReshapeMat2Mat.cpp @@ -0,0 +1,6 @@ +MatrixXf M1(2,6); // Column-major storage +M1 << 1, 2, 3, 4, 5, 6, + 7, 8, 9, 10, 11, 12; + +Map M2(M1.data(), 6,2); +cout << "M2:" << endl << M2 << endl; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_ReshapeMat2Vec.cpp b/thirdparty/eigen/doc/snippets/Tutorial_ReshapeMat2Vec.cpp new file mode 100644 index 00000000..32980a79 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_ReshapeMat2Vec.cpp @@ -0,0 +1,11 @@ +MatrixXf M1(3,3); // Column-major storage +M1 << 1, 2, 3, + 4, 5, 6, + 7, 8, 9; + +Map v1(M1.data(), M1.size()); +cout << "v1:" << endl << v1 << endl; + +Matrix M2(M1); +Map v2(M2.data(), M2.size()); +cout << "v2:" << endl << v2 << endl; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_SlicingCol.cpp b/thirdparty/eigen/doc/snippets/Tutorial_SlicingCol.cpp new file mode 100644 index 00000000..695d1301 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_SlicingCol.cpp @@ -0,0 +1,11 @@ +MatrixXf M1 = MatrixXf::Random(3,8); +cout << "Column major input:" << endl << M1 << "\n"; +Map > M2(M1.data(), M1.rows(), (M1.cols()+2)/3, OuterStride<>(M1.outerStride()*3)); +cout << "1 column over 3:" << endl << M2 << "\n"; + +typedef Matrix RowMajorMatrixXf; +RowMajorMatrixXf M3(M1); +cout << "Row major input:" << endl << M3 << "\n"; +Map > M4(M3.data(), M3.rows(), (M3.cols()+2)/3, + Stride(M3.outerStride(),3)); +cout << "1 column over 3:" << endl << M4 << "\n"; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_SlicingVec.cpp b/thirdparty/eigen/doc/snippets/Tutorial_SlicingVec.cpp new file mode 100644 index 00000000..9b822464 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_SlicingVec.cpp @@ -0,0 +1,4 @@ +RowVectorXf v = RowVectorXf::LinSpaced(20,0,19); +cout << "Input:" << endl << v << endl; +Map > v2(v.data(), v.size()/2); +cout << "Even:" << v2 << endl; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_range_for_loop_1d_cxx11.cpp b/thirdparty/eigen/doc/snippets/Tutorial_range_for_loop_1d_cxx11.cpp new file mode 100644 index 00000000..e72e715d --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_range_for_loop_1d_cxx11.cpp @@ -0,0 +1,4 @@ +VectorXi v = VectorXi::Random(4); +cout << "Here is the vector v:\n"; +for(auto x : v) cout << x << " "; +cout << "\n"; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_range_for_loop_2d_cxx11.cpp b/thirdparty/eigen/doc/snippets/Tutorial_range_for_loop_2d_cxx11.cpp new file mode 100644 index 00000000..4a12d26c --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_range_for_loop_2d_cxx11.cpp @@ -0,0 +1,5 @@ +Matrix2i A = Matrix2i::Random(); +cout << "Here are the coeffs of the 2x2 matrix A:\n"; +for(auto x : A.reshaped()) + cout << x << " "; +cout << "\n"; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_reshaped_vs_resize_1.cpp b/thirdparty/eigen/doc/snippets/Tutorial_reshaped_vs_resize_1.cpp new file mode 100644 index 00000000..e520e8e6 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_reshaped_vs_resize_1.cpp @@ -0,0 +1,5 @@ +MatrixXi m = Matrix4i::Random(); +cout << "Here is the matrix m:" << endl << m << endl; +cout << "Here is m.reshaped(2, 8):" << endl << m.reshaped(2, 8) << endl; +m.resize(2,8); +cout << "Here is the matrix m after m.resize(2,8):" << endl << m << endl; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_reshaped_vs_resize_2.cpp b/thirdparty/eigen/doc/snippets/Tutorial_reshaped_vs_resize_2.cpp new file mode 100644 index 00000000..50dc4548 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_reshaped_vs_resize_2.cpp @@ -0,0 +1,6 @@ +Matrix m = Matrix4i::Random(); +cout << "Here is the matrix m:" << endl << m << endl; +cout << "Here is m.reshaped(2, 8):" << endl << m.reshaped(2, 8) << endl; +cout << "Here is m.reshaped(2, 8):" << endl << m.reshaped(2, 8) << endl; +m.resize(2,8); +cout << "Here is the matrix m after m.resize(2,8):" << endl << m << endl; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_std_sort.cpp b/thirdparty/eigen/doc/snippets/Tutorial_std_sort.cpp new file mode 100644 index 00000000..cde2a6f1 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_std_sort.cpp @@ -0,0 +1,4 @@ +Array4i v = Array4i::Random().abs(); +cout << "Here is the initial vector v:\n" << v.transpose() << "\n"; +std::sort(v.begin(), v.end()); +cout << "Here is the sorted vector v:\n" << v.transpose() << "\n"; diff --git a/thirdparty/eigen/doc/snippets/Tutorial_std_sort_rows_cxx11.cpp b/thirdparty/eigen/doc/snippets/Tutorial_std_sort_rows_cxx11.cpp new file mode 100644 index 00000000..03641603 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/Tutorial_std_sort_rows_cxx11.cpp @@ -0,0 +1,5 @@ +ArrayXXi A = ArrayXXi::Random(4,4).abs(); +cout << "Here is the initial matrix A:\n" << A << "\n"; +for(auto row : A.rowwise()) + std::sort(row.begin(), row.end()); +cout << "Here is the sorted matrix A:\n" << A << "\n"; diff --git a/thirdparty/eigen/doc/snippets/VectorwiseOp_homogeneous.cpp b/thirdparty/eigen/doc/snippets/VectorwiseOp_homogeneous.cpp new file mode 100644 index 00000000..67cf5737 --- /dev/null +++ b/thirdparty/eigen/doc/snippets/VectorwiseOp_homogeneous.cpp @@ -0,0 +1,6 @@ +Matrix3Xd M = Matrix3Xd::Random(3,5); +Projective3d P(Matrix4d::Random()); +cout << "The matrix M is:" << endl << M << endl << endl; +cout << "M.colwise().homogeneous():" << endl << M.colwise().homogeneous() << endl << endl; +cout << "P * M.colwise().homogeneous():" << endl << P * M.colwise().homogeneous() << endl << endl; +cout << "P * M.colwise().homogeneous().hnormalized(): " << endl << (P * M.colwise().homogeneous()).colwise().hnormalized() << endl << endl; diff --git a/thirdparty/eigen/doc/snippets/compile_snippet.cpp.in b/thirdparty/eigen/doc/snippets/compile_snippet.cpp.in index 894cd526..04f276d0 100644 --- a/thirdparty/eigen/doc/snippets/compile_snippet.cpp.in +++ b/thirdparty/eigen/doc/snippets/compile_snippet.cpp.in @@ -1,5 +1,14 @@ -#include +static bool eigen_did_assert = false; +#define eigen_assert(X) if(!eigen_did_assert && !(X)){ std::cout << "### Assertion raised in " << __FILE__ << ":" << __LINE__ << ":\n" #X << "\n### The following would happen without assertions:\n"; eigen_did_assert = true;} + #include +#include +#include + +#ifndef M_PI +#define M_PI 3.1415926535897932384626433832795 +#endif + using namespace Eigen; using namespace std; @@ -7,6 +16,9 @@ using namespace std; int main(int, char**) { cout.precision(3); - ${snippet_source_code} +// intentionally remove indentation of snippet +{ +${snippet_source_code} +} return 0; } diff --git a/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_aliasing.cpp b/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_aliasing.cpp index c8e4746d..f82e6f2a 100644 --- a/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_aliasing.cpp +++ b/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_aliasing.cpp @@ -2,4 +2,4 @@ Matrix2i a; a << 1, 2, 3, 4; cout << "Here is the matrix a:\n" << a << endl; a = a.transpose(); // !!! do NOT do this !!! -cout << "and the result of the aliasing effect:\n" << a << endl; \ No newline at end of file +cout << "and the result of the aliasing effect:\n" << a << endl; diff --git a/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_inplace.cpp b/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_inplace.cpp index 7a069ff2..5c81c9e0 100644 --- a/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_inplace.cpp +++ b/thirdparty/eigen/doc/snippets/tut_arithmetic_transpose_inplace.cpp @@ -3,4 +3,4 @@ cout << "Here is the initial matrix a:\n" << a << endl; a.transposeInPlace(); -cout << "and after being transposed:\n" << a << endl; \ No newline at end of file +cout << "and after being transposed:\n" << a << endl; diff --git a/thirdparty/eigen/doc/special_examples/CMakeLists.txt b/thirdparty/eigen/doc/special_examples/CMakeLists.txt index 3ab75dbf..a0267a51 100644 --- a/thirdparty/eigen/doc/special_examples/CMakeLists.txt +++ b/thirdparty/eigen/doc/special_examples/CMakeLists.txt @@ -3,11 +3,11 @@ if(NOT EIGEN_TEST_NOQT) if(QT4_FOUND) include(${QT_USE_FILE}) endif() -endif(NOT EIGEN_TEST_NOQT) +endif() if(QT4_FOUND) add_executable(Tutorial_sparse_example Tutorial_sparse_example.cpp Tutorial_sparse_example_details.cpp) - target_link_libraries(Tutorial_sparse_example ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO} ${QT_QTCORE_LIBRARY} ${QT_QTGUI_LIBRARY}) + target_link_libraries(Tutorial_sparse_example ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO} ${QT_QTCORE_LIBRARY} ${QT_QTGUI_LIBRARY} Eigen3::Eigen) add_custom_command( TARGET Tutorial_sparse_example @@ -17,5 +17,18 @@ if(QT4_FOUND) ) add_dependencies(all_examples Tutorial_sparse_example) -endif(QT4_FOUND) +endif() +if(EIGEN_COMPILER_SUPPORT_CPP11) + add_executable(random_cpp11 random_cpp11.cpp) + target_link_libraries(random_cpp11 ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO} Eigen3::Eigen) + add_dependencies(all_examples random_cpp11) + target_compile_options(random_cpp11 PRIVATE "-std=c++11") + + add_custom_command( + TARGET random_cpp11 + POST_BUILD + COMMAND random_cpp11 + ARGS >${CMAKE_CURRENT_BINARY_DIR}/random_cpp11.out + ) +endif() diff --git a/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example.cpp b/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example.cpp index 002f19f0..8850db05 100644 --- a/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example.cpp +++ b/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example.cpp @@ -1,5 +1,6 @@ #include #include +#include typedef Eigen::SparseMatrix SpMat; // declares a column-major sparse matrix type of double typedef Eigen::Triplet T; @@ -9,8 +10,13 @@ void saveAsBitmap(const Eigen::VectorXd& x, int n, const char* filename); int main(int argc, char** argv) { + if(argc!=2) { + std::cerr << "Error: expected one and only one argument.\n"; + return -1; + } + int n = 300; // size of the image - int m = n*n; // number of unknows (=number of pixels) + int m = n*n; // number of unknowns (=number of pixels) // Assembly: std::vector coefficients; // list of non-zeros coefficients diff --git a/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example_details.cpp b/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example_details.cpp index 7d820b44..bc18b018 100644 --- a/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example_details.cpp +++ b/thirdparty/eigen/doc/special_examples/Tutorial_sparse_example_details.cpp @@ -8,7 +8,7 @@ typedef Eigen::Triplet T; void insertCoefficient(int id, int i, int j, double w, std::vector& coeffs, Eigen::VectorXd& b, const Eigen::VectorXd& boundary) { - int n = boundary.size(); + int n = int(boundary.size()); int id1 = i+j*n; if(i==-1 || i==n) b(id) -= w * boundary(j); // constrained coefficient diff --git a/thirdparty/eigen/doc/special_examples/random_cpp11.cpp b/thirdparty/eigen/doc/special_examples/random_cpp11.cpp new file mode 100644 index 00000000..33744c05 --- /dev/null +++ b/thirdparty/eigen/doc/special_examples/random_cpp11.cpp @@ -0,0 +1,14 @@ +#include +#include +#include + +using namespace Eigen; + +int main() { + std::default_random_engine generator; + std::poisson_distribution distribution(4.1); + auto poisson = [&] () {return distribution(generator);}; + + RowVectorXi v = RowVectorXi::NullaryExpr(10, poisson ); + std::cout << v << "\n"; +} diff --git a/thirdparty/eigen/failtest/CMakeLists.txt b/thirdparty/eigen/failtest/CMakeLists.txt index cadc6a25..256e541e 100644 --- a/thirdparty/eigen/failtest/CMakeLists.txt +++ b/thirdparty/eigen/failtest/CMakeLists.txt @@ -1,4 +1,3 @@ -message(STATUS "Running the failtests") ei_add_failtest("failtest_sanity_check") @@ -7,6 +6,9 @@ ei_add_failtest("block_nonconst_ctor_on_const_xpr_1") ei_add_failtest("block_nonconst_ctor_on_const_xpr_2") ei_add_failtest("transpose_nonconst_ctor_on_const_xpr") ei_add_failtest("diagonal_nonconst_ctor_on_const_xpr") +ei_add_failtest("cwiseunaryview_nonconst_ctor_on_const_xpr") +ei_add_failtest("triangularview_nonconst_ctor_on_const_xpr") +ei_add_failtest("selfadjointview_nonconst_ctor_on_const_xpr") ei_add_failtest("const_qualified_block_method_retval_0") ei_add_failtest("const_qualified_block_method_retval_1") @@ -25,6 +27,9 @@ ei_add_failtest("block_on_const_type_actually_const_0") ei_add_failtest("block_on_const_type_actually_const_1") ei_add_failtest("transpose_on_const_type_actually_const") ei_add_failtest("diagonal_on_const_type_actually_const") +ei_add_failtest("cwiseunaryview_on_const_type_actually_const") +ei_add_failtest("triangularview_on_const_type_actually_const") +ei_add_failtest("selfadjointview_on_const_type_actually_const") ei_add_failtest("ref_1") ei_add_failtest("ref_2") @@ -32,6 +37,20 @@ ei_add_failtest("ref_3") ei_add_failtest("ref_4") ei_add_failtest("ref_5") +ei_add_failtest("swap_1") +ei_add_failtest("swap_2") + +ei_add_failtest("ternary_1") +ei_add_failtest("ternary_2") + +ei_add_failtest("sparse_ref_1") +ei_add_failtest("sparse_ref_2") +ei_add_failtest("sparse_ref_3") +ei_add_failtest("sparse_ref_4") +ei_add_failtest("sparse_ref_5") + +ei_add_failtest("sparse_storage_mismatch") + ei_add_failtest("partialpivlu_int") ei_add_failtest("fullpivlu_int") ei_add_failtest("llt_int") @@ -40,15 +59,12 @@ ei_add_failtest("qr_int") ei_add_failtest("colpivqr_int") ei_add_failtest("fullpivqr_int") ei_add_failtest("jacobisvd_int") +ei_add_failtest("bdcsvd_int") ei_add_failtest("eigensolver_int") ei_add_failtest("eigensolver_cplx") -if (EIGEN_FAILTEST_FAILURE_COUNT) - message(FATAL_ERROR - "${EIGEN_FAILTEST_FAILURE_COUNT} out of ${EIGEN_FAILTEST_COUNT} failtests FAILED. " - "To debug these failures, manually compile these programs in ${CMAKE_CURRENT_SOURCE_DIR}, " - "with and without #define EIGEN_SHOULD_FAIL_TO_BUILD.") -else() - message(STATUS "Failtest SUCCESS: all ${EIGEN_FAILTEST_COUNT} failtests passed.") - message(STATUS "") +if(EIGEN_TEST_CXX11) + ei_add_failtest("initializer_list_1") + ei_add_failtest("initializer_list_2") endif() + diff --git a/thirdparty/eigen/failtest/bdcsvd_int.cpp b/thirdparty/eigen/failtest/bdcsvd_int.cpp new file mode 100644 index 00000000..670752cf --- /dev/null +++ b/thirdparty/eigen/failtest/bdcsvd_int.cpp @@ -0,0 +1,14 @@ +#include "../Eigen/SVD" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define SCALAR int +#else +#define SCALAR float +#endif + +using namespace Eigen; + +int main() +{ + BDCSVD > qr(Matrix::Random(10,10)); +} diff --git a/thirdparty/eigen/failtest/cwiseunaryview_nonconst_ctor_on_const_xpr.cpp b/thirdparty/eigen/failtest/cwiseunaryview_nonconst_ctor_on_const_xpr.cpp new file mode 100644 index 00000000..e23cf8fd --- /dev/null +++ b/thirdparty/eigen/failtest/cwiseunaryview_nonconst_ctor_on_const_xpr.cpp @@ -0,0 +1,15 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define CV_QUALIFIER const +#else +#define CV_QUALIFIER +#endif + +using namespace Eigen; + +void foo(CV_QUALIFIER Matrix3d &m){ + CwiseUnaryView,Matrix3d> t(m); +} + +int main() {} diff --git a/thirdparty/eigen/failtest/cwiseunaryview_on_const_type_actually_const.cpp b/thirdparty/eigen/failtest/cwiseunaryview_on_const_type_actually_const.cpp new file mode 100644 index 00000000..fcd41dfd --- /dev/null +++ b/thirdparty/eigen/failtest/cwiseunaryview_on_const_type_actually_const.cpp @@ -0,0 +1,16 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define CV_QUALIFIER const +#else +#define CV_QUALIFIER +#endif + +using namespace Eigen; + +void foo(){ + MatrixXf m; + CwiseUnaryView,CV_QUALIFIER MatrixXf>(m).coeffRef(0, 0) = 1.0f; +} + +int main() {} diff --git a/thirdparty/eigen/failtest/initializer_list_1.cpp b/thirdparty/eigen/failtest/initializer_list_1.cpp new file mode 100644 index 00000000..92dfd1f6 --- /dev/null +++ b/thirdparty/eigen/failtest/initializer_list_1.cpp @@ -0,0 +1,14 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define ROWS Dynamic +#else +#define ROWS 3 +#endif + +using namespace Eigen; + +int main() +{ + Matrix {1, 2, 3}; +} diff --git a/thirdparty/eigen/failtest/initializer_list_2.cpp b/thirdparty/eigen/failtest/initializer_list_2.cpp new file mode 100644 index 00000000..1996050a --- /dev/null +++ b/thirdparty/eigen/failtest/initializer_list_2.cpp @@ -0,0 +1,16 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define ROWS Dynamic +#define COLS Dynamic +#else +#define ROWS 3 +#define COLS 1 +#endif + +using namespace Eigen; + +int main() +{ + Matrix {1, 2, 3}; +} diff --git a/thirdparty/eigen/failtest/selfadjointview_nonconst_ctor_on_const_xpr.cpp b/thirdparty/eigen/failtest/selfadjointview_nonconst_ctor_on_const_xpr.cpp new file mode 100644 index 00000000..a240f818 --- /dev/null +++ b/thirdparty/eigen/failtest/selfadjointview_nonconst_ctor_on_const_xpr.cpp @@ -0,0 +1,15 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define CV_QUALIFIER const +#else +#define CV_QUALIFIER +#endif + +using namespace Eigen; + +void foo(CV_QUALIFIER Matrix3d &m){ + SelfAdjointView t(m); +} + +int main() {} diff --git a/thirdparty/eigen/failtest/selfadjointview_on_const_type_actually_const.cpp b/thirdparty/eigen/failtest/selfadjointview_on_const_type_actually_const.cpp new file mode 100644 index 00000000..19aaad6d --- /dev/null +++ b/thirdparty/eigen/failtest/selfadjointview_on_const_type_actually_const.cpp @@ -0,0 +1,16 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define CV_QUALIFIER const +#else +#define CV_QUALIFIER +#endif + +using namespace Eigen; + +void foo(){ + MatrixXf m; + SelfAdjointView(m).coeffRef(0, 0) = 1.0f; +} + +int main() {} diff --git a/thirdparty/eigen/failtest/sparse_ref_1.cpp b/thirdparty/eigen/failtest/sparse_ref_1.cpp new file mode 100644 index 00000000..d78d1f9b --- /dev/null +++ b/thirdparty/eigen/failtest/sparse_ref_1.cpp @@ -0,0 +1,18 @@ +#include "../Eigen/Sparse" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define CV_QUALIFIER const +#else +#define CV_QUALIFIER +#endif + +using namespace Eigen; + +void call_ref(Ref > a) { } + +int main() +{ + SparseMatrix a(10,10); + CV_QUALIFIER SparseMatrix& ac(a); + call_ref(ac); +} diff --git a/thirdparty/eigen/failtest/sparse_ref_2.cpp b/thirdparty/eigen/failtest/sparse_ref_2.cpp new file mode 100644 index 00000000..46c9440c --- /dev/null +++ b/thirdparty/eigen/failtest/sparse_ref_2.cpp @@ -0,0 +1,15 @@ +#include "../Eigen/Sparse" + +using namespace Eigen; + +void call_ref(Ref > a) { } + +int main() +{ + SparseMatrix A(10,10); +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD + call_ref(A.row(3)); +#else + call_ref(A.col(3)); +#endif +} diff --git a/thirdparty/eigen/failtest/sparse_ref_3.cpp b/thirdparty/eigen/failtest/sparse_ref_3.cpp new file mode 100644 index 00000000..a9949b55 --- /dev/null +++ b/thirdparty/eigen/failtest/sparse_ref_3.cpp @@ -0,0 +1,15 @@ +#include "../Eigen/Sparse" + +using namespace Eigen; + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +void call_ref(Ref > a) { } +#else +void call_ref(const Ref > &a) { } +#endif + +int main() +{ + SparseMatrix a(10,10); + call_ref(a+a); +} diff --git a/thirdparty/eigen/failtest/sparse_ref_4.cpp b/thirdparty/eigen/failtest/sparse_ref_4.cpp new file mode 100644 index 00000000..57bb6a1f --- /dev/null +++ b/thirdparty/eigen/failtest/sparse_ref_4.cpp @@ -0,0 +1,15 @@ +#include "../Eigen/Sparse" + +using namespace Eigen; + +void call_ref(Ref > a) {} + +int main() +{ + SparseMatrix A(10,10); +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD + call_ref(A.transpose()); +#else + call_ref(A); +#endif +} diff --git a/thirdparty/eigen/failtest/sparse_ref_5.cpp b/thirdparty/eigen/failtest/sparse_ref_5.cpp new file mode 100644 index 00000000..4478f6f2 --- /dev/null +++ b/thirdparty/eigen/failtest/sparse_ref_5.cpp @@ -0,0 +1,16 @@ +#include "../Eigen/Sparse" + +using namespace Eigen; + +void call_ref(Ref > a) { } + +int main() +{ + SparseMatrix a(10,10); + SparseMatrixBase > &ac(a); +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD + call_ref(ac); +#else + call_ref(ac.derived()); +#endif +} diff --git a/thirdparty/eigen/failtest/sparse_storage_mismatch.cpp b/thirdparty/eigen/failtest/sparse_storage_mismatch.cpp new file mode 100644 index 00000000..51840d41 --- /dev/null +++ b/thirdparty/eigen/failtest/sparse_storage_mismatch.cpp @@ -0,0 +1,16 @@ +#include "../Eigen/Sparse" +using namespace Eigen; + +typedef SparseMatrix Mat1; +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +typedef SparseMatrix Mat2; +#else +typedef SparseMatrix Mat2; +#endif + +int main() +{ + Mat1 a(10,10); + Mat2 b(10,10); + a += b; +} diff --git a/thirdparty/eigen/failtest/swap_1.cpp b/thirdparty/eigen/failtest/swap_1.cpp new file mode 100644 index 00000000..10637972 --- /dev/null +++ b/thirdparty/eigen/failtest/swap_1.cpp @@ -0,0 +1,14 @@ +#include "../Eigen/Core" + +using namespace Eigen; + +int main() +{ + VectorXf a(10), b(10); +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD + const DenseBase &ac(a); +#else + DenseBase &ac(a); +#endif + b.swap(ac); +} diff --git a/thirdparty/eigen/failtest/swap_2.cpp b/thirdparty/eigen/failtest/swap_2.cpp new file mode 100644 index 00000000..b386cf41 --- /dev/null +++ b/thirdparty/eigen/failtest/swap_2.cpp @@ -0,0 +1,14 @@ +#include "../Eigen/Core" + +using namespace Eigen; + +int main() +{ + VectorXf a(10), b(10); + VectorXf const &ac(a); +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD + b.swap(ac); +#else + b.swap(ac.const_cast_derived()); +#endif +} diff --git a/thirdparty/eigen/failtest/ternary_1.cpp b/thirdparty/eigen/failtest/ternary_1.cpp new file mode 100644 index 00000000..b40bcb0c --- /dev/null +++ b/thirdparty/eigen/failtest/ternary_1.cpp @@ -0,0 +1,13 @@ +#include "../Eigen/Core" + +using namespace Eigen; + +int main(int argc,char **) +{ + VectorXf a(10), b(10); +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD + b = argc>1 ? 2*a : -a; +#else + b = argc>1 ? 2*a : VectorXf(-a); +#endif +} diff --git a/thirdparty/eigen/failtest/ternary_2.cpp b/thirdparty/eigen/failtest/ternary_2.cpp new file mode 100644 index 00000000..a46b12b2 --- /dev/null +++ b/thirdparty/eigen/failtest/ternary_2.cpp @@ -0,0 +1,13 @@ +#include "../Eigen/Core" + +using namespace Eigen; + +int main(int argc,char **) +{ + VectorXf a(10), b(10); +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD + b = argc>1 ? 2*a : a+a; +#else + b = argc>1 ? VectorXf(2*a) : VectorXf(a+a); +#endif +} diff --git a/thirdparty/eigen/failtest/triangularview_nonconst_ctor_on_const_xpr.cpp b/thirdparty/eigen/failtest/triangularview_nonconst_ctor_on_const_xpr.cpp new file mode 100644 index 00000000..807447e4 --- /dev/null +++ b/thirdparty/eigen/failtest/triangularview_nonconst_ctor_on_const_xpr.cpp @@ -0,0 +1,15 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define CV_QUALIFIER const +#else +#define CV_QUALIFIER +#endif + +using namespace Eigen; + +void foo(CV_QUALIFIER Matrix3d &m){ + TriangularView t(m); +} + +int main() {} diff --git a/thirdparty/eigen/failtest/triangularview_on_const_type_actually_const.cpp b/thirdparty/eigen/failtest/triangularview_on_const_type_actually_const.cpp new file mode 100644 index 00000000..0a381a61 --- /dev/null +++ b/thirdparty/eigen/failtest/triangularview_on_const_type_actually_const.cpp @@ -0,0 +1,16 @@ +#include "../Eigen/Core" + +#ifdef EIGEN_SHOULD_FAIL_TO_BUILD +#define CV_QUALIFIER const +#else +#define CV_QUALIFIER +#endif + +using namespace Eigen; + +void foo(){ + MatrixXf m; + TriangularView(m).coeffRef(0, 0) = 1.0f; +} + +int main() {} diff --git a/thirdparty/eigen/lapack/CMakeLists.txt b/thirdparty/eigen/lapack/CMakeLists.txt index 9883d4c7..196835cd 100644 --- a/thirdparty/eigen/lapack/CMakeLists.txt +++ b/thirdparty/eigen/lapack/CMakeLists.txt @@ -1,15 +1,21 @@ - project(EigenLapack CXX) -include("../cmake/language_support.cmake") - -workaround_9220(Fortran EIGEN_Fortran_COMPILER_WORKS) - -if(EIGEN_Fortran_COMPILER_WORKS) - enable_language(Fortran OPTIONAL) - if(NOT CMAKE_Fortran_COMPILER) - set(EIGEN_Fortran_COMPILER_WORKS OFF) +if(EIGEN_BUILD_LAPACK AND EIGEN_BUILD_BLAS) + +include(CheckLanguage) +check_language(Fortran) +if(CMAKE_Fortran_COMPILER) + enable_language(Fortran) + if("${CMAKE_Fortran_COMPILER_ID}" STREQUAL "GNU") + if ("${CMAKE_Fortran_COMPILER_VERSION}" VERSION_GREATER_EQUAL 10.0) + # We use an old version of LAPACK with argument type mismatches. + # Allow them to compile anyway with newer GNU versions. + set(CMAKE_Fortran_FLAGS "${CMAKE_Fortran_FLAGS} -fallow-argument-mismatch") + endif() endif() + set(EIGEN_Fortran_COMPILER_WORKS ON) +else() + set(EIGEN_Fortran_COMPILER_WORKS OFF) endif() add_custom_target(lapack) @@ -35,7 +41,7 @@ set(EigenLapack_SRCS ${EigenLapack_SRCS} second_NONE.f dsecnd_NONE.f ) -option(EIGEN_ENABLE_LAPACK_TESTS OFF "Enbale the Lapack unit tests") +option(EIGEN_ENABLE_LAPACK_TESTS OFF "Enable the Lapack unit tests") if(EIGEN_ENABLE_LAPACK_TESTS) @@ -49,7 +55,7 @@ if(EIGEN_ENABLE_LAPACK_TESTS) INACTIVITY_TIMEOUT 15 TIMEOUT 240 STATUS download_status - EXPECTED_MD5 5758ce55afcf79da98de8b9de1615ad5 + EXPECTED_MD5 ab5742640617e3221a873aba44bbdc93 SHOW_PROGRESS) message(STATUS ${download_status}) @@ -59,7 +65,7 @@ if(EIGEN_ENABLE_LAPACK_TESTS) message(STATUS "Setup lapack reference and lapack unit tests") execute_process(COMMAND tar xzf "lapack_addons_3.4.1.tgz" WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}) else() - message(STATUS "Download of lapack_addons_3.4.1.tgz failed, LAPACK unit tests wont be enabled") + message(STATUS "Download of lapack_addons_3.4.1.tgz failed, LAPACK unit tests won't be enabled") set(EIGEN_ENABLE_LAPACK_TESTS false) endif() @@ -74,7 +80,7 @@ if(EIGEN_ENABLE_LAPACK_TESTS) sgetrf.f dgetrf.f cgetrf.f zgetrf.f sgetrs.f dgetrs.f cgetrs.f zgetrs.f) - FILE(GLOB ReferenceLapack_SRCS0 RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} "reference/*.f") + file(GLOB ReferenceLapack_SRCS0 RELATIVE ${CMAKE_CURRENT_SOURCE_DIR} "reference/*.f") foreach(filename1 IN LISTS ReferenceLapack_SRCS0) string(REPLACE "reference/" "" filename ${filename1}) list(FIND EigenLapack_SRCS ${filename} id1) @@ -86,29 +92,34 @@ if(EIGEN_ENABLE_LAPACK_TESTS) endif() -endif(EIGEN_ENABLE_LAPACK_TESTS) +endif() -endif(EIGEN_Fortran_COMPILER_WORKS) +endif() -add_library(eigen_lapack_static ${EigenLapack_SRCS} ${ReferenceLapack_SRCS}) -add_library(eigen_lapack SHARED ${EigenLapack_SRCS}) +set(EIGEN_LAPACK_TARGETS "") -target_link_libraries(eigen_lapack eigen_blas) +add_library(eigen_lapack_static ${EigenLapack_SRCS} ${ReferenceLapack_SRCS}) +list(APPEND EIGEN_LAPACK_TARGETS eigen_lapack_static) -if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) - target_link_libraries(eigen_lapack_static ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}) - target_link_libraries(eigen_lapack ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}) +if (EIGEN_BUILD_SHARED_LIBS) + add_library(eigen_lapack SHARED ${EigenLapack_SRCS}) + list(APPEND EIGEN_LAPACK_TARGETS eigen_lapack) + target_link_libraries(eigen_lapack eigen_blas) endif() -add_dependencies(lapack eigen_lapack eigen_lapack_static) +foreach(target IN LISTS EIGEN_LAPACK_TARGETS) + if(EIGEN_STANDARD_LIBRARIES_TO_LINK_TO) + target_link_libraries(${target} ${EIGEN_STANDARD_LIBRARIES_TO_LINK_TO}) + endif() + target_link_libraries(${target} Eigen3::Eigen) + add_dependencies(lapack ${target}) + install(TARGETS ${target} + RUNTIME DESTINATION bin + LIBRARY DESTINATION lib + ARCHIVE DESTINATION lib) +endforeach() -install(TARGETS eigen_lapack eigen_lapack_static - RUNTIME DESTINATION bin - LIBRARY DESTINATION lib - ARCHIVE DESTINATION lib) - - get_filename_component(eigen_full_path_to_testing_lapack "./testing/" ABSOLUTE) if(EXISTS ${eigen_full_path_to_testing_lapack}) @@ -133,24 +144,26 @@ if(EXISTS ${eigen_full_path_to_testing_lapack}) string(REGEX REPLACE "(.*)/STACK:(.*) (.*)" "\\1/STACK:900000000000000000 \\3" CMAKE_EXE_LINKER_FLAGS "${CMAKE_EXE_LINKER_FLAGS}") endif() + file(MAKE_DIRECTORY "${LAPACK_BINARY_DIR}/TESTING") add_subdirectory(testing/MATGEN) add_subdirectory(testing/LIN) add_subdirectory(testing/EIG) macro(add_lapack_test output input target) set(TEST_INPUT "${LAPACK_SOURCE_DIR}/testing/${input}") - set(TEST_OUTPUT "${LAPACK_BINARY_DIR}/testing/${output}") - get_target_property(TEST_LOC ${target} LOCATION) + set(TEST_OUTPUT "${LAPACK_BINARY_DIR}/TESTING/${output}") string(REPLACE "." "_" input_name ${input}) set(testName "${target}_${input_name}") if(EXISTS "${TEST_INPUT}") - add_test(LAPACK-${testName} "${CMAKE_COMMAND}" - -DTEST=${TEST_LOC} + add_dependencies(buildtests ${target}) + add_test(NAME LAPACK-${testName} + COMMAND "${CMAKE_COMMAND}" + -DTEST=$ -DINPUT=${TEST_INPUT} -DOUTPUT=${TEST_OUTPUT} -DINTDIR=${CMAKE_CFG_INTDIR} -P "${LAPACK_SOURCE_DIR}/testing/runtest.cmake") endif() - endmacro(add_lapack_test) + endmacro() if (BUILD_SINGLE) add_lapack_test(stest.out stest.in xlintsts) @@ -447,3 +460,6 @@ if(EXISTS ${eigen_full_path_to_testing_lapack}) endif() +elseif(EIGEN_BUILD_LAPACK AND NOT EIGEN_BUILD_BLAS) + message(FATAL_ERROR "EIGEN_BUILD_LAPACK requires EIGEN_BUILD_BLAS") +endif() #EIGEN_BUILD_LAPACK diff --git a/thirdparty/eigen/lapack/complex_double.cpp b/thirdparty/eigen/lapack/complex_double.cpp index 424d2b8c..c9c57527 100644 --- a/thirdparty/eigen/lapack/complex_double.cpp +++ b/thirdparty/eigen/lapack/complex_double.cpp @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2009-2011 Gael Guennebaud +// Copyright (C) 2009-2014 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -15,3 +15,4 @@ #include "cholesky.cpp" #include "lu.cpp" +#include "svd.cpp" diff --git a/thirdparty/eigen/lapack/complex_single.cpp b/thirdparty/eigen/lapack/complex_single.cpp index c0b2d29a..6d11b26c 100644 --- a/thirdparty/eigen/lapack/complex_single.cpp +++ b/thirdparty/eigen/lapack/complex_single.cpp @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2009-2011 Gael Guennebaud +// Copyright (C) 2009-2014 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -15,3 +15,4 @@ #include "cholesky.cpp" #include "lu.cpp" +#include "svd.cpp" diff --git a/thirdparty/eigen/lapack/double.cpp b/thirdparty/eigen/lapack/double.cpp index d86549e1..ea78bb66 100644 --- a/thirdparty/eigen/lapack/double.cpp +++ b/thirdparty/eigen/lapack/double.cpp @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2009-2011 Gael Guennebaud +// Copyright (C) 2009-2014 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -15,3 +15,4 @@ #include "cholesky.cpp" #include "lu.cpp" #include "eigenvalues.cpp" +#include "svd.cpp" diff --git a/thirdparty/eigen/lapack/eigenvalues.cpp b/thirdparty/eigen/lapack/eigenvalues.cpp index a1526ebc..921c5156 100644 --- a/thirdparty/eigen/lapack/eigenvalues.cpp +++ b/thirdparty/eigen/lapack/eigenvalues.cpp @@ -7,10 +7,10 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#include "common.h" +#include "lapack_common.h" #include -// computes an LU factorization of a general M-by-N matrix A using partial pivoting with row interchanges +// computes eigen values and vectors of a general N-by-N matrix A EIGEN_LAPACK_FUNC(syev,(char *jobz, char *uplo, int* n, Scalar* a, int *lda, Scalar* w, Scalar* /*work*/, int* lwork, int *info)) { // TODO exploit the work buffer @@ -22,24 +22,7 @@ EIGEN_LAPACK_FUNC(syev,(char *jobz, char *uplo, int* n, Scalar* a, int *lda, Sca else if(*n<0) *info = -3; else if(*lda +// Copyright (C) 2010-2014 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -11,6 +11,7 @@ #define EIGEN_LAPACK_COMMON_H #include "../blas/common.h" +#include "../Eigen/src/misc/lapack.h" #define EIGEN_LAPACK_FUNC(FUNC,ARGLIST) \ extern "C" { int EIGEN_BLAS_FUNC(FUNC) ARGLIST; } \ @@ -18,6 +19,11 @@ typedef Eigen::Map > PivotsType; +#if ISCOMPLEX +#define EIGEN_LAPACK_ARG_IF_COMPLEX(X) X, +#else +#define EIGEN_LAPACK_ARG_IF_COMPLEX(X) +#endif #endif // EIGEN_LAPACK_COMMON_H diff --git a/thirdparty/eigen/lapack/single.cpp b/thirdparty/eigen/lapack/single.cpp index a64ed44e..c7da3eff 100644 --- a/thirdparty/eigen/lapack/single.cpp +++ b/thirdparty/eigen/lapack/single.cpp @@ -1,7 +1,7 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2009-2011 Gael Guennebaud +// Copyright (C) 2009-2014 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -15,3 +15,4 @@ #include "cholesky.cpp" #include "lu.cpp" #include "eigenvalues.cpp" +#include "svd.cpp" diff --git a/thirdparty/eigen/lapack/svd.cpp b/thirdparty/eigen/lapack/svd.cpp new file mode 100644 index 00000000..77b302b6 --- /dev/null +++ b/thirdparty/eigen/lapack/svd.cpp @@ -0,0 +1,138 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2014 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "lapack_common.h" +#include + +// computes the singular values/vectors a general M-by-N matrix A using divide-and-conquer +EIGEN_LAPACK_FUNC(gesdd,(char *jobz, int *m, int* n, Scalar* a, int *lda, RealScalar *s, Scalar *u, int *ldu, Scalar *vt, int *ldvt, Scalar* /*work*/, int* lwork, + EIGEN_LAPACK_ARG_IF_COMPLEX(RealScalar */*rwork*/) int * /*iwork*/, int *info)) +{ + // TODO exploit the work buffer + bool query_size = *lwork==-1; + int diag_size = (std::min)(*m,*n); + + *info = 0; + if(*jobz!='A' && *jobz!='S' && *jobz!='O' && *jobz!='N') *info = -1; + else if(*m<0) *info = -2; + else if(*n<0) *info = -3; + else if(*lda=*n && *ldvt<*n)) *info = -10; + + if(*info!=0) + { + int e = -*info; + return xerbla_(SCALAR_SUFFIX_UP"GESDD ", &e, 6); + } + + if(query_size) + { + *lwork = 0; + return 0; + } + + if(*n==0 || *m==0) + return 0; + + PlainMatrixType mat(*m,*n); + mat = matrix(a,*m,*n,*lda); + + int option = *jobz=='A' ? ComputeFullU|ComputeFullV + : *jobz=='S' ? ComputeThinU|ComputeThinV + : *jobz=='O' ? ComputeThinU|ComputeThinV + : 0; + + BDCSVD svd(mat,option); + + make_vector(s,diag_size) = svd.singularValues().head(diag_size); + + if(*jobz=='A') + { + matrix(u,*m,*m,*ldu) = svd.matrixU(); + matrix(vt,*n,*n,*ldvt) = svd.matrixV().adjoint(); + } + else if(*jobz=='S') + { + matrix(u,*m,diag_size,*ldu) = svd.matrixU(); + matrix(vt,diag_size,*n,*ldvt) = svd.matrixV().adjoint(); + } + else if(*jobz=='O' && *m>=*n) + { + matrix(a,*m,*n,*lda) = svd.matrixU(); + matrix(vt,*n,*n,*ldvt) = svd.matrixV().adjoint(); + } + else if(*jobz=='O') + { + matrix(u,*m,*m,*ldu) = svd.matrixU(); + matrix(a,diag_size,*n,*lda) = svd.matrixV().adjoint(); + } + + return 0; +} + +// computes the singular values/vectors a general M-by-N matrix A using two sided jacobi algorithm +EIGEN_LAPACK_FUNC(gesvd,(char *jobu, char *jobv, int *m, int* n, Scalar* a, int *lda, RealScalar *s, Scalar *u, int *ldu, Scalar *vt, int *ldvt, Scalar* /*work*/, int* lwork, + EIGEN_LAPACK_ARG_IF_COMPLEX(RealScalar */*rwork*/) int *info)) +{ + // TODO exploit the work buffer + bool query_size = *lwork==-1; + int diag_size = (std::min)(*m,*n); + + *info = 0; + if( *jobu!='A' && *jobu!='S' && *jobu!='O' && *jobu!='N') *info = -1; + else if((*jobv!='A' && *jobv!='S' && *jobv!='O' && *jobv!='N') + || (*jobu=='O' && *jobv=='O')) *info = -2; + else if(*m<0) *info = -3; + else if(*n<0) *info = -4; + else if(*lda svd(mat,option); + + make_vector(s,diag_size) = svd.singularValues().head(diag_size); + { + if(*jobu=='A') matrix(u,*m,*m,*ldu) = svd.matrixU(); + else if(*jobu=='S') matrix(u,*m,diag_size,*ldu) = svd.matrixU(); + else if(*jobu=='O') matrix(a,*m,diag_size,*lda) = svd.matrixU(); + } + { + if(*jobv=='A') matrix(vt,*n,*n,*ldvt) = svd.matrixV().adjoint(); + else if(*jobv=='S') matrix(vt,diag_size,*n,*ldvt) = svd.matrixV().adjoint(); + else if(*jobv=='O') matrix(a,diag_size,*n,*lda) = svd.matrixV().adjoint(); + } + return 0; +} diff --git a/thirdparty/eigen/scripts/buildtests.in b/thirdparty/eigen/scripts/buildtests.in index 7026373c..ab9c18fb 100755 --- a/thirdparty/eigen/scripts/buildtests.in +++ b/thirdparty/eigen/scripts/buildtests.in @@ -2,7 +2,7 @@ if [[ $# != 1 || $1 == *help ]] then - echo "usage: ./check regexp" + echo "usage: $0 regexp" echo " Builds tests matching the regexp." echo " The EIGEN_MAKE_ARGS environment variable allows to pass args to 'make'." echo " For example, to launch 5 concurrent builds, use EIGEN_MAKE_ARGS='-j5'" @@ -10,13 +10,13 @@ then fi TESTSLIST="@EIGEN_TESTS_LIST@" -targets_to_make=`echo "$TESTSLIST" | egrep "$1" | xargs echo` +targets_to_make=$(echo "$TESTSLIST" | grep -E "$1" | xargs echo) if [ -n "${EIGEN_MAKE_ARGS:+x}" ] then - make $targets_to_make ${EIGEN_MAKE_ARGS} + @CMAKE_MAKE_PROGRAM@ $targets_to_make ${EIGEN_MAKE_ARGS} else - make $targets_to_make + @CMAKE_MAKE_PROGRAM@ $targets_to_make @EIGEN_TEST_BUILD_FLAGS@ fi exit $? diff --git a/thirdparty/eigen/scripts/cdashtesting.cmake.in b/thirdparty/eigen/scripts/cdashtesting.cmake.in index 59cf5332..0bf0fac2 100644 --- a/thirdparty/eigen/scripts/cdashtesting.cmake.in +++ b/thirdparty/eigen/scripts/cdashtesting.cmake.in @@ -12,8 +12,8 @@ elseif(${CTEST_SCRIPT_ARG} MATCHES Continuous) set(MODEL Continuous) endif() -find_program(CTEST_HG_COMMAND NAMES hg) -set(CTEST_UPDATE_COMMAND "${CTEST_HG_COMMAND}") +find_program(CTEST_GIT_COMMAND NAMES git) +set(CTEST_UPDATE_COMMAND "${CTEST_GIT_COMMAND}") ctest_start(${MODEL} ${CTEST_SOURCE_DIRECTORY} ${CTEST_BINARY_DIRECTORY}) diff --git a/thirdparty/eigen/scripts/check.in b/thirdparty/eigen/scripts/check.in index a90061a5..7717e2d9 100755 --- a/thirdparty/eigen/scripts/check.in +++ b/thirdparty/eigen/scripts/check.in @@ -3,7 +3,7 @@ if [[ $# != 1 || $1 == *help ]] then - echo "usage: ./check regexp" + echo "usage: $0 regexp" echo " Builds and runs tests matching the regexp." echo " The EIGEN_MAKE_ARGS environment variable allows to pass args to 'make'." echo " For example, to launch 5 concurrent builds, use EIGEN_MAKE_ARGS='-j5'" diff --git a/thirdparty/eigen/scripts/eigen_gen_docs b/thirdparty/eigen/scripts/eigen_gen_docs index 0e6f9ada..787dcb32 100644 --- a/thirdparty/eigen/scripts/eigen_gen_docs +++ b/thirdparty/eigen/scripts/eigen_gen_docs @@ -4,7 +4,7 @@ # You should call this script with USER set as you want, else some default # will be used USER=${USER:-'orzel'} -UPLOAD_DIR=dox +UPLOAD_DIR=dox-devel #ulimit -v 1024000 diff --git a/thirdparty/eigen/scripts/eigen_gen_split_test_help.cmake b/thirdparty/eigen/scripts/eigen_gen_split_test_help.cmake new file mode 100644 index 00000000..e43f5aab --- /dev/null +++ b/thirdparty/eigen/scripts/eigen_gen_split_test_help.cmake @@ -0,0 +1,11 @@ +#!cmake -P +file(WRITE split_test_helper.h "") +foreach(i RANGE 1 999) + file(APPEND split_test_helper.h + "#if defined(EIGEN_TEST_PART_${i}) || defined(EIGEN_TEST_PART_ALL)\n" + "#define CALL_SUBTEST_${i}(FUNC) CALL_SUBTEST(FUNC)\n" + "#else\n" + "#define CALL_SUBTEST_${i}(FUNC)\n" + "#endif\n\n" + ) +endforeach() \ No newline at end of file diff --git a/thirdparty/eigen/scripts/eigen_monitor_perf.sh b/thirdparty/eigen/scripts/eigen_monitor_perf.sh new file mode 100755 index 00000000..8f3425da --- /dev/null +++ b/thirdparty/eigen/scripts/eigen_monitor_perf.sh @@ -0,0 +1,25 @@ +#!/bin/bash + +# This is a script example to automatically update and upload performance unit tests. +# The following five variables must be adjusted to match your settings. + +USER='ggael' +UPLOAD_DIR=perf_monitoring/ggaelmacbook26 +EIGEN_SOURCE_PATH=$HOME/Eigen/eigen +export PREFIX="haswell-fma" +export CXX_FLAGS="-mfma -w" + +#### + +BENCH_PATH=$EIGEN_SOURCE_PATH/bench/perf_monitoring/$PREFIX +PREVPATH=$(pwd) +cd $EIGEN_SOURCE_PATH/bench/perf_monitoring && ./runall.sh "Haswell 2.6GHz, FMA, Apple's clang" "$@" +cd $PREVPATH || exit 1 + +ALLFILES="$BENCH_PATH/*.png $BENCH_PATH/*.html $BENCH_PATH/index.html $BENCH_PATH/s1.js $BENCH_PATH/s2.js" + +# (the '/' at the end of path is very important, see rsync documentation) +rsync -az --no-p --delete $ALLFILES $USER@ssh.tuxfamily.org:eigen/eigen.tuxfamily.org-web/htdocs/$UPLOAD_DIR/ || { echo "upload failed"; exit 1; } + +# fix the perm +ssh $USER@ssh.tuxfamily.org "chmod -R g+w /home/eigen/eigen.tuxfamily.org-web/htdocs/perf_monitoring" || { echo "perm failed"; exit 1; } diff --git a/thirdparty/eigen/test/AnnoyingScalar.h b/thirdparty/eigen/test/AnnoyingScalar.h new file mode 100644 index 00000000..b6218872 --- /dev/null +++ b/thirdparty/eigen/test/AnnoyingScalar.h @@ -0,0 +1,165 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2011-2018 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_TEST_ANNOYING_SCALAR_H +#define EIGEN_TEST_ANNOYING_SCALAR_H + +#include + +#if EIGEN_COMP_GNUC +#pragma GCC diagnostic ignored "-Wshadow" +#endif + +#ifndef EIGEN_TEST_ANNOYING_SCALAR_DONT_THROW +struct my_exception +{ + my_exception() {} + ~my_exception() {} +}; +#endif + +// An AnnoyingScalar is a pseudo scalar type that: +// - can randomly through an exception in operator + +// - randomly allocate on the heap or initialize a reference to itself making it non trivially copyable, nor movable, nor relocatable. + +class AnnoyingScalar +{ + public: + AnnoyingScalar() { init(); *v = 0; } + AnnoyingScalar(long double _v) { init(); *v = _v; } + AnnoyingScalar(double _v) { init(); *v = _v; } + AnnoyingScalar(float _v) { init(); *v = _v; } + AnnoyingScalar(int _v) { init(); *v = _v; } + AnnoyingScalar(long _v) { init(); *v = _v; } + #if EIGEN_HAS_CXX11 + AnnoyingScalar(long long _v) { init(); *v = _v; } + #endif + AnnoyingScalar(const AnnoyingScalar& other) { init(); *v = *(other.v); } + ~AnnoyingScalar() { + if(v!=&data) + delete v; + instances--; + } + + void init() { + if(internal::random()) + v = new float; + else + v = &data; + instances++; + } + + AnnoyingScalar operator+(const AnnoyingScalar& other) const + { + #ifndef EIGEN_TEST_ANNOYING_SCALAR_DONT_THROW + countdown--; + if(countdown<=0 && !dont_throw) + throw my_exception(); + #endif + return AnnoyingScalar(*v+*other.v); + } + + AnnoyingScalar operator-() const + { return AnnoyingScalar(-*v); } + + AnnoyingScalar operator-(const AnnoyingScalar& other) const + { return AnnoyingScalar(*v-*other.v); } + + AnnoyingScalar operator*(const AnnoyingScalar& other) const + { return AnnoyingScalar((*v)*(*other.v)); } + + AnnoyingScalar operator/(const AnnoyingScalar& other) const + { return AnnoyingScalar((*v)/(*other.v)); } + + AnnoyingScalar& operator+=(const AnnoyingScalar& other) { *v += *other.v; return *this; } + AnnoyingScalar& operator-=(const AnnoyingScalar& other) { *v -= *other.v; return *this; } + AnnoyingScalar& operator*=(const AnnoyingScalar& other) { *v *= *other.v; return *this; } + AnnoyingScalar& operator/=(const AnnoyingScalar& other) { *v /= *other.v; return *this; } + AnnoyingScalar& operator= (const AnnoyingScalar& other) { *v = *other.v; return *this; } + + bool operator==(const AnnoyingScalar& other) const { return *v == *other.v; } + bool operator!=(const AnnoyingScalar& other) const { return *v != *other.v; } + bool operator<=(const AnnoyingScalar& other) const { return *v <= *other.v; } + bool operator< (const AnnoyingScalar& other) const { return *v < *other.v; } + bool operator>=(const AnnoyingScalar& other) const { return *v >= *other.v; } + bool operator> (const AnnoyingScalar& other) const { return *v > *other.v; } + + float* v; + float data; + static int instances; +#ifndef EIGEN_TEST_ANNOYING_SCALAR_DONT_THROW + static int countdown; + static bool dont_throw; +#endif +}; + +AnnoyingScalar real(const AnnoyingScalar &x) { return x; } +AnnoyingScalar imag(const AnnoyingScalar & ) { return 0; } +AnnoyingScalar conj(const AnnoyingScalar &x) { return x; } +AnnoyingScalar sqrt(const AnnoyingScalar &x) { return std::sqrt(*x.v); } +AnnoyingScalar abs (const AnnoyingScalar &x) { return std::abs(*x.v); } +AnnoyingScalar cos (const AnnoyingScalar &x) { return std::cos(*x.v); } +AnnoyingScalar sin (const AnnoyingScalar &x) { return std::sin(*x.v); } +AnnoyingScalar acos(const AnnoyingScalar &x) { return std::acos(*x.v); } +AnnoyingScalar atan2(const AnnoyingScalar &y,const AnnoyingScalar &x) { return std::atan2(*y.v,*x.v); } + +std::ostream& operator<<(std::ostream& stream,const AnnoyingScalar& x) { + stream << (*(x.v)); + return stream; +} + +int AnnoyingScalar::instances = 0; + +#ifndef EIGEN_TEST_ANNOYING_SCALAR_DONT_THROW +int AnnoyingScalar::countdown = 0; +bool AnnoyingScalar::dont_throw = false; +#endif + +namespace Eigen { +template<> +struct NumTraits : NumTraits +{ + enum { + RequireInitialization = 1 + }; + typedef AnnoyingScalar Real; + typedef AnnoyingScalar Nested; + typedef AnnoyingScalar Literal; + typedef AnnoyingScalar NonInteger; +}; + +template<> inline AnnoyingScalar test_precision() { return test_precision(); } + +namespace numext { +template<> +EIGEN_DEVICE_FUNC EIGEN_ALWAYS_INLINE +bool (isfinite)(const AnnoyingScalar& x) { + return (numext::isfinite)(*x.v); +} +} + +namespace internal { + template<> EIGEN_STRONG_INLINE double cast(const AnnoyingScalar& x) { return double(*x.v); } + template<> EIGEN_STRONG_INLINE float cast(const AnnoyingScalar& x) { return *x.v; } +} +} // namespace Eigen + +AnnoyingScalar get_test_precision(const AnnoyingScalar&) +{ return Eigen::test_precision(); } + +AnnoyingScalar test_relative_error(const AnnoyingScalar &a, const AnnoyingScalar &b) +{ return test_relative_error(*a.v, *b.v); } + +inline bool test_isApprox(const AnnoyingScalar &a, const AnnoyingScalar &b) +{ return internal::isApprox(*a.v, *b.v, test_precision()); } + +inline bool test_isMuchSmallerThan(const AnnoyingScalar &a, const AnnoyingScalar &b) +{ return test_isMuchSmallerThan(*a.v, *b.v); } + +#endif // EIGEN_TEST_ANNOYING_SCALAR_H diff --git a/thirdparty/eigen/test/CMakeLists.txt b/thirdparty/eigen/test/CMakeLists.txt index 40c8f669..7ce260e0 100644 --- a/thirdparty/eigen/test/CMakeLists.txt +++ b/thirdparty/eigen/test/CMakeLists.txt @@ -1,86 +1,123 @@ +# The file split_test_helper.h was generated at first run, +# it is now included in test/ +if(EXISTS ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h) + file(REMOVE ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h) +endif() -# generate split test header file -message(STATUS ${CMAKE_CURRENT_BINARY_DIR}) -file(WRITE ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h "") -foreach(i RANGE 1 999) - file(APPEND ${CMAKE_CURRENT_BINARY_DIR}/split_test_helper.h - "#ifdef EIGEN_TEST_PART_${i}\n" - "#define CALL_SUBTEST_${i}(FUNC) CALL_SUBTEST(FUNC)\n" - "#else\n" - "#define CALL_SUBTEST_${i}(FUNC)\n" - "#endif\n\n" - ) -endforeach() +# check if we have a Fortran compiler +include(CheckLanguage) +check_language(Fortran) +if(CMAKE_Fortran_COMPILER) + enable_language(Fortran) + set(EIGEN_Fortran_COMPILER_WORKS ON) +else() + set(EIGEN_Fortran_COMPILER_WORKS OFF) + # search for a default Lapack library to complete Eigen's one + find_package(LAPACK QUIET) +endif() + +# TODO do the same for EXTERNAL_LAPACK +option(EIGEN_TEST_EXTERNAL_BLAS "Use external BLAS library for testsuite" OFF) +if(EIGEN_TEST_EXTERNAL_BLAS) + find_package(BLAS REQUIRED) + message(STATUS "BLAS_COMPILER_FLAGS: ${BLAS_COMPILER_FLAGS}") + add_definitions("-DEIGEN_USE_BLAS") # is adding ${BLAS_COMPILER_FLAGS} necessary? + list(APPEND EXTERNAL_LIBS "${BLAS_LIBRARIES}") +endif() # configure blas/lapack (use Eigen's ones) -set(BLAS_FOUND TRUE) -set(LAPACK_FOUND TRUE) -set(BLAS_LIBRARIES eigen_blas) -set(LAPACK_LIBRARIES eigen_lapack) +set(EIGEN_BLAS_LIBRARIES eigen_blas) +set(EIGEN_LAPACK_LIBRARIES eigen_lapack) set(EIGEN_TEST_MATRIX_DIR "" CACHE STRING "Enable testing of realword sparse matrices contained in the specified path") if(EIGEN_TEST_MATRIX_DIR) if(NOT WIN32) message(STATUS "Test realworld sparse matrices: ${EIGEN_TEST_MATRIX_DIR}") add_definitions( -DTEST_REAL_CASES="${EIGEN_TEST_MATRIX_DIR}" ) - else(NOT WIN32) + else() message(STATUS "REAL CASES CAN NOT BE CURRENTLY TESTED ON WIN32") - endif(NOT WIN32) -endif(EIGEN_TEST_MATRIX_DIR) + endif() +endif() set(SPARSE_LIBS " ") -find_package(Cholmod) -if(CHOLMOD_FOUND AND BLAS_FOUND AND LAPACK_FOUND) +find_package(CHOLMOD) +if(CHOLMOD_FOUND AND EIGEN_BUILD_BLAS AND EIGEN_BUILD_LAPACK) add_definitions("-DEIGEN_CHOLMOD_SUPPORT") include_directories(${CHOLMOD_INCLUDES}) - set(SPARSE_LIBS ${SPARSE_LIBS} ${CHOLMOD_LIBRARIES} ${BLAS_LIBRARIES} ${LAPACK_LIBRARIES}) - set(CHOLMOD_ALL_LIBS ${CHOLMOD_LIBRARIES} ${BLAS_LIBRARIES} ${LAPACK_LIBRARIES}) - ei_add_property(EIGEN_TESTED_BACKENDS "Cholmod, ") + set(SPARSE_LIBS ${SPARSE_LIBS} ${CHOLMOD_LIBRARIES} ${EIGEN_BLAS_LIBRARIES} ${EIGEN_LAPACK_LIBRARIES}) + set(CHOLMOD_ALL_LIBS ${CHOLMOD_LIBRARIES} ${EIGEN_BLAS_LIBRARIES} ${EIGEN_LAPACK_LIBRARIES}) + ei_add_property(EIGEN_TESTED_BACKENDS "CHOLMOD, ") + + ei_add_test(cholmod_support "" "${CHOLMOD_ALL_LIBS}") else() - ei_add_property(EIGEN_MISSING_BACKENDS "Cholmod, ") + ei_add_property(EIGEN_MISSING_BACKENDS "CHOLMOD, ") endif() -find_package(Umfpack) -if(UMFPACK_FOUND AND BLAS_FOUND) +find_package(UMFPACK) +if(UMFPACK_FOUND AND EIGEN_BUILD_BLAS) add_definitions("-DEIGEN_UMFPACK_SUPPORT") include_directories(${UMFPACK_INCLUDES}) - set(SPARSE_LIBS ${SPARSE_LIBS} ${UMFPACK_LIBRARIES} ${BLAS_LIBRARIES}) - set(UMFPACK_ALL_LIBS ${UMFPACK_LIBRARIES} ${BLAS_LIBRARIES}) - ei_add_property(EIGEN_TESTED_BACKENDS "UmfPack, ") + set(SPARSE_LIBS ${SPARSE_LIBS} ${UMFPACK_LIBRARIES} ${EIGEN_BLAS_LIBRARIES}) + set(UMFPACK_ALL_LIBS ${UMFPACK_LIBRARIES} ${EIGEN_BLAS_LIBRARIES}) + ei_add_property(EIGEN_TESTED_BACKENDS "UMFPACK, ") + + ei_add_test(umfpack_support "" "${UMFPACK_ALL_LIBS}") else() - ei_add_property(EIGEN_MISSING_BACKENDS "UmfPack, ") + ei_add_property(EIGEN_MISSING_BACKENDS "UMFPACK, ") endif() -find_package(SuperLU) -if(SUPERLU_FOUND AND BLAS_FOUND) +find_package(KLU) +if(KLU_FOUND AND EIGEN_BUILD_BLAS) + add_definitions("-DEIGEN_KLU_SUPPORT") + include_directories(${KLU_INCLUDES}) + set(SPARSE_LIBS ${SPARSE_LIBS} ${KLU_LIBRARIES} ${EIGEN_BLAS_LIBRARIES}) + set(KLU_ALL_LIBS ${KLU_LIBRARIES} ${EIGEN_BLAS_LIBRARIES}) + ei_add_property(EIGEN_TESTED_BACKENDS "KLU, ") + + ei_add_test(klu_support "" "${KLU_ALL_LIBS}") +else() + ei_add_property(EIGEN_MISSING_BACKENDS "KLU, ") +endif() + +find_package(SuperLU 4.0) +if(SuperLU_FOUND AND EIGEN_BUILD_BLAS) add_definitions("-DEIGEN_SUPERLU_SUPPORT") include_directories(${SUPERLU_INCLUDES}) - set(SPARSE_LIBS ${SPARSE_LIBS} ${SUPERLU_LIBRARIES} ${BLAS_LIBRARIES}) - set(SUPERLU_ALL_LIBS ${SUPERLU_LIBRARIES} ${BLAS_LIBRARIES}) + set(SPARSE_LIBS ${SPARSE_LIBS} ${SUPERLU_LIBRARIES} ${EIGEN_BLAS_LIBRARIES}) + set(SUPERLU_ALL_LIBS ${SUPERLU_LIBRARIES} ${EIGEN_BLAS_LIBRARIES}) ei_add_property(EIGEN_TESTED_BACKENDS "SuperLU, ") + + ei_add_test(superlu_support "" "${SUPERLU_ALL_LIBS}") else() ei_add_property(EIGEN_MISSING_BACKENDS "SuperLU, ") endif() -find_package(Pastix) -find_package(Scotch) -find_package(Metis 5.0 REQUIRED) -if(PASTIX_FOUND AND BLAS_FOUND) +find_package(PASTIX QUIET COMPONENTS METIS SEQ) +# check that the PASTIX found is a version without MPI +find_path(PASTIX_pastix_nompi.h_INCLUDE_DIRS + NAMES pastix_nompi.h + HINTS ${PASTIX_INCLUDE_DIRS} +) +if (NOT PASTIX_pastix_nompi.h_INCLUDE_DIRS) + message(STATUS "A version of Pastix has been found but pastix_nompi.h does not exist in the include directory." + " Because Eigen tests require a version without MPI, we disable the Pastix backend.") +endif() +if(PASTIX_FOUND AND PASTIX_pastix_nompi.h_INCLUDE_DIRS) add_definitions("-DEIGEN_PASTIX_SUPPORT") - include_directories(${PASTIX_INCLUDES}) + include_directories(${PASTIX_INCLUDE_DIRS_DEP}) if(SCOTCH_FOUND) - include_directories(${SCOTCH_INCLUDES}) + include_directories(${SCOTCH_INCLUDE_DIRS}) set(PASTIX_LIBRARIES ${PASTIX_LIBRARIES} ${SCOTCH_LIBRARIES}) elseif(METIS_FOUND) - include_directories(${METIS_INCLUDES}) + include_directories(${METIS_INCLUDE_DIRS}) set(PASTIX_LIBRARIES ${PASTIX_LIBRARIES} ${METIS_LIBRARIES}) - else(SCOTCH_FOUND) + else() ei_add_property(EIGEN_MISSING_BACKENDS "PaStiX, ") - endif(SCOTCH_FOUND) - set(SPARSE_LIBS ${SPARSE_LIBS} ${PASTIX_LIBRARIES} ${ORDERING_LIBRARIES} ${BLAS_LIBRARIES}) - set(PASTIX_ALL_LIBS ${PASTIX_LIBRARIES} ${BLAS_LIBRARIES}) + endif() + set(SPARSE_LIBS ${SPARSE_LIBS} ${PASTIX_LIBRARIES_DEP} ${ORDERING_LIBRARIES}) + set(PASTIX_ALL_LIBS ${PASTIX_LIBRARIES_DEP}) ei_add_property(EIGEN_TESTED_BACKENDS "PaStiX, ") else() ei_add_property(EIGEN_MISSING_BACKENDS "PaStiX, ") @@ -88,23 +125,21 @@ endif() if(METIS_FOUND) add_definitions("-DEIGEN_METIS_SUPPORT") - include_directories(${METIS_INCLUDES}) + include_directories(${METIS_INCLUDE_DIRS}) ei_add_property(EIGEN_TESTED_BACKENDS "METIS, ") else() ei_add_property(EIGEN_MISSING_BACKENDS "METIS, ") endif() find_package(SPQR) -if(SPQR_FOUND AND BLAS_FOUND AND LAPACK_FOUND) - if(CHOLMOD_FOUND) - add_definitions("-DEIGEN_SPQR_SUPPORT") - include_directories(${SPQR_INCLUDES}) - set(SPQR_ALL_LIBS ${SPQR_LIBRARIES} ${CHOLMOD_LIBRARIES} ${LAPACK_LIBRARIES} ${BLAS_LIBRARIES}) - set(SPARSE_LIBS ${SPARSE_LIBS} ${SPQR_ALL_LIBS}) - ei_add_property(EIGEN_TESTED_BACKENDS "SPQR, ") - else(CHOLMOD_FOUND) - ei_add_property(EIGEN_MISSING_BACKENDS "SPQR, ") - endif(CHOLMOD_FOUND) +if(SPQR_FOUND AND EIGEN_TEST_CXX11 AND CHOLMOD_FOUND AND EIGEN_BUILD_BLAS AND EIGEN_BUILD_LAPACK AND (EIGEN_Fortran_COMPILER_WORKS OR LAPACK_FOUND) ) + add_definitions("-DEIGEN_SPQR_SUPPORT") + include_directories(${SPQR_INCLUDES}) + set(SPQR_ALL_LIBS ${SPQR_LIBRARIES} ${CHOLMOD_LIBRARIES} ${EIGEN_LAPACK_LIBRARIES} ${EIGEN_BLAS_LIBRARIES} ${LAPACK_LIBRARIES}) + set(SPARSE_LIBS ${SPARSE_LIBS} ${SPQR_ALL_LIBS}) + ei_add_property(EIGEN_TESTED_BACKENDS "SPQR, ") +else() + ei_add_property(EIGEN_MISSING_BACKENDS "SPQR, ") endif() option(EIGEN_TEST_NOQT "Disable Qt support in unit tests" OFF) @@ -116,35 +151,47 @@ if(NOT EIGEN_TEST_NOQT) else() ei_add_property(EIGEN_MISSING_BACKENDS "Qt4 support, ") endif() -endif(NOT EIGEN_TEST_NOQT) +endif() if(TEST_LIB) add_definitions("-DEIGEN_EXTERN_INSTANTIATIONS=1") -endif(TEST_LIB) +endif() set_property(GLOBAL PROPERTY EIGEN_CURRENT_SUBPROJECT "Official") add_custom_target(BuildOfficial) +ei_add_test(clz) ei_add_test(rand) ei_add_test(meta) +ei_add_test(numext) ei_add_test(sizeof) ei_add_test(dynalloc) ei_add_test(nomalloc) ei_add_test(first_aligned) +ei_add_test(type_alias) +ei_add_test(nullary) ei_add_test(mixingtypes) -ei_add_test(packetmath) -ei_add_test(unalignedassert) +ei_add_test(io) +ei_add_test(packetmath "-DEIGEN_FAST_MATH=1") ei_add_test(vectorization_logic) ei_add_test(basicstuff) +ei_add_test(constructor) ei_add_test(linearstructure) ei_add_test(integer_types) -ei_add_test(cwiseop) ei_add_test(unalignedcount) -ei_add_test(exceptions) +if(NOT EIGEN_TEST_NO_EXCEPTIONS AND NOT EIGEN_TEST_OPENMP) + ei_add_test(exceptions) +endif() ei_add_test(redux) ei_add_test(visitor) ei_add_test(block) ei_add_test(corners) +ei_add_test(symbolic_index) +ei_add_test(indexed_view) +ei_add_test(reshape) +ei_add_test(swap) +ei_add_test(resize) +ei_add_test(conservative_resize) ei_add_test(product_small) ei_add_test(product_large) ei_add_test(product_extra) @@ -157,11 +204,12 @@ ei_add_test(smallvectors) ei_add_test(mapped_matrix) ei_add_test(mapstride) ei_add_test(mapstaticmethods) -ei_add_test(array) +ei_add_test(array_cwise) ei_add_test(array_for_matrix) ei_add_test(array_replicate) ei_add_test(array_reverse) ei_add_test(ref) +ei_add_test(is_same_dense) ei_add_test(triangular) ei_add_test(selfadjoint) ei_add_test(product_selfadjoint) @@ -173,6 +221,7 @@ ei_add_test(product_trsolve) ei_add_test(product_mmtr) ei_add_test(product_notemporary) ei_add_test(stable_norm) +ei_add_test(permutationmatrices) ei_add_test(bandmatrix) ei_add_test(cholesky) ei_add_test(lu) @@ -192,66 +241,81 @@ ei_add_test(real_qz) ei_add_test(eigensolver_generalized_real) ei_add_test(jacobi) ei_add_test(jacobisvd) +ei_add_test(bdcsvd) +ei_add_test(householder) ei_add_test(geo_orthomethods) -ei_add_test(geo_homogeneous) ei_add_test(geo_quaternion) -ei_add_test(geo_transformations) ei_add_test(geo_eulerangles) -ei_add_test(geo_hyperplane) ei_add_test(geo_parametrizedline) ei_add_test(geo_alignedbox) +ei_add_test(geo_hyperplane) +ei_add_test(geo_transformations) +ei_add_test(geo_homogeneous) ei_add_test(stdvector) ei_add_test(stdvector_overload) ei_add_test(stdlist) ei_add_test(stdlist_overload) ei_add_test(stddeque) ei_add_test(stddeque_overload) -ei_add_test(resize) -ei_add_test(sparse_vector) ei_add_test(sparse_basic) +ei_add_test(sparse_block) +ei_add_test(sparse_vector) ei_add_test(sparse_product) +ei_add_test(sparse_ref) ei_add_test(sparse_solvers) -ei_add_test(umeyama) -ei_add_test(householder) -ei_add_test(swap) -ei_add_test(conservative_resize) -ei_add_test(permutationmatrices) ei_add_test(sparse_permutations) -ei_add_test(nullary) +ei_add_test(simplicial_cholesky) +ei_add_test(conjugate_gradient) +ei_add_test(incomplete_cholesky) +ei_add_test(bicgstab) +ei_add_test(lscg) +ei_add_test(sparselu) +ei_add_test(sparseqr) +ei_add_test(umeyama) ei_add_test(nesting_ops "${CMAKE_CXX_FLAGS_DEBUG}") +ei_add_test(nestbyvalue) ei_add_test(zerosized) ei_add_test(dontalign) -ei_add_test(sizeoverflow) +ei_add_test(evaluators) +if(NOT EIGEN_TEST_NO_EXCEPTIONS) + ei_add_test(sizeoverflow) +endif() ei_add_test(prec_inverse_4x4) ei_add_test(vectorwiseop) ei_add_test(special_numbers) ei_add_test(rvalue_types) -ei_add_test(mpl2only) - -ei_add_test(simplicial_cholesky) -ei_add_test(conjugate_gradient) -ei_add_test(bicgstab) -ei_add_test(sparselu) -ei_add_test(sparseqr) - -# ei_add_test(denseLM) - -if(QT4_FOUND) - ei_add_test(qtvector "" "${QT_QTCORE_LIBRARY}") -endif(QT4_FOUND) +ei_add_test(dense_storage) +ei_add_test(ctorleak) +ei_add_test(inplace_decomposition) +ei_add_test(half_float) +ei_add_test(bfloat16_float) +ei_add_test(array_of_string) +ei_add_test(num_dimensions) +ei_add_test(stl_iterators) +ei_add_test(blasutil) +if(EIGEN_TEST_CXX11) + ei_add_test(initializer_list_construction) + ei_add_test(diagonal_matrix_variadic_ctor) +endif() -ei_add_test(eigen2support) +add_executable(bug1213 bug1213.cpp bug1213_main.cpp) -if(UMFPACK_FOUND) - ei_add_test(umfpack_support "" "${UMFPACK_ALL_LIBS}") +check_cxx_compiler_flag("-ffast-math" COMPILER_SUPPORT_FASTMATH) +if(COMPILER_SUPPORT_FASTMATH) + set(EIGEN_FASTMATH_FLAGS "-ffast-math") +else() + check_cxx_compiler_flag("/fp:fast" COMPILER_SUPPORT_FPFAST) + if(COMPILER_SUPPORT_FPFAST) + set(EIGEN_FASTMATH_FLAGS "/fp:fast") + endif() endif() -if(SUPERLU_FOUND) - ei_add_test(superlu_support "" "${SUPERLU_ALL_LIBS}") -endif() +ei_add_test(fastmath "${EIGEN_FASTMATH_FLAGS}") -if(CHOLMOD_FOUND) - ei_add_test(cholmod_support "" "${CHOLMOD_ALL_LIBS}") +# # ei_add_test(denseLM) + +if(QT4_FOUND) + ei_add_test(qtvector "" "${QT_QTCORE_LIBRARY}") endif() if(PARDISO_FOUND) @@ -262,7 +326,7 @@ if(PASTIX_FOUND AND (SCOTCH_FOUND OR METIS_FOUND)) ei_add_test(pastix_support "" "${PASTIX_ALL_LIBS}") endif() -if(SPQR_FOUND AND CHOLMOD_FOUND) +if(SPQR_FOUND AND EIGEN_TEST_CXX11 AND CHOLMOD_FOUND AND EIGEN_BUILD_BLAS AND EIGEN_BUILD_LAPACK) ei_add_test(spqr_support "" "${SPQR_ALL_LIBS}") endif() @@ -281,16 +345,131 @@ if(CMAKE_COMPILER_IS_GNUCXX AND NOT CXX_IS_QCC) ei_add_property(EIGEN_TESTING_SUMMARY "CXX_VERSION: ${EIGEN_CXX_VERSION_STRING}\n") endif() ei_add_property(EIGEN_TESTING_SUMMARY "CXX_FLAGS: ${CMAKE_CXX_FLAGS}\n") +if (EIGEN_TEST_CUSTOM_CXX_FLAGS) + ei_add_property(EIGEN_TESTING_SUMMARY "Custom CXX flags: ${EIGEN_TEST_CUSTOM_CXX_FLAGS}\n") +endif() ei_add_property(EIGEN_TESTING_SUMMARY "Sparse lib flags: ${SPARSE_LIBS}\n") option(EIGEN_TEST_EIGEN2 "Run whole Eigen2 test suite against EIGEN2_SUPPORT" OFF) mark_as_advanced(EIGEN_TEST_EIGEN2) if(EIGEN_TEST_EIGEN2) - add_subdirectory(eigen2) + message(WARNING "The Eigen2 test suite has been removed") endif() +# boost MP unit test +find_package(Boost 1.53.0) +if(Boost_FOUND AND EIGEN_TEST_CXX11) + include_directories(${Boost_INCLUDE_DIRS}) + ei_add_test(boostmultiprec "" "${Boost_LIBRARIES}") + ei_add_property(EIGEN_TESTED_BACKENDS "Boost.Multiprecision, ") +else() + ei_add_property(EIGEN_MISSING_BACKENDS "Boost.Multiprecision, ") +endif() + + +# CUDA unit tests +option(EIGEN_TEST_CUDA "Enable CUDA support in unit tests" OFF) +option(EIGEN_TEST_CUDA_CLANG "Use clang instead of nvcc to compile the CUDA tests" OFF) + +if(EIGEN_TEST_CUDA_CLANG AND NOT CMAKE_CXX_COMPILER MATCHES "clang") + message(WARNING "EIGEN_TEST_CUDA_CLANG is set, but CMAKE_CXX_COMPILER does not appear to be clang.") +endif() -option(EIGEN_TEST_BUILD_DOCUMENTATION "Test building the doxygen documentation" OFF) -IF(EIGEN_TEST_BUILD_DOCUMENTATION) +find_package(CUDA 9.0) +if(CUDA_FOUND AND EIGEN_TEST_CUDA) + # Make sure to compile without the -pedantic, -Wundef, -Wnon-virtual-dtor + # and -fno-check-new flags since they trigger thousands of compilation warnings + # in the CUDA runtime + string(REPLACE "-pedantic" "" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + string(REPLACE "-Wundef" "" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + string(REPLACE "-Wnon-virtual-dtor" "" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + string(REPLACE "-fno-check-new" "" CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS}") + + if(EIGEN_TEST_CUDA_CLANG) + set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11") + string(APPEND CMAKE_CXX_FLAGS " --cuda-path=${CUDA_TOOLKIT_ROOT_DIR}") + foreach(GPU IN LISTS EIGEN_CUDA_COMPUTE_ARCH) + string(APPEND CMAKE_CXX_FLAGS " --cuda-gpu-arch=sm_${GPU}") + endforeach() + string(APPEND CMAKE_CXX_FLAGS " ${EIGEN_CUDA_CXX_FLAGS}") + else() + set(CUDA_PROPAGATE_HOST_FLAGS OFF) + set(NVCC_ARCH_FLAGS) + # Define an -arch=sm_, otherwise if GPU does not exactly match one of + # those in the arch list for -gencode, the kernels will fail to run with + # cudaErrorNoKernelImageForDevice + # This can happen with newer cards (e.g. sm_75) and compiling with older + # versions of nvcc (e.g. 9.2) that do not support their specific arch. + list(LENGTH EIGEN_CUDA_COMPUTE_ARCH EIGEN_CUDA_COMPUTE_ARCH_SIZE) + if(EIGEN_CUDA_COMPUTE_ARCH_SIZE) + list(GET EIGEN_CUDA_COMPUTE_ARCH 0 EIGEN_CUDA_COMPUTE_DEFAULT) + set(NVCC_ARCH_FLAGS " -arch=sm_${EIGEN_CUDA_COMPUTE_DEFAULT}") + endif() + foreach(ARCH IN LISTS EIGEN_CUDA_COMPUTE_ARCH) + string(APPEND NVCC_ARCH_FLAGS " -gencode arch=compute_${ARCH},code=sm_${ARCH}") + endforeach() + set(CUDA_NVCC_FLAGS "--expt-relaxed-constexpr -Xcudafe \"--display_error_number\" ${NVCC_ARCH_FLAGS} ${CUDA_NVCC_FLAGS} ${EIGEN_CUDA_CXX_FLAGS}") + cuda_include_directories("${CMAKE_CURRENT_BINARY_DIR}" "${CUDA_TOOLKIT_ROOT_DIR}/include") + endif() + + set(EIGEN_ADD_TEST_FILENAME_EXTENSION "cu") + + ei_add_test(gpu_basic) + + unset(EIGEN_ADD_TEST_FILENAME_EXTENSION) + +endif() + + +# HIP unit tests +option(EIGEN_TEST_HIP "Add HIP support." OFF) +if (EIGEN_TEST_HIP) + + set(ROCM_PATH "/opt/rocm" CACHE STRING "Path to the ROCm installation.") + + if (EXISTS ${ROCM_PATH}/hip) + set(HIP_PATH ${ROCM_PATH}/hip) + list(APPEND CMAKE_MODULE_PATH ${HIP_PATH}/cmake) + elseif (EXISTS ${ROCM_PATH}/lib/cmake/hip) + set(HIP_PATH ${ROCM_PATH}) + list(APPEND CMAKE_MODULE_PATH ${HIP_PATH}/lib/cmake/hip) + else () + message(FATAL_ERROR "EIGEN_TEST_HIP is ON, but could not find the ROCm installation under ${ROCM_PATH}") + endif() + + find_package(HIP REQUIRED) + if (HIP_FOUND) + execute_process(COMMAND ${HIP_PATH}/bin/hipconfig --platform OUTPUT_VARIABLE HIP_PLATFORM) + + if ((${HIP_PLATFORM} STREQUAL "hcc") OR (${HIP_PLATFORM} STREQUAL "amd")) + + include_directories(${HIP_PATH}/include) + + set(EIGEN_ADD_TEST_FILENAME_EXTENSION "cu") + ei_add_test(gpu_basic) + unset(EIGEN_ADD_TEST_FILENAME_EXTENSION) + + elseif ((${HIP_PLATFORM} STREQUAL "nvcc") OR (${HIP_PLATFORM} STREQUAL "nvidia")) + message(FATAL_ERROR "HIP_PLATFORM = nvcc is not supported within Eigen") + else () + message(FATAL_ERROR "Unknown HIP_PLATFORM = ${HIP_PLATFORM}") + endif() + endif() +endif() + +if(EIGEN_TEST_SYCL) + set(EIGEN_SYCL ON) + include(SyclConfigureTesting) + + ei_add_test(sycl_basic) + set(EIGEN_SYCL OFF) +endif() + +cmake_dependent_option(EIGEN_TEST_BUILD_DOCUMENTATION "Test building the doxygen documentation" OFF "EIGEN_BUILD_DOC" OFF) +if(EIGEN_TEST_BUILD_DOCUMENTATION) add_dependencies(buildtests doc) -ENDIF() +endif() + +# Register all smoke tests +include("EigenSmokeTestList") +ei_add_smoke_tests("${ei_smoke_test_list}") diff --git a/thirdparty/eigen/test/MovableScalar.h b/thirdparty/eigen/test/MovableScalar.h new file mode 100644 index 00000000..6a90d037 --- /dev/null +++ b/thirdparty/eigen/test/MovableScalar.h @@ -0,0 +1,35 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2020 Sebastien Boisvert +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifndef EIGEN_MISC_MOVABLE_SCALAR_H +#define EIGEN_MISC_MOVABLE_SCALAR_H + +#include + +namespace Eigen +{ +template > +struct MovableScalar : public Base +{ + MovableScalar() = default; + ~MovableScalar() = default; + MovableScalar(const MovableScalar&) = default; + MovableScalar(MovableScalar&& other) = default; + MovableScalar& operator=(const MovableScalar&) = default; + MovableScalar& operator=(MovableScalar&& other) = default; + MovableScalar(Scalar scalar) : Base(100, scalar) {} + + operator Scalar() const { return this->size() > 0 ? this->back() : Scalar(); } +}; + +template<> struct NumTraits> : GenericNumTraits {}; +} + +#endif + diff --git a/thirdparty/eigen/test/SafeScalar.h b/thirdparty/eigen/test/SafeScalar.h new file mode 100644 index 00000000..c5cb7571 --- /dev/null +++ b/thirdparty/eigen/test/SafeScalar.h @@ -0,0 +1,30 @@ + +// A Scalar that asserts for uninitialized access. +template +class SafeScalar { + public: + SafeScalar() : initialized_(false) {} + SafeScalar(const SafeScalar& other) { + *this = other; + } + SafeScalar& operator=(const SafeScalar& other) { + val_ = T(other); + initialized_ = true; + return *this; + } + + SafeScalar(T val) : val_(val), initialized_(true) {} + SafeScalar& operator=(T val) { + val_ = val; + initialized_ = true; + } + + operator T() const { + VERIFY(initialized_ && "Uninitialized access."); + return val_; + } + + private: + T val_; + bool initialized_; +}; diff --git a/thirdparty/eigen/test/adjoint.cpp b/thirdparty/eigen/test/adjoint.cpp index ea36f784..4c4f98bb 100644 --- a/thirdparty/eigen/test/adjoint.cpp +++ b/thirdparty/eigen/test/adjoint.cpp @@ -42,6 +42,17 @@ template<> struct adjoint_specific { VERIFY_IS_APPROX(v1, v1.norm() * v3); VERIFY_IS_APPROX(v3, v1.normalized()); VERIFY_IS_APPROX(v3.norm(), RealScalar(1)); + + // check null inputs + VERIFY_IS_APPROX((v1*0).normalized(), (v1*0)); +#if (!EIGEN_ARCH_i386) || defined(EIGEN_VECTORIZE) + RealScalar very_small = (std::numeric_limits::min)(); + VERIFY( (v1*very_small).norm() == 0 ); + VERIFY_IS_APPROX((v1*very_small).normalized(), (v1*very_small)); + v3 = v1*very_small; + v3.normalize(); + VERIFY_IS_APPROX(v3, (v1*very_small)); +#endif // check compatibility of dot and adjoint ref = NumTraits::IsInteger ? 0 : (std::max)((std::max)(v1.norm(),v2.norm()),(std::max)((square * v2).norm(),(square.adjoint() * v1).norm())); @@ -59,11 +70,11 @@ template void adjoint(const MatrixType& m) Transpose.h Conjugate.h Dot.h */ using std::abs; - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef typename NumTraits::Real RealScalar; typedef Matrix VectorType; typedef Matrix SquareMatrixType; + const Index PacketSize = internal::packet_traits::size; Index rows = m.rows(); Index cols = m.cols(); @@ -108,6 +119,17 @@ template void adjoint(const MatrixType& m) VERIFY_IS_APPROX(m3,m1.transpose()); m3.transposeInPlace(); VERIFY_IS_APPROX(m3,m1); + + if(PacketSize(0,m3.rows()-PacketSize); + Index j = internal::random(0,m3.cols()-PacketSize); + m3.template block(i,j).transposeInPlace(); + VERIFY_IS_APPROX( (m3.template block(i,j)), (m1.template block(i,j).transpose()) ); + m3.template block(i,j).transposeInPlace(); + VERIFY_IS_APPROX(m3,m1); + } // check inplace adjoint m3 = m1; @@ -121,40 +143,77 @@ template void adjoint(const MatrixType& m) RealVectorType rv1 = RealVectorType::Random(rows); VERIFY_IS_APPROX(v1.dot(rv1.template cast()), v1.dot(rv1)); VERIFY_IS_APPROX(rv1.template cast().dot(v1), rv1.dot(v1)); + + VERIFY( is_same_type(m1,m1.template conjugateIf()) ); + VERIFY( is_same_type(m1.conjugate(),m1.template conjugateIf()) ); } -void test_adjoint() +template +void adjoint_extra() +{ + MatrixXcf a(10,10), b(10,10); + VERIFY_RAISES_ASSERT(a = a.transpose()); + VERIFY_RAISES_ASSERT(a = a.transpose() + b); + VERIFY_RAISES_ASSERT(a = b + a.transpose()); + VERIFY_RAISES_ASSERT(a = a.conjugate().transpose()); + VERIFY_RAISES_ASSERT(a = a.adjoint()); + VERIFY_RAISES_ASSERT(a = a.adjoint() + b); + VERIFY_RAISES_ASSERT(a = b + a.adjoint()); + + // no assertion should be triggered for these cases: + a.transpose() = a.transpose(); + a.transpose() += a.transpose(); + a.transpose() += a.transpose() + b; + a.transpose() = a.adjoint(); + a.transpose() += a.adjoint(); + a.transpose() += a.adjoint() + b; + + // regression tests for check_for_aliasing + MatrixXd c(10,10); + c = 1.0 * MatrixXd::Ones(10,10) + c; + c = MatrixXd::Ones(10,10) * 1.0 + c; + c = c + MatrixXd::Ones(10,10) .cwiseProduct( MatrixXd::Zero(10,10) ); + c = MatrixXd::Ones(10,10) * MatrixXd::Zero(10,10); + + // regression for bug 1646 + for (int j = 0; j < 10; ++j) { + c.col(j).head(j) = c.row(j).head(j); + } + + for (int j = 0; j < 10; ++j) { + c.col(j) = c.row(j); + } + + a.conservativeResize(1,1); + a = a.transpose(); + + a.conservativeResize(0,0); + a = a.transpose(); +} + +EIGEN_DECLARE_TEST(adjoint) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( adjoint(Matrix()) ); CALL_SUBTEST_2( adjoint(Matrix3d()) ); CALL_SUBTEST_3( adjoint(Matrix4f()) ); + CALL_SUBTEST_4( adjoint(MatrixXcf(internal::random(1,EIGEN_TEST_MAX_SIZE/2), internal::random(1,EIGEN_TEST_MAX_SIZE/2))) ); CALL_SUBTEST_5( adjoint(MatrixXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_6( adjoint(MatrixXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + + // Complement for 128 bits vectorization: + CALL_SUBTEST_8( adjoint(Matrix2d()) ); + CALL_SUBTEST_9( adjoint(Matrix()) ); + + // 256 bits vectorization: + CALL_SUBTEST_10( adjoint(Matrix()) ); + CALL_SUBTEST_11( adjoint(Matrix()) ); + CALL_SUBTEST_12( adjoint(Matrix()) ); } // test a large static matrix only once CALL_SUBTEST_7( adjoint(Matrix()) ); -#ifdef EIGEN_TEST_PART_4 - { - MatrixXcf a(10,10), b(10,10); - VERIFY_RAISES_ASSERT(a = a.transpose()); - VERIFY_RAISES_ASSERT(a = a.transpose() + b); - VERIFY_RAISES_ASSERT(a = b + a.transpose()); - VERIFY_RAISES_ASSERT(a = a.conjugate().transpose()); - VERIFY_RAISES_ASSERT(a = a.adjoint()); - VERIFY_RAISES_ASSERT(a = a.adjoint() + b); - VERIFY_RAISES_ASSERT(a = b + a.adjoint()); - - // no assertion should be triggered for these cases: - a.transpose() = a.transpose(); - a.transpose() += a.transpose(); - a.transpose() += a.transpose() + b; - a.transpose() = a.adjoint(); - a.transpose() += a.adjoint(); - a.transpose() += a.adjoint() + b; - } -#endif + CALL_SUBTEST_13( adjoint_extra<0>() ); } diff --git a/thirdparty/eigen/test/array.cpp b/thirdparty/eigen/test/array.cpp deleted file mode 100644 index 68f6b3d7..00000000 --- a/thirdparty/eigen/test/array.cpp +++ /dev/null @@ -1,308 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2008-2009 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void array(const ArrayType& m) -{ - typedef typename ArrayType::Index Index; - typedef typename ArrayType::Scalar Scalar; - typedef Array ColVectorType; - typedef Array RowVectorType; - - Index rows = m.rows(); - Index cols = m.cols(); - - ArrayType m1 = ArrayType::Random(rows, cols), - m2 = ArrayType::Random(rows, cols), - m3(rows, cols); - - ColVectorType cv1 = ColVectorType::Random(rows); - RowVectorType rv1 = RowVectorType::Random(cols); - - Scalar s1 = internal::random(), - s2 = internal::random(); - - // scalar addition - VERIFY_IS_APPROX(m1 + s1, s1 + m1); - VERIFY_IS_APPROX(m1 + s1, ArrayType::Constant(rows,cols,s1) + m1); - VERIFY_IS_APPROX(s1 - m1, (-m1)+s1 ); - VERIFY_IS_APPROX(m1 - s1, m1 - ArrayType::Constant(rows,cols,s1)); - VERIFY_IS_APPROX(s1 - m1, ArrayType::Constant(rows,cols,s1) - m1); - VERIFY_IS_APPROX((m1*Scalar(2)) - s2, (m1+m1) - ArrayType::Constant(rows,cols,s2) ); - m3 = m1; - m3 += s2; - VERIFY_IS_APPROX(m3, m1 + s2); - m3 = m1; - m3 -= s1; - VERIFY_IS_APPROX(m3, m1 - s1); - - // scalar operators via Maps - m3 = m1; - ArrayType::Map(m1.data(), m1.rows(), m1.cols()) -= ArrayType::Map(m2.data(), m2.rows(), m2.cols()); - VERIFY_IS_APPROX(m1, m3 - m2); - - m3 = m1; - ArrayType::Map(m1.data(), m1.rows(), m1.cols()) += ArrayType::Map(m2.data(), m2.rows(), m2.cols()); - VERIFY_IS_APPROX(m1, m3 + m2); - - m3 = m1; - ArrayType::Map(m1.data(), m1.rows(), m1.cols()) *= ArrayType::Map(m2.data(), m2.rows(), m2.cols()); - VERIFY_IS_APPROX(m1, m3 * m2); - - m3 = m1; - m2 = ArrayType::Random(rows,cols); - m2 = (m2==0).select(1,m2); - ArrayType::Map(m1.data(), m1.rows(), m1.cols()) /= ArrayType::Map(m2.data(), m2.rows(), m2.cols()); - VERIFY_IS_APPROX(m1, m3 / m2); - - // reductions - VERIFY_IS_APPROX(m1.abs().colwise().sum().sum(), m1.abs().sum()); - VERIFY_IS_APPROX(m1.abs().rowwise().sum().sum(), m1.abs().sum()); - using std::abs; - VERIFY_IS_MUCH_SMALLER_THAN(abs(m1.colwise().sum().sum() - m1.sum()), m1.abs().sum()); - VERIFY_IS_MUCH_SMALLER_THAN(abs(m1.rowwise().sum().sum() - m1.sum()), m1.abs().sum()); - if (!internal::isMuchSmallerThan(abs(m1.sum() - (m1+m2).sum()), m1.abs().sum(), test_precision())) - VERIFY_IS_NOT_APPROX(((m1+m2).rowwise().sum()).sum(), m1.sum()); - VERIFY_IS_APPROX(m1.colwise().sum(), m1.colwise().redux(internal::scalar_sum_op())); - - // vector-wise ops - m3 = m1; - VERIFY_IS_APPROX(m3.colwise() += cv1, m1.colwise() + cv1); - m3 = m1; - VERIFY_IS_APPROX(m3.colwise() -= cv1, m1.colwise() - cv1); - m3 = m1; - VERIFY_IS_APPROX(m3.rowwise() += rv1, m1.rowwise() + rv1); - m3 = m1; - VERIFY_IS_APPROX(m3.rowwise() -= rv1, m1.rowwise() - rv1); -} - -template void comparisons(const ArrayType& m) -{ - using std::abs; - typedef typename ArrayType::Index Index; - typedef typename ArrayType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - - Index rows = m.rows(); - Index cols = m.cols(); - - Index r = internal::random(0, rows-1), - c = internal::random(0, cols-1); - - ArrayType m1 = ArrayType::Random(rows, cols), - m2 = ArrayType::Random(rows, cols), - m3(rows, cols); - - VERIFY(((m1 + Scalar(1)) > m1).all()); - VERIFY(((m1 - Scalar(1)) < m1).all()); - if (rows*cols>1) - { - m3 = m1; - m3(r,c) += 1; - VERIFY(! (m1 < m3).all() ); - VERIFY(! (m1 > m3).all() ); - } - VERIFY(!(m1 > m2 && m1 < m2).any()); - VERIFY((m1 <= m2 || m1 >= m2).all()); - - // comparisons to scalar - VERIFY( (m1 != (m1(r,c)+1) ).any() ); - VERIFY( (m1 > (m1(r,c)-1) ).any() ); - VERIFY( (m1 < (m1(r,c)+1) ).any() ); - VERIFY( (m1 == m1(r,c) ).any() ); - - // test Select - VERIFY_IS_APPROX( (m1m2).select(m1,m2), m1.cwiseMax(m2) ); - Scalar mid = (m1.cwiseAbs().minCoeff() + m1.cwiseAbs().maxCoeff())/Scalar(2); - for (int j=0; j=ArrayType::Constant(rows,cols,mid)) - .select(m1,0), m3); - // even shorter version: - VERIFY_IS_APPROX( (m1.abs()RealScalar(0.1)).count() == rows*cols); - - // and/or - VERIFY( (m1RealScalar(0)).count() == 0); - VERIFY( (m1=RealScalar(0)).count() == rows*cols); - RealScalar a = m1.abs().mean(); - VERIFY( (m1<-a || m1>a).count() == (m1.abs()>a).count()); - - typedef Array ArrayOfIndices; - - // TODO allows colwise/rowwise for array - VERIFY_IS_APPROX(((m1.abs()+1)>RealScalar(0.1)).colwise().count(), ArrayOfIndices::Constant(cols,rows).transpose()); - VERIFY_IS_APPROX(((m1.abs()+1)>RealScalar(0.1)).rowwise().count(), ArrayOfIndices::Constant(rows, cols)); -} - -template void array_real(const ArrayType& m) -{ - using std::abs; - using std::sqrt; - typedef typename ArrayType::Index Index; - typedef typename ArrayType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - - Index rows = m.rows(); - Index cols = m.cols(); - - ArrayType m1 = ArrayType::Random(rows, cols), - m2 = ArrayType::Random(rows, cols), - m3(rows, cols); - - Scalar s1 = internal::random(); - - // these tests are mostly to check possible compilation issues. - VERIFY_IS_APPROX(m1.sin(), sin(m1)); - VERIFY_IS_APPROX(m1.cos(), cos(m1)); - VERIFY_IS_APPROX(m1.asin(), asin(m1)); - VERIFY_IS_APPROX(m1.acos(), acos(m1)); - VERIFY_IS_APPROX(m1.tan(), tan(m1)); - - VERIFY_IS_APPROX(cos(m1+RealScalar(3)*m2), cos((m1+RealScalar(3)*m2).eval())); - - VERIFY_IS_APPROX(m1.abs().sqrt(), sqrt(abs(m1))); - VERIFY_IS_APPROX(m1.abs(), sqrt(numext::abs2(m1))); - - VERIFY_IS_APPROX(numext::abs2(numext::real(m1)) + numext::abs2(numext::imag(m1)), numext::abs2(m1)); - VERIFY_IS_APPROX(numext::abs2(real(m1)) + numext::abs2(imag(m1)), numext::abs2(m1)); - if(!NumTraits::IsComplex) - VERIFY_IS_APPROX(numext::real(m1), m1); - - // shift argument of logarithm so that it is not zero - Scalar smallNumber = NumTraits::dummy_precision(); - VERIFY_IS_APPROX((m1.abs() + smallNumber).log() , log(abs(m1) + smallNumber)); - - VERIFY_IS_APPROX(m1.exp() * m2.exp(), exp(m1+m2)); - VERIFY_IS_APPROX(m1.exp(), exp(m1)); - VERIFY_IS_APPROX(m1.exp() / m2.exp(),(m1-m2).exp()); - - VERIFY_IS_APPROX(m1.pow(2), m1.square()); - VERIFY_IS_APPROX(pow(m1,2), m1.square()); - - ArrayType exponents = ArrayType::Constant(rows, cols, RealScalar(2)); - VERIFY_IS_APPROX(Eigen::pow(m1,exponents), m1.square()); - - m3 = m1.abs(); - VERIFY_IS_APPROX(m3.pow(RealScalar(0.5)), m3.sqrt()); - VERIFY_IS_APPROX(pow(m3,RealScalar(0.5)), m3.sqrt()); - - // scalar by array division - const RealScalar tiny = sqrt(std::numeric_limits::epsilon()); - s1 += Scalar(tiny); - m1 += ArrayType::Constant(rows,cols,Scalar(tiny)); - VERIFY_IS_APPROX(s1/m1, s1 * m1.inverse()); - - // check inplace transpose - m3 = m1; - m3.transposeInPlace(); - VERIFY_IS_APPROX(m3,m1.transpose()); - m3.transposeInPlace(); - VERIFY_IS_APPROX(m3,m1); -} - -template void array_complex(const ArrayType& m) -{ - typedef typename ArrayType::Index Index; - - Index rows = m.rows(); - Index cols = m.cols(); - - ArrayType m1 = ArrayType::Random(rows, cols), - m2(rows, cols); - - for (Index i = 0; i < m.rows(); ++i) - for (Index j = 0; j < m.cols(); ++j) - m2(i,j) = sqrt(m1(i,j)); - - VERIFY_IS_APPROX(m1.sqrt(), m2); - VERIFY_IS_APPROX(m1.sqrt(), Eigen::sqrt(m1)); -} - -template void min_max(const ArrayType& m) -{ - typedef typename ArrayType::Index Index; - typedef typename ArrayType::Scalar Scalar; - - Index rows = m.rows(); - Index cols = m.cols(); - - ArrayType m1 = ArrayType::Random(rows, cols); - - // min/max with array - Scalar maxM1 = m1.maxCoeff(); - Scalar minM1 = m1.minCoeff(); - - VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, minM1), (m1.min)(ArrayType::Constant(rows,cols, minM1))); - VERIFY_IS_APPROX(m1, (m1.min)(ArrayType::Constant(rows,cols, maxM1))); - - VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, maxM1), (m1.max)(ArrayType::Constant(rows,cols, maxM1))); - VERIFY_IS_APPROX(m1, (m1.max)(ArrayType::Constant(rows,cols, minM1))); - - // min/max with scalar input - VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, minM1), (m1.min)( minM1)); - VERIFY_IS_APPROX(m1, (m1.min)( maxM1)); - - VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, maxM1), (m1.max)( maxM1)); - VERIFY_IS_APPROX(m1, (m1.max)( minM1)); - -} - -void test_array() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( array(Array()) ); - CALL_SUBTEST_2( array(Array22f()) ); - CALL_SUBTEST_3( array(Array44d()) ); - CALL_SUBTEST_4( array(ArrayXXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - CALL_SUBTEST_5( array(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - CALL_SUBTEST_6( array(ArrayXXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( comparisons(Array()) ); - CALL_SUBTEST_2( comparisons(Array22f()) ); - CALL_SUBTEST_3( comparisons(Array44d()) ); - CALL_SUBTEST_5( comparisons(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - CALL_SUBTEST_6( comparisons(ArrayXXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( min_max(Array()) ); - CALL_SUBTEST_2( min_max(Array22f()) ); - CALL_SUBTEST_3( min_max(Array44d()) ); - CALL_SUBTEST_5( min_max(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - CALL_SUBTEST_6( min_max(ArrayXXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( array_real(Array()) ); - CALL_SUBTEST_2( array_real(Array22f()) ); - CALL_SUBTEST_3( array_real(Array44d()) ); - CALL_SUBTEST_5( array_real(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_4( array_complex(ArrayXXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - } - - VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, int >::value)); - VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, float >::value)); - VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, ArrayBase >::value)); - typedef CwiseUnaryOp, ArrayXd > Xpr; - VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, - ArrayBase - >::value)); -} diff --git a/thirdparty/eigen/test/array_cwise.cpp b/thirdparty/eigen/test/array_cwise.cpp new file mode 100644 index 00000000..bbb74b1a --- /dev/null +++ b/thirdparty/eigen/test/array_cwise.cpp @@ -0,0 +1,726 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2008-2009 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" + + +// Test the corner cases of pow(x, y) for real types. +template +void pow_test() { + const Scalar zero = Scalar(0); + const Scalar eps = Eigen::NumTraits::epsilon(); + const Scalar one = Scalar(1); + const Scalar two = Scalar(2); + const Scalar three = Scalar(3); + const Scalar sqrt_half = Scalar(std::sqrt(0.5)); + const Scalar sqrt2 = Scalar(std::sqrt(2)); + const Scalar inf = Eigen::NumTraits::infinity(); + const Scalar nan = Eigen::NumTraits::quiet_NaN(); + const Scalar denorm_min = EIGEN_ARCH_ARM ? zero : std::numeric_limits::denorm_min(); + const Scalar min = (std::numeric_limits::min)(); + const Scalar max = (std::numeric_limits::max)(); + const Scalar max_exp = (static_cast(int(Eigen::NumTraits::max_exponent())) * Scalar(EIGEN_LN2)) / eps; + + const static Scalar abs_vals[] = {zero, + denorm_min, + min, + eps, + sqrt_half, + one, + sqrt2, + two, + three, + max_exp, + max, + inf, + nan}; + const int abs_cases = 13; + const int num_cases = 2*abs_cases * 2*abs_cases; + // Repeat the same value to make sure we hit the vectorized path. + const int num_repeats = 32; + Array x(num_repeats, num_cases); + Array y(num_repeats, num_cases); + int count = 0; + for (int i = 0; i < abs_cases; ++i) { + const Scalar abs_x = abs_vals[i]; + for (int sign_x = 0; sign_x < 2; ++sign_x) { + Scalar x_case = sign_x == 0 ? -abs_x : abs_x; + for (int j = 0; j < abs_cases; ++j) { + const Scalar abs_y = abs_vals[j]; + for (int sign_y = 0; sign_y < 2; ++sign_y) { + Scalar y_case = sign_y == 0 ? -abs_y : abs_y; + for (int repeat = 0; repeat < num_repeats; ++repeat) { + x(repeat, count) = x_case; + y(repeat, count) = y_case; + } + ++count; + } + } + } + } + + Array actual = x.pow(y); + const Scalar tol = test_precision(); + bool all_pass = true; + for (int i = 0; i < 1; ++i) { + for (int j = 0; j < num_cases; ++j) { + Scalar e = static_cast(std::pow(x(i,j), y(i,j))); + Scalar a = actual(i, j); +#if EIGEN_ARCH_ARM + // Work around NEON flush-to-zero mode + // if ref returns a subnormal value and Eigen returns 0, then skip the test + if (a == Scalar(0) && + (e > -(std::numeric_limits::min)() && e < (std::numeric_limits::min)() && + e >= -std::numeric_limits::denorm_min() && e <= std::numeric_limits::denorm_min())) { + continue; + } +#endif + bool success = (a == e) || ((numext::isfinite)(e) && internal::isApprox(a, e, tol)) || + ((numext::isnan)(a) && (numext::isnan)(e)); + all_pass &= success; + if (!success) { + std::cout << "pow(" << x(i,j) << "," << y(i,j) << ") = " << a << " != " << e << std::endl; + } + } + } + VERIFY(all_pass); +} + +template void array(const ArrayType& m) +{ + typedef typename ArrayType::Scalar Scalar; + typedef typename ArrayType::RealScalar RealScalar; + typedef Array ColVectorType; + typedef Array RowVectorType; + + Index rows = m.rows(); + Index cols = m.cols(); + + ArrayType m1 = ArrayType::Random(rows, cols), + m2 = ArrayType::Random(rows, cols), + m3(rows, cols); + ArrayType m4 = m1; // copy constructor + VERIFY_IS_APPROX(m1, m4); + + ColVectorType cv1 = ColVectorType::Random(rows); + RowVectorType rv1 = RowVectorType::Random(cols); + + Scalar s1 = internal::random(), + s2 = internal::random(); + + // scalar addition + VERIFY_IS_APPROX(m1 + s1, s1 + m1); + VERIFY_IS_APPROX(m1 + s1, ArrayType::Constant(rows,cols,s1) + m1); + VERIFY_IS_APPROX(s1 - m1, (-m1)+s1 ); + VERIFY_IS_APPROX(m1 - s1, m1 - ArrayType::Constant(rows,cols,s1)); + VERIFY_IS_APPROX(s1 - m1, ArrayType::Constant(rows,cols,s1) - m1); + VERIFY_IS_APPROX((m1*Scalar(2)) - s2, (m1+m1) - ArrayType::Constant(rows,cols,s2) ); + m3 = m1; + m3 += s2; + VERIFY_IS_APPROX(m3, m1 + s2); + m3 = m1; + m3 -= s1; + VERIFY_IS_APPROX(m3, m1 - s1); + + // scalar operators via Maps + m3 = m1; + ArrayType::Map(m1.data(), m1.rows(), m1.cols()) -= ArrayType::Map(m2.data(), m2.rows(), m2.cols()); + VERIFY_IS_APPROX(m1, m3 - m2); + + m3 = m1; + ArrayType::Map(m1.data(), m1.rows(), m1.cols()) += ArrayType::Map(m2.data(), m2.rows(), m2.cols()); + VERIFY_IS_APPROX(m1, m3 + m2); + + m3 = m1; + ArrayType::Map(m1.data(), m1.rows(), m1.cols()) *= ArrayType::Map(m2.data(), m2.rows(), m2.cols()); + VERIFY_IS_APPROX(m1, m3 * m2); + + m3 = m1; + m2 = ArrayType::Random(rows,cols); + m2 = (m2==0).select(1,m2); + ArrayType::Map(m1.data(), m1.rows(), m1.cols()) /= ArrayType::Map(m2.data(), m2.rows(), m2.cols()); + VERIFY_IS_APPROX(m1, m3 / m2); + + // reductions + VERIFY_IS_APPROX(m1.abs().colwise().sum().sum(), m1.abs().sum()); + VERIFY_IS_APPROX(m1.abs().rowwise().sum().sum(), m1.abs().sum()); + using std::abs; + VERIFY_IS_MUCH_SMALLER_THAN(abs(m1.colwise().sum().sum() - m1.sum()), m1.abs().sum()); + VERIFY_IS_MUCH_SMALLER_THAN(abs(m1.rowwise().sum().sum() - m1.sum()), m1.abs().sum()); + if (!internal::isMuchSmallerThan(abs(m1.sum() - (m1+m2).sum()), m1.abs().sum(), test_precision())) + VERIFY_IS_NOT_APPROX(((m1+m2).rowwise().sum()).sum(), m1.sum()); + VERIFY_IS_APPROX(m1.colwise().sum(), m1.colwise().redux(internal::scalar_sum_op())); + + // vector-wise ops + m3 = m1; + VERIFY_IS_APPROX(m3.colwise() += cv1, m1.colwise() + cv1); + m3 = m1; + VERIFY_IS_APPROX(m3.colwise() -= cv1, m1.colwise() - cv1); + m3 = m1; + VERIFY_IS_APPROX(m3.rowwise() += rv1, m1.rowwise() + rv1); + m3 = m1; + VERIFY_IS_APPROX(m3.rowwise() -= rv1, m1.rowwise() - rv1); + + // Conversion from scalar + VERIFY_IS_APPROX((m3 = s1), ArrayType::Constant(rows,cols,s1)); + VERIFY_IS_APPROX((m3 = 1), ArrayType::Constant(rows,cols,1)); + VERIFY_IS_APPROX((m3.topLeftCorner(rows,cols) = 1), ArrayType::Constant(rows,cols,1)); + typedef Array FixedArrayType; + { + FixedArrayType f1(s1); + VERIFY_IS_APPROX(f1, FixedArrayType::Constant(s1)); + FixedArrayType f2(numext::real(s1)); + VERIFY_IS_APPROX(f2, FixedArrayType::Constant(numext::real(s1))); + FixedArrayType f3((int)100*numext::real(s1)); + VERIFY_IS_APPROX(f3, FixedArrayType::Constant((int)100*numext::real(s1))); + f1.setRandom(); + FixedArrayType f4(f1.data()); + VERIFY_IS_APPROX(f4, f1); + } + #if EIGEN_HAS_CXX11 + { + FixedArrayType f1{s1}; + VERIFY_IS_APPROX(f1, FixedArrayType::Constant(s1)); + FixedArrayType f2{numext::real(s1)}; + VERIFY_IS_APPROX(f2, FixedArrayType::Constant(numext::real(s1))); + FixedArrayType f3{(int)100*numext::real(s1)}; + VERIFY_IS_APPROX(f3, FixedArrayType::Constant((int)100*numext::real(s1))); + f1.setRandom(); + FixedArrayType f4{f1.data()}; + VERIFY_IS_APPROX(f4, f1); + } + #endif + + // pow + VERIFY_IS_APPROX(m1.pow(2), m1.square()); + VERIFY_IS_APPROX(pow(m1,2), m1.square()); + VERIFY_IS_APPROX(m1.pow(3), m1.cube()); + VERIFY_IS_APPROX(pow(m1,3), m1.cube()); + VERIFY_IS_APPROX((-m1).pow(3), -m1.cube()); + VERIFY_IS_APPROX(pow(2*m1,3), 8*m1.cube()); + ArrayType exponents = ArrayType::Constant(rows, cols, RealScalar(2)); + VERIFY_IS_APPROX(Eigen::pow(m1,exponents), m1.square()); + VERIFY_IS_APPROX(m1.pow(exponents), m1.square()); + VERIFY_IS_APPROX(Eigen::pow(2*m1,exponents), 4*m1.square()); + VERIFY_IS_APPROX((2*m1).pow(exponents), 4*m1.square()); + VERIFY_IS_APPROX(Eigen::pow(m1,2*exponents), m1.square().square()); + VERIFY_IS_APPROX(m1.pow(2*exponents), m1.square().square()); + VERIFY_IS_APPROX(Eigen::pow(m1(0,0), exponents), ArrayType::Constant(rows,cols,m1(0,0)*m1(0,0))); + + // Check possible conflicts with 1D ctor + typedef Array OneDArrayType; + { + OneDArrayType o1(rows); + VERIFY(o1.size()==rows); + OneDArrayType o2(static_cast(rows)); + VERIFY(o2.size()==rows); + } + #if EIGEN_HAS_CXX11 + { + OneDArrayType o1{rows}; + VERIFY(o1.size()==rows); + OneDArrayType o4{int(rows)}; + VERIFY(o4.size()==rows); + } + #endif + // Check possible conflicts with 2D ctor + typedef Array TwoDArrayType; + typedef Array ArrayType2; + { + TwoDArrayType o1(rows,cols); + VERIFY(o1.rows()==rows); + VERIFY(o1.cols()==cols); + TwoDArrayType o2(static_cast(rows),static_cast(cols)); + VERIFY(o2.rows()==rows); + VERIFY(o2.cols()==cols); + + ArrayType2 o3(rows,cols); + VERIFY(o3(0)==Scalar(rows) && o3(1)==Scalar(cols)); + ArrayType2 o4(static_cast(rows),static_cast(cols)); + VERIFY(o4(0)==Scalar(rows) && o4(1)==Scalar(cols)); + } + #if EIGEN_HAS_CXX11 + { + TwoDArrayType o1{rows,cols}; + VERIFY(o1.rows()==rows); + VERIFY(o1.cols()==cols); + TwoDArrayType o2{int(rows),int(cols)}; + VERIFY(o2.rows()==rows); + VERIFY(o2.cols()==cols); + + ArrayType2 o3{rows,cols}; + VERIFY(o3(0)==Scalar(rows) && o3(1)==Scalar(cols)); + ArrayType2 o4{int(rows),int(cols)}; + VERIFY(o4(0)==Scalar(rows) && o4(1)==Scalar(cols)); + } + #endif +} + +template void comparisons(const ArrayType& m) +{ + using std::abs; + typedef typename ArrayType::Scalar Scalar; + typedef typename NumTraits::Real RealScalar; + + Index rows = m.rows(); + Index cols = m.cols(); + + Index r = internal::random(0, rows-1), + c = internal::random(0, cols-1); + + ArrayType m1 = ArrayType::Random(rows, cols), + m2 = ArrayType::Random(rows, cols), + m3(rows, cols), + m4 = m1; + + m4 = (m4.abs()==Scalar(0)).select(1,m4); + + VERIFY(((m1 + Scalar(1)) > m1).all()); + VERIFY(((m1 - Scalar(1)) < m1).all()); + if (rows*cols>1) + { + m3 = m1; + m3(r,c) += 1; + VERIFY(! (m1 < m3).all() ); + VERIFY(! (m1 > m3).all() ); + } + VERIFY(!(m1 > m2 && m1 < m2).any()); + VERIFY((m1 <= m2 || m1 >= m2).all()); + + // comparisons array to scalar + VERIFY( (m1 != (m1(r,c)+1) ).any() ); + VERIFY( (m1 > (m1(r,c)-1) ).any() ); + VERIFY( (m1 < (m1(r,c)+1) ).any() ); + VERIFY( (m1 == m1(r,c) ).any() ); + + // comparisons scalar to array + VERIFY( ( (m1(r,c)+1) != m1).any() ); + VERIFY( ( (m1(r,c)-1) < m1).any() ); + VERIFY( ( (m1(r,c)+1) > m1).any() ); + VERIFY( ( m1(r,c) == m1).any() ); + + // test Select + VERIFY_IS_APPROX( (m1m2).select(m1,m2), m1.cwiseMax(m2) ); + Scalar mid = (m1.cwiseAbs().minCoeff() + m1.cwiseAbs().maxCoeff())/Scalar(2); + for (int j=0; j=ArrayType::Constant(rows,cols,mid)) + .select(m1,0), m3); + // even shorter version: + VERIFY_IS_APPROX( (m1.abs()RealScalar(0.1)).count() == rows*cols); + + // and/or + VERIFY( (m1RealScalar(0)).count() == 0); + VERIFY( (m1=RealScalar(0)).count() == rows*cols); + RealScalar a = m1.abs().mean(); + VERIFY( (m1<-a || m1>a).count() == (m1.abs()>a).count()); + + typedef Array ArrayOfIndices; + + // TODO allows colwise/rowwise for array + VERIFY_IS_APPROX(((m1.abs()+1)>RealScalar(0.1)).colwise().count(), ArrayOfIndices::Constant(cols,rows).transpose()); + VERIFY_IS_APPROX(((m1.abs()+1)>RealScalar(0.1)).rowwise().count(), ArrayOfIndices::Constant(rows, cols)); +} + +template void array_real(const ArrayType& m) +{ + using std::abs; + using std::sqrt; + typedef typename ArrayType::Scalar Scalar; + typedef typename NumTraits::Real RealScalar; + + Index rows = m.rows(); + Index cols = m.cols(); + + ArrayType m1 = ArrayType::Random(rows, cols), + m2 = ArrayType::Random(rows, cols), + m3(rows, cols), + m4 = m1; + + // avoid denormalized values so verification doesn't fail on platforms that don't support them + // denormalized behavior is tested elsewhere (unary_op_test, binary_ops_test) + const Scalar min = (std::numeric_limits::min)(); + m1 = (m1.abs()(); + + // these tests are mostly to check possible compilation issues with free-functions. + VERIFY_IS_APPROX(m1.sin(), sin(m1)); + VERIFY_IS_APPROX(m1.cos(), cos(m1)); + VERIFY_IS_APPROX(m1.tan(), tan(m1)); + VERIFY_IS_APPROX(m1.asin(), asin(m1)); + VERIFY_IS_APPROX(m1.acos(), acos(m1)); + VERIFY_IS_APPROX(m1.atan(), atan(m1)); + VERIFY_IS_APPROX(m1.sinh(), sinh(m1)); + VERIFY_IS_APPROX(m1.cosh(), cosh(m1)); + VERIFY_IS_APPROX(m1.tanh(), tanh(m1)); +#if EIGEN_HAS_CXX11_MATH + VERIFY_IS_APPROX(m1.tanh().atanh(), atanh(tanh(m1))); + VERIFY_IS_APPROX(m1.sinh().asinh(), asinh(sinh(m1))); + VERIFY_IS_APPROX(m1.cosh().acosh(), acosh(cosh(m1))); +#endif + VERIFY_IS_APPROX(m1.logistic(), logistic(m1)); + + VERIFY_IS_APPROX(m1.arg(), arg(m1)); + VERIFY_IS_APPROX(m1.round(), round(m1)); + VERIFY_IS_APPROX(m1.rint(), rint(m1)); + VERIFY_IS_APPROX(m1.floor(), floor(m1)); + VERIFY_IS_APPROX(m1.ceil(), ceil(m1)); + VERIFY((m1.isNaN() == (Eigen::isnan)(m1)).all()); + VERIFY((m1.isInf() == (Eigen::isinf)(m1)).all()); + VERIFY((m1.isFinite() == (Eigen::isfinite)(m1)).all()); + VERIFY_IS_APPROX(m4.inverse(), inverse(m4)); + VERIFY_IS_APPROX(m1.abs(), abs(m1)); + VERIFY_IS_APPROX(m1.abs2(), abs2(m1)); + VERIFY_IS_APPROX(m1.square(), square(m1)); + VERIFY_IS_APPROX(m1.cube(), cube(m1)); + VERIFY_IS_APPROX(cos(m1+RealScalar(3)*m2), cos((m1+RealScalar(3)*m2).eval())); + VERIFY_IS_APPROX(m1.sign(), sign(m1)); + VERIFY((m1.sqrt().sign().isNaN() == (Eigen::isnan)(sign(sqrt(m1)))).all()); + + // avoid inf and NaNs so verification doesn't fail + m3 = m4.abs(); + + VERIFY_IS_APPROX(m3.sqrt(), sqrt(abs(m3))); + VERIFY_IS_APPROX(m3.rsqrt(), Scalar(1)/sqrt(abs(m3))); + VERIFY_IS_APPROX(rsqrt(m3), Scalar(1)/sqrt(abs(m3))); + VERIFY_IS_APPROX(m3.log(), log(m3)); + VERIFY_IS_APPROX(m3.log1p(), log1p(m3)); + VERIFY_IS_APPROX(m3.log10(), log10(m3)); + VERIFY_IS_APPROX(m3.log2(), log2(m3)); + + + VERIFY((!(m1>m2) == (m1<=m2)).all()); + + VERIFY_IS_APPROX(sin(m1.asin()), m1); + VERIFY_IS_APPROX(cos(m1.acos()), m1); + VERIFY_IS_APPROX(tan(m1.atan()), m1); + VERIFY_IS_APPROX(sinh(m1), Scalar(0.5)*(exp(m1)-exp(-m1))); + VERIFY_IS_APPROX(cosh(m1), Scalar(0.5)*(exp(m1)+exp(-m1))); + VERIFY_IS_APPROX(tanh(m1), (Scalar(0.5)*(exp(m1)-exp(-m1)))/(Scalar(0.5)*(exp(m1)+exp(-m1)))); + VERIFY_IS_APPROX(logistic(m1), (Scalar(1)/(Scalar(1)+exp(-m1)))); + VERIFY_IS_APPROX(arg(m1), ((m1())*Scalar(std::acos(Scalar(-1)))); + VERIFY((round(m1) <= ceil(m1) && round(m1) >= floor(m1)).all()); + VERIFY((rint(m1) <= ceil(m1) && rint(m1) >= floor(m1)).all()); + VERIFY(((ceil(m1) - round(m1)) <= Scalar(0.5) || (round(m1) - floor(m1)) <= Scalar(0.5)).all()); + VERIFY(((ceil(m1) - round(m1)) <= Scalar(1.0) && (round(m1) - floor(m1)) <= Scalar(1.0)).all()); + VERIFY(((ceil(m1) - rint(m1)) <= Scalar(0.5) || (rint(m1) - floor(m1)) <= Scalar(0.5)).all()); + VERIFY(((ceil(m1) - rint(m1)) <= Scalar(1.0) && (rint(m1) - floor(m1)) <= Scalar(1.0)).all()); + VERIFY((Eigen::isnan)((m1*Scalar(0))/Scalar(0)).all()); + VERIFY((Eigen::isinf)(m4/Scalar(0)).all()); + VERIFY(((Eigen::isfinite)(m1) && (!(Eigen::isfinite)(m1*Scalar(0)/Scalar(0))) && (!(Eigen::isfinite)(m4/Scalar(0)))).all()); + VERIFY_IS_APPROX(inverse(inverse(m4)),m4); + VERIFY((abs(m1) == m1 || abs(m1) == -m1).all()); + VERIFY_IS_APPROX(m3, sqrt(abs2(m3))); + VERIFY_IS_APPROX(m1.absolute_difference(m2), (m1 > m2).select(m1 - m2, m2 - m1)); + VERIFY_IS_APPROX( m1.sign(), -(-m1).sign() ); + VERIFY_IS_APPROX( m1*m1.sign(),m1.abs()); + VERIFY_IS_APPROX(m1.sign() * m1.abs(), m1); + + VERIFY_IS_APPROX(numext::abs2(numext::real(m1)) + numext::abs2(numext::imag(m1)), numext::abs2(m1)); + VERIFY_IS_APPROX(numext::abs2(Eigen::real(m1)) + numext::abs2(Eigen::imag(m1)), numext::abs2(m1)); + if(!NumTraits::IsComplex) + VERIFY_IS_APPROX(numext::real(m1), m1); + + // shift argument of logarithm so that it is not zero + Scalar smallNumber = NumTraits::dummy_precision(); + VERIFY_IS_APPROX((m3 + smallNumber).log() , log(abs(m3) + smallNumber)); + VERIFY_IS_APPROX((m3 + smallNumber + Scalar(1)).log() , log1p(abs(m3) + smallNumber)); + + VERIFY_IS_APPROX(m1.exp() * m2.exp(), exp(m1+m2)); + VERIFY_IS_APPROX(m1.exp(), exp(m1)); + VERIFY_IS_APPROX(m1.exp() / m2.exp(),(m1-m2).exp()); + + VERIFY_IS_APPROX(m1.expm1(), expm1(m1)); + VERIFY_IS_APPROX((m3 + smallNumber).exp() - Scalar(1), expm1(abs(m3) + smallNumber)); + + VERIFY_IS_APPROX(m3.pow(RealScalar(0.5)), m3.sqrt()); + VERIFY_IS_APPROX(pow(m3,RealScalar(0.5)), m3.sqrt()); + + VERIFY_IS_APPROX(m3.pow(RealScalar(-0.5)), m3.rsqrt()); + VERIFY_IS_APPROX(pow(m3,RealScalar(-0.5)), m3.rsqrt()); + + // Avoid inf and NaN. + m3 = (m1.square()::epsilon()).select(Scalar(1),m3); + VERIFY_IS_APPROX(m3.pow(RealScalar(-2)), m3.square().inverse()); + pow_test(); + + VERIFY_IS_APPROX(log10(m3), log(m3)/numext::log(Scalar(10))); + VERIFY_IS_APPROX(log2(m3), log(m3)/numext::log(Scalar(2))); + + // scalar by array division + const RealScalar tiny = sqrt(std::numeric_limits::epsilon()); + s1 += Scalar(tiny); + m1 += ArrayType::Constant(rows,cols,Scalar(tiny)); + VERIFY_IS_APPROX(s1/m1, s1 * m1.inverse()); + + // check inplace transpose + m3 = m1; + m3.transposeInPlace(); + VERIFY_IS_APPROX(m3, m1.transpose()); + m3.transposeInPlace(); + VERIFY_IS_APPROX(m3, m1); +} + +template void array_complex(const ArrayType& m) +{ + typedef typename ArrayType::Scalar Scalar; + typedef typename NumTraits::Real RealScalar; + + Index rows = m.rows(); + Index cols = m.cols(); + + ArrayType m1 = ArrayType::Random(rows, cols), + m2(rows, cols), + m4 = m1; + + m4.real() = (m4.real().abs()==RealScalar(0)).select(RealScalar(1),m4.real()); + m4.imag() = (m4.imag().abs()==RealScalar(0)).select(RealScalar(1),m4.imag()); + + Array m3(rows, cols); + + for (Index i = 0; i < m.rows(); ++i) + for (Index j = 0; j < m.cols(); ++j) + m2(i,j) = sqrt(m1(i,j)); + + // these tests are mostly to check possible compilation issues with free-functions. + VERIFY_IS_APPROX(m1.sin(), sin(m1)); + VERIFY_IS_APPROX(m1.cos(), cos(m1)); + VERIFY_IS_APPROX(m1.tan(), tan(m1)); + VERIFY_IS_APPROX(m1.sinh(), sinh(m1)); + VERIFY_IS_APPROX(m1.cosh(), cosh(m1)); + VERIFY_IS_APPROX(m1.tanh(), tanh(m1)); + VERIFY_IS_APPROX(m1.logistic(), logistic(m1)); + VERIFY_IS_APPROX(m1.arg(), arg(m1)); + VERIFY((m1.isNaN() == (Eigen::isnan)(m1)).all()); + VERIFY((m1.isInf() == (Eigen::isinf)(m1)).all()); + VERIFY((m1.isFinite() == (Eigen::isfinite)(m1)).all()); + VERIFY_IS_APPROX(m4.inverse(), inverse(m4)); + VERIFY_IS_APPROX(m1.log(), log(m1)); + VERIFY_IS_APPROX(m1.log10(), log10(m1)); + VERIFY_IS_APPROX(m1.log2(), log2(m1)); + VERIFY_IS_APPROX(m1.abs(), abs(m1)); + VERIFY_IS_APPROX(m1.abs2(), abs2(m1)); + VERIFY_IS_APPROX(m1.sqrt(), sqrt(m1)); + VERIFY_IS_APPROX(m1.square(), square(m1)); + VERIFY_IS_APPROX(m1.cube(), cube(m1)); + VERIFY_IS_APPROX(cos(m1+RealScalar(3)*m2), cos((m1+RealScalar(3)*m2).eval())); + VERIFY_IS_APPROX(m1.sign(), sign(m1)); + + + VERIFY_IS_APPROX(m1.exp() * m2.exp(), exp(m1+m2)); + VERIFY_IS_APPROX(m1.exp(), exp(m1)); + VERIFY_IS_APPROX(m1.exp() / m2.exp(),(m1-m2).exp()); + + VERIFY_IS_APPROX(m1.expm1(), expm1(m1)); + VERIFY_IS_APPROX(expm1(m1), exp(m1) - 1.); + // Check for larger magnitude complex numbers that expm1 matches exp - 1. + VERIFY_IS_APPROX(expm1(10. * m1), exp(10. * m1) - 1.); + + VERIFY_IS_APPROX(sinh(m1), 0.5*(exp(m1)-exp(-m1))); + VERIFY_IS_APPROX(cosh(m1), 0.5*(exp(m1)+exp(-m1))); + VERIFY_IS_APPROX(tanh(m1), (0.5*(exp(m1)-exp(-m1)))/(0.5*(exp(m1)+exp(-m1)))); + VERIFY_IS_APPROX(logistic(m1), (1.0/(1.0 + exp(-m1)))); + + for (Index i = 0; i < m.rows(); ++i) + for (Index j = 0; j < m.cols(); ++j) + m3(i,j) = std::atan2(m1(i,j).imag(), m1(i,j).real()); + VERIFY_IS_APPROX(arg(m1), m3); + + std::complex zero(0.0,0.0); + VERIFY((Eigen::isnan)(m1*zero/zero).all()); +#if EIGEN_COMP_MSVC + // msvc complex division is not robust + VERIFY((Eigen::isinf)(m4/RealScalar(0)).all()); +#else +#if EIGEN_COMP_CLANG + // clang's complex division is notoriously broken too + if((numext::isinf)(m4(0,0)/RealScalar(0))) { +#endif + VERIFY((Eigen::isinf)(m4/zero).all()); +#if EIGEN_COMP_CLANG + } + else + { + VERIFY((Eigen::isinf)(m4.real()/zero.real()).all()); + } +#endif +#endif // MSVC + + VERIFY(((Eigen::isfinite)(m1) && (!(Eigen::isfinite)(m1*zero/zero)) && (!(Eigen::isfinite)(m1/zero))).all()); + + VERIFY_IS_APPROX(inverse(inverse(m4)),m4); + VERIFY_IS_APPROX(conj(m1.conjugate()), m1); + VERIFY_IS_APPROX(abs(m1), sqrt(square(m1.real())+square(m1.imag()))); + VERIFY_IS_APPROX(abs(m1), sqrt(abs2(m1))); + VERIFY_IS_APPROX(log10(m1), log(m1)/log(10)); + VERIFY_IS_APPROX(log2(m1), log(m1)/log(2)); + + VERIFY_IS_APPROX( m1.sign(), -(-m1).sign() ); + VERIFY_IS_APPROX( m1.sign() * m1.abs(), m1); + + // scalar by array division + Scalar s1 = internal::random(); + const RealScalar tiny = std::sqrt(std::numeric_limits::epsilon()); + s1 += Scalar(tiny); + m1 += ArrayType::Constant(rows,cols,Scalar(tiny)); + VERIFY_IS_APPROX(s1/m1, s1 * m1.inverse()); + + // check inplace transpose + m2 = m1; + m2.transposeInPlace(); + VERIFY_IS_APPROX(m2, m1.transpose()); + m2.transposeInPlace(); + VERIFY_IS_APPROX(m2, m1); + // Check vectorized inplace transpose. + ArrayType m5 = ArrayType::Random(131, 131); + ArrayType m6 = m5; + m6.transposeInPlace(); + VERIFY_IS_APPROX(m6, m5.transpose()); +} + +template void min_max(const ArrayType& m) +{ + typedef typename ArrayType::Scalar Scalar; + + Index rows = m.rows(); + Index cols = m.cols(); + + ArrayType m1 = ArrayType::Random(rows, cols); + + // min/max with array + Scalar maxM1 = m1.maxCoeff(); + Scalar minM1 = m1.minCoeff(); + + VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, minM1), (m1.min)(ArrayType::Constant(rows,cols, minM1))); + VERIFY_IS_APPROX(m1, (m1.min)(ArrayType::Constant(rows,cols, maxM1))); + + VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, maxM1), (m1.max)(ArrayType::Constant(rows,cols, maxM1))); + VERIFY_IS_APPROX(m1, (m1.max)(ArrayType::Constant(rows,cols, minM1))); + + // min/max with scalar input + VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, minM1), (m1.min)( minM1)); + VERIFY_IS_APPROX(m1, (m1.min)( maxM1)); + + VERIFY_IS_APPROX(ArrayType::Constant(rows,cols, maxM1), (m1.max)( maxM1)); + VERIFY_IS_APPROX(m1, (m1.max)( minM1)); + + + // min/max with various NaN propagation options. + if (m1.size() > 1 && !NumTraits::IsInteger) { + m1(0,0) = NumTraits::quiet_NaN(); + maxM1 = m1.template maxCoeff(); + minM1 = m1.template minCoeff(); + VERIFY((numext::isnan)(maxM1)); + VERIFY((numext::isnan)(minM1)); + + maxM1 = m1.template maxCoeff(); + minM1 = m1.template minCoeff(); + VERIFY(!(numext::isnan)(maxM1)); + VERIFY(!(numext::isnan)(minM1)); + } +} + +template +struct shift_left { + template + Scalar operator()(const Scalar& v) const { + return v << N; + } +}; + +template +struct arithmetic_shift_right { + template + Scalar operator()(const Scalar& v) const { + return v >> N; + } +}; + +template void array_integer(const ArrayType& m) +{ + Index rows = m.rows(); + Index cols = m.cols(); + + ArrayType m1 = ArrayType::Random(rows, cols), + m2(rows, cols); + + m2 = m1.template shiftLeft<2>(); + VERIFY( (m2 == m1.unaryExpr(shift_left<2>())).all() ); + m2 = m1.template shiftLeft<9>(); + VERIFY( (m2 == m1.unaryExpr(shift_left<9>())).all() ); + + m2 = m1.template shiftRight<2>(); + VERIFY( (m2 == m1.unaryExpr(arithmetic_shift_right<2>())).all() ); + m2 = m1.template shiftRight<9>(); + VERIFY( (m2 == m1.unaryExpr(arithmetic_shift_right<9>())).all() ); +} + +EIGEN_DECLARE_TEST(array_cwise) +{ + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1( array(Array()) ); + CALL_SUBTEST_2( array(Array22f()) ); + CALL_SUBTEST_3( array(Array44d()) ); + CALL_SUBTEST_4( array(ArrayXXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_5( array(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( array(ArrayXXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( array(Array(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( array_integer(ArrayXXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( array_integer(Array(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + } + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1( comparisons(Array()) ); + CALL_SUBTEST_2( comparisons(Array22f()) ); + CALL_SUBTEST_3( comparisons(Array44d()) ); + CALL_SUBTEST_5( comparisons(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( comparisons(ArrayXXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + } + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1( min_max(Array()) ); + CALL_SUBTEST_2( min_max(Array22f()) ); + CALL_SUBTEST_3( min_max(Array44d()) ); + CALL_SUBTEST_5( min_max(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( min_max(ArrayXXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + } + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1( array_real(Array()) ); + CALL_SUBTEST_2( array_real(Array22f()) ); + CALL_SUBTEST_3( array_real(Array44d()) ); + CALL_SUBTEST_5( array_real(ArrayXXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_7( array_real(Array()) ); + CALL_SUBTEST_8( array_real(Array()) ); + } + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_4( array_complex(ArrayXXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + } + + VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, int >::value)); + VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, float >::value)); + VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, ArrayBase >::value)); + typedef CwiseUnaryOp, ArrayXd > Xpr; + VERIFY((internal::is_same< internal::global_math_functions_filtering_base::type, + ArrayBase + >::value)); +} diff --git a/thirdparty/eigen/test/array_for_matrix.cpp b/thirdparty/eigen/test/array_for_matrix.cpp index 9667e1f1..06e04a2f 100644 --- a/thirdparty/eigen/test/array_for_matrix.cpp +++ b/thirdparty/eigen/test/array_for_matrix.cpp @@ -11,7 +11,6 @@ template void array_for_matrix(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef Matrix ColVectorType; typedef Matrix RowVectorType; @@ -45,7 +44,7 @@ template void array_for_matrix(const MatrixType& m) VERIFY_IS_MUCH_SMALLER_THAN(m1.rowwise().sum().sum() - m1.sum(), m1.squaredNorm()); VERIFY_IS_MUCH_SMALLER_THAN(m1.colwise().sum() + m2.colwise().sum() - (m1+m2).colwise().sum(), (m1+m2).squaredNorm()); VERIFY_IS_MUCH_SMALLER_THAN(m1.rowwise().sum() - m2.rowwise().sum() - (m1-m2).rowwise().sum(), (m1-m2).squaredNorm()); - VERIFY_IS_APPROX(m1.colwise().sum(), m1.colwise().redux(internal::scalar_sum_op())); + VERIFY_IS_APPROX(m1.colwise().sum(), m1.colwise().redux(internal::scalar_sum_op())); // vector-wise ops m3 = m1; @@ -58,7 +57,14 @@ template void array_for_matrix(const MatrixType& m) VERIFY_IS_APPROX(m3.rowwise() -= rv1, m1.rowwise() - rv1); // empty objects - VERIFY_IS_APPROX(m1.block(0,0,0,cols).colwise().sum(), RowVectorType::Zero(cols)); + VERIFY_IS_APPROX((m1.template block<0,Dynamic>(0,0,0,cols).colwise().sum()), RowVectorType::Zero(cols)); + VERIFY_IS_APPROX((m1.template block(0,0,rows,0).rowwise().sum()), ColVectorType::Zero(rows)); + VERIFY_IS_APPROX((m1.template block<0,Dynamic>(0,0,0,cols).colwise().prod()), RowVectorType::Ones(cols)); + VERIFY_IS_APPROX((m1.template block(0,0,rows,0).rowwise().prod()), ColVectorType::Ones(rows)); + + VERIFY_IS_APPROX(m1.block(0,0,0,cols).colwise().sum(), RowVectorType::Zero(cols)); + VERIFY_IS_APPROX(m1.block(0,0,rows,0).rowwise().sum(), ColVectorType::Zero(rows)); + VERIFY_IS_APPROX(m1.block(0,0,0,cols).colwise().prod(), RowVectorType::Ones(cols)); VERIFY_IS_APPROX(m1.block(0,0,rows,0).rowwise().prod(), ColVectorType::Ones(rows)); // verify the const accessors exist @@ -68,12 +74,21 @@ template void array_for_matrix(const MatrixType& m) const Scalar& ref_a2 = m.array().matrix().coeffRef(0,0); VERIFY(&ref_a1 == &ref_m1); VERIFY(&ref_a2 == &ref_m2); + + // Check write accessors: + m1.array().coeffRef(0,0) = 1; + VERIFY_IS_APPROX(m1(0,0),Scalar(1)); + m1.array()(0,0) = 2; + VERIFY_IS_APPROX(m1(0,0),Scalar(2)); + m1.array().matrix().coeffRef(0,0) = 3; + VERIFY_IS_APPROX(m1(0,0),Scalar(3)); + m1.array().matrix()(0,0) = 4; + VERIFY_IS_APPROX(m1(0,0),Scalar(4)); } template void comparisons(const MatrixType& m) { using std::abs; - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef typename NumTraits::Real RealScalar; @@ -124,7 +139,13 @@ template void comparisons(const MatrixType& m) // count VERIFY(((m1.array().abs()+1)>RealScalar(0.1)).count() == rows*cols); - typedef Matrix VectorOfIndices; + // and/or + VERIFY( ((m1.array()RealScalar(0)).matrix()).count() == 0); + VERIFY( ((m1.array()=RealScalar(0)).matrix()).count() == rows*cols); + RealScalar a = m1.cwiseAbs().mean(); + VERIFY( ((m1.array()<-a).matrix() || (m1.array()>a).matrix()).count() == (m1.cwiseAbs().array()>a).count()); + + typedef Matrix VectorOfIndices; // TODO allows colwise/rowwise for array VERIFY_IS_APPROX(((m1.array().abs()+1)>RealScalar(0.1)).matrix().colwise().count(), VectorOfIndices::Constant(cols,rows).transpose()); @@ -134,9 +155,21 @@ template void comparisons(const MatrixType& m) template void lpNorm(const VectorType& v) { using std::sqrt; + typedef typename VectorType::RealScalar RealScalar; VectorType u = VectorType::Random(v.size()); - VERIFY_IS_APPROX(u.template lpNorm(), u.cwiseAbs().maxCoeff()); + if(v.size()==0) + { + VERIFY_IS_APPROX(u.template lpNorm(), RealScalar(0)); + VERIFY_IS_APPROX(u.template lpNorm<1>(), RealScalar(0)); + VERIFY_IS_APPROX(u.template lpNorm<2>(), RealScalar(0)); + VERIFY_IS_APPROX(u.template lpNorm<5>(), RealScalar(0)); + } + else + { + VERIFY_IS_APPROX(u.template lpNorm(), u.cwiseAbs().maxCoeff()); + } + VERIFY_IS_APPROX(u.template lpNorm<1>(), u.cwiseAbs().sum()); VERIFY_IS_APPROX(u.template lpNorm<2>(), sqrt(u.array().abs().square().sum())); VERIFY_IS_APPROX(numext::pow(u.template lpNorm<5>(), typename VectorType::RealScalar(5)), u.array().abs().pow(5).sum()); @@ -144,7 +177,6 @@ template void lpNorm(const VectorType& v) template void cwise_min_max(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; Index rows = m.rows(); @@ -179,11 +211,44 @@ template void cwise_min_max(const MatrixType& m) VERIFY_IS_APPROX(MatrixType::Constant(rows,cols, maxM1).array(), (m1.array().max)( maxM1)); VERIFY_IS_APPROX(m1.array(), (m1.array().max)( minM1)); + // Test NaN propagation for min/max. + if (!NumTraits::IsInteger) { + m1(0,0) = NumTraits::quiet_NaN(); + // Elementwise. + VERIFY((numext::isnan)(m1.template cwiseMax(MatrixType::Constant(rows,cols, Scalar(1)))(0,0))); + VERIFY((numext::isnan)(m1.template cwiseMin(MatrixType::Constant(rows,cols, Scalar(1)))(0,0))); + VERIFY(!(numext::isnan)(m1.template cwiseMax(MatrixType::Constant(rows,cols, Scalar(1)))(0,0))); + VERIFY(!(numext::isnan)(m1.template cwiseMin(MatrixType::Constant(rows,cols, Scalar(1)))(0,0))); + VERIFY((numext::isnan)(m1.template cwiseMax(Scalar(1))(0,0))); + VERIFY((numext::isnan)(m1.template cwiseMin(Scalar(1))(0,0))); + VERIFY(!(numext::isnan)(m1.template cwiseMax(Scalar(1))(0,0))); + VERIFY(!(numext::isnan)(m1.template cwiseMin(Scalar(1))(0,0))); + + + VERIFY((numext::isnan)(m1.array().template max(MatrixType::Constant(rows,cols, Scalar(1)).array())(0,0))); + VERIFY((numext::isnan)(m1.array().template min(MatrixType::Constant(rows,cols, Scalar(1)).array())(0,0))); + VERIFY(!(numext::isnan)(m1.array().template max(MatrixType::Constant(rows,cols, Scalar(1)).array())(0,0))); + VERIFY(!(numext::isnan)(m1.array().template min(MatrixType::Constant(rows,cols, Scalar(1)).array())(0,0))); + VERIFY((numext::isnan)(m1.array().template max(Scalar(1))(0,0))); + VERIFY((numext::isnan)(m1.array().template min(Scalar(1))(0,0))); + VERIFY(!(numext::isnan)(m1.array().template max(Scalar(1))(0,0))); + VERIFY(!(numext::isnan)(m1.array().template min(Scalar(1))(0,0))); + + // Reductions. + VERIFY((numext::isnan)(m1.template maxCoeff())); + VERIFY((numext::isnan)(m1.template minCoeff())); + if (m1.size() > 1) { + VERIFY(!(numext::isnan)(m1.template maxCoeff())); + VERIFY(!(numext::isnan)(m1.template minCoeff())); + } else { + VERIFY((numext::isnan)(m1.template maxCoeff())); + VERIFY((numext::isnan)(m1.template minCoeff())); + } + } } template void resize(const MatrixTraits& t) { - typedef typename MatrixTraits::Index Index; typedef typename MatrixTraits::Scalar Scalar; typedef Matrix MatrixType; typedef Array Array2DType; @@ -207,13 +272,32 @@ template void resize(const MatrixTraits& t) VERIFY(a1.size()==cols); } +template void regression_bug_654() { ArrayXf a = RowVectorXf(3); VectorXf v = Array(3); } -void test_array_for_matrix() +// Check propagation of LvalueBit through Array/Matrix-Wrapper +template +void regrrssion_bug_1410() +{ + const Matrix4i M; + const Array4i A; + ArrayWrapper MA = M.array(); + MA.row(0); + MatrixWrapper AM = A.matrix(); + AM.row(0); + + VERIFY((internal::traits >::Flags&LvalueBit)==0); + VERIFY((internal::traits >::Flags&LvalueBit)==0); + + VERIFY((internal::traits >::Flags&LvalueBit)==LvalueBit); + VERIFY((internal::traits >::Flags&LvalueBit)==LvalueBit); +} + +EIGEN_DECLARE_TEST(array_for_matrix) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( array_for_matrix(Matrix()) ); @@ -245,10 +329,13 @@ void test_array_for_matrix() CALL_SUBTEST_5( lpNorm(VectorXf(internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_4( lpNorm(VectorXcf(internal::random(1,EIGEN_TEST_MAX_SIZE))) ); } + CALL_SUBTEST_5( lpNorm(VectorXf(0)) ); + CALL_SUBTEST_4( lpNorm(VectorXcf(0)) ); for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_4( resize(MatrixXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_5( resize(MatrixXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_6( resize(MatrixXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); } - CALL_SUBTEST_6( regression_bug_654() ); + CALL_SUBTEST_6( regression_bug_654<0>() ); + CALL_SUBTEST_6( regrrssion_bug_1410<0>() ); } diff --git a/thirdparty/eigen/test/array_of_string.cpp b/thirdparty/eigen/test/array_of_string.cpp new file mode 100644 index 00000000..23e51529 --- /dev/null +++ b/thirdparty/eigen/test/array_of_string.cpp @@ -0,0 +1,32 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2016 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" + +EIGEN_DECLARE_TEST(array_of_string) +{ + typedef Array ArrayXs; + ArrayXs a1(3), a2(3), a3(3), a3ref(3); + a1 << "one", "two", "three"; + a2 << "1", "2", "3"; + a3ref << "one (1)", "two (2)", "three (3)"; + std::stringstream s1; + s1 << a1; + VERIFY_IS_EQUAL(s1.str(), std::string(" one two three")); + a3 = a1 + std::string(" (") + a2 + std::string(")"); + VERIFY((a3==a3ref).all()); + + a3 = a1; + a3 += std::string(" (") + a2 + std::string(")"); + VERIFY((a3==a3ref).all()); + + a1.swap(a3); + VERIFY((a1==a3ref).all()); + VERIFY((a3!=a3ref).all()); +} diff --git a/thirdparty/eigen/test/array_replicate.cpp b/thirdparty/eigen/test/array_replicate.cpp index f412d1ae..057c3c77 100644 --- a/thirdparty/eigen/test/array_replicate.cpp +++ b/thirdparty/eigen/test/array_replicate.cpp @@ -14,7 +14,6 @@ template void replicate(const MatrixType& m) /* this test covers the following files: Replicate.cpp */ - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef Matrix VectorType; typedef Matrix MatrixX; @@ -44,6 +43,19 @@ template void replicate(const MatrixType& m) x2 << m2, m2, m2, m2, m2, m2; VERIFY_IS_APPROX(x2, (m2.template replicate<2,3>())); + + x2.resize(rows,3*cols); + x2 << m2, m2, m2; + VERIFY_IS_APPROX(x2, (m2.template replicate<1,3>())); + + vx1.resize(3*rows,cols); + vx1 << m2, m2, m2; + VERIFY_IS_APPROX(vx1+vx1, vx1+(m2.template replicate<3,1>())); + + vx1=m2+(m2.colwise().replicate(1)); + + if(m2.cols()==1) + VERIFY_IS_APPROX(m2.coeff(0), (m2.template replicate<3,1>().coeff(m2.rows()))); x2.resize(rows,f1); for (int j=0; j void replicate(const MatrixType& m) VERIFY_IS_APPROX(vx1, v1.colwise().replicate(f2)); } -void test_array_replicate() +EIGEN_DECLARE_TEST(array_replicate) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( replicate(Matrix()) ); diff --git a/thirdparty/eigen/test/array_reverse.cpp b/thirdparty/eigen/test/array_reverse.cpp index fbe7a990..c77528a5 100644 --- a/thirdparty/eigen/test/array_reverse.cpp +++ b/thirdparty/eigen/test/array_reverse.cpp @@ -15,7 +15,6 @@ using namespace std; template void reverse(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef Matrix VectorType; @@ -24,7 +23,7 @@ template void reverse(const MatrixType& m) // this test relies a lot on Random.h, and there's not much more that we can do // to test it, hence I consider that we will have tested Random.h - MatrixType m1 = MatrixType::Random(rows, cols); + MatrixType m1 = MatrixType::Random(rows, cols), m2; VectorType v1 = VectorType::Random(rows); MatrixType m1_r = m1.reverse(); @@ -96,33 +95,110 @@ template void reverse(const MatrixType& m) m1.reverse()(r, c) = x; VERIFY_IS_APPROX(x, m1(rows - 1 - r, cols - 1 - c)); + + m2 = m1; + m2.reverseInPlace(); + VERIFY_IS_APPROX(m2,m1.reverse().eval()); + + m2 = m1; + m2.col(0).reverseInPlace(); + VERIFY_IS_APPROX(m2.col(0),m1.col(0).reverse().eval()); + + m2 = m1; + m2.row(0).reverseInPlace(); + VERIFY_IS_APPROX(m2.row(0),m1.row(0).reverse().eval()); + + m2 = m1; + m2.rowwise().reverseInPlace(); + VERIFY_IS_APPROX(m2,m1.rowwise().reverse().eval()); + + m2 = m1; + m2.colwise().reverseInPlace(); + VERIFY_IS_APPROX(m2,m1.colwise().reverse().eval()); - /* m1.colwise().reverse()(r, c) = x; VERIFY_IS_APPROX(x, m1(rows - 1 - r, c)); m1.rowwise().reverse()(r, c) = x; VERIFY_IS_APPROX(x, m1(r, cols - 1 - c)); - */ } -void test_array_reverse() +template +void array_reverse_extra() +{ + Vector4f x; x << 1, 2, 3, 4; + Vector4f y; y << 4, 3, 2, 1; + VERIFY(x.reverse()[1] == 3); + VERIFY(x.reverse() == y); +} + +// Simpler version of reverseInPlace leveraging a bug +// in clang 6/7 with -O2 and AVX or AVX512 enabled. +// This simpler version ensure that the clang bug is not simply hidden +// through mis-inlining of reverseInPlace or other minor changes. +template +EIGEN_DONT_INLINE +void bug1684_job1(MatrixType& m1, MatrixType& m2) +{ + m2 = m1; + m2.col(0).swap(m2.col(3)); + m2.col(1).swap(m2.col(2)); +} + +template +EIGEN_DONT_INLINE +void bug1684_job2(MatrixType& m1, MatrixType& m2) +{ + m2 = m1; // load m1/m2 in AVX registers + m1.col(0) = m2.col(3); // perform 128 bits moves + m1.col(1) = m2.col(2); + m1.col(2) = m2.col(1); + m1.col(3) = m2.col(0); +} + +template +EIGEN_DONT_INLINE +void bug1684_job3(MatrixType& m1, MatrixType& m2) +{ + m2 = m1; + Vector4f tmp; + tmp = m2.col(0); + m2.col(0) = m2.col(3); + m2.col(3) = tmp; + tmp = m2.col(1); + m2.col(1) = m2.col(2); + m2.col(2) = tmp; + +} + +template +void bug1684() +{ + Matrix4f m1 = Matrix4f::Random(); + Matrix4f m2 = Matrix4f::Random(); + bug1684_job1(m1,m2); + VERIFY_IS_APPROX(m2, m1.rowwise().reverse().eval()); + bug1684_job2(m1,m2); + VERIFY_IS_APPROX(m2, m1.rowwise().reverse().eval()); + // This one still fail after our swap's workaround, + // but I expect users not to implement their own swap. + // bug1684_job3(m1,m2); + // VERIFY_IS_APPROX(m2, m1.rowwise().reverse().eval()); +} + +EIGEN_DECLARE_TEST(array_reverse) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( reverse(Matrix()) ); CALL_SUBTEST_2( reverse(Matrix2f()) ); CALL_SUBTEST_3( reverse(Matrix4f()) ); CALL_SUBTEST_4( reverse(Matrix4d()) ); - CALL_SUBTEST_5( reverse(MatrixXcf(3, 3)) ); - CALL_SUBTEST_6( reverse(MatrixXi(6, 3)) ); - CALL_SUBTEST_7( reverse(MatrixXcd(20, 20)) ); + CALL_SUBTEST_5( reverse(MatrixXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( reverse(MatrixXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_7( reverse(MatrixXcd(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_8( reverse(Matrix()) ); - CALL_SUBTEST_9( reverse(Matrix(6,3)) ); + CALL_SUBTEST_9( reverse(Matrix(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_3( bug1684<0>() ); } -#ifdef EIGEN_TEST_PART_3 - Vector4f x; x << 1, 2, 3, 4; - Vector4f y; y << 4, 3, 2, 1; - VERIFY(x.reverse()[1] == 3); - VERIFY(x.reverse() == y); -#endif + CALL_SUBTEST_3( array_reverse_extra<0>() ); } diff --git a/thirdparty/eigen/test/bandmatrix.cpp b/thirdparty/eigen/test/bandmatrix.cpp index 5e4e8e07..66a1b0db 100644 --- a/thirdparty/eigen/test/bandmatrix.cpp +++ b/thirdparty/eigen/test/bandmatrix.cpp @@ -11,7 +11,6 @@ template void bandmatrix(const MatrixType& _m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef typename NumTraits::Real RealScalar; typedef Matrix DenseMatrixType; @@ -60,10 +59,8 @@ template void bandmatrix(const MatrixType& _m) using Eigen::internal::BandMatrix; -void test_bandmatrix() +EIGEN_DECLARE_TEST(bandmatrix) { - typedef BandMatrix::Index Index; - for(int i = 0; i < 10*g_repeat ; i++) { Index rows = internal::random(1,10); Index cols = internal::random(1,10); diff --git a/thirdparty/eigen/test/basicstuff.cpp b/thirdparty/eigen/test/basicstuff.cpp index 8c0621ec..4ca607c8 100644 --- a/thirdparty/eigen/test/basicstuff.cpp +++ b/thirdparty/eigen/test/basicstuff.cpp @@ -10,10 +10,10 @@ #define EIGEN_NO_STATIC_ASSERT #include "main.h" +#include "random_without_cast_overflow.h" template void basicStuff(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef Matrix VectorType; typedef Matrix SquareMatrixType; @@ -49,6 +49,22 @@ template void basicStuff(const MatrixType& m) v1[r] = x; VERIFY_IS_APPROX(x, v1[r]); + // test fetching with various index types. + Index r1 = internal::random(0, numext::mini(Index(127),rows-1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); +#if EIGEN_HAS_CXX11 + x = v1(static_cast(r1)); + x = v1(static_cast(r1)); +#endif + VERIFY_IS_APPROX( v1, v1); VERIFY_IS_NOT_APPROX( v1, 2*v1); VERIFY_IS_MUCH_SMALLER_THAN( vzero, v1); @@ -75,7 +91,7 @@ template void basicStuff(const MatrixType& m) Matrix cv(rows); rv = square.row(r); cv = square.col(r); - + VERIFY_IS_APPROX(rv, cv.transpose()); if(cols!=1 && rows!=1 && MatrixType::SizeAtCompileTime!=Dynamic) @@ -105,32 +121,45 @@ template void basicStuff(const MatrixType& m) m1 = m2; VERIFY(m1==m2); VERIFY(!(m1!=m2)); - + // check automatic transposition sm2.setZero(); - for(typename MatrixType::Index i=0;i(0,10)>5; + m3 = b ? m1 : m2; + if(b) VERIFY_IS_APPROX(m3,m1); + else VERIFY_IS_APPROX(m3,m2); + m3 = b ? -m1 : m2; + if(b) VERIFY_IS_APPROX(m3,-m1); + else VERIFY_IS_APPROX(m3,m2); + m3 = b ? m1 : -m2; + if(b) VERIFY_IS_APPROX(m3,m1); + else VERIFY_IS_APPROX(m3,-m2); + } } template void basicStuffComplex(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef typename NumTraits::Real RealScalar; typedef Matrix RealMatrixType; @@ -166,32 +195,143 @@ template void basicStuffComplex(const MatrixType& m) VERIFY(!static_cast(cm).imag().isZero()); } -#ifdef EIGEN_TEST_PART_2 -void casting() +template +struct casting_test { + static void run() { + Matrix m; + for (int i=0; i::value(); + } + } + Matrix n = m.template cast(); + for (int i=0; i(m(i, j)))); + } + } + } +}; + +template +struct casting_test_runner { + static void run() { + casting_test::run(); + casting_test::run(); + casting_test::run(); + casting_test::run(); + casting_test::run(); + casting_test::run(); + casting_test::run(); +#if EIGEN_HAS_CXX11 + casting_test::run(); + casting_test::run(); +#endif + casting_test::run(); + casting_test::run(); + casting_test::run(); + casting_test::run(); + casting_test >::run(); + casting_test >::run(); + } +}; + +template +struct casting_test_runner::IsComplex)>::type> { - Matrix4f m = Matrix4f::Random(), m2; - Matrix4d n = m.cast(); - VERIFY(m.isApprox(n.cast())); - m2 = m.cast(); // check the specialization when NewType == Type - VERIFY(m.isApprox(m2)); -} + static void run() { + // Only a few casts from std::complex are defined. + casting_test::run(); + casting_test::run(); + casting_test >::run(); + casting_test >::run(); + } +}; + +void casting_all() { + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner::run(); +#if EIGEN_HAS_CXX11 + casting_test_runner::run(); + casting_test_runner::run(); #endif + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner::run(); + casting_test_runner >::run(); + casting_test_runner >::run(); +} template void fixedSizeMatrixConstruction() { - const Scalar raw[3] = {1,2,3}; - Matrix m(raw); - Array a(raw); - VERIFY(m(0) == 1); - VERIFY(m(1) == 2); - VERIFY(m(2) == 3); - VERIFY(a(0) == 1); - VERIFY(a(1) == 2); - VERIFY(a(2) == 3); + Scalar raw[4]; + for(int k=0; k<4; ++k) + raw[k] = internal::random(); + + { + Matrix m(raw); + Array a(raw); + for(int k=0; k<4; ++k) VERIFY(m(k) == raw[k]); + for(int k=0; k<4; ++k) VERIFY(a(k) == raw[k]); + VERIFY_IS_EQUAL(m,(Matrix(raw[0],raw[1],raw[2],raw[3]))); + VERIFY((a==(Array(raw[0],raw[1],raw[2],raw[3]))).all()); + } + { + Matrix m(raw); + Array a(raw); + for(int k=0; k<3; ++k) VERIFY(m(k) == raw[k]); + for(int k=0; k<3; ++k) VERIFY(a(k) == raw[k]); + VERIFY_IS_EQUAL(m,(Matrix(raw[0],raw[1],raw[2]))); + VERIFY((a==Array(raw[0],raw[1],raw[2])).all()); + } + { + Matrix m(raw), m2( (DenseIndex(raw[0])), (DenseIndex(raw[1])) ); + Array a(raw), a2( (DenseIndex(raw[0])), (DenseIndex(raw[1])) ); + for(int k=0; k<2; ++k) VERIFY(m(k) == raw[k]); + for(int k=0; k<2; ++k) VERIFY(a(k) == raw[k]); + VERIFY_IS_EQUAL(m,(Matrix(raw[0],raw[1]))); + VERIFY((a==Array(raw[0],raw[1])).all()); + for(int k=0; k<2; ++k) VERIFY(m2(k) == DenseIndex(raw[k])); + for(int k=0; k<2; ++k) VERIFY(a2(k) == DenseIndex(raw[k])); + } + { + Matrix m(raw), + m2( (DenseIndex(raw[0])), (DenseIndex(raw[1])) ), + m3( (int(raw[0])), (int(raw[1])) ), + m4( (float(raw[0])), (float(raw[1])) ); + Array a(raw), a2( (DenseIndex(raw[0])), (DenseIndex(raw[1])) ); + for(int k=0; k<2; ++k) VERIFY(m(k) == raw[k]); + for(int k=0; k<2; ++k) VERIFY(a(k) == raw[k]); + VERIFY_IS_EQUAL(m,(Matrix(raw[0],raw[1]))); + VERIFY((a==Array(raw[0],raw[1])).all()); + for(int k=0; k<2; ++k) VERIFY(m2(k) == DenseIndex(raw[k])); + for(int k=0; k<2; ++k) VERIFY(a2(k) == DenseIndex(raw[k])); + for(int k=0; k<2; ++k) VERIFY(m3(k) == int(raw[k])); + for(int k=0; k<2; ++k) VERIFY((m4(k)) == Scalar(float(raw[k]))); + } + { + Matrix m(raw), m1(raw[0]), m2( (DenseIndex(raw[0])) ), m3( (int(raw[0])) ); + Array a(raw), a1(raw[0]), a2( (DenseIndex(raw[0])) ); + VERIFY(m(0) == raw[0]); + VERIFY(a(0) == raw[0]); + VERIFY(m1(0) == raw[0]); + VERIFY(a1(0) == raw[0]); + VERIFY(m2(0) == DenseIndex(raw[0])); + VERIFY(a2(0) == DenseIndex(raw[0])); + VERIFY(m3(0) == int(raw[0])); + VERIFY_IS_EQUAL(m,(Matrix(raw[0]))); + VERIFY((a==Array(raw[0])).all()); + } } -void test_basicstuff() +EIGEN_DECLARE_TEST(basicstuff) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( basicStuff(Matrix()) ); @@ -201,14 +341,16 @@ void test_basicstuff() CALL_SUBTEST_5( basicStuff(MatrixXcd(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_6( basicStuff(Matrix()) ); CALL_SUBTEST_7( basicStuff(Matrix(internal::random(1,EIGEN_TEST_MAX_SIZE),internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_8( casting_all() ); CALL_SUBTEST_3( basicStuffComplex(MatrixXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_5( basicStuffComplex(MatrixXcd(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); } CALL_SUBTEST_1(fixedSizeMatrixConstruction()); + CALL_SUBTEST_1(fixedSizeMatrixConstruction()); CALL_SUBTEST_1(fixedSizeMatrixConstruction()); - CALL_SUBTEST_1(fixedSizeMatrixConstruction()); - - CALL_SUBTEST_2(casting()); + CALL_SUBTEST_1(fixedSizeMatrixConstruction()); + CALL_SUBTEST_1(fixedSizeMatrixConstruction()); + CALL_SUBTEST_1(fixedSizeMatrixConstruction()); } diff --git a/thirdparty/eigen/test/bdcsvd.cpp b/thirdparty/eigen/test/bdcsvd.cpp new file mode 100644 index 00000000..3ac83f52 --- /dev/null +++ b/thirdparty/eigen/test/bdcsvd.cpp @@ -0,0 +1,163 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2013 Gauthier Brun +// Copyright (C) 2013 Nicolas Carre +// Copyright (C) 2013 Jean Ceccato +// Copyright (C) 2013 Pierre Zoppitelli +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/ + +// We explicitly disable deprecated declarations for this set of tests +// because we purposely verify assertions for the deprecated SVD runtime +// option behavior. +#if defined(__GNUC__) +#pragma GCC diagnostic ignored "-Wdeprecated-declarations" +#elif defined(_MSC_VER) +#pragma warning( disable : 4996 ) +#endif + +// discard stack allocation as that too bypasses malloc +#define EIGEN_STACK_ALLOCATION_LIMIT 0 +#define EIGEN_RUNTIME_NO_MALLOC + +#include "main.h" +#include +#include +#include + + +#define SVD_DEFAULT(M) BDCSVD +#define SVD_FOR_MIN_NORM(M) BDCSVD +#include "svd_common.h" + +// Check all variants of JacobiSVD +template +void bdcsvd(const MatrixType& a = MatrixType(), bool pickrandom = true) +{ + MatrixType m; + if(pickrandom) { + m.resizeLike(a); + svd_fill_random(m); + } + else + m = a; + + CALL_SUBTEST(( svd_test_all_computation_options >(m, false) )); +} + +template +void bdcsvd_method() +{ + enum { Size = MatrixType::RowsAtCompileTime }; + typedef typename MatrixType::RealScalar RealScalar; + typedef Matrix RealVecType; + MatrixType m = MatrixType::Identity(); + VERIFY_IS_APPROX(m.bdcSvd().singularValues(), RealVecType::Ones()); + VERIFY_RAISES_ASSERT(m.bdcSvd().matrixU()); + VERIFY_RAISES_ASSERT(m.bdcSvd().matrixV()); + + // Deprecated behavior. + VERIFY_IS_APPROX(m.bdcSvd(ComputeFullU|ComputeFullV).solve(m), m); + VERIFY_IS_APPROX(m.bdcSvd(ComputeFullU|ComputeFullV).transpose().solve(m), m); + VERIFY_IS_APPROX(m.bdcSvd(ComputeFullU|ComputeFullV).adjoint().solve(m), m); +} + +// Compare the Singular values returned with Jacobi and Bdc. +template +void compare_bdc_jacobi(const MatrixType& a = MatrixType(), unsigned int computationOptions = 0, int algoswap = 16, bool random = true) +{ + MatrixType m = random ? MatrixType::Random(a.rows(), a.cols()) : a; + + BDCSVD bdc_svd(m.rows(), m.cols(), computationOptions); + bdc_svd.setSwitchSize(algoswap); + bdc_svd.compute(m); + + JacobiSVD jacobi_svd(m); + VERIFY_IS_APPROX(bdc_svd.singularValues(), jacobi_svd.singularValues()); + + if(computationOptions & ComputeFullU) VERIFY_IS_APPROX(bdc_svd.matrixU(), jacobi_svd.matrixU()); + if(computationOptions & ComputeThinU) VERIFY_IS_APPROX(bdc_svd.matrixU(), jacobi_svd.matrixU()); + if(computationOptions & ComputeFullV) VERIFY_IS_APPROX(bdc_svd.matrixV(), jacobi_svd.matrixV()); + if(computationOptions & ComputeThinV) VERIFY_IS_APPROX(bdc_svd.matrixV(), jacobi_svd.matrixV()); +} + +// Verifies total deflation is **not** triggered. +void compare_bdc_jacobi_instance(bool structure_as_m, int algoswap = 16) +{ + MatrixXd m(4, 3); + if (structure_as_m) { + // The first 3 rows are the reduced form of Matrix 1 as shown below, and it + // has nonzero elements in the first column and diagonals only. + m << 1.056293, 0, 0, + -0.336468, 0.907359, 0, + -1.566245, 0, 0.149150, + -0.1, 0, 0; + } else { + // Matrix 1. + m << 0.882336, 18.3914, -26.7921, + -5.58135, 17.1931, -24.0892, + -20.794, 8.68496, -4.83103, + -8.4981, -10.5451, 23.9072; + } + compare_bdc_jacobi(m, 0, algoswap, false); +} + +EIGEN_DECLARE_TEST(bdcsvd) +{ + CALL_SUBTEST_1(( svd_verify_assert >(Matrix3f()) )); + CALL_SUBTEST_2(( svd_verify_assert >(Matrix4d()) )); + CALL_SUBTEST_3(( svd_verify_assert >(MatrixXf(10,12)) )); + CALL_SUBTEST_4(( svd_verify_assert >(MatrixXcd(7,5)) )); + + CALL_SUBTEST_5(( svd_all_trivial_2x2(bdcsvd) )); + CALL_SUBTEST_6(( svd_all_trivial_2x2(bdcsvd) )); + + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1(( bdcsvd() )); + CALL_SUBTEST_2(( bdcsvd() )); + CALL_SUBTEST_7(( bdcsvd >() )); + + int r = internal::random(1, EIGEN_TEST_MAX_SIZE/2), + c = internal::random(1, EIGEN_TEST_MAX_SIZE/2); + + TEST_SET_BUT_UNUSED_VARIABLE(r) + TEST_SET_BUT_UNUSED_VARIABLE(c) + + CALL_SUBTEST_8(( bdcsvd(Matrix(r,2)) )); + CALL_SUBTEST_9(( bdcsvd(MatrixXf(r,c)) )); + CALL_SUBTEST_10(( compare_bdc_jacobi(MatrixXf(r,c)) )); + CALL_SUBTEST_11(( bdcsvd(MatrixXd(r,c)) )); + CALL_SUBTEST_12(( compare_bdc_jacobi(MatrixXd(r,c)) )); + CALL_SUBTEST_13(( bdcsvd(MatrixXcd(r,c)) )); + CALL_SUBTEST_14(( compare_bdc_jacobi(MatrixXcd(r,c)) )); + + // Test on inf/nan matrix + CALL_SUBTEST_15( (svd_inf_nan, MatrixXf>()) ); + CALL_SUBTEST_16( (svd_inf_nan, MatrixXd>()) ); + } + + // test matrixbase method + CALL_SUBTEST_17(( bdcsvd_method() )); + CALL_SUBTEST_18(( bdcsvd_method() )); + + // Test problem size constructors + CALL_SUBTEST_19( BDCSVD(10,10) ); + + // Check that preallocation avoids subsequent mallocs + // Disabled because not supported by BDCSVD + // CALL_SUBTEST_9( svd_preallocate() ); + + CALL_SUBTEST_20( svd_underoverflow() ); + + // Without total deflation issues. + CALL_SUBTEST_21(( compare_bdc_jacobi_instance(true) )); + CALL_SUBTEST_22(( compare_bdc_jacobi_instance(false) )); + + // With total deflation issues before, when it shouldn't be triggered. + CALL_SUBTEST_23(( compare_bdc_jacobi_instance(true, 3) )); + CALL_SUBTEST_24(( compare_bdc_jacobi_instance(false, 3) )); +} + diff --git a/thirdparty/eigen/test/bfloat16_float.cpp b/thirdparty/eigen/test/bfloat16_float.cpp new file mode 100644 index 00000000..c3de0b19 --- /dev/null +++ b/thirdparty/eigen/test/bfloat16_float.cpp @@ -0,0 +1,378 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include +#include +#include + +#include "main.h" + +#include + +#define VERIFY_BFLOAT16_BITS_EQUAL(h, bits) \ + VERIFY_IS_EQUAL((numext::bit_cast(h)), (static_cast(bits))) + +// Make sure it's possible to forward declare Eigen::bfloat16 +namespace Eigen { +struct bfloat16; +} + +using Eigen::bfloat16; + +float BinaryToFloat(uint32_t sign, uint32_t exponent, uint32_t high_mantissa, + uint32_t low_mantissa) { + float dest; + uint32_t src = (sign << 31) + (exponent << 23) + (high_mantissa << 16) + low_mantissa; + memcpy(static_cast(&dest), + static_cast(&src), sizeof(dest)); + return dest; +} + +template + void test_roundtrip() { + // Representable T round trip via bfloat16 + VERIFY_IS_EQUAL((internal::cast(internal::cast(-std::numeric_limits::infinity()))), -std::numeric_limits::infinity()); + VERIFY_IS_EQUAL((internal::cast(internal::cast(std::numeric_limits::infinity()))), std::numeric_limits::infinity()); + VERIFY_IS_EQUAL((internal::cast(internal::cast(T(-1.0)))), T(-1.0)); + VERIFY_IS_EQUAL((internal::cast(internal::cast(T(-0.5)))), T(-0.5)); + VERIFY_IS_EQUAL((internal::cast(internal::cast(T(-0.0)))), T(-0.0)); + VERIFY_IS_EQUAL((internal::cast(internal::cast(T(1.0)))), T(1.0)); + VERIFY_IS_EQUAL((internal::cast(internal::cast(T(0.5)))), T(0.5)); + VERIFY_IS_EQUAL((internal::cast(internal::cast(T(0.0)))), T(0.0)); +} + +void test_conversion() +{ + using Eigen::bfloat16_impl::__bfloat16_raw; + + // Round-trip casts + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(bfloat16(1.0f))), + bfloat16(1.0f)); + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(bfloat16(0.5f))), + bfloat16(0.5f)); + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(bfloat16(-0.33333f))), + bfloat16(-0.33333f)); + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(bfloat16(0.0f))), + bfloat16(0.0f)); + + // Conversion from float. + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(1.0f), 0x3f80); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.5f), 0x3f00); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.33333f), 0x3eab); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(3.38e38f), 0x7f7e); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(3.40e38f), 0x7f80); // Becomes infinity. + + // Verify round-to-nearest-even behavior. + float val1 = static_cast(bfloat16(__bfloat16_raw(0x3c00))); + float val2 = static_cast(bfloat16(__bfloat16_raw(0x3c01))); + float val3 = static_cast(bfloat16(__bfloat16_raw(0x3c02))); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.5f * (val1 + val2)), 0x3c00); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.5f * (val2 + val3)), 0x3c02); + + // Conversion from int. + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(-1), 0xbf80); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0), 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(1), 0x3f80); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(2), 0x4000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(3), 0x4040); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(12), 0x4140); + + // Conversion from bool. + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(false), 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(true), 0x3f80); + + // Conversion to bool + VERIFY_IS_EQUAL(static_cast(bfloat16(3)), true); + VERIFY_IS_EQUAL(static_cast(bfloat16(0.33333f)), true); + VERIFY_IS_EQUAL(bfloat16(-0.0), false); + VERIFY_IS_EQUAL(static_cast(bfloat16(0.0)), false); + + // Explicit conversion to float. + VERIFY_IS_EQUAL(static_cast(bfloat16(__bfloat16_raw(0x0000))), 0.0f); + VERIFY_IS_EQUAL(static_cast(bfloat16(__bfloat16_raw(0x3f80))), 1.0f); + + // Implicit conversion to float + VERIFY_IS_EQUAL(bfloat16(__bfloat16_raw(0x0000)), 0.0f); + VERIFY_IS_EQUAL(bfloat16(__bfloat16_raw(0x3f80)), 1.0f); + + // Zero representations + VERIFY_IS_EQUAL(bfloat16(0.0f), bfloat16(0.0f)); + VERIFY_IS_EQUAL(bfloat16(-0.0f), bfloat16(0.0f)); + VERIFY_IS_EQUAL(bfloat16(-0.0f), bfloat16(-0.0f)); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(0.0f), 0x0000); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(-0.0f), 0x8000); + + // Default is zero + VERIFY_IS_EQUAL(static_cast(bfloat16()), 0.0f); + + // Representable floats round trip via bfloat16 + test_roundtrip(); + test_roundtrip(); + test_roundtrip >(); + test_roundtrip >(); + + // Conversion + Array a; + for (int i = 0; i < 100; i++) a(i) = i + 1.25; + Array b = a.cast(); + Array c = b.cast(); + for (int i = 0; i < 100; ++i) { + VERIFY_LE(numext::abs(c(i) - a(i)), a(i) / 128); + } + + // Epsilon + VERIFY_LE(1.0f, static_cast((std::numeric_limits::epsilon)() + bfloat16(1.0f))); + VERIFY_IS_EQUAL(1.0f, static_cast((std::numeric_limits::epsilon)() / bfloat16(2.0f) + bfloat16(1.0f))); + + // Negate + VERIFY_IS_EQUAL(static_cast(-bfloat16(3.0f)), -3.0f); + VERIFY_IS_EQUAL(static_cast(-bfloat16(-4.5f)), 4.5f); + + +#if !EIGEN_COMP_MSVC + // Visual Studio errors out on divisions by 0 + VERIFY((numext::isnan)(static_cast(bfloat16(0.0 / 0.0)))); + VERIFY((numext::isinf)(static_cast(bfloat16(1.0 / 0.0)))); + VERIFY((numext::isinf)(static_cast(bfloat16(-1.0 / 0.0)))); + + // Visual Studio errors out on divisions by 0 + VERIFY((numext::isnan)(bfloat16(0.0 / 0.0))); + VERIFY((numext::isinf)(bfloat16(1.0 / 0.0))); + VERIFY((numext::isinf)(bfloat16(-1.0 / 0.0))); +#endif + + // NaNs and infinities. + VERIFY(!(numext::isinf)(static_cast(bfloat16(3.38e38f)))); // Largest finite number. + VERIFY(!(numext::isnan)(static_cast(bfloat16(0.0f)))); + VERIFY((numext::isinf)(static_cast(bfloat16(__bfloat16_raw(0xff80))))); + VERIFY((numext::isnan)(static_cast(bfloat16(__bfloat16_raw(0xffc0))))); + VERIFY((numext::isinf)(static_cast(bfloat16(__bfloat16_raw(0x7f80))))); + VERIFY((numext::isnan)(static_cast(bfloat16(__bfloat16_raw(0x7fc0))))); + + // Exactly same checks as above, just directly on the bfloat16 representation. + VERIFY(!(numext::isinf)(bfloat16(__bfloat16_raw(0x7bff)))); + VERIFY(!(numext::isnan)(bfloat16(__bfloat16_raw(0x0000)))); + VERIFY((numext::isinf)(bfloat16(__bfloat16_raw(0xff80)))); + VERIFY((numext::isnan)(bfloat16(__bfloat16_raw(0xffc0)))); + VERIFY((numext::isinf)(bfloat16(__bfloat16_raw(0x7f80)))); + VERIFY((numext::isnan)(bfloat16(__bfloat16_raw(0x7fc0)))); + + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(BinaryToFloat(0x0, 0xff, 0x40, 0x0)), 0x7fc0); + VERIFY_BFLOAT16_BITS_EQUAL(bfloat16(BinaryToFloat(0x1, 0xff, 0x40, 0x0)), 0xffc0); +} + +void test_numtraits() +{ + std::cout << "epsilon = " << NumTraits::epsilon() << " (0x" << std::hex << numext::bit_cast(NumTraits::epsilon()) << ")" << std::endl; + std::cout << "highest = " << NumTraits::highest() << " (0x" << std::hex << numext::bit_cast(NumTraits::highest()) << ")" << std::endl; + std::cout << "lowest = " << NumTraits::lowest() << " (0x" << std::hex << numext::bit_cast(NumTraits::lowest()) << ")" << std::endl; + std::cout << "min = " << (std::numeric_limits::min)() << " (0x" << std::hex << numext::bit_cast((std::numeric_limits::min)()) << ")" << std::endl; + std::cout << "denorm min = " << (std::numeric_limits::denorm_min)() << " (0x" << std::hex << numext::bit_cast((std::numeric_limits::denorm_min)()) << ")" << std::endl; + std::cout << "infinity = " << NumTraits::infinity() << " (0x" << std::hex << numext::bit_cast(NumTraits::infinity()) << ")" << std::endl; + std::cout << "quiet nan = " << NumTraits::quiet_NaN() << " (0x" << std::hex << numext::bit_cast(NumTraits::quiet_NaN()) << ")" << std::endl; + std::cout << "signaling nan = " << std::numeric_limits::signaling_NaN() << " (0x" << std::hex << numext::bit_cast(std::numeric_limits::signaling_NaN()) << ")" << std::endl; + + VERIFY(NumTraits::IsSigned); + + VERIFY_IS_EQUAL( + numext::bit_cast(std::numeric_limits::infinity()), + numext::bit_cast(bfloat16(std::numeric_limits::infinity())) ); + // There is no guarantee that casting a 32-bit NaN to bfloat16 has a precise + // bit pattern. We test that it is in fact a NaN, then test the signaling + // bit (msb of significand is 1 for quiet, 0 for signaling). + const numext::uint16_t BFLOAT16_QUIET_BIT = 0x0040; + VERIFY( + (numext::isnan)(std::numeric_limits::quiet_NaN()) + && (numext::isnan)(bfloat16(std::numeric_limits::quiet_NaN())) + && ((numext::bit_cast(std::numeric_limits::quiet_NaN()) & BFLOAT16_QUIET_BIT) > 0) + && ((numext::bit_cast(bfloat16(std::numeric_limits::quiet_NaN())) & BFLOAT16_QUIET_BIT) > 0) ); + // After a cast to bfloat16, a signaling NaN may become non-signaling. Thus, + // we check that both are NaN, and that only the `numeric_limits` version is + // signaling. + VERIFY( + (numext::isnan)(std::numeric_limits::signaling_NaN()) + && (numext::isnan)(bfloat16(std::numeric_limits::signaling_NaN())) + && ((numext::bit_cast(std::numeric_limits::signaling_NaN()) & BFLOAT16_QUIET_BIT) == 0) ); + + VERIFY( (std::numeric_limits::min)() > bfloat16(0.f) ); + VERIFY( (std::numeric_limits::denorm_min)() > bfloat16(0.f) ); + VERIFY_IS_EQUAL( (std::numeric_limits::denorm_min)()/bfloat16(2), bfloat16(0.f) ); +} + +void test_arithmetic() +{ + VERIFY_IS_EQUAL(static_cast(bfloat16(2) + bfloat16(2)), 4); + VERIFY_IS_EQUAL(static_cast(bfloat16(2) + bfloat16(-2)), 0); + VERIFY_IS_APPROX(static_cast(bfloat16(0.33333f) + bfloat16(0.66667f)), 1.0f); + VERIFY_IS_EQUAL(static_cast(bfloat16(2.0f) * bfloat16(-5.5f)), -11.0f); + VERIFY_IS_APPROX(static_cast(bfloat16(1.0f) / bfloat16(3.0f)), 0.3339f); + VERIFY_IS_EQUAL(static_cast(-bfloat16(4096.0f)), -4096.0f); + VERIFY_IS_EQUAL(static_cast(-bfloat16(-4096.0f)), 4096.0f); +} + +void test_comparison() +{ + VERIFY(bfloat16(1.0f) > bfloat16(0.5f)); + VERIFY(bfloat16(0.5f) < bfloat16(1.0f)); + VERIFY(!(bfloat16(1.0f) < bfloat16(0.5f))); + VERIFY(!(bfloat16(0.5f) > bfloat16(1.0f))); + + VERIFY(!(bfloat16(4.0f) > bfloat16(4.0f))); + VERIFY(!(bfloat16(4.0f) < bfloat16(4.0f))); + + VERIFY(!(bfloat16(0.0f) < bfloat16(-0.0f))); + VERIFY(!(bfloat16(-0.0f) < bfloat16(0.0f))); + VERIFY(!(bfloat16(0.0f) > bfloat16(-0.0f))); + VERIFY(!(bfloat16(-0.0f) > bfloat16(0.0f))); + + VERIFY(bfloat16(0.2f) > bfloat16(-1.0f)); + VERIFY(bfloat16(-1.0f) < bfloat16(0.2f)); + VERIFY(bfloat16(-16.0f) < bfloat16(-15.0f)); + + VERIFY(bfloat16(1.0f) == bfloat16(1.0f)); + VERIFY(bfloat16(1.0f) != bfloat16(2.0f)); + + // Comparisons with NaNs and infinities. +#if !EIGEN_COMP_MSVC + // Visual Studio errors out on divisions by 0 + VERIFY(!(bfloat16(0.0 / 0.0) == bfloat16(0.0 / 0.0))); + VERIFY(bfloat16(0.0 / 0.0) != bfloat16(0.0 / 0.0)); + + VERIFY(!(bfloat16(1.0) == bfloat16(0.0 / 0.0))); + VERIFY(!(bfloat16(1.0) < bfloat16(0.0 / 0.0))); + VERIFY(!(bfloat16(1.0) > bfloat16(0.0 / 0.0))); + VERIFY(bfloat16(1.0) != bfloat16(0.0 / 0.0)); + + VERIFY(bfloat16(1.0) < bfloat16(1.0 / 0.0)); + VERIFY(bfloat16(1.0) > bfloat16(-1.0 / 0.0)); +#endif +} + +void test_basic_functions() +{ + VERIFY_IS_EQUAL(static_cast(numext::abs(bfloat16(3.5f))), 3.5f); + VERIFY_IS_EQUAL(static_cast(abs(bfloat16(3.5f))), 3.5f); + VERIFY_IS_EQUAL(static_cast(numext::abs(bfloat16(-3.5f))), 3.5f); + VERIFY_IS_EQUAL(static_cast(abs(bfloat16(-3.5f))), 3.5f); + + VERIFY_IS_EQUAL(static_cast(numext::floor(bfloat16(3.5f))), 3.0f); + VERIFY_IS_EQUAL(static_cast(floor(bfloat16(3.5f))), 3.0f); + VERIFY_IS_EQUAL(static_cast(numext::floor(bfloat16(-3.5f))), -4.0f); + VERIFY_IS_EQUAL(static_cast(floor(bfloat16(-3.5f))), -4.0f); + + VERIFY_IS_EQUAL(static_cast(numext::ceil(bfloat16(3.5f))), 4.0f); + VERIFY_IS_EQUAL(static_cast(ceil(bfloat16(3.5f))), 4.0f); + VERIFY_IS_EQUAL(static_cast(numext::ceil(bfloat16(-3.5f))), -3.0f); + VERIFY_IS_EQUAL(static_cast(ceil(bfloat16(-3.5f))), -3.0f); + + VERIFY_IS_APPROX(static_cast(numext::sqrt(bfloat16(0.0f))), 0.0f); + VERIFY_IS_APPROX(static_cast(sqrt(bfloat16(0.0f))), 0.0f); + VERIFY_IS_APPROX(static_cast(numext::sqrt(bfloat16(4.0f))), 2.0f); + VERIFY_IS_APPROX(static_cast(sqrt(bfloat16(4.0f))), 2.0f); + + VERIFY_IS_APPROX(static_cast(numext::pow(bfloat16(0.0f), bfloat16(1.0f))), 0.0f); + VERIFY_IS_APPROX(static_cast(pow(bfloat16(0.0f), bfloat16(1.0f))), 0.0f); + VERIFY_IS_APPROX(static_cast(numext::pow(bfloat16(2.0f), bfloat16(2.0f))), 4.0f); + VERIFY_IS_APPROX(static_cast(pow(bfloat16(2.0f), bfloat16(2.0f))), 4.0f); + + VERIFY_IS_EQUAL(static_cast(numext::exp(bfloat16(0.0f))), 1.0f); + VERIFY_IS_EQUAL(static_cast(exp(bfloat16(0.0f))), 1.0f); + VERIFY_IS_APPROX(static_cast(numext::exp(bfloat16(EIGEN_PI))), 20.f + static_cast(EIGEN_PI)); + VERIFY_IS_APPROX(static_cast(exp(bfloat16(EIGEN_PI))), 20.f + static_cast(EIGEN_PI)); + + VERIFY_IS_EQUAL(static_cast(numext::expm1(bfloat16(0.0f))), 0.0f); + VERIFY_IS_EQUAL(static_cast(expm1(bfloat16(0.0f))), 0.0f); + VERIFY_IS_APPROX(static_cast(numext::expm1(bfloat16(2.0f))), 6.375f); + VERIFY_IS_APPROX(static_cast(expm1(bfloat16(2.0f))), 6.375f); + + VERIFY_IS_EQUAL(static_cast(numext::log(bfloat16(1.0f))), 0.0f); + VERIFY_IS_EQUAL(static_cast(log(bfloat16(1.0f))), 0.0f); + VERIFY_IS_APPROX(static_cast(numext::log(bfloat16(10.0f))), 2.296875f); + VERIFY_IS_APPROX(static_cast(log(bfloat16(10.0f))), 2.296875f); + + VERIFY_IS_EQUAL(static_cast(numext::log1p(bfloat16(0.0f))), 0.0f); + VERIFY_IS_EQUAL(static_cast(log1p(bfloat16(0.0f))), 0.0f); + VERIFY_IS_APPROX(static_cast(numext::log1p(bfloat16(10.0f))), 2.390625f); + VERIFY_IS_APPROX(static_cast(log1p(bfloat16(10.0f))), 2.390625f); +} + +void test_trigonometric_functions() +{ + VERIFY_IS_APPROX(numext::cos(bfloat16(0.0f)), bfloat16(cosf(0.0f))); + VERIFY_IS_APPROX(cos(bfloat16(0.0f)), bfloat16(cosf(0.0f))); + VERIFY_IS_APPROX(numext::cos(bfloat16(EIGEN_PI)), bfloat16(cosf(EIGEN_PI))); + // VERIFY_IS_APPROX(numext::cos(bfloat16(EIGEN_PI/2)), bfloat16(cosf(EIGEN_PI/2))); + // VERIFY_IS_APPROX(numext::cos(bfloat16(3*EIGEN_PI/2)), bfloat16(cosf(3*EIGEN_PI/2))); + VERIFY_IS_APPROX(numext::cos(bfloat16(3.5f)), bfloat16(cosf(3.5f))); + + VERIFY_IS_APPROX(numext::sin(bfloat16(0.0f)), bfloat16(sinf(0.0f))); + VERIFY_IS_APPROX(sin(bfloat16(0.0f)), bfloat16(sinf(0.0f))); + // VERIFY_IS_APPROX(numext::sin(bfloat16(EIGEN_PI)), bfloat16(sinf(EIGEN_PI))); + VERIFY_IS_APPROX(numext::sin(bfloat16(EIGEN_PI/2)), bfloat16(sinf(EIGEN_PI/2))); + VERIFY_IS_APPROX(numext::sin(bfloat16(3*EIGEN_PI/2)), bfloat16(sinf(3*EIGEN_PI/2))); + VERIFY_IS_APPROX(numext::sin(bfloat16(3.5f)), bfloat16(sinf(3.5f))); + + VERIFY_IS_APPROX(numext::tan(bfloat16(0.0f)), bfloat16(tanf(0.0f))); + VERIFY_IS_APPROX(tan(bfloat16(0.0f)), bfloat16(tanf(0.0f))); + // VERIFY_IS_APPROX(numext::tan(bfloat16(EIGEN_PI)), bfloat16(tanf(EIGEN_PI))); + // VERIFY_IS_APPROX(numext::tan(bfloat16(EIGEN_PI/2)), bfloat16(tanf(EIGEN_PI/2))); + // VERIFY_IS_APPROX(numext::tan(bfloat16(3*EIGEN_PI/2)), bfloat16(tanf(3*EIGEN_PI/2))); + VERIFY_IS_APPROX(numext::tan(bfloat16(3.5f)), bfloat16(tanf(3.5f))); +} + +void test_array() +{ + typedef Array ArrayXh; + Index size = internal::random(1,10); + Index i = internal::random(0,size-1); + ArrayXh a1 = ArrayXh::Random(size), a2 = ArrayXh::Random(size); + VERIFY_IS_APPROX( a1+a1, bfloat16(2)*a1 ); + VERIFY( (a1.abs() >= bfloat16(0)).all() ); + VERIFY_IS_APPROX( (a1*a1).sqrt(), a1.abs() ); + + VERIFY( ((a1.min)(a2) <= (a1.max)(a2)).all() ); + a1(i) = bfloat16(-10.); + VERIFY_IS_EQUAL( a1.minCoeff(), bfloat16(-10.) ); + a1(i) = bfloat16(10.); + VERIFY_IS_EQUAL( a1.maxCoeff(), bfloat16(10.) ); + + std::stringstream ss; + ss << a1; +} + +void test_product() +{ + typedef Matrix MatrixXh; + Index rows = internal::random(1,EIGEN_TEST_MAX_SIZE); + Index cols = internal::random(1,EIGEN_TEST_MAX_SIZE); + Index depth = internal::random(1,EIGEN_TEST_MAX_SIZE); + MatrixXh Ah = MatrixXh::Random(rows,depth); + MatrixXh Bh = MatrixXh::Random(depth,cols); + MatrixXh Ch = MatrixXh::Random(rows,cols); + MatrixXf Af = Ah.cast(); + MatrixXf Bf = Bh.cast(); + MatrixXf Cf = Ch.cast(); + VERIFY_IS_APPROX(Ch.noalias()+=Ah*Bh, (Cf.noalias()+=Af*Bf).cast()); +} + +EIGEN_DECLARE_TEST(bfloat16_float) +{ + CALL_SUBTEST(test_numtraits()); + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST(test_conversion()); + CALL_SUBTEST(test_arithmetic()); + CALL_SUBTEST(test_comparison()); + CALL_SUBTEST(test_basic_functions()); + CALL_SUBTEST(test_trigonometric_functions()); + CALL_SUBTEST(test_array()); + CALL_SUBTEST(test_product()); + } +} diff --git a/thirdparty/eigen/test/bicgstab.cpp b/thirdparty/eigen/test/bicgstab.cpp index f327e2fa..59c4b501 100644 --- a/thirdparty/eigen/test/bicgstab.cpp +++ b/thirdparty/eigen/test/bicgstab.cpp @@ -10,21 +10,25 @@ #include "sparse_solver.h" #include -template void test_bicgstab_T() +template void test_bicgstab_T() { - BiCGSTAB, DiagonalPreconditioner > bicgstab_colmajor_diag; - BiCGSTAB, IdentityPreconditioner > bicgstab_colmajor_I; - BiCGSTAB, IncompleteLUT > bicgstab_colmajor_ilut; + BiCGSTAB, DiagonalPreconditioner > bicgstab_colmajor_diag; + BiCGSTAB, IdentityPreconditioner > bicgstab_colmajor_I; + BiCGSTAB, IncompleteLUT > bicgstab_colmajor_ilut; //BiCGSTAB, SSORPreconditioner > bicgstab_colmajor_ssor; + bicgstab_colmajor_diag.setTolerance(NumTraits::epsilon()*4); + bicgstab_colmajor_ilut.setTolerance(NumTraits::epsilon()*4); + CALL_SUBTEST( check_sparse_square_solving(bicgstab_colmajor_diag) ); // CALL_SUBTEST( check_sparse_square_solving(bicgstab_colmajor_I) ); CALL_SUBTEST( check_sparse_square_solving(bicgstab_colmajor_ilut) ); //CALL_SUBTEST( check_sparse_square_solving(bicgstab_colmajor_ssor) ); } -void test_bicgstab() +EIGEN_DECLARE_TEST(bicgstab) { - CALL_SUBTEST_1(test_bicgstab_T()); - CALL_SUBTEST_2(test_bicgstab_T >()); + CALL_SUBTEST_1((test_bicgstab_T()) ); + CALL_SUBTEST_2((test_bicgstab_T, int>())); + CALL_SUBTEST_3((test_bicgstab_T())); } diff --git a/thirdparty/eigen/test/blasutil.cpp b/thirdparty/eigen/test/blasutil.cpp new file mode 100644 index 00000000..845a498d --- /dev/null +++ b/thirdparty/eigen/test/blasutil.cpp @@ -0,0 +1,210 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2020 Everton Constantino +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/ + +#include "main.h" + +// Disable "ignoring attributes on template argument" +// for packet_traits +// => The only workaround would be to wrap _m128 and the likes +// within wrappers. +#if EIGEN_GNUC_AT_LEAST(6,0) + #pragma GCC diagnostic ignored "-Wignored-attributes" +#endif + +#define GET(i,j) (StorageOrder == RowMajor ? (i)*stride + (j) : (i) + (j)*stride) +#define SCATTER(i,j,k) (StorageOrder == RowMajor ? ((i)+(k))*stride + (j) : (i) + ((j)+(k))*stride) + +template +void compare(const Packet& a, const Packet& b) +{ + int pktsz = internal::packet_traits::size; + Scalar *buffA = new Scalar[pktsz]; + Scalar *buffB = new Scalar[pktsz]; + + internal::pstoreu(buffA, a); + internal::pstoreu(buffB, b); + + for(int i = 0; i < pktsz; i++) + { + VERIFY_IS_EQUAL(buffA[i], buffB[i]); + } + + delete[] buffA; + delete[] buffB; +} + +template +struct PacketBlockSet +{ + typedef typename internal::packet_traits::type Packet; + + void setPacketBlock(internal::PacketBlock& block, Scalar value) + { + for(int idx = 0; idx < n; idx++) + { + block.packet[idx] = internal::pset1(value); + } + } + + void comparePacketBlock(Scalar *data, int i, int j, int stride, internal::PacketBlock& block) + { + for(int idx = 0; idx < n; idx++) + { + Packet line = internal::ploadu(data + SCATTER(i,j,idx)); + compare(block.packet[idx], line); + } + } +}; + +template +void run_bdmp_spec_1() +{ + typedef internal::blas_data_mapper BlasDataMapper; + int packetSize = internal::packet_traits::size; + int minSize = std::max(packetSize, BlockSize); + typedef typename internal::packet_traits::type Packet; + + int szm = internal::random(minSize,500), szn = internal::random(minSize,500); + int stride = StorageOrder == RowMajor ? szn : szm; + Scalar *d = new Scalar[szn*szm]; + + // Initializing with random entries + for(int i = 0; i < szm*szn; i++) + { + d[i] = internal::random(static_cast(3), static_cast(10)); + } + + BlasDataMapper bdm(d, stride); + + // Testing operator() + for(int i = 0; i < szm; i++) + { + for(int j = 0; j < szn; j++) + { + VERIFY_IS_EQUAL(d[GET(i,j)], bdm(i,j)); + } + } + + // Testing getSubMapper and getLinearMapper + int i0 = internal::random(0,szm-2); + int j0 = internal::random(0,szn-2); + for(int i = i0; i < szm; i++) + { + for(int j = j0; j < szn; j++) + { + const BlasDataMapper& bdmSM = bdm.getSubMapper(i0,j0); + const internal::BlasLinearMapper& bdmLM = bdm.getLinearMapper(i0,j0); + + Scalar v = bdmSM(i - i0, j - j0); + Scalar vd = d[GET(i,j)]; + VERIFY_IS_EQUAL(vd, v); + VERIFY_IS_EQUAL(vd, bdmLM(GET(i-i0, j-j0))); + } + } + + // Testing loadPacket + for(int i = 0; i < szm - minSize; i++) + { + for(int j = 0; j < szn - minSize; j++) + { + Packet pktBDM = bdm.template loadPacket(i,j); + Packet pktD = internal::ploadu(d + GET(i,j)); + + compare(pktBDM, pktD); + } + } + + // Testing gatherPacket + Scalar *buff = new Scalar[packetSize]; + for(int i = 0; i < szm - minSize; i++) + { + for(int j = 0; j < szn - minSize; j++) + { + Packet p = bdm.template gatherPacket(i,j); + internal::pstoreu(buff, p); + + for(int k = 0; k < packetSize; k++) + { + VERIFY_IS_EQUAL(d[SCATTER(i,j,k)], buff[k]); + } + + } + } + delete[] buff; + + // Testing scatterPacket + for(int i = 0; i < szm - minSize; i++) + { + for(int j = 0; j < szn - minSize; j++) + { + Packet p = internal::pset1(static_cast(1)); + bdm.template scatterPacket(i,j,p); + for(int k = 0; k < packetSize; k++) + { + VERIFY_IS_EQUAL(d[SCATTER(i,j,k)], static_cast(1)); + } + } + } + + //Testing storePacketBlock + internal::PacketBlock block; + + PacketBlockSet pbs; + pbs.setPacketBlock(block, static_cast(2)); + + for(int i = 0; i < szm - minSize; i++) + { + for(int j = 0; j < szn - minSize; j++) + { + bdm.template storePacketBlock(i, j, block); + + pbs.comparePacketBlock(d, i, j, stride, block); + } + } + + delete[] d; +} + +template +void run_test() +{ + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); + run_bdmp_spec_1(); +} + +EIGEN_DECLARE_TEST(blasutil) +{ + for(int i = 0; i < g_repeat; i++) + { + CALL_SUBTEST_1(run_test()); + CALL_SUBTEST_2(run_test()); + CALL_SUBTEST_3(run_test()); + +// TODO: Replace this by a call to numext::int64_t as soon as we have a way to +// detect the typedef for int64_t on all platforms +#if EIGEN_HAS_CXX11 + CALL_SUBTEST_4(run_test()); +#else + CALL_SUBTEST_4(run_test()); +#endif + + CALL_SUBTEST_5(run_test()); + CALL_SUBTEST_6(run_test()); + CALL_SUBTEST_7(run_test >()); + CALL_SUBTEST_8(run_test >()); + } +} diff --git a/thirdparty/eigen/test/block.cpp b/thirdparty/eigen/test/block.cpp index 9ed5d7bc..667a3be3 100644 --- a/thirdparty/eigen/test/block.cpp +++ b/thirdparty/eigen/test/block.cpp @@ -29,15 +29,21 @@ block_real_only(const MatrixType &, Index, Index, Index, Index, const Scalar&) { return Scalar(0); } +// Check at compile-time that T1==T2, and at runtime-time that a==b +template +typename internal::enable_if::value,bool>::type +is_same_block(const T1& a, const T2& b) +{ + return a.isApprox(b); +} template void block(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef typename MatrixType::RealScalar RealScalar; typedef Matrix VectorType; typedef Matrix RowVectorType; - typedef Matrix DynamicMatrixType; + typedef Matrix DynamicMatrixType; typedef Matrix DynamicVectorType; Index rows = m.rows(); @@ -87,10 +93,9 @@ template void block(const MatrixType& m) m1.block(r1,c1,r2-r1+1,c2-c1+1) = s1 * m2.block(0, 0, r2-r1+1,c2-c1+1); m1.block(r1,c1,r2-r1+1,c2-c1+1)(r2-r1,c2-c1) = m2.block(0, 0, r2-r1+1,c2-c1+1)(0,0); - enum { - BlockRows = 2, - BlockCols = 5 - }; + const Index BlockRows = 2; + const Index BlockCols = 5; + if (rows>=5 && cols>=8) { // test fixed block() as lvalue @@ -106,6 +111,11 @@ template void block(const MatrixType& m) m1.template block(1,1,BlockRows,BlockCols)(0,3) = m1.template block<2,5>(1,1)(1,2); Matrix b2 = m1.template block(3,3,2,5); VERIFY_IS_EQUAL(b2, m1.block(3,3,BlockRows,BlockCols)); + + VERIFY(is_same_block(m1.block(3,3,BlockRows,BlockCols), m1.block(3,3,fix(BlockRows),fix(BlockCols)))); + VERIFY(is_same_block(m1.template block(1,1,BlockRows,BlockCols), m1.block(1,1,fix,BlockCols))); + VERIFY(is_same_block(m1.template block(1,1,BlockRows,BlockCols), m1.block(1,1,fix(),fix))); + VERIFY(is_same_block(m1.template block(1,1,BlockRows,BlockCols), m1.block(1,1,fix,fix(BlockCols)))); } if (rows>2) @@ -130,6 +140,15 @@ template void block(const MatrixType& m) VERIFY(numext::real(ones.col(c1).dot(ones.col(c2))) == RealScalar(rows)); VERIFY(numext::real(ones.row(r1).dot(ones.row(r2))) == RealScalar(cols)); + + // check that linear acccessors works on blocks + m1 = m1_copy; + if (c1 > 0 && r1 > 0) { + if ((MatrixType::Flags & RowMajorBit) == 0) + VERIFY_IS_EQUAL(m1.leftCols(c1).coeff(r1 + c1 * rows), m1(r1, c1)); + else + VERIFY_IS_EQUAL(m1.topRows(r1).coeff(c1 + r1 * cols), m1(r1, c1)); + } // now test some block-inside-of-block. @@ -141,11 +160,27 @@ template void block(const MatrixType& m) VERIFY_IS_EQUAL( (m1.transpose().block(c1,r1,c2-c1+1,r2-r1+1).col(0)) , (m1.row(r1).segment(c1,c2-c1+1)).transpose() ); // expressions without direct access - VERIFY_IS_EQUAL( ((m1+m2).block(r1,c1,rows-r1,cols-c1).block(r2-r1,c2-c1,rows-r2,cols-c2)) , ((m1+m2).block(r2,c2,rows-r2,cols-c2)) ); - VERIFY_IS_EQUAL( ((m1+m2).block(r1,c1,r2-r1+1,c2-c1+1).row(0)) , ((m1+m2).row(r1).segment(c1,c2-c1+1)) ); - VERIFY_IS_EQUAL( ((m1+m2).block(r1,c1,r2-r1+1,c2-c1+1).col(0)) , ((m1+m2).col(c1).segment(r1,r2-r1+1)) ); - VERIFY_IS_EQUAL( ((m1+m2).block(r1,c1,r2-r1+1,c2-c1+1).transpose().col(0)) , ((m1+m2).row(r1).segment(c1,c2-c1+1)).transpose() ); - VERIFY_IS_EQUAL( ((m1+m2).transpose().block(c1,r1,c2-c1+1,r2-r1+1).col(0)) , ((m1+m2).row(r1).segment(c1,c2-c1+1)).transpose() ); + VERIFY_IS_APPROX( ((m1+m2).block(r1,c1,rows-r1,cols-c1).block(r2-r1,c2-c1,rows-r2,cols-c2)) , ((m1+m2).block(r2,c2,rows-r2,cols-c2)) ); + VERIFY_IS_APPROX( ((m1+m2).block(r1,c1,r2-r1+1,c2-c1+1).row(0)) , ((m1+m2).row(r1).segment(c1,c2-c1+1)) ); + VERIFY_IS_APPROX( ((m1+m2).block(r1,c1,r2-r1+1,c2-c1+1).row(0)) , ((m1+m2).eval().row(r1).segment(c1,c2-c1+1)) ); + VERIFY_IS_APPROX( ((m1+m2).block(r1,c1,r2-r1+1,c2-c1+1).col(0)) , ((m1+m2).col(c1).segment(r1,r2-r1+1)) ); + VERIFY_IS_APPROX( ((m1+m2).block(r1,c1,r2-r1+1,c2-c1+1).transpose().col(0)) , ((m1+m2).row(r1).segment(c1,c2-c1+1)).transpose() ); + VERIFY_IS_APPROX( ((m1+m2).transpose().block(c1,r1,c2-c1+1,r2-r1+1).col(0)) , ((m1+m2).row(r1).segment(c1,c2-c1+1)).transpose() ); + VERIFY_IS_APPROX( ((m1+m2).template block(r1,c1,r2-r1+1,1)) , ((m1+m2).eval().col(c1).eval().segment(r1,r2-r1+1)) ); + VERIFY_IS_APPROX( ((m1+m2).template block<1,Dynamic>(r1,c1,1,c2-c1+1)) , ((m1+m2).eval().row(r1).eval().segment(c1,c2-c1+1)) ); + VERIFY_IS_APPROX( ((m1+m2).transpose().template block<1,Dynamic>(c1,r1,1,r2-r1+1)) , ((m1+m2).eval().col(c1).eval().segment(r1,r2-r1+1)).transpose() ); + VERIFY_IS_APPROX( (m1+m2).row(r1).eval(), (m1+m2).eval().row(r1) ); + VERIFY_IS_APPROX( (m1+m2).adjoint().col(r1).eval(), (m1+m2).adjoint().eval().col(r1) ); + VERIFY_IS_APPROX( (m1+m2).adjoint().row(c1).eval(), (m1+m2).adjoint().eval().row(c1) ); + VERIFY_IS_APPROX( (m1*1).row(r1).segment(c1,c2-c1+1).eval(), m1.row(r1).eval().segment(c1,c2-c1+1).eval() ); + VERIFY_IS_APPROX( m1.col(c1).reverse().segment(r1,r2-r1+1).eval(),m1.col(c1).reverse().eval().segment(r1,r2-r1+1).eval() ); + + VERIFY_IS_APPROX( (m1*1).topRows(r1), m1.topRows(r1) ); + VERIFY_IS_APPROX( (m1*1).leftCols(c1), m1.leftCols(c1) ); + VERIFY_IS_APPROX( (m1*1).transpose().topRows(c1), m1.transpose().topRows(c1) ); + VERIFY_IS_APPROX( (m1*1).transpose().leftCols(r1), m1.transpose().leftCols(r1) ); + VERIFY_IS_APPROX( (m1*1).transpose().middleRows(c1,c2-c1+1), m1.transpose().middleRows(c1,c2-c1+1) ); + VERIFY_IS_APPROX( (m1*1).transpose().middleCols(r1,r2-r1+1), m1.transpose().middleCols(r1,r2-r1+1) ); // evaluation into plain matrices from expressions with direct access (stress MapBase) DynamicMatrixType dm; @@ -173,13 +208,42 @@ template void block(const MatrixType& m) dm = m1.row(r1).segment(c1,c2-c1+1).transpose(); dv = m1.transpose().block(c1,r1,c2-c1+1,r2-r1+1).col(0); VERIFY_IS_EQUAL(dv, dm); + + VERIFY_IS_EQUAL( (m1.template block(1,0,0,1)), m1.block(1,0,0,1)); + VERIFY_IS_EQUAL( (m1.template block<1,Dynamic>(0,1,1,0)), m1.block(0,1,1,0)); + VERIFY_IS_EQUAL( ((m1*1).template block(1,0,0,1)), m1.block(1,0,0,1)); + VERIFY_IS_EQUAL( ((m1*1).template block<1,Dynamic>(0,1,1,0)), m1.block(0,1,1,0)); + + if (rows>=2 && cols>=2) + { + VERIFY_RAISES_ASSERT( m1 += m1.col(0) ); + VERIFY_RAISES_ASSERT( m1 -= m1.col(0) ); + VERIFY_RAISES_ASSERT( m1.array() *= m1.col(0).array() ); + VERIFY_RAISES_ASSERT( m1.array() /= m1.col(0).array() ); + } + + VERIFY_IS_EQUAL( m1.template subVector(r1), m1.row(r1) ); + VERIFY_IS_APPROX( (m1+m1).template subVector(r1), (m1+m1).row(r1) ); + VERIFY_IS_EQUAL( m1.template subVector(c1), m1.col(c1) ); + VERIFY_IS_APPROX( (m1+m1).template subVector(c1), (m1+m1).col(c1) ); + VERIFY_IS_EQUAL( m1.template subVectors(), m1.rows() ); + VERIFY_IS_EQUAL( m1.template subVectors(), m1.cols() ); + + if (rows>=2 || cols>=2) { + VERIFY_IS_EQUAL( int(m1.middleCols(0,0).IsRowMajor), int(m1.IsRowMajor) ); + VERIFY_IS_EQUAL( m1.middleCols(0,0).outerSize(), m1.IsRowMajor ? rows : 0); + VERIFY_IS_EQUAL( m1.middleCols(0,0).innerSize(), m1.IsRowMajor ? 0 : rows); + + VERIFY_IS_EQUAL( int(m1.middleRows(0,0).IsRowMajor), int(m1.IsRowMajor) ); + VERIFY_IS_EQUAL( m1.middleRows(0,0).outerSize(), m1.IsRowMajor ? 0 : cols); + VERIFY_IS_EQUAL( m1.middleRows(0,0).innerSize(), m1.IsRowMajor ? cols : 0); + } } template void compare_using_data_and_stride(const MatrixType& m) { - typedef typename MatrixType::Index Index; Index rows = m.rows(); Index cols = m.cols(); Index size = m.size(); @@ -213,7 +277,6 @@ void compare_using_data_and_stride(const MatrixType& m) template void data_and_stride(const MatrixType& m) { - typedef typename MatrixType::Index Index; Index rows = m.rows(); Index cols = m.cols(); @@ -231,15 +294,18 @@ void data_and_stride(const MatrixType& m) compare_using_data_and_stride(m1.col(c1).transpose()); } -void test_block() +EIGEN_DECLARE_TEST(block) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( block(Matrix()) ); + CALL_SUBTEST_1( block(Matrix(internal::random(2,50))) ); + CALL_SUBTEST_1( block(Matrix(internal::random(2,50))) ); CALL_SUBTEST_2( block(Matrix4d()) ); - CALL_SUBTEST_3( block(MatrixXcf(3, 3)) ); - CALL_SUBTEST_4( block(MatrixXi(8, 12)) ); - CALL_SUBTEST_5( block(MatrixXcd(20, 20)) ); - CALL_SUBTEST_6( block(MatrixXf(20, 20)) ); + CALL_SUBTEST_3( block(MatrixXcf(internal::random(2,50), internal::random(2,50))) ); + CALL_SUBTEST_4( block(MatrixXi(internal::random(2,50), internal::random(2,50))) ); + CALL_SUBTEST_5( block(MatrixXcd(internal::random(2,50), internal::random(2,50))) ); + CALL_SUBTEST_6( block(MatrixXf(internal::random(2,50), internal::random(2,50))) ); + CALL_SUBTEST_7( block(Matrix(internal::random(2,50), internal::random(2,50))) ); CALL_SUBTEST_8( block(Matrix(3, 4)) ); diff --git a/thirdparty/eigen/test/boostmultiprec.cpp b/thirdparty/eigen/test/boostmultiprec.cpp new file mode 100644 index 00000000..e83e9704 --- /dev/null +++ b/thirdparty/eigen/test/boostmultiprec.cpp @@ -0,0 +1,207 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2016 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include + +#ifdef EIGEN_TEST_MAX_SIZE +#undef EIGEN_TEST_MAX_SIZE +#endif + +#define EIGEN_TEST_MAX_SIZE 50 + +#ifdef EIGEN_TEST_PART_1 +#include "cholesky.cpp" +#endif + +#ifdef EIGEN_TEST_PART_2 +#include "lu.cpp" +#endif + +#ifdef EIGEN_TEST_PART_3 +#include "qr.cpp" +#endif + +#ifdef EIGEN_TEST_PART_4 +#include "qr_colpivoting.cpp" +#endif + +#ifdef EIGEN_TEST_PART_5 +#include "qr_fullpivoting.cpp" +#endif + +#ifdef EIGEN_TEST_PART_6 +#include "eigensolver_selfadjoint.cpp" +#endif + +#ifdef EIGEN_TEST_PART_7 +#include "eigensolver_generic.cpp" +#endif + +#ifdef EIGEN_TEST_PART_8 +#include "eigensolver_generalized_real.cpp" +#endif + +#ifdef EIGEN_TEST_PART_9 +#include "jacobisvd.cpp" +#endif + +#ifdef EIGEN_TEST_PART_10 +#include "bdcsvd.cpp" +#endif + +#ifdef EIGEN_TEST_PART_11 +#include "simplicial_cholesky.cpp" +#endif + +#include + +#undef min +#undef max +#undef isnan +#undef isinf +#undef isfinite +#undef I + +#include +#include +#include +#include +#include + +typedef boost::multiprecision::number, boost::multiprecision::et_on> Real; + +namespace Eigen { + template<> struct NumTraits : GenericNumTraits { + static inline Real dummy_precision() { return 1e-50; } + }; + + template + struct NumTraits > : NumTraits {}; + + template<> + Real test_precision() { return 1e-50; } + + // needed in C++93 mode where number does not support explicit cast. + namespace internal { + template + struct cast_impl { + static inline NewType run(const Real& x) { + return x.template convert_to(); + } + }; + + template<> + struct cast_impl > { + static inline std::complex run(const Real& x) { + return std::complex(x); + } + }; + } +} + +namespace boost { +namespace multiprecision { + // to make ADL works as expected: + using boost::math::isfinite; + using boost::math::isnan; + using boost::math::isinf; + using boost::math::copysign; + using boost::math::hypot; + + // The following is needed for std::complex: + Real fabs(const Real& a) { return abs EIGEN_NOT_A_MACRO (a); } + Real fmax(const Real& a, const Real& b) { using std::max; return max(a,b); } + + // some specialization for the unit tests: + inline bool test_isMuchSmallerThan(const Real& a, const Real& b) { + return internal::isMuchSmallerThan(a, b, test_precision()); + } + + inline bool test_isApprox(const Real& a, const Real& b) { + return internal::isApprox(a, b, test_precision()); + } + + inline bool test_isApproxOrLessThan(const Real& a, const Real& b) { + return internal::isApproxOrLessThan(a, b, test_precision()); + } + + Real get_test_precision(const Real&) { + return test_precision(); + } + + Real test_relative_error(const Real &a, const Real &b) { + using Eigen::numext::abs2; + return sqrt(abs2(a-b)/Eigen::numext::mini(abs2(a),abs2(b))); + } +} +} + +namespace Eigen { + +} + +EIGEN_DECLARE_TEST(boostmultiprec) +{ + typedef Matrix Mat; + typedef Matrix,Dynamic,Dynamic> MatC; + + std::cout << "NumTraits::epsilon() = " << NumTraits::epsilon() << std::endl; + std::cout << "NumTraits::dummy_precision() = " << NumTraits::dummy_precision() << std::endl; + std::cout << "NumTraits::lowest() = " << NumTraits::lowest() << std::endl; + std::cout << "NumTraits::highest() = " << NumTraits::highest() << std::endl; + std::cout << "NumTraits::digits10() = " << NumTraits::digits10() << std::endl; + + // check stream output + { + Mat A(10,10); + A.setRandom(); + std::stringstream ss; + ss << A; + } + { + MatC A(10,10); + A.setRandom(); + std::stringstream ss; + ss << A; + } + + for(int i = 0; i < g_repeat; i++) { + int s = internal::random(1,EIGEN_TEST_MAX_SIZE); + + CALL_SUBTEST_1( cholesky(Mat(s,s)) ); + + CALL_SUBTEST_2( lu_non_invertible() ); + CALL_SUBTEST_2( lu_invertible() ); + CALL_SUBTEST_2( lu_non_invertible() ); + CALL_SUBTEST_2( lu_invertible() ); + + CALL_SUBTEST_3( qr(Mat(internal::random(1,EIGEN_TEST_MAX_SIZE),internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_3( qr_invertible() ); + + CALL_SUBTEST_4( qr() ); + CALL_SUBTEST_4( cod() ); + CALL_SUBTEST_4( qr_invertible() ); + + CALL_SUBTEST_5( qr() ); + CALL_SUBTEST_5( qr_invertible() ); + + CALL_SUBTEST_6( selfadjointeigensolver(Mat(s,s)) ); + + CALL_SUBTEST_7( eigensolver(Mat(s,s)) ); + + CALL_SUBTEST_8( generalized_eigensolver_real(Mat(s,s)) ); + + TEST_SET_BUT_UNUSED_VARIABLE(s) + } + + CALL_SUBTEST_9(( jacobisvd(Mat(internal::random(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE), internal::random(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/2))) )); + CALL_SUBTEST_10(( bdcsvd(Mat(internal::random(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE), internal::random(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/2))) )); + + CALL_SUBTEST_11(( test_simplicial_cholesky_T() )); +} diff --git a/thirdparty/eigen/test/bug1213.cpp b/thirdparty/eigen/test/bug1213.cpp new file mode 100644 index 00000000..581760c1 --- /dev/null +++ b/thirdparty/eigen/test/bug1213.cpp @@ -0,0 +1,13 @@ + +// This anonymous enum is essential to trigger the linking issue +enum { + Foo +}; + +#include "bug1213.h" + +bool bug1213_1(const Eigen::Vector3f& x) +{ + return bug1213_2(x); +} + diff --git a/thirdparty/eigen/test/bug1213.h b/thirdparty/eigen/test/bug1213.h new file mode 100644 index 00000000..040e5a47 --- /dev/null +++ b/thirdparty/eigen/test/bug1213.h @@ -0,0 +1,8 @@ + +#include + +template +bool bug1213_2(const Eigen::Matrix& x); + +bool bug1213_1(const Eigen::Vector3f& x); + diff --git a/thirdparty/eigen/test/bug1213_main.cpp b/thirdparty/eigen/test/bug1213_main.cpp new file mode 100644 index 00000000..4802c000 --- /dev/null +++ b/thirdparty/eigen/test/bug1213_main.cpp @@ -0,0 +1,18 @@ + +// This is a regression unit regarding a weird linking issue with gcc. + +#include "bug1213.h" + +int main() +{ + return 0; +} + + +template +bool bug1213_2(const Eigen::Matrix& ) +{ + return true; +} + +template bool bug1213_2(const Eigen::Vector3f&); diff --git a/thirdparty/eigen/test/cholesky.cpp b/thirdparty/eigen/test/cholesky.cpp index 56885deb..0b1a7b45 100644 --- a/thirdparty/eigen/test/cholesky.cpp +++ b/thirdparty/eigen/test/cholesky.cpp @@ -7,24 +7,19 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -#ifndef EIGEN_NO_ASSERTION_CHECKING -#define EIGEN_NO_ASSERTION_CHECKING -#endif - -static int nb_temporaries; - -#define EIGEN_DENSE_STORAGE_CTOR_PLUGIN { if(size!=0) nb_temporaries++; } +#define TEST_ENABLE_TEMPORARY_TRACKING #include "main.h" #include #include +#include "solverbase.h" -#define VERIFY_EVALUATION_COUNT(XPR,N) {\ - nb_temporaries = 0; \ - XPR; \ - if(nb_temporaries!=N) std::cerr << "nb_temporaries == " << nb_temporaries << "\n"; \ - VERIFY( (#XPR) && nb_temporaries==N ); \ - } +template +typename MatrixType::RealScalar matrix_l1_norm(const MatrixType& m) { + if(m.cols()==0) return typename MatrixType::RealScalar(0); + MatrixType symm = m.template selfadjointView(); + return symm.cwiseAbs().colwise().sum().maxCoeff(); +} template class CholType> void test_chol_update(const MatrixType& symm) { @@ -60,7 +55,6 @@ template class CholType> void test_c template void cholesky(const MatrixType& m) { - typedef typename MatrixType::Index Index; /* this test covers the following files: LLT.h LDLT.h */ @@ -83,20 +77,26 @@ template void cholesky(const MatrixType& m) symm += a1 * a1.adjoint(); } - // to test if really Cholesky only uses the upper triangular part, uncomment the following - // FIXME: currently that fails !! - //symm.template part().setZero(); - { + STATIC_CHECK(( internal::is_same::StorageIndex,int>::value )); + STATIC_CHECK(( internal::is_same::StorageIndex,int>::value )); + SquareMatrixType symmUp = symm.template triangularView(); SquareMatrixType symmLo = symm.template triangularView(); - + LLT chollo(symmLo); VERIFY_IS_APPROX(symm, chollo.reconstructedMatrix()); - vecX = chollo.solve(vecB); - VERIFY_IS_APPROX(symm * vecX, vecB); - matX = chollo.solve(matB); - VERIFY_IS_APPROX(symm * matX, matB); + + check_solverbase(symm, chollo, rows, rows, 1); + check_solverbase(symm, chollo, rows, cols, rows); + + const MatrixType symmLo_inverse = chollo.solve(MatrixType::Identity(rows,cols)); + RealScalar rcond = (RealScalar(1) / matrix_l1_norm(symmLo)) / + matrix_l1_norm(symmLo_inverse); + RealScalar rcond_est = chollo.rcond(); + // Verify that the estimated condition number is within a factor of 10 of the + // truth. + VERIFY(rcond_est >= rcond / 10 && rcond_est <= rcond * 10); // test the upper mode LLT cholup(symmUp); @@ -106,15 +106,24 @@ template void cholesky(const MatrixType& m) matX = cholup.solve(matB); VERIFY_IS_APPROX(symm * matX, matB); + // Verify that the estimated condition number is within a factor of 10 of the + // truth. + const MatrixType symmUp_inverse = cholup.solve(MatrixType::Identity(rows,cols)); + rcond = (RealScalar(1) / matrix_l1_norm(symmUp)) / + matrix_l1_norm(symmUp_inverse); + rcond_est = cholup.rcond(); + VERIFY(rcond_est >= rcond / 10 && rcond_est <= rcond * 10); + + MatrixType neg = -symmLo; chollo.compute(neg); - VERIFY(chollo.info()==NumericalIssue); + VERIFY(neg.size()==0 || chollo.info()==NumericalIssue); VERIFY_IS_APPROX(MatrixType(chollo.matrixL().transpose().conjugate()), MatrixType(chollo.matrixU())); VERIFY_IS_APPROX(MatrixType(chollo.matrixU().transpose().conjugate()), MatrixType(chollo.matrixL())); VERIFY_IS_APPROX(MatrixType(cholup.matrixL().transpose().conjugate()), MatrixType(cholup.matrixU())); VERIFY_IS_APPROX(MatrixType(cholup.matrixU().transpose().conjugate()), MatrixType(cholup.matrixL())); - + // test some special use cases of SelfCwiseBinaryOp: MatrixType m1 = MatrixType::Random(rows,cols), m2(rows,cols); m2 = m1; @@ -133,6 +142,9 @@ template void cholesky(const MatrixType& m) // LDLT { + STATIC_CHECK(( internal::is_same::StorageIndex,int>::value )); + STATIC_CHECK(( internal::is_same::StorageIndex,int>::value )); + int sign = internal::random()%2 ? 1 : -1; if(sign == -1) @@ -144,19 +156,37 @@ template void cholesky(const MatrixType& m) SquareMatrixType symmLo = symm.template triangularView(); LDLT ldltlo(symmLo); + VERIFY(ldltlo.info()==Success); VERIFY_IS_APPROX(symm, ldltlo.reconstructedMatrix()); - vecX = ldltlo.solve(vecB); - VERIFY_IS_APPROX(symm * vecX, vecB); - matX = ldltlo.solve(matB); - VERIFY_IS_APPROX(symm * matX, matB); + + check_solverbase(symm, ldltlo, rows, rows, 1); + check_solverbase(symm, ldltlo, rows, cols, rows); + + const MatrixType symmLo_inverse = ldltlo.solve(MatrixType::Identity(rows,cols)); + RealScalar rcond = (RealScalar(1) / matrix_l1_norm(symmLo)) / + matrix_l1_norm(symmLo_inverse); + RealScalar rcond_est = ldltlo.rcond(); + // Verify that the estimated condition number is within a factor of 10 of the + // truth. + VERIFY(rcond_est >= rcond / 10 && rcond_est <= rcond * 10); + LDLT ldltup(symmUp); + VERIFY(ldltup.info()==Success); VERIFY_IS_APPROX(symm, ldltup.reconstructedMatrix()); vecX = ldltup.solve(vecB); VERIFY_IS_APPROX(symm * vecX, vecB); matX = ldltup.solve(matB); VERIFY_IS_APPROX(symm * matX, matB); + // Verify that the estimated condition number is within a factor of 10 of the + // truth. + const MatrixType symmUp_inverse = ldltup.solve(MatrixType::Identity(rows,cols)); + rcond = (RealScalar(1) / matrix_l1_norm(symmUp)) / + matrix_l1_norm(symmUp_inverse); + rcond_est = ldltup.rcond(); + VERIFY(rcond_est >= rcond / 10 && rcond_est <= rcond * 10); + VERIFY_IS_APPROX(MatrixType(ldltlo.matrixL().transpose().conjugate()), MatrixType(ldltlo.matrixU())); VERIFY_IS_APPROX(MatrixType(ldltlo.matrixU().transpose().conjugate()), MatrixType(ldltlo.matrixL())); VERIFY_IS_APPROX(MatrixType(ldltup.matrixL().transpose().conjugate()), MatrixType(ldltup.matrixU())); @@ -185,7 +215,7 @@ template void cholesky(const MatrixType& m) if(rows>=3) { SquareMatrixType A = symm; - int c = internal::random(0,rows-2); + Index c = internal::random(0,rows-2); A.bottomRightCorner(c,c).setZero(); // Make sure a solution exists: vecX.setRandom(); @@ -196,11 +226,11 @@ template void cholesky(const MatrixType& m) vecX = ldltlo.solve(vecB); VERIFY_IS_APPROX(A * vecX, vecB); } - + // check non-full rank matrices if(rows>=3) { - int r = internal::random(1,rows-1); + Index r = internal::random(1,rows-1); Matrix a = Matrix::Random(rows,r); SquareMatrixType A = a * a.adjoint(); // Make sure a solution exists: @@ -212,15 +242,17 @@ template void cholesky(const MatrixType& m) vecX = ldltlo.solve(vecB); VERIFY_IS_APPROX(A * vecX, vecB); } - + // check matrices with a wide spectrum if(rows>=3) { + using std::pow; + using std::sqrt; RealScalar s = (std::min)(16,std::numeric_limits::max_exponent10/8); Matrix a = Matrix::Random(rows,rows); Matrix d = Matrix::Random(rows); - for(int k=0; k(-s,s)); + for(Index k=0; k(-s,s)); SquareMatrixType A = a * d.asDiagonal() * a.adjoint(); // Make sure a solution exists: vecX.setRandom(); @@ -229,7 +261,20 @@ template void cholesky(const MatrixType& m) ldltlo.compute(A); VERIFY_IS_APPROX(A, ldltlo.reconstructedMatrix()); vecX = ldltlo.solve(vecB); - VERIFY_IS_APPROX(A * vecX, vecB); + + if(ldltlo.vectorD().real().cwiseAbs().minCoeff()>RealScalar(0)) + { + VERIFY_IS_APPROX(A * vecX,vecB); + } + else + { + RealScalar large_tol = sqrt(test_precision()); + VERIFY((A * vecX).isApprox(vecB, large_tol)); + + ++g_test_level; + VERIFY_IS_APPROX(A * vecX,vecB); + --g_test_level; + } } } @@ -245,8 +290,6 @@ template void cholesky_cplx(const MatrixType& m) // test mixing real/scalar types - typedef typename MatrixType::Index Index; - Index rows = m.rows(); Index cols = m.cols(); @@ -271,10 +314,9 @@ template void cholesky_cplx(const MatrixType& m) LLT chollo(symmLo); VERIFY_IS_APPROX(symm, chollo.reconstructedMatrix()); - vecX = chollo.solve(vecB); - VERIFY_IS_APPROX(symm * vecX, vecB); -// matX = chollo.solve(matB); -// VERIFY_IS_APPROX(symm * matX, matB); + + check_solverbase(symm, chollo, rows, rows, 1); + //check_solverbase(symm, chollo, rows, cols, rows); } // LDLT @@ -289,11 +331,11 @@ template void cholesky_cplx(const MatrixType& m) RealMatrixType symmLo = symm.template triangularView(); LDLT ldltlo(symmLo); + VERIFY(ldltlo.info()==Success); VERIFY_IS_APPROX(symm, ldltlo.reconstructedMatrix()); - vecX = ldltlo.solve(vecB); - VERIFY_IS_APPROX(symm * vecX, vecB); -// matX = ldltlo.solve(matB); -// VERIFY_IS_APPROX(symm * matX, matB); + + check_solverbase(symm, ldltlo, rows, rows, 1); + //check_solverbase(symm, ldltlo, rows, cols, rows); } } @@ -314,43 +356,115 @@ template void cholesky_bug241(const MatrixType& m) } // LDLT is not guaranteed to work for indefinite matrices, but happens to work fine if matrix is diagonal. -// This test checks that LDLT reports correctly that matrix is indefinite. +// This test checks that LDLT reports correctly that matrix is indefinite. // See http://forum.kde.org/viewtopic.php?f=74&t=106942 and bug 736 template void cholesky_definiteness(const MatrixType& m) { eigen_assert(m.rows() == 2 && m.cols() == 2); MatrixType mat; LDLT ldlt(2); - + { mat << 1, 0, 0, -1; ldlt.compute(mat); + VERIFY(ldlt.info()==Success); VERIFY(!ldlt.isNegative()); VERIFY(!ldlt.isPositive()); + VERIFY_IS_APPROX(mat,ldlt.reconstructedMatrix()); } { mat << 1, 2, 2, 1; ldlt.compute(mat); + VERIFY(ldlt.info()==Success); VERIFY(!ldlt.isNegative()); VERIFY(!ldlt.isPositive()); + VERIFY_IS_APPROX(mat,ldlt.reconstructedMatrix()); } { mat << 0, 0, 0, 0; ldlt.compute(mat); + VERIFY(ldlt.info()==Success); VERIFY(ldlt.isNegative()); VERIFY(ldlt.isPositive()); + VERIFY_IS_APPROX(mat,ldlt.reconstructedMatrix()); } { mat << 0, 0, 0, 1; ldlt.compute(mat); + VERIFY(ldlt.info()==Success); VERIFY(!ldlt.isNegative()); VERIFY(ldlt.isPositive()); + VERIFY_IS_APPROX(mat,ldlt.reconstructedMatrix()); } { mat << -1, 0, 0, 0; ldlt.compute(mat); + VERIFY(ldlt.info()==Success); VERIFY(ldlt.isNegative()); VERIFY(!ldlt.isPositive()); + VERIFY_IS_APPROX(mat,ldlt.reconstructedMatrix()); + } +} + +template +void cholesky_faillure_cases() +{ + MatrixXd mat; + LDLT ldlt; + + { + mat.resize(2,2); + mat << 0, 1, 1, 0; + ldlt.compute(mat); + VERIFY_IS_NOT_APPROX(mat,ldlt.reconstructedMatrix()); + VERIFY(ldlt.info()==NumericalIssue); + } +#if (!EIGEN_ARCH_i386) || defined(EIGEN_VECTORIZE_SSE2) + { + mat.resize(3,3); + mat << -1, -3, 3, + -3, -8.9999999999999999999, 1, + 3, 1, 0; + ldlt.compute(mat); + VERIFY(ldlt.info()==NumericalIssue); + VERIFY_IS_NOT_APPROX(mat,ldlt.reconstructedMatrix()); + } +#endif + { + mat.resize(3,3); + mat << 1, 2, 3, + 2, 4, 1, + 3, 1, 0; + ldlt.compute(mat); + VERIFY(ldlt.info()==NumericalIssue); + VERIFY_IS_NOT_APPROX(mat,ldlt.reconstructedMatrix()); + } + + { + mat.resize(8,8); + mat << 0.1, 0, -0.1, 0, 0, 0, 1, 0, + 0, 4.24667, 0, 2.00333, 0, 0, 0, 0, + -0.1, 0, 0.2, 0, -0.1, 0, 0, 0, + 0, 2.00333, 0, 8.49333, 0, 2.00333, 0, 0, + 0, 0, -0.1, 0, 0.1, 0, 0, 1, + 0, 0, 0, 2.00333, 0, 4.24667, 0, 0, + 1, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 1, 0, 0, 0; + ldlt.compute(mat); + VERIFY(ldlt.info()==NumericalIssue); + VERIFY_IS_NOT_APPROX(mat,ldlt.reconstructedMatrix()); + } + + // bug 1479 + { + mat.resize(4,4); + mat << 1, 2, 0, 1, + 2, 4, 0, 2, + 0, 0, 0, 1, + 1, 2, 1, 1; + ldlt.compute(mat); + VERIFY(ldlt.info()==NumericalIssue); + VERIFY_IS_NOT_APPROX(mat,ldlt.reconstructedMatrix()); } } @@ -362,19 +476,23 @@ template void cholesky_verify_assert() VERIFY_RAISES_ASSERT(llt.matrixL()) VERIFY_RAISES_ASSERT(llt.matrixU()) VERIFY_RAISES_ASSERT(llt.solve(tmp)) - VERIFY_RAISES_ASSERT(llt.solveInPlace(&tmp)) + VERIFY_RAISES_ASSERT(llt.transpose().solve(tmp)) + VERIFY_RAISES_ASSERT(llt.adjoint().solve(tmp)) + VERIFY_RAISES_ASSERT(llt.solveInPlace(tmp)) LDLT ldlt; VERIFY_RAISES_ASSERT(ldlt.matrixL()) - VERIFY_RAISES_ASSERT(ldlt.permutationP()) + VERIFY_RAISES_ASSERT(ldlt.transpositionsP()) VERIFY_RAISES_ASSERT(ldlt.vectorD()) VERIFY_RAISES_ASSERT(ldlt.isPositive()) VERIFY_RAISES_ASSERT(ldlt.isNegative()) VERIFY_RAISES_ASSERT(ldlt.solve(tmp)) - VERIFY_RAISES_ASSERT(ldlt.solveInPlace(&tmp)) + VERIFY_RAISES_ASSERT(ldlt.transpose().solve(tmp)) + VERIFY_RAISES_ASSERT(ldlt.adjoint().solve(tmp)) + VERIFY_RAISES_ASSERT(ldlt.solveInPlace(tmp)) } -void test_cholesky() +EIGEN_DECLARE_TEST(cholesky) { int s = 0; for(int i = 0; i < g_repeat; i++) { @@ -384,11 +502,20 @@ void test_cholesky() CALL_SUBTEST_3( cholesky_definiteness(Matrix2d()) ); CALL_SUBTEST_4( cholesky(Matrix3f()) ); CALL_SUBTEST_5( cholesky(Matrix4d()) ); + s = internal::random(1,EIGEN_TEST_MAX_SIZE); CALL_SUBTEST_2( cholesky(MatrixXd(s,s)) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) + s = internal::random(1,EIGEN_TEST_MAX_SIZE/2); CALL_SUBTEST_6( cholesky_cplx(MatrixXcd(s,s)) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) } + // empty matrix, regression test for Bug 785: + CALL_SUBTEST_2( cholesky(MatrixXd(0,0)) ); + + // This does not work yet: + // CALL_SUBTEST_2( cholesky(Matrix()) ); CALL_SUBTEST_4( cholesky_verify_assert() ); CALL_SUBTEST_7( cholesky_verify_assert() ); @@ -398,7 +525,8 @@ void test_cholesky() // Test problem size constructors CALL_SUBTEST_9( LLT(10) ); CALL_SUBTEST_9( LDLT(10) ); - - TEST_SET_BUT_UNUSED_VARIABLE(s) + + CALL_SUBTEST_2( cholesky_faillure_cases() ); + TEST_SET_BUT_UNUSED_VARIABLE(nb_temporaries) } diff --git a/thirdparty/eigen/test/cholmod_support.cpp b/thirdparty/eigen/test/cholmod_support.cpp index 8f8be3c0..89b9cf41 100644 --- a/thirdparty/eigen/test/cholmod_support.cpp +++ b/thirdparty/eigen/test/cholmod_support.cpp @@ -7,25 +7,26 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +#define EIGEN_NO_DEBUG_SMALL_PRODUCT_BLOCKS #include "sparse_solver.h" #include -template void test_cholmod_T() +template void test_cholmod_ST() { - CholmodDecomposition, Lower> g_chol_colmajor_lower; g_chol_colmajor_lower.setMode(CholmodSupernodalLLt); - CholmodDecomposition, Upper> g_chol_colmajor_upper; g_chol_colmajor_upper.setMode(CholmodSupernodalLLt); - CholmodDecomposition, Lower> g_llt_colmajor_lower; g_llt_colmajor_lower.setMode(CholmodSimplicialLLt); - CholmodDecomposition, Upper> g_llt_colmajor_upper; g_llt_colmajor_upper.setMode(CholmodSimplicialLLt); - CholmodDecomposition, Lower> g_ldlt_colmajor_lower; g_ldlt_colmajor_lower.setMode(CholmodLDLt); - CholmodDecomposition, Upper> g_ldlt_colmajor_upper; g_ldlt_colmajor_upper.setMode(CholmodLDLt); + CholmodDecomposition g_chol_colmajor_lower; g_chol_colmajor_lower.setMode(CholmodSupernodalLLt); + CholmodDecomposition g_chol_colmajor_upper; g_chol_colmajor_upper.setMode(CholmodSupernodalLLt); + CholmodDecomposition g_llt_colmajor_lower; g_llt_colmajor_lower.setMode(CholmodSimplicialLLt); + CholmodDecomposition g_llt_colmajor_upper; g_llt_colmajor_upper.setMode(CholmodSimplicialLLt); + CholmodDecomposition g_ldlt_colmajor_lower; g_ldlt_colmajor_lower.setMode(CholmodLDLt); + CholmodDecomposition g_ldlt_colmajor_upper; g_ldlt_colmajor_upper.setMode(CholmodLDLt); - CholmodSupernodalLLT, Lower> chol_colmajor_lower; - CholmodSupernodalLLT, Upper> chol_colmajor_upper; - CholmodSimplicialLLT, Lower> llt_colmajor_lower; - CholmodSimplicialLLT, Upper> llt_colmajor_upper; - CholmodSimplicialLDLT, Lower> ldlt_colmajor_lower; - CholmodSimplicialLDLT, Upper> ldlt_colmajor_upper; + CholmodSupernodalLLT chol_colmajor_lower; + CholmodSupernodalLLT chol_colmajor_upper; + CholmodSimplicialLLT llt_colmajor_lower; + CholmodSimplicialLLT llt_colmajor_upper; + CholmodSimplicialLDLT ldlt_colmajor_lower; + CholmodSimplicialLDLT ldlt_colmajor_upper; check_sparse_spd_solving(g_chol_colmajor_lower); check_sparse_spd_solving(g_chol_colmajor_upper); @@ -40,17 +41,29 @@ template void test_cholmod_T() check_sparse_spd_solving(llt_colmajor_upper); check_sparse_spd_solving(ldlt_colmajor_lower); check_sparse_spd_solving(ldlt_colmajor_upper); - -// check_sparse_spd_determinant(chol_colmajor_lower); -// check_sparse_spd_determinant(chol_colmajor_upper); -// check_sparse_spd_determinant(llt_colmajor_lower); -// check_sparse_spd_determinant(llt_colmajor_upper); -// check_sparse_spd_determinant(ldlt_colmajor_lower); -// check_sparse_spd_determinant(ldlt_colmajor_upper); + + check_sparse_spd_determinant(chol_colmajor_lower); + check_sparse_spd_determinant(chol_colmajor_upper); + check_sparse_spd_determinant(llt_colmajor_lower); + check_sparse_spd_determinant(llt_colmajor_upper); + check_sparse_spd_determinant(ldlt_colmajor_lower); + check_sparse_spd_determinant(ldlt_colmajor_upper); +} + +template void test_cholmod_T() +{ + test_cholmod_ST >(); } -void test_cholmod_support() +EIGEN_DECLARE_TEST(cholmod_support) { - CALL_SUBTEST_1(test_cholmod_T()); - CALL_SUBTEST_2(test_cholmod_T >()); + CALL_SUBTEST_11( (test_cholmod_T()) ); + CALL_SUBTEST_12( (test_cholmod_T()) ); + CALL_SUBTEST_13( (test_cholmod_T()) ); + CALL_SUBTEST_14( (test_cholmod_T()) ); + CALL_SUBTEST_21( (test_cholmod_T, ColMajor, int >()) ); + CALL_SUBTEST_22( (test_cholmod_T, ColMajor, long>()) ); + // TODO complex row-major matrices do not work at the moment: + // CALL_SUBTEST_23( (test_cholmod_T, RowMajor, int >()) ); + // CALL_SUBTEST_24( (test_cholmod_T, RowMajor, long>()) ); } diff --git a/thirdparty/eigen/test/clz.cpp b/thirdparty/eigen/test/clz.cpp new file mode 100644 index 00000000..1d08b471 --- /dev/null +++ b/thirdparty/eigen/test/clz.cpp @@ -0,0 +1,74 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2023 The Eigen Authors +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" + +template +int ref_clz(T val) { + static const int kNumBits = sizeof(T) * CHAR_BIT; + T kMsbMask = T(1) << (kNumBits - 1); + int z = 0; + for (; z < kNumBits && ((val & kMsbMask) == 0); ++z) { + val <<= 1; + } + return z; +} + +template +int ref_ctz(T val) { + static const int kNumBits = sizeof(T) * CHAR_BIT; + T kLsbMask = T(1); + int z = 0; + for (; z < kNumBits && ((val & kLsbMask) == 0); ++z) { + val >>= 1; + } + return z; +} + +template +void test_clz_ctz() { + T step = sizeof(T) <= 2 ? 1 : (Eigen::NumTraits::highest() / (T(1) << 16)); + T iters = Eigen::NumTraits::highest() / step; + for (T i = 0; i < iters; ++i) { + T val = i * step; + int expected_clz = ref_clz(val); + int actual_clz = Eigen::internal::clz(val); + VERIFY(expected_clz == actual_clz); + + int expected_ctz = ref_ctz(val); + int actual_ctz = Eigen::internal::ctz(val); + VERIFY(expected_ctz == actual_ctz); + } +} + +template +void test_clz_ctz_random() { + for (int i = 0; i < 1024 * 1024; ++i) { + T val = Eigen::internal::random(); + int expected_clz = ref_clz(val); + int actual_clz = Eigen::internal::clz(val); + VERIFY(expected_clz == actual_clz); + + int expected_ctz = ref_ctz(val); + int actual_ctz = Eigen::internal::ctz(val); + VERIFY(expected_ctz == actual_ctz); + } +} + +EIGEN_DECLARE_TEST(clz) { + CALL_SUBTEST_1(test_clz_ctz()); + CALL_SUBTEST_2(test_clz_ctz()); + CALL_SUBTEST_3(test_clz_ctz()); + CALL_SUBTEST_4(test_clz_ctz()); + + for (int i = 0; i < g_repeat; i++) { + test_clz_ctz_random(); + test_clz_ctz_random(); + } +} diff --git a/thirdparty/eigen/test/commainitializer.cpp b/thirdparty/eigen/test/commainitializer.cpp index 29659234..eb275be9 100644 --- a/thirdparty/eigen/test/commainitializer.cpp +++ b/thirdparty/eigen/test/commainitializer.cpp @@ -34,8 +34,14 @@ void test_blocks() if(N1 > 0) { - VERIFY_RAISES_ASSERT((m_fixed << mat11, mat12, mat11, mat21, mat22)); - VERIFY_RAISES_ASSERT((m_fixed << mat11, mat12, mat21, mat21, mat22)); + if(M1 > 0) + { + VERIFY_RAISES_ASSERT((m_fixed << mat11, mat12, mat11, mat21, mat22)); + } + if(M2 > 0) + { + VERIFY_RAISES_ASSERT((m_fixed << mat11, mat12, mat21, mat21, mat22)); + } } else { @@ -49,29 +55,31 @@ void test_blocks() } -template +template struct test_block_recursion { static void run() { - test_blocks<(N>>6)&3, (N>>4)&3, (N>>2)&3, N & 3>(); - test_block_recursion::run(); + test_block_recursion::run(); + test_block_recursion::run(); } }; -template<> -struct test_block_recursion<-1> +template +struct test_block_recursion<0,N> { - static void run() { } + static void run() { + test_blocks<(N>>6)&3, (N>>4)&3, (N>>2)&3, N & 3>(); + } }; -void test_commainitializer() -{ +void test_basics() { Matrix3d m3; Matrix4d m4; - #ifndef _MSC_VER VERIFY_RAISES_ASSERT( (m3 << 1, 2, 3, 4, 5, 6, 7, 8) ); + + #ifndef _MSC_VER VERIFY_RAISES_ASSERT( (m3 << 1, 2, 3, 4, 5, 6, 7, 8, 9, 10) ); #endif @@ -98,7 +106,13 @@ void test_commainitializer() 4, 5, 6, vec[2].transpose(); VERIFY_IS_APPROX(m3, ref); +} + +EIGEN_DECLARE_TEST(commainitializer) +{ + + CALL_SUBTEST_1(test_basics()); // recursively test all block-sizes from 0 to 3: - test_block_recursion<(1<<8) - 1>(); + CALL_SUBTEST_2(test_block_recursion<8>::run()); } diff --git a/thirdparty/eigen/test/conjugate_gradient.cpp b/thirdparty/eigen/test/conjugate_gradient.cpp index 019cc4d6..b076a126 100644 --- a/thirdparty/eigen/test/conjugate_gradient.cpp +++ b/thirdparty/eigen/test/conjugate_gradient.cpp @@ -10,13 +10,14 @@ #include "sparse_solver.h" #include -template void test_conjugate_gradient_T() +template void test_conjugate_gradient_T() { - ConjugateGradient, Lower > cg_colmajor_lower_diag; - ConjugateGradient, Upper > cg_colmajor_upper_diag; - ConjugateGradient, Lower|Upper> cg_colmajor_loup_diag; - ConjugateGradient, Lower, IdentityPreconditioner> cg_colmajor_lower_I; - ConjugateGradient, Upper, IdentityPreconditioner> cg_colmajor_upper_I; + typedef SparseMatrix SparseMatrixType; + ConjugateGradient cg_colmajor_lower_diag; + ConjugateGradient cg_colmajor_upper_diag; + ConjugateGradient cg_colmajor_loup_diag; + ConjugateGradient cg_colmajor_lower_I; + ConjugateGradient cg_colmajor_upper_I; CALL_SUBTEST( check_sparse_spd_solving(cg_colmajor_lower_diag) ); CALL_SUBTEST( check_sparse_spd_solving(cg_colmajor_upper_diag) ); @@ -25,8 +26,9 @@ template void test_conjugate_gradient_T() CALL_SUBTEST( check_sparse_spd_solving(cg_colmajor_upper_I) ); } -void test_conjugate_gradient() +EIGEN_DECLARE_TEST(conjugate_gradient) { - CALL_SUBTEST_1(test_conjugate_gradient_T()); - CALL_SUBTEST_2(test_conjugate_gradient_T >()); + CALL_SUBTEST_1(( test_conjugate_gradient_T() )); + CALL_SUBTEST_2(( test_conjugate_gradient_T, int>() )); + CALL_SUBTEST_3(( test_conjugate_gradient_T() )); } diff --git a/thirdparty/eigen/test/conservative_resize.cpp b/thirdparty/eigen/test/conservative_resize.cpp index 498421b4..d48eb126 100644 --- a/thirdparty/eigen/test/conservative_resize.cpp +++ b/thirdparty/eigen/test/conservative_resize.cpp @@ -10,6 +10,7 @@ #include "main.h" #include +#include "AnnoyingScalar.h" using namespace Eigen; @@ -17,7 +18,6 @@ template void run_matrix_tests() { typedef Matrix MatrixType; - typedef typename MatrixType::Index Index; MatrixType m, n; @@ -110,7 +110,33 @@ void run_vector_tests() } } -void test_conservative_resize() +// Basic memory leak check with a non-copyable scalar type +template void noncopyable() +{ + typedef Eigen::Matrix VectorType; + typedef Eigen::Matrix MatrixType; + + { +#ifndef EIGEN_TEST_ANNOYING_SCALAR_DONT_THROW + AnnoyingScalar::dont_throw = true; +#endif + int n = 50; + VectorType v0(n), v1(n); + MatrixType m0(n,n), m1(n,n), m2(n,n); + v0.setOnes(); v1.setOnes(); + m0.setOnes(); m1.setOnes(); m2.setOnes(); + VERIFY(m0==m1); + m0.conservativeResize(2*n,2*n); + VERIFY(m0.topLeftCorner(n,n) == m1); + + VERIFY(v0.head(n) == v1); + v0.conservativeResize(2*n); + VERIFY(v0.head(n) == v1); + } + VERIFY(AnnoyingScalar::instances==0 && "global memory leak detected in noncopyable"); +} + +EIGEN_DECLARE_TEST(conservative_resize) { for(int i=0; i, Eigen::RowMajor>())); CALL_SUBTEST_4((run_matrix_tests, Eigen::ColMajor>())); CALL_SUBTEST_5((run_matrix_tests, Eigen::RowMajor>())); - CALL_SUBTEST_6((run_matrix_tests, Eigen::ColMajor>())); + CALL_SUBTEST_5((run_matrix_tests, Eigen::ColMajor>())); + CALL_SUBTEST_1((run_matrix_tests())); CALL_SUBTEST_1((run_vector_tests())); CALL_SUBTEST_2((run_vector_tests())); CALL_SUBTEST_3((run_vector_tests())); CALL_SUBTEST_4((run_vector_tests >())); CALL_SUBTEST_5((run_vector_tests >())); + +#ifndef EIGEN_TEST_ANNOYING_SCALAR_DONT_THROW + AnnoyingScalar::dont_throw = true; +#endif + CALL_SUBTEST_6(( run_vector_tests() )); + CALL_SUBTEST_6(( noncopyable<0>() )); } } diff --git a/thirdparty/eigen/test/constructor.cpp b/thirdparty/eigen/test/constructor.cpp new file mode 100644 index 00000000..ffd5e802 --- /dev/null +++ b/thirdparty/eigen/test/constructor.cpp @@ -0,0 +1,98 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2017 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + + +#define TEST_ENABLE_TEMPORARY_TRACKING + +#include "main.h" + +template struct Wrapper +{ + MatrixType m_mat; + inline Wrapper(const MatrixType &x) : m_mat(x) {} + inline operator const MatrixType& () const { return m_mat; } + inline operator MatrixType& () { return m_mat; } +}; + +enum my_sizes { M = 12, N = 7}; + +template void ctor_init1(const MatrixType& m) +{ + // Check logic in PlainObjectBase::_init1 + Index rows = m.rows(); + Index cols = m.cols(); + + MatrixType m0 = MatrixType::Random(rows,cols); + + VERIFY_EVALUATION_COUNT( MatrixType m1(m0), 1); + VERIFY_EVALUATION_COUNT( MatrixType m2(m0+m0), 1); + VERIFY_EVALUATION_COUNT( MatrixType m2(m0.block(0,0,rows,cols)) , 1); + + Wrapper wrapper(m0); + VERIFY_EVALUATION_COUNT( MatrixType m3(wrapper) , 1); +} + + +EIGEN_DECLARE_TEST(constructor) +{ + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1( ctor_init1(Matrix()) ); + CALL_SUBTEST_1( ctor_init1(Matrix4d()) ); + CALL_SUBTEST_1( ctor_init1(MatrixXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_1( ctor_init1(MatrixXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + } + { + Matrix a(123); + VERIFY_IS_EQUAL(a[0], 123); + } + { + Matrix a(123.0); + VERIFY_IS_EQUAL(a[0], 123); + } + { + Matrix a(123); + VERIFY_IS_EQUAL(a[0], 123.f); + } + { + Array a(123); + VERIFY_IS_EQUAL(a[0], 123); + } + { + Array a(123.0); + VERIFY_IS_EQUAL(a[0], 123); + } + { + Array a(123); + VERIFY_IS_EQUAL(a[0], 123.f); + } + { + Array a(123); + VERIFY_IS_EQUAL(a(4), 123); + } + { + Array a(123.0); + VERIFY_IS_EQUAL(a(4), 123); + } + { + Array a(123); + VERIFY_IS_EQUAL(a(4), 123.f); + } + { + MatrixXi m1(M,N); + VERIFY_IS_EQUAL(m1.rows(),M); + VERIFY_IS_EQUAL(m1.cols(),N); + ArrayXXi a1(M,N); + VERIFY_IS_EQUAL(a1.rows(),M); + VERIFY_IS_EQUAL(a1.cols(),N); + VectorXi v1(M); + VERIFY_IS_EQUAL(v1.size(),M); + ArrayXi a2(M); + VERIFY_IS_EQUAL(a2.size(),M); + } +} diff --git a/thirdparty/eigen/test/corners.cpp b/thirdparty/eigen/test/corners.cpp index 3c64c32a..73342a8d 100644 --- a/thirdparty/eigen/test/corners.cpp +++ b/thirdparty/eigen/test/corners.cpp @@ -15,7 +15,6 @@ template void corners(const MatrixType& m) { - typedef typename MatrixType::Index Index; Index rows = m.rows(); Index cols = m.cols(); @@ -102,7 +101,7 @@ template void c VERIFY_IS_EQUAL((const_matrix.template rightCols()), (const_matrix.template block(0,cols-c))); } -void test_corners() +EIGEN_DECLARE_TEST(corners) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( corners(Matrix()) ); diff --git a/thirdparty/eigen/test/ctorleak.cpp b/thirdparty/eigen/test/ctorleak.cpp new file mode 100644 index 00000000..73904176 --- /dev/null +++ b/thirdparty/eigen/test/ctorleak.cpp @@ -0,0 +1,81 @@ +#include "main.h" + +#include // std::exception + +struct Foo +{ + static Index object_count; + static Index object_limit; + int dummy; + + Foo() : dummy(0) + { +#ifdef EIGEN_EXCEPTIONS + // TODO: Is this the correct way to handle this? + if (Foo::object_count > Foo::object_limit) { std::cout << "\nThrow!\n"; throw Foo::Fail(); } +#endif + std::cout << '+'; + ++Foo::object_count; + } + + ~Foo() + { + std::cout << '-'; + --Foo::object_count; + } + + class Fail : public std::exception {}; +}; + +Index Foo::object_count = 0; +Index Foo::object_limit = 0; + +#undef EIGEN_TEST_MAX_SIZE +#define EIGEN_TEST_MAX_SIZE 3 + +EIGEN_DECLARE_TEST(ctorleak) +{ + typedef Matrix MatrixX; + typedef Matrix VectorX; + + Foo::object_count = 0; + for(int i = 0; i < g_repeat; i++) { + Index rows = internal::random(2,EIGEN_TEST_MAX_SIZE), cols = internal::random(2,EIGEN_TEST_MAX_SIZE); + Foo::object_limit = rows*cols; + { + MatrixX r(rows, cols); + Foo::object_limit = r.size()+internal::random(0, rows*cols - 2); + std::cout << "object_limit =" << Foo::object_limit << std::endl; +#ifdef EIGEN_EXCEPTIONS + try + { +#endif + if(internal::random()) { + std::cout << "\nMatrixX m(" << rows << ", " << cols << ");\n"; + MatrixX m(rows, cols); + } + else { + std::cout << "\nMatrixX m(r);\n"; + MatrixX m(r); + } +#ifdef EIGEN_EXCEPTIONS + VERIFY(false); // not reached if exceptions are enabled + } + catch (const Foo::Fail&) { /* ignore */ } +#endif + } + VERIFY_IS_EQUAL(Index(0), Foo::object_count); + + { + Foo::object_limit = (rows+1)*(cols+1); + MatrixX A(rows, cols); + VERIFY_IS_EQUAL(Foo::object_count, rows*cols); + VectorX v=A.row(0); + VERIFY_IS_EQUAL(Foo::object_count, (rows+1)*cols); + v = A.col(0); + VERIFY_IS_EQUAL(Foo::object_count, rows*(cols+1)); + } + VERIFY_IS_EQUAL(Index(0), Foo::object_count); + } + std::cout << "\n"; +} diff --git a/thirdparty/eigen/test/cwiseop.cpp b/thirdparty/eigen/test/cwiseop.cpp deleted file mode 100644 index d13002ca..00000000 --- a/thirdparty/eigen/test/cwiseop.cpp +++ /dev/null @@ -1,187 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2008 Gael Guennebaud -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#define EIGEN2_SUPPORT -#define EIGEN_NO_EIGEN2_DEPRECATED_WARNING - -#define EIGEN_NO_STATIC_ASSERT -#include "main.h" -#include - -#ifdef min -#undef min -#endif - -#ifdef max -#undef max -#endif - -using namespace std; - -template struct AddIfNull { - const Scalar operator() (const Scalar a, const Scalar b) const {return a<=1e-3 ? b : a;} - enum { Cost = NumTraits::AddCost }; -}; - -template -typename Eigen::internal::enable_if::IsInteger,typename MatrixType::Scalar>::type -cwiseops_real_only(MatrixType& m1, MatrixType& m2, MatrixType& m3, MatrixType& mones) -{ - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - - VERIFY_IS_APPROX(m1.cwise() / m2, m1.cwise() * (m2.cwise().inverse())); - m3 = m1.cwise().abs().cwise().sqrt(); - VERIFY_IS_APPROX(m3.cwise().square(), m1.cwise().abs()); - VERIFY_IS_APPROX(m1.cwise().square().cwise().sqrt(), m1.cwise().abs()); - VERIFY_IS_APPROX(m1.cwise().abs().cwise().log().cwise().exp() , m1.cwise().abs()); - - VERIFY_IS_APPROX(m1.cwise().pow(2), m1.cwise().square()); - m3 = (m1.cwise().abs().cwise()<=RealScalar(0.01)).select(mones,m1); - VERIFY_IS_APPROX(m3.cwise().pow(-1), m3.cwise().inverse()); - m3 = m1.cwise().abs(); - VERIFY_IS_APPROX(m3.cwise().pow(RealScalar(0.5)), m3.cwise().sqrt()); - -// VERIFY_IS_APPROX(m1.cwise().tan(), m1.cwise().sin().cwise() / m1.cwise().cos()); - VERIFY_IS_APPROX(mones, m1.cwise().sin().cwise().square() + m1.cwise().cos().cwise().square()); - m3 = m1; - m3.cwise() /= m2; - VERIFY_IS_APPROX(m3, m1.cwise() / m2); - - return Scalar(0); -} - -template -typename Eigen::internal::enable_if::IsInteger,typename MatrixType::Scalar>::type -cwiseops_real_only(MatrixType& , MatrixType& , MatrixType& , MatrixType& ) -{ - return 0; -} - -template void cwiseops(const MatrixType& m) -{ - typedef typename MatrixType::Index Index; - typedef typename MatrixType::Scalar Scalar; - typedef Matrix VectorType; - - Index rows = m.rows(); - Index cols = m.cols(); - - MatrixType m1 = MatrixType::Random(rows, cols), - m1bis = m1, - m2 = MatrixType::Random(rows, cols), - m3(rows, cols), - m4(rows, cols), - mzero = MatrixType::Zero(rows, cols), - mones = MatrixType::Ones(rows, cols), - identity = Matrix - ::Identity(rows, rows); - VectorType vzero = VectorType::Zero(rows), - vones = VectorType::Ones(rows), - v3(rows); - - Index r = internal::random(0, rows-1), - c = internal::random(0, cols-1); - - Scalar s1 = internal::random(); - - // test Zero, Ones, Constant, and the set* variants - m3 = MatrixType::Constant(rows, cols, s1); - for (int j=0; j >(mones); - - VERIFY_IS_APPROX(m1.cwise().pow(2), m1.cwise().abs2()); - VERIFY_IS_APPROX(m1.cwise().pow(2), m1.cwise().square()); - VERIFY_IS_APPROX(m1.cwise().pow(3), m1.cwise().cube()); - - VERIFY_IS_APPROX(m1 + mones, m1.cwise()+Scalar(1)); - VERIFY_IS_APPROX(m1 - mones, m1.cwise()-Scalar(1)); - m3 = m1; m3.cwise() += 1; - VERIFY_IS_APPROX(m1 + mones, m3); - m3 = m1; m3.cwise() -= 1; - VERIFY_IS_APPROX(m1 - mones, m3); - - VERIFY_IS_APPROX(m2, m2.cwise() * mones); - VERIFY_IS_APPROX(m1.cwise() * m2, m2.cwise() * m1); - m3 = m1; - m3.cwise() *= m2; - VERIFY_IS_APPROX(m3, m1.cwise() * m2); - - VERIFY_IS_APPROX(mones, m2.cwise()/m2); - - // check min - VERIFY_IS_APPROX( m1.cwise().min(m2), m2.cwise().min(m1) ); - VERIFY_IS_APPROX( m1.cwise().min(m1+mones), m1 ); - VERIFY_IS_APPROX( m1.cwise().min(m1-mones), m1-mones ); - - // check max - VERIFY_IS_APPROX( m1.cwise().max(m2), m2.cwise().max(m1) ); - VERIFY_IS_APPROX( m1.cwise().max(m1-mones), m1 ); - VERIFY_IS_APPROX( m1.cwise().max(m1+mones), m1+mones ); - - VERIFY( (m1.cwise() == m1).all() ); - VERIFY( (m1.cwise() != m2).any() ); - VERIFY(!(m1.cwise() == (m1+mones)).any() ); - if (rows*cols>1) - { - m3 = m1; - m3(r,c) += 1; - VERIFY( (m1.cwise() == m3).any() ); - VERIFY( !(m1.cwise() == m3).all() ); - } - VERIFY( (m1.cwise().min(m2).cwise() <= m2).all() ); - VERIFY( (m1.cwise().max(m2).cwise() >= m2).all() ); - VERIFY( (m1.cwise().min(m2).cwise() < (m1+mones)).all() ); - VERIFY( (m1.cwise().max(m2).cwise() > (m1-mones)).all() ); - -#if(__cplusplus < 201103L) -// std::binder* are deprecated since c++11 and will be removed in c++17 - VERIFY( (m1.cwise()(), Scalar(1)))).all() ); - VERIFY( !(m1.cwise()(), Scalar(1)))).all() ); - VERIFY( !(m1.cwise()>m1bis.unaryExpr(bind2nd(plus(), Scalar(1)))).any() ); -#endif - - cwiseops_real_only(m1, m2, m3, mones); -} - -void test_cwiseop() -{ - for(int i = 0; i < g_repeat ; i++) { - CALL_SUBTEST_1( cwiseops(Matrix()) ); - CALL_SUBTEST_2( cwiseops(Matrix4d()) ); - CALL_SUBTEST_3( cwiseops(MatrixXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - CALL_SUBTEST_4( cwiseops(MatrixXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - CALL_SUBTEST_5( cwiseops(MatrixXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - CALL_SUBTEST_6( cwiseops(MatrixXd(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); - } -} diff --git a/thirdparty/eigen/test/denseLM.cpp b/thirdparty/eigen/test/denseLM.cpp index 0aa736ea..afb8004b 100644 --- a/thirdparty/eigen/test/denseLM.cpp +++ b/thirdparty/eigen/test/denseLM.cpp @@ -182,7 +182,7 @@ void test_denseLM_T() } -void test_denseLM() +EIGEN_DECLARE_TEST(denseLM) { CALL_SUBTEST_2(test_denseLM_T()); diff --git a/thirdparty/eigen/test/dense_storage.cpp b/thirdparty/eigen/test/dense_storage.cpp new file mode 100644 index 00000000..45c2bd72 --- /dev/null +++ b/thirdparty/eigen/test/dense_storage.cpp @@ -0,0 +1,190 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2013 Hauke Heibel +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" +#include "AnnoyingScalar.h" +#include "SafeScalar.h" + +#include + +#if EIGEN_HAS_TYPE_TRAITS && EIGEN_HAS_CXX11 +using DenseStorageD3x3 = Eigen::DenseStorage; +static_assert(std::is_trivially_move_constructible::value, "DenseStorage not trivially_move_constructible"); +static_assert(std::is_trivially_move_assignable::value, "DenseStorage not trivially_move_assignable"); +#if !defined(EIGEN_DENSE_STORAGE_CTOR_PLUGIN) +static_assert(std::is_trivially_copy_constructible::value, "DenseStorage not trivially_copy_constructible"); +static_assert(std::is_trivially_copy_assignable::value, "DenseStorage not trivially_copy_assignable"); +static_assert(std::is_trivially_copyable::value, "DenseStorage not trivially_copyable"); +#endif +#endif + +template +void dense_storage_copy(int rows, int cols) +{ + typedef DenseStorage DenseStorageType; + + const int size = rows*cols; + DenseStorageType reference(size, rows, cols); + T* raw_reference = reference.data(); + for (int i=0; i(i); + + DenseStorageType copied_reference(reference); + const T* raw_copied_reference = copied_reference.data(); + for (int i=0; i +void dense_storage_assignment(int rows, int cols) +{ + typedef DenseStorage DenseStorageType; + + const int size = rows*cols; + DenseStorageType reference(size, rows, cols); + T* raw_reference = reference.data(); + for (int i=0; i(i); + + DenseStorageType copied_reference; + copied_reference = reference; + const T* raw_copied_reference = copied_reference.data(); + for (int i=0; i +void dense_storage_swap(int rows0, int cols0, int rows1, int cols1) +{ + typedef DenseStorage DenseStorageType; + + const int size0 = rows0*cols0; + DenseStorageType a(size0, rows0, cols0); + for (int i=0; i(i); + } + + const int size1 = rows1*cols1; + DenseStorageType b(size1, rows1, cols1); + for (int i=0; i(-i); + } + + a.swap(b); + + for (int i=0; i(i)); + } + + for (int i=0; i(-i)); + } +} + +template +void dense_storage_alignment() +{ + #if EIGEN_HAS_ALIGNAS + + struct alignas(Alignment) Empty1 {}; + VERIFY_IS_EQUAL(std::alignment_of::value, Alignment); + + struct EIGEN_ALIGN_TO_BOUNDARY(Alignment) Empty2 {}; + VERIFY_IS_EQUAL(std::alignment_of::value, Alignment); + + struct Nested1 { EIGEN_ALIGN_TO_BOUNDARY(Alignment) T data[Size]; }; + VERIFY_IS_EQUAL(std::alignment_of::value, Alignment); + + VERIFY_IS_EQUAL( (std::alignment_of >::value), Alignment); + + const std::size_t default_alignment = internal::compute_default_alignment::value; + + VERIFY_IS_EQUAL( (std::alignment_of >::value), default_alignment); + VERIFY_IS_EQUAL( (std::alignment_of >::value), default_alignment); + struct Nested2 { Matrix mat; }; + VERIFY_IS_EQUAL(std::alignment_of::value, default_alignment); + + #endif +} + +template +void dense_storage_tests() { + // Dynamic Storage. + dense_storage_copy(4, 3); + dense_storage_copy(4, 3); + dense_storage_copy(4, 3); + // Fixed Storage. + dense_storage_copy(4, 3); + dense_storage_copy(4, 3); + dense_storage_copy(4, 3); + dense_storage_copy(4, 3); + // Fixed Storage with Uninitialized Elements. + dense_storage_copy(4, 3); + dense_storage_copy(4, 3); + dense_storage_copy(4, 3); + + // Dynamic Storage. + dense_storage_assignment(4, 3); + dense_storage_assignment(4, 3); + dense_storage_assignment(4, 3); + // Fixed Storage. + dense_storage_assignment(4, 3); + dense_storage_assignment(4, 3); + dense_storage_assignment(4, 3); + dense_storage_assignment(4, 3); + // Fixed Storage with Uninitialized Elements. + dense_storage_assignment(4, 3); + dense_storage_assignment(4, 3); + dense_storage_assignment(4, 3); + + // Dynamic Storage. + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 2, 1); + dense_storage_swap(2, 1, 4, 3); + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 2, 3); + dense_storage_swap(2, 3, 4, 3); + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 4, 1); + dense_storage_swap(4, 1, 4, 3); + // Fixed Storage. + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 2, 1); + dense_storage_swap(2, 1, 4, 3); + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 4, 1); + dense_storage_swap(4, 1, 4, 3); + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 2, 3); + dense_storage_swap(2, 3, 4, 3); + // Fixed Storage with Uninitialized Elements. + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 2, 1); + dense_storage_swap(2, 1, 4, 3); + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 4, 1); + dense_storage_swap(4, 1, 4, 3); + dense_storage_swap(4, 3, 4, 3); + dense_storage_swap(4, 3, 2, 3); + dense_storage_swap(2, 3, 4, 3); + + dense_storage_alignment(); + dense_storage_alignment(); + dense_storage_alignment(); + dense_storage_alignment(); +} + +EIGEN_DECLARE_TEST(dense_storage) +{ + dense_storage_tests(); + dense_storage_tests(); + dense_storage_tests >(); + dense_storage_tests(); +} diff --git a/thirdparty/eigen/test/determinant.cpp b/thirdparty/eigen/test/determinant.cpp index 758f3afb..7dd33c37 100644 --- a/thirdparty/eigen/test/determinant.cpp +++ b/thirdparty/eigen/test/determinant.cpp @@ -16,7 +16,6 @@ template void determinant(const MatrixType& m) /* this test covers the following files: Determinant.h */ - typedef typename MatrixType::Index Index; Index size = m.rows(); MatrixType m1(size, size), m2(size, size); @@ -51,7 +50,7 @@ template void determinant(const MatrixType& m) VERIFY_IS_APPROX(m2.block(0,0,0,0).determinant(), Scalar(1)); } -void test_determinant() +EIGEN_DECLARE_TEST(determinant) { for(int i = 0; i < g_repeat; i++) { int s = 0; diff --git a/thirdparty/eigen/test/diagonal.cpp b/thirdparty/eigen/test/diagonal.cpp index 53814a58..4e8c4b3c 100644 --- a/thirdparty/eigen/test/diagonal.cpp +++ b/thirdparty/eigen/test/diagonal.cpp @@ -11,7 +11,6 @@ template void diagonal(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; Index rows = m.rows(); @@ -20,6 +19,8 @@ template void diagonal(const MatrixType& m) MatrixType m1 = MatrixType::Random(rows, cols), m2 = MatrixType::Random(rows, cols); + Scalar s1 = internal::random(); + //check diagonal() VERIFY_IS_APPROX(m1.diagonal(), m1.transpose().diagonal()); m2.diagonal() = 2 * m1.diagonal(); @@ -58,10 +59,36 @@ template void diagonal(const MatrixType& m) VERIFY_IS_APPROX(m2.template diagonal(), static_cast(2) * m1.diagonal(N2)); m2.diagonal(N2)[0] *= 3; VERIFY_IS_APPROX(m2.diagonal(N2)[0], static_cast(6) * m1.diagonal(N2)[0]); + + m2.diagonal(N2).x() = s1; + VERIFY_IS_APPROX(m2.diagonal(N2).x(), s1); + m2.diagonal(N2).coeffRef(0) = Scalar(2)*s1; + VERIFY_IS_APPROX(m2.diagonal(N2).coeff(0), Scalar(2)*s1); + } + + VERIFY( m1.diagonal( cols).size()==0 ); + VERIFY( m1.diagonal(-rows).size()==0 ); +} + +template void diagonal_assert(const MatrixType& m) { + Index rows = m.rows(); + Index cols = m.cols(); + + MatrixType m1 = MatrixType::Random(rows, cols); + + if (rows>=2 && cols>=2) + { + VERIFY_RAISES_ASSERT( m1 += m1.diagonal() ); + VERIFY_RAISES_ASSERT( m1 -= m1.diagonal() ); + VERIFY_RAISES_ASSERT( m1.array() *= m1.diagonal().array() ); + VERIFY_RAISES_ASSERT( m1.array() /= m1.diagonal().array() ); } + + VERIFY_RAISES_ASSERT( m1.diagonal(cols+1) ); + VERIFY_RAISES_ASSERT( m1.diagonal(-(rows+1)) ); } -void test_diagonal() +EIGEN_DECLARE_TEST(diagonal) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( diagonal(Matrix()) ); @@ -73,5 +100,6 @@ void test_diagonal() CALL_SUBTEST_2( diagonal(MatrixXcd(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_1( diagonal(MatrixXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_1( diagonal(Matrix(3, 4)) ); + CALL_SUBTEST_1( diagonal_assert(MatrixXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); } } diff --git a/thirdparty/eigen/test/diagonal_matrix_variadic_ctor.cpp b/thirdparty/eigen/test/diagonal_matrix_variadic_ctor.cpp new file mode 100644 index 00000000..fbc8f847 --- /dev/null +++ b/thirdparty/eigen/test/diagonal_matrix_variadic_ctor.cpp @@ -0,0 +1,185 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2019 David Tellenbach +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#define EIGEN_NO_STATIC_ASSERT + +#include "main.h" + +template +void assertionTest() +{ + typedef DiagonalMatrix DiagMatrix5; + typedef DiagonalMatrix DiagMatrix7; + typedef DiagonalMatrix DiagMatrixX; + + Scalar raw[6]; + for (int i = 0; i < 6; ++i) { + raw[i] = internal::random(); + } + + VERIFY_RAISES_ASSERT((DiagMatrix5{raw[0], raw[1], raw[2], raw[3]})); + VERIFY_RAISES_ASSERT((DiagMatrix5{raw[0], raw[1], raw[3]})); + VERIFY_RAISES_ASSERT((DiagMatrix7{raw[0], raw[1], raw[2], raw[3]})); + + VERIFY_RAISES_ASSERT((DiagMatrixX { + {raw[0], raw[1], raw[2]}, + {raw[3], raw[4], raw[5]} + })); +} + +#define VERIFY_IMPLICIT_CONVERSION_3(DIAGTYPE, V0, V1, V2) \ + DIAGTYPE d(V0, V1, V2); \ + DIAGTYPE::DenseMatrixType Dense = d.toDenseMatrix(); \ + VERIFY_IS_APPROX(Dense(0, 0), (Scalar)V0); \ + VERIFY_IS_APPROX(Dense(1, 1), (Scalar)V1); \ + VERIFY_IS_APPROX(Dense(2, 2), (Scalar)V2); + +#define VERIFY_IMPLICIT_CONVERSION_4(DIAGTYPE, V0, V1, V2, V3) \ + DIAGTYPE d(V0, V1, V2, V3); \ + DIAGTYPE::DenseMatrixType Dense = d.toDenseMatrix(); \ + VERIFY_IS_APPROX(Dense(0, 0), (Scalar)V0); \ + VERIFY_IS_APPROX(Dense(1, 1), (Scalar)V1); \ + VERIFY_IS_APPROX(Dense(2, 2), (Scalar)V2); \ + VERIFY_IS_APPROX(Dense(3, 3), (Scalar)V3); + +#define VERIFY_IMPLICIT_CONVERSION_5(DIAGTYPE, V0, V1, V2, V3, V4) \ + DIAGTYPE d(V0, V1, V2, V3, V4); \ + DIAGTYPE::DenseMatrixType Dense = d.toDenseMatrix(); \ + VERIFY_IS_APPROX(Dense(0, 0), (Scalar)V0); \ + VERIFY_IS_APPROX(Dense(1, 1), (Scalar)V1); \ + VERIFY_IS_APPROX(Dense(2, 2), (Scalar)V2); \ + VERIFY_IS_APPROX(Dense(3, 3), (Scalar)V3); \ + VERIFY_IS_APPROX(Dense(4, 4), (Scalar)V4); + +template +void constructorTest() +{ + typedef DiagonalMatrix DiagonalMatrix0; + typedef DiagonalMatrix DiagonalMatrix3; + typedef DiagonalMatrix DiagonalMatrix4; + typedef DiagonalMatrix DiagonalMatrixX; + + Scalar raw[7]; + for (int k = 0; k < 7; ++k) raw[k] = internal::random(); + + // Fixed-sized matrices + { + DiagonalMatrix0 a {{}}; + VERIFY(a.rows() == 0); + VERIFY(a.cols() == 0); + typename DiagonalMatrix0::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + { + DiagonalMatrix3 a {{raw[0], raw[1], raw[2]}}; + VERIFY(a.rows() == 3); + VERIFY(a.cols() == 3); + typename DiagonalMatrix3::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + { + DiagonalMatrix4 a {{raw[0], raw[1], raw[2], raw[3]}}; + VERIFY(a.rows() == 4); + VERIFY(a.cols() == 4); + typename DiagonalMatrix4::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + + // dynamically sized matrices + { + DiagonalMatrixX a{{}}; + VERIFY(a.rows() == 0); + VERIFY(a.rows() == 0); + typename DiagonalMatrixX::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + { + DiagonalMatrixX a{{raw[0], raw[1], raw[2], raw[3], raw[4], raw[5], raw[6]}}; + VERIFY(a.rows() == 7); + VERIFY(a.rows() == 7); + typename DiagonalMatrixX::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } +} + +template<> +void constructorTest() +{ + typedef float Scalar; + + typedef DiagonalMatrix DiagonalMatrix0; + typedef DiagonalMatrix DiagonalMatrix3; + typedef DiagonalMatrix DiagonalMatrix4; + typedef DiagonalMatrix DiagonalMatrix5; + typedef DiagonalMatrix DiagonalMatrixX; + + Scalar raw[7]; + for (int k = 0; k < 7; ++k) raw[k] = internal::random(); + + // Fixed-sized matrices + { + DiagonalMatrix0 a {{}}; + VERIFY(a.rows() == 0); + VERIFY(a.cols() == 0); + typename DiagonalMatrix0::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + { + DiagonalMatrix3 a {{raw[0], raw[1], raw[2]}}; + VERIFY(a.rows() == 3); + VERIFY(a.cols() == 3); + typename DiagonalMatrix3::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + { + DiagonalMatrix4 a {{raw[0], raw[1], raw[2], raw[3]}}; + VERIFY(a.rows() == 4); + VERIFY(a.cols() == 4); + typename DiagonalMatrix4::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + + // dynamically sized matrices + { + DiagonalMatrixX a{{}}; + VERIFY(a.rows() == 0); + VERIFY(a.rows() == 0); + typename DiagonalMatrixX::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + { + DiagonalMatrixX a{{raw[0], raw[1], raw[2], raw[3], raw[4], raw[5], raw[6]}}; + VERIFY(a.rows() == 7); + VERIFY(a.rows() == 7); + typename DiagonalMatrixX::DenseMatrixType m = a.toDenseMatrix(); + for (Index k = 0; k < a.rows(); ++k) VERIFY(m(k, k) == raw[k]); + } + { VERIFY_IMPLICIT_CONVERSION_3(DiagonalMatrix3, 1.2647, 2.56f, -3); } + { VERIFY_IMPLICIT_CONVERSION_4(DiagonalMatrix4, 1.2647, 2.56f, -3, 3.23f); } + { VERIFY_IMPLICIT_CONVERSION_5(DiagonalMatrix5, 1.2647, 2.56f, -3, 3.23f, 2); } +} + +EIGEN_DECLARE_TEST(diagonal_matrix_variadic_ctor) +{ + CALL_SUBTEST_1(assertionTest()); + CALL_SUBTEST_1(assertionTest()); + CALL_SUBTEST_1(assertionTest()); + CALL_SUBTEST_1(assertionTest()); + CALL_SUBTEST_1(assertionTest()); + CALL_SUBTEST_1(assertionTest()); + CALL_SUBTEST_1(assertionTest>()); + + CALL_SUBTEST_2(constructorTest()); + CALL_SUBTEST_2(constructorTest()); + CALL_SUBTEST_2(constructorTest()); + CALL_SUBTEST_2(constructorTest()); + CALL_SUBTEST_2(constructorTest()); + CALL_SUBTEST_2(constructorTest()); + CALL_SUBTEST_2(constructorTest>()); +} diff --git a/thirdparty/eigen/test/diagonalmatrices.cpp b/thirdparty/eigen/test/diagonalmatrices.cpp index 149f1db2..276beade 100644 --- a/thirdparty/eigen/test/diagonalmatrices.cpp +++ b/thirdparty/eigen/test/diagonalmatrices.cpp @@ -11,12 +11,12 @@ using namespace std; template void diagonalmatrices(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; enum { Rows = MatrixType::RowsAtCompileTime, Cols = MatrixType::ColsAtCompileTime }; typedef Matrix VectorType; typedef Matrix RowVectorType; typedef Matrix SquareMatrixType; + typedef Matrix DynMatrixType; typedef DiagonalMatrix LeftDiagonalMatrix; typedef DiagonalMatrix RightDiagonalMatrix; typedef Matrix BigMatrix; @@ -29,6 +29,7 @@ template void diagonalmatrices(const MatrixType& m) v2 = VectorType::Random(rows); RowVectorType rv1 = RowVectorType::Random(cols), rv2 = RowVectorType::Random(cols); + LeftDiagonalMatrix ldm1(v1), ldm2(v2); RightDiagonalMatrix rdm1(rv1), rdm2(rv2); @@ -64,6 +65,13 @@ template void diagonalmatrices(const MatrixType& m) VERIFY_IS_APPROX( (((v1+v2).asDiagonal() * (m1+m2))(i,j)) , (v1+v2)(i) * (m1+m2)(i,j) ); VERIFY_IS_APPROX( ((m1 * (rv1+rv2).asDiagonal())(i,j)) , (rv1+rv2)(j) * m1(i,j) ); VERIFY_IS_APPROX( (((m1+m2) * (rv1+rv2).asDiagonal())(i,j)) , (rv1+rv2)(j) * (m1+m2)(i,j) ); + + if(rows>1) + { + DynMatrixType tmp = m1.topRows(rows/2), res; + VERIFY_IS_APPROX( (res = m1.topRows(rows/2) * rv1.asDiagonal()), tmp * rv1.asDiagonal() ); + VERIFY_IS_APPROX( (res = v1.head(rows/2).asDiagonal()*m1.topRows(rows/2)), v1.head(rows/2).asDiagonal()*tmp ); + } BigMatrix big; big.setZero(2*rows, 2*cols); @@ -84,19 +92,82 @@ template void diagonalmatrices(const MatrixType& m) VERIFY_IS_APPROX(m1 * (rdm1 * s1), (m1 * rdm1) * s1); VERIFY_IS_APPROX(m1 * (s1 * rdm1), (m1 * rdm1) * s1); + + // Diagonal to dense + sq_m1.setRandom(); + sq_m2 = sq_m1; + VERIFY_IS_APPROX( (sq_m1 += (s1*v1).asDiagonal()), sq_m2 += (s1*v1).asDiagonal().toDenseMatrix() ); + VERIFY_IS_APPROX( (sq_m1 -= (s1*v1).asDiagonal()), sq_m2 -= (s1*v1).asDiagonal().toDenseMatrix() ); + VERIFY_IS_APPROX( (sq_m1 = (s1*v1).asDiagonal()), (s1*v1).asDiagonal().toDenseMatrix() ); + + sq_m1.setRandom(); + sq_m2 = v1.asDiagonal(); + sq_m2 = sq_m1 * sq_m2; + VERIFY_IS_APPROX( (sq_m1*v1.asDiagonal()).col(i), sq_m2.col(i) ); + VERIFY_IS_APPROX( (sq_m1*v1.asDiagonal()).row(i), sq_m2.row(i) ); + + sq_m1 = v1.asDiagonal(); + sq_m2 = v2.asDiagonal(); + SquareMatrixType sq_m3 = v1.asDiagonal(); + VERIFY_IS_APPROX( sq_m3 = v1.asDiagonal() + v2.asDiagonal(), sq_m1 + sq_m2); + VERIFY_IS_APPROX( sq_m3 = v1.asDiagonal() - v2.asDiagonal(), sq_m1 - sq_m2); + VERIFY_IS_APPROX( sq_m3 = v1.asDiagonal() - 2*v2.asDiagonal() + v1.asDiagonal(), sq_m1 - 2*sq_m2 + sq_m1); } -void test_diagonalmatrices() +template void as_scalar_product(const MatrixType& m) +{ + typedef typename MatrixType::Scalar Scalar; + typedef Matrix VectorType; + typedef Matrix DynMatrixType; + typedef Matrix DynVectorType; + typedef Matrix DynRowVectorType; + + Index rows = m.rows(); + Index depth = internal::random(1,EIGEN_TEST_MAX_SIZE); + + VectorType v1 = VectorType::Random(rows); + DynVectorType dv1 = DynVectorType::Random(depth); + DynRowVectorType drv1 = DynRowVectorType::Random(depth); + DynMatrixType dm1 = dv1; + DynMatrixType drm1 = drv1; + + Scalar s = v1(0); + + VERIFY_IS_APPROX( v1.asDiagonal() * drv1, s*drv1 ); + VERIFY_IS_APPROX( dv1 * v1.asDiagonal(), dv1*s ); + + VERIFY_IS_APPROX( v1.asDiagonal() * drm1, s*drm1 ); + VERIFY_IS_APPROX( dm1 * v1.asDiagonal(), dm1*s ); +} + +template +void bug987() +{ + Matrix3Xd points = Matrix3Xd::Random(3, 3); + Vector2d diag = Vector2d::Random(); + Matrix2Xd tmp1 = points.topRows<2>(), res1, res2; + VERIFY_IS_APPROX( res1 = diag.asDiagonal() * points.topRows<2>(), res2 = diag.asDiagonal() * tmp1 ); + Matrix2d tmp2 = points.topLeftCorner<2,2>(); + VERIFY_IS_APPROX(( res1 = points.topLeftCorner<2,2>()*diag.asDiagonal()) , res2 = tmp2*diag.asDiagonal() ); +} + +EIGEN_DECLARE_TEST(diagonalmatrices) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( diagonalmatrices(Matrix()) ); + CALL_SUBTEST_1( as_scalar_product(Matrix()) ); + CALL_SUBTEST_2( diagonalmatrices(Matrix3f()) ); CALL_SUBTEST_3( diagonalmatrices(Matrix()) ); CALL_SUBTEST_4( diagonalmatrices(Matrix4d()) ); CALL_SUBTEST_5( diagonalmatrices(Matrix()) ); CALL_SUBTEST_6( diagonalmatrices(MatrixXcf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_6( as_scalar_product(MatrixXcf(1,1)) ); CALL_SUBTEST_7( diagonalmatrices(MatrixXi(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_8( diagonalmatrices(Matrix(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_9( diagonalmatrices(MatrixXf(internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_9( diagonalmatrices(MatrixXf(1,1)) ); + CALL_SUBTEST_9( as_scalar_product(MatrixXf(1,1)) ); } + CALL_SUBTEST_10( bug987<0>() ); } diff --git a/thirdparty/eigen/test/dontalign.cpp b/thirdparty/eigen/test/dontalign.cpp index 4643cfed..2e4102b8 100644 --- a/thirdparty/eigen/test/dontalign.cpp +++ b/thirdparty/eigen/test/dontalign.cpp @@ -19,7 +19,6 @@ template void dontalign(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef Matrix VectorType; typedef Matrix SquareMatrixType; @@ -45,7 +44,7 @@ void dontalign(const MatrixType& m) internal::aligned_delete(array, rows); } -void test_dontalign() +EIGEN_DECLARE_TEST(dontalign) { #if defined EIGEN_TEST_PART_1 || defined EIGEN_TEST_PART_5 dontalign(Matrix3d()); diff --git a/thirdparty/eigen/test/dynalloc.cpp b/thirdparty/eigen/test/dynalloc.cpp index ef92c050..23c90a7b 100644 --- a/thirdparty/eigen/test/dynalloc.cpp +++ b/thirdparty/eigen/test/dynalloc.cpp @@ -9,18 +9,21 @@ #include "main.h" -#if EIGEN_ALIGN -#define ALIGNMENT 16 +#if EIGEN_MAX_ALIGN_BYTES>0 +#define ALIGNMENT EIGEN_MAX_ALIGN_BYTES #else #define ALIGNMENT 1 #endif +typedef Matrix Vector16f; +typedef Matrix Vector8f; + void check_handmade_aligned_malloc() { for(int i = 1; i < 1000; i++) { char *p = (char*)internal::handmade_aligned_malloc(i); - VERIFY(size_t(p)%ALIGNMENT==0); + VERIFY(internal::UIntPtr(p)%ALIGNMENT==0); // if the buffer is wrongly allocated this will give a bad write --> check with valgrind for(int j = 0; j < i; j++) p[j]=0; internal::handmade_aligned_free(p); @@ -29,10 +32,10 @@ void check_handmade_aligned_malloc() void check_aligned_malloc() { - for(int i = 1; i < 1000; i++) + for(int i = ALIGNMENT; i < 1000; i++) { char *p = (char*)internal::aligned_malloc(i); - VERIFY(size_t(p)%ALIGNMENT==0); + VERIFY(internal::UIntPtr(p)%ALIGNMENT==0); // if the buffer is wrongly allocated this will give a bad write --> check with valgrind for(int j = 0; j < i; j++) p[j]=0; internal::aligned_free(p); @@ -41,10 +44,10 @@ void check_aligned_malloc() void check_aligned_new() { - for(int i = 1; i < 1000; i++) + for(int i = ALIGNMENT; i < 1000; i++) { float *p = internal::aligned_new(i); - VERIFY(size_t(p)%ALIGNMENT==0); + VERIFY(internal::UIntPtr(p)%ALIGNMENT==0); // if the buffer is wrongly allocated this will give a bad write --> check with valgrind for(int j = 0; j < i; j++) p[j]=0; internal::aligned_delete(p,i); @@ -53,10 +56,10 @@ void check_aligned_new() void check_aligned_stack_alloc() { - for(int i = 1; i < 400; i++) + for(int i = ALIGNMENT; i < 400; i++) { ei_declare_aligned_stack_constructed_variable(float,p,i,0); - VERIFY(size_t(p)%ALIGNMENT==0); + VERIFY(internal::UIntPtr(p)%ALIGNMENT==0); // if the buffer is wrongly allocated this will give a bad write --> check with valgrind for(int j = 0; j < i; j++) p[j]=0; } @@ -68,7 +71,7 @@ struct MyStruct { EIGEN_MAKE_ALIGNED_OPERATOR_NEW char dummychar; - Vector4f avec; + Vector16f avec; }; class MyClassA @@ -76,15 +79,19 @@ class MyClassA public: EIGEN_MAKE_ALIGNED_OPERATOR_NEW char dummychar; - Vector4f avec; + Vector16f avec; }; template void check_dynaligned() { - T* obj = new T; - VERIFY(T::NeedsToAlign==1); - VERIFY(size_t(obj)%ALIGNMENT==0); - delete obj; + // TODO have to be updated once we support multiple alignment values + if(T::SizeAtCompileTime % ALIGNMENT == 0) + { + T* obj = new T; + VERIFY(T::NeedsToAlign==1); + VERIFY(internal::UIntPtr(obj)%ALIGNMENT==0); + delete obj; + } } template void check_custom_new_delete() @@ -100,7 +107,7 @@ template void check_custom_new_delete() delete[] t; } -#ifdef EIGEN_ALIGN +#if EIGEN_MAX_ALIGN_BYTES>0 && (!EIGEN_HAS_CXX17_OVERALIGN) { T* t = static_cast((T::operator new)(sizeof(T))); (T::operator delete)(t, sizeof(T)); @@ -113,16 +120,24 @@ template void check_custom_new_delete() #endif } -void test_dynalloc() +EIGEN_DECLARE_TEST(dynalloc) { // low level dynamic memory allocation CALL_SUBTEST(check_handmade_aligned_malloc()); CALL_SUBTEST(check_aligned_malloc()); CALL_SUBTEST(check_aligned_new()); CALL_SUBTEST(check_aligned_stack_alloc()); + + for (int i=0; i() ); + CALL_SUBTEST( check_custom_new_delete() ); + CALL_SUBTEST( check_custom_new_delete() ); + CALL_SUBTEST( check_custom_new_delete() ); + } // check static allocation, who knows ? - #if EIGEN_ALIGN_STATICALLY + #if EIGEN_MAX_STATIC_ALIGN_BYTES for (int i=0; i() ); @@ -130,23 +145,20 @@ void test_dynalloc() CALL_SUBTEST(check_dynaligned() ); CALL_SUBTEST(check_dynaligned() ); CALL_SUBTEST(check_dynaligned() ); - - CALL_SUBTEST( check_custom_new_delete() ); - CALL_SUBTEST( check_custom_new_delete() ); - CALL_SUBTEST( check_custom_new_delete() ); - CALL_SUBTEST( check_custom_new_delete() ); + CALL_SUBTEST(check_dynaligned() ); + CALL_SUBTEST(check_dynaligned() ); } { - MyStruct foo0; VERIFY(size_t(foo0.avec.data())%ALIGNMENT==0); - MyClassA fooA; VERIFY(size_t(fooA.avec.data())%ALIGNMENT==0); + MyStruct foo0; VERIFY(internal::UIntPtr(foo0.avec.data())%ALIGNMENT==0); + MyClassA fooA; VERIFY(internal::UIntPtr(fooA.avec.data())%ALIGNMENT==0); } // dynamic allocation, single object for (int i=0; iavec.data())%ALIGNMENT==0); - MyClassA *fooA = new MyClassA(); VERIFY(size_t(fooA->avec.data())%ALIGNMENT==0); + MyStruct *foo0 = new MyStruct(); VERIFY(internal::UIntPtr(foo0->avec.data())%ALIGNMENT==0); + MyClassA *fooA = new MyClassA(); VERIFY(internal::UIntPtr(fooA->avec.data())%ALIGNMENT==0); delete foo0; delete fooA; } @@ -155,8 +167,8 @@ void test_dynalloc() const int N = 10; for (int i=0; iavec.data())%ALIGNMENT==0); - MyClassA *fooA = new MyClassA[N]; VERIFY(size_t(fooA->avec.data())%ALIGNMENT==0); + MyStruct *foo0 = new MyStruct[N]; VERIFY(internal::UIntPtr(foo0->avec.data())%ALIGNMENT==0); + MyClassA *fooA = new MyClassA[N]; VERIFY(internal::UIntPtr(fooA->avec.data())%ALIGNMENT==0); delete[] foo0; delete[] fooA; } diff --git a/thirdparty/eigen/test/eigen2/CMakeLists.txt b/thirdparty/eigen/test/eigen2/CMakeLists.txt deleted file mode 100644 index 9615a602..00000000 --- a/thirdparty/eigen/test/eigen2/CMakeLists.txt +++ /dev/null @@ -1,61 +0,0 @@ -add_custom_target(eigen2_buildtests) -add_custom_target(eigen2_check COMMAND "ctest -R eigen2") -add_dependencies(eigen2_check eigen2_buildtests) -add_dependencies(buildtests eigen2_buildtests) - -add_definitions("-DEIGEN2_SUPPORT_STAGE10_FULL_EIGEN2_API") -add_definitions("-DEIGEN_NO_EIGEN2_DEPRECATED_WARNING") - -ei_add_test(eigen2_meta) -ei_add_test(eigen2_sizeof) -ei_add_test(eigen2_dynalloc) -ei_add_test(eigen2_nomalloc) -#ei_add_test(eigen2_first_aligned) -ei_add_test(eigen2_mixingtypes) -#ei_add_test(eigen2_packetmath) -ei_add_test(eigen2_unalignedassert) -#ei_add_test(eigen2_vectorization_logic) -ei_add_test(eigen2_basicstuff) -ei_add_test(eigen2_linearstructure) -ei_add_test(eigen2_cwiseop) -ei_add_test(eigen2_sum) -ei_add_test(eigen2_product_small) -ei_add_test(eigen2_product_large ${EI_OFLAG}) -ei_add_test(eigen2_adjoint) -ei_add_test(eigen2_submatrices) -ei_add_test(eigen2_miscmatrices) -ei_add_test(eigen2_commainitializer) -ei_add_test(eigen2_smallvectors) -ei_add_test(eigen2_map) -ei_add_test(eigen2_array) -ei_add_test(eigen2_triangular) -ei_add_test(eigen2_cholesky " " "${GSL_LIBRARIES}") -ei_add_test(eigen2_lu ${EI_OFLAG}) -ei_add_test(eigen2_determinant ${EI_OFLAG}) -ei_add_test(eigen2_inverse) -ei_add_test(eigen2_qr) -ei_add_test(eigen2_eigensolver " " "${GSL_LIBRARIES}") -ei_add_test(eigen2_svd) -ei_add_test(eigen2_geometry) -ei_add_test(eigen2_geometry_with_eigen2_prefix) -ei_add_test(eigen2_hyperplane) -ei_add_test(eigen2_parametrizedline) -ei_add_test(eigen2_alignedbox) -ei_add_test(eigen2_regression) -ei_add_test(eigen2_stdvector) -ei_add_test(eigen2_newstdvector) -if(QT4_FOUND) - ei_add_test(eigen2_qtvector " " "${QT_QTCORE_LIBRARY}") -endif(QT4_FOUND) -# no support for eigen2 sparse module -# if(NOT EIGEN_DEFAULT_TO_ROW_MAJOR) -# ei_add_test(eigen2_sparse_vector) -# ei_add_test(eigen2_sparse_basic) -# ei_add_test(eigen2_sparse_solvers " " "${SPARSE_LIBS}") -# ei_add_test(eigen2_sparse_product) -# endif() -ei_add_test(eigen2_swap) -ei_add_test(eigen2_visitor) -ei_add_test(eigen2_bug_132) - -ei_add_test(eigen2_prec_inverse_4x4 ${EI_OFLAG}) diff --git a/thirdparty/eigen/test/eigen2/eigen2_adjoint.cpp b/thirdparty/eigen/test/eigen2/eigen2_adjoint.cpp deleted file mode 100644 index c0f81145..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_adjoint.cpp +++ /dev/null @@ -1,99 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void adjoint(const MatrixType& m) -{ - /* this test covers the following files: - Transpose.h Conjugate.h Dot.h - */ - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - typedef Matrix SquareMatrixType; - int rows = m.rows(); - int cols = m.cols(); - - RealScalar largerEps = test_precision(); - if (ei_is_same_type::ret) - largerEps = RealScalar(1e-3f); - - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols), - square = SquareMatrixType::Random(rows, rows); - VectorType v1 = VectorType::Random(rows), - v2 = VectorType::Random(rows), - v3 = VectorType::Random(rows), - vzero = VectorType::Zero(rows); - - Scalar s1 = ei_random(), - s2 = ei_random(); - - // check basic compatibility of adjoint, transpose, conjugate - VERIFY_IS_APPROX(m1.transpose().conjugate().adjoint(), m1); - VERIFY_IS_APPROX(m1.adjoint().conjugate().transpose(), m1); - - // check multiplicative behavior - VERIFY_IS_APPROX((m1.adjoint() * m2).adjoint(), m2.adjoint() * m1); - VERIFY_IS_APPROX((s1 * m1).adjoint(), ei_conj(s1) * m1.adjoint()); - - // check basic properties of dot, norm, norm2 - typedef typename NumTraits::Real RealScalar; - VERIFY(ei_isApprox((s1 * v1 + s2 * v2).eigen2_dot(v3), s1 * v1.eigen2_dot(v3) + s2 * v2.eigen2_dot(v3), largerEps)); - VERIFY(ei_isApprox(v3.eigen2_dot(s1 * v1 + s2 * v2), ei_conj(s1)*v3.eigen2_dot(v1)+ei_conj(s2)*v3.eigen2_dot(v2), largerEps)); - VERIFY_IS_APPROX(ei_conj(v1.eigen2_dot(v2)), v2.eigen2_dot(v1)); - VERIFY_IS_APPROX(ei_real(v1.eigen2_dot(v1)), v1.squaredNorm()); - if(NumTraits::HasFloatingPoint) - VERIFY_IS_APPROX(v1.squaredNorm(), v1.norm() * v1.norm()); - VERIFY_IS_MUCH_SMALLER_THAN(ei_abs(vzero.eigen2_dot(v1)), static_cast(1)); - if(NumTraits::HasFloatingPoint) - VERIFY_IS_MUCH_SMALLER_THAN(vzero.norm(), static_cast(1)); - - // check compatibility of dot and adjoint - VERIFY(ei_isApprox(v1.eigen2_dot(square * v2), (square.adjoint() * v1).eigen2_dot(v2), largerEps)); - - // like in testBasicStuff, test operator() to check const-qualification - int r = ei_random(0, rows-1), - c = ei_random(0, cols-1); - VERIFY_IS_APPROX(m1.conjugate()(r,c), ei_conj(m1(r,c))); - VERIFY_IS_APPROX(m1.adjoint()(c,r), ei_conj(m1(r,c))); - - if(NumTraits::HasFloatingPoint) - { - // check that Random().normalized() works: tricky as the random xpr must be evaluated by - // normalized() in order to produce a consistent result. - VERIFY_IS_APPROX(VectorType::Random(rows).normalized().norm(), RealScalar(1)); - } - - // check inplace transpose - m3 = m1; - m3.transposeInPlace(); - VERIFY_IS_APPROX(m3,m1.transpose()); - m3.transposeInPlace(); - VERIFY_IS_APPROX(m3,m1); - -} - -void test_eigen2_adjoint() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( adjoint(Matrix()) ); - CALL_SUBTEST_2( adjoint(Matrix3d()) ); - CALL_SUBTEST_3( adjoint(Matrix4f()) ); - CALL_SUBTEST_4( adjoint(MatrixXcf(4, 4)) ); - CALL_SUBTEST_5( adjoint(MatrixXi(8, 12)) ); - CALL_SUBTEST_6( adjoint(MatrixXf(21, 21)) ); - } - // test a large matrix only once - CALL_SUBTEST_7( adjoint(Matrix()) ); -} - diff --git a/thirdparty/eigen/test/eigen2/eigen2_alignedbox.cpp b/thirdparty/eigen/test/eigen2/eigen2_alignedbox.cpp deleted file mode 100644 index 35043b95..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_alignedbox.cpp +++ /dev/null @@ -1,60 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include -#include -#include - -template void alignedbox(const BoxType& _box) -{ - /* this test covers the following files: - AlignedBox.h - */ - - const int dim = _box.dim(); - typedef typename BoxType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - - VectorType p0 = VectorType::Random(dim); - VectorType p1 = VectorType::Random(dim); - RealScalar s1 = ei_random(0,1); - - BoxType b0(dim); - BoxType b1(VectorType::Random(dim),VectorType::Random(dim)); - BoxType b2; - - b0.extend(p0); - b0.extend(p1); - VERIFY(b0.contains(p0*s1+(Scalar(1)-s1)*p1)); - VERIFY(!b0.contains(p0 + (1+s1)*(p1-p0))); - - (b2 = b0).extend(b1); - VERIFY(b2.contains(b0)); - VERIFY(b2.contains(b1)); - VERIFY_IS_APPROX(b2.clamp(b0), b0); - - // casting - const int Dim = BoxType::AmbientDimAtCompileTime; - typedef typename GetDifferentType::type OtherScalar; - AlignedBox hp1f = b0.template cast(); - VERIFY_IS_APPROX(hp1f.template cast(),b0); - AlignedBox hp1d = b0.template cast(); - VERIFY_IS_APPROX(hp1d.template cast(),b0); -} - -void test_eigen2_alignedbox() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( alignedbox(AlignedBox()) ); - CALL_SUBTEST_2( alignedbox(AlignedBox()) ); - CALL_SUBTEST_3( alignedbox(AlignedBox()) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_array.cpp b/thirdparty/eigen/test/eigen2/eigen2_array.cpp deleted file mode 100644 index c1ff40ce..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_array.cpp +++ /dev/null @@ -1,142 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -template void array(const MatrixType& m) -{ - /* this test covers the following files: - Array.cpp - */ - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - - int rows = m.rows(); - int cols = m.cols(); - - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols); - - Scalar s1 = ei_random(), - s2 = ei_random(); - - // scalar addition - VERIFY_IS_APPROX(m1.cwise() + s1, s1 + m1.cwise()); - VERIFY_IS_APPROX(m1.cwise() + s1, MatrixType::Constant(rows,cols,s1) + m1); - VERIFY_IS_APPROX((m1*Scalar(2)).cwise() - s2, (m1+m1) - MatrixType::Constant(rows,cols,s2) ); - m3 = m1; - m3.cwise() += s2; - VERIFY_IS_APPROX(m3, m1.cwise() + s2); - m3 = m1; - m3.cwise() -= s1; - VERIFY_IS_APPROX(m3, m1.cwise() - s1); - - // reductions - VERIFY_IS_APPROX(m1.colwise().sum().sum(), m1.sum()); - VERIFY_IS_APPROX(m1.rowwise().sum().sum(), m1.sum()); - if (!ei_isApprox(m1.sum(), (m1+m2).sum())) - VERIFY_IS_NOT_APPROX(((m1+m2).rowwise().sum()).sum(), m1.sum()); - VERIFY_IS_APPROX(m1.colwise().sum(), m1.colwise().redux(internal::scalar_sum_op())); -} - -template void comparisons(const MatrixType& m) -{ - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - - int rows = m.rows(); - int cols = m.cols(); - - int r = ei_random(0, rows-1), - c = ei_random(0, cols-1); - - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols); - - VERIFY(((m1.cwise() + Scalar(1)).cwise() > m1).all()); - VERIFY(((m1.cwise() - Scalar(1)).cwise() < m1).all()); - if (rows*cols>1) - { - m3 = m1; - m3(r,c) += 1; - VERIFY(! (m1.cwise() < m3).all() ); - VERIFY(! (m1.cwise() > m3).all() ); - } - - // comparisons to scalar - VERIFY( (m1.cwise() != (m1(r,c)+1) ).any() ); - VERIFY( (m1.cwise() > (m1(r,c)-1) ).any() ); - VERIFY( (m1.cwise() < (m1(r,c)+1) ).any() ); - VERIFY( (m1.cwise() == m1(r,c) ).any() ); - - // test Select - VERIFY_IS_APPROX( (m1.cwise()m2).select(m1,m2), m1.cwise().max(m2) ); - Scalar mid = (m1.cwise().abs().minCoeff() + m1.cwise().abs().maxCoeff())/Scalar(2); - for (int j=0; j=MatrixType::Constant(rows,cols,mid)) - .select(m1,0), m3); - // even shorter version: - VERIFY_IS_APPROX( (m1.cwise().abs().cwise()RealScalar(0.1)).count() == rows*cols); - VERIFY_IS_APPROX(((m1.cwise().abs().cwise()+1).cwise()>RealScalar(0.1)).colwise().count().template cast(), RowVectorXi::Constant(cols,rows)); - VERIFY_IS_APPROX(((m1.cwise().abs().cwise()+1).cwise()>RealScalar(0.1)).rowwise().count().template cast(), VectorXi::Constant(rows, cols)); -} - -template void lpNorm(const VectorType& v) -{ - VectorType u = VectorType::Random(v.size()); - - VERIFY_IS_APPROX(u.template lpNorm(), u.cwise().abs().maxCoeff()); - VERIFY_IS_APPROX(u.template lpNorm<1>(), u.cwise().abs().sum()); - VERIFY_IS_APPROX(u.template lpNorm<2>(), ei_sqrt(u.cwise().abs().cwise().square().sum())); - VERIFY_IS_APPROX(ei_pow(u.template lpNorm<5>(), typename VectorType::RealScalar(5)), u.cwise().abs().cwise().pow(5).sum()); -} - -void test_eigen2_array() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( array(Matrix()) ); - CALL_SUBTEST_2( array(Matrix2f()) ); - CALL_SUBTEST_3( array(Matrix4d()) ); - CALL_SUBTEST_4( array(MatrixXcf(3, 3)) ); - CALL_SUBTEST_5( array(MatrixXf(8, 12)) ); - CALL_SUBTEST_6( array(MatrixXi(8, 12)) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( comparisons(Matrix()) ); - CALL_SUBTEST_2( comparisons(Matrix2f()) ); - CALL_SUBTEST_3( comparisons(Matrix4d()) ); - CALL_SUBTEST_5( comparisons(MatrixXf(8, 12)) ); - CALL_SUBTEST_6( comparisons(MatrixXi(8, 12)) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( lpNorm(Matrix()) ); - CALL_SUBTEST_2( lpNorm(Vector2f()) ); - CALL_SUBTEST_3( lpNorm(Vector3d()) ); - CALL_SUBTEST_4( lpNorm(Vector4f()) ); - CALL_SUBTEST_5( lpNorm(VectorXf(16)) ); - CALL_SUBTEST_7( lpNorm(VectorXcd(10)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_basicstuff.cpp b/thirdparty/eigen/test/eigen2/eigen2_basicstuff.cpp deleted file mode 100644 index dd2dec1e..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_basicstuff.cpp +++ /dev/null @@ -1,105 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void basicStuff(const MatrixType& m) -{ - typedef typename MatrixType::Scalar Scalar; - typedef Matrix VectorType; - - int rows = m.rows(); - int cols = m.cols(); - - // this test relies a lot on Random.h, and there's not much more that we can do - // to test it, hence I consider that we will have tested Random.h - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols), - mzero = MatrixType::Zero(rows, cols), - square = Matrix::Random(rows, rows); - VectorType v1 = VectorType::Random(rows), - vzero = VectorType::Zero(rows); - - Scalar x = ei_random(); - - int r = ei_random(0, rows-1), - c = ei_random(0, cols-1); - - m1.coeffRef(r,c) = x; - VERIFY_IS_APPROX(x, m1.coeff(r,c)); - m1(r,c) = x; - VERIFY_IS_APPROX(x, m1(r,c)); - v1.coeffRef(r) = x; - VERIFY_IS_APPROX(x, v1.coeff(r)); - v1(r) = x; - VERIFY_IS_APPROX(x, v1(r)); - v1[r] = x; - VERIFY_IS_APPROX(x, v1[r]); - - VERIFY_IS_APPROX( v1, v1); - VERIFY_IS_NOT_APPROX( v1, 2*v1); - VERIFY_IS_MUCH_SMALLER_THAN( vzero, v1); - if(NumTraits::HasFloatingPoint) - VERIFY_IS_MUCH_SMALLER_THAN( vzero, v1.norm()); - VERIFY_IS_NOT_MUCH_SMALLER_THAN(v1, v1); - VERIFY_IS_APPROX( vzero, v1-v1); - VERIFY_IS_APPROX( m1, m1); - VERIFY_IS_NOT_APPROX( m1, 2*m1); - VERIFY_IS_MUCH_SMALLER_THAN( mzero, m1); - VERIFY_IS_NOT_MUCH_SMALLER_THAN(m1, m1); - VERIFY_IS_APPROX( mzero, m1-m1); - - // always test operator() on each read-only expression class, - // in order to check const-qualifiers. - // indeed, if an expression class (here Zero) is meant to be read-only, - // hence has no _write() method, the corresponding MatrixBase method (here zero()) - // should return a const-qualified object so that it is the const-qualified - // operator() that gets called, which in turn calls _read(). - VERIFY_IS_MUCH_SMALLER_THAN(MatrixType::Zero(rows,cols)(r,c), static_cast(1)); - - // now test copying a row-vector into a (column-)vector and conversely. - square.col(r) = square.row(r).eval(); - Matrix rv(rows); - Matrix cv(rows); - rv = square.row(r); - cv = square.col(r); - VERIFY_IS_APPROX(rv, cv.transpose()); - - if(cols!=1 && rows!=1 && MatrixType::SizeAtCompileTime!=Dynamic) - { - VERIFY_RAISES_ASSERT(m1 = (m2.block(0,0, rows-1, cols-1))); - } - - VERIFY_IS_APPROX(m3 = m1,m1); - MatrixType m4; - VERIFY_IS_APPROX(m4 = m1,m1); - - // test swap - m3 = m1; - m1.swap(m2); - VERIFY_IS_APPROX(m3, m2); - if(rows*cols>=3) - { - VERIFY_IS_NOT_APPROX(m3, m1); - } -} - -void test_eigen2_basicstuff() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( basicStuff(Matrix()) ); - CALL_SUBTEST_2( basicStuff(Matrix4d()) ); - CALL_SUBTEST_3( basicStuff(MatrixXcf(3, 3)) ); - CALL_SUBTEST_4( basicStuff(MatrixXi(8, 12)) ); - CALL_SUBTEST_5( basicStuff(MatrixXcd(20, 20)) ); - CALL_SUBTEST_6( basicStuff(Matrix()) ); - CALL_SUBTEST_7( basicStuff(Matrix(10,10)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_bug_132.cpp b/thirdparty/eigen/test/eigen2/eigen2_bug_132.cpp deleted file mode 100644 index 7fe3610c..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_bug_132.cpp +++ /dev/null @@ -1,26 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2010 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -void test_eigen2_bug_132() { - int size = 100; - MatrixXd A(size, size); - VectorXd b(size), c(size); - { - VectorXd y = A.transpose() * (b-c); // bug 132: infinite recursion in coeffRef - VectorXd z = (b-c).transpose() * A; // bug 132: infinite recursion in coeffRef - } - - // the following ones weren't failing, but let's include them for completeness: - { - VectorXd y = A * (b-c); - VectorXd z = (b-c).transpose() * A.transpose(); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_cholesky.cpp b/thirdparty/eigen/test/eigen2/eigen2_cholesky.cpp deleted file mode 100644 index 9c4b6f56..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_cholesky.cpp +++ /dev/null @@ -1,113 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#define EIGEN_NO_ASSERTION_CHECKING -#include "main.h" -#include -#include - -#ifdef HAS_GSL -#include "gsl_helper.h" -#endif - -template void cholesky(const MatrixType& m) -{ - /* this test covers the following files: - LLT.h LDLT.h - */ - int rows = m.rows(); - int cols = m.cols(); - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix SquareMatrixType; - typedef Matrix VectorType; - - MatrixType a0 = MatrixType::Random(rows,cols); - VectorType vecB = VectorType::Random(rows), vecX(rows); - MatrixType matB = MatrixType::Random(rows,cols), matX(rows,cols); - SquareMatrixType symm = a0 * a0.adjoint(); - // let's make sure the matrix is not singular or near singular - MatrixType a1 = MatrixType::Random(rows,cols); - symm += a1 * a1.adjoint(); - - #ifdef HAS_GSL - if (ei_is_same_type::ret) - { - typedef GslTraits Gsl; - typename Gsl::Matrix gMatA=0, gSymm=0; - typename Gsl::Vector gVecB=0, gVecX=0; - convert(symm, gSymm); - convert(symm, gMatA); - convert(vecB, gVecB); - convert(vecB, gVecX); - Gsl::cholesky(gMatA); - Gsl::cholesky_solve(gMatA, gVecB, gVecX); - VectorType vecX(rows), _vecX, _vecB; - convert(gVecX, _vecX); - symm.llt().solve(vecB, &vecX); - Gsl::prod(gSymm, gVecX, gVecB); - convert(gVecB, _vecB); - // test gsl itself ! - VERIFY_IS_APPROX(vecB, _vecB); - VERIFY_IS_APPROX(vecX, _vecX); - - Gsl::free(gMatA); - Gsl::free(gSymm); - Gsl::free(gVecB); - Gsl::free(gVecX); - } - #endif - - { - LDLT ldlt(symm); - VERIFY(ldlt.isPositiveDefinite()); - // in eigen3, LDLT is pivoting - //VERIFY_IS_APPROX(symm, ldlt.matrixL() * ldlt.vectorD().asDiagonal() * ldlt.matrixL().adjoint()); - ldlt.solve(vecB, &vecX); - VERIFY_IS_APPROX(symm * vecX, vecB); - ldlt.solve(matB, &matX); - VERIFY_IS_APPROX(symm * matX, matB); - } - - { - LLT chol(symm); - VERIFY(chol.isPositiveDefinite()); - VERIFY_IS_APPROX(symm, chol.matrixL() * chol.matrixL().adjoint()); - chol.solve(vecB, &vecX); - VERIFY_IS_APPROX(symm * vecX, vecB); - chol.solve(matB, &matX); - VERIFY_IS_APPROX(symm * matX, matB); - } - -#if 0 // cholesky is not rank-revealing anyway - // test isPositiveDefinite on non definite matrix - if (rows>4) - { - SquareMatrixType symm = a0.block(0,0,rows,cols-4) * a0.block(0,0,rows,cols-4).adjoint(); - LLT chol(symm); - VERIFY(!chol.isPositiveDefinite()); - LDLT cholnosqrt(symm); - VERIFY(!cholnosqrt.isPositiveDefinite()); - } -#endif -} - -void test_eigen2_cholesky() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( cholesky(Matrix()) ); - CALL_SUBTEST_2( cholesky(Matrix2d()) ); - CALL_SUBTEST_3( cholesky(Matrix3f()) ); - CALL_SUBTEST_4( cholesky(Matrix4d()) ); - CALL_SUBTEST_5( cholesky(MatrixXcd(7,7)) ); - CALL_SUBTEST_6( cholesky(MatrixXf(17,17)) ); - CALL_SUBTEST_7( cholesky(MatrixXd(33,33)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_commainitializer.cpp b/thirdparty/eigen/test/eigen2/eigen2_commainitializer.cpp deleted file mode 100644 index e0f901e0..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_commainitializer.cpp +++ /dev/null @@ -1,46 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -void test_eigen2_commainitializer() -{ - Matrix3d m3; - Matrix4d m4; - - VERIFY_RAISES_ASSERT( (m3 << 1, 2, 3, 4, 5, 6, 7, 8) ); - - #ifndef _MSC_VER - VERIFY_RAISES_ASSERT( (m3 << 1, 2, 3, 4, 5, 6, 7, 8, 9, 10) ); - #endif - - double data[] = {1, 2, 3, 4, 5, 6, 7, 8, 9}; - Matrix3d ref = Map >(data); - - m3 = Matrix3d::Random(); - m3 << 1, 2, 3, 4, 5, 6, 7, 8, 9; - VERIFY_IS_APPROX(m3, ref ); - - Vector3d vec[3]; - vec[0] << 1, 4, 7; - vec[1] << 2, 5, 8; - vec[2] << 3, 6, 9; - m3 = Matrix3d::Random(); - m3 << vec[0], vec[1], vec[2]; - VERIFY_IS_APPROX(m3, ref); - - vec[0] << 1, 2, 3; - vec[1] << 4, 5, 6; - vec[2] << 7, 8, 9; - m3 = Matrix3d::Random(); - m3 << vec[0].transpose(), - 4, 5, 6, - vec[2].transpose(); - VERIFY_IS_APPROX(m3, ref); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_cwiseop.cpp b/thirdparty/eigen/test/eigen2/eigen2_cwiseop.cpp deleted file mode 100644 index a36edd47..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_cwiseop.cpp +++ /dev/null @@ -1,158 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include -#include - -using namespace std; - -template struct AddIfNull { - const Scalar operator() (const Scalar a, const Scalar b) const {return a<=1e-3 ? b : a;} - enum { Cost = NumTraits::AddCost }; -}; - -template void cwiseops(const MatrixType& m) -{ - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - - int rows = m.rows(); - int cols = m.cols(); - - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols), - m4(rows, cols), - mzero = MatrixType::Zero(rows, cols), - mones = MatrixType::Ones(rows, cols), - identity = Matrix - ::Identity(rows, rows); - VectorType vzero = VectorType::Zero(rows), - vones = VectorType::Ones(rows), - v3(rows); - - int r = ei_random(0, rows-1), - c = ei_random(0, cols-1); - - Scalar s1 = ei_random(); - - // test Zero, Ones, Constant, and the set* variants - m3 = MatrixType::Constant(rows, cols, s1); - for (int j=0; j >(mones); - - VERIFY_IS_APPROX(m1.cwise().pow(2), m1.cwise().abs2()); - VERIFY_IS_APPROX(m1.cwise().pow(2), m1.cwise().square()); - VERIFY_IS_APPROX(m1.cwise().pow(3), m1.cwise().cube()); - - VERIFY_IS_APPROX(m1 + mones, m1.cwise()+Scalar(1)); - VERIFY_IS_APPROX(m1 - mones, m1.cwise()-Scalar(1)); - m3 = m1; m3.cwise() += 1; - VERIFY_IS_APPROX(m1 + mones, m3); - m3 = m1; m3.cwise() -= 1; - VERIFY_IS_APPROX(m1 - mones, m3); - - VERIFY_IS_APPROX(m2, m2.cwise() * mones); - VERIFY_IS_APPROX(m1.cwise() * m2, m2.cwise() * m1); - m3 = m1; - m3.cwise() *= m2; - VERIFY_IS_APPROX(m3, m1.cwise() * m2); - - VERIFY_IS_APPROX(mones, m2.cwise()/m2); - if(NumTraits::HasFloatingPoint) - { - VERIFY_IS_APPROX(m1.cwise() / m2, m1.cwise() * (m2.cwise().inverse())); - m3 = m1.cwise().abs().cwise().sqrt(); - VERIFY_IS_APPROX(m3.cwise().square(), m1.cwise().abs()); - VERIFY_IS_APPROX(m1.cwise().square().cwise().sqrt(), m1.cwise().abs()); - VERIFY_IS_APPROX(m1.cwise().abs().cwise().log().cwise().exp() , m1.cwise().abs()); - - VERIFY_IS_APPROX(m1.cwise().pow(2), m1.cwise().square()); - m3 = (m1.cwise().abs().cwise()<=RealScalar(0.01)).select(mones,m1); - VERIFY_IS_APPROX(m3.cwise().pow(-1), m3.cwise().inverse()); - m3 = m1.cwise().abs(); - VERIFY_IS_APPROX(m3.cwise().pow(RealScalar(0.5)), m3.cwise().sqrt()); - -// VERIFY_IS_APPROX(m1.cwise().tan(), m1.cwise().sin().cwise() / m1.cwise().cos()); - VERIFY_IS_APPROX(mones, m1.cwise().sin().cwise().square() + m1.cwise().cos().cwise().square()); - m3 = m1; - m3.cwise() /= m2; - VERIFY_IS_APPROX(m3, m1.cwise() / m2); - } - - // check min - VERIFY_IS_APPROX( m1.cwise().min(m2), m2.cwise().min(m1) ); - VERIFY_IS_APPROX( m1.cwise().min(m1+mones), m1 ); - VERIFY_IS_APPROX( m1.cwise().min(m1-mones), m1-mones ); - - // check max - VERIFY_IS_APPROX( m1.cwise().max(m2), m2.cwise().max(m1) ); - VERIFY_IS_APPROX( m1.cwise().max(m1-mones), m1 ); - VERIFY_IS_APPROX( m1.cwise().max(m1+mones), m1+mones ); - - VERIFY( (m1.cwise() == m1).all() ); - VERIFY( (m1.cwise() != m2).any() ); - VERIFY(!(m1.cwise() == (m1+mones)).any() ); - if (rows*cols>1) - { - m3 = m1; - m3(r,c) += 1; - VERIFY( (m1.cwise() == m3).any() ); - VERIFY( !(m1.cwise() == m3).all() ); - } - VERIFY( (m1.cwise().min(m2).cwise() <= m2).all() ); - VERIFY( (m1.cwise().max(m2).cwise() >= m2).all() ); - VERIFY( (m1.cwise().min(m2).cwise() < (m1+mones)).all() ); - VERIFY( (m1.cwise().max(m2).cwise() > (m1-mones)).all() ); - -#if(__cplusplus < 201103L) -// std::binder* are deprecated since c++11 and will be removed in c++17 - VERIFY( (m1.cwise()(), Scalar(1)))).all() ); - VERIFY( !(m1.cwise()(), Scalar(1)))).all() ); - VERIFY( !(m1.cwise()>m1.unaryExpr(bind2nd(plus(), Scalar(1)))).any() ); -#endif -} - -void test_eigen2_cwiseop() -{ - for(int i = 0; i < g_repeat ; i++) { - CALL_SUBTEST_1( cwiseops(Matrix()) ); - CALL_SUBTEST_2( cwiseops(Matrix4d()) ); - CALL_SUBTEST_3( cwiseops(MatrixXf(3, 3)) ); - CALL_SUBTEST_3( cwiseops(MatrixXf(22, 22)) ); - CALL_SUBTEST_4( cwiseops(MatrixXi(8, 12)) ); - CALL_SUBTEST_5( cwiseops(MatrixXd(20, 20)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_determinant.cpp b/thirdparty/eigen/test/eigen2/eigen2_determinant.cpp deleted file mode 100644 index c7b4ad05..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_determinant.cpp +++ /dev/null @@ -1,61 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Benoit Jacob -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -template void determinant(const MatrixType& m) -{ - /* this test covers the following files: - Determinant.h - */ - int size = m.rows(); - - MatrixType m1(size, size), m2(size, size); - m1.setRandom(); - m2.setRandom(); - typedef typename MatrixType::Scalar Scalar; - Scalar x = ei_random(); - VERIFY_IS_APPROX(MatrixType::Identity(size, size).determinant(), Scalar(1)); - VERIFY_IS_APPROX((m1*m2).determinant(), m1.determinant() * m2.determinant()); - if(size==1) return; - int i = ei_random(0, size-1); - int j; - do { - j = ei_random(0, size-1); - } while(j==i); - m2 = m1; - m2.row(i).swap(m2.row(j)); - VERIFY_IS_APPROX(m2.determinant(), -m1.determinant()); - m2 = m1; - m2.col(i).swap(m2.col(j)); - VERIFY_IS_APPROX(m2.determinant(), -m1.determinant()); - VERIFY_IS_APPROX(m2.determinant(), m2.transpose().determinant()); - VERIFY_IS_APPROX(ei_conj(m2.determinant()), m2.adjoint().determinant()); - m2 = m1; - m2.row(i) += x*m2.row(j); - VERIFY_IS_APPROX(m2.determinant(), m1.determinant()); - m2 = m1; - m2.row(i) *= x; - VERIFY_IS_APPROX(m2.determinant(), m1.determinant() * x); -} - -void test_eigen2_determinant() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( determinant(Matrix()) ); - CALL_SUBTEST_2( determinant(Matrix()) ); - CALL_SUBTEST_3( determinant(Matrix()) ); - CALL_SUBTEST_4( determinant(Matrix()) ); - CALL_SUBTEST_5( determinant(Matrix, 10, 10>()) ); - CALL_SUBTEST_6( determinant(MatrixXd(20, 20)) ); - } - CALL_SUBTEST_6( determinant(MatrixXd(200, 200)) ); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_dynalloc.cpp b/thirdparty/eigen/test/eigen2/eigen2_dynalloc.cpp deleted file mode 100644 index 1891a9e3..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_dynalloc.cpp +++ /dev/null @@ -1,131 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -#if EIGEN_ARCH_WANTS_ALIGNMENT -#define ALIGNMENT 16 -#else -#define ALIGNMENT 1 -#endif - -void check_handmade_aligned_malloc() -{ - for(int i = 1; i < 1000; i++) - { - char *p = (char*)ei_handmade_aligned_malloc(i); - VERIFY(std::size_t(p)%ALIGNMENT==0); - // if the buffer is wrongly allocated this will give a bad write --> check with valgrind - for(int j = 0; j < i; j++) p[j]=0; - ei_handmade_aligned_free(p); - } -} - -void check_aligned_malloc() -{ - for(int i = 1; i < 1000; i++) - { - char *p = (char*)ei_aligned_malloc(i); - VERIFY(std::size_t(p)%ALIGNMENT==0); - // if the buffer is wrongly allocated this will give a bad write --> check with valgrind - for(int j = 0; j < i; j++) p[j]=0; - ei_aligned_free(p); - } -} - -void check_aligned_new() -{ - for(int i = 1; i < 1000; i++) - { - float *p = ei_aligned_new(i); - VERIFY(std::size_t(p)%ALIGNMENT==0); - // if the buffer is wrongly allocated this will give a bad write --> check with valgrind - for(int j = 0; j < i; j++) p[j]=0; - ei_aligned_delete(p,i); - } -} - -void check_aligned_stack_alloc() -{ - for(int i = 1; i < 1000; i++) - { - ei_declare_aligned_stack_constructed_variable(float, p, i, 0); - VERIFY(std::size_t(p)%ALIGNMENT==0); - // if the buffer is wrongly allocated this will give a bad write --> check with valgrind - for(int j = 0; j < i; j++) p[j]=0; - } -} - - -// test compilation with both a struct and a class... -struct MyStruct -{ - EIGEN_MAKE_ALIGNED_OPERATOR_NEW - char dummychar; - Vector4f avec; -}; - -class MyClassA -{ - public: - EIGEN_MAKE_ALIGNED_OPERATOR_NEW - char dummychar; - Vector4f avec; -}; - -template void check_dynaligned() -{ - T* obj = new T; - VERIFY(std::size_t(obj)%ALIGNMENT==0); - delete obj; -} - -void test_eigen2_dynalloc() -{ - // low level dynamic memory allocation - CALL_SUBTEST(check_handmade_aligned_malloc()); - CALL_SUBTEST(check_aligned_malloc()); - CALL_SUBTEST(check_aligned_new()); - CALL_SUBTEST(check_aligned_stack_alloc()); - - for (int i=0; i() ); - CALL_SUBTEST( check_dynaligned() ); - CALL_SUBTEST( check_dynaligned() ); - CALL_SUBTEST( check_dynaligned() ); - CALL_SUBTEST( check_dynaligned() ); - } - - // check static allocation, who knows ? - { - MyStruct foo0; VERIFY(std::size_t(foo0.avec.data())%ALIGNMENT==0); - MyClassA fooA; VERIFY(std::size_t(fooA.avec.data())%ALIGNMENT==0); - } - - // dynamic allocation, single object - for (int i=0; iavec.data())%ALIGNMENT==0); - MyClassA *fooA = new MyClassA(); VERIFY(std::size_t(fooA->avec.data())%ALIGNMENT==0); - delete foo0; - delete fooA; - } - - // dynamic allocation, array - const int N = 10; - for (int i=0; iavec.data())%ALIGNMENT==0); - MyClassA *fooA = new MyClassA[N]; VERIFY(std::size_t(fooA->avec.data())%ALIGNMENT==0); - delete[] foo0; - delete[] fooA; - } - -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_eigensolver.cpp b/thirdparty/eigen/test/eigen2/eigen2_eigensolver.cpp deleted file mode 100644 index 48b4ace4..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_eigensolver.cpp +++ /dev/null @@ -1,146 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -#ifdef HAS_GSL -#include "gsl_helper.h" -#endif - -template void selfadjointeigensolver(const MatrixType& m) -{ - /* this test covers the following files: - EigenSolver.h, SelfAdjointEigenSolver.h (and indirectly: Tridiagonalization.h) - */ - int rows = m.rows(); - int cols = m.cols(); - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - typedef Matrix RealVectorType; - typedef typename std::complex::Real> Complex; - - RealScalar largerEps = 10*test_precision(); - - MatrixType a = MatrixType::Random(rows,cols); - MatrixType a1 = MatrixType::Random(rows,cols); - MatrixType symmA = a.adjoint() * a + a1.adjoint() * a1; - - MatrixType b = MatrixType::Random(rows,cols); - MatrixType b1 = MatrixType::Random(rows,cols); - MatrixType symmB = b.adjoint() * b + b1.adjoint() * b1; - - SelfAdjointEigenSolver eiSymm(symmA); - // generalized eigen pb - SelfAdjointEigenSolver eiSymmGen(symmA, symmB); - - #ifdef HAS_GSL - if (ei_is_same_type::ret) - { - typedef GslTraits Gsl; - typename Gsl::Matrix gEvec=0, gSymmA=0, gSymmB=0; - typename GslTraits::Vector gEval=0; - RealVectorType _eval; - MatrixType _evec; - convert(symmA, gSymmA); - convert(symmB, gSymmB); - convert(symmA, gEvec); - gEval = GslTraits::createVector(rows); - - Gsl::eigen_symm(gSymmA, gEval, gEvec); - convert(gEval, _eval); - convert(gEvec, _evec); - - // test gsl itself ! - VERIFY((symmA * _evec).isApprox(_evec * _eval.asDiagonal(), largerEps)); - - // compare with eigen - VERIFY_IS_APPROX(_eval, eiSymm.eigenvalues()); - VERIFY_IS_APPROX(_evec.cwise().abs(), eiSymm.eigenvectors().cwise().abs()); - - // generalized pb - Gsl::eigen_symm_gen(gSymmA, gSymmB, gEval, gEvec); - convert(gEval, _eval); - convert(gEvec, _evec); - // test GSL itself: - VERIFY((symmA * _evec).isApprox(symmB * (_evec * _eval.asDiagonal()), largerEps)); - - // compare with eigen - MatrixType normalized_eivec = eiSymmGen.eigenvectors()*eiSymmGen.eigenvectors().colwise().norm().asDiagonal().inverse(); - VERIFY_IS_APPROX(_eval, eiSymmGen.eigenvalues()); - VERIFY_IS_APPROX(_evec.cwiseAbs(), normalized_eivec.cwiseAbs()); - - Gsl::free(gSymmA); - Gsl::free(gSymmB); - GslTraits::free(gEval); - Gsl::free(gEvec); - } - #endif - - VERIFY((symmA * eiSymm.eigenvectors()).isApprox( - eiSymm.eigenvectors() * eiSymm.eigenvalues().asDiagonal(), largerEps)); - - // generalized eigen problem Ax = lBx - VERIFY((symmA * eiSymmGen.eigenvectors()).isApprox( - symmB * (eiSymmGen.eigenvectors() * eiSymmGen.eigenvalues().asDiagonal()), largerEps)); - - MatrixType sqrtSymmA = eiSymm.operatorSqrt(); - VERIFY_IS_APPROX(symmA, sqrtSymmA*sqrtSymmA); - VERIFY_IS_APPROX(sqrtSymmA, symmA*eiSymm.operatorInverseSqrt()); -} - -template void eigensolver(const MatrixType& m) -{ - /* this test covers the following files: - EigenSolver.h - */ - int rows = m.rows(); - int cols = m.cols(); - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - typedef Matrix RealVectorType; - typedef typename std::complex::Real> Complex; - - // RealScalar largerEps = 10*test_precision(); - - MatrixType a = MatrixType::Random(rows,cols); - MatrixType a1 = MatrixType::Random(rows,cols); - MatrixType symmA = a.adjoint() * a + a1.adjoint() * a1; - - EigenSolver ei0(symmA); - VERIFY_IS_APPROX(symmA * ei0.pseudoEigenvectors(), ei0.pseudoEigenvectors() * ei0.pseudoEigenvalueMatrix()); - VERIFY_IS_APPROX((symmA.template cast()) * (ei0.pseudoEigenvectors().template cast()), - (ei0.pseudoEigenvectors().template cast()) * (ei0.eigenvalues().asDiagonal())); - - EigenSolver ei1(a); - VERIFY_IS_APPROX(a * ei1.pseudoEigenvectors(), ei1.pseudoEigenvectors() * ei1.pseudoEigenvalueMatrix()); - VERIFY_IS_APPROX(a.template cast() * ei1.eigenvectors(), - ei1.eigenvectors() * ei1.eigenvalues().asDiagonal()); - -} - -void test_eigen2_eigensolver() -{ - for(int i = 0; i < g_repeat; i++) { - // very important to test a 3x3 matrix since we provide a special path for it - CALL_SUBTEST_1( selfadjointeigensolver(Matrix3f()) ); - CALL_SUBTEST_2( selfadjointeigensolver(Matrix4d()) ); - CALL_SUBTEST_3( selfadjointeigensolver(MatrixXf(7,7)) ); - CALL_SUBTEST_4( selfadjointeigensolver(MatrixXcd(5,5)) ); - CALL_SUBTEST_5( selfadjointeigensolver(MatrixXd(19,19)) ); - - CALL_SUBTEST_6( eigensolver(Matrix4f()) ); - CALL_SUBTEST_5( eigensolver(MatrixXd(17,17)) ); - } -} - diff --git a/thirdparty/eigen/test/eigen2/eigen2_first_aligned.cpp b/thirdparty/eigen/test/eigen2/eigen2_first_aligned.cpp deleted file mode 100644 index 51bb3cad..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_first_aligned.cpp +++ /dev/null @@ -1,49 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2009 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template -void test_eigen2_first_aligned_helper(Scalar *array, int size) -{ - const int packet_size = sizeof(Scalar) * ei_packet_traits::size; - VERIFY(((std::size_t(array) + sizeof(Scalar) * ei_alignmentOffset(array, size)) % packet_size) == 0); -} - -template -void test_eigen2_none_aligned_helper(Scalar *array, int size) -{ - VERIFY(ei_packet_traits::size == 1 || ei_alignmentOffset(array, size) == size); -} - -struct some_non_vectorizable_type { float x; }; - -void test_eigen2_first_aligned() -{ - EIGEN_ALIGN_128 float array_float[100]; - test_first_aligned_helper(array_float, 50); - test_first_aligned_helper(array_float+1, 50); - test_first_aligned_helper(array_float+2, 50); - test_first_aligned_helper(array_float+3, 50); - test_first_aligned_helper(array_float+4, 50); - test_first_aligned_helper(array_float+5, 50); - - EIGEN_ALIGN_128 double array_double[100]; - test_first_aligned_helper(array_double, 50); - test_first_aligned_helper(array_double+1, 50); - test_first_aligned_helper(array_double+2, 50); - - double *array_double_plus_4_bytes = (double*)(std::size_t(array_double)+4); - test_none_aligned_helper(array_double_plus_4_bytes, 50); - test_none_aligned_helper(array_double_plus_4_bytes+1, 50); - - some_non_vectorizable_type array_nonvec[100]; - test_first_aligned_helper(array_nonvec, 100); - test_none_aligned_helper(array_nonvec, 100); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_geometry.cpp b/thirdparty/eigen/test/eigen2/eigen2_geometry.cpp deleted file mode 100644 index 51404077..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_geometry.cpp +++ /dev/null @@ -1,432 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include -#include -#include - -template void geometry(void) -{ - /* this test covers the following files: - Cross.h Quaternion.h, Transform.cpp - */ - - typedef Matrix Matrix2; - typedef Matrix Matrix3; - typedef Matrix Matrix4; - typedef Matrix Vector2; - typedef Matrix Vector3; - typedef Matrix Vector4; - typedef Quaternion Quaternionx; - typedef AngleAxis AngleAxisx; - typedef Transform Transform2; - typedef Transform Transform3; - typedef Scaling Scaling2; - typedef Scaling Scaling3; - typedef Translation Translation2; - typedef Translation Translation3; - - Scalar largeEps = test_precision(); - if (ei_is_same_type::ret) - largeEps = 1e-2f; - - Vector3 v0 = Vector3::Random(), - v1 = Vector3::Random(), - v2 = Vector3::Random(); - Vector2 u0 = Vector2::Random(); - Matrix3 matrot1; - - Scalar a = ei_random(-Scalar(M_PI), Scalar(M_PI)); - - // cross product - VERIFY_IS_MUCH_SMALLER_THAN(v1.cross(v2).eigen2_dot(v1), Scalar(1)); - Matrix3 m; - m << v0.normalized(), - (v0.cross(v1)).normalized(), - (v0.cross(v1).cross(v0)).normalized(); - VERIFY(m.isUnitary()); - - // Quaternion: Identity(), setIdentity(); - Quaternionx q1, q2; - q2.setIdentity(); - VERIFY_IS_APPROX(Quaternionx(Quaternionx::Identity()).coeffs(), q2.coeffs()); - q1.coeffs().setRandom(); - VERIFY_IS_APPROX(q1.coeffs(), (q1*q2).coeffs()); - - // unitOrthogonal - VERIFY_IS_MUCH_SMALLER_THAN(u0.unitOrthogonal().eigen2_dot(u0), Scalar(1)); - VERIFY_IS_MUCH_SMALLER_THAN(v0.unitOrthogonal().eigen2_dot(v0), Scalar(1)); - VERIFY_IS_APPROX(u0.unitOrthogonal().norm(), Scalar(1)); - VERIFY_IS_APPROX(v0.unitOrthogonal().norm(), Scalar(1)); - - - VERIFY_IS_APPROX(v0, AngleAxisx(a, v0.normalized()) * v0); - VERIFY_IS_APPROX(-v0, AngleAxisx(Scalar(M_PI), v0.unitOrthogonal()) * v0); - VERIFY_IS_APPROX(ei_cos(a)*v0.squaredNorm(), v0.eigen2_dot(AngleAxisx(a, v0.unitOrthogonal()) * v0)); - m = AngleAxisx(a, v0.normalized()).toRotationMatrix().adjoint(); - VERIFY_IS_APPROX(Matrix3::Identity(), m * AngleAxisx(a, v0.normalized())); - VERIFY_IS_APPROX(Matrix3::Identity(), AngleAxisx(a, v0.normalized()) * m); - - q1 = AngleAxisx(a, v0.normalized()); - q2 = AngleAxisx(a, v1.normalized()); - - // angular distance - Scalar refangle = ei_abs(AngleAxisx(q1.inverse()*q2).angle()); - if (refangle>Scalar(M_PI)) - refangle = Scalar(2)*Scalar(M_PI) - refangle; - - if((q1.coeffs()-q2.coeffs()).norm() > 10*largeEps) - { - VERIFY(ei_isApprox(q1.angularDistance(q2), refangle, largeEps)); - } - - // rotation matrix conversion - VERIFY_IS_APPROX(q1 * v2, q1.toRotationMatrix() * v2); - VERIFY_IS_APPROX(q1 * q2 * v2, - q1.toRotationMatrix() * q2.toRotationMatrix() * v2); - - VERIFY( (q2*q1).isApprox(q1*q2, largeEps) || !(q2 * q1 * v2).isApprox( - q1.toRotationMatrix() * q2.toRotationMatrix() * v2)); - - q2 = q1.toRotationMatrix(); - VERIFY_IS_APPROX(q1*v1,q2*v1); - - matrot1 = AngleAxisx(Scalar(0.1), Vector3::UnitX()) - * AngleAxisx(Scalar(0.2), Vector3::UnitY()) - * AngleAxisx(Scalar(0.3), Vector3::UnitZ()); - VERIFY_IS_APPROX(matrot1 * v1, - AngleAxisx(Scalar(0.1), Vector3(1,0,0)).toRotationMatrix() - * (AngleAxisx(Scalar(0.2), Vector3(0,1,0)).toRotationMatrix() - * (AngleAxisx(Scalar(0.3), Vector3(0,0,1)).toRotationMatrix() * v1))); - - // angle-axis conversion - AngleAxisx aa = q1; - VERIFY_IS_APPROX(q1 * v1, Quaternionx(aa) * v1); - VERIFY_IS_NOT_APPROX(q1 * v1, Quaternionx(AngleAxisx(aa.angle()*2,aa.axis())) * v1); - - // from two vector creation - VERIFY_IS_APPROX(v2.normalized(),(q2.setFromTwoVectors(v1,v2)*v1).normalized()); - VERIFY_IS_APPROX(v2.normalized(),(q2.setFromTwoVectors(v1,v2)*v1).normalized()); - - // inverse and conjugate - VERIFY_IS_APPROX(q1 * (q1.inverse() * v1), v1); - VERIFY_IS_APPROX(q1 * (q1.conjugate() * v1), v1); - - // AngleAxis - VERIFY_IS_APPROX(AngleAxisx(a,v1.normalized()).toRotationMatrix(), - Quaternionx(AngleAxisx(a,v1.normalized())).toRotationMatrix()); - - AngleAxisx aa1; - m = q1.toRotationMatrix(); - aa1 = m; - VERIFY_IS_APPROX(AngleAxisx(m).toRotationMatrix(), - Quaternionx(m).toRotationMatrix()); - - // Transform - // TODO complete the tests ! - a = 0; - while (ei_abs(a)(-Scalar(0.4)*Scalar(M_PI), Scalar(0.4)*Scalar(M_PI)); - q1 = AngleAxisx(a, v0.normalized()); - Transform3 t0, t1, t2; - // first test setIdentity() and Identity() - t0.setIdentity(); - VERIFY_IS_APPROX(t0.matrix(), Transform3::MatrixType::Identity()); - t0.matrix().setZero(); - t0 = Transform3::Identity(); - VERIFY_IS_APPROX(t0.matrix(), Transform3::MatrixType::Identity()); - - t0.linear() = q1.toRotationMatrix(); - t1.setIdentity(); - t1.linear() = q1.toRotationMatrix(); - - v0 << 50, 2, 1;//= ei_random_matrix().cwiseProduct(Vector3(10,2,0.5)); - t0.scale(v0); - t1.prescale(v0); - - VERIFY_IS_APPROX( (t0 * Vector3(1,0,0)).norm(), v0.x()); - //VERIFY(!ei_isApprox((t1 * Vector3(1,0,0)).norm(), v0.x())); - - t0.setIdentity(); - t1.setIdentity(); - v1 << 1, 2, 3; - t0.linear() = q1.toRotationMatrix(); - t0.pretranslate(v0); - t0.scale(v1); - t1.linear() = q1.conjugate().toRotationMatrix(); - t1.prescale(v1.cwise().inverse()); - t1.translate(-v0); - - VERIFY((t0.matrix() * t1.matrix()).isIdentity(test_precision())); - - t1.fromPositionOrientationScale(v0, q1, v1); - VERIFY_IS_APPROX(t1.matrix(), t0.matrix()); - VERIFY_IS_APPROX(t1*v1, t0*v1); - - t0.setIdentity(); t0.scale(v0).rotate(q1.toRotationMatrix()); - t1.setIdentity(); t1.scale(v0).rotate(q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - t0.setIdentity(); t0.scale(v0).rotate(AngleAxisx(q1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - VERIFY_IS_APPROX(t0.scale(a).matrix(), t1.scale(Vector3::Constant(a)).matrix()); - VERIFY_IS_APPROX(t0.prescale(a).matrix(), t1.prescale(Vector3::Constant(a)).matrix()); - - // More transform constructors, operator=, operator*= - - Matrix3 mat3 = Matrix3::Random(); - Matrix4 mat4; - mat4 << mat3 , Vector3::Zero() , Vector4::Zero().transpose(); - Transform3 tmat3(mat3), tmat4(mat4); - tmat4.matrix()(3,3) = Scalar(1); - VERIFY_IS_APPROX(tmat3.matrix(), tmat4.matrix()); - - Scalar a3 = ei_random(-Scalar(M_PI), Scalar(M_PI)); - Vector3 v3 = Vector3::Random().normalized(); - AngleAxisx aa3(a3, v3); - Transform3 t3(aa3); - Transform3 t4; - t4 = aa3; - VERIFY_IS_APPROX(t3.matrix(), t4.matrix()); - t4.rotate(AngleAxisx(-a3,v3)); - VERIFY_IS_APPROX(t4.matrix(), Matrix4::Identity()); - t4 *= aa3; - VERIFY_IS_APPROX(t3.matrix(), t4.matrix()); - - v3 = Vector3::Random(); - Translation3 tv3(v3); - Transform3 t5(tv3); - t4 = tv3; - VERIFY_IS_APPROX(t5.matrix(), t4.matrix()); - t4.translate(-v3); - VERIFY_IS_APPROX(t4.matrix(), Matrix4::Identity()); - t4 *= tv3; - VERIFY_IS_APPROX(t5.matrix(), t4.matrix()); - - Scaling3 sv3(v3); - Transform3 t6(sv3); - t4 = sv3; - VERIFY_IS_APPROX(t6.matrix(), t4.matrix()); - t4.scale(v3.cwise().inverse()); - VERIFY_IS_APPROX(t4.matrix(), Matrix4::Identity()); - t4 *= sv3; - VERIFY_IS_APPROX(t6.matrix(), t4.matrix()); - - // matrix * transform - VERIFY_IS_APPROX(Transform3(t3.matrix()*t4).matrix(), Transform3(t3*t4).matrix()); - - // chained Transform product - VERIFY_IS_APPROX(((t3*t4)*t5).matrix(), (t3*(t4*t5)).matrix()); - - // check that Transform product doesn't have aliasing problems - t5 = t4; - t5 = t5*t5; - VERIFY_IS_APPROX(t5, t4*t4); - - // 2D transformation - Transform2 t20, t21; - Vector2 v20 = Vector2::Random(); - Vector2 v21 = Vector2::Random(); - for (int k=0; k<2; ++k) - if (ei_abs(v21[k])(a).toRotationMatrix(); - VERIFY_IS_APPROX(t20.fromPositionOrientationScale(v20,a,v21).matrix(), - t21.pretranslate(v20).scale(v21).matrix()); - - t21.setIdentity(); - t21.linear() = Rotation2D(-a).toRotationMatrix(); - VERIFY( (t20.fromPositionOrientationScale(v20,a,v21) - * (t21.prescale(v21.cwise().inverse()).translate(-v20))).matrix().isIdentity(test_precision()) ); - - // Transform - new API - // 3D - t0.setIdentity(); - t0.rotate(q1).scale(v0).translate(v0); - // mat * scaling and mat * translation - t1 = (Matrix3(q1) * Scaling3(v0)) * Translation3(v0); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // mat * transformation and scaling * translation - t1 = Matrix3(q1) * (Scaling3(v0) * Translation3(v0)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - t0.setIdentity(); - t0.prerotate(q1).prescale(v0).pretranslate(v0); - // translation * scaling and transformation * mat - t1 = (Translation3(v0) * Scaling3(v0)) * Matrix3(q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // scaling * mat and translation * mat - t1 = Translation3(v0) * (Scaling3(v0) * Matrix3(q1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - t0.setIdentity(); - t0.scale(v0).translate(v0).rotate(q1); - // translation * mat and scaling * transformation - t1 = Scaling3(v0) * (Translation3(v0) * Matrix3(q1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // transformation * scaling - t0.scale(v0); - t1 = t1 * Scaling3(v0); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // transformation * translation - t0.translate(v0); - t1 = t1 * Translation3(v0); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // translation * transformation - t0.pretranslate(v0); - t1 = Translation3(v0) * t1; - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // transform * quaternion - t0.rotate(q1); - t1 = t1 * q1; - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // translation * quaternion - t0.translate(v1).rotate(q1); - t1 = t1 * (Translation3(v1) * q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // scaling * quaternion - t0.scale(v1).rotate(q1); - t1 = t1 * (Scaling3(v1) * q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // quaternion * transform - t0.prerotate(q1); - t1 = q1 * t1; - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // quaternion * translation - t0.rotate(q1).translate(v1); - t1 = t1 * (q1 * Translation3(v1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // quaternion * scaling - t0.rotate(q1).scale(v1); - t1 = t1 * (q1 * Scaling3(v1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // translation * vector - t0.setIdentity(); - t0.translate(v0); - VERIFY_IS_APPROX(t0 * v1, Translation3(v0) * v1); - - // scaling * vector - t0.setIdentity(); - t0.scale(v0); - VERIFY_IS_APPROX(t0 * v1, Scaling3(v0) * v1); - - // test transform inversion - t0.setIdentity(); - t0.translate(v0); - t0.linear().setRandom(); - VERIFY_IS_APPROX(t0.inverse(Affine), t0.matrix().inverse()); - t0.setIdentity(); - t0.translate(v0).rotate(q1); - VERIFY_IS_APPROX(t0.inverse(Isometry), t0.matrix().inverse()); - - // test extract rotation and scaling - t0.setIdentity(); - t0.translate(v0).rotate(q1).scale(v1); - VERIFY_IS_APPROX(t0.rotation() * v1, Matrix3(q1) * v1); - - Matrix3 mat_rotation, mat_scaling; - t0.setIdentity(); - t0.translate(v0).rotate(q1).scale(v1); - t0.computeRotationScaling(&mat_rotation, &mat_scaling); - VERIFY_IS_APPROX(t0.linear(), mat_rotation * mat_scaling); - VERIFY_IS_APPROX(mat_rotation*mat_rotation.adjoint(), Matrix3::Identity()); - VERIFY_IS_APPROX(mat_rotation.determinant(), Scalar(1)); - t0.computeScalingRotation(&mat_scaling, &mat_rotation); - VERIFY_IS_APPROX(t0.linear(), mat_scaling * mat_rotation); - VERIFY_IS_APPROX(mat_rotation*mat_rotation.adjoint(), Matrix3::Identity()); - VERIFY_IS_APPROX(mat_rotation.determinant(), Scalar(1)); - - // test casting - Transform t1f = t1.template cast(); - VERIFY_IS_APPROX(t1f.template cast(),t1); - Transform t1d = t1.template cast(); - VERIFY_IS_APPROX(t1d.template cast(),t1); - - Translation3 tr1(v0); - Translation tr1f = tr1.template cast(); - VERIFY_IS_APPROX(tr1f.template cast(),tr1); - Translation tr1d = tr1.template cast(); - VERIFY_IS_APPROX(tr1d.template cast(),tr1); - - Scaling3 sc1(v0); - Scaling sc1f = sc1.template cast(); - VERIFY_IS_APPROX(sc1f.template cast(),sc1); - Scaling sc1d = sc1.template cast(); - VERIFY_IS_APPROX(sc1d.template cast(),sc1); - - Quaternion q1f = q1.template cast(); - VERIFY_IS_APPROX(q1f.template cast(),q1); - Quaternion q1d = q1.template cast(); - VERIFY_IS_APPROX(q1d.template cast(),q1); - - AngleAxis aa1f = aa1.template cast(); - VERIFY_IS_APPROX(aa1f.template cast(),aa1); - AngleAxis aa1d = aa1.template cast(); - VERIFY_IS_APPROX(aa1d.template cast(),aa1); - - Rotation2D r2d1(ei_random()); - Rotation2D r2d1f = r2d1.template cast(); - VERIFY_IS_APPROX(r2d1f.template cast(),r2d1); - Rotation2D r2d1d = r2d1.template cast(); - VERIFY_IS_APPROX(r2d1d.template cast(),r2d1); - - m = q1; -// m.col(1) = Vector3(0,ei_random(),ei_random()).normalized(); -// m.col(0) = Vector3(-1,0,0).normalized(); -// m.col(2) = m.col(0).cross(m.col(1)); - #define VERIFY_EULER(I,J,K, X,Y,Z) { \ - Vector3 ea = m.eulerAngles(I,J,K); \ - Matrix3 m1 = Matrix3(AngleAxisx(ea[0], Vector3::Unit##X()) * AngleAxisx(ea[1], Vector3::Unit##Y()) * AngleAxisx(ea[2], Vector3::Unit##Z())); \ - VERIFY_IS_APPROX(m, m1); \ - VERIFY_IS_APPROX(m, Matrix3(AngleAxisx(ea[0], Vector3::Unit##X()) * AngleAxisx(ea[1], Vector3::Unit##Y()) * AngleAxisx(ea[2], Vector3::Unit##Z()))); \ - } - VERIFY_EULER(0,1,2, X,Y,Z); - VERIFY_EULER(0,1,0, X,Y,X); - VERIFY_EULER(0,2,1, X,Z,Y); - VERIFY_EULER(0,2,0, X,Z,X); - - VERIFY_EULER(1,2,0, Y,Z,X); - VERIFY_EULER(1,2,1, Y,Z,Y); - VERIFY_EULER(1,0,2, Y,X,Z); - VERIFY_EULER(1,0,1, Y,X,Y); - - VERIFY_EULER(2,0,1, Z,X,Y); - VERIFY_EULER(2,0,2, Z,X,Z); - VERIFY_EULER(2,1,0, Z,Y,X); - VERIFY_EULER(2,1,2, Z,Y,Z); - - // colwise/rowwise cross product - mat3.setRandom(); - Vector3 vec3 = Vector3::Random(); - Matrix3 mcross; - int i = ei_random(0,2); - mcross = mat3.colwise().cross(vec3); - VERIFY_IS_APPROX(mcross.col(i), mat3.col(i).cross(vec3)); - mcross = mat3.rowwise().cross(vec3); - VERIFY_IS_APPROX(mcross.row(i), mat3.row(i).cross(vec3)); - - -} - -void test_eigen2_geometry() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( geometry() ); - CALL_SUBTEST_2( geometry() ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_geometry_with_eigen2_prefix.cpp b/thirdparty/eigen/test/eigen2/eigen2_geometry_with_eigen2_prefix.cpp deleted file mode 100644 index 12d4a71c..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_geometry_with_eigen2_prefix.cpp +++ /dev/null @@ -1,435 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#define EIGEN2_SUPPORT_STAGE15_RESOLVE_API_CONFLICTS_WARN - -#include "main.h" -#include -#include -#include - -template void geometry(void) -{ - /* this test covers the following files: - Cross.h Quaternion.h, Transform.cpp - */ - - typedef Matrix Matrix2; - typedef Matrix Matrix3; - typedef Matrix Matrix4; - typedef Matrix Vector2; - typedef Matrix Vector3; - typedef Matrix Vector4; - typedef eigen2_Quaternion Quaternionx; - typedef eigen2_AngleAxis AngleAxisx; - typedef eigen2_Transform Transform2; - typedef eigen2_Transform Transform3; - typedef eigen2_Scaling Scaling2; - typedef eigen2_Scaling Scaling3; - typedef eigen2_Translation Translation2; - typedef eigen2_Translation Translation3; - - Scalar largeEps = test_precision(); - if (ei_is_same_type::ret) - largeEps = 1e-2f; - - Vector3 v0 = Vector3::Random(), - v1 = Vector3::Random(), - v2 = Vector3::Random(); - Vector2 u0 = Vector2::Random(); - Matrix3 matrot1; - - Scalar a = ei_random(-Scalar(M_PI), Scalar(M_PI)); - - // cross product - VERIFY_IS_MUCH_SMALLER_THAN(v1.cross(v2).eigen2_dot(v1), Scalar(1)); - Matrix3 m; - m << v0.normalized(), - (v0.cross(v1)).normalized(), - (v0.cross(v1).cross(v0)).normalized(); - VERIFY(m.isUnitary()); - - // Quaternion: Identity(), setIdentity(); - Quaternionx q1, q2; - q2.setIdentity(); - VERIFY_IS_APPROX(Quaternionx(Quaternionx::Identity()).coeffs(), q2.coeffs()); - q1.coeffs().setRandom(); - VERIFY_IS_APPROX(q1.coeffs(), (q1*q2).coeffs()); - - // unitOrthogonal - VERIFY_IS_MUCH_SMALLER_THAN(u0.unitOrthogonal().eigen2_dot(u0), Scalar(1)); - VERIFY_IS_MUCH_SMALLER_THAN(v0.unitOrthogonal().eigen2_dot(v0), Scalar(1)); - VERIFY_IS_APPROX(u0.unitOrthogonal().norm(), Scalar(1)); - VERIFY_IS_APPROX(v0.unitOrthogonal().norm(), Scalar(1)); - - - VERIFY_IS_APPROX(v0, AngleAxisx(a, v0.normalized()) * v0); - VERIFY_IS_APPROX(-v0, AngleAxisx(Scalar(M_PI), v0.unitOrthogonal()) * v0); - VERIFY_IS_APPROX(ei_cos(a)*v0.squaredNorm(), v0.eigen2_dot(AngleAxisx(a, v0.unitOrthogonal()) * v0)); - m = AngleAxisx(a, v0.normalized()).toRotationMatrix().adjoint(); - VERIFY_IS_APPROX(Matrix3::Identity(), m * AngleAxisx(a, v0.normalized())); - VERIFY_IS_APPROX(Matrix3::Identity(), AngleAxisx(a, v0.normalized()) * m); - - q1 = AngleAxisx(a, v0.normalized()); - q2 = AngleAxisx(a, v1.normalized()); - - // angular distance - Scalar refangle = ei_abs(AngleAxisx(q1.inverse()*q2).angle()); - if (refangle>Scalar(M_PI)) - refangle = Scalar(2)*Scalar(M_PI) - refangle; - - if((q1.coeffs()-q2.coeffs()).norm() > 10*largeEps) - { - VERIFY(ei_isApprox(q1.angularDistance(q2), refangle, largeEps)); - } - - // rotation matrix conversion - VERIFY_IS_APPROX(q1 * v2, q1.toRotationMatrix() * v2); - VERIFY_IS_APPROX(q1 * q2 * v2, - q1.toRotationMatrix() * q2.toRotationMatrix() * v2); - - VERIFY( (q2*q1).isApprox(q1*q2, largeEps) || !(q2 * q1 * v2).isApprox( - q1.toRotationMatrix() * q2.toRotationMatrix() * v2)); - - q2 = q1.toRotationMatrix(); - VERIFY_IS_APPROX(q1*v1,q2*v1); - - matrot1 = AngleAxisx(Scalar(0.1), Vector3::UnitX()) - * AngleAxisx(Scalar(0.2), Vector3::UnitY()) - * AngleAxisx(Scalar(0.3), Vector3::UnitZ()); - VERIFY_IS_APPROX(matrot1 * v1, - AngleAxisx(Scalar(0.1), Vector3(1,0,0)).toRotationMatrix() - * (AngleAxisx(Scalar(0.2), Vector3(0,1,0)).toRotationMatrix() - * (AngleAxisx(Scalar(0.3), Vector3(0,0,1)).toRotationMatrix() * v1))); - - // angle-axis conversion - AngleAxisx aa = q1; - VERIFY_IS_APPROX(q1 * v1, Quaternionx(aa) * v1); - VERIFY_IS_NOT_APPROX(q1 * v1, Quaternionx(AngleAxisx(aa.angle()*2,aa.axis())) * v1); - - // from two vector creation - VERIFY_IS_APPROX(v2.normalized(),(q2.setFromTwoVectors(v1,v2)*v1).normalized()); - VERIFY_IS_APPROX(v2.normalized(),(q2.setFromTwoVectors(v1,v2)*v1).normalized()); - - // inverse and conjugate - VERIFY_IS_APPROX(q1 * (q1.inverse() * v1), v1); - VERIFY_IS_APPROX(q1 * (q1.conjugate() * v1), v1); - - // AngleAxis - VERIFY_IS_APPROX(AngleAxisx(a,v1.normalized()).toRotationMatrix(), - Quaternionx(AngleAxisx(a,v1.normalized())).toRotationMatrix()); - - AngleAxisx aa1; - m = q1.toRotationMatrix(); - aa1 = m; - VERIFY_IS_APPROX(AngleAxisx(m).toRotationMatrix(), - Quaternionx(m).toRotationMatrix()); - - // Transform - // TODO complete the tests ! - a = 0; - while (ei_abs(a)(-Scalar(0.4)*Scalar(M_PI), Scalar(0.4)*Scalar(M_PI)); - q1 = AngleAxisx(a, v0.normalized()); - Transform3 t0, t1, t2; - // first test setIdentity() and Identity() - t0.setIdentity(); - VERIFY_IS_APPROX(t0.matrix(), Transform3::MatrixType::Identity()); - t0.matrix().setZero(); - t0 = Transform3::Identity(); - VERIFY_IS_APPROX(t0.matrix(), Transform3::MatrixType::Identity()); - - t0.linear() = q1.toRotationMatrix(); - t1.setIdentity(); - t1.linear() = q1.toRotationMatrix(); - - v0 << 50, 2, 1;//= ei_random_matrix().cwiseProduct(Vector3(10,2,0.5)); - t0.scale(v0); - t1.prescale(v0); - - VERIFY_IS_APPROX( (t0 * Vector3(1,0,0)).norm(), v0.x()); - //VERIFY(!ei_isApprox((t1 * Vector3(1,0,0)).norm(), v0.x())); - - t0.setIdentity(); - t1.setIdentity(); - v1 << 1, 2, 3; - t0.linear() = q1.toRotationMatrix(); - t0.pretranslate(v0); - t0.scale(v1); - t1.linear() = q1.conjugate().toRotationMatrix(); - t1.prescale(v1.cwise().inverse()); - t1.translate(-v0); - - VERIFY((t0.matrix() * t1.matrix()).isIdentity(test_precision())); - - t1.fromPositionOrientationScale(v0, q1, v1); - VERIFY_IS_APPROX(t1.matrix(), t0.matrix()); - VERIFY_IS_APPROX(t1*v1, t0*v1); - - t0.setIdentity(); t0.scale(v0).rotate(q1.toRotationMatrix()); - t1.setIdentity(); t1.scale(v0).rotate(q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - t0.setIdentity(); t0.scale(v0).rotate(AngleAxisx(q1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - VERIFY_IS_APPROX(t0.scale(a).matrix(), t1.scale(Vector3::Constant(a)).matrix()); - VERIFY_IS_APPROX(t0.prescale(a).matrix(), t1.prescale(Vector3::Constant(a)).matrix()); - - // More transform constructors, operator=, operator*= - - Matrix3 mat3 = Matrix3::Random(); - Matrix4 mat4; - mat4 << mat3 , Vector3::Zero() , Vector4::Zero().transpose(); - Transform3 tmat3(mat3), tmat4(mat4); - tmat4.matrix()(3,3) = Scalar(1); - VERIFY_IS_APPROX(tmat3.matrix(), tmat4.matrix()); - - Scalar a3 = ei_random(-Scalar(M_PI), Scalar(M_PI)); - Vector3 v3 = Vector3::Random().normalized(); - AngleAxisx aa3(a3, v3); - Transform3 t3(aa3); - Transform3 t4; - t4 = aa3; - VERIFY_IS_APPROX(t3.matrix(), t4.matrix()); - t4.rotate(AngleAxisx(-a3,v3)); - VERIFY_IS_APPROX(t4.matrix(), Matrix4::Identity()); - t4 *= aa3; - VERIFY_IS_APPROX(t3.matrix(), t4.matrix()); - - v3 = Vector3::Random(); - Translation3 tv3(v3); - Transform3 t5(tv3); - t4 = tv3; - VERIFY_IS_APPROX(t5.matrix(), t4.matrix()); - t4.translate(-v3); - VERIFY_IS_APPROX(t4.matrix(), Matrix4::Identity()); - t4 *= tv3; - VERIFY_IS_APPROX(t5.matrix(), t4.matrix()); - - Scaling3 sv3(v3); - Transform3 t6(sv3); - t4 = sv3; - VERIFY_IS_APPROX(t6.matrix(), t4.matrix()); - t4.scale(v3.cwise().inverse()); - VERIFY_IS_APPROX(t4.matrix(), Matrix4::Identity()); - t4 *= sv3; - VERIFY_IS_APPROX(t6.matrix(), t4.matrix()); - - // matrix * transform - VERIFY_IS_APPROX(Transform3(t3.matrix()*t4).matrix(), Transform3(t3*t4).matrix()); - - // chained Transform product - VERIFY_IS_APPROX(((t3*t4)*t5).matrix(), (t3*(t4*t5)).matrix()); - - // check that Transform product doesn't have aliasing problems - t5 = t4; - t5 = t5*t5; - VERIFY_IS_APPROX(t5, t4*t4); - - // 2D transformation - Transform2 t20, t21; - Vector2 v20 = Vector2::Random(); - Vector2 v21 = Vector2::Random(); - for (int k=0; k<2; ++k) - if (ei_abs(v21[k])(a).toRotationMatrix(); - VERIFY_IS_APPROX(t20.fromPositionOrientationScale(v20,a,v21).matrix(), - t21.pretranslate(v20).scale(v21).matrix()); - - t21.setIdentity(); - t21.linear() = Rotation2D(-a).toRotationMatrix(); - VERIFY( (t20.fromPositionOrientationScale(v20,a,v21) - * (t21.prescale(v21.cwise().inverse()).translate(-v20))).matrix().isIdentity(test_precision()) ); - - // Transform - new API - // 3D - t0.setIdentity(); - t0.rotate(q1).scale(v0).translate(v0); - // mat * scaling and mat * translation - t1 = (Matrix3(q1) * Scaling3(v0)) * Translation3(v0); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // mat * transformation and scaling * translation - t1 = Matrix3(q1) * (Scaling3(v0) * Translation3(v0)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - t0.setIdentity(); - t0.prerotate(q1).prescale(v0).pretranslate(v0); - // translation * scaling and transformation * mat - t1 = (Translation3(v0) * Scaling3(v0)) * Matrix3(q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // scaling * mat and translation * mat - t1 = Translation3(v0) * (Scaling3(v0) * Matrix3(q1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - t0.setIdentity(); - t0.scale(v0).translate(v0).rotate(q1); - // translation * mat and scaling * transformation - t1 = Scaling3(v0) * (Translation3(v0) * Matrix3(q1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // transformation * scaling - t0.scale(v0); - t1 = t1 * Scaling3(v0); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // transformation * translation - t0.translate(v0); - t1 = t1 * Translation3(v0); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - // translation * transformation - t0.pretranslate(v0); - t1 = Translation3(v0) * t1; - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // transform * quaternion - t0.rotate(q1); - t1 = t1 * q1; - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // translation * quaternion - t0.translate(v1).rotate(q1); - t1 = t1 * (Translation3(v1) * q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // scaling * quaternion - t0.scale(v1).rotate(q1); - t1 = t1 * (Scaling3(v1) * q1); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // quaternion * transform - t0.prerotate(q1); - t1 = q1 * t1; - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // quaternion * translation - t0.rotate(q1).translate(v1); - t1 = t1 * (q1 * Translation3(v1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // quaternion * scaling - t0.rotate(q1).scale(v1); - t1 = t1 * (q1 * Scaling3(v1)); - VERIFY_IS_APPROX(t0.matrix(), t1.matrix()); - - // translation * vector - t0.setIdentity(); - t0.translate(v0); - VERIFY_IS_APPROX(t0 * v1, Translation3(v0) * v1); - - // scaling * vector - t0.setIdentity(); - t0.scale(v0); - VERIFY_IS_APPROX(t0 * v1, Scaling3(v0) * v1); - - // test transform inversion - t0.setIdentity(); - t0.translate(v0); - t0.linear().setRandom(); - VERIFY_IS_APPROX(t0.inverse(Affine), t0.matrix().inverse()); - t0.setIdentity(); - t0.translate(v0).rotate(q1); - VERIFY_IS_APPROX(t0.inverse(Isometry), t0.matrix().inverse()); - - // test extract rotation and scaling - t0.setIdentity(); - t0.translate(v0).rotate(q1).scale(v1); - VERIFY_IS_APPROX(t0.rotation() * v1, Matrix3(q1) * v1); - - Matrix3 mat_rotation, mat_scaling; - t0.setIdentity(); - t0.translate(v0).rotate(q1).scale(v1); - t0.computeRotationScaling(&mat_rotation, &mat_scaling); - VERIFY_IS_APPROX(t0.linear(), mat_rotation * mat_scaling); - VERIFY_IS_APPROX(mat_rotation*mat_rotation.adjoint(), Matrix3::Identity()); - VERIFY_IS_APPROX(mat_rotation.determinant(), Scalar(1)); - t0.computeScalingRotation(&mat_scaling, &mat_rotation); - VERIFY_IS_APPROX(t0.linear(), mat_scaling * mat_rotation); - VERIFY_IS_APPROX(mat_rotation*mat_rotation.adjoint(), Matrix3::Identity()); - VERIFY_IS_APPROX(mat_rotation.determinant(), Scalar(1)); - - // test casting - eigen2_Transform t1f = t1.template cast(); - VERIFY_IS_APPROX(t1f.template cast(),t1); - eigen2_Transform t1d = t1.template cast(); - VERIFY_IS_APPROX(t1d.template cast(),t1); - - Translation3 tr1(v0); - eigen2_Translation tr1f = tr1.template cast(); - VERIFY_IS_APPROX(tr1f.template cast(),tr1); - eigen2_Translation tr1d = tr1.template cast(); - VERIFY_IS_APPROX(tr1d.template cast(),tr1); - - Scaling3 sc1(v0); - eigen2_Scaling sc1f = sc1.template cast(); - VERIFY_IS_APPROX(sc1f.template cast(),sc1); - eigen2_Scaling sc1d = sc1.template cast(); - VERIFY_IS_APPROX(sc1d.template cast(),sc1); - - eigen2_Quaternion q1f = q1.template cast(); - VERIFY_IS_APPROX(q1f.template cast(),q1); - eigen2_Quaternion q1d = q1.template cast(); - VERIFY_IS_APPROX(q1d.template cast(),q1); - - eigen2_AngleAxis aa1f = aa1.template cast(); - VERIFY_IS_APPROX(aa1f.template cast(),aa1); - eigen2_AngleAxis aa1d = aa1.template cast(); - VERIFY_IS_APPROX(aa1d.template cast(),aa1); - - eigen2_Rotation2D r2d1(ei_random()); - eigen2_Rotation2D r2d1f = r2d1.template cast(); - VERIFY_IS_APPROX(r2d1f.template cast(),r2d1); - eigen2_Rotation2D r2d1d = r2d1.template cast(); - VERIFY_IS_APPROX(r2d1d.template cast(),r2d1); - - m = q1; -// m.col(1) = Vector3(0,ei_random(),ei_random()).normalized(); -// m.col(0) = Vector3(-1,0,0).normalized(); -// m.col(2) = m.col(0).cross(m.col(1)); - #define VERIFY_EULER(I,J,K, X,Y,Z) { \ - Vector3 ea = m.eulerAngles(I,J,K); \ - Matrix3 m1 = Matrix3(AngleAxisx(ea[0], Vector3::Unit##X()) * AngleAxisx(ea[1], Vector3::Unit##Y()) * AngleAxisx(ea[2], Vector3::Unit##Z())); \ - VERIFY_IS_APPROX(m, m1); \ - VERIFY_IS_APPROX(m, Matrix3(AngleAxisx(ea[0], Vector3::Unit##X()) * AngleAxisx(ea[1], Vector3::Unit##Y()) * AngleAxisx(ea[2], Vector3::Unit##Z()))); \ - } - VERIFY_EULER(0,1,2, X,Y,Z); - VERIFY_EULER(0,1,0, X,Y,X); - VERIFY_EULER(0,2,1, X,Z,Y); - VERIFY_EULER(0,2,0, X,Z,X); - - VERIFY_EULER(1,2,0, Y,Z,X); - VERIFY_EULER(1,2,1, Y,Z,Y); - VERIFY_EULER(1,0,2, Y,X,Z); - VERIFY_EULER(1,0,1, Y,X,Y); - - VERIFY_EULER(2,0,1, Z,X,Y); - VERIFY_EULER(2,0,2, Z,X,Z); - VERIFY_EULER(2,1,0, Z,Y,X); - VERIFY_EULER(2,1,2, Z,Y,Z); - - // colwise/rowwise cross product - mat3.setRandom(); - Vector3 vec3 = Vector3::Random(); - Matrix3 mcross; - int i = ei_random(0,2); - mcross = mat3.colwise().cross(vec3); - VERIFY_IS_APPROX(mcross.col(i), mat3.col(i).cross(vec3)); - mcross = mat3.rowwise().cross(vec3); - VERIFY_IS_APPROX(mcross.row(i), mat3.row(i).cross(vec3)); - - -} - -void test_eigen2_geometry_with_eigen2_prefix() -{ - std::cout << "eigen2 support: " << EIGEN2_SUPPORT_STAGE << std::endl; - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( geometry() ); - CALL_SUBTEST_2( geometry() ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_hyperplane.cpp b/thirdparty/eigen/test/eigen2/eigen2_hyperplane.cpp deleted file mode 100644 index f3f85e14..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_hyperplane.cpp +++ /dev/null @@ -1,126 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include -#include -#include - -template void hyperplane(const HyperplaneType& _plane) -{ - /* this test covers the following files: - Hyperplane.h - */ - - const int dim = _plane.dim(); - typedef typename HyperplaneType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - typedef Matrix MatrixType; - - VectorType p0 = VectorType::Random(dim); - VectorType p1 = VectorType::Random(dim); - - VectorType n0 = VectorType::Random(dim).normalized(); - VectorType n1 = VectorType::Random(dim).normalized(); - - HyperplaneType pl0(n0, p0); - HyperplaneType pl1(n1, p1); - HyperplaneType pl2 = pl1; - - Scalar s0 = ei_random(); - Scalar s1 = ei_random(); - - VERIFY_IS_APPROX( n1.eigen2_dot(n1), Scalar(1) ); - - VERIFY_IS_MUCH_SMALLER_THAN( pl0.absDistance(p0), Scalar(1) ); - VERIFY_IS_APPROX( pl1.signedDistance(p1 + n1 * s0), s0 ); - VERIFY_IS_MUCH_SMALLER_THAN( pl1.signedDistance(pl1.projection(p0)), Scalar(1) ); - VERIFY_IS_MUCH_SMALLER_THAN( pl1.absDistance(p1 + pl1.normal().unitOrthogonal() * s1), Scalar(1) ); - - // transform - if (!NumTraits::IsComplex) - { - MatrixType rot = MatrixType::Random(dim,dim).qr().matrixQ(); - Scaling scaling(VectorType::Random()); - Translation translation(VectorType::Random()); - - pl2 = pl1; - VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot).absDistance(rot * p1), Scalar(1) ); - pl2 = pl1; - VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot,Isometry).absDistance(rot * p1), Scalar(1) ); - pl2 = pl1; - VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot*scaling).absDistance((rot*scaling) * p1), Scalar(1) ); - pl2 = pl1; - VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot*scaling*translation) - .absDistance((rot*scaling*translation) * p1), Scalar(1) ); - pl2 = pl1; - VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot*translation,Isometry) - .absDistance((rot*translation) * p1), Scalar(1) ); - } - - // casting - const int Dim = HyperplaneType::AmbientDimAtCompileTime; - typedef typename GetDifferentType::type OtherScalar; - Hyperplane hp1f = pl1.template cast(); - VERIFY_IS_APPROX(hp1f.template cast(),pl1); - Hyperplane hp1d = pl1.template cast(); - VERIFY_IS_APPROX(hp1d.template cast(),pl1); -} - -template void lines() -{ - typedef Hyperplane HLine; - typedef ParametrizedLine PLine; - typedef Matrix Vector; - typedef Matrix CoeffsType; - - for(int i = 0; i < 10; i++) - { - Vector center = Vector::Random(); - Vector u = Vector::Random(); - Vector v = Vector::Random(); - Scalar a = ei_random(); - while (ei_abs(a-1) < 1e-4) a = ei_random(); - while (u.norm() < 1e-4) u = Vector::Random(); - while (v.norm() < 1e-4) v = Vector::Random(); - - HLine line_u = HLine::Through(center + u, center + a*u); - HLine line_v = HLine::Through(center + v, center + a*v); - - // the line equations should be normalized so that a^2+b^2=1 - VERIFY_IS_APPROX(line_u.normal().norm(), Scalar(1)); - VERIFY_IS_APPROX(line_v.normal().norm(), Scalar(1)); - - Vector result = line_u.intersection(line_v); - - // the lines should intersect at the point we called "center" - VERIFY_IS_APPROX(result, center); - - // check conversions between two types of lines - PLine pl(line_u); // gcc 3.3 will commit suicide if we don't name this variable - CoeffsType converted_coeffs(HLine(pl).coeffs()); - converted_coeffs *= line_u.coeffs()(0)/converted_coeffs(0); - VERIFY(line_u.coeffs().isApprox(converted_coeffs)); - } -} - -void test_eigen2_hyperplane() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( hyperplane(Hyperplane()) ); - CALL_SUBTEST_2( hyperplane(Hyperplane()) ); - CALL_SUBTEST_3( hyperplane(Hyperplane()) ); - CALL_SUBTEST_4( hyperplane(Hyperplane,5>()) ); - CALL_SUBTEST_5( lines() ); - CALL_SUBTEST_6( lines() ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_inverse.cpp b/thirdparty/eigen/test/eigen2/eigen2_inverse.cpp deleted file mode 100644 index ccd24a19..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_inverse.cpp +++ /dev/null @@ -1,62 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -template void inverse(const MatrixType& m) -{ - /* this test covers the following files: - Inverse.h - */ - int rows = m.rows(); - int cols = m.cols(); - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - - MatrixType m1 = MatrixType::Random(rows, cols), - m2(rows, cols), - identity = MatrixType::Identity(rows, rows); - - while(ei_abs(m1.determinant()) < RealScalar(0.1) && rows <= 8) - { - m1 = MatrixType::Random(rows, cols); - } - - m2 = m1.inverse(); - VERIFY_IS_APPROX(m1, m2.inverse() ); - - m1.computeInverse(&m2); - VERIFY_IS_APPROX(m1, m2.inverse() ); - - VERIFY_IS_APPROX((Scalar(2)*m2).inverse(), m2.inverse()*Scalar(0.5)); - - VERIFY_IS_APPROX(identity, m1.inverse() * m1 ); - VERIFY_IS_APPROX(identity, m1 * m1.inverse() ); - - VERIFY_IS_APPROX(m1, m1.inverse().inverse() ); - - // since for the general case we implement separately row-major and col-major, test that - VERIFY_IS_APPROX(m1.transpose().inverse(), m1.inverse().transpose()); -} - -void test_eigen2_inverse() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( inverse(Matrix()) ); - CALL_SUBTEST_2( inverse(Matrix2d()) ); - CALL_SUBTEST_3( inverse(Matrix3f()) ); - CALL_SUBTEST_4( inverse(Matrix4f()) ); - CALL_SUBTEST_5( inverse(MatrixXf(8,8)) ); - CALL_SUBTEST_6( inverse(MatrixXcd(7,7)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_linearstructure.cpp b/thirdparty/eigen/test/eigen2/eigen2_linearstructure.cpp deleted file mode 100644 index 488f4c48..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_linearstructure.cpp +++ /dev/null @@ -1,83 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void linearStructure(const MatrixType& m) -{ - /* this test covers the following files: - Sum.h Difference.h Opposite.h ScalarMultiple.h - */ - - typedef typename MatrixType::Scalar Scalar; - typedef Matrix VectorType; - - int rows = m.rows(); - int cols = m.cols(); - - // this test relies a lot on Random.h, and there's not much more that we can do - // to test it, hence I consider that we will have tested Random.h - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols); - - Scalar s1 = ei_random(); - while (ei_abs(s1)<1e-3) s1 = ei_random(); - - int r = ei_random(0, rows-1), - c = ei_random(0, cols-1); - - VERIFY_IS_APPROX(-(-m1), m1); - VERIFY_IS_APPROX(m1+m1, 2*m1); - VERIFY_IS_APPROX(m1+m2-m1, m2); - VERIFY_IS_APPROX(-m2+m1+m2, m1); - VERIFY_IS_APPROX(m1*s1, s1*m1); - VERIFY_IS_APPROX((m1+m2)*s1, s1*m1+s1*m2); - VERIFY_IS_APPROX((-m1+m2)*s1, -s1*m1+s1*m2); - m3 = m2; m3 += m1; - VERIFY_IS_APPROX(m3, m1+m2); - m3 = m2; m3 -= m1; - VERIFY_IS_APPROX(m3, m2-m1); - m3 = m2; m3 *= s1; - VERIFY_IS_APPROX(m3, s1*m2); - if(NumTraits::HasFloatingPoint) - { - m3 = m2; m3 /= s1; - VERIFY_IS_APPROX(m3, m2/s1); - } - - // again, test operator() to check const-qualification - VERIFY_IS_APPROX((-m1)(r,c), -(m1(r,c))); - VERIFY_IS_APPROX((m1-m2)(r,c), (m1(r,c))-(m2(r,c))); - VERIFY_IS_APPROX((m1+m2)(r,c), (m1(r,c))+(m2(r,c))); - VERIFY_IS_APPROX((s1*m1)(r,c), s1*(m1(r,c))); - VERIFY_IS_APPROX((m1*s1)(r,c), (m1(r,c))*s1); - if(NumTraits::HasFloatingPoint) - VERIFY_IS_APPROX((m1/s1)(r,c), (m1(r,c))/s1); - - // use .block to disable vectorization and compare to the vectorized version - VERIFY_IS_APPROX(m1+m1.block(0,0,rows,cols), m1+m1); - VERIFY_IS_APPROX(m1.cwise() * m1.block(0,0,rows,cols), m1.cwise() * m1); - VERIFY_IS_APPROX(m1 - m1.block(0,0,rows,cols), m1 - m1); - VERIFY_IS_APPROX(m1.block(0,0,rows,cols) * s1, m1 * s1); -} - -void test_eigen2_linearstructure() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( linearStructure(Matrix()) ); - CALL_SUBTEST_2( linearStructure(Matrix2f()) ); - CALL_SUBTEST_3( linearStructure(Vector3d()) ); - CALL_SUBTEST_4( linearStructure(Matrix4d()) ); - CALL_SUBTEST_5( linearStructure(MatrixXcf(3, 3)) ); - CALL_SUBTEST_6( linearStructure(MatrixXf(8, 12)) ); - CALL_SUBTEST_7( linearStructure(MatrixXi(8, 12)) ); - CALL_SUBTEST_8( linearStructure(MatrixXcd(20, 20)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_lu.cpp b/thirdparty/eigen/test/eigen2/eigen2_lu.cpp deleted file mode 100644 index e993b1c7..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_lu.cpp +++ /dev/null @@ -1,122 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -template -void doSomeRankPreservingOperations(Eigen::MatrixBase& m) -{ - typedef typename Derived::RealScalar RealScalar; - for(int a = 0; a < 3*(m.rows()+m.cols()); a++) - { - RealScalar d = Eigen::ei_random(-1,1); - int i = Eigen::ei_random(0,m.rows()-1); // i is a random row number - int j; - do { - j = Eigen::ei_random(0,m.rows()-1); - } while (i==j); // j is another one (must be different) - m.row(i) += d * m.row(j); - - i = Eigen::ei_random(0,m.cols()-1); // i is a random column number - do { - j = Eigen::ei_random(0,m.cols()-1); - } while (i==j); // j is another one (must be different) - m.col(i) += d * m.col(j); - } -} - -template void lu_non_invertible() -{ - /* this test covers the following files: - LU.h - */ - // NOTE there seems to be a problem with too small sizes -- could easily lie in the doSomeRankPreservingOperations function - int rows = ei_random(20,200), cols = ei_random(20,200), cols2 = ei_random(20,200); - int rank = ei_random(1, std::min(rows, cols)-1); - - MatrixType m1(rows, cols), m2(cols, cols2), m3(rows, cols2), k(1,1); - m1 = MatrixType::Random(rows,cols); - if(rows <= cols) - for(int i = rank; i < rows; i++) m1.row(i).setZero(); - else - for(int i = rank; i < cols; i++) m1.col(i).setZero(); - doSomeRankPreservingOperations(m1); - - LU lu(m1); - typename LU::KernelResultType m1kernel = lu.kernel(); - typename LU::ImageResultType m1image = lu.image(); - - VERIFY(rank == lu.rank()); - VERIFY(cols - lu.rank() == lu.dimensionOfKernel()); - VERIFY(!lu.isInjective()); - VERIFY(!lu.isInvertible()); - VERIFY(lu.isSurjective() == (lu.rank() == rows)); - VERIFY((m1 * m1kernel).isMuchSmallerThan(m1)); - VERIFY(m1image.lu().rank() == rank); - MatrixType sidebyside(m1.rows(), m1.cols() + m1image.cols()); - sidebyside << m1, m1image; - VERIFY(sidebyside.lu().rank() == rank); - m2 = MatrixType::Random(cols,cols2); - m3 = m1*m2; - m2 = MatrixType::Random(cols,cols2); - lu.solve(m3, &m2); - VERIFY_IS_APPROX(m3, m1*m2); - /* solve now always returns true - m3 = MatrixType::Random(rows,cols2); - VERIFY(!lu.solve(m3, &m2)); - */ -} - -template void lu_invertible() -{ - /* this test covers the following files: - LU.h - */ - typedef typename NumTraits::Real RealScalar; - int size = ei_random(10,200); - - MatrixType m1(size, size), m2(size, size), m3(size, size); - m1 = MatrixType::Random(size,size); - - if (ei_is_same_type::ret) - { - // let's build a matrix more stable to inverse - MatrixType a = MatrixType::Random(size,size*2); - m1 += a * a.adjoint(); - } - - LU lu(m1); - VERIFY(0 == lu.dimensionOfKernel()); - VERIFY(size == lu.rank()); - VERIFY(lu.isInjective()); - VERIFY(lu.isSurjective()); - VERIFY(lu.isInvertible()); - VERIFY(lu.image().lu().isInvertible()); - m3 = MatrixType::Random(size,size); - lu.solve(m3, &m2); - VERIFY_IS_APPROX(m3, m1*m2); - VERIFY_IS_APPROX(m2, lu.inverse()*m3); - m3 = MatrixType::Random(size,size); - VERIFY(lu.solve(m3, &m2)); -} - -void test_eigen2_lu() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( lu_non_invertible() ); - CALL_SUBTEST_2( lu_non_invertible() ); - CALL_SUBTEST_3( lu_non_invertible() ); - CALL_SUBTEST_4( lu_non_invertible() ); - CALL_SUBTEST_1( lu_invertible() ); - CALL_SUBTEST_2( lu_invertible() ); - CALL_SUBTEST_3( lu_invertible() ); - CALL_SUBTEST_4( lu_invertible() ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_map.cpp b/thirdparty/eigen/test/eigen2/eigen2_map.cpp deleted file mode 100644 index 4a1c4e11..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_map.cpp +++ /dev/null @@ -1,114 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2007-2010 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void map_class_vector(const VectorType& m) -{ - typedef typename VectorType::Scalar Scalar; - - int size = m.size(); - - // test Map.h - Scalar* array1 = ei_aligned_new(size); - Scalar* array2 = ei_aligned_new(size); - Scalar* array3 = new Scalar[size+1]; - Scalar* array3unaligned = std::size_t(array3)%16 == 0 ? array3+1 : array3; - - Map(array1, size) = VectorType::Random(size); - Map(array2, size) = Map(array1, size); - Map(array3unaligned, size) = Map((const Scalar*)array1, size); // test non-const-correctness support in eigen2 - VectorType ma1 = Map(array1, size); - VectorType ma2 = Map(array2, size); - VectorType ma3 = Map(array3unaligned, size); - VERIFY_IS_APPROX(ma1, ma2); - VERIFY_IS_APPROX(ma1, ma3); - - ei_aligned_delete(array1, size); - ei_aligned_delete(array2, size); - delete[] array3; -} - -template void map_class_matrix(const MatrixType& m) -{ - typedef typename MatrixType::Scalar Scalar; - - int rows = m.rows(), cols = m.cols(), size = rows*cols; - - // test Map.h - Scalar* array1 = ei_aligned_new(size); - for(int i = 0; i < size; i++) array1[i] = Scalar(1); - Scalar* array2 = ei_aligned_new(size); - for(int i = 0; i < size; i++) array2[i] = Scalar(1); - Scalar* array3 = new Scalar[size+1]; - for(int i = 0; i < size+1; i++) array3[i] = Scalar(1); - Scalar* array3unaligned = std::size_t(array3)%16 == 0 ? array3+1 : array3; - Map(array1, rows, cols) = MatrixType::Ones(rows,cols); - Map(array2, rows, cols) = Map((const Scalar*)array1, rows, cols); // test non-const-correctness support in eigen2 - Map(array3unaligned, rows, cols) = Map(array1, rows, cols); - MatrixType ma1 = Map(array1, rows, cols); - MatrixType ma2 = Map(array2, rows, cols); - VERIFY_IS_APPROX(ma1, ma2); - MatrixType ma3 = Map(array3unaligned, rows, cols); - VERIFY_IS_APPROX(ma1, ma3); - - ei_aligned_delete(array1, size); - ei_aligned_delete(array2, size); - delete[] array3; -} - -template void map_static_methods(const VectorType& m) -{ - typedef typename VectorType::Scalar Scalar; - - int size = m.size(); - - // test Map.h - Scalar* array1 = ei_aligned_new(size); - Scalar* array2 = ei_aligned_new(size); - Scalar* array3 = new Scalar[size+1]; - Scalar* array3unaligned = std::size_t(array3)%16 == 0 ? array3+1 : array3; - - VectorType::MapAligned(array1, size) = VectorType::Random(size); - VectorType::Map(array2, size) = VectorType::Map(array1, size); - VectorType::Map(array3unaligned, size) = VectorType::Map(array1, size); - VectorType ma1 = VectorType::Map(array1, size); - VectorType ma2 = VectorType::MapAligned(array2, size); - VectorType ma3 = VectorType::Map(array3unaligned, size); - VERIFY_IS_APPROX(ma1, ma2); - VERIFY_IS_APPROX(ma1, ma3); - - ei_aligned_delete(array1, size); - ei_aligned_delete(array2, size); - delete[] array3; -} - - -void test_eigen2_map() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( map_class_vector(Matrix()) ); - CALL_SUBTEST_2( map_class_vector(Vector4d()) ); - CALL_SUBTEST_3( map_class_vector(RowVector4f()) ); - CALL_SUBTEST_4( map_class_vector(VectorXcf(8)) ); - CALL_SUBTEST_5( map_class_vector(VectorXi(12)) ); - - CALL_SUBTEST_1( map_class_matrix(Matrix()) ); - CALL_SUBTEST_2( map_class_matrix(Matrix4d()) ); - CALL_SUBTEST_6( map_class_matrix(Matrix()) ); - CALL_SUBTEST_4( map_class_matrix(MatrixXcf(ei_random(1,10),ei_random(1,10))) ); - CALL_SUBTEST_5( map_class_matrix(MatrixXi(ei_random(1,10),ei_random(1,10))) ); - - CALL_SUBTEST_1( map_static_methods(Matrix()) ); - CALL_SUBTEST_2( map_static_methods(Vector3f()) ); - CALL_SUBTEST_7( map_static_methods(RowVector3d()) ); - CALL_SUBTEST_4( map_static_methods(VectorXcd(8)) ); - CALL_SUBTEST_5( map_static_methods(VectorXf(12)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_meta.cpp b/thirdparty/eigen/test/eigen2/eigen2_meta.cpp deleted file mode 100644 index 1d01bd84..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_meta.cpp +++ /dev/null @@ -1,60 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -void test_eigen2_meta() -{ - typedef float & FloatRef; - typedef const float & ConstFloatRef; - - VERIFY((ei_meta_if<(3<4),ei_meta_true, ei_meta_false>::ret::ret)); - VERIFY(( ei_is_same_type::ret)); - VERIFY((!ei_is_same_type::ret)); - VERIFY((!ei_is_same_type::ret)); - VERIFY((!ei_is_same_type::ret)); - - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - VERIFY(( ei_is_same_type::type >::ret)); - - VERIFY(ei_meta_sqrt<1>::ret == 1); - #define VERIFY_META_SQRT(X) VERIFY(ei_meta_sqrt::ret == int(ei_sqrt(double(X)))) - VERIFY_META_SQRT(2); - VERIFY_META_SQRT(3); - VERIFY_META_SQRT(4); - VERIFY_META_SQRT(5); - VERIFY_META_SQRT(6); - VERIFY_META_SQRT(8); - VERIFY_META_SQRT(9); - VERIFY_META_SQRT(15); - VERIFY_META_SQRT(16); - VERIFY_META_SQRT(17); - VERIFY_META_SQRT(255); - VERIFY_META_SQRT(256); - VERIFY_META_SQRT(257); - VERIFY_META_SQRT(1023); - VERIFY_META_SQRT(1024); - VERIFY_META_SQRT(1025); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_miscmatrices.cpp b/thirdparty/eigen/test/eigen2/eigen2_miscmatrices.cpp deleted file mode 100644 index 8bbb20cc..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_miscmatrices.cpp +++ /dev/null @@ -1,48 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void miscMatrices(const MatrixType& m) -{ - /* this test covers the following files: - DiagonalMatrix.h Ones.h - */ - - typedef typename MatrixType::Scalar Scalar; - typedef Matrix VectorType; - typedef Matrix RowVectorType; - int rows = m.rows(); - int cols = m.cols(); - - int r = ei_random(0, rows-1), r2 = ei_random(0, rows-1), c = ei_random(0, cols-1); - VERIFY_IS_APPROX(MatrixType::Ones(rows,cols)(r,c), static_cast(1)); - MatrixType m1 = MatrixType::Ones(rows,cols); - VERIFY_IS_APPROX(m1(r,c), static_cast(1)); - VectorType v1 = VectorType::Random(rows); - v1[0]; - Matrix - square = v1.asDiagonal(); - if(r==r2) VERIFY_IS_APPROX(square(r,r2), v1[r]); - else VERIFY_IS_MUCH_SMALLER_THAN(square(r,r2), static_cast(1)); - square = MatrixType::Zero(rows, rows); - square.diagonal() = VectorType::Ones(rows); - VERIFY_IS_APPROX(square, MatrixType::Identity(rows, rows)); -} - -void test_eigen2_miscmatrices() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( miscMatrices(Matrix()) ); - CALL_SUBTEST_2( miscMatrices(Matrix4d()) ); - CALL_SUBTEST_3( miscMatrices(MatrixXcf(3, 3)) ); - CALL_SUBTEST_4( miscMatrices(MatrixXi(8, 12)) ); - CALL_SUBTEST_5( miscMatrices(MatrixXcd(20, 20)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_mixingtypes.cpp b/thirdparty/eigen/test/eigen2/eigen2_mixingtypes.cpp deleted file mode 100644 index fb5ac5dd..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_mixingtypes.cpp +++ /dev/null @@ -1,77 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_NO_STATIC_ASSERT -#define EIGEN_NO_STATIC_ASSERT // turn static asserts into runtime asserts in order to check them -#endif - -#ifndef EIGEN_DONT_VECTORIZE -#define EIGEN_DONT_VECTORIZE // SSE intrinsics aren't designed to allow mixing types -#endif - -#include "main.h" - - -template void mixingtypes(int size = SizeAtCompileType) -{ - typedef Matrix Mat_f; - typedef Matrix Mat_d; - typedef Matrix, SizeAtCompileType, SizeAtCompileType> Mat_cf; - typedef Matrix, SizeAtCompileType, SizeAtCompileType> Mat_cd; - typedef Matrix Vec_f; - typedef Matrix Vec_d; - typedef Matrix, SizeAtCompileType, 1> Vec_cf; - typedef Matrix, SizeAtCompileType, 1> Vec_cd; - - Mat_f mf(size,size); - Mat_d md(size,size); - Mat_cf mcf(size,size); - Mat_cd mcd(size,size); - Vec_f vf(size,1); - Vec_d vd(size,1); - Vec_cf vcf(size,1); - Vec_cd vcd(size,1); - - mf+mf; - VERIFY_RAISES_ASSERT(mf+md); - VERIFY_RAISES_ASSERT(mf+mcf); - VERIFY_RAISES_ASSERT(vf=vd); - VERIFY_RAISES_ASSERT(vf+=vd); - VERIFY_RAISES_ASSERT(mcd=md); - - mf*mf; - md*mcd; - mcd*md; - mf*vcf; - mcf*vf; - mcf *= mf; - vcd = md*vcd; - vcf = mcf*vf; -#if 0 - // these are know generating hard build errors in eigen3 - VERIFY_RAISES_ASSERT(mf*md); - VERIFY_RAISES_ASSERT(mcf*mcd); - VERIFY_RAISES_ASSERT(mcf*vcd); - VERIFY_RAISES_ASSERT(vcf = mf*vf); - - vf.eigen2_dot(vf); - VERIFY_RAISES_ASSERT(vd.eigen2_dot(vf)); - VERIFY_RAISES_ASSERT(vcf.eigen2_dot(vf)); // yeah eventually we should allow this but i'm too lazy to make that change now in Dot.h - // especially as that might be rewritten as cwise product .sum() which would make that automatic. -#endif -} - -void test_eigen2_mixingtypes() -{ - // check that our operator new is indeed called: - CALL_SUBTEST_1(mixingtypes<3>()); - CALL_SUBTEST_2(mixingtypes<4>()); - CALL_SUBTEST_3(mixingtypes(20)); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_newstdvector.cpp b/thirdparty/eigen/test/eigen2/eigen2_newstdvector.cpp deleted file mode 100644 index 5f900990..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_newstdvector.cpp +++ /dev/null @@ -1,149 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#define EIGEN_USE_NEW_STDVECTOR -#include "main.h" -#include -#include - -template -void check_stdvector_matrix(const MatrixType& m) -{ - int rows = m.rows(); - int cols = m.cols(); - MatrixType x = MatrixType::Random(rows,cols), y = MatrixType::Random(rows,cols); - std::vector > v(10, MatrixType(rows,cols)), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.resize(22,y); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((std::size_t)&(v[22]) == (std::size_t)&(v[21]) + sizeof(MatrixType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - MatrixType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; i -void check_stdvector_transform(const TransformType&) -{ - typedef typename TransformType::MatrixType MatrixType; - TransformType x(MatrixType::Random()), y(MatrixType::Random()); - std::vector > v(10), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.resize(22,y); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((std::size_t)&(v[22]) == (std::size_t)&(v[21]) + sizeof(TransformType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - TransformType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; i -void check_stdvector_quaternion(const QuaternionType&) -{ - typedef typename QuaternionType::Coefficients Coefficients; - QuaternionType x(Coefficients::Random()), y(Coefficients::Random()); - std::vector > v(10), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.resize(22,y); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((std::size_t)&(v[22]) == (std::size_t)&(v[21]) + sizeof(QuaternionType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - QuaternionType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; i -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -// this hack is needed to make this file compiles with -pedantic (gcc) -#ifdef __GNUC__ -#define throw(X) -#endif -// discard stack allocation as that too bypasses malloc -#define EIGEN_STACK_ALLOCATION_LIMIT 0 -// any heap allocation will raise an assert -#define EIGEN_NO_MALLOC - -#include "main.h" - -template void nomalloc(const MatrixType& m) -{ - /* this test check no dynamic memory allocation are issued with fixed-size matrices - */ - - typedef typename MatrixType::Scalar Scalar; - - int rows = m.rows(); - int cols = m.cols(); - - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols); - - Scalar s1 = ei_random(); - - int r = ei_random(0, rows-1), - c = ei_random(0, cols-1); - - VERIFY_IS_APPROX((m1+m2)*s1, s1*m1+s1*m2); - VERIFY_IS_APPROX((m1+m2)(r,c), (m1(r,c))+(m2(r,c))); - VERIFY_IS_APPROX(m1.cwise() * m1.block(0,0,rows,cols), m1.cwise() * m1); - VERIFY_IS_APPROX((m1*m1.transpose())*m2, m1*(m1.transpose()*m2)); -} - -void test_eigen2_nomalloc() -{ - // check that our operator new is indeed called: - VERIFY_RAISES_ASSERT(MatrixXd dummy = MatrixXd::Random(3,3)); - CALL_SUBTEST_1( nomalloc(Matrix()) ); - CALL_SUBTEST_2( nomalloc(Matrix4d()) ); - CALL_SUBTEST_3( nomalloc(Matrix()) ); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_packetmath.cpp b/thirdparty/eigen/test/eigen2/eigen2_packetmath.cpp deleted file mode 100644 index b1f325fe..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_packetmath.cpp +++ /dev/null @@ -1,132 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -// using namespace Eigen; - -template bool areApprox(const Scalar* a, const Scalar* b, int size) -{ - for (int i=0; i const complex& min(const complex& a, const complex& b) -{ return a.real() < b.real() ? a : b; } - -template<> const complex& max(const complex& a, const complex& b) -{ return a.real() < b.real() ? b : a; } - -} - -template void packetmath() -{ - typedef typename ei_packet_traits::type Packet; - const int PacketSize = ei_packet_traits::size; - - const int size = PacketSize*4; - EIGEN_ALIGN_128 Scalar data1[ei_packet_traits::size*4]; - EIGEN_ALIGN_128 Scalar data2[ei_packet_traits::size*4]; - EIGEN_ALIGN_128 Packet packets[PacketSize*2]; - EIGEN_ALIGN_128 Scalar ref[ei_packet_traits::size*4]; - for (int i=0; i(); - data2[i] = ei_random(); - } - - ei_pstore(data2, ei_pload(data1)); - VERIFY(areApprox(data1, data2, PacketSize) && "aligned load/store"); - - for (int offset=0; offset(packets[0], packets[1]); - else if (offset==1) ei_palign<1>(packets[0], packets[1]); - else if (offset==2) ei_palign<2>(packets[0], packets[1]); - else if (offset==3) ei_palign<3>(packets[0], packets[1]); - ei_pstore(data2, packets[0]); - - for (int i=0; i Vector; - VERIFY(areApprox(ref, data2, PacketSize) && "ei_palign"); - } - - CHECK_CWISE(REF_ADD, ei_padd); - CHECK_CWISE(REF_SUB, ei_psub); - CHECK_CWISE(REF_MUL, ei_pmul); - #ifndef EIGEN_VECTORIZE_ALTIVEC - if (!ei_is_same_type::ret) - CHECK_CWISE(REF_DIV, ei_pdiv); - #endif - CHECK_CWISE(std::min, ei_pmin); - CHECK_CWISE(std::max, ei_pmax); - - for (int i=0; i() ); - CALL_SUBTEST_2( packetmath() ); - CALL_SUBTEST_3( packetmath() ); - CALL_SUBTEST_4( packetmath >() ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_parametrizedline.cpp b/thirdparty/eigen/test/eigen2/eigen2_parametrizedline.cpp deleted file mode 100644 index 81472887..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_parametrizedline.cpp +++ /dev/null @@ -1,62 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include -#include -#include - -template void parametrizedline(const LineType& _line) -{ - /* this test covers the following files: - ParametrizedLine.h - */ - - const int dim = _line.dim(); - typedef typename LineType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - typedef Matrix MatrixType; - - VectorType p0 = VectorType::Random(dim); - VectorType p1 = VectorType::Random(dim); - - VectorType d0 = VectorType::Random(dim).normalized(); - - LineType l0(p0, d0); - - Scalar s0 = ei_random(); - Scalar s1 = ei_abs(ei_random()); - - VERIFY_IS_MUCH_SMALLER_THAN( l0.distance(p0), RealScalar(1) ); - VERIFY_IS_MUCH_SMALLER_THAN( l0.distance(p0+s0*d0), RealScalar(1) ); - VERIFY_IS_APPROX( (l0.projection(p1)-p1).norm(), l0.distance(p1) ); - VERIFY_IS_MUCH_SMALLER_THAN( l0.distance(l0.projection(p1)), RealScalar(1) ); - VERIFY_IS_APPROX( Scalar(l0.distance((p0+s0*d0) + d0.unitOrthogonal() * s1)), s1 ); - - // casting - const int Dim = LineType::AmbientDimAtCompileTime; - typedef typename GetDifferentType::type OtherScalar; - ParametrizedLine hp1f = l0.template cast(); - VERIFY_IS_APPROX(hp1f.template cast(),l0); - ParametrizedLine hp1d = l0.template cast(); - VERIFY_IS_APPROX(hp1d.template cast(),l0); -} - -void test_eigen2_parametrizedline() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( parametrizedline(ParametrizedLine()) ); - CALL_SUBTEST_2( parametrizedline(ParametrizedLine()) ); - CALL_SUBTEST_3( parametrizedline(ParametrizedLine()) ); - CALL_SUBTEST_4( parametrizedline(ParametrizedLine,5>()) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_prec_inverse_4x4.cpp b/thirdparty/eigen/test/eigen2/eigen2_prec_inverse_4x4.cpp deleted file mode 100644 index 8bfa5569..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_prec_inverse_4x4.cpp +++ /dev/null @@ -1,84 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2009 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include -#include - -template std::string type_name() { return "other"; } -template<> std::string type_name() { return "float"; } -template<> std::string type_name() { return "double"; } -template<> std::string type_name() { return "int"; } -template<> std::string type_name >() { return "complex"; } -template<> std::string type_name >() { return "complex"; } -template<> std::string type_name >() { return "complex"; } - -#define EIGEN_DEBUG_VAR(x) std::cerr << #x << " = " << x << std::endl; - -template inline typename NumTraits::Real epsilon() -{ - return std::numeric_limits::Real>::epsilon(); -} - -template void inverse_permutation_4x4() -{ - typedef typename MatrixType::Scalar Scalar; - typedef typename MatrixType::RealScalar RealScalar; - Vector4i indices(0,1,2,3); - for(int i = 0; i < 24; ++i) - { - MatrixType m = MatrixType::Zero(); - m(indices(0),0) = 1; - m(indices(1),1) = 1; - m(indices(2),2) = 1; - m(indices(3),3) = 1; - MatrixType inv = m.inverse(); - double error = double( (m*inv-MatrixType::Identity()).norm() / epsilon() ); - VERIFY(error == 0.0); - std::next_permutation(indices.data(),indices.data()+4); - } -} - -template void inverse_general_4x4(int repeat) -{ - typedef typename MatrixType::Scalar Scalar; - typedef typename MatrixType::RealScalar RealScalar; - double error_sum = 0., error_max = 0.; - for(int i = 0; i < repeat; ++i) - { - MatrixType m; - RealScalar absdet; - do { - m = MatrixType::Random(); - absdet = ei_abs(m.determinant()); - } while(absdet < 10 * epsilon()); - MatrixType inv = m.inverse(); - double error = double( (m*inv-MatrixType::Identity()).norm() * absdet / epsilon() ); - error_sum += error; - error_max = std::max(error_max, error); - } - std::cerr << "inverse_general_4x4, Scalar = " << type_name() << std::endl; - double error_avg = error_sum / repeat; - EIGEN_DEBUG_VAR(error_avg); - EIGEN_DEBUG_VAR(error_max); - VERIFY(error_avg < (NumTraits::IsComplex ? 8.0 : 1.25)); - VERIFY(error_max < (NumTraits::IsComplex ? 64.0 : 20.0)); -} - -void test_eigen2_prec_inverse_4x4() -{ - CALL_SUBTEST_1((inverse_permutation_4x4())); - CALL_SUBTEST_1(( inverse_general_4x4(200000 * g_repeat) )); - - CALL_SUBTEST_2((inverse_permutation_4x4 >())); - CALL_SUBTEST_2(( inverse_general_4x4 >(200000 * g_repeat) )); - - CALL_SUBTEST_3((inverse_permutation_4x4())); - CALL_SUBTEST_3((inverse_general_4x4(50000 * g_repeat))); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_product_large.cpp b/thirdparty/eigen/test/eigen2/eigen2_product_large.cpp deleted file mode 100644 index 5149ef74..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_product_large.cpp +++ /dev/null @@ -1,45 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "product.h" - -void test_eigen2_product_large() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( product(MatrixXf(ei_random(1,320), ei_random(1,320))) ); - CALL_SUBTEST_2( product(MatrixXd(ei_random(1,320), ei_random(1,320))) ); - CALL_SUBTEST_3( product(MatrixXi(ei_random(1,320), ei_random(1,320))) ); - CALL_SUBTEST_4( product(MatrixXcf(ei_random(1,50), ei_random(1,50))) ); - CALL_SUBTEST_5( product(Matrix(ei_random(1,320), ei_random(1,320))) ); - } - -#ifdef EIGEN_TEST_PART_6 - { - // test a specific issue in DiagonalProduct - int N = 1000000; - VectorXf v = VectorXf::Ones(N); - MatrixXf m = MatrixXf::Ones(N,3); - m = (v+v).asDiagonal() * m; - VERIFY_IS_APPROX(m, MatrixXf::Constant(N,3,2)); - } - - { - // test deferred resizing in Matrix::operator= - MatrixXf a = MatrixXf::Random(10,4), b = MatrixXf::Random(4,10), c = a; - VERIFY_IS_APPROX((a = a * b), (c * b).eval()); - } - - { - MatrixXf mat1(10,10); mat1.setRandom(); - MatrixXf mat2(32,10); mat2.setRandom(); - MatrixXf result = mat1.row(2)*mat2.transpose(); - VERIFY_IS_APPROX(result, (mat1.row(2)*mat2.transpose()).eval()); - } -#endif -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_product_small.cpp b/thirdparty/eigen/test/eigen2/eigen2_product_small.cpp deleted file mode 100644 index 4cd8c102..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_product_small.cpp +++ /dev/null @@ -1,22 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#define EIGEN_NO_STATIC_ASSERT -#include "product.h" - -void test_eigen2_product_small() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( product(Matrix()) ); - CALL_SUBTEST_2( product(Matrix()) ); - CALL_SUBTEST_3( product(Matrix3d()) ); - CALL_SUBTEST_4( product(Matrix4d()) ); - CALL_SUBTEST_5( product(Matrix4f()) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_qr.cpp b/thirdparty/eigen/test/eigen2/eigen2_qr.cpp deleted file mode 100644 index 76977e4c..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_qr.cpp +++ /dev/null @@ -1,69 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -template void qr(const MatrixType& m) -{ - /* this test covers the following files: - QR.h - */ - int rows = m.rows(); - int cols = m.cols(); - - typedef typename MatrixType::Scalar Scalar; - typedef Matrix SquareMatrixType; - typedef Matrix VectorType; - - MatrixType a = MatrixType::Random(rows,cols); - QR qrOfA(a); - VERIFY_IS_APPROX(a, qrOfA.matrixQ() * qrOfA.matrixR()); - VERIFY_IS_NOT_APPROX(a+MatrixType::Identity(rows, cols), qrOfA.matrixQ() * qrOfA.matrixR()); - - #if 0 // eigenvalues module not yet ready - SquareMatrixType b = a.adjoint() * a; - - // check tridiagonalization - Tridiagonalization tridiag(b); - VERIFY_IS_APPROX(b, tridiag.matrixQ() * tridiag.matrixT() * tridiag.matrixQ().adjoint()); - - // check hessenberg decomposition - HessenbergDecomposition hess(b); - VERIFY_IS_APPROX(b, hess.matrixQ() * hess.matrixH() * hess.matrixQ().adjoint()); - VERIFY_IS_APPROX(tridiag.matrixT(), hess.matrixH()); - b = SquareMatrixType::Random(cols,cols); - hess.compute(b); - VERIFY_IS_APPROX(b, hess.matrixQ() * hess.matrixH() * hess.matrixQ().adjoint()); - #endif -} - -void test_eigen2_qr() -{ - for(int i = 0; i < 1; i++) { - CALL_SUBTEST_1( qr(Matrix2f()) ); - CALL_SUBTEST_2( qr(Matrix4d()) ); - CALL_SUBTEST_3( qr(MatrixXf(12,8)) ); - CALL_SUBTEST_4( qr(MatrixXcd(5,5)) ); - CALL_SUBTEST_4( qr(MatrixXcd(7,3)) ); - } - -#ifdef EIGEN_TEST_PART_5 - // small isFullRank test - { - Matrix3d mat; - mat << 1, 45, 1, 2, 2, 2, 1, 2, 3; - VERIFY(mat.qr().isFullRank()); - mat << 1, 1, 1, 2, 2, 2, 1, 2, 3; - //always returns true in eigen2support - //VERIFY(!mat.qr().isFullRank()); - } - -#endif -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_qtvector.cpp b/thirdparty/eigen/test/eigen2/eigen2_qtvector.cpp deleted file mode 100644 index 6cfb58a2..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_qtvector.cpp +++ /dev/null @@ -1,158 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#define EIGEN_WORK_AROUND_QT_BUG_CALLING_WRONG_OPERATOR_NEW_FIXED_IN_QT_4_5 - -#include "main.h" - -#include -#include - -#include - -template -void check_qtvector_matrix(const MatrixType& m) -{ - int rows = m.rows(); - int cols = m.cols(); - MatrixType x = MatrixType::Random(rows,cols), y = MatrixType::Random(rows,cols); - QVector v(10, MatrixType(rows,cols)), w(20, y); - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], y); - } - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.fill(y,22); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((size_t)&(v[22]) == (size_t)&(v[21]) + sizeof(MatrixType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - MatrixType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(int i=23; i -void check_qtvector_transform(const TransformType&) -{ - typedef typename TransformType::MatrixType MatrixType; - TransformType x(MatrixType::Random()), y(MatrixType::Random()); - QVector v(10), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.fill(y,22); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((size_t)&(v[22]) == (size_t)&(v[21]) + sizeof(TransformType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - TransformType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; int(i) -void check_qtvector_quaternion(const QuaternionType&) -{ - typedef typename QuaternionType::Coefficients Coefficients; - QuaternionType x(Coefficients::Random()), y(Coefficients::Random()); - QVector v(10), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.fill(y,22); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((size_t)&(v[22]) == (size_t)&(v[21]) + sizeof(QuaternionType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - QuaternionType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; int(i) -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -template -void makeNoisyCohyperplanarPoints(int numPoints, - VectorType **points, - HyperplaneType *hyperplane, - typename VectorType::Scalar noiseAmplitude) -{ - typedef typename VectorType::Scalar Scalar; - const int size = points[0]->size(); - // pick a random hyperplane, store the coefficients of its equation - hyperplane->coeffs().resize(size + 1); - for(int j = 0; j < size + 1; j++) - { - do { - hyperplane->coeffs().coeffRef(j) = ei_random(); - } while(ei_abs(hyperplane->coeffs().coeff(j)) < 0.5); - } - - // now pick numPoints random points on this hyperplane - for(int i = 0; i < numPoints; i++) - { - VectorType& cur_point = *(points[i]); - do - { - cur_point = VectorType::Random(size)/*.normalized()*/; - // project cur_point onto the hyperplane - Scalar x = - (hyperplane->coeffs().start(size).cwise()*cur_point).sum(); - cur_point *= hyperplane->coeffs().coeff(size) / x; - } while( cur_point.norm() < 0.5 - || cur_point.norm() > 2.0 ); - } - - // add some noise to these points - for(int i = 0; i < numPoints; i++ ) - *(points[i]) += noiseAmplitude * VectorType::Random(size); -} - -template -void check_linearRegression(int numPoints, - VectorType **points, - const VectorType& original, - typename VectorType::Scalar tolerance) -{ - int size = points[0]->size(); - assert(size==2); - VectorType result(size); - linearRegression(numPoints, points, &result, 1); - typename VectorType::Scalar error = (result - original).norm() / original.norm(); - VERIFY(ei_abs(error) < ei_abs(tolerance)); -} - -template -void check_fitHyperplane(int numPoints, - VectorType **points, - const HyperplaneType& original, - typename VectorType::Scalar tolerance) -{ - int size = points[0]->size(); - HyperplaneType result(size); - fitHyperplane(numPoints, points, &result); - result.coeffs() *= original.coeffs().coeff(size)/result.coeffs().coeff(size); - typename VectorType::Scalar error = (result.coeffs() - original.coeffs()).norm() / original.coeffs().norm(); - std::cout << ei_abs(error) << " xxx " << ei_abs(tolerance) << std::endl; - VERIFY(ei_abs(error) < ei_abs(tolerance)); -} - -void test_eigen2_regression() -{ - for(int i = 0; i < g_repeat; i++) - { -#ifdef EIGEN_TEST_PART_1 - { - Vector2f points2f [1000]; - Vector2f *points2f_ptrs [1000]; - for(int i = 0; i < 1000; i++) points2f_ptrs[i] = &(points2f[i]); - Vector2f coeffs2f; - Hyperplane coeffs3f; - makeNoisyCohyperplanarPoints(1000, points2f_ptrs, &coeffs3f, 0.01f); - coeffs2f[0] = -coeffs3f.coeffs()[0]/coeffs3f.coeffs()[1]; - coeffs2f[1] = -coeffs3f.coeffs()[2]/coeffs3f.coeffs()[1]; - CALL_SUBTEST(check_linearRegression(10, points2f_ptrs, coeffs2f, 0.05f)); - CALL_SUBTEST(check_linearRegression(100, points2f_ptrs, coeffs2f, 0.01f)); - CALL_SUBTEST(check_linearRegression(1000, points2f_ptrs, coeffs2f, 0.002f)); - } -#endif -#ifdef EIGEN_TEST_PART_2 - { - Vector2f points2f [1000]; - Vector2f *points2f_ptrs [1000]; - for(int i = 0; i < 1000; i++) points2f_ptrs[i] = &(points2f[i]); - Hyperplane coeffs3f; - makeNoisyCohyperplanarPoints(1000, points2f_ptrs, &coeffs3f, 0.01f); - CALL_SUBTEST(check_fitHyperplane(10, points2f_ptrs, coeffs3f, 0.05f)); - CALL_SUBTEST(check_fitHyperplane(100, points2f_ptrs, coeffs3f, 0.01f)); - CALL_SUBTEST(check_fitHyperplane(1000, points2f_ptrs, coeffs3f, 0.002f)); - } -#endif -#ifdef EIGEN_TEST_PART_3 - { - Vector4d points4d [1000]; - Vector4d *points4d_ptrs [1000]; - for(int i = 0; i < 1000; i++) points4d_ptrs[i] = &(points4d[i]); - Hyperplane coeffs5d; - makeNoisyCohyperplanarPoints(1000, points4d_ptrs, &coeffs5d, 0.01); - CALL_SUBTEST(check_fitHyperplane(10, points4d_ptrs, coeffs5d, 0.05)); - CALL_SUBTEST(check_fitHyperplane(100, points4d_ptrs, coeffs5d, 0.01)); - CALL_SUBTEST(check_fitHyperplane(1000, points4d_ptrs, coeffs5d, 0.002)); - } -#endif -#ifdef EIGEN_TEST_PART_4 - { - VectorXcd *points11cd_ptrs[1000]; - for(int i = 0; i < 1000; i++) points11cd_ptrs[i] = new VectorXcd(11); - Hyperplane,Dynamic> *coeffs12cd = new Hyperplane,Dynamic>(11); - makeNoisyCohyperplanarPoints(1000, points11cd_ptrs, coeffs12cd, 0.01); - CALL_SUBTEST(check_fitHyperplane(100, points11cd_ptrs, *coeffs12cd, 0.025)); - CALL_SUBTEST(check_fitHyperplane(1000, points11cd_ptrs, *coeffs12cd, 0.006)); - delete coeffs12cd; - for(int i = 0; i < 1000; i++) delete points11cd_ptrs[i]; - } -#endif - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_sizeof.cpp b/thirdparty/eigen/test/eigen2/eigen2_sizeof.cpp deleted file mode 100644 index ec1af5a0..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_sizeof.cpp +++ /dev/null @@ -1,31 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void verifySizeOf(const MatrixType&) -{ - typedef typename MatrixType::Scalar Scalar; - if (MatrixType::RowsAtCompileTime!=Dynamic && MatrixType::ColsAtCompileTime!=Dynamic) - VERIFY(sizeof(MatrixType)==sizeof(Scalar)*MatrixType::SizeAtCompileTime); - else - VERIFY(sizeof(MatrixType)==sizeof(Scalar*) + 2 * sizeof(typename MatrixType::Index)); -} - -void test_eigen2_sizeof() -{ - CALL_SUBTEST( verifySizeOf(Matrix()) ); - CALL_SUBTEST( verifySizeOf(Matrix4d()) ); - CALL_SUBTEST( verifySizeOf(Matrix()) ); - CALL_SUBTEST( verifySizeOf(Matrix()) ); - CALL_SUBTEST( verifySizeOf(MatrixXcf(3, 3)) ); - CALL_SUBTEST( verifySizeOf(MatrixXi(8, 12)) ); - CALL_SUBTEST( verifySizeOf(MatrixXcd(20, 20)) ); - CALL_SUBTEST( verifySizeOf(Matrix()) ); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_smallvectors.cpp b/thirdparty/eigen/test/eigen2/eigen2_smallvectors.cpp deleted file mode 100644 index 03962b17..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_smallvectors.cpp +++ /dev/null @@ -1,42 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void smallVectors() -{ - typedef Matrix V2; - typedef Matrix V3; - typedef Matrix V4; - Scalar x1 = ei_random(), - x2 = ei_random(), - x3 = ei_random(), - x4 = ei_random(); - V2 v2(x1, x2); - V3 v3(x1, x2, x3); - V4 v4(x1, x2, x3, x4); - VERIFY_IS_APPROX(x1, v2.x()); - VERIFY_IS_APPROX(x1, v3.x()); - VERIFY_IS_APPROX(x1, v4.x()); - VERIFY_IS_APPROX(x2, v2.y()); - VERIFY_IS_APPROX(x2, v3.y()); - VERIFY_IS_APPROX(x2, v4.y()); - VERIFY_IS_APPROX(x3, v3.z()); - VERIFY_IS_APPROX(x3, v4.z()); - VERIFY_IS_APPROX(x4, v4.w()); -} - -void test_eigen2_smallvectors() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST( smallVectors() ); - CALL_SUBTEST( smallVectors() ); - CALL_SUBTEST( smallVectors() ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_sparse_basic.cpp b/thirdparty/eigen/test/eigen2/eigen2_sparse_basic.cpp deleted file mode 100644 index 04907767..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_sparse_basic.cpp +++ /dev/null @@ -1,317 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Daniel Gomez Ferro -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "sparse.h" - -template -bool test_random_setter(SparseMatrix& sm, const DenseType& ref, const std::vector& nonzeroCoords) -{ - typedef SparseMatrix SparseType; - { - sm.setZero(); - SetterType w(sm); - std::vector remaining = nonzeroCoords; - while(!remaining.empty()) - { - int i = ei_random(0,remaining.size()-1); - w(remaining[i].x(),remaining[i].y()) = ref.coeff(remaining[i].x(),remaining[i].y()); - remaining[i] = remaining.back(); - remaining.pop_back(); - } - } - return sm.isApprox(ref); -} - -template -bool test_random_setter(DynamicSparseMatrix& sm, const DenseType& ref, const std::vector& nonzeroCoords) -{ - sm.setZero(); - std::vector remaining = nonzeroCoords; - while(!remaining.empty()) - { - int i = ei_random(0,remaining.size()-1); - sm.coeffRef(remaining[i].x(),remaining[i].y()) = ref.coeff(remaining[i].x(),remaining[i].y()); - remaining[i] = remaining.back(); - remaining.pop_back(); - } - return sm.isApprox(ref); -} - -template void sparse_basic(const SparseMatrixType& ref) -{ - const int rows = ref.rows(); - const int cols = ref.cols(); - typedef typename SparseMatrixType::Scalar Scalar; - enum { Flags = SparseMatrixType::Flags }; - - double density = std::max(8./(rows*cols), 0.01); - typedef Matrix DenseMatrix; - typedef Matrix DenseVector; - Scalar eps = 1e-6; - - SparseMatrixType m(rows, cols); - DenseMatrix refMat = DenseMatrix::Zero(rows, cols); - DenseVector vec1 = DenseVector::Random(rows); - Scalar s1 = ei_random(); - - std::vector zeroCoords; - std::vector nonzeroCoords; - initSparse(density, refMat, m, 0, &zeroCoords, &nonzeroCoords); - - if (zeroCoords.size()==0 || nonzeroCoords.size()==0) - return; - - // test coeff and coeffRef - for (int i=0; i<(int)zeroCoords.size(); ++i) - { - VERIFY_IS_MUCH_SMALLER_THAN( m.coeff(zeroCoords[i].x(),zeroCoords[i].y()), eps ); - if(ei_is_same_type >::ret) - VERIFY_RAISES_ASSERT( m.coeffRef(zeroCoords[0].x(),zeroCoords[0].y()) = 5 ); - } - VERIFY_IS_APPROX(m, refMat); - - m.coeffRef(nonzeroCoords[0].x(), nonzeroCoords[0].y()) = Scalar(5); - refMat.coeffRef(nonzeroCoords[0].x(), nonzeroCoords[0].y()) = Scalar(5); - - VERIFY_IS_APPROX(m, refMat); - /* - // test InnerIterators and Block expressions - for (int t=0; t<10; ++t) - { - int j = ei_random(0,cols-1); - int i = ei_random(0,rows-1); - int w = ei_random(1,cols-j-1); - int h = ei_random(1,rows-i-1); - -// VERIFY_IS_APPROX(m.block(i,j,h,w), refMat.block(i,j,h,w)); - for(int c=0; c w(m); -// for (int i=0; icoeffRef(nonzeroCoords[i].x(),nonzeroCoords[i].y()) = refMat.coeff(nonzeroCoords[i].x(),nonzeroCoords[i].y()); -// } -// } -// VERIFY_IS_APPROX(m, refMat); - - // random setter -// { -// m.setZero(); -// VERIFY_IS_NOT_APPROX(m, refMat); -// SparseSetter w(m); -// std::vector remaining = nonzeroCoords; -// while(!remaining.empty()) -// { -// int i = ei_random(0,remaining.size()-1); -// w->coeffRef(remaining[i].x(),remaining[i].y()) = refMat.coeff(remaining[i].x(),remaining[i].y()); -// remaining[i] = remaining.back(); -// remaining.pop_back(); -// } -// } -// VERIFY_IS_APPROX(m, refMat); - - VERIFY(( test_random_setter >(m,refMat,nonzeroCoords) )); - #ifdef EIGEN_UNORDERED_MAP_SUPPORT - VERIFY(( test_random_setter >(m,refMat,nonzeroCoords) )); - #endif - #ifdef _DENSE_HASH_MAP_H_ - VERIFY(( test_random_setter >(m,refMat,nonzeroCoords) )); - #endif - #ifdef _SPARSE_HASH_MAP_H_ - VERIFY(( test_random_setter >(m,refMat,nonzeroCoords) )); - #endif - - // test fillrand - { - DenseMatrix m1(rows,cols); - m1.setZero(); - SparseMatrixType m2(rows,cols); - m2.startFill(); - for (int j=0; j(0,rows-1); - if (m1.coeff(i,j)==Scalar(0)) - m2.fillrand(i,j) = m1(i,j) = ei_random(); - } - } - m2.endFill(); - VERIFY_IS_APPROX(m2,m1); - } - - // test RandomSetter - /*{ - SparseMatrixType m1(rows,cols), m2(rows,cols); - DenseMatrix refM1 = DenseMatrix::Zero(rows, rows); - initSparse(density, refM1, m1); - { - Eigen::RandomSetter setter(m2); - for (int j=0; j(density, refM1, m1); - initSparse(density, refM2, m2); - initSparse(density, refM3, m3); - initSparse(density, refM4, m4); - - VERIFY_IS_APPROX(m1+m2, refM1+refM2); - VERIFY_IS_APPROX(m1+m2+m3, refM1+refM2+refM3); - VERIFY_IS_APPROX(m3.cwise()*(m1+m2), refM3.cwise()*(refM1+refM2)); - VERIFY_IS_APPROX(m1*s1-m2, refM1*s1-refM2); - - VERIFY_IS_APPROX(m1*=s1, refM1*=s1); - VERIFY_IS_APPROX(m1/=s1, refM1/=s1); - - VERIFY_IS_APPROX(m1+=m2, refM1+=refM2); - VERIFY_IS_APPROX(m1-=m2, refM1-=refM2); - - VERIFY_IS_APPROX(m1.col(0).eigen2_dot(refM2.row(0)), refM1.col(0).eigen2_dot(refM2.row(0))); - - refM4.setRandom(); - // sparse cwise* dense - VERIFY_IS_APPROX(m3.cwise()*refM4, refM3.cwise()*refM4); -// VERIFY_IS_APPROX(m3.cwise()/refM4, refM3.cwise()/refM4); - } - - // test innerVector() - { - DenseMatrix refMat2 = DenseMatrix::Zero(rows, rows); - SparseMatrixType m2(rows, rows); - initSparse(density, refMat2, m2); - int j0 = ei_random(0,rows-1); - int j1 = ei_random(0,rows-1); - VERIFY_IS_APPROX(m2.innerVector(j0), refMat2.col(j0)); - VERIFY_IS_APPROX(m2.innerVector(j0)+m2.innerVector(j1), refMat2.col(j0)+refMat2.col(j1)); - //m2.innerVector(j0) = 2*m2.innerVector(j1); - //refMat2.col(j0) = 2*refMat2.col(j1); - //VERIFY_IS_APPROX(m2, refMat2); - } - - // test innerVectors() - { - DenseMatrix refMat2 = DenseMatrix::Zero(rows, rows); - SparseMatrixType m2(rows, rows); - initSparse(density, refMat2, m2); - int j0 = ei_random(0,rows-2); - int j1 = ei_random(0,rows-2); - int n0 = ei_random(1,rows-std::max(j0,j1)); - VERIFY_IS_APPROX(m2.innerVectors(j0,n0), refMat2.block(0,j0,rows,n0)); - VERIFY_IS_APPROX(m2.innerVectors(j0,n0)+m2.innerVectors(j1,n0), - refMat2.block(0,j0,rows,n0)+refMat2.block(0,j1,rows,n0)); - //m2.innerVectors(j0,n0) = m2.innerVectors(j0,n0) + m2.innerVectors(j1,n0); - //refMat2.block(0,j0,rows,n0) = refMat2.block(0,j0,rows,n0) + refMat2.block(0,j1,rows,n0); - } - - // test transpose - { - DenseMatrix refMat2 = DenseMatrix::Zero(rows, rows); - SparseMatrixType m2(rows, rows); - initSparse(density, refMat2, m2); - VERIFY_IS_APPROX(m2.transpose().eval(), refMat2.transpose().eval()); - VERIFY_IS_APPROX(m2.transpose(), refMat2.transpose()); - } - - // test prune - { - SparseMatrixType m2(rows, rows); - DenseMatrix refM2(rows, rows); - refM2.setZero(); - int countFalseNonZero = 0; - int countTrueNonZero = 0; - m2.startFill(); - for (int j=0; j(0,1); - if (x<0.1) - { - // do nothing - } - else if (x<0.5) - { - countFalseNonZero++; - m2.fill(i,j) = Scalar(0); - } - else - { - countTrueNonZero++; - m2.fill(i,j) = refM2(i,j) = Scalar(1); - } - } - m2.endFill(); - VERIFY(countFalseNonZero+countTrueNonZero == m2.nonZeros()); - VERIFY_IS_APPROX(m2, refM2); - m2.prune(1); - VERIFY(countTrueNonZero==m2.nonZeros()); - VERIFY_IS_APPROX(m2, refM2); - } -} - -void test_eigen2_sparse_basic() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( sparse_basic(SparseMatrix(8, 8)) ); - CALL_SUBTEST_2( sparse_basic(SparseMatrix >(16, 16)) ); - CALL_SUBTEST_1( sparse_basic(SparseMatrix(33, 33)) ); - - CALL_SUBTEST_3( sparse_basic(DynamicSparseMatrix(8, 8)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_sparse_product.cpp b/thirdparty/eigen/test/eigen2/eigen2_sparse_product.cpp deleted file mode 100644 index d28e76df..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_sparse_product.cpp +++ /dev/null @@ -1,115 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Daniel Gomez Ferro -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "sparse.h" - -template void sparse_product(const SparseMatrixType& ref) -{ - const int rows = ref.rows(); - const int cols = ref.cols(); - typedef typename SparseMatrixType::Scalar Scalar; - enum { Flags = SparseMatrixType::Flags }; - - double density = std::max(8./(rows*cols), 0.01); - typedef Matrix DenseMatrix; - typedef Matrix DenseVector; - - // test matrix-matrix product - { - DenseMatrix refMat2 = DenseMatrix::Zero(rows, rows); - DenseMatrix refMat3 = DenseMatrix::Zero(rows, rows); - DenseMatrix refMat4 = DenseMatrix::Zero(rows, rows); - DenseMatrix dm4 = DenseMatrix::Zero(rows, rows); - SparseMatrixType m2(rows, rows); - SparseMatrixType m3(rows, rows); - SparseMatrixType m4(rows, rows); - initSparse(density, refMat2, m2); - initSparse(density, refMat3, m3); - initSparse(density, refMat4, m4); - VERIFY_IS_APPROX(m4=m2*m3, refMat4=refMat2*refMat3); - VERIFY_IS_APPROX(m4=m2.transpose()*m3, refMat4=refMat2.transpose()*refMat3); - VERIFY_IS_APPROX(m4=m2.transpose()*m3.transpose(), refMat4=refMat2.transpose()*refMat3.transpose()); - VERIFY_IS_APPROX(m4=m2*m3.transpose(), refMat4=refMat2*refMat3.transpose()); - - // sparse * dense - VERIFY_IS_APPROX(dm4=m2*refMat3, refMat4=refMat2*refMat3); - VERIFY_IS_APPROX(dm4=m2*refMat3.transpose(), refMat4=refMat2*refMat3.transpose()); - VERIFY_IS_APPROX(dm4=m2.transpose()*refMat3, refMat4=refMat2.transpose()*refMat3); - VERIFY_IS_APPROX(dm4=m2.transpose()*refMat3.transpose(), refMat4=refMat2.transpose()*refMat3.transpose()); - - // dense * sparse - VERIFY_IS_APPROX(dm4=refMat2*m3, refMat4=refMat2*refMat3); - VERIFY_IS_APPROX(dm4=refMat2*m3.transpose(), refMat4=refMat2*refMat3.transpose()); - VERIFY_IS_APPROX(dm4=refMat2.transpose()*m3, refMat4=refMat2.transpose()*refMat3); - VERIFY_IS_APPROX(dm4=refMat2.transpose()*m3.transpose(), refMat4=refMat2.transpose()*refMat3.transpose()); - - VERIFY_IS_APPROX(m3=m3*m3, refMat3=refMat3*refMat3); - } - - // test matrix - diagonal product - if(false) // it compiles, but the precision is terrible. probably doesn't matter in this branch.... - { - DenseMatrix refM2 = DenseMatrix::Zero(rows, rows); - DenseMatrix refM3 = DenseMatrix::Zero(rows, rows); - DiagonalMatrix d1(DenseVector::Random(rows)); - SparseMatrixType m2(rows, rows); - SparseMatrixType m3(rows, rows); - initSparse(density, refM2, m2); - initSparse(density, refM3, m3); - VERIFY_IS_APPROX(m3=m2*d1, refM3=refM2*d1); - VERIFY_IS_APPROX(m3=m2.transpose()*d1, refM3=refM2.transpose()*d1); - VERIFY_IS_APPROX(m3=d1*m2, refM3=d1*refM2); - VERIFY_IS_APPROX(m3=d1*m2.transpose(), refM3=d1 * refM2.transpose()); - } - - // test self adjoint products - { - DenseMatrix b = DenseMatrix::Random(rows, rows); - DenseMatrix x = DenseMatrix::Random(rows, rows); - DenseMatrix refX = DenseMatrix::Random(rows, rows); - DenseMatrix refUp = DenseMatrix::Zero(rows, rows); - DenseMatrix refLo = DenseMatrix::Zero(rows, rows); - DenseMatrix refS = DenseMatrix::Zero(rows, rows); - SparseMatrixType mUp(rows, rows); - SparseMatrixType mLo(rows, rows); - SparseMatrixType mS(rows, rows); - do { - initSparse(density, refUp, mUp, ForceRealDiag|/*ForceNonZeroDiag|*/MakeUpperTriangular); - } while (refUp.isZero()); - refLo = refUp.transpose().conjugate(); - mLo = mUp.transpose().conjugate(); - refS = refUp + refLo; - refS.diagonal() *= 0.5; - mS = mUp + mLo; - for (int k=0; k()*b, refX=refS*b); - VERIFY_IS_APPROX(x=mLo.template marked()*b, refX=refS*b); - VERIFY_IS_APPROX(x=mS.template marked()*b, refX=refS*b); - } - -} - -void test_eigen2_sparse_product() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( sparse_product(SparseMatrix(8, 8)) ); - CALL_SUBTEST_2( sparse_product(SparseMatrix >(16, 16)) ); - CALL_SUBTEST_1( sparse_product(SparseMatrix(33, 33)) ); - - CALL_SUBTEST_3( sparse_product(DynamicSparseMatrix(8, 8)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_sparse_solvers.cpp b/thirdparty/eigen/test/eigen2/eigen2_sparse_solvers.cpp deleted file mode 100644 index 3aef27ab..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_sparse_solvers.cpp +++ /dev/null @@ -1,200 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Daniel Gomez Ferro -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "sparse.h" - -template void -initSPD(double density, - Matrix& refMat, - SparseMatrix& sparseMat) -{ - Matrix aux(refMat.rows(),refMat.cols()); - initSparse(density,refMat,sparseMat); - refMat = refMat * refMat.adjoint(); - for (int k=0; k<2; ++k) - { - initSparse(density,aux,sparseMat,ForceNonZeroDiag); - refMat += aux * aux.adjoint(); - } - sparseMat.startFill(); - for (int j=0 ; j void sparse_solvers(int rows, int cols) -{ - double density = std::max(8./(rows*cols), 0.01); - typedef Matrix DenseMatrix; - typedef Matrix DenseVector; - // Scalar eps = 1e-6; - - DenseVector vec1 = DenseVector::Random(rows); - - std::vector zeroCoords; - std::vector nonzeroCoords; - - // test triangular solver - { - DenseVector vec2 = vec1, vec3 = vec1; - SparseMatrix m2(rows, cols); - DenseMatrix refMat2 = DenseMatrix::Zero(rows, cols); - - // lower - initSparse(density, refMat2, m2, ForceNonZeroDiag|MakeLowerTriangular, &zeroCoords, &nonzeroCoords); - VERIFY_IS_APPROX(refMat2.template marked().solveTriangular(vec2), - m2.template marked().solveTriangular(vec3)); - - // lower - transpose - initSparse(density, refMat2, m2, ForceNonZeroDiag|MakeLowerTriangular, &zeroCoords, &nonzeroCoords); - VERIFY_IS_APPROX(refMat2.template marked().transpose().solveTriangular(vec2), - m2.template marked().transpose().solveTriangular(vec3)); - - // upper - initSparse(density, refMat2, m2, ForceNonZeroDiag|MakeUpperTriangular, &zeroCoords, &nonzeroCoords); - VERIFY_IS_APPROX(refMat2.template marked().solveTriangular(vec2), - m2.template marked().solveTriangular(vec3)); - - // upper - transpose - initSparse(density, refMat2, m2, ForceNonZeroDiag|MakeUpperTriangular, &zeroCoords, &nonzeroCoords); - VERIFY_IS_APPROX(refMat2.template marked().transpose().solveTriangular(vec2), - m2.template marked().transpose().solveTriangular(vec3)); - } - - // test LLT - { - // TODO fix the issue with complex (see SparseLLT::solveInPlace) - SparseMatrix m2(rows, cols); - DenseMatrix refMat2(rows, cols); - - DenseVector b = DenseVector::Random(cols); - DenseVector refX(cols), x(cols); - - initSPD(density, refMat2, m2); - - refMat2.llt().solve(b, &refX); - typedef SparseMatrix SparseSelfAdjointMatrix; - if (!NumTraits::IsComplex) - { - x = b; - SparseLLT (m2).solveInPlace(x); - VERIFY(refX.isApprox(x,test_precision()) && "LLT: default"); - } - #ifdef EIGEN_CHOLMOD_SUPPORT - x = b; - SparseLLT(m2).solveInPlace(x); - VERIFY(refX.isApprox(x,test_precision()) && "LLT: cholmod"); - #endif - if (!NumTraits::IsComplex) - { - #ifdef EIGEN_TAUCS_SUPPORT - x = b; - SparseLLT(m2,IncompleteFactorization).solveInPlace(x); - VERIFY(refX.isApprox(x,test_precision()) && "LLT: taucs (IncompleteFactorization)"); - x = b; - SparseLLT(m2,SupernodalMultifrontal).solveInPlace(x); - VERIFY(refX.isApprox(x,test_precision()) && "LLT: taucs (SupernodalMultifrontal)"); - x = b; - SparseLLT(m2,SupernodalLeftLooking).solveInPlace(x); - VERIFY(refX.isApprox(x,test_precision()) && "LLT: taucs (SupernodalLeftLooking)"); - #endif - } - } - - // test LDLT - if (!NumTraits::IsComplex) - { - // TODO fix the issue with complex (see SparseLDLT::solveInPlace) - SparseMatrix m2(rows, cols); - DenseMatrix refMat2(rows, cols); - - DenseVector b = DenseVector::Random(cols); - DenseVector refX(cols), x(cols); - - //initSPD(density, refMat2, m2); - initSparse(density, refMat2, m2, ForceNonZeroDiag|MakeUpperTriangular, 0, 0); - refMat2 += refMat2.adjoint(); - refMat2.diagonal() *= 0.5; - - refMat2.ldlt().solve(b, &refX); - typedef SparseMatrix SparseSelfAdjointMatrix; - x = b; - SparseLDLT ldlt(m2); - if (ldlt.succeeded()) - ldlt.solveInPlace(x); - VERIFY(refX.isApprox(x,test_precision()) && "LDLT: default"); - } - - // test LU - { - static int count = 0; - SparseMatrix m2(rows, cols); - DenseMatrix refMat2(rows, cols); - - DenseVector b = DenseVector::Random(cols); - DenseVector refX(cols), x(cols); - - initSparse(density, refMat2, m2, ForceNonZeroDiag, &zeroCoords, &nonzeroCoords); - - LU refLu(refMat2); - refLu.solve(b, &refX); - #if defined(EIGEN_SUPERLU_SUPPORT) || defined(EIGEN_UMFPACK_SUPPORT) - Scalar refDet = refLu.determinant(); - #endif - x.setZero(); - // // SparseLU > (m2).solve(b,&x); - // // VERIFY(refX.isApprox(x,test_precision()) && "LU: default"); - #ifdef EIGEN_SUPERLU_SUPPORT - { - x.setZero(); - SparseLU,SuperLU> slu(m2); - if (slu.succeeded()) - { - if (slu.solve(b,&x)) { - VERIFY(refX.isApprox(x,test_precision()) && "LU: SuperLU"); - } - // std::cerr << refDet << " == " << slu.determinant() << "\n"; - if (count==0) { - VERIFY_IS_APPROX(refDet,slu.determinant()); // FIXME det is not very stable for complex - } - } - } - #endif - #ifdef EIGEN_UMFPACK_SUPPORT - { - // check solve - x.setZero(); - SparseLU,UmfPack> slu(m2); - if (slu.succeeded()) { - if (slu.solve(b,&x)) { - if (count==0) { - VERIFY(refX.isApprox(x,test_precision()) && "LU: umfpack"); // FIXME solve is not very stable for complex - } - } - VERIFY_IS_APPROX(refDet,slu.determinant()); - // TODO check the extracted data - //std::cerr << slu.matrixL() << "\n"; - } - } - #endif - count++; - } - -} - -void test_eigen2_sparse_solvers() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( sparse_solvers(8, 8) ); - CALL_SUBTEST_2( sparse_solvers >(16, 16) ); - CALL_SUBTEST_1( sparse_solvers(101, 101) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_sparse_vector.cpp b/thirdparty/eigen/test/eigen2/eigen2_sparse_vector.cpp deleted file mode 100644 index e6d2d77a..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_sparse_vector.cpp +++ /dev/null @@ -1,84 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Daniel Gomez Ferro -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "sparse.h" - -template void sparse_vector(int rows, int cols) -{ - double densityMat = std::max(8./(rows*cols), 0.01); - double densityVec = std::max(8./float(rows), 0.1); - typedef Matrix DenseMatrix; - typedef Matrix DenseVector; - typedef SparseVector SparseVectorType; - typedef SparseMatrix SparseMatrixType; - Scalar eps = 1e-6; - - SparseMatrixType m1(rows,cols); - SparseVectorType v1(rows), v2(rows), v3(rows); - DenseMatrix refM1 = DenseMatrix::Zero(rows, cols); - DenseVector refV1 = DenseVector::Random(rows), - refV2 = DenseVector::Random(rows), - refV3 = DenseVector::Random(rows); - - std::vector zerocoords, nonzerocoords; - initSparse(densityVec, refV1, v1, &zerocoords, &nonzerocoords); - initSparse(densityMat, refM1, m1); - - initSparse(densityVec, refV2, v2); - initSparse(densityVec, refV3, v3); - - Scalar s1 = ei_random(); - - // test coeff and coeffRef - for (unsigned int i=0; i(8, 8) ); - CALL_SUBTEST_2( sparse_vector >(16, 16) ); - CALL_SUBTEST_1( sparse_vector(299, 535) ); - } -} - diff --git a/thirdparty/eigen/test/eigen2/eigen2_stdvector.cpp b/thirdparty/eigen/test/eigen2/eigen2_stdvector.cpp deleted file mode 100644 index 6ab05c20..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_stdvector.cpp +++ /dev/null @@ -1,148 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include -#include "main.h" -#include - -template -void check_stdvector_matrix(const MatrixType& m) -{ - int rows = m.rows(); - int cols = m.cols(); - MatrixType x = MatrixType::Random(rows,cols), y = MatrixType::Random(rows,cols); - std::vector > v(10, MatrixType(rows,cols)), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.resize(22,y); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((std::size_t)&(v[22]) == (std::size_t)&(v[21]) + sizeof(MatrixType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - MatrixType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; i -void check_stdvector_transform(const TransformType&) -{ - typedef typename TransformType::MatrixType MatrixType; - TransformType x(MatrixType::Random()), y(MatrixType::Random()); - std::vector > v(10), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.resize(22,y); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((std::size_t)&(v[22]) == (std::size_t)&(v[21]) + sizeof(TransformType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - TransformType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; i -void check_stdvector_quaternion(const QuaternionType&) -{ - typedef typename QuaternionType::Coefficients Coefficients; - QuaternionType x(Coefficients::Random()), y(Coefficients::Random()); - std::vector > v(10), w(20, y); - v[5] = x; - w[6] = v[5]; - VERIFY_IS_APPROX(w[6], v[5]); - v = w; - for(int i = 0; i < 20; i++) - { - VERIFY_IS_APPROX(w[i], v[i]); - } - - v.resize(21); - v[20] = x; - VERIFY_IS_APPROX(v[20], x); - v.resize(22,y); - VERIFY_IS_APPROX(v[21], y); - v.push_back(x); - VERIFY_IS_APPROX(v[22], x); - VERIFY((std::size_t)&(v[22]) == (std::size_t)&(v[21]) + sizeof(QuaternionType)); - - // do a lot of push_back such that the vector gets internally resized - // (with memory reallocation) - QuaternionType* ref = &w[0]; - for(int i=0; i<30 || ((ref==&w[0]) && i<300); ++i) - v.push_back(w[i%w.size()]); - for(unsigned int i=23; i -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -// check minor separately in order to avoid the possible creation of a zero-sized -// array. Comes from a compilation error with gcc-3.4 or gcc-4 with -ansi -pedantic. -// Another solution would be to declare the array like this: T m_data[Size==0?1:Size]; in ei_matrix_storage -// but this is probably not bad to raise such an error at compile time... -template struct CheckMinor -{ - typedef Matrix MatrixType; - CheckMinor(MatrixType& m1, int r1, int c1) - { - int rows = m1.rows(); - int cols = m1.cols(); - - Matrix mi = m1.minor(0,0).eval(); - VERIFY_IS_APPROX(mi, m1.block(1,1,rows-1,cols-1)); - mi = m1.minor(r1,c1); - VERIFY_IS_APPROX(mi.transpose(), m1.transpose().minor(c1,r1)); - //check operator(), both constant and non-constant, on minor() - m1.minor(r1,c1)(0,0) = m1.minor(0,0)(0,0); - } -}; - -template struct CheckMinor -{ - typedef Matrix MatrixType; - CheckMinor(MatrixType&, int, int) {} -}; - -template void submatrices(const MatrixType& m) -{ - /* this test covers the following files: - Row.h Column.h Block.h Minor.h DiagonalCoeffs.h - */ - typedef typename MatrixType::Scalar Scalar; - typedef typename MatrixType::RealScalar RealScalar; - typedef Matrix VectorType; - typedef Matrix RowVectorType; - int rows = m.rows(); - int cols = m.cols(); - - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols), - ones = MatrixType::Ones(rows, cols), - square = Matrix - ::Random(rows, rows); - VectorType v1 = VectorType::Random(rows); - - Scalar s1 = ei_random(); - - int r1 = ei_random(0,rows-1); - int r2 = ei_random(r1,rows-1); - int c1 = ei_random(0,cols-1); - int c2 = ei_random(c1,cols-1); - - //check row() and col() - VERIFY_IS_APPROX(m1.col(c1).transpose(), m1.transpose().row(c1)); - VERIFY_IS_APPROX(square.row(r1).eigen2_dot(m1.col(c1)), (square.lazy() * m1.conjugate())(r1,c1)); - //check operator(), both constant and non-constant, on row() and col() - m1.row(r1) += s1 * m1.row(r2); - m1.col(c1) += s1 * m1.col(c2); - - //check block() - Matrix b1(1,1); b1(0,0) = m1(r1,c1); - RowVectorType br1(m1.block(r1,0,1,cols)); - VectorType bc1(m1.block(0,c1,rows,1)); - VERIFY_IS_APPROX(b1, m1.block(r1,c1,1,1)); - VERIFY_IS_APPROX(m1.row(r1), br1); - VERIFY_IS_APPROX(m1.col(c1), bc1); - //check operator(), both constant and non-constant, on block() - m1.block(r1,c1,r2-r1+1,c2-c1+1) = s1 * m2.block(0, 0, r2-r1+1,c2-c1+1); - m1.block(r1,c1,r2-r1+1,c2-c1+1)(r2-r1,c2-c1) = m2.block(0, 0, r2-r1+1,c2-c1+1)(0,0); - - //check minor() - CheckMinor checkminor(m1,r1,c1); - - //check diagonal() - VERIFY_IS_APPROX(m1.diagonal(), m1.transpose().diagonal()); - m2.diagonal() = 2 * m1.diagonal(); - m2.diagonal()[0] *= 3; - VERIFY_IS_APPROX(m2.diagonal()[0], static_cast(6) * m1.diagonal()[0]); - - enum { - BlockRows = EIGEN_SIZE_MIN_PREFER_FIXED(MatrixType::RowsAtCompileTime,2), - BlockCols = EIGEN_SIZE_MIN_PREFER_FIXED(MatrixType::ColsAtCompileTime,5) - }; - if (rows>=5 && cols>=8) - { - // test fixed block() as lvalue - m1.template block(1,1) *= s1; - // test operator() on fixed block() both as constant and non-constant - m1.template block(1,1)(0, 3) = m1.template block<2,5>(1,1)(1,2); - // check that fixed block() and block() agree - Matrix b = m1.template block(3,3); - VERIFY_IS_APPROX(b, m1.block(3,3,BlockRows,BlockCols)); - } - - if (rows>2) - { - // test sub vectors - VERIFY_IS_APPROX(v1.template start<2>(), v1.block(0,0,2,1)); - VERIFY_IS_APPROX(v1.template start<2>(), v1.start(2)); - VERIFY_IS_APPROX(v1.template start<2>(), v1.segment(0,2)); - VERIFY_IS_APPROX(v1.template start<2>(), v1.template segment<2>(0)); - int i = rows-2; - VERIFY_IS_APPROX(v1.template end<2>(), v1.block(i,0,2,1)); - VERIFY_IS_APPROX(v1.template end<2>(), v1.end(2)); - VERIFY_IS_APPROX(v1.template end<2>(), v1.segment(i,2)); - VERIFY_IS_APPROX(v1.template end<2>(), v1.template segment<2>(i)); - i = ei_random(0,rows-2); - VERIFY_IS_APPROX(v1.segment(i,2), v1.template segment<2>(i)); - } - - // stress some basic stuffs with block matrices - VERIFY(ei_real(ones.col(c1).sum()) == RealScalar(rows)); - VERIFY(ei_real(ones.row(r1).sum()) == RealScalar(cols)); - - VERIFY(ei_real(ones.col(c1).eigen2_dot(ones.col(c2))) == RealScalar(rows)); - VERIFY(ei_real(ones.row(r1).eigen2_dot(ones.row(r2))) == RealScalar(cols)); -} - -void test_eigen2_submatrices() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( submatrices(Matrix()) ); - CALL_SUBTEST_2( submatrices(Matrix4d()) ); - CALL_SUBTEST_3( submatrices(MatrixXcf(3, 3)) ); - CALL_SUBTEST_4( submatrices(MatrixXi(8, 12)) ); - CALL_SUBTEST_5( submatrices(MatrixXcd(20, 20)) ); - CALL_SUBTEST_6( submatrices(MatrixXf(20, 20)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_sum.cpp b/thirdparty/eigen/test/eigen2/eigen2_sum.cpp deleted file mode 100644 index b47057ca..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_sum.cpp +++ /dev/null @@ -1,71 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void matrixSum(const MatrixType& m) -{ - typedef typename MatrixType::Scalar Scalar; - - int rows = m.rows(); - int cols = m.cols(); - - MatrixType m1 = MatrixType::Random(rows, cols); - - VERIFY_IS_MUCH_SMALLER_THAN(MatrixType::Zero(rows, cols).sum(), Scalar(1)); - VERIFY_IS_APPROX(MatrixType::Ones(rows, cols).sum(), Scalar(float(rows*cols))); // the float() here to shut up excessive MSVC warning about int->complex conversion being lossy - Scalar x = Scalar(0); - for(int i = 0; i < rows; i++) for(int j = 0; j < cols; j++) x += m1(i,j); - VERIFY_IS_APPROX(m1.sum(), x); -} - -template void vectorSum(const VectorType& w) -{ - typedef typename VectorType::Scalar Scalar; - int size = w.size(); - - VectorType v = VectorType::Random(size); - for(int i = 1; i < size; i++) - { - Scalar s = Scalar(0); - for(int j = 0; j < i; j++) s += v[j]; - VERIFY_IS_APPROX(s, v.start(i).sum()); - } - - for(int i = 0; i < size-1; i++) - { - Scalar s = Scalar(0); - for(int j = i; j < size; j++) s += v[j]; - VERIFY_IS_APPROX(s, v.end(size-i).sum()); - } - - for(int i = 0; i < size/2; i++) - { - Scalar s = Scalar(0); - for(int j = i; j < size-i; j++) s += v[j]; - VERIFY_IS_APPROX(s, v.segment(i, size-2*i).sum()); - } -} - -void test_eigen2_sum() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( matrixSum(Matrix()) ); - CALL_SUBTEST_2( matrixSum(Matrix2f()) ); - CALL_SUBTEST_3( matrixSum(Matrix4d()) ); - CALL_SUBTEST_4( matrixSum(MatrixXcf(3, 3)) ); - CALL_SUBTEST_5( matrixSum(MatrixXf(8, 12)) ); - CALL_SUBTEST_6( matrixSum(MatrixXi(8, 12)) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_5( vectorSum(VectorXf(5)) ); - CALL_SUBTEST_7( vectorSum(VectorXd(10)) ); - CALL_SUBTEST_5( vectorSum(VectorXf(33)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_svd.cpp b/thirdparty/eigen/test/eigen2/eigen2_svd.cpp deleted file mode 100644 index d4689a56..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_svd.cpp +++ /dev/null @@ -1,87 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include - -template void svd(const MatrixType& m) -{ - /* this test covers the following files: - SVD.h - */ - int rows = m.rows(); - int cols = m.cols(); - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - MatrixType a = MatrixType::Random(rows,cols); - Matrix b = - Matrix::Random(rows,1); - Matrix x(cols,1), x2(cols,1); - - RealScalar largerEps = test_precision(); - if (ei_is_same_type::ret) - largerEps = 1e-3f; - - { - SVD svd(a); - MatrixType sigma = MatrixType::Zero(rows,cols); - MatrixType matU = MatrixType::Zero(rows,rows); - sigma.block(0,0,cols,cols) = svd.singularValues().asDiagonal(); - matU.block(0,0,rows,cols) = svd.matrixU(); - VERIFY_IS_APPROX(a, matU * sigma * svd.matrixV().transpose()); - } - - - if (rows==cols) - { - if (ei_is_same_type::ret) - { - MatrixType a1 = MatrixType::Random(rows,cols); - a += a * a.adjoint() + a1 * a1.adjoint(); - } - SVD svd(a); - svd.solve(b, &x); - VERIFY_IS_APPROX(a * x,b); - } - - - if(rows==cols) - { - SVD svd(a); - MatrixType unitary, positive; - svd.computeUnitaryPositive(&unitary, &positive); - VERIFY_IS_APPROX(unitary * unitary.adjoint(), MatrixType::Identity(unitary.rows(),unitary.rows())); - VERIFY_IS_APPROX(positive, positive.adjoint()); - for(int i = 0; i < rows; i++) VERIFY(positive.diagonal()[i] >= 0); // cheap necessary (not sufficient) condition for positivity - VERIFY_IS_APPROX(unitary*positive, a); - - svd.computePositiveUnitary(&positive, &unitary); - VERIFY_IS_APPROX(unitary * unitary.adjoint(), MatrixType::Identity(unitary.rows(),unitary.rows())); - VERIFY_IS_APPROX(positive, positive.adjoint()); - for(int i = 0; i < rows; i++) VERIFY(positive.diagonal()[i] >= 0); // cheap necessary (not sufficient) condition for positivity - VERIFY_IS_APPROX(positive*unitary, a); - } -} - -void test_eigen2_svd() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( svd(Matrix3f()) ); - CALL_SUBTEST_2( svd(Matrix4d()) ); - CALL_SUBTEST_3( svd(MatrixXf(7,7)) ); - CALL_SUBTEST_4( svd(MatrixXd(14,7)) ); - // complex are not implemented yet -// CALL_SUBTEST( svd(MatrixXcd(6,6)) ); -// CALL_SUBTEST( svd(MatrixXcf(3,3)) ); - SVD s; - MatrixXf m = MatrixXf::Random(10,1); - s.compute(m); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_swap.cpp b/thirdparty/eigen/test/eigen2/eigen2_swap.cpp deleted file mode 100644 index f3a8846d..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_swap.cpp +++ /dev/null @@ -1,83 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2009 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#define EIGEN_NO_STATIC_ASSERT -#include "main.h" - -template -struct other_matrix_type -{ - typedef int type; -}; - -template -struct other_matrix_type > -{ - typedef Matrix<_Scalar, _Rows, _Cols, _Options^RowMajor, _MaxRows, _MaxCols> type; -}; - -template void swap(const MatrixType& m) -{ - typedef typename other_matrix_type::type OtherMatrixType; - typedef typename MatrixType::Scalar Scalar; - - ei_assert((!ei_is_same_type::ret)); - int rows = m.rows(); - int cols = m.cols(); - - // construct 3 matrix guaranteed to be distinct - MatrixType m1 = MatrixType::Random(rows,cols); - MatrixType m2 = MatrixType::Random(rows,cols) + Scalar(100) * MatrixType::Identity(rows,cols); - OtherMatrixType m3 = OtherMatrixType::Random(rows,cols) + Scalar(200) * OtherMatrixType::Identity(rows,cols); - - MatrixType m1_copy = m1; - MatrixType m2_copy = m2; - OtherMatrixType m3_copy = m3; - - // test swapping 2 matrices of same type - m1.swap(m2); - VERIFY_IS_APPROX(m1,m2_copy); - VERIFY_IS_APPROX(m2,m1_copy); - m1 = m1_copy; - m2 = m2_copy; - - // test swapping 2 matrices of different types - m1.swap(m3); - VERIFY_IS_APPROX(m1,m3_copy); - VERIFY_IS_APPROX(m3,m1_copy); - m1 = m1_copy; - m3 = m3_copy; - - // test swapping matrix with expression - m1.swap(m2.block(0,0,rows,cols)); - VERIFY_IS_APPROX(m1,m2_copy); - VERIFY_IS_APPROX(m2,m1_copy); - m1 = m1_copy; - m2 = m2_copy; - - // test swapping two expressions of different types - m1.transpose().swap(m3.transpose()); - VERIFY_IS_APPROX(m1,m3_copy); - VERIFY_IS_APPROX(m3,m1_copy); - m1 = m1_copy; - m3 = m3_copy; - - // test assertion on mismatching size -- matrix case - VERIFY_RAISES_ASSERT(m1.swap(m1.row(0))); - // test assertion on mismatching size -- xpr case - VERIFY_RAISES_ASSERT(m1.row(0).swap(m1)); -} - -void test_eigen2_swap() -{ - CALL_SUBTEST_1( swap(Matrix3f()) ); // fixed size, no vectorization - CALL_SUBTEST_1( swap(Matrix4d()) ); // fixed size, possible vectorization - CALL_SUBTEST_1( swap(MatrixXd(3,3)) ); // dyn size, no vectorization - CALL_SUBTEST_1( swap(MatrixXf(30,30)) ); // dyn size, possible vectorization -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_triangular.cpp b/thirdparty/eigen/test/eigen2/eigen2_triangular.cpp deleted file mode 100644 index 6f17b7df..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_triangular.cpp +++ /dev/null @@ -1,148 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void triangular(const MatrixType& m) -{ - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; - - RealScalar largerEps = 10*test_precision(); - - int rows = m.rows(); - int cols = m.cols(); - - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols), - m4(rows, cols), - r1(rows, cols), - r2(rows, cols); - - MatrixType m1up = m1.template part(); - MatrixType m2up = m2.template part(); - - if (rows*cols>1) - { - VERIFY(m1up.isUpperTriangular()); - VERIFY(m2up.transpose().isLowerTriangular()); - VERIFY(!m2.isLowerTriangular()); - } - -// VERIFY_IS_APPROX(m1up.transpose() * m2, m1.upper().transpose().lower() * m2); - - // test overloaded operator+= - r1.setZero(); - r2.setZero(); - r1.template part() += m1; - r2 += m1up; - VERIFY_IS_APPROX(r1,r2); - - // test overloaded operator= - m1.setZero(); - m1.template part() = (m2.transpose() * m2).lazy(); - m3 = m2.transpose() * m2; - VERIFY_IS_APPROX(m3.template part().transpose(), m1); - - // test overloaded operator= - m1.setZero(); - m1.template part() = (m2.transpose() * m2).lazy(); - VERIFY_IS_APPROX(m3.template part(), m1); - - VERIFY_IS_APPROX(m3.template part(), m3.diagonal().asDiagonal()); - - m1 = MatrixType::Random(rows, cols); - for (int i=0; i(); - - Transpose trm4(m4); - // test back and forward subsitution - m3 = m1.template part(); - VERIFY(m3.template marked().solveTriangular(m3).cwise().abs().isIdentity(test_precision())); - VERIFY(m3.transpose().template marked() - .solveTriangular(m3.transpose()).cwise().abs().isIdentity(test_precision())); - // check M * inv(L) using in place API - m4 = m3; - m3.transpose().template marked().solveTriangularInPlace(trm4); - VERIFY(m4.cwise().abs().isIdentity(test_precision())); - - m3 = m1.template part(); - VERIFY(m3.template marked().solveTriangular(m3).cwise().abs().isIdentity(test_precision())); - VERIFY(m3.transpose().template marked() - .solveTriangular(m3.transpose()).cwise().abs().isIdentity(test_precision())); - // check M * inv(U) using in place API - m4 = m3; - m3.transpose().template marked().solveTriangularInPlace(trm4); - VERIFY(m4.cwise().abs().isIdentity(test_precision())); - - m3 = m1.template part(); - VERIFY(m2.isApprox(m3 * (m3.template marked().solveTriangular(m2)), largerEps)); - m3 = m1.template part(); - VERIFY(m2.isApprox(m3 * (m3.template marked().solveTriangular(m2)), largerEps)); - - VERIFY((m1.template part() * m2.template part()).isUpperTriangular()); - - // test swap - m1.setOnes(); - m2.setZero(); - m2.template part().swap(m1); - m3.setZero(); - m3.template part().setOnes(); - VERIFY_IS_APPROX(m2,m3); - -} - -void selfadjoint() -{ - Matrix2i m; - m << 1, 2, - 3, 4; - - Matrix2i m1 = Matrix2i::Zero(); - m1.part() = m; - Matrix2i ref1; - ref1 << 1, 2, - 2, 4; - VERIFY(m1 == ref1); - - Matrix2i m2 = Matrix2i::Zero(); - m2.part() = m.part(); - Matrix2i ref2; - ref2 << 1, 2, - 2, 4; - VERIFY(m2 == ref2); - - Matrix2i m3 = Matrix2i::Zero(); - m3.part() = m.part(); - Matrix2i ref3; - ref3 << 1, 0, - 0, 4; - VERIFY(m3 == ref3); - - // example inspired from bug 159 - int array[] = {1, 2, 3, 4}; - Matrix2i::Map(array).part() = Matrix2i::Random().part(); - - std::cout << "hello\n" << array << std::endl; -} - -void test_eigen2_triangular() -{ - CALL_SUBTEST_8( selfadjoint() ); - for(int i = 0; i < g_repeat ; i++) { - CALL_SUBTEST_1( triangular(Matrix()) ); - CALL_SUBTEST_2( triangular(Matrix()) ); - CALL_SUBTEST_3( triangular(Matrix3d()) ); - CALL_SUBTEST_4( triangular(MatrixXcf(4, 4)) ); - CALL_SUBTEST_5( triangular(Matrix,8, 8>()) ); - CALL_SUBTEST_6( triangular(MatrixXd(17,17)) ); - CALL_SUBTEST_7( triangular(Matrix(5, 5)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_unalignedassert.cpp b/thirdparty/eigen/test/eigen2/eigen2_unalignedassert.cpp deleted file mode 100644 index d10b6f53..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_unalignedassert.cpp +++ /dev/null @@ -1,116 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -struct Good1 -{ - MatrixXd m; // good: m will allocate its own array, taking care of alignment. - Good1() : m(20,20) {} -}; - -struct Good2 -{ - Matrix3d m; // good: m's size isn't a multiple of 16 bytes, so m doesn't have to be aligned -}; - -struct Good3 -{ - Vector2f m; // good: same reason -}; - -struct Bad4 -{ - Vector2d m; // bad: sizeof(m)%16==0 so alignment is required -}; - -struct Bad5 -{ - Matrix m; // bad: same reason -}; - -struct Bad6 -{ - Matrix m; // bad: same reason -}; - -struct Good7 -{ - EIGEN_MAKE_ALIGNED_OPERATOR_NEW - Vector2d m; - float f; // make the struct have sizeof%16!=0 to make it a little more tricky when we allow an array of 2 such objects -}; - -struct Good8 -{ - EIGEN_MAKE_ALIGNED_OPERATOR_NEW - float f; // try the f at first -- the EIGEN_ALIGN_128 attribute of m should make that still work - Matrix4f m; -}; - -struct Good9 -{ - Matrix m; // good: no alignment requested - float f; -}; - -template struct Depends -{ - EIGEN_MAKE_ALIGNED_OPERATOR_NEW_IF(Align) - Vector2d m; - float f; -}; - -template -void check_unalignedassert_good() -{ - T *x, *y; - x = new T; - delete x; - y = new T[2]; - delete[] y; -} - -#if EIGEN_ARCH_WANTS_ALIGNMENT -template -void check_unalignedassert_bad() -{ - float buf[sizeof(T)+16]; - float *unaligned = buf; - while((reinterpret_cast(unaligned)&0xf)==0) ++unaligned; // make sure unaligned is really unaligned - T *x = ::new(static_cast(unaligned)) T; - x->~T(); -} -#endif - -void unalignedassert() -{ - check_unalignedassert_good(); - check_unalignedassert_good(); - check_unalignedassert_good(); -#if EIGEN_ARCH_WANTS_ALIGNMENT - VERIFY_RAISES_ASSERT(check_unalignedassert_bad()); - VERIFY_RAISES_ASSERT(check_unalignedassert_bad()); - VERIFY_RAISES_ASSERT(check_unalignedassert_bad()); -#endif - - check_unalignedassert_good(); - check_unalignedassert_good(); - check_unalignedassert_good(); - check_unalignedassert_good >(); - -#if EIGEN_ARCH_WANTS_ALIGNMENT - VERIFY_RAISES_ASSERT(check_unalignedassert_bad >()); -#endif -} - -void test_eigen2_unalignedassert() -{ - CALL_SUBTEST(unalignedassert()); -} diff --git a/thirdparty/eigen/test/eigen2/eigen2_visitor.cpp b/thirdparty/eigen/test/eigen2/eigen2_visitor.cpp deleted file mode 100644 index 4781991d..00000000 --- a/thirdparty/eigen/test/eigen2/eigen2_visitor.cpp +++ /dev/null @@ -1,116 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. -// -// Copyright (C) 2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" - -template void matrixVisitor(const MatrixType& p) -{ - typedef typename MatrixType::Scalar Scalar; - - int rows = p.rows(); - int cols = p.cols(); - - // construct a random matrix where all coefficients are different - MatrixType m; - m = MatrixType::Random(rows, cols); - for(int i = 0; i < m.size(); i++) - for(int i2 = 0; i2 < i; i2++) - while(m(i) == m(i2)) // yes, == - m(i) = ei_random(); - - Scalar minc = Scalar(1000), maxc = Scalar(-1000); - int minrow=0,mincol=0,maxrow=0,maxcol=0; - for(int j = 0; j < cols; j++) - for(int i = 0; i < rows; i++) - { - if(m(i,j) < minc) - { - minc = m(i,j); - minrow = i; - mincol = j; - } - if(m(i,j) > maxc) - { - maxc = m(i,j); - maxrow = i; - maxcol = j; - } - } - int eigen_minrow, eigen_mincol, eigen_maxrow, eigen_maxcol; - Scalar eigen_minc, eigen_maxc; - eigen_minc = m.minCoeff(&eigen_minrow,&eigen_mincol); - eigen_maxc = m.maxCoeff(&eigen_maxrow,&eigen_maxcol); - VERIFY(minrow == eigen_minrow); - VERIFY(maxrow == eigen_maxrow); - VERIFY(mincol == eigen_mincol); - VERIFY(maxcol == eigen_maxcol); - VERIFY_IS_APPROX(minc, eigen_minc); - VERIFY_IS_APPROX(maxc, eigen_maxc); - VERIFY_IS_APPROX(minc, m.minCoeff()); - VERIFY_IS_APPROX(maxc, m.maxCoeff()); -} - -template void vectorVisitor(const VectorType& w) -{ - typedef typename VectorType::Scalar Scalar; - - int size = w.size(); - - // construct a random vector where all coefficients are different - VectorType v; - v = VectorType::Random(size); - for(int i = 0; i < size; i++) - for(int i2 = 0; i2 < i; i2++) - while(v(i) == v(i2)) // yes, == - v(i) = ei_random(); - - Scalar minc = Scalar(1000), maxc = Scalar(-1000); - int minidx=0,maxidx=0; - for(int i = 0; i < size; i++) - { - if(v(i) < minc) - { - minc = v(i); - minidx = i; - } - if(v(i) > maxc) - { - maxc = v(i); - maxidx = i; - } - } - int eigen_minidx, eigen_maxidx; - Scalar eigen_minc, eigen_maxc; - eigen_minc = v.minCoeff(&eigen_minidx); - eigen_maxc = v.maxCoeff(&eigen_maxidx); - VERIFY(minidx == eigen_minidx); - VERIFY(maxidx == eigen_maxidx); - VERIFY_IS_APPROX(minc, eigen_minc); - VERIFY_IS_APPROX(maxc, eigen_maxc); - VERIFY_IS_APPROX(minc, v.minCoeff()); - VERIFY_IS_APPROX(maxc, v.maxCoeff()); -} - -void test_eigen2_visitor() -{ - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( matrixVisitor(Matrix()) ); - CALL_SUBTEST_2( matrixVisitor(Matrix2f()) ); - CALL_SUBTEST_3( matrixVisitor(Matrix4d()) ); - CALL_SUBTEST_4( matrixVisitor(MatrixXd(8, 12)) ); - CALL_SUBTEST_5( matrixVisitor(Matrix(20, 20)) ); - CALL_SUBTEST_6( matrixVisitor(MatrixXi(8, 12)) ); - } - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_7( vectorVisitor(Vector4f()) ); - CALL_SUBTEST_4( vectorVisitor(VectorXd(10)) ); - CALL_SUBTEST_4( vectorVisitor(RowVectorXd(10)) ); - CALL_SUBTEST_8( vectorVisitor(VectorXf(33)) ); - } -} diff --git a/thirdparty/eigen/test/eigen2/gsl_helper.h b/thirdparty/eigen/test/eigen2/gsl_helper.h deleted file mode 100644 index d1d85453..00000000 --- a/thirdparty/eigen/test/eigen2/gsl_helper.h +++ /dev/null @@ -1,175 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_GSL_HELPER -#define EIGEN_GSL_HELPER - -#include - -#include -#include -#include -#include -#include -#include - -namespace Eigen { - -template::IsComplex> struct GslTraits -{ - typedef gsl_matrix* Matrix; - typedef gsl_vector* Vector; - static Matrix createMatrix(int rows, int cols) { return gsl_matrix_alloc(rows,cols); } - static Vector createVector(int size) { return gsl_vector_alloc(size); } - static void free(Matrix& m) { gsl_matrix_free(m); m=0; } - static void free(Vector& m) { gsl_vector_free(m); m=0; } - static void prod(const Matrix& m, const Vector& v, Vector& x) { gsl_blas_dgemv(CblasNoTrans,1,m,v,0,x); } - static void cholesky(Matrix& m) { gsl_linalg_cholesky_decomp(m); } - static void cholesky_solve(const Matrix& m, const Vector& b, Vector& x) { gsl_linalg_cholesky_solve(m,b,x); } - static void eigen_symm(const Matrix& m, Vector& eval, Matrix& evec) - { - gsl_eigen_symmv_workspace * w = gsl_eigen_symmv_alloc(m->size1); - Matrix a = createMatrix(m->size1, m->size2); - gsl_matrix_memcpy(a, m); - gsl_eigen_symmv(a,eval,evec,w); - gsl_eigen_symmv_sort(eval, evec, GSL_EIGEN_SORT_VAL_ASC); - gsl_eigen_symmv_free(w); - free(a); - } - static void eigen_symm_gen(const Matrix& m, const Matrix& _b, Vector& eval, Matrix& evec) - { - gsl_eigen_gensymmv_workspace * w = gsl_eigen_gensymmv_alloc(m->size1); - Matrix a = createMatrix(m->size1, m->size2); - Matrix b = createMatrix(_b->size1, _b->size2); - gsl_matrix_memcpy(a, m); - gsl_matrix_memcpy(b, _b); - gsl_eigen_gensymmv(a,b,eval,evec,w); - gsl_eigen_symmv_sort(eval, evec, GSL_EIGEN_SORT_VAL_ASC); - gsl_eigen_gensymmv_free(w); - free(a); - } -}; - -template struct GslTraits -{ - typedef gsl_matrix_complex* Matrix; - typedef gsl_vector_complex* Vector; - static Matrix createMatrix(int rows, int cols) { return gsl_matrix_complex_alloc(rows,cols); } - static Vector createVector(int size) { return gsl_vector_complex_alloc(size); } - static void free(Matrix& m) { gsl_matrix_complex_free(m); m=0; } - static void free(Vector& m) { gsl_vector_complex_free(m); m=0; } - static void cholesky(Matrix& m) { gsl_linalg_complex_cholesky_decomp(m); } - static void cholesky_solve(const Matrix& m, const Vector& b, Vector& x) { gsl_linalg_complex_cholesky_solve(m,b,x); } - static void prod(const Matrix& m, const Vector& v, Vector& x) - { gsl_blas_zgemv(CblasNoTrans,gsl_complex_rect(1,0),m,v,gsl_complex_rect(0,0),x); } - static void eigen_symm(const Matrix& m, gsl_vector* &eval, Matrix& evec) - { - gsl_eigen_hermv_workspace * w = gsl_eigen_hermv_alloc(m->size1); - Matrix a = createMatrix(m->size1, m->size2); - gsl_matrix_complex_memcpy(a, m); - gsl_eigen_hermv(a,eval,evec,w); - gsl_eigen_hermv_sort(eval, evec, GSL_EIGEN_SORT_VAL_ASC); - gsl_eigen_hermv_free(w); - free(a); - } - static void eigen_symm_gen(const Matrix& m, const Matrix& _b, gsl_vector* &eval, Matrix& evec) - { - gsl_eigen_genhermv_workspace * w = gsl_eigen_genhermv_alloc(m->size1); - Matrix a = createMatrix(m->size1, m->size2); - Matrix b = createMatrix(_b->size1, _b->size2); - gsl_matrix_complex_memcpy(a, m); - gsl_matrix_complex_memcpy(b, _b); - gsl_eigen_genhermv(a,b,eval,evec,w); - gsl_eigen_hermv_sort(eval, evec, GSL_EIGEN_SORT_VAL_ASC); - gsl_eigen_genhermv_free(w); - free(a); - } -}; - -template -void convert(const MatrixType& m, gsl_matrix* &res) -{ -// if (res) -// gsl_matrix_free(res); - res = gsl_matrix_alloc(m.rows(), m.cols()); - for (int i=0 ; i -void convert(const gsl_matrix* m, MatrixType& res) -{ - res.resize(int(m->size1), int(m->size2)); - for (int i=0 ; i -void convert(const VectorType& m, gsl_vector* &res) -{ - if (res) gsl_vector_free(res); - res = gsl_vector_alloc(m.size()); - for (int i=0 ; i -void convert(const gsl_vector* m, VectorType& res) -{ - res.resize (m->size); - for (int i=0 ; i -void convert(const MatrixType& m, gsl_matrix_complex* &res) -{ - res = gsl_matrix_complex_alloc(m.rows(), m.cols()); - for (int i=0 ; i -void convert(const gsl_matrix_complex* m, MatrixType& res) -{ - res.resize(int(m->size1), int(m->size2)); - for (int i=0 ; i -void convert(const VectorType& m, gsl_vector_complex* &res) -{ - res = gsl_vector_complex_alloc(m.size()); - for (int i=0 ; i -void convert(const gsl_vector_complex* m, VectorType& res) -{ - res.resize(m->size); - for (int i=0 ; i -// Copyright (C) 2008 Gael Guennebaud -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include -#include -#include -#include -#include - -#ifndef EIGEN_TEST_FUNC -#error EIGEN_TEST_FUNC must be defined -#endif - -#define DEFAULT_REPEAT 10 - -namespace Eigen -{ - static std::vector g_test_stack; - static int g_repeat; -} - -#define EI_PP_MAKE_STRING2(S) #S -#define EI_PP_MAKE_STRING(S) EI_PP_MAKE_STRING2(S) - -#define EI_PP_CAT2(a,b) a ## b -#define EI_PP_CAT(a,b) EI_PP_CAT2(a,b) - -#ifndef EIGEN_NO_ASSERTION_CHECKING - - namespace Eigen - { - static const bool should_raise_an_assert = false; - - // Used to avoid to raise two exceptions at a time in which - // case the exception is not properly caught. - // This may happen when a second exceptions is raise in a destructor. - static bool no_more_assert = false; - - struct eigen_assert_exception - { - eigen_assert_exception(void) {} - ~eigen_assert_exception() { Eigen::no_more_assert = false; } - }; - } - - // If EIGEN_DEBUG_ASSERTS is defined and if no assertion is raised while - // one should have been, then the list of excecuted assertions is printed out. - // - // EIGEN_DEBUG_ASSERTS is not enabled by default as it - // significantly increases the compilation time - // and might even introduce side effects that would hide - // some memory errors. - #ifdef EIGEN_DEBUG_ASSERTS - - namespace Eigen - { - static bool ei_push_assert = false; - static std::vector eigen_assert_list; - } - - #define eigen_assert(a) \ - if( (!(a)) && (!no_more_assert) ) \ - { \ - Eigen::no_more_assert = true; \ - throw Eigen::eigen_assert_exception(); \ - } \ - else if (Eigen::ei_push_assert) \ - { \ - eigen_assert_list.push_back(std::string(EI_PP_MAKE_STRING(__FILE__)" ("EI_PP_MAKE_STRING(__LINE__)") : "#a) ); \ - } - - #define VERIFY_RAISES_ASSERT(a) \ - { \ - Eigen::no_more_assert = false; \ - try { \ - Eigen::eigen_assert_list.clear(); \ - Eigen::ei_push_assert = true; \ - a; \ - Eigen::ei_push_assert = false; \ - std::cerr << "One of the following asserts should have been raised:\n"; \ - for (uint ai=0 ; ai - - -#define VERIFY(a) do { if (!(a)) { \ - std::cerr << "Test " << g_test_stack.back() << " failed in "EI_PP_MAKE_STRING(__FILE__) << " (" << EI_PP_MAKE_STRING(__LINE__) << ")" \ - << std::endl << " " << EI_PP_MAKE_STRING(a) << std::endl << std::endl; \ - abort(); \ - } } while (0) - -#define VERIFY_IS_APPROX(a, b) VERIFY(test_ei_isApprox(a, b)) -#define VERIFY_IS_NOT_APPROX(a, b) VERIFY(!test_ei_isApprox(a, b)) -#define VERIFY_IS_MUCH_SMALLER_THAN(a, b) VERIFY(test_ei_isMuchSmallerThan(a, b)) -#define VERIFY_IS_NOT_MUCH_SMALLER_THAN(a, b) VERIFY(!test_ei_isMuchSmallerThan(a, b)) -#define VERIFY_IS_APPROX_OR_LESS_THAN(a, b) VERIFY(test_ei_isApproxOrLessThan(a, b)) -#define VERIFY_IS_NOT_APPROX_OR_LESS_THAN(a, b) VERIFY(!test_ei_isApproxOrLessThan(a, b)) - -#define CALL_SUBTEST(FUNC) do { \ - g_test_stack.push_back(EI_PP_MAKE_STRING(FUNC)); \ - FUNC; \ - g_test_stack.pop_back(); \ - } while (0) - -namespace Eigen { - -template inline typename NumTraits::Real test_precision(); -template<> inline int test_precision() { return 0; } -template<> inline float test_precision() { return 1e-3f; } -template<> inline double test_precision() { return 1e-6; } -template<> inline float test_precision >() { return test_precision(); } -template<> inline double test_precision >() { return test_precision(); } -template<> inline long double test_precision() { return 1e-6; } - -inline bool test_ei_isApprox(const int& a, const int& b) -{ return ei_isApprox(a, b, test_precision()); } -inline bool test_ei_isMuchSmallerThan(const int& a, const int& b) -{ return ei_isMuchSmallerThan(a, b, test_precision()); } -inline bool test_ei_isApproxOrLessThan(const int& a, const int& b) -{ return ei_isApproxOrLessThan(a, b, test_precision()); } - -inline bool test_ei_isApprox(const float& a, const float& b) -{ return ei_isApprox(a, b, test_precision()); } -inline bool test_ei_isMuchSmallerThan(const float& a, const float& b) -{ return ei_isMuchSmallerThan(a, b, test_precision()); } -inline bool test_ei_isApproxOrLessThan(const float& a, const float& b) -{ return ei_isApproxOrLessThan(a, b, test_precision()); } - -inline bool test_ei_isApprox(const double& a, const double& b) -{ return ei_isApprox(a, b, test_precision()); } -inline bool test_ei_isMuchSmallerThan(const double& a, const double& b) -{ return ei_isMuchSmallerThan(a, b, test_precision()); } -inline bool test_ei_isApproxOrLessThan(const double& a, const double& b) -{ return ei_isApproxOrLessThan(a, b, test_precision()); } - -inline bool test_ei_isApprox(const std::complex& a, const std::complex& b) -{ return ei_isApprox(a, b, test_precision >()); } -inline bool test_ei_isMuchSmallerThan(const std::complex& a, const std::complex& b) -{ return ei_isMuchSmallerThan(a, b, test_precision >()); } - -inline bool test_ei_isApprox(const std::complex& a, const std::complex& b) -{ return ei_isApprox(a, b, test_precision >()); } -inline bool test_ei_isMuchSmallerThan(const std::complex& a, const std::complex& b) -{ return ei_isMuchSmallerThan(a, b, test_precision >()); } - -inline bool test_ei_isApprox(const long double& a, const long double& b) -{ return ei_isApprox(a, b, test_precision()); } -inline bool test_ei_isMuchSmallerThan(const long double& a, const long double& b) -{ return ei_isMuchSmallerThan(a, b, test_precision()); } -inline bool test_ei_isApproxOrLessThan(const long double& a, const long double& b) -{ return ei_isApproxOrLessThan(a, b, test_precision()); } - -template -inline bool test_ei_isApprox(const Type1& a, const Type2& b) -{ - return a.isApprox(b, test_precision()); -} - -template -inline bool test_ei_isMuchSmallerThan(const MatrixBase& m1, - const MatrixBase& m2) -{ - return m1.isMuchSmallerThan(m2, test_precision::Scalar>()); -} - -template -inline bool test_ei_isMuchSmallerThan(const MatrixBase& m, - const typename NumTraits::Scalar>::Real& s) -{ - return m.isMuchSmallerThan(s, test_precision::Scalar>()); -} - -} // end namespace Eigen - -template struct GetDifferentType; - -template<> struct GetDifferentType { typedef double type; }; -template<> struct GetDifferentType { typedef float type; }; -template struct GetDifferentType > -{ typedef std::complex::type> type; }; - -// forward declaration of the main test function -void EI_PP_CAT(test_,EIGEN_TEST_FUNC)(); - -using namespace Eigen; - -#ifdef EIGEN_TEST_PART_1 -#define CALL_SUBTEST_1(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_1(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_2 -#define CALL_SUBTEST_2(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_2(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_3 -#define CALL_SUBTEST_3(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_3(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_4 -#define CALL_SUBTEST_4(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_4(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_5 -#define CALL_SUBTEST_5(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_5(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_6 -#define CALL_SUBTEST_6(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_6(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_7 -#define CALL_SUBTEST_7(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_7(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_8 -#define CALL_SUBTEST_8(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_8(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_9 -#define CALL_SUBTEST_9(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_9(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_10 -#define CALL_SUBTEST_10(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_10(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_11 -#define CALL_SUBTEST_11(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_11(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_12 -#define CALL_SUBTEST_12(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_12(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_13 -#define CALL_SUBTEST_13(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_13(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_14 -#define CALL_SUBTEST_14(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_14(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_15 -#define CALL_SUBTEST_15(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_15(FUNC) -#endif - -#ifdef EIGEN_TEST_PART_16 -#define CALL_SUBTEST_16(FUNC) CALL_SUBTEST(FUNC) -#else -#define CALL_SUBTEST_16(FUNC) -#endif - - - -int main(int argc, char *argv[]) -{ - bool has_set_repeat = false; - bool has_set_seed = false; - bool need_help = false; - unsigned int seed = 0; - int repeat = DEFAULT_REPEAT; - - for(int i = 1; i < argc; i++) - { - if(argv[i][0] == 'r') - { - if(has_set_repeat) - { - std::cout << "Argument " << argv[i] << " conflicting with a former argument" << std::endl; - return 1; - } - repeat = std::atoi(argv[i]+1); - has_set_repeat = true; - if(repeat <= 0) - { - std::cout << "Invalid \'repeat\' value " << argv[i]+1 << std::endl; - return 1; - } - } - else if(argv[i][0] == 's') - { - if(has_set_seed) - { - std::cout << "Argument " << argv[i] << " conflicting with a former argument" << std::endl; - return 1; - } - seed = int(std::strtoul(argv[i]+1, 0, 10)); - has_set_seed = true; - bool ok = seed!=0; - if(!ok) - { - std::cout << "Invalid \'seed\' value " << argv[i]+1 << std::endl; - return 1; - } - } - else - { - need_help = true; - } - } - - if(need_help) - { - std::cout << "This test application takes the following optional arguments:" << std::endl; - std::cout << " rN Repeat each test N times (default: " << DEFAULT_REPEAT << ")" << std::endl; - std::cout << " sN Use N as seed for random numbers (default: based on current time)" << std::endl; - return 1; - } - - if(!has_set_seed) seed = (unsigned int) std::time(NULL); - if(!has_set_repeat) repeat = DEFAULT_REPEAT; - - std::cout << "Initializing random number generator with seed " << seed << std::endl; - std::srand(seed); - std::cout << "Repeating each test " << repeat << " times" << std::endl; - - Eigen::g_repeat = repeat; - Eigen::g_test_stack.push_back(EI_PP_MAKE_STRING(EIGEN_TEST_FUNC)); - - EI_PP_CAT(test_,EIGEN_TEST_FUNC)(); - return 0; -} - - - diff --git a/thirdparty/eigen/test/eigen2/product.h b/thirdparty/eigen/test/eigen2/product.h deleted file mode 100644 index ae1b4bae..00000000 --- a/thirdparty/eigen/test/eigen2/product.h +++ /dev/null @@ -1,129 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2006-2008 Benoit Jacob -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#include "main.h" -#include -#include - -template -bool areNotApprox(const MatrixBase& m1, const MatrixBase& m2, typename Derived1::RealScalar epsilon = precision()) -{ - return !((m1-m2).cwise().abs2().maxCoeff() < epsilon * epsilon - * std::max(m1.cwise().abs2().maxCoeff(), m2.cwise().abs2().maxCoeff())); -} - -template void product(const MatrixType& m) -{ - /* this test covers the following files: - Identity.h Product.h - */ - - typedef typename MatrixType::Scalar Scalar; - typedef typename NumTraits::FloatingPoint FloatingPoint; - typedef Matrix RowVectorType; - typedef Matrix ColVectorType; - typedef Matrix RowSquareMatrixType; - typedef Matrix ColSquareMatrixType; - typedef Matrix OtherMajorMatrixType; - - int rows = m.rows(); - int cols = m.cols(); - - // this test relies a lot on Random.h, and there's not much more that we can do - // to test it, hence I consider that we will have tested Random.h - MatrixType m1 = MatrixType::Random(rows, cols), - m2 = MatrixType::Random(rows, cols), - m3(rows, cols); - RowSquareMatrixType - identity = RowSquareMatrixType::Identity(rows, rows), - square = RowSquareMatrixType::Random(rows, rows), - res = RowSquareMatrixType::Random(rows, rows); - ColSquareMatrixType - square2 = ColSquareMatrixType::Random(cols, cols), - res2 = ColSquareMatrixType::Random(cols, cols); - RowVectorType v1 = RowVectorType::Random(rows); - ColVectorType vc2 = ColVectorType::Random(cols), vcres(cols); - OtherMajorMatrixType tm1 = m1; - - Scalar s1 = ei_random(); - - int r = ei_random(0, rows-1), - c = ei_random(0, cols-1); - - // begin testing Product.h: only associativity for now - // (we use Transpose.h but this doesn't count as a test for it) - - VERIFY_IS_APPROX((m1*m1.transpose())*m2, m1*(m1.transpose()*m2)); - m3 = m1; - m3 *= m1.transpose() * m2; - VERIFY_IS_APPROX(m3, m1 * (m1.transpose()*m2)); - VERIFY_IS_APPROX(m3, m1.lazy() * (m1.transpose()*m2)); - - // continue testing Product.h: distributivity - VERIFY_IS_APPROX(square*(m1 + m2), square*m1+square*m2); - VERIFY_IS_APPROX(square*(m1 - m2), square*m1-square*m2); - - // continue testing Product.h: compatibility with ScalarMultiple.h - VERIFY_IS_APPROX(s1*(square*m1), (s1*square)*m1); - VERIFY_IS_APPROX(s1*(square*m1), square*(m1*s1)); - - // again, test operator() to check const-qualification - s1 += (square.lazy() * m1)(r,c); - - // test Product.h together with Identity.h - VERIFY_IS_APPROX(v1, identity*v1); - VERIFY_IS_APPROX(v1.transpose(), v1.transpose() * identity); - // again, test operator() to check const-qualification - VERIFY_IS_APPROX(MatrixType::Identity(rows, cols)(r,c), static_cast(r==c)); - - if (rows!=cols) - VERIFY_RAISES_ASSERT(m3 = m1*m1); - - // test the previous tests were not screwed up because operator* returns 0 - // (we use the more accurate default epsilon) - if (NumTraits::HasFloatingPoint && std::min(rows,cols)>1) - { - VERIFY(areNotApprox(m1.transpose()*m2,m2.transpose()*m1)); - } - - // test optimized operator+= path - res = square; - res += (m1 * m2.transpose()).lazy(); - VERIFY_IS_APPROX(res, square + m1 * m2.transpose()); - if (NumTraits::HasFloatingPoint && std::min(rows,cols)>1) - { - VERIFY(areNotApprox(res,square + m2 * m1.transpose())); - } - vcres = vc2; - vcres += (m1.transpose() * v1).lazy(); - VERIFY_IS_APPROX(vcres, vc2 + m1.transpose() * v1); - tm1 = m1; - VERIFY_IS_APPROX(tm1.transpose() * v1, m1.transpose() * v1); - VERIFY_IS_APPROX(v1.transpose() * tm1, v1.transpose() * m1); - - // test submatrix and matrix/vector product - for (int i=0; i::HasFloatingPoint && std::min(rows,cols)>1) - { - VERIFY(areNotApprox(res2,square2 + m2.transpose() * m1)); - } -} - diff --git a/thirdparty/eigen/test/eigen2/runtest.sh b/thirdparty/eigen/test/eigen2/runtest.sh deleted file mode 100755 index bc693af1..00000000 --- a/thirdparty/eigen/test/eigen2/runtest.sh +++ /dev/null @@ -1,28 +0,0 @@ -#!/bin/bash - -black='\E[30m' -red='\E[31m' -green='\E[32m' -yellow='\E[33m' -blue='\E[34m' -magenta='\E[35m' -cyan='\E[36m' -white='\E[37m' - -if make test_$1 > /dev/null 2> .runtest.log ; then - if ! ./test_$1 r20 > /dev/null 2> .runtest.log ; then - echo -e $red Test $1 failed: $black - echo -e $blue - cat .runtest.log - echo -e $black - exit 1 - else - echo -e $green Test $1 passed$black - fi -else - echo -e $red Build of target $1 failed: $black - echo -e $blue - cat .runtest.log - echo -e $black - exit 1 -fi diff --git a/thirdparty/eigen/test/eigen2/sparse.h b/thirdparty/eigen/test/eigen2/sparse.h deleted file mode 100644 index e12f8999..00000000 --- a/thirdparty/eigen/test/eigen2/sparse.h +++ /dev/null @@ -1,154 +0,0 @@ -// This file is part of Eigen, a lightweight C++ template library -// for linear algebra. Eigen itself is part of the KDE project. -// -// Copyright (C) 2008 Daniel Gomez Ferro -// -// This Source Code Form is subject to the terms of the Mozilla -// Public License v. 2.0. If a copy of the MPL was not distributed -// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. - -#ifndef EIGEN_TESTSPARSE_H - -#include "main.h" - -#if EIGEN_GNUC_AT_LEAST(4,0) && !defined __ICC -#include -#define EIGEN_UNORDERED_MAP_SUPPORT -namespace std { - using std::tr1::unordered_map; -} -#endif - -#ifdef EIGEN_GOOGLEHASH_SUPPORT - #include -#endif - -#include -#include -#include - -enum { - ForceNonZeroDiag = 1, - MakeLowerTriangular = 2, - MakeUpperTriangular = 4, - ForceRealDiag = 8 -}; - -/* Initializes both a sparse and dense matrix with same random values, - * and a ratio of \a density non zero entries. - * \param flags is a union of ForceNonZeroDiag, MakeLowerTriangular and MakeUpperTriangular - * allowing to control the shape of the matrix. - * \param zeroCoords and nonzeroCoords allows to get the coordinate lists of the non zero, - * and zero coefficients respectively. - */ -template void -initSparse(double density, - Matrix& refMat, - SparseMatrix& sparseMat, - int flags = 0, - std::vector* zeroCoords = 0, - std::vector* nonzeroCoords = 0) -{ - sparseMat.startFill(int(refMat.rows()*refMat.cols()*density)); - for(int j=0; j(0,1) < density) ? ei_random() : Scalar(0); - if ((flags&ForceNonZeroDiag) && (i==j)) - { - v = ei_random()*Scalar(3.); - v = v*v + Scalar(5.); - } - if ((flags & MakeLowerTriangular) && j>i) - v = Scalar(0); - else if ((flags & MakeUpperTriangular) && jpush_back(Vector2i(i,j)); - } - else if (zeroCoords) - { - zeroCoords->push_back(Vector2i(i,j)); - } - refMat(i,j) = v; - } - } - sparseMat.endFill(); -} - -template void -initSparse(double density, - Matrix& refMat, - DynamicSparseMatrix& sparseMat, - int flags = 0, - std::vector* zeroCoords = 0, - std::vector* nonzeroCoords = 0) -{ - sparseMat.startFill(int(refMat.rows()*refMat.cols()*density)); - for(int j=0; j(0,1) < density) ? ei_random() : Scalar(0); - if ((flags&ForceNonZeroDiag) && (i==j)) - { - v = ei_random()*Scalar(3.); - v = v*v + Scalar(5.); - } - if ((flags & MakeLowerTriangular) && j>i) - v = Scalar(0); - else if ((flags & MakeUpperTriangular) && jpush_back(Vector2i(i,j)); - } - else if (zeroCoords) - { - zeroCoords->push_back(Vector2i(i,j)); - } - refMat(i,j) = v; - } - } - sparseMat.endFill(); -} - -template void -initSparse(double density, - Matrix& refVec, - SparseVector& sparseVec, - std::vector* zeroCoords = 0, - std::vector* nonzeroCoords = 0) -{ - sparseVec.reserve(int(refVec.size()*density)); - sparseVec.setZero(); - for(int i=0; i(0,1) < density) ? ei_random() : Scalar(0); - if (v!=Scalar(0)) - { - sparseVec.fill(i) = v; - if (nonzeroCoords) - nonzeroCoords->push_back(i); - } - else if (zeroCoords) - zeroCoords->push_back(i); - refVec[i] = v; - } -} - -#endif // EIGEN_TESTSPARSE_H diff --git a/thirdparty/eigen/test/eigen2/testsuite.cmake b/thirdparty/eigen/test/eigen2/testsuite.cmake deleted file mode 100644 index 12b6bfa2..00000000 --- a/thirdparty/eigen/test/eigen2/testsuite.cmake +++ /dev/null @@ -1,197 +0,0 @@ - -#################################################################### -# -# Usage: -# - create a new folder, let's call it cdash -# - in that folder, do: -# ctest -S path/to/eigen2/test/testsuite.cmake[,option1=value1[,option2=value2]] -# -# Options: -# - EIGEN_CXX: compiler, eg.: g++-4.2 -# default: default c++ compiler -# - EIGEN_SITE: eg, INRIA-Bdx_pc-gael, or the name of the contributor, etc. -# default: hostname -# - EIGEN_BUILD_STRING: a string which identify the system/compiler. It should be formed like that: -# --- -# with: -# = opensuse, debian, osx, windows, cygwin, freebsd, solaris, etc. -# = 11.1, XP, vista, leopard, etc. -# = i386, x86_64, ia64, powerpc, etc. -# = gcc-4.3.2, icc-11.0, MSVC-2008, etc. -# - EIGEN_EXPLICIT_VECTORIZATION: novec, SSE2, Altivec -# default: SSE2 for x86_64 systems, novec otherwise -# Its value is automatically appended to EIGEN_BUILD_STRING -# - EIGEN_CMAKE_DIR: path to cmake executable -# - EIGEN_MODE: dashboard model, can be Experimental, Nightly, or Continuous -# default: Nightly -# - EIGEN_WORK_DIR: directory used to download the source files and make the builds -# default: folder which contains this script -# - EIGEN_CMAKE_ARGS: additional arguments passed to cmake -# - CTEST_SOURCE_DIRECTORY: path to eigen's src (use a new and empty folder, not the one you are working on) -# default: /src -# - CTEST_BINARY_DIRECTORY: build directory -# default: /nightly- -# -# Here is an example running several compilers on a linux system: -# #!/bin/bash -# ARCH=`uname -m` -# SITE=`hostname` -# VERSION=opensuse-11.1 -# WORK_DIR=/home/gael/Coding/eigen2/cdash -# # get the last version of the script -# wget http://bitbucket.org/eigen/eigen/raw/tip/test/testsuite.cmake -o $WORK_DIR/testsuite.cmake -# COMMON="ctest -S $WORK_DIR/testsuite.cmake,EIGEN_WORK_DIR=$WORK_DIR,EIGEN_SITE=$SITE,EIGEN_MODE=$1,EIGEN_BUILD_STRING=$OS_VERSION-$ARCH" -# $COMMON-gcc-3.4.6,EIGEN_CXX=g++-3.4 -# $COMMON-gcc-4.0.1,EIGEN_CXX=g++-4.0.1 -# $COMMON-gcc-4.3.2,EIGEN_CXX=g++-4.3,EIGEN_EXPLICIT_VECTORIZATION=novec -# $COMMON-gcc-4.3.2,EIGEN_CXX=g++-4.3,EIGEN_EXPLICIT_VECTORIZATION=SSE2 -# $COMMON-icc-11.0,EIGEN_CXX=icpc -# -#################################################################### - -# process the arguments - -set(ARGLIST ${CTEST_SCRIPT_ARG}) -while(${ARGLIST} MATCHES ".+.*") - - # pick first - string(REGEX MATCH "([^,]*)(,.*)?" DUMMY ${ARGLIST}) - SET(TOP ${CMAKE_MATCH_1}) - - # remove first - string(REGEX MATCHALL "[^,]*,(.*)" DUMMY ${ARGLIST}) - SET(ARGLIST ${CMAKE_MATCH_1}) - - # decompose as a pair key=value - string(REGEX MATCH "([^=]*)(=.*)?" DUMMY ${TOP}) - SET(KEY ${CMAKE_MATCH_1}) - - string(REGEX MATCH "[^=]*=(.*)" DUMMY ${TOP}) - SET(VALUE ${CMAKE_MATCH_1}) - - # set the variable to the specified value - if(VALUE) - SET(${KEY} ${VALUE}) - else(VALUE) - SET(${KEY} ON) - endif(VALUE) - -endwhile(${ARGLIST} MATCHES ".+.*") - -#################################################################### -# Automatically set some user variables if they have not been defined manually -#################################################################### -cmake_minimum_required(VERSION 2.6 FATAL_ERROR) - -if(NOT EIGEN_SITE) - site_name(EIGEN_SITE) -endif(NOT EIGEN_SITE) - -if(NOT EIGEN_CMAKE_DIR) - SET(EIGEN_CMAKE_DIR "") -endif(NOT EIGEN_CMAKE_DIR) - -if(NOT EIGEN_BUILD_STRING) - - # let's try to find all information we need to make the build string ourself - - # OS - build_name(EIGEN_OS_VERSION) - - # arch - set(EIGEN_ARCH ${CMAKE_SYSTEM_PROCESSOR}) - if(WIN32) - set(EIGEN_ARCH $ENV{PROCESSOR_ARCHITECTURE}) - else(WIN32) - execute_process(COMMAND uname -m OUTPUT_VARIABLE EIGEN_ARCH OUTPUT_STRIP_TRAILING_WHITESPACE) - endif(WIN32) - - set(EIGEN_BUILD_STRING ${EIGEN_OS_VERSION}${EIGEN_ARCH}-${EIGEN_CXX}) - -endif(NOT EIGEN_BUILD_STRING) - -if(DEFINED EIGEN_EXPLICIT_VECTORIZATION) - set(EIGEN_BUILD_STRING ${EIGEN_BUILD_STRING}-${EIGEN_EXPLICIT_VECTORIZATION}) -endif(DEFINED EIGEN_EXPLICIT_VECTORIZATION) - -if(NOT EIGEN_WORK_DIR) - set(EIGEN_WORK_DIR ${CTEST_SCRIPT_DIRECTORY}) -endif(NOT EIGEN_WORK_DIR) - -if(NOT CTEST_SOURCE_DIRECTORY) - SET (CTEST_SOURCE_DIRECTORY "${EIGEN_WORK_DIR}/src") -endif(NOT CTEST_SOURCE_DIRECTORY) - -if(NOT CTEST_BINARY_DIRECTORY) - SET (CTEST_BINARY_DIRECTORY "${EIGEN_WORK_DIR}/nightly_${EIGEN_CXX}") -endif(NOT CTEST_BINARY_DIRECTORY) - -if(NOT EIGEN_MODE) - set(EIGEN_MODE Nightly) -endif(NOT EIGEN_MODE) - -## mandatory variables (the default should be ok in most cases): - -SET (CTEST_CVS_COMMAND "hg") -SET (CTEST_CVS_CHECKOUT "${CTEST_CVS_COMMAND} clone -r 2.0 http://bitbucket.org/eigen/eigen \"${CTEST_SOURCE_DIRECTORY}\"") - -# which ctest command to use for running the dashboard -SET (CTEST_COMMAND "${EIGEN_CMAKE_DIR}ctest -D ${EIGEN_MODE}") - -# what cmake command to use for configuring this dashboard -SET (CTEST_CMAKE_COMMAND "${EIGEN_CMAKE_DIR}cmake -DEIGEN_BUILD_TESTS=on ") - -#################################################################### -# The values in this section are optional you can either -# have them or leave them commented out -#################################################################### - -# this make sure we get consistent outputs -SET($ENV{LC_MESSAGES} "en_EN") - -# should ctest wipe the binary tree before running -SET(CTEST_START_WITH_EMPTY_BINARY_DIRECTORY TRUE) -SET(CTEST_BACKUP_AND_RESTORE TRUE) - -# this is the initial cache to use for the binary tree, be careful to escape -# any quotes inside of this string if you use it -if(WIN32 AND NOT UNIX) - #message(SEND_ERROR "win32") - set(CTEST_CMAKE_COMMAND "${CTEST_CMAKE_COMMAND} -G \"NMake Makefiles\" -DCMAKE_MAKE_PROGRAM=nmake") - SET (CTEST_INITIAL_CACHE " - MAKECOMMAND:STRING=nmake -i - CMAKE_MAKE_PROGRAM:FILEPATH=nmake - CMAKE_GENERATOR:INTERNAL=NMake Makefiles - BUILDNAME:STRING=${EIGEN_BUILD_STRING} - SITE:STRING=${EIGEN_SITE} - ") -else(WIN32 AND NOT UNIX) - SET (CTEST_INITIAL_CACHE " - BUILDNAME:STRING=${EIGEN_BUILD_STRING} - SITE:STRING=${EIGEN_SITE} - ") -endif(WIN32 AND NOT UNIX) - -# set any extra environment variables to use during the execution of the script here: - -if(EIGEN_CXX) - set(CTEST_ENVIRONMENT "CXX=${EIGEN_CXX}") -endif(EIGEN_CXX) - -if(DEFINED EIGEN_EXPLICIT_VECTORIZATION) - if(EIGEN_EXPLICIT_VECTORIZATION MATCHES SSE2) - set(CTEST_CMAKE_COMMAND "${CTEST_CMAKE_COMMAND} -DEIGEN_TEST_SSE2=ON") - elseif(EIGEN_EXPLICIT_VECTORIZATION MATCHES SSE3) - set(CTEST_CMAKE_COMMAND "${CTEST_CMAKE_COMMAND} -DEIGEN_TEST_SSE2=ON -DEIGEN_TEST_SSE3=ON") - elseif(EIGEN_EXPLICIT_VECTORIZATION MATCHES Altivec) - set(CTEST_CMAKE_COMMAND "${CTEST_CMAKE_COMMAND} -DEIGEN_TEST_ALTIVEC=ON") - elseif(EIGEN_EXPLICIT_VECTORIZATION MATCHES novec) - set(CTEST_CMAKE_COMMAND "${CTEST_CMAKE_COMMAND} -DEIGEN_TEST_NO_EXPLICIT_VECTORIZATION=ON") - else(EIGEN_EXPLICIT_VECTORIZATION MATCHES SSE2) - message(FATAL_ERROR "Invalid value for EIGEN_EXPLICIT_VECTORIZATION (${EIGEN_EXPLICIT_VECTORIZATION}), must be: novec, SSE2, SSE3, Altivec") - endif(EIGEN_EXPLICIT_VECTORIZATION MATCHES SSE2) -endif(DEFINED EIGEN_EXPLICIT_VECTORIZATION) - -if(DEFINED EIGEN_CMAKE_ARGS) - set(CTEST_CMAKE_COMMAND "${CTEST_CMAKE_COMMAND} ${EIGEN_CMAKE_ARGS}") -endif(DEFINED EIGEN_CMAKE_ARGS) diff --git a/thirdparty/eigen/test/eigen2support.cpp b/thirdparty/eigen/test/eigen2support.cpp index 1fa49a8c..49d7328e 100644 --- a/thirdparty/eigen/test/eigen2support.cpp +++ b/thirdparty/eigen/test/eigen2support.cpp @@ -8,13 +8,11 @@ // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. #define EIGEN2_SUPPORT -#define EIGEN_NO_EIGEN2_DEPRECATED_WARNING #include "main.h" template void eigen2support(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; Index rows = m.rows(); @@ -54,7 +52,7 @@ template void eigen2support(const MatrixType& m) m1.minor(0,0); } -void test_eigen2support() +EIGEN_DECLARE_TEST(eigen2support) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( eigen2support(Matrix()) ); diff --git a/thirdparty/eigen/test/eigensolver_complex.cpp b/thirdparty/eigen/test/eigensolver_complex.cpp index c9d8c087..c5373f42 100644 --- a/thirdparty/eigen/test/eigensolver_complex.cpp +++ b/thirdparty/eigen/test/eigensolver_complex.cpp @@ -13,26 +13,64 @@ #include #include -/* Check that two column vectors are approximately equal upto permutations, - by checking that the k-th power sums are equal for k = 1, ..., vec1.rows() */ +template bool find_pivot(typename MatrixType::Scalar tol, MatrixType &diffs, Index col=0) +{ + bool match = diffs.diagonal().sum() <= tol; + if(match || col==diffs.cols()) + { + return match; + } + else + { + Index n = diffs.cols(); + std::vector > transpositions; + for(Index i=col; i tol) + break; + + best_index += col; + + diffs.row(col).swap(diffs.row(best_index)); + if(find_pivot(tol,diffs,col+1)) return true; + diffs.row(col).swap(diffs.row(best_index)); + + // move current pivot to the end + diffs.row(n-(i-col)-1).swap(diffs.row(best_index)); + transpositions.push_back(std::pair(n-(i-col)-1,best_index)); + } + // restore + for(Index k=transpositions.size()-1; k>=0; --k) + diffs.row(transpositions[k].first).swap(diffs.row(transpositions[k].second)); + } + return false; +} + +/* Check that two column vectors are approximately equal up to permutations. + * Initially, this method checked that the k-th power sums are equal for all k = 1, ..., vec1.rows(), + * however this strategy is numerically inacurate because of numerical cancellation issues. + */ template void verify_is_approx_upto_permutation(const VectorType& vec1, const VectorType& vec2) { - typedef typename NumTraits::Real RealScalar; + typedef typename VectorType::Scalar Scalar; + typedef typename NumTraits::Real RealScalar; VERIFY(vec1.cols() == 1); VERIFY(vec2.cols() == 1); VERIFY(vec1.rows() == vec2.rows()); - for (int k = 1; k <= vec1.rows(); ++k) - { - VERIFY_IS_APPROX(vec1.array().pow(RealScalar(k)).sum(), vec2.array().pow(RealScalar(k)).sum()); - } + + Index n = vec1.rows(); + RealScalar tol = test_precision()*test_precision()*numext::maxi(vec1.squaredNorm(),vec2.squaredNorm()); + Matrix diffs = (vec1.rowwise().replicate(n) - vec2.rowwise().replicate(n).transpose()).cwiseAbs2(); + + VERIFY( find_pivot(tol, diffs) ); } template void eigensolver(const MatrixType& m) { - typedef typename MatrixType::Index Index; /* this test covers the following files: ComplexEigenSolver.h, and indirectly ComplexSchur.h */ @@ -79,13 +117,28 @@ template void eigensolver(const MatrixType& m) MatrixType id = MatrixType::Identity(rows, cols); VERIFY_IS_APPROX(id.operatorNorm(), RealScalar(1)); - if (rows > 1) + if (rows > 1 && rows < 20) { // Test matrix with NaN a(0,0) = std::numeric_limits::quiet_NaN(); ComplexEigenSolver eiNaN(a); VERIFY_IS_EQUAL(eiNaN.info(), NoConvergence); } + + // regression test for bug 1098 + { + ComplexEigenSolver eig(a.adjoint() * a); + eig.compute(a.adjoint() * a); + } + + // regression test for bug 478 + { + a.setZero(); + ComplexEigenSolver ei3(a); + VERIFY_IS_EQUAL(ei3.info(), Success); + VERIFY_IS_MUCH_SMALLER_THAN(ei3.eigenvalues().norm(),RealScalar(1)); + VERIFY((ei3.eigenvectors().transpose()*ei3.eigenvectors().transpose()).eval().isIdentity()); + } } template void eigensolver_verify_assert(const MatrixType& m) @@ -99,7 +152,7 @@ template void eigensolver_verify_assert(const MatrixType& m VERIFY_RAISES_ASSERT(eig.eigenvectors()); } -void test_eigensolver_complex() +EIGEN_DECLARE_TEST(eigensolver_complex) { int s = 0; for(int i = 0; i < g_repeat; i++) { @@ -108,6 +161,7 @@ void test_eigensolver_complex() CALL_SUBTEST_2( eigensolver(MatrixXcd(s,s)) ); CALL_SUBTEST_3( eigensolver(Matrix, 1, 1>()) ); CALL_SUBTEST_4( eigensolver(Matrix3f()) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) } CALL_SUBTEST_1( eigensolver_verify_assert(Matrix4cf()) ); s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); diff --git a/thirdparty/eigen/test/eigensolver_generalized_real.cpp b/thirdparty/eigen/test/eigensolver_generalized_real.cpp index 566a4bdc..a0c99b18 100644 --- a/thirdparty/eigen/test/eigensolver_generalized_real.cpp +++ b/thirdparty/eigen/test/eigensolver_generalized_real.cpp @@ -1,19 +1,20 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2012 Gael Guennebaud +// Copyright (C) 2012-2016 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +#define EIGEN_RUNTIME_NO_MALLOC #include "main.h" #include #include +#include template void generalized_eigensolver_real(const MatrixType& m) { - typedef typename MatrixType::Index Index; /* this test covers the following files: GeneralizedEigenSolver.h */ @@ -21,6 +22,7 @@ template void generalized_eigensolver_real(const MatrixType Index cols = m.cols(); typedef typename MatrixType::Scalar Scalar; + typedef std::complex ComplexScalar; typedef Matrix VectorType; MatrixType a = MatrixType::Random(rows,cols); @@ -31,17 +33,95 @@ template void generalized_eigensolver_real(const MatrixType MatrixType spdB = b.adjoint() * b + b1.adjoint() * b1; // lets compare to GeneralizedSelfAdjointEigenSolver - GeneralizedSelfAdjointEigenSolver symmEig(spdA, spdB); - GeneralizedEigenSolver eig(spdA, spdB); + { + GeneralizedSelfAdjointEigenSolver symmEig(spdA, spdB); + GeneralizedEigenSolver eig(spdA, spdB); - VERIFY_IS_EQUAL(eig.eigenvalues().imag().cwiseAbs().maxCoeff(), 0); + VERIFY_IS_EQUAL(eig.eigenvalues().imag().cwiseAbs().maxCoeff(), 0); - VectorType realEigenvalues = eig.eigenvalues().real(); - std::sort(realEigenvalues.data(), realEigenvalues.data()+realEigenvalues.size()); - VERIFY_IS_APPROX(realEigenvalues, symmEig.eigenvalues()); + VectorType realEigenvalues = eig.eigenvalues().real(); + std::sort(realEigenvalues.data(), realEigenvalues.data()+realEigenvalues.size()); + VERIFY_IS_APPROX(realEigenvalues, symmEig.eigenvalues()); + + // check eigenvectors + typename GeneralizedEigenSolver::EigenvectorsType D = eig.eigenvalues().asDiagonal(); + typename GeneralizedEigenSolver::EigenvectorsType V = eig.eigenvectors(); + VERIFY_IS_APPROX(spdA*V, spdB*V*D); + } + + // non symmetric case: + { + GeneralizedEigenSolver eig(rows); + // TODO enable full-prealocation of required memory, this probably requires an in-place mode for HessenbergDecomposition + //Eigen::internal::set_is_malloc_allowed(false); + eig.compute(a,b); + //Eigen::internal::set_is_malloc_allowed(true); + for(Index k=0; k tmp = (eig.betas()(k)*a).template cast() - eig.alphas()(k)*b; + if(tmp.size()>1 && tmp.norm()>(std::numeric_limits::min)()) + tmp /= tmp.norm(); + VERIFY_IS_MUCH_SMALLER_THAN( std::abs(tmp.determinant()), Scalar(1) ); + } + // check eigenvectors + typename GeneralizedEigenSolver::EigenvectorsType D = eig.eigenvalues().asDiagonal(); + typename GeneralizedEigenSolver::EigenvectorsType V = eig.eigenvectors(); + VERIFY_IS_APPROX(a*V, b*V*D); + } + + // regression test for bug 1098 + { + GeneralizedSelfAdjointEigenSolver eig1(a.adjoint() * a,b.adjoint() * b); + eig1.compute(a.adjoint() * a,b.adjoint() * b); + GeneralizedEigenSolver eig2(a.adjoint() * a,b.adjoint() * b); + eig2.compute(a.adjoint() * a,b.adjoint() * b); + } + + // check without eigenvectors + { + GeneralizedEigenSolver eig1(spdA, spdB, true); + GeneralizedEigenSolver eig2(spdA, spdB, false); + VERIFY_IS_APPROX(eig1.eigenvalues(), eig2.eigenvalues()); + } +} + +template +void generalized_eigensolver_assert() { + GeneralizedEigenSolver eig; + // all raise assert if uninitialized + VERIFY_RAISES_ASSERT(eig.info()); + VERIFY_RAISES_ASSERT(eig.eigenvectors()); + VERIFY_RAISES_ASSERT(eig.eigenvalues()); + VERIFY_RAISES_ASSERT(eig.alphas()); + VERIFY_RAISES_ASSERT(eig.betas()); + + // none raise assert after compute called + eig.compute(MatrixType::Random(20, 20), MatrixType::Random(20, 20)); + VERIFY(eig.info() == Success); + eig.eigenvectors(); + eig.eigenvalues(); + eig.alphas(); + eig.betas(); + + // eigenvectors() raises assert, if eigenvectors were not requested + eig.compute(MatrixType::Random(20, 20), MatrixType::Random(20, 20), false); + VERIFY(eig.info() == Success); + VERIFY_RAISES_ASSERT(eig.eigenvectors()); + eig.eigenvalues(); + eig.alphas(); + eig.betas(); + + // all except info raise assert if realQZ did not converge + eig.setMaxIterations(0); // force real QZ to fail. + eig.compute(MatrixType::Random(20, 20), MatrixType::Random(20, 20)); + VERIFY(eig.info() == NoConvergence); + VERIFY_RAISES_ASSERT(eig.eigenvectors()); + VERIFY_RAISES_ASSERT(eig.eigenvalues()); + VERIFY_RAISES_ASSERT(eig.alphas()); + VERIFY_RAISES_ASSERT(eig.betas()); } -void test_eigensolver_generalized_real() +EIGEN_DECLARE_TEST(eigensolver_generalized_real) { for(int i = 0; i < g_repeat; i++) { int s = 0; @@ -49,11 +129,12 @@ void test_eigensolver_generalized_real() s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); CALL_SUBTEST_2( generalized_eigensolver_real(MatrixXd(s,s)) ); - // some trivial but implementation-wise tricky cases + // some trivial but implementation-wise special cases CALL_SUBTEST_2( generalized_eigensolver_real(MatrixXd(1,1)) ); CALL_SUBTEST_2( generalized_eigensolver_real(MatrixXd(2,2)) ); CALL_SUBTEST_3( generalized_eigensolver_real(Matrix()) ); CALL_SUBTEST_4( generalized_eigensolver_real(Matrix2d()) ); + CALL_SUBTEST_5( generalized_eigensolver_assert() ); TEST_SET_BUT_UNUSED_VARIABLE(s) } } diff --git a/thirdparty/eigen/test/eigensolver_generic.cpp b/thirdparty/eigen/test/eigensolver_generic.cpp index 005af81e..7adb9866 100644 --- a/thirdparty/eigen/test/eigensolver_generic.cpp +++ b/thirdparty/eigen/test/eigensolver_generic.cpp @@ -12,9 +12,23 @@ #include #include +template +void check_eigensolver_for_given_mat(const EigType &eig, const MatType& a) +{ + typedef typename NumTraits::Real RealScalar; + typedef Matrix RealVectorType; + typedef typename std::complex Complex; + Index n = a.rows(); + VERIFY_IS_EQUAL(eig.info(), Success); + VERIFY_IS_APPROX(a * eig.pseudoEigenvectors(), eig.pseudoEigenvectors() * eig.pseudoEigenvalueMatrix()); + VERIFY_IS_APPROX(a.template cast() * eig.eigenvectors(), + eig.eigenvectors() * eig.eigenvalues().asDiagonal()); + VERIFY_IS_APPROX(eig.eigenvectors().colwise().norm(), RealVectorType::Ones(n).transpose()); + VERIFY_IS_APPROX(a.eigenvalues(), eig.eigenvalues()); +} + template void eigensolver(const MatrixType& m) { - typedef typename MatrixType::Index Index; /* this test covers the following files: EigenSolver.h */ @@ -23,8 +37,7 @@ template void eigensolver(const MatrixType& m) typedef typename MatrixType::Scalar Scalar; typedef typename NumTraits::Real RealScalar; - typedef Matrix RealVectorType; - typedef typename std::complex::Real> Complex; + typedef typename std::complex Complex; MatrixType a = MatrixType::Random(rows,cols); MatrixType a1 = MatrixType::Random(rows,cols); @@ -37,12 +50,7 @@ template void eigensolver(const MatrixType& m) (ei0.pseudoEigenvectors().template cast()) * (ei0.eigenvalues().asDiagonal())); EigenSolver ei1(a); - VERIFY_IS_EQUAL(ei1.info(), Success); - VERIFY_IS_APPROX(a * ei1.pseudoEigenvectors(), ei1.pseudoEigenvectors() * ei1.pseudoEigenvalueMatrix()); - VERIFY_IS_APPROX(a.template cast() * ei1.eigenvectors(), - ei1.eigenvectors() * ei1.eigenvalues().asDiagonal()); - VERIFY_IS_APPROX(ei1.eigenvectors().colwise().norm(), RealVectorType::Ones(rows).transpose()); - VERIFY_IS_APPROX(a.eigenvalues(), ei1.eigenvalues()); + CALL_SUBTEST( check_eigensolver_for_given_mat(ei1,a) ); EigenSolver ei2; ei2.setMaxIterations(RealSchur::m_maxIterationsPerRow * rows).compute(a); @@ -63,12 +71,27 @@ template void eigensolver(const MatrixType& m) MatrixType id = MatrixType::Identity(rows, cols); VERIFY_IS_APPROX(id.operatorNorm(), RealScalar(1)); - if (rows > 2) + if (rows > 2 && rows < 20) { // Test matrix with NaN a(0,0) = std::numeric_limits::quiet_NaN(); EigenSolver eiNaN(a); - VERIFY_IS_EQUAL(eiNaN.info(), NoConvergence); + VERIFY_IS_NOT_EQUAL(eiNaN.info(), Success); + } + + // regression test for bug 1098 + { + EigenSolver eig(a.adjoint() * a); + eig.compute(a.adjoint() * a); + } + + // regression test for bug 478 + { + a.setZero(); + EigenSolver ei3(a); + VERIFY_IS_EQUAL(ei3.info(), Success); + VERIFY_IS_MUCH_SMALLER_THAN(ei3.eigenvalues().norm(),RealScalar(1)); + VERIFY((ei3.eigenvectors().transpose()*ei3.eigenvectors().transpose()).eval().isIdentity()); } } @@ -86,13 +109,111 @@ template void eigensolver_verify_assert(const MatrixType& m VERIFY_RAISES_ASSERT(eig.pseudoEigenvectors()); } -void test_eigensolver_generic() + +template +Matrix +make_companion(const CoeffType& coeffs) +{ + Index n = coeffs.size()-1; + Matrix res(n,n); + res.setZero(); + res.row(0) = -coeffs.tail(n) / coeffs(0); + res.diagonal(-1).setOnes(); + return res; +} + +template +void eigensolver_generic_extra() +{ + { + // regression test for bug 793 + MatrixXd a(3,3); + a << 0, 0, 1, + 1, 1, 1, + 1, 1e+200, 1; + Eigen::EigenSolver eig(a); + double scale = 1e-200; // scale to avoid overflow during the comparisons + VERIFY_IS_APPROX(a * eig.pseudoEigenvectors()*scale, eig.pseudoEigenvectors() * eig.pseudoEigenvalueMatrix()*scale); + VERIFY_IS_APPROX(a * eig.eigenvectors()*scale, eig.eigenvectors() * eig.eigenvalues().asDiagonal()*scale); + } + { + // check a case where all eigenvalues are null. + MatrixXd a(2,2); + a << 1, 1, + -1, -1; + Eigen::EigenSolver eig(a); + VERIFY_IS_APPROX(eig.pseudoEigenvectors().squaredNorm(), 2.); + VERIFY_IS_APPROX((a * eig.pseudoEigenvectors()).norm()+1., 1.); + VERIFY_IS_APPROX((eig.pseudoEigenvectors() * eig.pseudoEigenvalueMatrix()).norm()+1., 1.); + VERIFY_IS_APPROX((a * eig.eigenvectors()).norm()+1., 1.); + VERIFY_IS_APPROX((eig.eigenvectors() * eig.eigenvalues().asDiagonal()).norm()+1., 1.); + } + + // regression test for bug 933 + { + { + VectorXd coeffs(5); coeffs << 1, -3, -175, -225, 2250; + MatrixXd C = make_companion(coeffs); + EigenSolver eig(C); + CALL_SUBTEST( check_eigensolver_for_given_mat(eig,C) ); + } + { + // this test is tricky because it requires high accuracy in smallest eigenvalues + VectorXd coeffs(5); coeffs << 6.154671e-15, -1.003870e-10, -9.819570e-01, 3.995715e+03, 2.211511e+08; + MatrixXd C = make_companion(coeffs); + EigenSolver eig(C); + CALL_SUBTEST( check_eigensolver_for_given_mat(eig,C) ); + Index n = C.rows(); + for(Index i=0;i Complex; + MatrixXcd ac = C.cast(); + ac.diagonal().array() -= eig.eigenvalues()(i); + VectorXd sv = ac.jacobiSvd().singularValues(); + // comparing to sv(0) is not enough here to catch the "bug", + // the hard-coded 1.0 is important! + VERIFY_IS_MUCH_SMALLER_THAN(sv(n-1), 1.0); + } + } + } + // regression test for bug 1557 + { + // this test is interesting because it contains zeros on the diagonal. + MatrixXd A_bug1557(3,3); + A_bug1557 << 0, 0, 0, 1, 0, 0.5887907064808635127, 0, 1, 0; + EigenSolver eig(A_bug1557); + CALL_SUBTEST( check_eigensolver_for_given_mat(eig,A_bug1557) ); + } + + // regression test for bug 1174 + { + Index n = 12; + MatrixXf A_bug1174(n,n); + A_bug1174 << 262144, 0, 0, 262144, 786432, 0, 0, 0, 0, 0, 0, 786432, + 262144, 0, 0, 262144, 786432, 0, 0, 0, 0, 0, 0, 786432, + 262144, 0, 0, 262144, 786432, 0, 0, 0, 0, 0, 0, 786432, + 262144, 0, 0, 262144, 786432, 0, 0, 0, 0, 0, 0, 786432, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0, + 0, 262144, 262144, 0, 0, 262144, 262144, 262144, 262144, 262144, 262144, 0; + EigenSolver eig(A_bug1174); + CALL_SUBTEST( check_eigensolver_for_given_mat(eig,A_bug1174) ); + } +} + +EIGEN_DECLARE_TEST(eigensolver_generic) { int s = 0; for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( eigensolver(Matrix4f()) ); s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); CALL_SUBTEST_2( eigensolver(MatrixXd(s,s)) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) // some trivial but implementation-wise tricky cases CALL_SUBTEST_2( eigensolver(MatrixXd(1,1)) ); @@ -114,12 +235,13 @@ void test_eigensolver_generic() CALL_SUBTEST_2( { MatrixXd A(1,1); - A(0,0) = std::sqrt(-1.); + A(0,0) = std::sqrt(-1.); // is Not-a-Number Eigen::EigenSolver solver(A); - MatrixXd V(1, 1); - V(0,0) = solver.eigenvectors()(0,0).real(); + VERIFY_IS_EQUAL(solver.info(), NumericalIssue); } ); + CALL_SUBTEST_2( eigensolver_generic_extra<0>() ); + TEST_SET_BUT_UNUSED_VARIABLE(s) } diff --git a/thirdparty/eigen/test/eigensolver_selfadjoint.cpp b/thirdparty/eigen/test/eigensolver_selfadjoint.cpp index 38689cfb..0fb2f4da 100644 --- a/thirdparty/eigen/test/eigensolver_selfadjoint.cpp +++ b/thirdparty/eigen/test/eigensolver_selfadjoint.cpp @@ -9,12 +9,65 @@ // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. #include "main.h" +#include "svd_fill.h" #include #include +#include + + +template void selfadjointeigensolver_essential_check(const MatrixType& m) +{ + typedef typename MatrixType::Scalar Scalar; + typedef typename NumTraits::Real RealScalar; + RealScalar eival_eps = numext::mini(test_precision(), NumTraits::dummy_precision()*20000); + + SelfAdjointEigenSolver eiSymm(m); + VERIFY_IS_EQUAL(eiSymm.info(), Success); + + RealScalar scaling = m.cwiseAbs().maxCoeff(); + + if(scaling<(std::numeric_limits::min)()) + { + VERIFY(eiSymm.eigenvalues().cwiseAbs().maxCoeff() <= (std::numeric_limits::min)()); + } + else + { + VERIFY_IS_APPROX((m.template selfadjointView() * eiSymm.eigenvectors())/scaling, + (eiSymm.eigenvectors() * eiSymm.eigenvalues().asDiagonal())/scaling); + } + VERIFY_IS_APPROX(m.template selfadjointView().eigenvalues(), eiSymm.eigenvalues()); + VERIFY_IS_UNITARY(eiSymm.eigenvectors()); + + if(m.cols()<=4) + { + SelfAdjointEigenSolver eiDirect; + eiDirect.computeDirect(m); + VERIFY_IS_EQUAL(eiDirect.info(), Success); + if(! eiSymm.eigenvalues().isApprox(eiDirect.eigenvalues(), eival_eps) ) + { + std::cerr << "reference eigenvalues: " << eiSymm.eigenvalues().transpose() << "\n" + << "obtained eigenvalues: " << eiDirect.eigenvalues().transpose() << "\n" + << "diff: " << (eiSymm.eigenvalues()-eiDirect.eigenvalues()).transpose() << "\n" + << "error (eps): " << (eiSymm.eigenvalues()-eiDirect.eigenvalues()).norm() / eiSymm.eigenvalues().norm() << " (" << eival_eps << ")\n"; + } + if(scaling<(std::numeric_limits::min)()) + { + VERIFY(eiDirect.eigenvalues().cwiseAbs().maxCoeff() <= (std::numeric_limits::min)()); + } + else + { + VERIFY_IS_APPROX(eiSymm.eigenvalues()/scaling, eiDirect.eigenvalues()/scaling); + VERIFY_IS_APPROX((m.template selfadjointView() * eiDirect.eigenvectors())/scaling, + (eiDirect.eigenvectors() * eiDirect.eigenvalues().asDiagonal())/scaling); + VERIFY_IS_APPROX(m.template selfadjointView().eigenvalues()/scaling, eiDirect.eigenvalues()/scaling); + } + + VERIFY_IS_UNITARY(eiDirect.eigenvectors()); + } +} template void selfadjointeigensolver(const MatrixType& m) { - typedef typename MatrixType::Index Index; /* this test covers the following files: EigenSolver.h, SelfAdjointEigenSolver.h (and indirectly: Tridiagonalization.h) */ @@ -31,17 +84,8 @@ template void selfadjointeigensolver(const MatrixType& m) MatrixType symmA = a.adjoint() * a + a1.adjoint() * a1; MatrixType symmC = symmA; - // randomly nullify some rows/columns - { - Index count = 1;//internal::random(-cols,cols); - for(Index k=0; k(0,cols-1); - symmA.row(i).setZero(); - symmA.col(i).setZero(); - } - } - + svd_fill_random(symmA,Symmetric); + symmA.template triangularView().setZero(); symmC.template triangularView().setZero(); @@ -49,23 +93,13 @@ template void selfadjointeigensolver(const MatrixType& m) MatrixType b1 = MatrixType::Random(rows,cols); MatrixType symmB = b.adjoint() * b + b1.adjoint() * b1; symmB.template triangularView().setZero(); + + CALL_SUBTEST( selfadjointeigensolver_essential_check(symmA) ); SelfAdjointEigenSolver eiSymm(symmA); - SelfAdjointEigenSolver eiDirect; - eiDirect.computeDirect(symmA); // generalized eigen pb GeneralizedSelfAdjointEigenSolver eiSymmGen(symmC, symmB); - VERIFY_IS_EQUAL(eiSymm.info(), Success); - VERIFY((symmA.template selfadjointView() * eiSymm.eigenvectors()).isApprox( - eiSymm.eigenvectors() * eiSymm.eigenvalues().asDiagonal(), largerEps)); - VERIFY_IS_APPROX(symmA.template selfadjointView().eigenvalues(), eiSymm.eigenvalues()); - - VERIFY_IS_EQUAL(eiDirect.info(), Success); - VERIFY((symmA.template selfadjointView() * eiDirect.eigenvectors()).isApprox( - eiDirect.eigenvectors() * eiDirect.eigenvalues().asDiagonal(), largerEps)); - VERIFY_IS_APPROX(symmA.template selfadjointView().eigenvalues(), eiDirect.eigenvalues()); - SelfAdjointEigenSolver eiSymmNoEivecs(symmA, false); VERIFY_IS_EQUAL(eiSymmNoEivecs.info(), Success); VERIFY_IS_APPROX(eiSymm.eigenvalues(), eiSymmNoEivecs.eigenvalues()); @@ -111,44 +145,131 @@ template void selfadjointeigensolver(const MatrixType& m) // test Tridiagonalization's methods Tridiagonalization tridiag(symmC); - // FIXME tridiag.matrixQ().adjoint() does not work + VERIFY_IS_APPROX(tridiag.diagonal(), tridiag.matrixT().diagonal()); + VERIFY_IS_APPROX(tridiag.subDiagonal(), tridiag.matrixT().template diagonal<-1>()); + Matrix T = tridiag.matrixT(); + if(rows>1 && cols>1) { + // FIXME check that upper and lower part are 0: + //VERIFY(T.topRightCorner(rows-2, cols-2).template triangularView().isZero()); + } + VERIFY_IS_APPROX(tridiag.diagonal(), T.diagonal()); + VERIFY_IS_APPROX(tridiag.subDiagonal(), T.template diagonal<1>()); VERIFY_IS_APPROX(MatrixType(symmC.template selfadjointView()), tridiag.matrixQ() * tridiag.matrixT().eval() * MatrixType(tridiag.matrixQ()).adjoint()); + VERIFY_IS_APPROX(MatrixType(symmC.template selfadjointView()), tridiag.matrixQ() * tridiag.matrixT() * tridiag.matrixQ().adjoint()); - if (rows > 1) + // Test computation of eigenvalues from tridiagonal matrix + if(rows > 1) + { + SelfAdjointEigenSolver eiSymmTridiag; + eiSymmTridiag.computeFromTridiagonal(tridiag.matrixT().diagonal(), tridiag.matrixT().diagonal(-1), ComputeEigenvectors); + VERIFY_IS_APPROX(eiSymm.eigenvalues(), eiSymmTridiag.eigenvalues()); + VERIFY_IS_APPROX(tridiag.matrixT(), eiSymmTridiag.eigenvectors().real() * eiSymmTridiag.eigenvalues().asDiagonal() * eiSymmTridiag.eigenvectors().real().transpose()); + } + + if (rows > 1 && rows < 20) { // Test matrix with NaN symmC(0,0) = std::numeric_limits::quiet_NaN(); SelfAdjointEigenSolver eiSymmNaN(symmC); VERIFY_IS_EQUAL(eiSymmNaN.info(), NoConvergence); } + + // regression test for bug 1098 + { + SelfAdjointEigenSolver eig(a.adjoint() * a); + eig.compute(a.adjoint() * a); + } + + // regression test for bug 478 + { + a.setZero(); + SelfAdjointEigenSolver ei3(a); + VERIFY_IS_EQUAL(ei3.info(), Success); + VERIFY_IS_MUCH_SMALLER_THAN(ei3.eigenvalues().norm(),RealScalar(1)); + VERIFY((ei3.eigenvectors().transpose()*ei3.eigenvectors().transpose()).eval().isIdentity()); + } +} + +template +void bug_854() +{ + Matrix3d m; + m << 850.961, 51.966, 0, + 51.966, 254.841, 0, + 0, 0, 0; + selfadjointeigensolver_essential_check(m); +} + +template +void bug_1014() +{ + Matrix3d m; + m << 0.11111111111111114658, 0, 0, + 0, 0.11111111111111109107, 0, + 0, 0, 0.11111111111111107719; + selfadjointeigensolver_essential_check(m); +} + +template +void bug_1225() +{ + Matrix3d m1, m2; + m1.setRandom(); + m1 = m1*m1.transpose(); + m2 = m1.triangularView(); + SelfAdjointEigenSolver eig1(m1); + SelfAdjointEigenSolver eig2(m2.selfadjointView()); + VERIFY_IS_APPROX(eig1.eigenvalues(), eig2.eigenvalues()); +} + +template +void bug_1204() +{ + SparseMatrix A(2,2); + A.setIdentity(); + SelfAdjointEigenSolver > eig(A); } -void test_eigensolver_selfadjoint() +EIGEN_DECLARE_TEST(eigensolver_selfadjoint) { int s = 0; for(int i = 0; i < g_repeat; i++) { + + // trivial test for 1x1 matrices: + CALL_SUBTEST_1( selfadjointeigensolver(Matrix())); + CALL_SUBTEST_1( selfadjointeigensolver(Matrix())); + CALL_SUBTEST_1( selfadjointeigensolver(Matrix, 1, 1>())); + // very important to test 3x3 and 2x2 matrices since we provide special paths for them - CALL_SUBTEST_1( selfadjointeigensolver(Matrix2f()) ); - CALL_SUBTEST_1( selfadjointeigensolver(Matrix2d()) ); - CALL_SUBTEST_1( selfadjointeigensolver(Matrix3f()) ); - CALL_SUBTEST_1( selfadjointeigensolver(Matrix3d()) ); + CALL_SUBTEST_12( selfadjointeigensolver(Matrix2f()) ); + CALL_SUBTEST_12( selfadjointeigensolver(Matrix2d()) ); + CALL_SUBTEST_12( selfadjointeigensolver(Matrix2cd()) ); + CALL_SUBTEST_13( selfadjointeigensolver(Matrix3f()) ); + CALL_SUBTEST_13( selfadjointeigensolver(Matrix3d()) ); + CALL_SUBTEST_13( selfadjointeigensolver(Matrix3cd()) ); CALL_SUBTEST_2( selfadjointeigensolver(Matrix4d()) ); + CALL_SUBTEST_2( selfadjointeigensolver(Matrix4cd()) ); + s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); CALL_SUBTEST_3( selfadjointeigensolver(MatrixXf(s,s)) ); - s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); CALL_SUBTEST_4( selfadjointeigensolver(MatrixXd(s,s)) ); - s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); CALL_SUBTEST_5( selfadjointeigensolver(MatrixXcd(s,s)) ); - - s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); CALL_SUBTEST_9( selfadjointeigensolver(Matrix,Dynamic,Dynamic,RowMajor>(s,s)) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) // some trivial but implementation-wise tricky cases CALL_SUBTEST_4( selfadjointeigensolver(MatrixXd(1,1)) ); CALL_SUBTEST_4( selfadjointeigensolver(MatrixXd(2,2)) ); + CALL_SUBTEST_5( selfadjointeigensolver(MatrixXcd(1,1)) ); + CALL_SUBTEST_5( selfadjointeigensolver(MatrixXcd(2,2)) ); CALL_SUBTEST_6( selfadjointeigensolver(Matrix()) ); CALL_SUBTEST_7( selfadjointeigensolver(Matrix()) ); } + + CALL_SUBTEST_13( bug_854<0>() ); + CALL_SUBTEST_13( bug_1014<0>() ); + CALL_SUBTEST_13( bug_1204<0>() ); + CALL_SUBTEST_13( bug_1225<0>() ); // Test problem size constructors s = internal::random(1,EIGEN_TEST_MAX_SIZE/4); diff --git a/thirdparty/eigen/test/evaluator_common.h b/thirdparty/eigen/test/evaluator_common.h new file mode 100644 index 00000000..e69de29b diff --git a/thirdparty/eigen/test/evaluators.cpp b/thirdparty/eigen/test/evaluators.cpp new file mode 100644 index 00000000..2810cd26 --- /dev/null +++ b/thirdparty/eigen/test/evaluators.cpp @@ -0,0 +1,525 @@ + +#include "main.h" + +namespace Eigen { + + template + const Product + prod(const Lhs& lhs, const Rhs& rhs) + { + return Product(lhs,rhs); + } + + template + const Product + lazyprod(const Lhs& lhs, const Rhs& rhs) + { + return Product(lhs,rhs); + } + + template + EIGEN_STRONG_INLINE + DstXprType& copy_using_evaluator(const EigenBase &dst, const SrcXprType &src) + { + call_assignment(dst.const_cast_derived(), src.derived(), internal::assign_op()); + return dst.const_cast_derived(); + } + + template class StorageBase, typename SrcXprType> + EIGEN_STRONG_INLINE + const DstXprType& copy_using_evaluator(const NoAlias& dst, const SrcXprType &src) + { + call_assignment(dst, src.derived(), internal::assign_op()); + return dst.expression(); + } + + template + EIGEN_STRONG_INLINE + DstXprType& copy_using_evaluator(const PlainObjectBase &dst, const SrcXprType &src) + { + #ifdef EIGEN_NO_AUTOMATIC_RESIZING + eigen_assert((dst.size()==0 || (IsVectorAtCompileTime ? (dst.size() == src.size()) + : (dst.rows() == src.rows() && dst.cols() == src.cols()))) + && "Size mismatch. Automatic resizing is disabled because EIGEN_NO_AUTOMATIC_RESIZING is defined"); + #else + dst.const_cast_derived().resizeLike(src.derived()); + #endif + + call_assignment(dst.const_cast_derived(), src.derived(), internal::assign_op()); + return dst.const_cast_derived(); + } + + template + void add_assign_using_evaluator(const DstXprType& dst, const SrcXprType& src) + { + typedef typename DstXprType::Scalar Scalar; + call_assignment(const_cast(dst), src.derived(), internal::add_assign_op()); + } + + template + void subtract_assign_using_evaluator(const DstXprType& dst, const SrcXprType& src) + { + typedef typename DstXprType::Scalar Scalar; + call_assignment(const_cast(dst), src.derived(), internal::sub_assign_op()); + } + + template + void multiply_assign_using_evaluator(const DstXprType& dst, const SrcXprType& src) + { + typedef typename DstXprType::Scalar Scalar; + call_assignment(dst.const_cast_derived(), src.derived(), internal::mul_assign_op()); + } + + template + void divide_assign_using_evaluator(const DstXprType& dst, const SrcXprType& src) + { + typedef typename DstXprType::Scalar Scalar; + call_assignment(dst.const_cast_derived(), src.derived(), internal::div_assign_op()); + } + + template + void swap_using_evaluator(const DstXprType& dst, const SrcXprType& src) + { + typedef typename DstXprType::Scalar Scalar; + call_assignment(dst.const_cast_derived(), src.const_cast_derived(), internal::swap_assign_op()); + } + + namespace internal { + template class StorageBase, typename Src, typename Func> + EIGEN_DEVICE_FUNC void call_assignment(const NoAlias& dst, const Src& src, const Func& func) + { + call_assignment_no_alias(dst.expression(), src, func); + } + + template class StorageBase, typename Src, typename Func> + EIGEN_DEVICE_FUNC void call_restricted_packet_assignment(const NoAlias& dst, const Src& src, const Func& func) + { + call_restricted_packet_assignment_no_alias(dst.expression(), src, func); + } + } + +} + +template long get_cost(const XprType& ) { return Eigen::internal::evaluator::CoeffReadCost; } + +using namespace std; + +#define VERIFY_IS_APPROX_EVALUATOR(DEST,EXPR) VERIFY_IS_APPROX(copy_using_evaluator(DEST,(EXPR)), (EXPR).eval()); +#define VERIFY_IS_APPROX_EVALUATOR2(DEST,EXPR,REF) VERIFY_IS_APPROX(copy_using_evaluator(DEST,(EXPR)), (REF).eval()); + +EIGEN_DECLARE_TEST(evaluators) +{ + // Testing Matrix evaluator and Transpose + Vector2d v = Vector2d::Random(); + const Vector2d v_const(v); + Vector2d v2; + RowVector2d w; + + VERIFY_IS_APPROX_EVALUATOR(v2, v); + VERIFY_IS_APPROX_EVALUATOR(v2, v_const); + + // Testing Transpose + VERIFY_IS_APPROX_EVALUATOR(w, v.transpose()); // Transpose as rvalue + VERIFY_IS_APPROX_EVALUATOR(w, v_const.transpose()); + + copy_using_evaluator(w.transpose(), v); // Transpose as lvalue + VERIFY_IS_APPROX(w,v.transpose().eval()); + + copy_using_evaluator(w.transpose(), v_const); + VERIFY_IS_APPROX(w,v_const.transpose().eval()); + + // Testing Array evaluator + { + ArrayXXf a(2,3); + ArrayXXf b(3,2); + a << 1,2,3, 4,5,6; + const ArrayXXf a_const(a); + + VERIFY_IS_APPROX_EVALUATOR(b, a.transpose()); + + VERIFY_IS_APPROX_EVALUATOR(b, a_const.transpose()); + + // Testing CwiseNullaryOp evaluator + copy_using_evaluator(w, RowVector2d::Random()); + VERIFY((w.array() >= -1).all() && (w.array() <= 1).all()); // not easy to test ... + + VERIFY_IS_APPROX_EVALUATOR(w, RowVector2d::Zero()); + + VERIFY_IS_APPROX_EVALUATOR(w, RowVector2d::Constant(3)); + + // mix CwiseNullaryOp and transpose + VERIFY_IS_APPROX_EVALUATOR(w, Vector2d::Zero().transpose()); + } + + { + // test product expressions + int s = internal::random(1,100); + MatrixXf a(s,s), b(s,s), c(s,s), d(s,s); + a.setRandom(); + b.setRandom(); + c.setRandom(); + d.setRandom(); + VERIFY_IS_APPROX_EVALUATOR(d, (a + b)); + VERIFY_IS_APPROX_EVALUATOR(d, (a + b).transpose()); + VERIFY_IS_APPROX_EVALUATOR2(d, prod(a,b), a*b); + VERIFY_IS_APPROX_EVALUATOR2(d.noalias(), prod(a,b), a*b); + VERIFY_IS_APPROX_EVALUATOR2(d, prod(a,b) + c, a*b + c); + VERIFY_IS_APPROX_EVALUATOR2(d, s * prod(a,b), s * a*b); + VERIFY_IS_APPROX_EVALUATOR2(d, prod(a,b).transpose(), (a*b).transpose()); + VERIFY_IS_APPROX_EVALUATOR2(d, prod(a,b) + prod(b,c), a*b + b*c); + + // check that prod works even with aliasing present + c = a*a; + copy_using_evaluator(a, prod(a,a)); + VERIFY_IS_APPROX(a,c); + + // check compound assignment of products + d = c; + add_assign_using_evaluator(c.noalias(), prod(a,b)); + d.noalias() += a*b; + VERIFY_IS_APPROX(c, d); + + d = c; + subtract_assign_using_evaluator(c.noalias(), prod(a,b)); + d.noalias() -= a*b; + VERIFY_IS_APPROX(c, d); + } + + { + // test product with all possible sizes + int s = internal::random(1,100); + Matrix m11, res11; m11.setRandom(1,1); + Matrix m14, res14; m14.setRandom(1,4); + Matrix m1X, res1X; m1X.setRandom(1,s); + Matrix m41, res41; m41.setRandom(4,1); + Matrix m44, res44; m44.setRandom(4,4); + Matrix m4X, res4X; m4X.setRandom(4,s); + Matrix mX1, resX1; mX1.setRandom(s,1); + Matrix mX4, resX4; mX4.setRandom(s,4); + Matrix mXX, resXX; mXX.setRandom(s,s); + + VERIFY_IS_APPROX_EVALUATOR2(res11, prod(m11,m11), m11*m11); + VERIFY_IS_APPROX_EVALUATOR2(res11, prod(m14,m41), m14*m41); + VERIFY_IS_APPROX_EVALUATOR2(res11, prod(m1X,mX1), m1X*mX1); + VERIFY_IS_APPROX_EVALUATOR2(res14, prod(m11,m14), m11*m14); + VERIFY_IS_APPROX_EVALUATOR2(res14, prod(m14,m44), m14*m44); + VERIFY_IS_APPROX_EVALUATOR2(res14, prod(m1X,mX4), m1X*mX4); + VERIFY_IS_APPROX_EVALUATOR2(res1X, prod(m11,m1X), m11*m1X); + VERIFY_IS_APPROX_EVALUATOR2(res1X, prod(m14,m4X), m14*m4X); + VERIFY_IS_APPROX_EVALUATOR2(res1X, prod(m1X,mXX), m1X*mXX); + VERIFY_IS_APPROX_EVALUATOR2(res41, prod(m41,m11), m41*m11); + VERIFY_IS_APPROX_EVALUATOR2(res41, prod(m44,m41), m44*m41); + VERIFY_IS_APPROX_EVALUATOR2(res41, prod(m4X,mX1), m4X*mX1); + VERIFY_IS_APPROX_EVALUATOR2(res44, prod(m41,m14), m41*m14); + VERIFY_IS_APPROX_EVALUATOR2(res44, prod(m44,m44), m44*m44); + VERIFY_IS_APPROX_EVALUATOR2(res44, prod(m4X,mX4), m4X*mX4); + VERIFY_IS_APPROX_EVALUATOR2(res4X, prod(m41,m1X), m41*m1X); + VERIFY_IS_APPROX_EVALUATOR2(res4X, prod(m44,m4X), m44*m4X); + VERIFY_IS_APPROX_EVALUATOR2(res4X, prod(m4X,mXX), m4X*mXX); + VERIFY_IS_APPROX_EVALUATOR2(resX1, prod(mX1,m11), mX1*m11); + VERIFY_IS_APPROX_EVALUATOR2(resX1, prod(mX4,m41), mX4*m41); + VERIFY_IS_APPROX_EVALUATOR2(resX1, prod(mXX,mX1), mXX*mX1); + VERIFY_IS_APPROX_EVALUATOR2(resX4, prod(mX1,m14), mX1*m14); + VERIFY_IS_APPROX_EVALUATOR2(resX4, prod(mX4,m44), mX4*m44); + VERIFY_IS_APPROX_EVALUATOR2(resX4, prod(mXX,mX4), mXX*mX4); + VERIFY_IS_APPROX_EVALUATOR2(resXX, prod(mX1,m1X), mX1*m1X); + VERIFY_IS_APPROX_EVALUATOR2(resXX, prod(mX4,m4X), mX4*m4X); + VERIFY_IS_APPROX_EVALUATOR2(resXX, prod(mXX,mXX), mXX*mXX); + } + + { + ArrayXXf a(2,3); + ArrayXXf b(3,2); + a << 1,2,3, 4,5,6; + const ArrayXXf a_const(a); + + // this does not work because Random is eval-before-nested: + // copy_using_evaluator(w, Vector2d::Random().transpose()); + + // test CwiseUnaryOp + VERIFY_IS_APPROX_EVALUATOR(v2, 3 * v); + VERIFY_IS_APPROX_EVALUATOR(w, (3 * v).transpose()); + VERIFY_IS_APPROX_EVALUATOR(b, (a + 3).transpose()); + VERIFY_IS_APPROX_EVALUATOR(b, (2 * a_const + 3).transpose()); + + // test CwiseBinaryOp + VERIFY_IS_APPROX_EVALUATOR(v2, v + Vector2d::Ones()); + VERIFY_IS_APPROX_EVALUATOR(w, (v + Vector2d::Ones()).transpose().cwiseProduct(RowVector2d::Constant(3))); + + // dynamic matrices and arrays + MatrixXd mat1(6,6), mat2(6,6); + VERIFY_IS_APPROX_EVALUATOR(mat1, MatrixXd::Identity(6,6)); + VERIFY_IS_APPROX_EVALUATOR(mat2, mat1); + copy_using_evaluator(mat2.transpose(), mat1); + VERIFY_IS_APPROX(mat2.transpose(), mat1); + + ArrayXXd arr1(6,6), arr2(6,6); + VERIFY_IS_APPROX_EVALUATOR(arr1, ArrayXXd::Constant(6,6, 3.0)); + VERIFY_IS_APPROX_EVALUATOR(arr2, arr1); + + // test automatic resizing + mat2.resize(3,3); + VERIFY_IS_APPROX_EVALUATOR(mat2, mat1); + arr2.resize(9,9); + VERIFY_IS_APPROX_EVALUATOR(arr2, arr1); + + // test direct traversal + Matrix3f m3; + Array33f a3; + VERIFY_IS_APPROX_EVALUATOR(m3, Matrix3f::Identity()); // matrix, nullary + // TODO: find a way to test direct traversal with array + VERIFY_IS_APPROX_EVALUATOR(m3.transpose(), Matrix3f::Identity().transpose()); // transpose + VERIFY_IS_APPROX_EVALUATOR(m3, 2 * Matrix3f::Identity()); // unary + VERIFY_IS_APPROX_EVALUATOR(m3, Matrix3f::Identity() + Matrix3f::Zero()); // binary + VERIFY_IS_APPROX_EVALUATOR(m3.block(0,0,2,2), Matrix3f::Identity().block(1,1,2,2)); // block + + // test linear traversal + VERIFY_IS_APPROX_EVALUATOR(m3, Matrix3f::Zero()); // matrix, nullary + VERIFY_IS_APPROX_EVALUATOR(a3, Array33f::Zero()); // array + VERIFY_IS_APPROX_EVALUATOR(m3.transpose(), Matrix3f::Zero().transpose()); // transpose + VERIFY_IS_APPROX_EVALUATOR(m3, 2 * Matrix3f::Zero()); // unary + VERIFY_IS_APPROX_EVALUATOR(m3, Matrix3f::Zero() + m3); // binary + + // test inner vectorization + Matrix4f m4, m4src = Matrix4f::Random(); + Array44f a4, a4src = Matrix4f::Random(); + VERIFY_IS_APPROX_EVALUATOR(m4, m4src); // matrix + VERIFY_IS_APPROX_EVALUATOR(a4, a4src); // array + VERIFY_IS_APPROX_EVALUATOR(m4.transpose(), m4src.transpose()); // transpose + // TODO: find out why Matrix4f::Zero() does not allow inner vectorization + VERIFY_IS_APPROX_EVALUATOR(m4, 2 * m4src); // unary + VERIFY_IS_APPROX_EVALUATOR(m4, m4src + m4src); // binary + + // test linear vectorization + MatrixXf mX(6,6), mXsrc = MatrixXf::Random(6,6); + ArrayXXf aX(6,6), aXsrc = ArrayXXf::Random(6,6); + VERIFY_IS_APPROX_EVALUATOR(mX, mXsrc); // matrix + VERIFY_IS_APPROX_EVALUATOR(aX, aXsrc); // array + VERIFY_IS_APPROX_EVALUATOR(mX.transpose(), mXsrc.transpose()); // transpose + VERIFY_IS_APPROX_EVALUATOR(mX, MatrixXf::Zero(6,6)); // nullary + VERIFY_IS_APPROX_EVALUATOR(mX, 2 * mXsrc); // unary + VERIFY_IS_APPROX_EVALUATOR(mX, mXsrc + mXsrc); // binary + + // test blocks and slice vectorization + VERIFY_IS_APPROX_EVALUATOR(m4, (mXsrc.block<4,4>(1,0))); + VERIFY_IS_APPROX_EVALUATOR(aX, ArrayXXf::Constant(10, 10, 3.0).block(2, 3, 6, 6)); + + Matrix4f m4ref = m4; + copy_using_evaluator(m4.block(1, 1, 2, 3), m3.bottomRows(2)); + m4ref.block(1, 1, 2, 3) = m3.bottomRows(2); + VERIFY_IS_APPROX(m4, m4ref); + + mX.setIdentity(20,20); + MatrixXf mXref = MatrixXf::Identity(20,20); + mXsrc = MatrixXf::Random(9,12); + copy_using_evaluator(mX.block(4, 4, 9, 12), mXsrc); + mXref.block(4, 4, 9, 12) = mXsrc; + VERIFY_IS_APPROX(mX, mXref); + + // test Map + const float raw[3] = {1,2,3}; + float buffer[3] = {0,0,0}; + Vector3f v3; + Array3f a3f; + VERIFY_IS_APPROX_EVALUATOR(v3, Map(raw)); + VERIFY_IS_APPROX_EVALUATOR(a3f, Map(raw)); + Vector3f::Map(buffer) = 2*v3; + VERIFY(buffer[0] == 2); + VERIFY(buffer[1] == 4); + VERIFY(buffer[2] == 6); + + // test CwiseUnaryView + mat1.setRandom(); + mat2.setIdentity(); + MatrixXcd matXcd(6,6), matXcd_ref(6,6); + copy_using_evaluator(matXcd.real(), mat1); + copy_using_evaluator(matXcd.imag(), mat2); + matXcd_ref.real() = mat1; + matXcd_ref.imag() = mat2; + VERIFY_IS_APPROX(matXcd, matXcd_ref); + + // test Select + VERIFY_IS_APPROX_EVALUATOR(aX, (aXsrc > 0).select(aXsrc, -aXsrc)); + + // test Replicate + mXsrc = MatrixXf::Random(6, 6); + VectorXf vX = VectorXf::Random(6); + mX.resize(6, 6); + VERIFY_IS_APPROX_EVALUATOR(mX, mXsrc.colwise() + vX); + matXcd.resize(12, 12); + VERIFY_IS_APPROX_EVALUATOR(matXcd, matXcd_ref.replicate(2,2)); + VERIFY_IS_APPROX_EVALUATOR(matXcd, (matXcd_ref.replicate<2,2>())); + + // test partial reductions + VectorXd vec1(6); + VERIFY_IS_APPROX_EVALUATOR(vec1, mat1.rowwise().sum()); + VERIFY_IS_APPROX_EVALUATOR(vec1, mat1.colwise().sum().transpose()); + + // test MatrixWrapper and ArrayWrapper + mat1.setRandom(6,6); + arr1.setRandom(6,6); + VERIFY_IS_APPROX_EVALUATOR(mat2, arr1.matrix()); + VERIFY_IS_APPROX_EVALUATOR(arr2, mat1.array()); + VERIFY_IS_APPROX_EVALUATOR(mat2, (arr1 + 2).matrix()); + VERIFY_IS_APPROX_EVALUATOR(arr2, mat1.array() + 2); + mat2.array() = arr1 * arr1; + VERIFY_IS_APPROX(mat2, (arr1 * arr1).matrix()); + arr2.matrix() = MatrixXd::Identity(6,6); + VERIFY_IS_APPROX(arr2, MatrixXd::Identity(6,6).array()); + + // test Reverse + VERIFY_IS_APPROX_EVALUATOR(arr2, arr1.reverse()); + VERIFY_IS_APPROX_EVALUATOR(arr2, arr1.colwise().reverse()); + VERIFY_IS_APPROX_EVALUATOR(arr2, arr1.rowwise().reverse()); + arr2.reverse() = arr1; + VERIFY_IS_APPROX(arr2, arr1.reverse()); + mat2.array() = mat1.array().reverse(); + VERIFY_IS_APPROX(mat2.array(), mat1.array().reverse()); + + // test Diagonal + VERIFY_IS_APPROX_EVALUATOR(vec1, mat1.diagonal()); + vec1.resize(5); + VERIFY_IS_APPROX_EVALUATOR(vec1, mat1.diagonal(1)); + VERIFY_IS_APPROX_EVALUATOR(vec1, mat1.diagonal<-1>()); + vec1.setRandom(); + + mat2 = mat1; + copy_using_evaluator(mat1.diagonal(1), vec1); + mat2.diagonal(1) = vec1; + VERIFY_IS_APPROX(mat1, mat2); + + copy_using_evaluator(mat1.diagonal<-1>(), mat1.diagonal(1)); + mat2.diagonal<-1>() = mat2.diagonal(1); + VERIFY_IS_APPROX(mat1, mat2); + } + + { + // test swapping + MatrixXd mat1, mat2, mat1ref, mat2ref; + mat1ref = mat1 = MatrixXd::Random(6, 6); + mat2ref = mat2 = 2 * mat1 + MatrixXd::Identity(6, 6); + swap_using_evaluator(mat1, mat2); + mat1ref.swap(mat2ref); + VERIFY_IS_APPROX(mat1, mat1ref); + VERIFY_IS_APPROX(mat2, mat2ref); + + swap_using_evaluator(mat1.block(0, 0, 3, 3), mat2.block(3, 3, 3, 3)); + mat1ref.block(0, 0, 3, 3).swap(mat2ref.block(3, 3, 3, 3)); + VERIFY_IS_APPROX(mat1, mat1ref); + VERIFY_IS_APPROX(mat2, mat2ref); + + swap_using_evaluator(mat1.row(2), mat2.col(3).transpose()); + mat1.row(2).swap(mat2.col(3).transpose()); + VERIFY_IS_APPROX(mat1, mat1ref); + VERIFY_IS_APPROX(mat2, mat2ref); + } + + { + // test compound assignment + const Matrix4d mat_const = Matrix4d::Random(); + Matrix4d mat, mat_ref; + mat = mat_ref = Matrix4d::Identity(); + add_assign_using_evaluator(mat, mat_const); + mat_ref += mat_const; + VERIFY_IS_APPROX(mat, mat_ref); + + subtract_assign_using_evaluator(mat.row(1), 2*mat.row(2)); + mat_ref.row(1) -= 2*mat_ref.row(2); + VERIFY_IS_APPROX(mat, mat_ref); + + const ArrayXXf arr_const = ArrayXXf::Random(5,3); + ArrayXXf arr, arr_ref; + arr = arr_ref = ArrayXXf::Constant(5, 3, 0.5); + multiply_assign_using_evaluator(arr, arr_const); + arr_ref *= arr_const; + VERIFY_IS_APPROX(arr, arr_ref); + + divide_assign_using_evaluator(arr.row(1), arr.row(2) + 1); + arr_ref.row(1) /= (arr_ref.row(2) + 1); + VERIFY_IS_APPROX(arr, arr_ref); + } + + { + // test triangular shapes + MatrixXd A = MatrixXd::Random(6,6), B(6,6), C(6,6), D(6,6); + A.setRandom();B.setRandom(); + VERIFY_IS_APPROX_EVALUATOR2(B, A.triangularView(), MatrixXd(A.triangularView())); + + A.setRandom();B.setRandom(); + VERIFY_IS_APPROX_EVALUATOR2(B, A.triangularView(), MatrixXd(A.triangularView())); + + A.setRandom();B.setRandom(); + VERIFY_IS_APPROX_EVALUATOR2(B, A.triangularView(), MatrixXd(A.triangularView())); + + A.setRandom();B.setRandom(); + C = B; C.triangularView() = A; + copy_using_evaluator(B.triangularView(), A); + VERIFY(B.isApprox(C) && "copy_using_evaluator(B.triangularView(), A)"); + + A.setRandom();B.setRandom(); + C = B; C.triangularView() = A.triangularView(); + copy_using_evaluator(B.triangularView(), A.triangularView()); + VERIFY(B.isApprox(C) && "copy_using_evaluator(B.triangularView(), A.triangularView())"); + + + A.setRandom();B.setRandom(); + C = B; C.triangularView() = A.triangularView().transpose(); + copy_using_evaluator(B.triangularView(), A.triangularView().transpose()); + VERIFY(B.isApprox(C) && "copy_using_evaluator(B.triangularView(), A.triangularView().transpose())"); + + + A.setRandom();B.setRandom(); C = B; D = A; + C.triangularView().swap(D.triangularView()); + swap_using_evaluator(B.triangularView(), A.triangularView()); + VERIFY(B.isApprox(C) && "swap_using_evaluator(B.triangularView(), A.triangularView())"); + + + VERIFY_IS_APPROX_EVALUATOR2(B, prod(A.triangularView(),A), MatrixXd(A.triangularView()*A)); + + VERIFY_IS_APPROX_EVALUATOR2(B, prod(A.selfadjointView(),A), MatrixXd(A.selfadjointView()*A)); + } + + { + // test diagonal shapes + VectorXd d = VectorXd::Random(6); + MatrixXd A = MatrixXd::Random(6,6), B(6,6); + A.setRandom();B.setRandom(); + + VERIFY_IS_APPROX_EVALUATOR2(B, lazyprod(d.asDiagonal(),A), MatrixXd(d.asDiagonal()*A)); + VERIFY_IS_APPROX_EVALUATOR2(B, lazyprod(A,d.asDiagonal()), MatrixXd(A*d.asDiagonal())); + } + + { + // test CoeffReadCost + Matrix4d a, b; + VERIFY_IS_EQUAL( get_cost(a), 1 ); + VERIFY_IS_EQUAL( get_cost(a+b), 3); + VERIFY_IS_EQUAL( get_cost(2*a+b), 4); + VERIFY_IS_EQUAL( get_cost(a*b), 1); + VERIFY_IS_EQUAL( get_cost(a.lazyProduct(b)), 15); + VERIFY_IS_EQUAL( get_cost(a*(a*b)), 1); + VERIFY_IS_EQUAL( get_cost(a.lazyProduct(a*b)), 15); + VERIFY_IS_EQUAL( get_cost(a*(a+b)), 1); + VERIFY_IS_EQUAL( get_cost(a.lazyProduct(a+b)), 15); + } + + // regression test for PR 544 and bug 1622 (introduced in #71609c4) + { + // test restricted_packet_assignment with an unaligned destination + const size_t M = 2; + const size_t K = 2; + const size_t N = 5; + float *destMem = new float[(M*N) + 1]; + float *dest = (internal::UIntPtr(destMem)%EIGEN_MAX_ALIGN_BYTES) == 0 ? destMem+1 : destMem; + + const Matrix a = Matrix::Random(M, K); + const Matrix b = Matrix::Random(K, N); + + Map > z(dest, M, N);; + Product, Matrix, LazyProduct> tmp(a,b); + internal::call_restricted_packet_assignment(z.noalias(), tmp.derived(), internal::assign_op()); + + VERIFY_IS_APPROX(z, a*b); + delete[] destMem; + } +} diff --git a/thirdparty/eigen/test/exceptions.cpp b/thirdparty/eigen/test/exceptions.cpp index b83fb82b..3d93060a 100644 --- a/thirdparty/eigen/test/exceptions.cpp +++ b/thirdparty/eigen/test/exceptions.cpp @@ -8,93 +8,34 @@ // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -// Various sanity tests with exceptions: +// Various sanity tests with exceptions and non trivially copyable scalar type. // - no memory leak when a custom scalar type trow an exceptions // - todo: complete the list of tests! #define EIGEN_STACK_ALLOCATION_LIMIT 100000000 #include "main.h" - -struct my_exception -{ - my_exception() {} - ~my_exception() {} -}; - -class ScalarWithExceptions -{ - public: - ScalarWithExceptions() { init(); } - ScalarWithExceptions(const float& _v) { init(); *v = _v; } - ScalarWithExceptions(const ScalarWithExceptions& other) { init(); *v = *(other.v); } - ~ScalarWithExceptions() { - delete v; - instances--; - } - - void init() { - v = new float; - instances++; - } - - ScalarWithExceptions operator+(const ScalarWithExceptions& other) const - { - countdown--; - if(countdown<=0) - throw my_exception(); - return ScalarWithExceptions(*v+*other.v); - } - - ScalarWithExceptions operator-(const ScalarWithExceptions& other) const - { return ScalarWithExceptions(*v-*other.v); } - - ScalarWithExceptions operator*(const ScalarWithExceptions& other) const - { return ScalarWithExceptions((*v)*(*other.v)); } - - ScalarWithExceptions& operator+=(const ScalarWithExceptions& other) - { *v+=*other.v; return *this; } - ScalarWithExceptions& operator-=(const ScalarWithExceptions& other) - { *v-=*other.v; return *this; } - ScalarWithExceptions& operator=(const ScalarWithExceptions& other) - { *v = *(other.v); return *this; } - - bool operator==(const ScalarWithExceptions& other) const - { return *v==*other.v; } - bool operator!=(const ScalarWithExceptions& other) const - { return *v!=*other.v; } - - float* v; - static int instances; - static int countdown; -}; - -ScalarWithExceptions real(const ScalarWithExceptions &x) { return x; } -ScalarWithExceptions imag(const ScalarWithExceptions & ) { return 0; } -ScalarWithExceptions conj(const ScalarWithExceptions &x) { return x; } - -int ScalarWithExceptions::instances = 0; -int ScalarWithExceptions::countdown = 0; - +#include "AnnoyingScalar.h" #define CHECK_MEMLEAK(OP) { \ - ScalarWithExceptions::countdown = 100; \ - int before = ScalarWithExceptions::instances; \ - bool exception_thrown = false; \ - try { OP; } \ + AnnoyingScalar::countdown = 100; \ + int before = AnnoyingScalar::instances; \ + bool exception_thrown = false; \ + try { OP; } \ catch (my_exception) { \ exception_thrown = true; \ - VERIFY(ScalarWithExceptions::instances==before && "memory leak detected in " && EIGEN_MAKESTRING(OP)); \ + VERIFY(AnnoyingScalar::instances==before && "memory leak detected in " && EIGEN_MAKESTRING(OP)); \ } \ - VERIFY(exception_thrown && " no exception thrown in " && EIGEN_MAKESTRING(OP)); \ + VERIFY( (AnnoyingScalar::dont_throw) || (exception_thrown && " no exception thrown in " && EIGEN_MAKESTRING(OP)) ); \ } -void memoryleak() +EIGEN_DECLARE_TEST(exceptions) { - typedef Eigen::Matrix VectorType; - typedef Eigen::Matrix MatrixType; + typedef Eigen::Matrix VectorType; + typedef Eigen::Matrix MatrixType; { + AnnoyingScalar::dont_throw = false; int n = 50; VectorType v0(n), v1(n); MatrixType m0(n,n), m1(n,n), m2(n,n); @@ -104,10 +45,5 @@ void memoryleak() CHECK_MEMLEAK(m2 = m0 * m1 * m2); CHECK_MEMLEAK((v0+v1).dot(v0+v1)); } - VERIFY(ScalarWithExceptions::instances==0 && "global memory leak detected in " && EIGEN_MAKESTRING(OP)); \ -} - -void test_exceptions() -{ - CALL_SUBTEST( memoryleak() ); + VERIFY(AnnoyingScalar::instances==0 && "global memory leak detected in " && EIGEN_MAKESTRING(OP)); } diff --git a/thirdparty/eigen/test/fastmath.cpp b/thirdparty/eigen/test/fastmath.cpp new file mode 100644 index 00000000..00a1a59b --- /dev/null +++ b/thirdparty/eigen/test/fastmath.cpp @@ -0,0 +1,99 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2015 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" + +void check(bool b, bool ref) +{ + std::cout << b; + if(b==ref) + std::cout << " OK "; + else + std::cout << " BAD "; +} + +#if EIGEN_COMP_MSVC && EIGEN_COMP_MSVC < 1800 +namespace std { + template bool (isfinite)(T x) { return _finite(x); } + template bool (isnan)(T x) { return _isnan(x); } + template bool (isinf)(T x) { return _fpclass(x)==_FPCLASS_NINF || _fpclass(x)==_FPCLASS_PINF; } +} +#endif + +template +void check_inf_nan(bool dryrun) { + Matrix m(10); + m.setRandom(); + m(3) = std::numeric_limits::quiet_NaN(); + + if(dryrun) + { + std::cout << "std::isfinite(" << m(3) << ") = "; check((std::isfinite)(m(3)),false); std::cout << " ; numext::isfinite = "; check((numext::isfinite)(m(3)), false); std::cout << "\n"; + std::cout << "std::isinf(" << m(3) << ") = "; check((std::isinf)(m(3)),false); std::cout << " ; numext::isinf = "; check((numext::isinf)(m(3)), false); std::cout << "\n"; + std::cout << "std::isnan(" << m(3) << ") = "; check((std::isnan)(m(3)),true); std::cout << " ; numext::isnan = "; check((numext::isnan)(m(3)), true); std::cout << "\n"; + std::cout << "allFinite: "; check(m.allFinite(), 0); std::cout << "\n"; + std::cout << "hasNaN: "; check(m.hasNaN(), 1); std::cout << "\n"; + std::cout << "\n"; + } + else + { + if( (std::isfinite)(m(3))) g_test_level=1; VERIFY( !(numext::isfinite)(m(3)) ); g_test_level=0; + if( (std::isinf) (m(3))) g_test_level=1; VERIFY( !(numext::isinf)(m(3)) ); g_test_level=0; + if(!(std::isnan) (m(3))) g_test_level=1; VERIFY( (numext::isnan)(m(3)) ); g_test_level=0; + if( (std::isfinite)(m(3))) g_test_level=1; VERIFY( !m.allFinite() ); g_test_level=0; + if(!(std::isnan) (m(3))) g_test_level=1; VERIFY( m.hasNaN() ); g_test_level=0; + } + T hidden_zero = (std::numeric_limits::min)()*(std::numeric_limits::min)(); + m(4) /= hidden_zero; + if(dryrun) + { + std::cout << "std::isfinite(" << m(4) << ") = "; check((std::isfinite)(m(4)),false); std::cout << " ; numext::isfinite = "; check((numext::isfinite)(m(4)), false); std::cout << "\n"; + std::cout << "std::isinf(" << m(4) << ") = "; check((std::isinf)(m(4)),true); std::cout << " ; numext::isinf = "; check((numext::isinf)(m(4)), true); std::cout << "\n"; + std::cout << "std::isnan(" << m(4) << ") = "; check((std::isnan)(m(4)),false); std::cout << " ; numext::isnan = "; check((numext::isnan)(m(4)), false); std::cout << "\n"; + std::cout << "allFinite: "; check(m.allFinite(), 0); std::cout << "\n"; + std::cout << "hasNaN: "; check(m.hasNaN(), 1); std::cout << "\n"; + std::cout << "\n"; + } + else + { + if( (std::isfinite)(m(3))) g_test_level=1; VERIFY( !(numext::isfinite)(m(4)) ); g_test_level=0; + if(!(std::isinf) (m(3))) g_test_level=1; VERIFY( (numext::isinf)(m(4)) ); g_test_level=0; + if( (std::isnan) (m(3))) g_test_level=1; VERIFY( !(numext::isnan)(m(4)) ); g_test_level=0; + if( (std::isfinite)(m(3))) g_test_level=1; VERIFY( !m.allFinite() ); g_test_level=0; + if(!(std::isnan) (m(3))) g_test_level=1; VERIFY( m.hasNaN() ); g_test_level=0; + } + m(3) = 0; + if(dryrun) + { + std::cout << "std::isfinite(" << m(3) << ") = "; check((std::isfinite)(m(3)),true); std::cout << " ; numext::isfinite = "; check((numext::isfinite)(m(3)), true); std::cout << "\n"; + std::cout << "std::isinf(" << m(3) << ") = "; check((std::isinf)(m(3)),false); std::cout << " ; numext::isinf = "; check((numext::isinf)(m(3)), false); std::cout << "\n"; + std::cout << "std::isnan(" << m(3) << ") = "; check((std::isnan)(m(3)),false); std::cout << " ; numext::isnan = "; check((numext::isnan)(m(3)), false); std::cout << "\n"; + std::cout << "allFinite: "; check(m.allFinite(), 0); std::cout << "\n"; + std::cout << "hasNaN: "; check(m.hasNaN(), 0); std::cout << "\n"; + std::cout << "\n\n"; + } + else + { + if(!(std::isfinite)(m(3))) g_test_level=1; VERIFY( (numext::isfinite)(m(3)) ); g_test_level=0; + if( (std::isinf) (m(3))) g_test_level=1; VERIFY( !(numext::isinf)(m(3)) ); g_test_level=0; + if( (std::isnan) (m(3))) g_test_level=1; VERIFY( !(numext::isnan)(m(3)) ); g_test_level=0; + if( (std::isfinite)(m(3))) g_test_level=1; VERIFY( !m.allFinite() ); g_test_level=0; + if( (std::isnan) (m(3))) g_test_level=1; VERIFY( !m.hasNaN() ); g_test_level=0; + } +} + +EIGEN_DECLARE_TEST(fastmath) { + std::cout << "*** float *** \n\n"; check_inf_nan(true); + std::cout << "*** double ***\n\n"; check_inf_nan(true); + std::cout << "*** long double *** \n\n"; check_inf_nan(true); + + check_inf_nan(false); + check_inf_nan(false); + check_inf_nan(false); +} diff --git a/thirdparty/eigen/test/first_aligned.cpp b/thirdparty/eigen/test/first_aligned.cpp index 467f9451..ed994507 100644 --- a/thirdparty/eigen/test/first_aligned.cpp +++ b/thirdparty/eigen/test/first_aligned.cpp @@ -13,7 +13,7 @@ template void test_first_aligned_helper(Scalar *array, int size) { const int packet_size = sizeof(Scalar) * internal::packet_traits::size; - VERIFY(((size_t(array) + sizeof(Scalar) * internal::first_aligned(array, size)) % packet_size) == 0); + VERIFY(((size_t(array) + sizeof(Scalar) * internal::first_default_aligned(array, size)) % packet_size) == 0); } template @@ -21,12 +21,12 @@ void test_none_aligned_helper(Scalar *array, int size) { EIGEN_UNUSED_VARIABLE(array); EIGEN_UNUSED_VARIABLE(size); - VERIFY(internal::packet_traits::size == 1 || internal::first_aligned(array, size) == size); + VERIFY(internal::packet_traits::size == 1 || internal::first_default_aligned(array, size) == size); } struct some_non_vectorizable_type { float x; }; -void test_first_aligned() +EIGEN_DECLARE_TEST(first_aligned) { EIGEN_ALIGN16 float array_float[100]; test_first_aligned_helper(array_float, 50); @@ -41,7 +41,7 @@ void test_first_aligned() test_first_aligned_helper(array_double+1, 50); test_first_aligned_helper(array_double+2, 50); - double *array_double_plus_4_bytes = (double*)(size_t(array_double)+4); + double *array_double_plus_4_bytes = (double*)(internal::UIntPtr(array_double)+4); test_none_aligned_helper(array_double_plus_4_bytes, 50); test_none_aligned_helper(array_double_plus_4_bytes+1, 50); diff --git a/thirdparty/eigen/test/geo_alignedbox.cpp b/thirdparty/eigen/test/geo_alignedbox.cpp index 84663ad1..7b1684f2 100644 --- a/thirdparty/eigen/test/geo_alignedbox.cpp +++ b/thirdparty/eigen/test/geo_alignedbox.cpp @@ -9,27 +9,33 @@ #include "main.h" #include -#include -#include -#include using namespace std; +// NOTE the following workaround was needed on some 32 bits builds to kill extra precision of x87 registers. +// It seems that it is not needed anymore, but let's keep it here, just in case... + template EIGEN_DONT_INLINE -void kill_extra_precision(T& x) { eigen_assert(&x != 0); } +void kill_extra_precision(T& /* x */) { + // This one worked but triggered a warning: + /* eigen_assert((void*)(&x) != (void*)0); */ + // An alternative could be: + /* volatile T tmp = x; */ + /* x = tmp; */ +} -template void alignedbox(const BoxType& _box) +template void alignedbox(const BoxType& box) { /* this test covers the following files: AlignedBox.h */ - typedef typename BoxType::Index Index; typedef typename BoxType::Scalar Scalar; - typedef typename NumTraits::Real RealScalar; + typedef NumTraits ScalarTraits; + typedef typename ScalarTraits::Real RealScalar; typedef Matrix VectorType; - const Index dim = _box.dim(); + const Index dim = box.dim(); VectorType p0 = VectorType::Random(dim); VectorType p1 = VectorType::Random(dim); @@ -40,7 +46,7 @@ template void alignedbox(const BoxType& _box) BoxType b0(dim); BoxType b1(VectorType::Random(dim),VectorType::Random(dim)); BoxType b2; - + kill_extra_precision(b1); kill_extra_precision(p0); kill_extra_precision(p1); @@ -48,12 +54,21 @@ template void alignedbox(const BoxType& _box) b0.extend(p0); b0.extend(p1); VERIFY(b0.contains(p0*s1+(Scalar(1)-s1)*p1)); + VERIFY(b0.contains(b0.center())); + VERIFY_IS_APPROX(b0.center(),(p0+p1)/Scalar(2)); (b2 = b0).extend(b1); VERIFY(b2.contains(b0)); VERIFY(b2.contains(b1)); VERIFY_IS_APPROX(b2.clamp(b0), b0); + // intersection + BoxType box1(VectorType::Random(dim)); + box1.extend(VectorType::Random(dim)); + BoxType box2(VectorType::Random(dim)); + box2.extend(VectorType::Random(dim)); + + VERIFY(box1.intersects(box2) == !box1.intersection(box2).isEmpty()); // alignment -- make sure there is no memory alignment assertion BoxType *bp0 = new BoxType(dim); @@ -71,17 +86,353 @@ template void alignedbox(const BoxType& _box) } +template void alignedboxTranslatable(const BoxType& box) +{ + typedef typename BoxType::Scalar Scalar; + typedef Matrix VectorType; + typedef Transform IsometryTransform; + typedef Transform AffineTransform; + + alignedbox(box); + + const VectorType Ones = VectorType::Ones(); + const VectorType UnitX = VectorType::UnitX(); + const Index dim = box.dim(); + + // box((-1, -1, -1), (1, 1, 1)) + BoxType a(-Ones, Ones); + + VERIFY_IS_APPROX(a.sizes(), Ones * Scalar(2)); + + BoxType b = a; + VectorType translate = Ones; + translate[0] = Scalar(2); + b.translate(translate); + // translate by (2, 1, 1) -> box((1, 0, 0), (3, 2, 2)) + + VERIFY_IS_APPROX(b.sizes(), Ones * Scalar(2)); + VERIFY_IS_APPROX((b.min)(), UnitX); + VERIFY_IS_APPROX((b.max)(), Ones * Scalar(2) + UnitX); + + // Test transform + + IsometryTransform tf = IsometryTransform::Identity(); + tf.translation() = -translate; + + BoxType c = b.transformed(tf); + // translate by (-2, -1, -1) -> box((-1, -1, -1), (1, 1, 1)) + VERIFY_IS_APPROX(c.sizes(), a.sizes()); + VERIFY_IS_APPROX((c.min)(), (a.min)()); + VERIFY_IS_APPROX((c.max)(), (a.max)()); + + c.transform(tf); + // translate by (-2, -1, -1) -> box((-3, -2, -2), (-1, 0, 0)) + VERIFY_IS_APPROX(c.sizes(), a.sizes()); + VERIFY_IS_APPROX((c.min)(), Ones * Scalar(-2) - UnitX); + VERIFY_IS_APPROX((c.max)(), -UnitX); + + // Scaling + + AffineTransform atf = AffineTransform::Identity(); + atf.scale(Scalar(3)); + c.transform(atf); + // scale by 3 -> box((-9, -6, -6), (-3, 0, 0)) + VERIFY_IS_APPROX(c.sizes(), Scalar(3) * a.sizes()); + VERIFY_IS_APPROX((c.min)(), Ones * Scalar(-6) - UnitX * Scalar(3)); + VERIFY_IS_APPROX((c.max)(), UnitX * Scalar(-3)); + + atf = AffineTransform::Identity(); + atf.scale(Scalar(-3)); + c.transform(atf); + // scale by -3 -> box((27, 18, 18), (9, 0, 0)) + VERIFY_IS_APPROX(c.sizes(), Scalar(9) * a.sizes()); + VERIFY_IS_APPROX((c.min)(), UnitX * Scalar(9)); + VERIFY_IS_APPROX((c.max)(), Ones * Scalar(18) + UnitX * Scalar(9)); + + // Check identity transform within numerical precision. + BoxType transformedC = c.transformed(IsometryTransform::Identity()); + VERIFY_IS_APPROX(transformedC, c); + + for (size_t i = 0; i < 10; ++i) + { + VectorType minCorner; + VectorType maxCorner; + for (Index d = 0; d < dim; ++d) + { + minCorner[d] = internal::random(-10,10); + maxCorner[d] = minCorner[d] + internal::random(0, 10); + } + + c = BoxType(minCorner, maxCorner); + + translate = VectorType::Random(); + c.translate(translate); + + VERIFY_IS_APPROX((c.min)(), minCorner + translate); + VERIFY_IS_APPROX((c.max)(), maxCorner + translate); + } +} + +template +Rotation rotate2D(Scalar angle) { + return Rotation2D(angle); +} + +template +Rotation rotate2DIntegral(typename NumTraits::NonInteger angle) { + typedef typename NumTraits::NonInteger NonInteger; + return Rotation2D(angle).toRotationMatrix(). + template cast(); +} + +template +Rotation rotate3DZAxis(Scalar angle) { + return AngleAxis(angle, Matrix(0, 0, 1)); +} + +template +Rotation rotate3DZAxisIntegral(typename NumTraits::NonInteger angle) { + typedef typename NumTraits::NonInteger NonInteger; + return AngleAxis(angle, Matrix(0, 0, 1)). + toRotationMatrix().template cast(); +} + +template +Rotation rotate4DZWAxis(Scalar angle) { + Rotation result = Matrix::Identity(); + result.block(0, 0, 3, 3) = rotate3DZAxis(angle).toRotationMatrix(); + return result; +} +template +MatrixType randomRotationMatrix() +{ + // algorithm from + // https://www.isprs-ann-photogramm-remote-sens-spatial-inf-sci.net/III-7/103/2016/isprs-annals-III-7-103-2016.pdf + const MatrixType rand = MatrixType::Random(); + const MatrixType q = rand.householderQr().householderQ(); + const JacobiSVD svd = q.jacobiSvd(ComputeFullU | ComputeFullV); + const typename MatrixType::Scalar det = (svd.matrixU() * svd.matrixV().transpose()).determinant(); + MatrixType diag = rand.Identity(); + diag(MatrixType::RowsAtCompileTime - 1, MatrixType::ColsAtCompileTime - 1) = det; + const MatrixType rotation = svd.matrixU() * diag * svd.matrixV().transpose(); + return rotation; +} + +template +Matrix boxGetCorners(const Matrix& min_, const Matrix& max_) +{ + Matrix result; + for(Index i=0; i<(1< void alignedboxRotatable( + const BoxType& box, + Rotation (*rotate)(typename NumTraits::NonInteger /*_angle*/)) +{ + alignedboxTranslatable(box); + + typedef typename BoxType::Scalar Scalar; + typedef typename NumTraits::NonInteger NonInteger; + typedef Matrix VectorType; + typedef Transform IsometryTransform; + typedef Transform AffineTransform; + + const VectorType Zero = VectorType::Zero(); + const VectorType Ones = VectorType::Ones(); + const VectorType UnitX = VectorType::UnitX(); + const VectorType UnitY = VectorType::UnitY(); + // this is vector (0, 0, -1, -1, -1, ...), i.e. with zeros at first and second dimensions + const VectorType UnitZ = Ones - UnitX - UnitY; + + // in this kind of comments the 3D case values will be illustrated + // box((-1, -1, -1), (1, 1, 1)) + BoxType a(-Ones, Ones); + + // to allow templating this test for both 2D and 3D cases, we always set all + // but the first coordinate to the same value; so basically 3D case works as + // if you were looking at the scene from top + + VectorType minPoint = -2 * Ones; + minPoint[0] = -3; + VectorType maxPoint = Zero; + maxPoint[0] = -1; + BoxType c(minPoint, maxPoint); + // box((-3, -2, -2), (-1, 0, 0)) + + IsometryTransform tf2 = IsometryTransform::Identity(); + // for some weird reason the following statement has to be put separate from + // the following rotate call, otherwise precision problems arise... + Rotation rot = rotate(NonInteger(EIGEN_PI)); + tf2.rotate(rot); + + c.transform(tf2); + // rotate by 180 deg around origin -> box((1, 0, -2), (3, 2, 0)) + + VERIFY_IS_APPROX(c.sizes(), a.sizes()); + VERIFY_IS_APPROX((c.min)(), UnitX - UnitZ * Scalar(2)); + VERIFY_IS_APPROX((c.max)(), UnitX * Scalar(3) + UnitY * Scalar(2)); + + rot = rotate(NonInteger(EIGEN_PI / 2)); + tf2.setIdentity(); + tf2.rotate(rot); + + c.transform(tf2); + // rotate by 90 deg around origin -> box((-2, 1, -2), (0, 3, 0)) + + VERIFY_IS_APPROX(c.sizes(), a.sizes()); + VERIFY_IS_APPROX((c.min)(), Ones * Scalar(-2) + UnitY * Scalar(3)); + VERIFY_IS_APPROX((c.max)(), UnitY * Scalar(3)); + + // box((-1, -1, -1), (1, 1, 1)) + AffineTransform atf = AffineTransform::Identity(); + atf.linearExt()(0, 1) = Scalar(1); + c = BoxType(-Ones, Ones); + c.transform(atf); + // 45 deg shear in x direction -> box((-2, -1, -1), (2, 1, 1)) + + VERIFY_IS_APPROX(c.sizes(), Ones * Scalar(2) + UnitX * Scalar(2)); + VERIFY_IS_APPROX((c.min)(), -Ones - UnitX); + VERIFY_IS_APPROX((c.max)(), Ones + UnitX); +} + +template void alignedboxNonIntegralRotatable( + const BoxType& box, + Rotation (*rotate)(typename NumTraits::NonInteger /*_angle*/)) +{ + alignedboxRotatable(box, rotate); + + typedef typename BoxType::Scalar Scalar; + typedef typename NumTraits::NonInteger NonInteger; + enum { Dim = BoxType::AmbientDimAtCompileTime }; + typedef Matrix VectorType; + typedef Matrix CornersType; + typedef Transform IsometryTransform; + typedef Transform AffineTransform; + + const Index dim = box.dim(); + const VectorType Zero = VectorType::Zero(); + const VectorType Ones = VectorType::Ones(); + + VectorType minPoint = -2 * Ones; + minPoint[1] = 1; + VectorType maxPoint = Zero; + maxPoint[1] = 3; + BoxType c(minPoint, maxPoint); + // ((-2, 1, -2), (0, 3, 0)) + + VectorType cornerBL = (c.min)(); + VectorType cornerTR = (c.max)(); + VectorType cornerBR = (c.min)(); cornerBR[0] = cornerTR[0]; + VectorType cornerTL = (c.max)(); cornerTL[0] = cornerBL[0]; + + NonInteger angle = NonInteger(EIGEN_PI/3); + Rotation rot = rotate(angle); + IsometryTransform tf2; + tf2.setIdentity(); + tf2.rotate(rot); + + c.transform(tf2); + // rotate by 60 deg -> box((-3.59, -1.23, -2), (-0.86, 1.5, 0)) + + cornerBL = tf2 * cornerBL; + cornerBR = tf2 * cornerBR; + cornerTL = tf2 * cornerTL; + cornerTR = tf2 * cornerTR; + + VectorType minCorner = Ones * Scalar(-2); + VectorType maxCorner = Zero; + minCorner[0] = (min)((min)(cornerBL[0], cornerBR[0]), (min)(cornerTL[0], cornerTR[0])); + maxCorner[0] = (max)((max)(cornerBL[0], cornerBR[0]), (max)(cornerTL[0], cornerTR[0])); + minCorner[1] = (min)((min)(cornerBL[1], cornerBR[1]), (min)(cornerTL[1], cornerTR[1])); + maxCorner[1] = (max)((max)(cornerBL[1], cornerBR[1]), (max)(cornerTL[1], cornerTR[1])); + + for (Index d = 2; d < dim; ++d) + VERIFY_IS_APPROX(c.sizes()[d], Scalar(2)); + + VERIFY_IS_APPROX((c.min)(), minCorner); + VERIFY_IS_APPROX((c.max)(), maxCorner); + + VectorType minCornerValue = Ones * Scalar(-2); + VectorType maxCornerValue = Zero; + minCornerValue[0] = Scalar(Scalar(-sqrt(2*2 + 3*3)) * Scalar(cos(Scalar(atan(2.0/3.0)) - angle/2))); + minCornerValue[1] = Scalar(Scalar(-sqrt(1*1 + 2*2)) * Scalar(sin(Scalar(atan(2.0/1.0)) - angle/2))); + maxCornerValue[0] = Scalar(-sin(angle)); + maxCornerValue[1] = Scalar(3 * cos(angle)); + VERIFY_IS_APPROX((c.min)(), minCornerValue); + VERIFY_IS_APPROX((c.max)(), maxCornerValue); + + // randomized test - translate and rotate the box and compare to a box made of transformed vertices + for (size_t i = 0; i < 10; ++i) + { + for (Index d = 0; d < dim; ++d) + { + minCorner[d] = internal::random(-10,10); + maxCorner[d] = minCorner[d] + internal::random(0, 10); + } + + c = BoxType(minCorner, maxCorner); + + CornersType corners = boxGetCorners(minCorner, maxCorner); + + typename AffineTransform::LinearMatrixType rotation = + randomRotationMatrix(); + + tf2.setIdentity(); + tf2.rotate(rotation); + tf2.translate(VectorType::Random()); + + c.transform(tf2); + corners = tf2 * corners; + + minCorner = corners.rowwise().minCoeff(); + maxCorner = corners.rowwise().maxCoeff(); + + VERIFY_IS_APPROX((c.min)(), minCorner); + VERIFY_IS_APPROX((c.max)(), maxCorner); + } + + // randomized test - transform the box with a random affine matrix and compare to a box made of transformed vertices + for (size_t i = 0; i < 10; ++i) + { + for (Index d = 0; d < dim; ++d) + { + minCorner[d] = internal::random(-10,10); + maxCorner[d] = minCorner[d] + internal::random(0, 10); + } + + c = BoxType(minCorner, maxCorner); + + CornersType corners = boxGetCorners(minCorner, maxCorner); + + AffineTransform atf = AffineTransform::Identity(); + atf.linearExt() = AffineTransform::LinearPart::Random(); + atf.translate(VectorType::Random()); + + c.transform(atf); + corners = atf * corners; + + minCorner = corners.rowwise().minCoeff(); + maxCorner = corners.rowwise().maxCoeff(); + + VERIFY_IS_APPROX((c.min)(), minCorner); + VERIFY_IS_APPROX((c.max)(), maxCorner); + } +} template -void alignedboxCastTests(const BoxType& _box) +void alignedboxCastTests(const BoxType& box) { - // casting - typedef typename BoxType::Index Index; + // casting typedef typename BoxType::Scalar Scalar; typedef Matrix VectorType; - const Index dim = _box.dim(); + const Index dim = box.dim(); VectorType p0 = VectorType::Random(dim); VectorType p1 = VectorType::Random(dim); @@ -153,25 +504,25 @@ void specificTest2() } -void test_geo_alignedbox() +EIGEN_DECLARE_TEST(geo_alignedbox) { for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_1( alignedbox(AlignedBox2f()) ); + CALL_SUBTEST_1( (alignedboxNonIntegralRotatable(AlignedBox2f(), &rotate2D)) ); CALL_SUBTEST_2( alignedboxCastTests(AlignedBox2f()) ); - CALL_SUBTEST_3( alignedbox(AlignedBox3f()) ); + CALL_SUBTEST_3( (alignedboxNonIntegralRotatable(AlignedBox3f(), &rotate3DZAxis)) ); CALL_SUBTEST_4( alignedboxCastTests(AlignedBox3f()) ); - CALL_SUBTEST_5( alignedbox(AlignedBox4d()) ); + CALL_SUBTEST_5( (alignedboxNonIntegralRotatable(AlignedBox4d(), &rotate4DZWAxis)) ); CALL_SUBTEST_6( alignedboxCastTests(AlignedBox4d()) ); - CALL_SUBTEST_7( alignedbox(AlignedBox1d()) ); + CALL_SUBTEST_7( alignedboxTranslatable(AlignedBox1d()) ); CALL_SUBTEST_8( alignedboxCastTests(AlignedBox1d()) ); - CALL_SUBTEST_9( alignedbox(AlignedBox1i()) ); - CALL_SUBTEST_10( alignedbox(AlignedBox2i()) ); - CALL_SUBTEST_11( alignedbox(AlignedBox3i()) ); + CALL_SUBTEST_9( alignedboxTranslatable(AlignedBox1i()) ); + CALL_SUBTEST_10( (alignedboxRotatable(AlignedBox2i(), &rotate2DIntegral)) ); + CALL_SUBTEST_11( (alignedboxRotatable(AlignedBox3i(), &rotate3DZAxisIntegral)) ); CALL_SUBTEST_14( alignedbox(AlignedBox(4)) ); } diff --git a/thirdparty/eigen/test/geo_eulerangles.cpp b/thirdparty/eigen/test/geo_eulerangles.cpp index b4830bd4..693c627a 100644 --- a/thirdparty/eigen/test/geo_eulerangles.cpp +++ b/thirdparty/eigen/test/geo_eulerangles.cpp @@ -26,16 +26,16 @@ void verify_euler(const Matrix& ea, int i, int j, int k) VERIFY_IS_APPROX(m, mbis); /* If I==K, and ea[1]==0, then there no unique solution. */ /* The remark apply in the case where I!=K, and |ea[1]| is close to pi/2. */ - if( (i!=k || ea[1]!=0) && (i==k || !internal::isApprox(abs(ea[1]),Scalar(M_PI/2),test_precision())) ) + if( (i!=k || ea[1]!=0) && (i==k || !internal::isApprox(abs(ea[1]),Scalar(EIGEN_PI/2),test_precision())) ) VERIFY((ea-eabis).norm() <= test_precision()); // approx_or_less_than does not work for 0 VERIFY(0 < eabis[0] || test_isMuchSmallerThan(eabis[0], Scalar(1))); - VERIFY_IS_APPROX_OR_LESS_THAN(eabis[0], Scalar(M_PI)); - VERIFY_IS_APPROX_OR_LESS_THAN(-Scalar(M_PI), eabis[1]); - VERIFY_IS_APPROX_OR_LESS_THAN(eabis[1], Scalar(M_PI)); - VERIFY_IS_APPROX_OR_LESS_THAN(-Scalar(M_PI), eabis[2]); - VERIFY_IS_APPROX_OR_LESS_THAN(eabis[2], Scalar(M_PI)); + VERIFY_IS_APPROX_OR_LESS_THAN(eabis[0], Scalar(EIGEN_PI)); + VERIFY_IS_APPROX_OR_LESS_THAN(-Scalar(EIGEN_PI), eabis[1]); + VERIFY_IS_APPROX_OR_LESS_THAN(eabis[1], Scalar(EIGEN_PI)); + VERIFY_IS_APPROX_OR_LESS_THAN(-Scalar(EIGEN_PI), eabis[2]); + VERIFY_IS_APPROX_OR_LESS_THAN(eabis[2], Scalar(EIGEN_PI)); } template void check_all_var(const Matrix& ea) @@ -64,7 +64,7 @@ template void eulerangles() typedef Quaternion Quaternionx; typedef AngleAxis AngleAxisx; - Scalar a = internal::random(-Scalar(M_PI), Scalar(M_PI)); + Scalar a = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)); Quaternionx q1; q1 = AngleAxisx(a, Vector3::Random().normalized()); Matrix3 m; @@ -84,13 +84,13 @@ template void eulerangles() check_all_var(ea); // Check with random angles in range [0:pi]x[-pi:pi]x[-pi:pi]. - ea = (Array3::Random() + Array3(1,0,0))*Scalar(M_PI)*Array3(0.5,1,1); + ea = (Array3::Random() + Array3(1,0,0))*Scalar(EIGEN_PI)*Array3(0.5,1,1); check_all_var(ea); - ea[2] = ea[0] = internal::random(0,Scalar(M_PI)); + ea[2] = ea[0] = internal::random(0,Scalar(EIGEN_PI)); check_all_var(ea); - ea[0] = ea[1] = internal::random(0,Scalar(M_PI)); + ea[0] = ea[1] = internal::random(0,Scalar(EIGEN_PI)); check_all_var(ea); ea[1] = 0; @@ -103,7 +103,7 @@ template void eulerangles() check_all_var(ea); } -void test_geo_eulerangles() +EIGEN_DECLARE_TEST(geo_eulerangles) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( eulerangles() ); diff --git a/thirdparty/eigen/test/geo_homogeneous.cpp b/thirdparty/eigen/test/geo_homogeneous.cpp index 01330308..9aebe622 100644 --- a/thirdparty/eigen/test/geo_homogeneous.cpp +++ b/thirdparty/eigen/test/geo_homogeneous.cpp @@ -38,6 +38,10 @@ template void homogeneous(void) hv0 << v0, 1; VERIFY_IS_APPROX(v0.homogeneous(), hv0); VERIFY_IS_APPROX(v0, hv0.hnormalized()); + + VERIFY_IS_APPROX(v0.homogeneous().sum(), hv0.sum()); + VERIFY_IS_APPROX(v0.homogeneous().minCoeff(), hv0.minCoeff()); + VERIFY_IS_APPROX(v0.homogeneous().maxCoeff(), hv0.maxCoeff()); hm0 << m0, ones.transpose(); VERIFY_IS_APPROX(m0.colwise().homogeneous(), hm0); @@ -59,7 +63,6 @@ template void homogeneous(void) VERIFY_IS_APPROX((v0.transpose().rowwise().homogeneous().eval()) * t2, v0.transpose().rowwise().homogeneous() * t2); - m0.transpose().rowwise().homogeneous().eval(); VERIFY_IS_APPROX((m0.transpose().rowwise().homogeneous().eval()) * t2, m0.transpose().rowwise().homogeneous() * t2); @@ -84,7 +87,7 @@ template void homogeneous(void) VERIFY_IS_APPROX(aff * pts.colwise().homogeneous(), (aff * pts1).colwise().hnormalized()); VERIFY_IS_APPROX(caff * pts.colwise().homogeneous(), (caff * pts1).colwise().hnormalized()); VERIFY_IS_APPROX(proj * pts.colwise().homogeneous(), (proj * pts1)); - + VERIFY_IS_APPROX((aff * pts1).colwise().hnormalized(), aff * pts); VERIFY_IS_APPROX((caff * pts1).colwise().hnormalized(), caff * pts); @@ -93,9 +96,26 @@ template void homogeneous(void) VERIFY_IS_APPROX((aff * pts2).colwise().hnormalized(), aff * pts2.colwise().hnormalized()); VERIFY_IS_APPROX((caff * pts2).colwise().hnormalized(), caff * pts2.colwise().hnormalized()); VERIFY_IS_APPROX((proj * pts2).colwise().hnormalized(), (proj * pts2.colwise().hnormalized().colwise().homogeneous()).colwise().hnormalized()); + + // Test combination of homogeneous + + VERIFY_IS_APPROX( (t2 * v0.homogeneous()).hnormalized(), + (t2.template topLeftCorner() * v0 + t2.template topRightCorner()) + / ((t2.template bottomLeftCorner<1,Size>()*v0).value() + t2(Size,Size)) ); + + VERIFY_IS_APPROX( (t2 * pts.colwise().homogeneous()).colwise().hnormalized(), + (Matrix(t2 * pts1).colwise().hnormalized()) ); + + VERIFY_IS_APPROX( (t2 .lazyProduct( v0.homogeneous() )).hnormalized(), (t2 * v0.homogeneous()).hnormalized() ); + VERIFY_IS_APPROX( (t2 .lazyProduct ( pts.colwise().homogeneous() )).colwise().hnormalized(), (t2 * pts1).colwise().hnormalized() ); + + VERIFY_IS_APPROX( (v0.transpose().homogeneous() .lazyProduct( t2 )).hnormalized(), (v0.transpose().homogeneous()*t2).hnormalized() ); + VERIFY_IS_APPROX( (pts.transpose().rowwise().homogeneous() .lazyProduct( t2 )).rowwise().hnormalized(), (pts1.transpose()*t2).rowwise().hnormalized() ); + + VERIFY_IS_APPROX( (t2.template triangularView() * v0.homogeneous()).eval(), (t2.template triangularView()*hv0) ); } -void test_geo_homogeneous() +EIGEN_DECLARE_TEST(geo_homogeneous) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1(( homogeneous() )); diff --git a/thirdparty/eigen/test/geo_hyperplane.cpp b/thirdparty/eigen/test/geo_hyperplane.cpp index 32753780..44b2f2ae 100644 --- a/thirdparty/eigen/test/geo_hyperplane.cpp +++ b/thirdparty/eigen/test/geo_hyperplane.cpp @@ -18,10 +18,11 @@ template void hyperplane(const HyperplaneType& _plane) /* this test covers the following files: Hyperplane.h */ - typedef typename HyperplaneType::Index Index; + using std::abs; const Index dim = _plane.dim(); enum { Options = HyperplaneType::Options }; typedef typename HyperplaneType::Scalar Scalar; + typedef typename HyperplaneType::RealScalar RealScalar; typedef Matrix VectorType; typedef Matrix MatrixType; @@ -42,7 +43,10 @@ template void hyperplane(const HyperplaneType& _plane) VERIFY_IS_APPROX( n1.dot(n1), Scalar(1) ); VERIFY_IS_MUCH_SMALLER_THAN( pl0.absDistance(p0), Scalar(1) ); - VERIFY_IS_APPROX( pl1.signedDistance(p1 + n1 * s0), s0 ); + if(numext::abs2(s0)>RealScalar(1e-6)) + VERIFY_IS_APPROX( pl1.signedDistance(p1 + n1 * s0), s0); + else + VERIFY_IS_MUCH_SMALLER_THAN( abs(pl1.signedDistance(p1 + n1 * s0) - s0), Scalar(1) ); VERIFY_IS_MUCH_SMALLER_THAN( pl1.signedDistance(pl1.projection(p0)), Scalar(1) ); VERIFY_IS_MUCH_SMALLER_THAN( pl1.absDistance(p1 + pl1.normal().unitOrthogonal() * s1), Scalar(1) ); @@ -52,6 +56,8 @@ template void hyperplane(const HyperplaneType& _plane) MatrixType rot = MatrixType::Random(dim,dim).householderQr().householderQ(); DiagonalMatrix scaling(VectorType::Random()); Translation translation(VectorType::Random()); + + while(scaling.diagonal().cwiseAbs().minCoeff() void hyperplane(const HyperplaneType& _plane) VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot,Isometry).absDistance(rot * p1), Scalar(1) ); pl2 = pl1; VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot*scaling).absDistance((rot*scaling) * p1), Scalar(1) ); + VERIFY_IS_APPROX( pl2.normal().norm(), RealScalar(1) ); pl2 = pl1; VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot*scaling*translation) - .absDistance((rot*scaling*translation) * p1), Scalar(1) ); + .absDistance((rot*scaling*translation) * p1), Scalar(1) ); + VERIFY_IS_APPROX( pl2.normal().norm(), RealScalar(1) ); pl2 = pl1; VERIFY_IS_MUCH_SMALLER_THAN( pl2.transform(rot*translation,Isometry) .absDistance((rot*translation) * p1), Scalar(1) ); + VERIFY_IS_APPROX( pl2.normal().norm(), RealScalar(1) ); } // casting @@ -90,9 +99,9 @@ template void lines() Vector u = Vector::Random(); Vector v = Vector::Random(); Scalar a = internal::random(); - while (abs(a-1) < 1e-4) a = internal::random(); - while (u.norm() < 1e-4) u = Vector::Random(); - while (v.norm() < 1e-4) v = Vector::Random(); + while (abs(a-1) < Scalar(1e-4)) a = internal::random(); + while (u.norm() < Scalar(1e-4)) u = Vector::Random(); + while (v.norm() < Scalar(1e-4)) v = Vector::Random(); HLine line_u = HLine::Through(center + u, center + a*u); HLine line_v = HLine::Through(center + v, center + a*v); @@ -104,12 +113,15 @@ template void lines() Vector result = line_u.intersection(line_v); // the lines should intersect at the point we called "center" - VERIFY_IS_APPROX(result, center); + if(abs(a-1) > Scalar(1e-2) && abs(v.normalized().dot(u.normalized())) void hyperplane_alignment() typedef Hyperplane Plane3a; typedef Hyperplane Plane3u; - EIGEN_ALIGN16 Scalar array1[4]; - EIGEN_ALIGN16 Scalar array2[4]; - EIGEN_ALIGN16 Scalar array3[4+1]; + EIGEN_ALIGN_MAX Scalar array1[4]; + EIGEN_ALIGN_MAX Scalar array2[4]; + EIGEN_ALIGN_MAX Scalar array3[4+1]; Scalar* array3u = array3+1; Plane3a *p1 = ::new(reinterpret_cast(array1)) Plane3a; @@ -160,15 +172,10 @@ template void hyperplane_alignment() VERIFY_IS_APPROX(p1->coeffs(), p2->coeffs()); VERIFY_IS_APPROX(p1->coeffs(), p3->coeffs()); - - #if defined(EIGEN_VECTORIZE) && EIGEN_ALIGN_STATICALLY - if(internal::packet_traits::Vectorizable) - VERIFY_RAISES_ASSERT((::new(reinterpret_cast(array3u)) Plane3a)); - #endif } -void test_geo_hyperplane() +EIGEN_DECLARE_TEST(geo_hyperplane) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( hyperplane(Hyperplane()) ); diff --git a/thirdparty/eigen/test/geo_orthomethods.cpp b/thirdparty/eigen/test/geo_orthomethods.cpp index c836dae4..5f7ddb91 100644 --- a/thirdparty/eigen/test/geo_orthomethods.cpp +++ b/thirdparty/eigen/test/geo_orthomethods.cpp @@ -33,12 +33,16 @@ template void orthomethods_3() VERIFY_IS_MUCH_SMALLER_THAN(v1.dot(v1.cross(v2)), Scalar(1)); VERIFY_IS_MUCH_SMALLER_THAN(v1.cross(v2).dot(v2), Scalar(1)); VERIFY_IS_MUCH_SMALLER_THAN(v2.dot(v1.cross(v2)), Scalar(1)); + VERIFY_IS_MUCH_SMALLER_THAN(v1.cross(Vector3::Random()).dot(v1), Scalar(1)); Matrix3 mat3; mat3 << v0.normalized(), (v0.cross(v1)).normalized(), (v0.cross(v1).cross(v0)).normalized(); VERIFY(mat3.isUnitary()); - + + mat3.setRandom(); + VERIFY_IS_APPROX(v0.cross(mat3*v1), -(mat3*v1).cross(v0)); + VERIFY_IS_APPROX(v0.cross(mat3.lazyProduct(v1)), -(mat3.lazyProduct(v1)).cross(v0)); // colwise/rowwise cross product mat3.setRandom(); @@ -47,6 +51,13 @@ template void orthomethods_3() int i = internal::random(0,2); mcross = mat3.colwise().cross(vec3); VERIFY_IS_APPROX(mcross.col(i), mat3.col(i).cross(vec3)); + + VERIFY_IS_MUCH_SMALLER_THAN((mat3.adjoint() * mat3.colwise().cross(vec3)).diagonal().cwiseAbs().sum(), Scalar(1)); + VERIFY_IS_MUCH_SMALLER_THAN((mat3.adjoint() * mat3.colwise().cross(Vector3::Random())).diagonal().cwiseAbs().sum(), Scalar(1)); + + VERIFY_IS_MUCH_SMALLER_THAN((vec3.adjoint() * mat3.colwise().cross(vec3)).cwiseAbs().sum(), Scalar(1)); + VERIFY_IS_MUCH_SMALLER_THAN((vec3.adjoint() * Matrix3::Random().colwise().cross(vec3)).cwiseAbs().sum(), Scalar(1)); + mcross = mat3.rowwise().cross(vec3); VERIFY_IS_APPROX(mcross.row(i), mat3.row(i).cross(vec3)); @@ -57,12 +68,14 @@ template void orthomethods_3() v40.w() = v41.w() = v42.w() = 0; v42.template head<3>() = v40.template head<3>().cross(v41.template head<3>()); VERIFY_IS_APPROX(v40.cross3(v41), v42); + VERIFY_IS_MUCH_SMALLER_THAN(v40.cross3(Vector4::Random()).dot(v40), Scalar(1)); // check mixed product typedef Matrix RealVector3; RealVector3 rv1 = RealVector3::Random(); - VERIFY_IS_APPROX(v1.cross(rv1.template cast()), v1.cross(rv1)); - VERIFY_IS_APPROX(rv1.template cast().cross(v1), rv1.cross(v1)); + v2 = rv1.template cast(); + VERIFY_IS_APPROX(v1.cross(v2), v1.cross(rv1)); + VERIFY_IS_APPROX(v2.cross(v1), rv1.cross(v1)); } template void orthomethods(int size=Size) @@ -103,7 +116,7 @@ template void orthomethods(int size=Size) VERIFY_IS_APPROX(mcrossN3.row(i), matN3.row(i).cross(vec3)); } -void test_geo_orthomethods() +EIGEN_DECLARE_TEST(geo_orthomethods) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( orthomethods_3() ); diff --git a/thirdparty/eigen/test/geo_parametrizedline.cpp b/thirdparty/eigen/test/geo_parametrizedline.cpp index f0462d40..e4b194ab 100644 --- a/thirdparty/eigen/test/geo_parametrizedline.cpp +++ b/thirdparty/eigen/test/geo_parametrizedline.cpp @@ -19,12 +19,13 @@ template void parametrizedline(const LineType& _line) ParametrizedLine.h */ using std::abs; - typedef typename LineType::Index Index; const Index dim = _line.dim(); typedef typename LineType::Scalar Scalar; typedef typename NumTraits::Real RealScalar; typedef Matrix VectorType; typedef Hyperplane HyperplaneType; + typedef Matrix MatrixType; VectorType p0 = VectorType::Random(dim); VectorType p1 = VectorType::Random(dim); @@ -59,6 +60,31 @@ template void parametrizedline(const LineType& _line) VERIFY_IS_MUCH_SMALLER_THAN(hp.signedDistance(pi), RealScalar(1)); VERIFY_IS_MUCH_SMALLER_THAN(l0.distance(pi), RealScalar(1)); VERIFY_IS_APPROX(l0.intersectionPoint(hp), pi); + + // transform + if (!NumTraits::IsComplex) + { + MatrixType rot = MatrixType::Random(dim,dim).householderQr().householderQ(); + DiagonalMatrix scaling(VectorType::Random()); + Translation translation(VectorType::Random()); + + while(scaling.diagonal().cwiseAbs().minCoeff() void parametrizedline_alignment() @@ -66,9 +92,9 @@ template void parametrizedline_alignment() typedef ParametrizedLine Line4a; typedef ParametrizedLine Line4u; - EIGEN_ALIGN16 Scalar array1[8]; - EIGEN_ALIGN16 Scalar array2[8]; - EIGEN_ALIGN16 Scalar array3[8+1]; + EIGEN_ALIGN_MAX Scalar array1[16]; + EIGEN_ALIGN_MAX Scalar array2[16]; + EIGEN_ALIGN_MAX Scalar array3[16+1]; Scalar* array3u = array3+1; Line4a *p1 = ::new(reinterpret_cast(array1)) Line4a; @@ -84,14 +110,9 @@ template void parametrizedline_alignment() VERIFY_IS_APPROX(p1->origin(), p3->origin()); VERIFY_IS_APPROX(p1->direction(), p2->direction()); VERIFY_IS_APPROX(p1->direction(), p3->direction()); - - #if defined(EIGEN_VECTORIZE) && EIGEN_ALIGN_STATICALLY - if(internal::packet_traits::Vectorizable) - VERIFY_RAISES_ASSERT((::new(reinterpret_cast(array3u)) Line4a)); - #endif } -void test_geo_parametrizedline() +EIGEN_DECLARE_TEST(geo_parametrizedline) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( parametrizedline(ParametrizedLine()) ); diff --git a/thirdparty/eigen/test/geo_quaternion.cpp b/thirdparty/eigen/test/geo_quaternion.cpp index 1694b32c..c561fc89 100644 --- a/thirdparty/eigen/test/geo_quaternion.cpp +++ b/thirdparty/eigen/test/geo_quaternion.cpp @@ -12,6 +12,7 @@ #include #include #include +#include "AnnoyingScalar.h" template T bounded_acos(T v) { @@ -30,8 +31,8 @@ template void check_slerp(const QuatType& q0, const QuatType& Scalar largeEps = test_precision(); Scalar theta_tot = AA(q1*q0.inverse()).angle(); - if(theta_tot>M_PI) - theta_tot = Scalar(2.*M_PI)-theta_tot; + if(theta_tot>Scalar(EIGEN_PI)) + theta_tot = Scalar(2.)*Scalar(EIGEN_PI)-theta_tot; for(Scalar t=0; t<=Scalar(1.001); t+=Scalar(0.1)) { QuatType q = q0.slerp(t,q1); @@ -49,13 +50,13 @@ template void quaternion(void) */ using std::abs; typedef Matrix Vector3; - typedef Matrix Vector4; + typedef Matrix Matrix3; typedef Quaternion Quaternionx; typedef AngleAxis AngleAxisx; Scalar largeEps = test_precision(); if (internal::is_same::value) - largeEps = 1e-3f; + largeEps = Scalar(1e-3); Scalar eps = internal::random() * Scalar(1e-2); @@ -64,8 +65,8 @@ template void quaternion(void) v2 = Vector3::Random(), v3 = Vector3::Random(); - Scalar a = internal::random(-Scalar(M_PI), Scalar(M_PI)), - b = internal::random(-Scalar(M_PI), Scalar(M_PI)); + Scalar a = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)), + b = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)); // Quaternion: Identity(), setIdentity(); Quaternionx q1, q2; @@ -74,6 +75,13 @@ template void quaternion(void) q1.coeffs().setRandom(); VERIFY_IS_APPROX(q1.coeffs(), (q1*q2).coeffs()); +#ifndef EIGEN_NO_IO + // Printing + std::ostringstream ss; + ss << q2; + VERIFY(ss.str() == "0i + 0j + 0k + 1"); +#endif + // concatenation q1 *= q2; @@ -82,10 +90,10 @@ template void quaternion(void) // angular distance Scalar refangle = abs(AngleAxisx(q1.inverse()*q2).angle()); - if (refangle>Scalar(M_PI)) - refangle = Scalar(2)*Scalar(M_PI) - refangle; + if (refangle>Scalar(EIGEN_PI)) + refangle = Scalar(2)*Scalar(EIGEN_PI) - refangle; - if((q1.coeffs()-q2.coeffs()).norm() > 10*largeEps) + if((q1.coeffs()-q2.coeffs()).norm() > Scalar(10)*largeEps) { VERIFY_IS_MUCH_SMALLER_THAN(abs(q1.angularDistance(q2) - refangle), Scalar(1)); } @@ -101,6 +109,11 @@ template void quaternion(void) q2 = q1.toRotationMatrix(); VERIFY_IS_APPROX(q1*v1,q2*v1); + Matrix3 rot1(q1); + VERIFY_IS_APPROX(q1*v1,rot1*v1); + Quaternionx q3(rot1.transpose()*rot1); + VERIFY_IS_APPROX(q3*v1,v1); + // angle-axis conversion AngleAxisx aa = AngleAxisx(q1); @@ -108,9 +121,9 @@ template void quaternion(void) // Do not execute the test if the rotation angle is almost zero, or // the rotation axis and v1 are almost parallel. - if (abs(aa.angle()) > 5*test_precision() - && (aa.axis() - v1.normalized()).norm() < 1.99 - && (aa.axis() + v1.normalized()).norm() < 1.99) + if (abs(aa.angle()) > Scalar(5)*test_precision() + && (aa.axis() - v1.normalized()).norm() < Scalar(1.99) + && (aa.axis() + v1.normalized()).norm() < Scalar(1.99)) { VERIFY_IS_NOT_APPROX(q1 * v1, Quaternionx(AngleAxisx(aa.angle()*2,aa.axis())) * v1); } @@ -151,19 +164,19 @@ template void quaternion(void) Quaternionx *q = new Quaternionx; delete q; - q1 = AngleAxisx(a, v0.normalized()); - q2 = AngleAxisx(b, v1.normalized()); + q1 = Quaternionx::UnitRandom(); + q2 = Quaternionx::UnitRandom(); check_slerp(q1,q2); q1 = AngleAxisx(b, v1.normalized()); - q2 = AngleAxisx(b+Scalar(M_PI), v1.normalized()); + q2 = AngleAxisx(b+Scalar(EIGEN_PI), v1.normalized()); check_slerp(q1,q2); q1 = AngleAxisx(b, v1.normalized()); q2 = AngleAxisx(-b, -v1.normalized()); check_slerp(q1,q2); - q1.coeffs() = Vector4::Random().normalized(); + q1 = Quaternionx::UnitRandom(); q2.coeffs() = -q1.coeffs(); check_slerp(q1,q2); } @@ -179,11 +192,11 @@ template void mapQuaternion(void){ Vector3 v0 = Vector3::Random(), v1 = Vector3::Random(); - Scalar a = internal::random(-Scalar(M_PI), Scalar(M_PI)); + Scalar a = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)); - EIGEN_ALIGN16 Scalar array1[4]; - EIGEN_ALIGN16 Scalar array2[4]; - EIGEN_ALIGN16 Scalar array3[4+1]; + EIGEN_ALIGN_MAX Scalar array1[4]; + EIGEN_ALIGN_MAX Scalar array2[4]; + EIGEN_ALIGN_MAX Scalar array3[4+1]; Scalar* array3unaligned = array3+1; MQuaternionA mq1(array1); @@ -205,10 +218,6 @@ template void mapQuaternion(void){ VERIFY_IS_APPROX(q1.coeffs(), q2.coeffs()); VERIFY_IS_APPROX(q1.coeffs(), q3.coeffs()); VERIFY_IS_APPROX(q4.coeffs(), q3.coeffs()); - #ifdef EIGEN_VECTORIZE - if(internal::packet_traits::Vectorizable) - VERIFY_RAISES_ASSERT((MQuaternionA(array3unaligned))); - #endif VERIFY_IS_APPROX(mq1 * (mq1.inverse() * v1), v1); VERIFY_IS_APPROX(mq1 * (mq1.conjugate() * v1), v1); @@ -226,15 +235,36 @@ template void mapQuaternion(void){ VERIFY_IS_APPROX(mq3*mq2, q3*q2); VERIFY_IS_APPROX(mcq1*mq2, q1*q2); VERIFY_IS_APPROX(mcq3*mq2, q3*q2); + + // Bug 1461, compilation issue with Map::w(), and other reference/constness checks: + VERIFY_IS_APPROX(mcq3.coeffs().x() + mcq3.coeffs().y() + mcq3.coeffs().z() + mcq3.coeffs().w(), mcq3.coeffs().sum()); + VERIFY_IS_APPROX(mcq3.x() + mcq3.y() + mcq3.z() + mcq3.w(), mcq3.coeffs().sum()); + mq3.w() = 1; + const Quaternionx& cq3(q3); + VERIFY( &cq3.x() == &q3.x() ); + const MQuaternionUA& cmq3(mq3); + VERIFY( &cmq3.x() == &mq3.x() ); + // FIXME the following should be ok. The problem is that currently the LValueBit flag + // is used to determine whether we can return a coeff by reference or not, which is not enough for Map. + //const MCQuaternionUA& cmcq3(mcq3); + //VERIFY( &cmcq3.x() == &mcq3.x() ); + + // test cast + { + Quaternion q1f = mq1.template cast(); + VERIFY_IS_APPROX(q1f.template cast(),mq1); + Quaternion q1d = mq1.template cast(); + VERIFY_IS_APPROX(q1d.template cast(),mq1); + } } template void quaternionAlignment(void){ typedef Quaternion QuaternionA; typedef Quaternion QuaternionUA; - EIGEN_ALIGN16 Scalar array1[4]; - EIGEN_ALIGN16 Scalar array2[4]; - EIGEN_ALIGN16 Scalar array3[4+1]; + EIGEN_ALIGN_MAX Scalar array1[4]; + EIGEN_ALIGN_MAX Scalar array2[4]; + EIGEN_ALIGN_MAX Scalar array3[4+1]; Scalar* arrayunaligned = array3+1; QuaternionA *q1 = ::new(reinterpret_cast(array1)) QuaternionA; @@ -247,10 +277,6 @@ template void quaternionAlignment(void){ VERIFY_IS_APPROX(q1->coeffs(), q2->coeffs()); VERIFY_IS_APPROX(q1->coeffs(), q3->coeffs()); - #if defined(EIGEN_VECTORIZE) && EIGEN_ALIGN_STATICALLY - if(internal::packet_traits::Vectorizable) - VERIFY_RAISES_ASSERT((::new(reinterpret_cast(arrayunaligned)) QuaternionA)); - #endif } template void check_const_correctness(const PlainObjectType&) @@ -267,18 +293,40 @@ template void check_const_correctness(const PlainObjec VERIFY( !(Map::Flags & LvalueBit) ); } -void test_geo_quaternion() +#if EIGEN_HAS_RVALUE_REFERENCES + +// Regression for bug 1573 +struct MovableClass { + // The following line is a workaround for gcc 4.7 and 4.8 (see bug 1573 comments). + static_assert(std::is_nothrow_move_constructible::value,""); + MovableClass() = default; + MovableClass(const MovableClass&) = default; + MovableClass(MovableClass&&) noexcept = default; + MovableClass& operator=(const MovableClass&) = default; + MovableClass& operator=(MovableClass&&) = default; + Quaternionf m_quat; +}; + +#endif + +EIGEN_DECLARE_TEST(geo_quaternion) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1(( quaternion() )); CALL_SUBTEST_1( check_const_correctness(Quaternionf()) ); + CALL_SUBTEST_1(( quaternion() )); + CALL_SUBTEST_1(( quaternionAlignment() )); + CALL_SUBTEST_1( mapQuaternion() ); + CALL_SUBTEST_2(( quaternion() )); CALL_SUBTEST_2( check_const_correctness(Quaterniond()) ); - CALL_SUBTEST_3(( quaternion() )); - CALL_SUBTEST_4(( quaternion() )); - CALL_SUBTEST_5(( quaternionAlignment() )); - CALL_SUBTEST_6(( quaternionAlignment() )); - CALL_SUBTEST_1( mapQuaternion() ); + CALL_SUBTEST_2(( quaternion() )); + CALL_SUBTEST_2(( quaternionAlignment() )); CALL_SUBTEST_2( mapQuaternion() ); + +#ifndef EIGEN_TEST_ANNOYING_SCALAR_DONT_THROW + AnnoyingScalar::dont_throw = true; +#endif + CALL_SUBTEST_3(( quaternion() )); } } diff --git a/thirdparty/eigen/test/geo_transformations.cpp b/thirdparty/eigen/test/geo_transformations.cpp index 383c42ba..ed31c304 100644 --- a/thirdparty/eigen/test/geo_transformations.cpp +++ b/thirdparty/eigen/test/geo_transformations.cpp @@ -12,6 +12,17 @@ #include #include +template +Matrix angleToVec(T a) +{ + return Matrix(std::cos(a), std::sin(a)); +} + +// This permits to workaround a bug in clang/llvm code generation. +template +EIGEN_DONT_INLINE +void dont_over_optimize(T& x) { volatile typename T::Scalar tmp = x(0); x(0) = tmp; } + template void non_projective_only() { /* this test covers the following files: @@ -29,7 +40,7 @@ template void non_projective_only() Transform3 t0, t1, t2; - Scalar a = internal::random(-Scalar(M_PI), Scalar(M_PI)); + Scalar a = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)); Quaternionx q1, q2; @@ -97,16 +108,14 @@ template void transformations() v1 = Vector3::Random(); Matrix3 matrot1, m; - Scalar a = internal::random(-Scalar(M_PI), Scalar(M_PI)); - Scalar s0 = internal::random(), - s1 = internal::random(); + Scalar a = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)); + Scalar s0 = internal::random(), s1 = internal::random(); while(v0.norm() < test_precision()) v0 = Vector3::Random(); while(v1.norm() < test_precision()) v1 = Vector3::Random(); - VERIFY_IS_APPROX(v0, AngleAxisx(a, v0.normalized()) * v0); - VERIFY_IS_APPROX(-v0, AngleAxisx(Scalar(M_PI), v0.unitOrthogonal()) * v0); + VERIFY_IS_APPROX(-v0, AngleAxisx(Scalar(EIGEN_PI), v0.unitOrthogonal()) * v0); if(abs(cos(a)) > test_precision()) { VERIFY_IS_APPROX(cos(a)*v0.squaredNorm(), v0.dot(AngleAxisx(a, v0.unitOrthogonal()) * v0)); @@ -132,14 +141,16 @@ template void transformations() AngleAxisx aa = AngleAxisx(q1); VERIFY_IS_APPROX(q1 * v1, Quaternionx(aa) * v1); - if(abs(aa.angle()) > NumTraits::dummy_precision()) + // The following test is stable only if 2*angle != angle and v1 is not colinear with axis + if( (abs(aa.angle()) > test_precision()) && (abs(aa.axis().dot(v1.normalized()))<(Scalar(1)-Scalar(4)*test_precision())) ) { VERIFY( !(q1 * v1).isApprox(Quaternionx(AngleAxisx(aa.angle()*2,aa.axis())) * v1) ); } aa.fromRotationMatrix(aa.toRotationMatrix()); VERIFY_IS_APPROX(q1 * v1, Quaternionx(aa) * v1); - if(abs(aa.angle()) > NumTraits::dummy_precision()) + // The following test is stable only if 2*angle != angle and v1 is not colinear with axis + if( (abs(aa.angle()) > test_precision()) && (abs(aa.axis().dot(v1.normalized()))<(Scalar(1)-Scalar(4)*test_precision())) ) { VERIFY( !(q1 * v1).isApprox(Quaternionx(AngleAxisx(aa.angle()*2,aa.axis())) * v1) ); } @@ -158,7 +169,7 @@ template void transformations() // TODO complete the tests ! a = 0; while (abs(a)(-Scalar(0.4)*Scalar(M_PI), Scalar(0.4)*Scalar(M_PI)); + a = internal::random(-Scalar(0.4)*Scalar(EIGEN_PI), Scalar(0.4)*Scalar(EIGEN_PI)); q1 = AngleAxisx(a, v0.normalized()); Transform3 t0, t1, t2; @@ -204,7 +215,7 @@ template void transformations() tmat4.matrix()(3,3) = Scalar(1); VERIFY_IS_APPROX(tmat3.matrix(), tmat4.matrix()); - Scalar a3 = internal::random(-Scalar(M_PI), Scalar(M_PI)); + Scalar a3 = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)); Vector3 v3 = Vector3::Random().normalized(); AngleAxisx aa3(a3, v3); Transform3 t3(aa3); @@ -216,12 +227,15 @@ template void transformations() t4 *= aa3; VERIFY_IS_APPROX(t3.matrix(), t4.matrix()); - v3 = Vector3::Random(); + do { + v3 = Vector3::Random(); + dont_over_optimize(v3); + } while (v3.cwiseAbs().minCoeff()::epsilon()); Translation3 tv3(v3); Transform3 t5(tv3); t4 = tv3; VERIFY_IS_APPROX(t5.matrix(), t4.matrix()); - t4.translate(-v3); + t4.translate((-v3).eval()); VERIFY_IS_APPROX(t4.matrix(), MatrixType::Identity()); t4 *= tv3; VERIFY_IS_APPROX(t5.matrix(), t4.matrix()); @@ -262,6 +276,12 @@ template void transformations() VERIFY( (t20.fromPositionOrientationScale(v20,a,v21) * (t21.prescale(v21.cwiseInverse()).translate(-v20))).matrix().isIdentity(test_precision()) ); + t20.setIdentity(); + t20.shear(Scalar(2), Scalar(3)); + Transform2 t23 = t20 * t21; + t21.preshear(Scalar(2), Scalar(3)); + VERIFY_IS_APPROX(t21, t23); + // Transform - new API // 3D t0.setIdentity(); @@ -413,12 +433,28 @@ template void transformations() VERIFY_IS_APPROX(r2d1f.template cast(),r2d1); Rotation2D r2d1d = r2d1.template cast(); VERIFY_IS_APPROX(r2d1d.template cast(),r2d1); - - t20 = Translation2(v20) * (Rotation2D(s0) * Eigen::Scaling(s0)); - t21 = Translation2(v20) * Rotation2D(s0) * Eigen::Scaling(s0); - VERIFY_IS_APPROX(t20,t21); + for(int k=0; k<100; ++k) + { + Scalar angle = internal::random(-100,100); + Rotation2D rot2(angle); + VERIFY( rot2.smallestPositiveAngle() >= 0 ); + VERIFY( rot2.smallestPositiveAngle() <= Scalar(2)*Scalar(EIGEN_PI) ); + VERIFY_IS_APPROX( angleToVec(rot2.smallestPositiveAngle()), angleToVec(rot2.angle()) ); + + VERIFY( rot2.smallestAngle() >= -Scalar(EIGEN_PI) ); + VERIFY( rot2.smallestAngle() <= Scalar(EIGEN_PI) ); + VERIFY_IS_APPROX( angleToVec(rot2.smallestAngle()), angleToVec(rot2.angle()) ); + + Matrix rot2_as_mat(rot2); + Rotation2D rot3(rot2_as_mat); + VERIFY_IS_APPROX( angleToVec(rot2.smallestAngle()), angleToVec(rot3.angle()) ); + } + + s0 = internal::random(-100,100); + s1 = internal::random(-100,100); Rotation2D R0(s0), R1(s1); + t20 = Translation2(v20) * (R0 * Eigen::Scaling(s0)); t21 = Translation2(v20) * R0 * Eigen::Scaling(s0); VERIFY_IS_APPROX(t20,t21); @@ -428,9 +464,24 @@ template void transformations() VERIFY_IS_APPROX(t20,t21); VERIFY_IS_APPROX(s0, (R0.slerp(0, R1)).angle()); - VERIFY_IS_APPROX(s1, (R0.slerp(1, R1)).angle()); - VERIFY_IS_APPROX(s0, (R0.slerp(0.5, R0)).angle()); - VERIFY_IS_APPROX(Scalar(0), (R0.slerp(0.5, R0.inverse())).angle()); + VERIFY_IS_APPROX( angleToVec(R1.smallestPositiveAngle()), angleToVec((R0.slerp(1, R1)).smallestPositiveAngle()) ); + VERIFY_IS_APPROX(R0.smallestPositiveAngle(), (R0.slerp(0.5, R0)).smallestPositiveAngle()); + + if(std::cos(s0)>0) + VERIFY_IS_MUCH_SMALLER_THAN((R0.slerp(0.5, R0.inverse())).smallestAngle(), Scalar(1)); + else + VERIFY_IS_APPROX(Scalar(EIGEN_PI), (R0.slerp(0.5, R0.inverse())).smallestPositiveAngle()); + + // Check path length + Scalar l = 0; + int path_steps = 100; + for(int k=0; k::epsilon()*Scalar(path_steps/2))); // check basic features { @@ -520,9 +571,9 @@ template void transform_alignment() typedef Transform Projective3a; typedef Transform Projective3u; - EIGEN_ALIGN16 Scalar array1[16]; - EIGEN_ALIGN16 Scalar array2[16]; - EIGEN_ALIGN16 Scalar array3[16+1]; + EIGEN_ALIGN_MAX Scalar array1[16]; + EIGEN_ALIGN_MAX Scalar array2[16]; + EIGEN_ALIGN_MAX Scalar array3[16+1]; Scalar* array3u = array3+1; Projective3a *p1 = ::new(reinterpret_cast(array1)) Projective3a; @@ -537,11 +588,6 @@ template void transform_alignment() VERIFY_IS_APPROX(p1->matrix(), p3->matrix()); VERIFY_IS_APPROX( (*p1) * (*p1), (*p2)*(*p3)); - - #if defined(EIGEN_VECTORIZE) && EIGEN_ALIGN_STATICALLY - if(internal::packet_traits::Vectorizable) - VERIFY_RAISES_ASSERT((::new(reinterpret_cast(array3u)) Projective3a)); - #endif } template void transform_products() @@ -567,11 +613,99 @@ template void transform_products() VERIFY_IS_APPROX((ac*p).matrix(), a_m*p_m); } -void test_geo_transformations() +template void transformations_no_scale() +{ + /* this test covers the following files: + Cross.h Quaternion.h, Transform.h + */ + typedef Matrix Vector3; + typedef Matrix Vector4; + typedef Quaternion Quaternionx; + typedef AngleAxis AngleAxisx; + typedef Transform Transform3; + typedef Translation Translation3; + typedef Matrix Matrix4; + + Vector3 v0 = Vector3::Random(), + v1 = Vector3::Random(); + + Transform3 t0, t1, t2; + + Scalar a = internal::random(-Scalar(EIGEN_PI), Scalar(EIGEN_PI)); + + Quaternionx q1, q2; + + q1 = AngleAxisx(a, v0.normalized()); + + t0 = Transform3::Identity(); + VERIFY_IS_APPROX(t0.matrix(), Transform3::MatrixType::Identity()); + + t0.setIdentity(); + t1.setIdentity(); + v1 = Vector3::Ones(); + t0.linear() = q1.toRotationMatrix(); + t0.pretranslate(v0); + t1.linear() = q1.conjugate().toRotationMatrix(); + t1.translate(-v0); + + VERIFY((t0 * t1).matrix().isIdentity(test_precision())); + + t1.fromPositionOrientationScale(v0, q1, v1); + VERIFY_IS_APPROX(t1.matrix(), t0.matrix()); + VERIFY_IS_APPROX(t1*v1, t0*v1); + + // translation * vector + t0.setIdentity(); + t0.translate(v0); + VERIFY_IS_APPROX((t0 * v1).template head<3>(), Translation3(v0) * v1); + + // Conversion to matrix. + Transform3 t3; + t3.linear() = q1.toRotationMatrix(); + t3.translation() = v1; + Matrix4 m3 = t3.matrix(); + VERIFY((m3 * m3.inverse()).isIdentity(test_precision())); + // Verify implicit last row is initialized. + VERIFY_IS_APPROX(Vector4(m3.row(3)), Vector4(0.0, 0.0, 0.0, 1.0)); + + VERIFY_IS_APPROX(t3.rotation(), t3.linear()); + if(Mode==Isometry) + VERIFY(t3.rotation().data()==t3.linear().data()); +} + +template void transformations_computed_scaling_continuity() +{ + typedef Matrix Vector3; + typedef Transform Transform3; + typedef Matrix Matrix3; + + // Given: two transforms that differ by '2*eps'. + Scalar eps(1e-3); + Vector3 v0 = Vector3::Random().normalized(), + v1 = Vector3::Random().normalized(), + v3 = Vector3::Random().normalized(); + Transform3 t0, t1; + // The interesting case is when their determinants have different signs. + Matrix3 rank2 = 50 * v0 * v0.adjoint() + 20 * v1 * v1.adjoint(); + t0.linear() = rank2 + eps * v3 * v3.adjoint(); + t1.linear() = rank2 - eps * v3 * v3.adjoint(); + + // When: computing the rotation-scaling parts + Matrix3 r0, s0, r1, s1; + t0.computeRotationScaling(&r0, &s0); + t1.computeRotationScaling(&r1, &s1); + + // Then: the scaling parts should differ by no more than '2*eps'. + const Scalar c(2.1); // 2 + room for rounding errors + VERIFY((s0 - s1).norm() < c * eps); +} + +EIGEN_DECLARE_TEST(geo_transformations) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1(( transformations() )); CALL_SUBTEST_1(( non_projective_only() )); + CALL_SUBTEST_1(( transformations_computed_scaling_continuity() )); CALL_SUBTEST_2(( transformations() )); CALL_SUBTEST_2(( non_projective_only() )); @@ -580,7 +714,7 @@ void test_geo_transformations() CALL_SUBTEST_3(( transformations() )); CALL_SUBTEST_3(( transformations() )); CALL_SUBTEST_3(( transform_alignment() )); - + CALL_SUBTEST_4(( transformations() )); CALL_SUBTEST_4(( non_projective_only() )); @@ -594,7 +728,10 @@ void test_geo_transformations() CALL_SUBTEST_7(( transform_products() )); CALL_SUBTEST_7(( transform_products() )); - CALL_SUBTEST_8(( transform_associativity(Rotation2D(internal::random()*double(3.14))) )); - CALL_SUBTEST_8(( transform_associativity(Quaterniond(Vector4d::Random().normalized())) )); + CALL_SUBTEST_8(( transform_associativity(Rotation2D(internal::random()*double(EIGEN_PI))) )); + CALL_SUBTEST_8(( transform_associativity(Quaterniond::UnitRandom()) )); + + CALL_SUBTEST_9(( transformations_no_scale() )); + CALL_SUBTEST_9(( transformations_no_scale() )); } } diff --git a/thirdparty/eigen/test/gpu_basic.cu b/thirdparty/eigen/test/gpu_basic.cu new file mode 100644 index 00000000..e424a93c --- /dev/null +++ b/thirdparty/eigen/test/gpu_basic.cu @@ -0,0 +1,465 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2015-2016 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +// workaround issue between gcc >= 4.7 and cuda 5.5 +#if (defined __GNUC__) && (__GNUC__>4 || __GNUC_MINOR__>=7) + #undef _GLIBCXX_ATOMIC_BUILTINS + #undef _GLIBCXX_USE_INT128 +#endif + +#define EIGEN_TEST_NO_LONGDOUBLE +#define EIGEN_DEFAULT_DENSE_INDEX_TYPE int + +#include "main.h" +#include "gpu_common.h" + +// Check that dense modules can be properly parsed by nvcc +#include + +// struct Foo{ +// EIGEN_DEVICE_FUNC +// void operator()(int i, const float* mats, float* vecs) const { +// using namespace Eigen; +// // Matrix3f M(data); +// // Vector3f x(data+9); +// // Map(data+9) = M.inverse() * x; +// Matrix3f M(mats+i/16); +// Vector3f x(vecs+i*3); +// // using std::min; +// // using std::sqrt; +// Map(vecs+i*3) << x.minCoeff(), 1, 2;// / x.dot(x);//(M.inverse() * x) / x.x(); +// //x = x*2 + x.y() * x + x * x.maxCoeff() - x / x.sum(); +// } +// }; + +template +struct coeff_wise { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + T x1(in+i); + T x2(in+i+1); + T x3(in+i+2); + Map res(out+i*T::MaxSizeAtCompileTime); + + res.array() += (in[0] * x1 + x2).array() * x3.array(); + } +}; + +template +struct complex_sqrt { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + typedef typename T::Scalar ComplexType; + typedef typename T::Scalar::value_type ValueType; + const int num_special_inputs = 18; + + if (i == 0) { + const ValueType nan = std::numeric_limits::quiet_NaN(); + typedef Eigen::Vector SpecialInputs; + SpecialInputs special_in; + special_in.setZero(); + int idx = 0; + special_in[idx++] = ComplexType(0, 0); + special_in[idx++] = ComplexType(-0, 0); + special_in[idx++] = ComplexType(0, -0); + special_in[idx++] = ComplexType(-0, -0); + // GCC's fallback sqrt implementation fails for inf inputs. + // It is called when _GLIBCXX_USE_C99_COMPLEX is false or if + // clang includes the GCC header (which temporarily disables + // _GLIBCXX_USE_C99_COMPLEX) + #if !defined(_GLIBCXX_COMPLEX) || \ + (_GLIBCXX_USE_C99_COMPLEX && !defined(__CLANG_CUDA_WRAPPERS_COMPLEX)) + const ValueType inf = std::numeric_limits::infinity(); + special_in[idx++] = ComplexType(1.0, inf); + special_in[idx++] = ComplexType(nan, inf); + special_in[idx++] = ComplexType(1.0, -inf); + special_in[idx++] = ComplexType(nan, -inf); + special_in[idx++] = ComplexType(-inf, 1.0); + special_in[idx++] = ComplexType(inf, 1.0); + special_in[idx++] = ComplexType(-inf, -1.0); + special_in[idx++] = ComplexType(inf, -1.0); + special_in[idx++] = ComplexType(-inf, nan); + special_in[idx++] = ComplexType(inf, nan); + #endif + special_in[idx++] = ComplexType(1.0, nan); + special_in[idx++] = ComplexType(nan, 1.0); + special_in[idx++] = ComplexType(nan, -1.0); + special_in[idx++] = ComplexType(nan, nan); + + Map special_out(out); + special_out = special_in.cwiseSqrt(); + } + + T x1(in + i); + Map res(out + num_special_inputs + i*T::MaxSizeAtCompileTime); + res = x1.cwiseSqrt(); + } +}; + +template +struct complex_operators { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + typedef typename T::Scalar ComplexType; + typedef typename T::Scalar::value_type ValueType; + const int num_scalar_operators = 24; + const int num_vector_operators = 23; // no unary + operator. + int out_idx = i * (num_scalar_operators + num_vector_operators * T::MaxSizeAtCompileTime); + + // Scalar operators. + const ComplexType a = in[i]; + const ComplexType b = in[i + 1]; + + out[out_idx++] = +a; + out[out_idx++] = -a; + + out[out_idx++] = a + b; + out[out_idx++] = a + numext::real(b); + out[out_idx++] = numext::real(a) + b; + out[out_idx++] = a - b; + out[out_idx++] = a - numext::real(b); + out[out_idx++] = numext::real(a) - b; + out[out_idx++] = a * b; + out[out_idx++] = a * numext::real(b); + out[out_idx++] = numext::real(a) * b; + out[out_idx++] = a / b; + out[out_idx++] = a / numext::real(b); + out[out_idx++] = numext::real(a) / b; + +#if !defined(EIGEN_COMP_MSVC) + out[out_idx] = a; out[out_idx++] += b; + out[out_idx] = a; out[out_idx++] -= b; + out[out_idx] = a; out[out_idx++] *= b; + out[out_idx] = a; out[out_idx++] /= b; +#endif + + const ComplexType true_value = ComplexType(ValueType(1), ValueType(0)); + const ComplexType false_value = ComplexType(ValueType(0), ValueType(0)); + out[out_idx++] = (a == b ? true_value : false_value); + out[out_idx++] = (a == numext::real(b) ? true_value : false_value); + out[out_idx++] = (numext::real(a) == b ? true_value : false_value); + out[out_idx++] = (a != b ? true_value : false_value); + out[out_idx++] = (a != numext::real(b) ? true_value : false_value); + out[out_idx++] = (numext::real(a) != b ? true_value : false_value); + + // Vector versions. + T x1(in + i); + T x2(in + i + 1); + const int res_size = T::MaxSizeAtCompileTime * num_scalar_operators; + const int size = T::MaxSizeAtCompileTime; + int block_idx = 0; + + Map> res(out + out_idx, res_size); + res.segment(block_idx, size) = -x1; + block_idx += size; + + res.segment(block_idx, size) = x1 + x2; + block_idx += size; + res.segment(block_idx, size) = x1 + x2.real(); + block_idx += size; + res.segment(block_idx, size) = x1.real() + x2; + block_idx += size; + res.segment(block_idx, size) = x1 - x2; + block_idx += size; + res.segment(block_idx, size) = x1 - x2.real(); + block_idx += size; + res.segment(block_idx, size) = x1.real() - x2; + block_idx += size; + res.segment(block_idx, size) = x1.array() * x2.array(); + block_idx += size; + res.segment(block_idx, size) = x1.array() * x2.real().array(); + block_idx += size; + res.segment(block_idx, size) = x1.real().array() * x2.array(); + block_idx += size; + res.segment(block_idx, size) = x1.array() / x2.array(); + block_idx += size; + res.segment(block_idx, size) = x1.array() / x2.real().array(); + block_idx += size; + res.segment(block_idx, size) = x1.real().array() / x2.array(); + block_idx += size; + +#if !defined(EIGEN_COMP_MSVC) + res.segment(block_idx, size) = x1; res.segment(block_idx, size) += x2; + block_idx += size; + res.segment(block_idx, size) = x1; res.segment(block_idx, size) -= x2; + block_idx += size; + res.segment(block_idx, size) = x1; res.segment(block_idx, size).array() *= x2.array(); + block_idx += size; + res.segment(block_idx, size) = x1; res.segment(block_idx, size).array() /= x2.array(); + block_idx += size; +#endif + + const T true_vector = T::Constant(true_value); + const T false_vector = T::Constant(false_value); + res.segment(block_idx, size) = (x1 == x2 ? true_vector : false_vector); + block_idx += size; + // Mixing types in equality comparison does not work. + // res.segment(block_idx, size) = (x1 == x2.real() ? true_vector : false_vector); + // block_idx += size; + // res.segment(block_idx, size) = (x1.real() == x2 ? true_vector : false_vector); + // block_idx += size; + res.segment(block_idx, size) = (x1 != x2 ? true_vector : false_vector); + block_idx += size; + // res.segment(block_idx, size) = (x1 != x2.real() ? true_vector : false_vector); + // block_idx += size; + // res.segment(block_idx, size) = (x1.real() != x2 ? true_vector : false_vector); + // block_idx += size; + } +}; + +template +struct replicate { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + T x1(in+i); + int step = x1.size() * 4; + int stride = 3 * step; + + typedef Map > MapType; + MapType(out+i*stride+0*step, x1.rows()*2, x1.cols()*2) = x1.replicate(2,2); + MapType(out+i*stride+1*step, x1.rows()*3, x1.cols()) = in[i] * x1.colwise().replicate(3); + MapType(out+i*stride+2*step, x1.rows(), x1.cols()*3) = in[i] * x1.rowwise().replicate(3); + } +}; + +template +struct alloc_new_delete { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + int offset = 2*i*T::MaxSizeAtCompileTime; + T* x = new T(in + offset); + Eigen::Map u(out + offset); + u = *x; + delete x; + + offset += T::MaxSizeAtCompileTime; + T* y = new T[1]; + y[0] = T(in + offset); + Eigen::Map v(out + offset); + v = y[0]; + delete[] y; + } +}; + +template +struct redux { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + int N = 10; + T x1(in+i); + out[i*N+0] = x1.minCoeff(); + out[i*N+1] = x1.maxCoeff(); + out[i*N+2] = x1.sum(); + out[i*N+3] = x1.prod(); + out[i*N+4] = x1.matrix().squaredNorm(); + out[i*N+5] = x1.matrix().norm(); + out[i*N+6] = x1.colwise().sum().maxCoeff(); + out[i*N+7] = x1.rowwise().maxCoeff().sum(); + out[i*N+8] = x1.matrix().colwise().squaredNorm().sum(); + } +}; + +template +struct prod_test { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T1::Scalar* in, typename T1::Scalar* out) const + { + using namespace Eigen; + typedef Matrix T3; + T1 x1(in+i); + T2 x2(in+i+1); + Map res(out+i*T3::MaxSizeAtCompileTime); + res += in[i] * x1 * x2; + } +}; + +template +struct diagonal { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T1::Scalar* in, typename T1::Scalar* out) const + { + using namespace Eigen; + T1 x1(in+i); + Map res(out+i*T2::MaxSizeAtCompileTime); + res += x1.diagonal(); + } +}; + +template +struct eigenvalues_direct { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + typedef Matrix Vec; + T M(in+i); + Map res(out+i*Vec::MaxSizeAtCompileTime); + T A = M*M.adjoint(); + SelfAdjointEigenSolver eig; + eig.computeDirect(A); + res = eig.eigenvalues(); + } +}; + +template +struct eigenvalues { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + typedef Matrix Vec; + T M(in+i); + Map res(out+i*Vec::MaxSizeAtCompileTime); + T A = M*M.adjoint(); + SelfAdjointEigenSolver eig; + eig.compute(A); + res = eig.eigenvalues(); + } +}; + +template +struct matrix_inverse { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + using namespace Eigen; + T M(in+i); + Map res(out+i*T::MaxSizeAtCompileTime); + res = M.inverse(); + } +}; + +template +struct numeric_limits_test { + EIGEN_DEVICE_FUNC + void operator()(int i, const typename T::Scalar* in, typename T::Scalar* out) const + { + EIGEN_UNUSED_VARIABLE(in) + int out_idx = i * 5; + out[out_idx++] = numext::numeric_limits::epsilon(); + out[out_idx++] = (numext::numeric_limits::max)(); + out[out_idx++] = (numext::numeric_limits::min)(); + out[out_idx++] = numext::numeric_limits::infinity(); + out[out_idx++] = numext::numeric_limits::quiet_NaN(); + } +}; + +template +bool verifyIsApproxWithInfsNans(const Type1& a, const Type2& b, typename Type1::Scalar* = 0) // Enabled for Eigen's type only +{ + if (a.rows() != b.rows()) { + return false; + } + if (a.cols() != b.cols()) { + return false; + } + for (Index r = 0; r < a.rows(); ++r) { + for (Index c = 0; c < a.cols(); ++c) { + if (a(r, c) != b(r, c) + && !((numext::isnan)(a(r, c)) && (numext::isnan)(b(r, c))) + && !test_isApprox(a(r, c), b(r, c))) { + return false; + } + } + } + return true; +} + +template +void test_with_infs_nans(const Kernel& ker, int n, const Input& in, Output& out) +{ + Output out_ref, out_gpu; + #if !defined(EIGEN_GPU_COMPILE_PHASE) + out_ref = out_gpu = out; + #else + EIGEN_UNUSED_VARIABLE(in); + EIGEN_UNUSED_VARIABLE(out); + #endif + run_on_cpu (ker, n, in, out_ref); + run_on_gpu(ker, n, in, out_gpu); + #if !defined(EIGEN_GPU_COMPILE_PHASE) + verifyIsApproxWithInfsNans(out_ref, out_gpu); + #endif +} + +EIGEN_DECLARE_TEST(gpu_basic) +{ + ei_test_init_gpu(); + + int nthreads = 100; + Eigen::VectorXf in, out; + Eigen::VectorXcf cfin, cfout; + + #if !defined(EIGEN_GPU_COMPILE_PHASE) + int data_size = nthreads * 512; + in.setRandom(data_size); + out.setConstant(data_size, -1); + cfin.setRandom(data_size); + cfout.setConstant(data_size, -1); + #endif + + CALL_SUBTEST( run_and_compare_to_gpu(coeff_wise(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(coeff_wise(), nthreads, in, out) ); + +#if !defined(EIGEN_USE_HIP) + // FIXME + // These subtests result in a compile failure on the HIP platform + // + // eigen-upstream/Eigen/src/Core/Replicate.h:61:65: error: + // base class 'internal::dense_xpr_base, -1, -1> >::type' + // (aka 'ArrayBase, -1, -1> >') has protected default constructor + CALL_SUBTEST( run_and_compare_to_gpu(replicate(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(replicate(), nthreads, in, out) ); + + // HIP does not support new/delete on device. + CALL_SUBTEST( run_and_compare_to_gpu(alloc_new_delete(), nthreads, in, out) ); +#endif + + CALL_SUBTEST( run_and_compare_to_gpu(redux(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(redux(), nthreads, in, out) ); + + CALL_SUBTEST( run_and_compare_to_gpu(prod_test(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(prod_test(), nthreads, in, out) ); + + CALL_SUBTEST( run_and_compare_to_gpu(diagonal(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(diagonal(), nthreads, in, out) ); + + CALL_SUBTEST( run_and_compare_to_gpu(matrix_inverse(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(matrix_inverse(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(matrix_inverse(), nthreads, in, out) ); + + CALL_SUBTEST( run_and_compare_to_gpu(eigenvalues_direct(), nthreads, in, out) ); + CALL_SUBTEST( run_and_compare_to_gpu(eigenvalues_direct(), nthreads, in, out) ); + + // Test std::complex. + CALL_SUBTEST( run_and_compare_to_gpu(complex_operators(), nthreads, cfin, cfout) ); + CALL_SUBTEST( test_with_infs_nans(complex_sqrt(), nthreads, cfin, cfout) ); + + // numeric_limits + CALL_SUBTEST( test_with_infs_nans(numeric_limits_test(), 1, in, out) ); + +#if defined(__NVCC__) + // FIXME + // These subtests compiles only with nvcc and fail with HIPCC and clang-cuda + CALL_SUBTEST( run_and_compare_to_gpu(eigenvalues(), nthreads, in, out) ); + typedef Matrix Matrix6f; + CALL_SUBTEST( run_and_compare_to_gpu(eigenvalues(), nthreads, in, out) ); +#endif +} diff --git a/thirdparty/eigen/test/gpu_common.h b/thirdparty/eigen/test/gpu_common.h new file mode 100644 index 00000000..c37eaa13 --- /dev/null +++ b/thirdparty/eigen/test/gpu_common.h @@ -0,0 +1,176 @@ +#ifndef EIGEN_TEST_GPU_COMMON_H +#define EIGEN_TEST_GPU_COMMON_H + +#ifdef EIGEN_USE_HIP + #include + #include +#else + #include + #include + #include +#endif + +#include + +#define EIGEN_USE_GPU +#include + +#if !defined(__CUDACC__) && !defined(__HIPCC__) +dim3 threadIdx, blockDim, blockIdx; +#endif + +template +void run_on_cpu(const Kernel& ker, int n, const Input& in, Output& out) +{ + for(int i=0; i +__global__ +EIGEN_HIP_LAUNCH_BOUNDS_1024 +void run_on_gpu_meta_kernel(const Kernel ker, int n, const Input* in, Output* out) +{ + int i = threadIdx.x + blockIdx.x*blockDim.x; + if(i +void run_on_gpu(const Kernel& ker, int n, const Input& in, Output& out) +{ + typename Input::Scalar* d_in; + typename Output::Scalar* d_out; + std::ptrdiff_t in_bytes = in.size() * sizeof(typename Input::Scalar); + std::ptrdiff_t out_bytes = out.size() * sizeof(typename Output::Scalar); + + gpuMalloc((void**)(&d_in), in_bytes); + gpuMalloc((void**)(&d_out), out_bytes); + + gpuMemcpy(d_in, in.data(), in_bytes, gpuMemcpyHostToDevice); + gpuMemcpy(d_out, out.data(), out_bytes, gpuMemcpyHostToDevice); + + // Simple and non-optimal 1D mapping assuming n is not too large + // That's only for unit testing! + dim3 Blocks(128); + dim3 Grids( (n+int(Blocks.x)-1)/int(Blocks.x) ); + + gpuDeviceSynchronize(); + +#ifdef EIGEN_USE_HIP + hipLaunchKernelGGL(HIP_KERNEL_NAME(run_on_gpu_meta_kernel::type, + typename std::decay::type>), + dim3(Grids), dim3(Blocks), 0, 0, ker, n, d_in, d_out); +#else + run_on_gpu_meta_kernel<<>>(ker, n, d_in, d_out); +#endif + // Pre-launch errors. + gpuError_t err = gpuGetLastError(); + if (err != gpuSuccess) { + printf("%s: %s\n", gpuGetErrorName(err), gpuGetErrorString(err)); + gpu_assert(false); + } + + // Kernel execution errors. + err = gpuDeviceSynchronize(); + if (err != gpuSuccess) { + printf("%s: %s\n", gpuGetErrorName(err), gpuGetErrorString(err)); + gpu_assert(false); + } + + + // check inputs have not been modified + gpuMemcpy(const_cast(in.data()), d_in, in_bytes, gpuMemcpyDeviceToHost); + gpuMemcpy(out.data(), d_out, out_bytes, gpuMemcpyDeviceToHost); + + gpuFree(d_in); + gpuFree(d_out); +} + + +template +void run_and_compare_to_gpu(const Kernel& ker, int n, const Input& in, Output& out) +{ + Input in_ref, in_gpu; + Output out_ref, out_gpu; + #if !defined(EIGEN_GPU_COMPILE_PHASE) + in_ref = in_gpu = in; + out_ref = out_gpu = out; + #else + EIGEN_UNUSED_VARIABLE(in); + EIGEN_UNUSED_VARIABLE(out); + #endif + run_on_cpu (ker, n, in_ref, out_ref); + run_on_gpu(ker, n, in_gpu, out_gpu); + #if !defined(EIGEN_GPU_COMPILE_PHASE) + VERIFY_IS_APPROX(in_ref, in_gpu); + VERIFY_IS_APPROX(out_ref, out_gpu); + #endif +} + +struct compile_time_device_info { + EIGEN_DEVICE_FUNC + void operator()(int i, const int* /*in*/, int* info) const + { + if (i == 0) { + EIGEN_UNUSED_VARIABLE(info) + #if defined(__CUDA_ARCH__) + info[0] = int(__CUDA_ARCH__ +0); + #endif + #if defined(EIGEN_HIP_DEVICE_COMPILE) + info[1] = int(EIGEN_HIP_DEVICE_COMPILE +0); + #endif + } + } +}; + +void ei_test_init_gpu() +{ + int device = 0; + gpuDeviceProp_t deviceProp; + gpuGetDeviceProperties(&deviceProp, device); + + ArrayXi dummy(1), info(10); + info = -1; + run_on_gpu(compile_time_device_info(),10,dummy,info); + + + std::cout << "GPU compile-time info:\n"; + + #ifdef EIGEN_CUDACC + std::cout << " EIGEN_CUDACC: " << int(EIGEN_CUDACC) << "\n"; + #endif + + #ifdef EIGEN_CUDA_SDK_VER + std::cout << " EIGEN_CUDA_SDK_VER: " << int(EIGEN_CUDA_SDK_VER) << "\n"; + #endif + + #ifdef EIGEN_COMP_NVCC + std::cout << " EIGEN_COMP_NVCC: " << int(EIGEN_COMP_NVCC) << "\n"; + #endif + + #ifdef EIGEN_HIPCC + std::cout << " EIGEN_HIPCC: " << int(EIGEN_HIPCC) << "\n"; + #endif + + std::cout << " EIGEN_CUDA_ARCH: " << info[0] << "\n"; + std::cout << " EIGEN_HIP_DEVICE_COMPILE: " << info[1] << "\n"; + + std::cout << "GPU device info:\n"; + std::cout << " name: " << deviceProp.name << "\n"; + std::cout << " capability: " << deviceProp.major << "." << deviceProp.minor << "\n"; + std::cout << " multiProcessorCount: " << deviceProp.multiProcessorCount << "\n"; + std::cout << " maxThreadsPerMultiProcessor: " << deviceProp.maxThreadsPerMultiProcessor << "\n"; + std::cout << " warpSize: " << deviceProp.warpSize << "\n"; + std::cout << " regsPerBlock: " << deviceProp.regsPerBlock << "\n"; + std::cout << " concurrentKernels: " << deviceProp.concurrentKernels << "\n"; + std::cout << " clockRate: " << deviceProp.clockRate << "\n"; + std::cout << " canMapHostMemory: " << deviceProp.canMapHostMemory << "\n"; + std::cout << " computeMode: " << deviceProp.computeMode << "\n"; +} + +#endif // EIGEN_TEST_GPU_COMMON_H diff --git a/thirdparty/eigen/test/half_float.cpp b/thirdparty/eigen/test/half_float.cpp new file mode 100644 index 00000000..ffb3215b --- /dev/null +++ b/thirdparty/eigen/test/half_float.cpp @@ -0,0 +1,352 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include + +#include "main.h" + +#include + +#define VERIFY_HALF_BITS_EQUAL(h, bits) \ + VERIFY_IS_EQUAL((numext::bit_cast(h)), (static_cast(bits))) + +// Make sure it's possible to forward declare Eigen::half +namespace Eigen { +struct half; +} + +using Eigen::half; + +void test_conversion() +{ + using Eigen::half_impl::__half_raw; + + // Round-trip bit-cast with uint16. + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(half(1.0f))), + half(1.0f)); + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(half(0.5f))), + half(0.5f)); + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(half(-0.33333f))), + half(-0.33333f)); + VERIFY_IS_EQUAL( + numext::bit_cast(numext::bit_cast(half(0.0f))), + half(0.0f)); + + // Conversion from float. + VERIFY_HALF_BITS_EQUAL(half(1.0f), 0x3c00); + VERIFY_HALF_BITS_EQUAL(half(0.5f), 0x3800); + VERIFY_HALF_BITS_EQUAL(half(0.33333f), 0x3555); + VERIFY_HALF_BITS_EQUAL(half(0.0f), 0x0000); + VERIFY_HALF_BITS_EQUAL(half(-0.0f), 0x8000); + VERIFY_HALF_BITS_EQUAL(half(65504.0f), 0x7bff); + VERIFY_HALF_BITS_EQUAL(half(65536.0f), 0x7c00); // Becomes infinity. + + // Denormals. + VERIFY_HALF_BITS_EQUAL(half(-5.96046e-08f), 0x8001); + VERIFY_HALF_BITS_EQUAL(half(5.96046e-08f), 0x0001); + VERIFY_HALF_BITS_EQUAL(half(1.19209e-07f), 0x0002); + + // Verify round-to-nearest-even behavior. + float val1 = float(half(__half_raw(0x3c00))); + float val2 = float(half(__half_raw(0x3c01))); + float val3 = float(half(__half_raw(0x3c02))); + VERIFY_HALF_BITS_EQUAL(half(0.5f * (val1 + val2)), 0x3c00); + VERIFY_HALF_BITS_EQUAL(half(0.5f * (val2 + val3)), 0x3c02); + + // Conversion from int. + VERIFY_HALF_BITS_EQUAL(half(-1), 0xbc00); + VERIFY_HALF_BITS_EQUAL(half(0), 0x0000); + VERIFY_HALF_BITS_EQUAL(half(1), 0x3c00); + VERIFY_HALF_BITS_EQUAL(half(2), 0x4000); + VERIFY_HALF_BITS_EQUAL(half(3), 0x4200); + + // Conversion from bool. + VERIFY_HALF_BITS_EQUAL(half(false), 0x0000); + VERIFY_HALF_BITS_EQUAL(half(true), 0x3c00); + + // Conversion to float. + VERIFY_IS_EQUAL(float(half(__half_raw(0x0000))), 0.0f); + VERIFY_IS_EQUAL(float(half(__half_raw(0x3c00))), 1.0f); + + // Denormals. + VERIFY_IS_APPROX(float(half(__half_raw(0x8001))), -5.96046e-08f); + VERIFY_IS_APPROX(float(half(__half_raw(0x0001))), 5.96046e-08f); + VERIFY_IS_APPROX(float(half(__half_raw(0x0002))), 1.19209e-07f); + + // NaNs and infinities. + VERIFY(!(numext::isinf)(float(half(65504.0f)))); // Largest finite number. + VERIFY(!(numext::isnan)(float(half(0.0f)))); + VERIFY((numext::isinf)(float(half(__half_raw(0xfc00))))); + VERIFY((numext::isnan)(float(half(__half_raw(0xfc01))))); + VERIFY((numext::isinf)(float(half(__half_raw(0x7c00))))); + VERIFY((numext::isnan)(float(half(__half_raw(0x7c01))))); + +#if !EIGEN_COMP_MSVC + // Visual Studio errors out on divisions by 0 + VERIFY((numext::isnan)(float(half(0.0 / 0.0)))); + VERIFY((numext::isinf)(float(half(1.0 / 0.0)))); + VERIFY((numext::isinf)(float(half(-1.0 / 0.0)))); +#endif + + // Exactly same checks as above, just directly on the half representation. + VERIFY(!(numext::isinf)(half(__half_raw(0x7bff)))); + VERIFY(!(numext::isnan)(half(__half_raw(0x0000)))); + VERIFY((numext::isinf)(half(__half_raw(0xfc00)))); + VERIFY((numext::isnan)(half(__half_raw(0xfc01)))); + VERIFY((numext::isinf)(half(__half_raw(0x7c00)))); + VERIFY((numext::isnan)(half(__half_raw(0x7c01)))); + +#if !EIGEN_COMP_MSVC + // Visual Studio errors out on divisions by 0 + VERIFY((numext::isnan)(half(0.0 / 0.0))); + VERIFY((numext::isinf)(half(1.0 / 0.0))); + VERIFY((numext::isinf)(half(-1.0 / 0.0))); +#endif + + // Conversion to bool + VERIFY(!static_cast(half(0.0))); + VERIFY(!static_cast(half(-0.0))); + VERIFY(static_cast(half(__half_raw(0x7bff)))); + VERIFY(static_cast(half(-0.33333))); + VERIFY(static_cast(half(1.0))); + VERIFY(static_cast(half(-1.0))); + VERIFY(static_cast(half(-5.96046e-08f))); +} + +void test_numtraits() +{ + std::cout << "epsilon = " << NumTraits::epsilon() << " (0x" << std::hex << numext::bit_cast(NumTraits::epsilon()) << ")" << std::endl; + std::cout << "highest = " << NumTraits::highest() << " (0x" << std::hex << numext::bit_cast(NumTraits::highest()) << ")" << std::endl; + std::cout << "lowest = " << NumTraits::lowest() << " (0x" << std::hex << numext::bit_cast(NumTraits::lowest()) << ")" << std::endl; + std::cout << "min = " << (std::numeric_limits::min)() << " (0x" << std::hex << numext::bit_cast(half((std::numeric_limits::min)())) << ")" << std::endl; + std::cout << "denorm min = " << (std::numeric_limits::denorm_min)() << " (0x" << std::hex << numext::bit_cast(half((std::numeric_limits::denorm_min)())) << ")" << std::endl; + std::cout << "infinity = " << NumTraits::infinity() << " (0x" << std::hex << numext::bit_cast(NumTraits::infinity()) << ")" << std::endl; + std::cout << "quiet nan = " << NumTraits::quiet_NaN() << " (0x" << std::hex << numext::bit_cast(NumTraits::quiet_NaN()) << ")" << std::endl; + std::cout << "signaling nan = " << std::numeric_limits::signaling_NaN() << " (0x" << std::hex << numext::bit_cast(std::numeric_limits::signaling_NaN()) << ")" << std::endl; + + VERIFY(NumTraits::IsSigned); + + VERIFY_IS_EQUAL( + numext::bit_cast(std::numeric_limits::infinity()), + numext::bit_cast(half(std::numeric_limits::infinity())) ); + // There is no guarantee that casting a 32-bit NaN to 16-bit has a precise + // bit pattern. We test that it is in fact a NaN, then test the signaling + // bit (msb of significand is 1 for quiet, 0 for signaling). + const numext::uint16_t HALF_QUIET_BIT = 0x0200; + VERIFY( + (numext::isnan)(std::numeric_limits::quiet_NaN()) + && (numext::isnan)(half(std::numeric_limits::quiet_NaN())) + && ((numext::bit_cast(std::numeric_limits::quiet_NaN()) & HALF_QUIET_BIT) > 0) + && ((numext::bit_cast(half(std::numeric_limits::quiet_NaN())) & HALF_QUIET_BIT) > 0) ); + // After a cast to half, a signaling NaN may become non-signaling + // (e.g. in the case of casting float to native __fp16). Thus, we check that + // both are NaN, and that only the `numeric_limits` version is signaling. + VERIFY( + (numext::isnan)(std::numeric_limits::signaling_NaN()) + && (numext::isnan)(half(std::numeric_limits::signaling_NaN())) + && ((numext::bit_cast(std::numeric_limits::signaling_NaN()) & HALF_QUIET_BIT) == 0) ); + + VERIFY( (std::numeric_limits::min)() > half(0.f) ); + VERIFY( (std::numeric_limits::denorm_min)() > half(0.f) ); + VERIFY( (std::numeric_limits::min)()/half(2) > half(0.f) ); + VERIFY_IS_EQUAL( (std::numeric_limits::denorm_min)()/half(2), half(0.f) ); +} + +void test_arithmetic() +{ + VERIFY_IS_EQUAL(float(half(2) + half(2)), 4); + VERIFY_IS_EQUAL(float(half(2) + half(-2)), 0); + VERIFY_IS_APPROX(float(half(0.33333f) + half(0.66667f)), 1.0f); + VERIFY_IS_EQUAL(float(half(2.0f) * half(-5.5f)), -11.0f); + VERIFY_IS_APPROX(float(half(1.0f) / half(3.0f)), 0.33333f); + VERIFY_IS_EQUAL(float(-half(4096.0f)), -4096.0f); + VERIFY_IS_EQUAL(float(-half(-4096.0f)), 4096.0f); + + half x(3); + half y = ++x; + VERIFY_IS_EQUAL(x, half(4)); + VERIFY_IS_EQUAL(y, half(4)); + y = --x; + VERIFY_IS_EQUAL(x, half(3)); + VERIFY_IS_EQUAL(y, half(3)); + y = x++; + VERIFY_IS_EQUAL(x, half(4)); + VERIFY_IS_EQUAL(y, half(3)); + y = x--; + VERIFY_IS_EQUAL(x, half(3)); + VERIFY_IS_EQUAL(y, half(4)); +} + +void test_comparison() +{ + VERIFY(half(1.0f) > half(0.5f)); + VERIFY(half(0.5f) < half(1.0f)); + VERIFY(!(half(1.0f) < half(0.5f))); + VERIFY(!(half(0.5f) > half(1.0f))); + + VERIFY(!(half(4.0f) > half(4.0f))); + VERIFY(!(half(4.0f) < half(4.0f))); + + VERIFY(!(half(0.0f) < half(-0.0f))); + VERIFY(!(half(-0.0f) < half(0.0f))); + VERIFY(!(half(0.0f) > half(-0.0f))); + VERIFY(!(half(-0.0f) > half(0.0f))); + + VERIFY(half(0.2f) > half(-1.0f)); + VERIFY(half(-1.0f) < half(0.2f)); + VERIFY(half(-16.0f) < half(-15.0f)); + + VERIFY(half(1.0f) == half(1.0f)); + VERIFY(half(1.0f) != half(2.0f)); + + // Comparisons with NaNs and infinities. +#if !EIGEN_COMP_MSVC + // Visual Studio errors out on divisions by 0 + VERIFY(!(half(0.0 / 0.0) == half(0.0 / 0.0))); + VERIFY(half(0.0 / 0.0) != half(0.0 / 0.0)); + + VERIFY(!(half(1.0) == half(0.0 / 0.0))); + VERIFY(!(half(1.0) < half(0.0 / 0.0))); + VERIFY(!(half(1.0) > half(0.0 / 0.0))); + VERIFY(half(1.0) != half(0.0 / 0.0)); + + VERIFY(half(1.0) < half(1.0 / 0.0)); + VERIFY(half(1.0) > half(-1.0 / 0.0)); +#endif +} + +void test_basic_functions() +{ + const float PI = static_cast(EIGEN_PI); + + VERIFY_IS_EQUAL(float(numext::abs(half(3.5f))), 3.5f); + VERIFY_IS_EQUAL(float(abs(half(3.5f))), 3.5f); + VERIFY_IS_EQUAL(float(numext::abs(half(-3.5f))), 3.5f); + VERIFY_IS_EQUAL(float(abs(half(-3.5f))), 3.5f); + + VERIFY_IS_EQUAL(float(numext::floor(half(3.5f))), 3.0f); + VERIFY_IS_EQUAL(float(floor(half(3.5f))), 3.0f); + VERIFY_IS_EQUAL(float(numext::floor(half(-3.5f))), -4.0f); + VERIFY_IS_EQUAL(float(floor(half(-3.5f))), -4.0f); + + VERIFY_IS_EQUAL(float(numext::ceil(half(3.5f))), 4.0f); + VERIFY_IS_EQUAL(float(ceil(half(3.5f))), 4.0f); + VERIFY_IS_EQUAL(float(numext::ceil(half(-3.5f))), -3.0f); + VERIFY_IS_EQUAL(float(ceil(half(-3.5f))), -3.0f); + + VERIFY_IS_APPROX(float(numext::sqrt(half(0.0f))), 0.0f); + VERIFY_IS_APPROX(float(sqrt(half(0.0f))), 0.0f); + VERIFY_IS_APPROX(float(numext::sqrt(half(4.0f))), 2.0f); + VERIFY_IS_APPROX(float(sqrt(half(4.0f))), 2.0f); + + VERIFY_IS_APPROX(float(numext::pow(half(0.0f), half(1.0f))), 0.0f); + VERIFY_IS_APPROX(float(pow(half(0.0f), half(1.0f))), 0.0f); + VERIFY_IS_APPROX(float(numext::pow(half(2.0f), half(2.0f))), 4.0f); + VERIFY_IS_APPROX(float(pow(half(2.0f), half(2.0f))), 4.0f); + + VERIFY_IS_EQUAL(float(numext::exp(half(0.0f))), 1.0f); + VERIFY_IS_EQUAL(float(exp(half(0.0f))), 1.0f); + VERIFY_IS_APPROX(float(numext::exp(half(PI))), 20.f + PI); + VERIFY_IS_APPROX(float(exp(half(PI))), 20.f + PI); + + VERIFY_IS_EQUAL(float(numext::expm1(half(0.0f))), 0.0f); + VERIFY_IS_EQUAL(float(expm1(half(0.0f))), 0.0f); + VERIFY_IS_APPROX(float(numext::expm1(half(2.0f))), 6.3890561f); + VERIFY_IS_APPROX(float(expm1(half(2.0f))), 6.3890561f); + + VERIFY_IS_EQUAL(float(numext::log(half(1.0f))), 0.0f); + VERIFY_IS_EQUAL(float(log(half(1.0f))), 0.0f); + VERIFY_IS_APPROX(float(numext::log(half(10.0f))), 2.30273f); + VERIFY_IS_APPROX(float(log(half(10.0f))), 2.30273f); + + VERIFY_IS_EQUAL(float(numext::log1p(half(0.0f))), 0.0f); + VERIFY_IS_EQUAL(float(log1p(half(0.0f))), 0.0f); + VERIFY_IS_APPROX(float(numext::log1p(half(10.0f))), 2.3978953f); + VERIFY_IS_APPROX(float(log1p(half(10.0f))), 2.3978953f); + + VERIFY_IS_APPROX(numext::fmod(half(5.3f), half(2.0f)), half(1.3f)); + VERIFY_IS_APPROX(fmod(half(5.3f), half(2.0f)), half(1.3f)); + VERIFY_IS_APPROX(numext::fmod(half(-18.5f), half(-4.2f)), half(-1.7f)); + VERIFY_IS_APPROX(fmod(half(-18.5f), half(-4.2f)), half(-1.7f)); +} + +void test_trigonometric_functions() +{ + const float PI = static_cast(EIGEN_PI); + VERIFY_IS_APPROX(numext::cos(half(0.0f)), half(cosf(0.0f))); + VERIFY_IS_APPROX(cos(half(0.0f)), half(cosf(0.0f))); + VERIFY_IS_APPROX(numext::cos(half(PI)), half(cosf(PI))); + // VERIFY_IS_APPROX(numext::cos(half(PI/2)), half(cosf(PI/2))); + // VERIFY_IS_APPROX(numext::cos(half(3*PI/2)), half(cosf(3*PI/2))); + VERIFY_IS_APPROX(numext::cos(half(3.5f)), half(cosf(3.5f))); + + VERIFY_IS_APPROX(numext::sin(half(0.0f)), half(sinf(0.0f))); + VERIFY_IS_APPROX(sin(half(0.0f)), half(sinf(0.0f))); + // VERIFY_IS_APPROX(numext::sin(half(PI)), half(sinf(PI))); + VERIFY_IS_APPROX(numext::sin(half(PI/2)), half(sinf(PI/2))); + VERIFY_IS_APPROX(numext::sin(half(3*PI/2)), half(sinf(3*PI/2))); + VERIFY_IS_APPROX(numext::sin(half(3.5f)), half(sinf(3.5f))); + + VERIFY_IS_APPROX(numext::tan(half(0.0f)), half(tanf(0.0f))); + VERIFY_IS_APPROX(tan(half(0.0f)), half(tanf(0.0f))); + // VERIFY_IS_APPROX(numext::tan(half(PI)), half(tanf(PI))); + // VERIFY_IS_APPROX(numext::tan(half(PI/2)), half(tanf(PI/2))); + //VERIFY_IS_APPROX(numext::tan(half(3*PI/2)), half(tanf(3*PI/2))); + VERIFY_IS_APPROX(numext::tan(half(3.5f)), half(tanf(3.5f))); +} + +void test_array() +{ + typedef Array ArrayXh; + Index size = internal::random(1,10); + Index i = internal::random(0,size-1); + ArrayXh a1 = ArrayXh::Random(size), a2 = ArrayXh::Random(size); + VERIFY_IS_APPROX( a1+a1, half(2)*a1 ); + VERIFY( (a1.abs() >= half(0)).all() ); + VERIFY_IS_APPROX( (a1*a1).sqrt(), a1.abs() ); + + VERIFY( ((a1.min)(a2) <= (a1.max)(a2)).all() ); + a1(i) = half(-10.); + VERIFY_IS_EQUAL( a1.minCoeff(), half(-10.) ); + a1(i) = half(10.); + VERIFY_IS_EQUAL( a1.maxCoeff(), half(10.) ); + + std::stringstream ss; + ss << a1; +} + +void test_product() +{ + typedef Matrix MatrixXh; + Index rows = internal::random(1,EIGEN_TEST_MAX_SIZE); + Index cols = internal::random(1,EIGEN_TEST_MAX_SIZE); + Index depth = internal::random(1,EIGEN_TEST_MAX_SIZE); + MatrixXh Ah = MatrixXh::Random(rows,depth); + MatrixXh Bh = MatrixXh::Random(depth,cols); + MatrixXh Ch = MatrixXh::Random(rows,cols); + MatrixXf Af = Ah.cast(); + MatrixXf Bf = Bh.cast(); + MatrixXf Cf = Ch.cast(); + VERIFY_IS_APPROX(Ch.noalias()+=Ah*Bh, (Cf.noalias()+=Af*Bf).cast()); +} + +EIGEN_DECLARE_TEST(half_float) +{ + CALL_SUBTEST(test_numtraits()); + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST(test_conversion()); + CALL_SUBTEST(test_arithmetic()); + CALL_SUBTEST(test_comparison()); + CALL_SUBTEST(test_basic_functions()); + CALL_SUBTEST(test_trigonometric_functions()); + CALL_SUBTEST(test_array()); + CALL_SUBTEST(test_product()); + } +} diff --git a/thirdparty/eigen/test/hessenberg.cpp b/thirdparty/eigen/test/hessenberg.cpp index 96bc19e2..0e1b0098 100644 --- a/thirdparty/eigen/test/hessenberg.cpp +++ b/thirdparty/eigen/test/hessenberg.cpp @@ -49,7 +49,7 @@ template void hessenberg(int size = Size) // TODO: Add tests for packedMatrix() and householderCoefficients() } -void test_hessenberg() +EIGEN_DECLARE_TEST(hessenberg) { CALL_SUBTEST_1(( hessenberg,1>() )); CALL_SUBTEST_2(( hessenberg,2>() )); diff --git a/thirdparty/eigen/test/householder.cpp b/thirdparty/eigen/test/householder.cpp index c5f6b5e4..cad8138a 100644 --- a/thirdparty/eigen/test/householder.cpp +++ b/thirdparty/eigen/test/householder.cpp @@ -12,7 +12,6 @@ template void householder(const MatrixType& m) { - typedef typename MatrixType::Index Index; static bool even = true; even = !even; /* this test covers the following files: @@ -49,6 +48,17 @@ template void householder(const MatrixType& m) v1.applyHouseholderOnTheLeft(essential,beta,tmp); VERIFY_IS_APPROX(v1.norm(), v2.norm()); + // reconstruct householder matrix: + SquareMatrixType id, H1, H2; + id.setIdentity(rows, rows); + H1 = H2 = id; + VectorType vv(rows); + vv << Scalar(1), essential; + H1.applyHouseholderOnTheLeft(essential, beta, tmp); + H2.applyHouseholderOnTheRight(essential, beta, tmp); + VERIFY_IS_APPROX(H1, H2); + VERIFY_IS_APPROX(H1, id - beta * vv*vv.adjoint()); + MatrixType m1(rows, cols), m2(rows, cols); @@ -69,7 +79,7 @@ template void householder(const MatrixType& m) m3.rowwise() = v1.transpose(); m4 = m3; m3.row(0).makeHouseholder(essential, beta, alpha); - m3.applyHouseholderOnTheRight(essential,beta,tmp); + m3.applyHouseholderOnTheRight(essential.conjugate(),beta,tmp); VERIFY_IS_APPROX(m3.norm(), m4.norm()); if(rows>=2) VERIFY_IS_MUCH_SMALLER_THAN(m3.block(0,1,rows,rows-1).norm(), m3.norm()); VERIFY_IS_MUCH_SMALLER_THAN(numext::imag(m3(0,0)), numext::real(m3(0,0))); @@ -104,14 +114,14 @@ template void householder(const MatrixType& m) VERIFY_IS_APPROX(hseq_mat.adjoint(), hseq_mat_adj); VERIFY_IS_APPROX(hseq_mat.conjugate(), hseq_mat_conj); VERIFY_IS_APPROX(hseq_mat.transpose(), hseq_mat_trans); - VERIFY_IS_APPROX(hseq_mat * m6, hseq_mat * m6); - VERIFY_IS_APPROX(hseq_mat.adjoint() * m6, hseq_mat_adj * m6); - VERIFY_IS_APPROX(hseq_mat.conjugate() * m6, hseq_mat_conj * m6); - VERIFY_IS_APPROX(hseq_mat.transpose() * m6, hseq_mat_trans * m6); - VERIFY_IS_APPROX(m6 * hseq_mat, m6 * hseq_mat); - VERIFY_IS_APPROX(m6 * hseq_mat.adjoint(), m6 * hseq_mat_adj); - VERIFY_IS_APPROX(m6 * hseq_mat.conjugate(), m6 * hseq_mat_conj); - VERIFY_IS_APPROX(m6 * hseq_mat.transpose(), m6 * hseq_mat_trans); + VERIFY_IS_APPROX(hseq * m6, hseq_mat * m6); + VERIFY_IS_APPROX(hseq.adjoint() * m6, hseq_mat_adj * m6); + VERIFY_IS_APPROX(hseq.conjugate() * m6, hseq_mat_conj * m6); + VERIFY_IS_APPROX(hseq.transpose() * m6, hseq_mat_trans * m6); + VERIFY_IS_APPROX(m6 * hseq, m6 * hseq_mat); + VERIFY_IS_APPROX(m6 * hseq.adjoint(), m6 * hseq_mat_adj); + VERIFY_IS_APPROX(m6 * hseq.conjugate(), m6 * hseq_mat_conj); + VERIFY_IS_APPROX(m6 * hseq.transpose(), m6 * hseq_mat_trans); // test householder sequence on the right with a shift @@ -123,7 +133,7 @@ template void householder(const MatrixType& m) VERIFY_IS_APPROX(m3 * m5, m1); // test evaluating rhseq to a dense matrix, then applying } -void test_householder() +EIGEN_DECLARE_TEST(householder) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( householder(Matrix()) ); diff --git a/thirdparty/eigen/test/incomplete_cholesky.cpp b/thirdparty/eigen/test/incomplete_cholesky.cpp new file mode 100644 index 00000000..ecc17f5c --- /dev/null +++ b/thirdparty/eigen/test/incomplete_cholesky.cpp @@ -0,0 +1,69 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2015-2016 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +// #define EIGEN_DONT_VECTORIZE +// #define EIGEN_MAX_ALIGN_BYTES 0 +#include "sparse_solver.h" +#include +#include + +template void test_incomplete_cholesky_T() +{ + typedef SparseMatrix SparseMatrixType; + ConjugateGradient > > cg_illt_lower_amd; + ConjugateGradient > > cg_illt_lower_nat; + ConjugateGradient > > cg_illt_upper_amd; + ConjugateGradient > > cg_illt_upper_nat; + ConjugateGradient > > cg_illt_uplo_amd; + + + CALL_SUBTEST( check_sparse_spd_solving(cg_illt_lower_amd) ); + CALL_SUBTEST( check_sparse_spd_solving(cg_illt_lower_nat) ); + CALL_SUBTEST( check_sparse_spd_solving(cg_illt_upper_amd) ); + CALL_SUBTEST( check_sparse_spd_solving(cg_illt_upper_nat) ); + CALL_SUBTEST( check_sparse_spd_solving(cg_illt_uplo_amd) ); +} + +template +void bug1150() +{ + // regression for bug 1150 + for(int N = 1; N<20; ++N) + { + Eigen::MatrixXd b( N, N ); + b.setOnes(); + + Eigen::SparseMatrix m( N, N ); + m.reserve(Eigen::VectorXi::Constant(N,4)); + for( int i = 0; i < N; ++i ) + { + m.insert( i, i ) = 1; + m.coeffRef( i, i / 2 ) = 2; + m.coeffRef( i, i / 3 ) = 2; + m.coeffRef( i, i / 4 ) = 2; + } + + Eigen::SparseMatrix A; + A = m * m.transpose(); + + Eigen::ConjugateGradient, + Eigen::Lower | Eigen::Upper, + Eigen::IncompleteCholesky > solver( A ); + VERIFY(solver.preconditioner().info() == Eigen::Success); + VERIFY(solver.info() == Eigen::Success); + } +} + +EIGEN_DECLARE_TEST(incomplete_cholesky) +{ + CALL_SUBTEST_1(( test_incomplete_cholesky_T() )); + CALL_SUBTEST_2(( test_incomplete_cholesky_T, int>() )); + CALL_SUBTEST_3(( test_incomplete_cholesky_T() )); + + CALL_SUBTEST_1(( bug1150<0>() )); +} diff --git a/thirdparty/eigen/test/indexed_view.cpp b/thirdparty/eigen/test/indexed_view.cpp new file mode 100644 index 00000000..72c54af6 --- /dev/null +++ b/thirdparty/eigen/test/indexed_view.cpp @@ -0,0 +1,473 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2017 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#ifdef EIGEN_TEST_PART_2 +// Make sure we also check c++11 max implementation +#define EIGEN_MAX_CPP_VER 11 +#endif + +#ifdef EIGEN_TEST_PART_3 +// Make sure we also check c++98 max implementation +#define EIGEN_MAX_CPP_VER 03 + +// We need to disable this warning when compiling with c++11 while limiting Eigen to c++98 +// Ideally we would rather configure the compiler to build in c++98 mode but this needs +// to be done at the CMakeLists.txt level. +#if defined(__GNUC__) && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 8)) + #pragma GCC diagnostic ignored "-Wdeprecated" +#endif + +#if defined(__GNUC__) && (__GNUC__ >=9) + #pragma GCC diagnostic ignored "-Wdeprecated-copy" +#endif +#if defined(__clang__) && (__clang_major__ >= 10) + #pragma clang diagnostic ignored "-Wdeprecated-copy" +#endif + +#endif + +#include +#include +#include "main.h" + +#if EIGEN_HAS_CXX11 +#include +#endif + +typedef std::pair IndexPair; + +int encode(Index i, Index j) { + return int(i*100 + j); +} + +IndexPair decode(Index ij) { + return IndexPair(ij / 100, ij % 100); +} + +template +bool match(const T& xpr, std::string ref, std::string str_xpr = "") { + EIGEN_UNUSED_VARIABLE(str_xpr); + std::stringstream str; + str << xpr; + if(!(str.str() == ref)) + std::cout << str_xpr << "\n" << xpr << "\n\n"; + return str.str() == ref; +} + +#define MATCH(X,R) match(X, R, #X) + +template +typename internal::enable_if::value,bool>::type +is_same_eq(const T1& a, const T2& b) +{ + return (a == b).all(); +} + +template +bool is_same_seq(const T1& a, const T2& b) +{ + bool ok = a.first()==b.first() && a.size() == b.size() && Index(a.incrObject())==Index(b.incrObject());; + if(!ok) + { + std::cerr << "seqN(" << a.first() << ", " << a.size() << ", " << Index(a.incrObject()) << ") != "; + std::cerr << "seqN(" << b.first() << ", " << b.size() << ", " << Index(b.incrObject()) << ")\n"; + } + return ok; +} + +template +typename internal::enable_if::value,bool>::type +is_same_seq_type(const T1& a, const T2& b) +{ + return is_same_seq(a,b); +} + + + +#define VERIFY_EQ_INT(A,B) VERIFY_IS_APPROX(int(A),int(B)) + +// C++03 does not allow local or unnamed enums as index +enum DummyEnum { XX=0, YY=1 }; + +void check_indexed_view() +{ + Index n = 10; + + ArrayXd a = ArrayXd::LinSpaced(n,0,n-1); + Array b = a.transpose(); + + #if EIGEN_COMP_CXXVER>=14 + ArrayXXi A = ArrayXXi::NullaryExpr(n,n, std::ref(encode)); + #else + ArrayXXi A = ArrayXXi::NullaryExpr(n,n, std::ptr_fun(&encode)); + #endif + + for(Index i=0; i vali(4); Map(&vali[0],4) = eii; + std::vector veci(4); Map(veci.data(),4) = eii; + + VERIFY( MATCH( A(3, seq(9,3,-1)), + "309 308 307 306 305 304 303") + ); + + VERIFY( MATCH( A(seqN(2,5), seq(9,3,-1)), + "209 208 207 206 205 204 203\n" + "309 308 307 306 305 304 303\n" + "409 408 407 406 405 404 403\n" + "509 508 507 506 505 504 503\n" + "609 608 607 606 605 604 603") + ); + + VERIFY( MATCH( A(seqN(2,5), 5), + "205\n" + "305\n" + "405\n" + "505\n" + "605") + ); + + VERIFY( MATCH( A(seqN(last,5,-1), seq(2,last)), + "902 903 904 905 906 907 908 909\n" + "802 803 804 805 806 807 808 809\n" + "702 703 704 705 706 707 708 709\n" + "602 603 604 605 606 607 608 609\n" + "502 503 504 505 506 507 508 509") + ); + + VERIFY( MATCH( A(eii, veci), + "303 301 306 305\n" + "103 101 106 105\n" + "603 601 606 605\n" + "503 501 506 505") + ); + + VERIFY( MATCH( A(eii, all), + "300 301 302 303 304 305 306 307 308 309\n" + "100 101 102 103 104 105 106 107 108 109\n" + "600 601 602 603 604 605 606 607 608 609\n" + "500 501 502 503 504 505 506 507 508 509") + ); + + // take row number 3, and repeat it 5 times + VERIFY( MATCH( A(seqN(3,5,0), all), + "300 301 302 303 304 305 306 307 308 309\n" + "300 301 302 303 304 305 306 307 308 309\n" + "300 301 302 303 304 305 306 307 308 309\n" + "300 301 302 303 304 305 306 307 308 309\n" + "300 301 302 303 304 305 306 307 308 309") + ); + + VERIFY( MATCH( a(seqN(3,3),0), "3\n4\n5" ) ); + VERIFY( MATCH( a(seq(3,5)), "3\n4\n5" ) ); + VERIFY( MATCH( a(seqN(3,3,1)), "3\n4\n5" ) ); + VERIFY( MATCH( a(seqN(5,3,-1)), "5\n4\n3" ) ); + + VERIFY( MATCH( b(0,seqN(3,3)), "3 4 5" ) ); + VERIFY( MATCH( b(seq(3,5)), "3 4 5" ) ); + VERIFY( MATCH( b(seqN(3,3,1)), "3 4 5" ) ); + VERIFY( MATCH( b(seqN(5,3,-1)), "5 4 3" ) ); + + VERIFY( MATCH( b(all), "0 1 2 3 4 5 6 7 8 9" ) ); + VERIFY( MATCH( b(eii), "3 1 6 5" ) ); + + Array44i B; + B.setRandom(); + VERIFY( (A(seqN(2,5), 5)).ColsAtCompileTime == 1); + VERIFY( (A(seqN(2,5), 5)).RowsAtCompileTime == Dynamic); + VERIFY_EQ_INT( (A(seqN(2,5), 5)).InnerStrideAtCompileTime , A.InnerStrideAtCompileTime); + VERIFY_EQ_INT( (A(seqN(2,5), 5)).OuterStrideAtCompileTime , A.col(5).OuterStrideAtCompileTime); + + VERIFY_EQ_INT( (A(5,seqN(2,5))).InnerStrideAtCompileTime , A.row(5).InnerStrideAtCompileTime); + VERIFY_EQ_INT( (A(5,seqN(2,5))).OuterStrideAtCompileTime , A.row(5).OuterStrideAtCompileTime); + VERIFY_EQ_INT( (B(1,seqN(1,2))).InnerStrideAtCompileTime , B.row(1).InnerStrideAtCompileTime); + VERIFY_EQ_INT( (B(1,seqN(1,2))).OuterStrideAtCompileTime , B.row(1).OuterStrideAtCompileTime); + + VERIFY_EQ_INT( (A(seqN(2,5), seq(1,3))).InnerStrideAtCompileTime , A.InnerStrideAtCompileTime); + VERIFY_EQ_INT( (A(seqN(2,5), seq(1,3))).OuterStrideAtCompileTime , A.OuterStrideAtCompileTime); + VERIFY_EQ_INT( (B(seqN(1,2), seq(1,3))).InnerStrideAtCompileTime , B.InnerStrideAtCompileTime); + VERIFY_EQ_INT( (B(seqN(1,2), seq(1,3))).OuterStrideAtCompileTime , B.OuterStrideAtCompileTime); + VERIFY_EQ_INT( (A(seqN(2,5,2), seq(1,3,2))).InnerStrideAtCompileTime , Dynamic); + VERIFY_EQ_INT( (A(seqN(2,5,2), seq(1,3,2))).OuterStrideAtCompileTime , Dynamic); + VERIFY_EQ_INT( (A(seqN(2,5,fix<2>), seq(1,3,fix<3>))).InnerStrideAtCompileTime , 2); + VERIFY_EQ_INT( (A(seqN(2,5,fix<2>), seq(1,3,fix<3>))).OuterStrideAtCompileTime , Dynamic); + VERIFY_EQ_INT( (B(seqN(1,2,fix<2>), seq(1,3,fix<3>))).InnerStrideAtCompileTime , 2); + VERIFY_EQ_INT( (B(seqN(1,2,fix<2>), seq(1,3,fix<3>))).OuterStrideAtCompileTime , 3*4); + + VERIFY_EQ_INT( (A(seqN(2,fix<5>), seqN(1,fix<3>))).RowsAtCompileTime, 5); + VERIFY_EQ_INT( (A(seqN(2,fix<5>), seqN(1,fix<3>))).ColsAtCompileTime, 3); + VERIFY_EQ_INT( (A(seqN(2,fix<5>(5)), seqN(1,fix<3>(3)))).RowsAtCompileTime, 5); + VERIFY_EQ_INT( (A(seqN(2,fix<5>(5)), seqN(1,fix<3>(3)))).ColsAtCompileTime, 3); + VERIFY_EQ_INT( (A(seqN(2,fix(5)), seqN(1,fix(3)))).RowsAtCompileTime, Dynamic); + VERIFY_EQ_INT( (A(seqN(2,fix(5)), seqN(1,fix(3)))).ColsAtCompileTime, Dynamic); + VERIFY_EQ_INT( (A(seqN(2,fix(5)), seqN(1,fix(3)))).rows(), 5); + VERIFY_EQ_INT( (A(seqN(2,fix(5)), seqN(1,fix(3)))).cols(), 3); + + VERIFY( is_same_seq_type( seqN(2,5,fix<-1>), seqN(2,5,fix<-1>(-1)) ) ); + VERIFY( is_same_seq_type( seqN(2,5), seqN(2,5,fix<1>(1)) ) ); + VERIFY( is_same_seq_type( seqN(2,5,3), seqN(2,5,fix(3)) ) ); + VERIFY( is_same_seq_type( seq(2,7,fix<3>), seqN(2,2,fix<3>) ) ); + VERIFY( is_same_seq_type( seqN(2,fix(5),3), seqN(2,5,fix(3)) ) ); + VERIFY( is_same_seq_type( seqN(2,fix<5>(5),fix<-2>), seqN(2,fix<5>,fix<-2>()) ) ); + + VERIFY( is_same_seq_type( seq(2,fix<5>), seqN(2,4) ) ); +#if EIGEN_HAS_CXX11 + VERIFY( is_same_seq_type( seq(fix<2>,fix<5>), seqN(fix<2>,fix<4>) ) ); + VERIFY( is_same_seq( seqN(2,std::integral_constant(),std::integral_constant()), seqN(2,fix<5>,fix<-2>()) ) ); + VERIFY( is_same_seq( seq(std::integral_constant(),std::integral_constant(),std::integral_constant()), + seq(fix<1>,fix<5>,fix<2>()) ) ); + VERIFY( is_same_seq_type( seqN(2,std::integral_constant(),std::integral_constant()), seqN(2,fix<5>,fix<-2>()) ) ); + VERIFY( is_same_seq_type( seq(std::integral_constant(),std::integral_constant(),std::integral_constant()), + seq(fix<1>,fix<5>,fix<2>()) ) ); + + VERIFY( is_same_seq_type( seqN(2,std::integral_constant()), seqN(2,fix<5>) ) ); + VERIFY( is_same_seq_type( seq(std::integral_constant(),std::integral_constant()), seq(fix<1>,fix<5>) ) ); +#else + // sorry, no compile-time size recovery in c++98/03 + VERIFY( is_same_seq( seq(fix<2>,fix<5>), seqN(fix<2>,fix<4>) ) ); +#endif + + VERIFY( (A(seqN(2,fix<5>), 5)).RowsAtCompileTime == 5); + VERIFY( (A(4, all)).ColsAtCompileTime == Dynamic); + VERIFY( (A(4, all)).RowsAtCompileTime == 1); + VERIFY( (B(1, all)).ColsAtCompileTime == 4); + VERIFY( (B(1, all)).RowsAtCompileTime == 1); + VERIFY( (B(all,1)).ColsAtCompileTime == 1); + VERIFY( (B(all,1)).RowsAtCompileTime == 4); + + VERIFY(int( (A(all, eii)).ColsAtCompileTime) == int(eii.SizeAtCompileTime)); + VERIFY_EQ_INT( (A(eii, eii)).Flags&DirectAccessBit, (unsigned int)(0)); + VERIFY_EQ_INT( (A(eii, eii)).InnerStrideAtCompileTime, 0); + VERIFY_EQ_INT( (A(eii, eii)).OuterStrideAtCompileTime, 0); + + VERIFY_IS_APPROX( A(seq(n-1,2,-2), seqN(n-1-6,3,-1)), A(seq(last,2,fix<-2>), seqN(last-6,3,fix<-1>)) ); + + VERIFY_IS_APPROX( A(seq(n-1,2,-2), seqN(n-1-6,4)), A(seq(last,2,-2), seqN(last-6,4)) ); + VERIFY_IS_APPROX( A(seq(n-1-6,n-1-2), seqN(n-1-6,4)), A(seq(last-6,last-2), seqN(6+last-6-6,4)) ); + VERIFY_IS_APPROX( A(seq((n-1)/2,(n)/2+3), seqN(2,4)), A(seq(last/2,(last+1)/2+3), seqN(last+2-last,4)) ); + VERIFY_IS_APPROX( A(seq(n-2,2,-2), seqN(n-8,4)), A(seq(lastp1-2,2,-2), seqN(lastp1-8,4)) ); + + // Check all combinations of seq: + VERIFY_IS_APPROX( A(seq(1,n-1-2,2), seq(1,n-1-2,2)), A(seq(1,last-2,2), seq(1,last-2,fix<2>)) ); + VERIFY_IS_APPROX( A(seq(n-1-5,n-1-2,2), seq(n-1-5,n-1-2,2)), A(seq(last-5,last-2,2), seq(last-5,last-2,fix<2>)) ); + VERIFY_IS_APPROX( A(seq(n-1-5,7,2), seq(n-1-5,7,2)), A(seq(last-5,7,2), seq(last-5,7,fix<2>)) ); + VERIFY_IS_APPROX( A(seq(1,n-1-2), seq(n-1-5,7)), A(seq(1,last-2), seq(last-5,7)) ); + VERIFY_IS_APPROX( A(seq(n-1-5,n-1-2), seq(n-1-5,n-1-2)), A(seq(last-5,last-2), seq(last-5,last-2)) ); + + VERIFY_IS_APPROX( A.col(A.cols()-1), A(all,last) ); + VERIFY_IS_APPROX( A(A.rows()-2, A.cols()/2), A(last-1, lastp1/2) ); + VERIFY_IS_APPROX( a(a.size()-2), a(last-1) ); + VERIFY_IS_APPROX( a(a.size()/2), a((last+1)/2) ); + + // Check fall-back to Block + { + VERIFY( is_same_eq(A.col(0), A(all,0)) ); + VERIFY( is_same_eq(A.row(0), A(0,all)) ); + VERIFY( is_same_eq(A.block(0,0,2,2), A(seqN(0,2),seq(0,1))) ); + VERIFY( is_same_eq(A.middleRows(2,4), A(seqN(2,4),all)) ); + VERIFY( is_same_eq(A.middleCols(2,4), A(all,seqN(2,4))) ); + + VERIFY( is_same_eq(A.col(A.cols()-1), A(all,last)) ); + + const ArrayXXi& cA(A); + VERIFY( is_same_eq(cA.col(0), cA(all,0)) ); + VERIFY( is_same_eq(cA.row(0), cA(0,all)) ); + VERIFY( is_same_eq(cA.block(0,0,2,2), cA(seqN(0,2),seq(0,1))) ); + VERIFY( is_same_eq(cA.middleRows(2,4), cA(seqN(2,4),all)) ); + VERIFY( is_same_eq(cA.middleCols(2,4), cA(all,seqN(2,4))) ); + + VERIFY( is_same_eq(a.head(4), a(seq(0,3))) ); + VERIFY( is_same_eq(a.tail(4), a(seqN(last-3,4))) ); + VERIFY( is_same_eq(a.tail(4), a(seq(lastp1-4,last))) ); + VERIFY( is_same_eq(a.segment<4>(3), a(seqN(3,fix<4>))) ); + } + + ArrayXXi A1=A, A2 = ArrayXXi::Random(4,4); + ArrayXi range25(4); range25 << 3,2,4,5; + A1(seqN(3,4),seq(2,5)) = A2; + VERIFY_IS_APPROX( A1.block(3,2,4,4), A2 ); + A1 = A; + A2.setOnes(); + A1(seq(6,3,-1),range25) = A2; + VERIFY_IS_APPROX( A1.block(3,2,4,4), A2 ); + + // check reverse + { + VERIFY( is_same_seq_type( seq(3,7).reverse(), seqN(7,5,fix<-1>) ) ); + VERIFY( is_same_seq_type( seq(7,3,fix<-2>).reverse(), seqN(3,3,fix<2>) ) ); + VERIFY_IS_APPROX( a(seqN(2,last/2).reverse()), a(seqN(2+(last/2-1)*1,last/2,fix<-1>)) ); + VERIFY_IS_APPROX( a(seqN(last/2,fix<4>).reverse()),a(seqN(last/2,fix<4>)).reverse() ); + VERIFY_IS_APPROX( A(seq(last-5,last-1,2).reverse(), seqN(last-3,3,fix<-2>).reverse()), + A(seq(last-5,last-1,2), seqN(last-3,3,fix<-2>)).reverse() ); + } + +#if EIGEN_HAS_CXX11 + // check lastN + VERIFY_IS_APPROX( a(lastN(3)), a.tail(3) ); + VERIFY( MATCH( a(lastN(3)), "7\n8\n9" ) ); + VERIFY_IS_APPROX( a(lastN(fix<3>())), a.tail<3>() ); + VERIFY( MATCH( a(lastN(3,2)), "5\n7\n9" ) ); + VERIFY( MATCH( a(lastN(3,fix<2>())), "5\n7\n9" ) ); + VERIFY( a(lastN(fix<3>())).SizeAtCompileTime == 3 ); + + VERIFY( (A(all, std::array{{1,3,2,4}})).ColsAtCompileTime == 4); + + VERIFY_IS_APPROX( (A(std::array{{1,3,5}}, std::array{{9,6,3,0}})), A(seqN(1,3,2), seqN(9,4,-3)) ); + +#if EIGEN_HAS_STATIC_ARRAY_TEMPLATE + VERIFY_IS_APPROX( A({3, 1, 6, 5}, all), A(std::array{{3, 1, 6, 5}}, all) ); + VERIFY_IS_APPROX( A(all,{3, 1, 6, 5}), A(all,std::array{{3, 1, 6, 5}}) ); + VERIFY_IS_APPROX( A({1,3,5},{3, 1, 6, 5}), A(std::array{{1,3,5}},std::array{{3, 1, 6, 5}}) ); + + VERIFY_IS_EQUAL( A({1,3,5},{3, 1, 6, 5}).RowsAtCompileTime, 3 ); + VERIFY_IS_EQUAL( A({1,3,5},{3, 1, 6, 5}).ColsAtCompileTime, 4 ); + + VERIFY_IS_APPROX( a({3, 1, 6, 5}), a(std::array{{3, 1, 6, 5}}) ); + VERIFY_IS_EQUAL( a({1,3,5}).SizeAtCompileTime, 3 ); + + VERIFY_IS_APPROX( b({3, 1, 6, 5}), b(std::array{{3, 1, 6, 5}}) ); + VERIFY_IS_EQUAL( b({1,3,5}).SizeAtCompileTime, 3 ); +#endif + +#endif + + // check mat(i,j) with weird types for i and j + { + VERIFY_IS_APPROX( A(B.RowsAtCompileTime-1, 1), A(3,1) ); + VERIFY_IS_APPROX( A(B.RowsAtCompileTime, 1), A(4,1) ); + VERIFY_IS_APPROX( A(B.RowsAtCompileTime-1, B.ColsAtCompileTime-1), A(3,3) ); + VERIFY_IS_APPROX( A(B.RowsAtCompileTime, B.ColsAtCompileTime), A(4,4) ); + const Index I_ = 3, J_ = 4; + VERIFY_IS_APPROX( A(I_,J_), A(3,4) ); + } + + // check extended block API + { + VERIFY( is_same_eq( A.block<3,4>(1,1), A.block(1,1,fix<3>,fix<4>)) ); + VERIFY( is_same_eq( A.block<3,4>(1,1,3,4), A.block(1,1,fix<3>(),fix<4>(4))) ); + VERIFY( is_same_eq( A.block<3,Dynamic>(1,1,3,4), A.block(1,1,fix<3>,4)) ); + VERIFY( is_same_eq( A.block(1,1,3,4), A.block(1,1,fix(3),fix<4>)) ); + VERIFY( is_same_eq( A.block(1,1,3,4), A.block(1,1,fix(3),fix(4))) ); + + VERIFY( is_same_eq( A.topLeftCorner<3,4>(), A.topLeftCorner(fix<3>,fix<4>)) ); + VERIFY( is_same_eq( A.bottomLeftCorner<3,4>(), A.bottomLeftCorner(fix<3>,fix<4>)) ); + VERIFY( is_same_eq( A.bottomRightCorner<3,4>(), A.bottomRightCorner(fix<3>,fix<4>)) ); + VERIFY( is_same_eq( A.topRightCorner<3,4>(), A.topRightCorner(fix<3>,fix<4>)) ); + + VERIFY( is_same_eq( A.leftCols<3>(), A.leftCols(fix<3>)) ); + VERIFY( is_same_eq( A.rightCols<3>(), A.rightCols(fix<3>)) ); + VERIFY( is_same_eq( A.middleCols<3>(1), A.middleCols(1,fix<3>)) ); + + VERIFY( is_same_eq( A.topRows<3>(), A.topRows(fix<3>)) ); + VERIFY( is_same_eq( A.bottomRows<3>(), A.bottomRows(fix<3>)) ); + VERIFY( is_same_eq( A.middleRows<3>(1), A.middleRows(1,fix<3>)) ); + + VERIFY( is_same_eq( a.segment<3>(1), a.segment(1,fix<3>)) ); + VERIFY( is_same_eq( a.head<3>(), a.head(fix<3>)) ); + VERIFY( is_same_eq( a.tail<3>(), a.tail(fix<3>)) ); + + const ArrayXXi& cA(A); + VERIFY( is_same_eq( cA.block(1,1,3,4), cA.block(1,1,fix(3),fix<4>)) ); + + VERIFY( is_same_eq( cA.topLeftCorner<3,4>(), cA.topLeftCorner(fix<3>,fix<4>)) ); + VERIFY( is_same_eq( cA.bottomLeftCorner<3,4>(), cA.bottomLeftCorner(fix<3>,fix<4>)) ); + VERIFY( is_same_eq( cA.bottomRightCorner<3,4>(), cA.bottomRightCorner(fix<3>,fix<4>)) ); + VERIFY( is_same_eq( cA.topRightCorner<3,4>(), cA.topRightCorner(fix<3>,fix<4>)) ); + + VERIFY( is_same_eq( cA.leftCols<3>(), cA.leftCols(fix<3>)) ); + VERIFY( is_same_eq( cA.rightCols<3>(), cA.rightCols(fix<3>)) ); + VERIFY( is_same_eq( cA.middleCols<3>(1), cA.middleCols(1,fix<3>)) ); + + VERIFY( is_same_eq( cA.topRows<3>(), cA.topRows(fix<3>)) ); + VERIFY( is_same_eq( cA.bottomRows<3>(), cA.bottomRows(fix<3>)) ); + VERIFY( is_same_eq( cA.middleRows<3>(1), cA.middleRows(1,fix<3>)) ); + } + + // Check compilation of enums as index type: + a(XX) = 1; + A(XX,YY) = 1; + // Anonymous enums only work with C++11 +#if EIGEN_HAS_CXX11 + enum { X=0, Y=1 }; + a(X) = 1; + A(X,Y) = 1; + A(XX,Y) = 1; + A(X,YY) = 1; +#endif + + // Check compilation of varying integer types as index types: + Index i = n/2; + short i_short(i); + std::size_t i_sizet(i); + VERIFY_IS_EQUAL( a(i), a.coeff(i_short) ); + VERIFY_IS_EQUAL( a(i), a.coeff(i_sizet) ); + + VERIFY_IS_EQUAL( A(i,i), A.coeff(i_short, i_short) ); + VERIFY_IS_EQUAL( A(i,i), A.coeff(i_short, i) ); + VERIFY_IS_EQUAL( A(i,i), A.coeff(i, i_short) ); + VERIFY_IS_EQUAL( A(i,i), A.coeff(i, i_sizet) ); + VERIFY_IS_EQUAL( A(i,i), A.coeff(i_sizet, i) ); + VERIFY_IS_EQUAL( A(i,i), A.coeff(i_sizet, i_short) ); + VERIFY_IS_EQUAL( A(i,i), A.coeff(5, i_sizet) ); + + // Regression test for Max{Rows,Cols}AtCompileTime + { + Matrix3i A3 = Matrix3i::Random(); + ArrayXi ind(5); ind << 1,1,1,1,1; + VERIFY_IS_EQUAL( A3(ind,ind).eval(), MatrixXi::Constant(5,5,A3(1,1)) ); + } + + // Regression for bug 1736 + { + VERIFY_IS_APPROX(A(all, eii).col(0).eval(), A.col(eii(0))); + A(all, eii).col(0) = A.col(eii(0)); + } + + // bug 1815: IndexedView should allow linear access + { + VERIFY( MATCH( b(eii)(0), "3" ) ); + VERIFY( MATCH( a(eii)(0), "3" ) ); + VERIFY( MATCH( A(1,eii)(0), "103")); + VERIFY( MATCH( A(eii,1)(0), "301")); + VERIFY( MATCH( A(1,all)(1), "101")); + VERIFY( MATCH( A(all,1)(1), "101")); + } + +#if EIGEN_HAS_CXX11 + //Bug IndexView with a single static row should be RowMajor: + { + // A(1, seq(0,2,1)).cwiseAbs().colwise().replicate(2).eval(); + STATIC_CHECK(( (internal::evaluator::Flags & RowMajorBit) == RowMajorBit )); + } +#endif + +} + +EIGEN_DECLARE_TEST(indexed_view) +{ +// for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1( check_indexed_view() ); + CALL_SUBTEST_2( check_indexed_view() ); + CALL_SUBTEST_3( check_indexed_view() ); +// } + + // static checks of some internals: + STATIC_CHECK(( internal::is_valid_index_type::value )); + STATIC_CHECK(( internal::is_valid_index_type::value )); + STATIC_CHECK(( internal::is_valid_index_type::value )); + STATIC_CHECK(( internal::is_valid_index_type::value )); + STATIC_CHECK(( internal::is_valid_index_type::value )); + STATIC_CHECK(( !internal::valid_indexed_view_overload::value )); + STATIC_CHECK(( !internal::valid_indexed_view_overload::value )); + STATIC_CHECK(( !internal::valid_indexed_view_overload::value )); + STATIC_CHECK(( !internal::valid_indexed_view_overload::value )); +} diff --git a/thirdparty/eigen/test/initializer_list_construction.cpp b/thirdparty/eigen/test/initializer_list_construction.cpp new file mode 100644 index 00000000..7a9c49e8 --- /dev/null +++ b/thirdparty/eigen/test/initializer_list_construction.cpp @@ -0,0 +1,385 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2019 David Tellenbach +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#define EIGEN_NO_STATIC_ASSERT + +#include "main.h" + +template::IsInteger> +struct TestMethodDispatching { + static void run() {} +}; + +template +struct TestMethodDispatching { + static void run() + { + { + Matrix m {3, 4}; + Array a {3, 4}; + VERIFY(m.rows() == 3); + VERIFY(m.cols() == 4); + VERIFY(a.rows() == 3); + VERIFY(a.cols() == 4); + } + { + Matrix m {3, 4}; + Array a {3, 4}; + VERIFY(m(0) == 3); + VERIFY(m(1) == 4); + VERIFY(a(0) == 3); + VERIFY(a(1) == 4); + } + { + Matrix m {3, 4}; + Array a {3, 4}; + VERIFY(m(0) == 3); + VERIFY(m(1) == 4); + VERIFY(a(0) == 3); + VERIFY(a(1) == 4); + } + } +}; + +template void fixedsizeVariadicVectorConstruction2() +{ + { + Vec4 ref = Vec4::Random(); + Vec4 v{ ref[0], ref[1], ref[2], ref[3] }; + VERIFY_IS_APPROX(v, ref); + VERIFY_IS_APPROX(v, (Vec4( ref[0], ref[1], ref[2], ref[3] ))); + VERIFY_IS_APPROX(v, (Vec4({ref[0], ref[1], ref[2], ref[3]}))); + + Vec4 v2 = { ref[0], ref[1], ref[2], ref[3] }; + VERIFY_IS_APPROX(v2, ref); + } + { + Vec5 ref = Vec5::Random(); + Vec5 v{ ref[0], ref[1], ref[2], ref[3], ref[4] }; + VERIFY_IS_APPROX(v, ref); + VERIFY_IS_APPROX(v, (Vec5( ref[0], ref[1], ref[2], ref[3], ref[4] ))); + VERIFY_IS_APPROX(v, (Vec5({ref[0], ref[1], ref[2], ref[3], ref[4]}))); + + Vec5 v2 = { ref[0], ref[1], ref[2], ref[3], ref[4] }; + VERIFY_IS_APPROX(v2, ref); + } +} + +#define CHECK_MIXSCALAR_V5_APPROX(V, A0, A1, A2, A3, A4) { \ + VERIFY_IS_APPROX(V[0], Scalar(A0) ); \ + VERIFY_IS_APPROX(V[1], Scalar(A1) ); \ + VERIFY_IS_APPROX(V[2], Scalar(A2) ); \ + VERIFY_IS_APPROX(V[3], Scalar(A3) ); \ + VERIFY_IS_APPROX(V[4], Scalar(A4) ); \ +} + +#define CHECK_MIXSCALAR_V5(VEC5, A0, A1, A2, A3, A4) { \ + typedef VEC5::Scalar Scalar; \ + VEC5 v = { A0 , A1 , A2 , A3 , A4 }; \ + CHECK_MIXSCALAR_V5_APPROX(v, A0 , A1 , A2 , A3 , A4); \ +} + +template void fixedsizeVariadicVectorConstruction3() +{ + typedef Matrix Vec5; + typedef Array Arr5; + CHECK_MIXSCALAR_V5(Vec5, 1, 2., -3, 4.121, 5.53252); + CHECK_MIXSCALAR_V5(Arr5, 1, 2., 3.12f, 4.121, 5.53252); +} + +template void fixedsizeVariadicVectorConstruction() +{ + CALL_SUBTEST(( fixedsizeVariadicVectorConstruction2, Matrix >() )); + CALL_SUBTEST(( fixedsizeVariadicVectorConstruction2, Matrix >() )); + CALL_SUBTEST(( fixedsizeVariadicVectorConstruction2, Array >() )); + CALL_SUBTEST(( fixedsizeVariadicVectorConstruction2, Array >() )); +} + + +template void initializerListVectorConstruction() +{ + Scalar raw[4]; + for(int k = 0; k < 4; ++k) { + raw[k] = internal::random(); + } + { + Matrix m { {raw[0]}, {raw[1]},{raw[2]},{raw[3]} }; + Array a { {raw[0]}, {raw[1]}, {raw[2]}, {raw[3]} }; + for(int k = 0; k < 4; ++k) { + VERIFY(m(k) == raw[k]); + } + for(int k = 0; k < 4; ++k) { + VERIFY(a(k) == raw[k]); + } + VERIFY_IS_EQUAL(m, (Matrix({ {raw[0]}, {raw[1]}, {raw[2]}, {raw[3]} }))); + VERIFY((a == (Array({ {raw[0]}, {raw[1]}, {raw[2]}, {raw[3]} }))).all()); + } + { + Matrix m { {raw[0], raw[1], raw[2], raw[3]} }; + Array a { {raw[0], raw[1], raw[2], raw[3]} }; + for(int k = 0; k < 4; ++k) { + VERIFY(m(k) == raw[k]); + } + for(int k = 0; k < 4; ++k) { + VERIFY(a(k) == raw[k]); + } + VERIFY_IS_EQUAL(m, (Matrix({{raw[0],raw[1],raw[2],raw[3]}}))); + VERIFY((a == (Array({{raw[0],raw[1],raw[2],raw[3]}}))).all()); + } + { + Matrix m { {raw[0]}, {raw[1]}, {raw[2]}, {raw[3]} }; + Array a { {raw[0]}, {raw[1]}, {raw[2]}, {raw[3]} }; + for(int k=0; k < 4; ++k) { + VERIFY(m(k) == raw[k]); + } + for(int k=0; k < 4; ++k) { + VERIFY(a(k) == raw[k]); + } + VERIFY_IS_EQUAL(m, (Matrix({ {raw[0]}, {raw[1]}, {raw[2]}, {raw[3]} }))); + VERIFY((a == (Array({ {raw[0]}, {raw[1]}, {raw[2]}, {raw[3]} }))).all()); + } + { + Matrix m {{raw[0],raw[1],raw[2],raw[3]}}; + Array a {{raw[0],raw[1],raw[2],raw[3]}}; + for(int k=0; k < 4; ++k) { + VERIFY(m(k) == raw[k]); + } + for(int k=0; k < 4; ++k) { + VERIFY(a(k) == raw[k]); + } + VERIFY_IS_EQUAL(m, (Matrix({{raw[0],raw[1],raw[2],raw[3]}}))); + VERIFY((a == (Array({{raw[0],raw[1],raw[2],raw[3]}}))).all()); + } +} + +template void initializerListMatrixConstruction() +{ + const Index RowsAtCompileTime = 5; + const Index ColsAtCompileTime = 4; + const Index SizeAtCompileTime = RowsAtCompileTime * ColsAtCompileTime; + + Scalar raw[SizeAtCompileTime]; + for (int i = 0; i < SizeAtCompileTime; ++i) { + raw[i] = internal::random(); + } + { + Matrix m {}; + VERIFY(m.cols() == 0); + VERIFY(m.rows() == 0); + VERIFY_IS_EQUAL(m, (Matrix())); + } + { + Matrix m { + {raw[0], raw[1], raw[2], raw[3]}, + {raw[4], raw[5], raw[6], raw[7]}, + {raw[8], raw[9], raw[10], raw[11]}, + {raw[12], raw[13], raw[14], raw[15]}, + {raw[16], raw[17], raw[18], raw[19]} + }; + + Matrix m2; + m2 << raw[0], raw[1], raw[2], raw[3], + raw[4], raw[5], raw[6], raw[7], + raw[8], raw[9], raw[10], raw[11], + raw[12], raw[13], raw[14], raw[15], + raw[16], raw[17], raw[18], raw[19]; + + int k = 0; + for(int i = 0; i < RowsAtCompileTime; ++i) { + for (int j = 0; j < ColsAtCompileTime; ++j) { + VERIFY(m(i, j) == raw[k]); + ++k; + } + } + VERIFY_IS_EQUAL(m, m2); + } + { + Matrix m{ + {raw[0], raw[1], raw[2], raw[3]}, + {raw[4], raw[5], raw[6], raw[7]}, + {raw[8], raw[9], raw[10], raw[11]}, + {raw[12], raw[13], raw[14], raw[15]}, + {raw[16], raw[17], raw[18], raw[19]} + }; + + VERIFY(m.cols() == 4); + VERIFY(m.rows() == 5); + int k = 0; + for(int i = 0; i < RowsAtCompileTime; ++i) { + for (int j = 0; j < ColsAtCompileTime; ++j) { + VERIFY(m(i, j) == raw[k]); + ++k; + } + } + + Matrix m2(RowsAtCompileTime, ColsAtCompileTime); + k = 0; + for(int i = 0; i < RowsAtCompileTime; ++i) { + for (int j = 0; j < ColsAtCompileTime; ++j) { + m2(i, j) = raw[k]; + ++k; + } + } + VERIFY_IS_EQUAL(m, m2); + } +} + +template void initializerListArrayConstruction() +{ + const Index RowsAtCompileTime = 5; + const Index ColsAtCompileTime = 4; + const Index SizeAtCompileTime = RowsAtCompileTime * ColsAtCompileTime; + + Scalar raw[SizeAtCompileTime]; + for (int i = 0; i < SizeAtCompileTime; ++i) { + raw[i] = internal::random(); + } + { + Array a {}; + VERIFY(a.cols() == 0); + VERIFY(a.rows() == 0); + } + { + Array m { + {raw[0], raw[1], raw[2], raw[3]}, + {raw[4], raw[5], raw[6], raw[7]}, + {raw[8], raw[9], raw[10], raw[11]}, + {raw[12], raw[13], raw[14], raw[15]}, + {raw[16], raw[17], raw[18], raw[19]} + }; + + Array m2; + m2 << raw[0], raw[1], raw[2], raw[3], + raw[4], raw[5], raw[6], raw[7], + raw[8], raw[9], raw[10], raw[11], + raw[12], raw[13], raw[14], raw[15], + raw[16], raw[17], raw[18], raw[19]; + + int k = 0; + for(int i = 0; i < RowsAtCompileTime; ++i) { + for (int j = 0; j < ColsAtCompileTime; ++j) { + VERIFY(m(i, j) == raw[k]); + ++k; + } + } + VERIFY_IS_APPROX(m, m2); + } + { + Array m { + {raw[0], raw[1], raw[2], raw[3]}, + {raw[4], raw[5], raw[6], raw[7]}, + {raw[8], raw[9], raw[10], raw[11]}, + {raw[12], raw[13], raw[14], raw[15]}, + {raw[16], raw[17], raw[18], raw[19]} + }; + + VERIFY(m.cols() == 4); + VERIFY(m.rows() == 5); + int k = 0; + for(int i = 0; i < RowsAtCompileTime; ++i) { + for (int j = 0; j < ColsAtCompileTime; ++j) { + VERIFY(m(i, j) == raw[k]); + ++k; + } + } + + Array m2(RowsAtCompileTime, ColsAtCompileTime); + k = 0; + for(int i = 0; i < RowsAtCompileTime; ++i) { + for (int j = 0; j < ColsAtCompileTime; ++j) { + m2(i, j) = raw[k]; + ++k; + } + } + VERIFY_IS_APPROX(m, m2); + } +} + +template void dynamicVectorConstruction() +{ + const Index size = 4; + Scalar raw[size]; + for (int i = 0; i < size; ++i) { + raw[i] = internal::random(); + } + + typedef Matrix VectorX; + + { + VectorX v {{raw[0], raw[1], raw[2], raw[3]}}; + for (int i = 0; i < size; ++i) { + VERIFY(v(i) == raw[i]); + } + VERIFY(v.rows() == size); + VERIFY(v.cols() == 1); + VERIFY_IS_EQUAL(v, (VectorX {{raw[0], raw[1], raw[2], raw[3]}})); + } + + { + VERIFY_RAISES_ASSERT((VectorX {raw[0], raw[1], raw[2], raw[3]})); + } + { + VERIFY_RAISES_ASSERT((VectorX { + {raw[0], raw[1], raw[2], raw[3]}, + {raw[0], raw[1], raw[2], raw[3]}, + })); + } +} + +EIGEN_DECLARE_TEST(initializer_list_construction) +{ + CALL_SUBTEST_1(initializerListVectorConstruction()); + CALL_SUBTEST_1(initializerListVectorConstruction()); + CALL_SUBTEST_1(initializerListVectorConstruction()); + CALL_SUBTEST_1(initializerListVectorConstruction()); + CALL_SUBTEST_1(initializerListVectorConstruction()); + CALL_SUBTEST_1(initializerListVectorConstruction()); + CALL_SUBTEST_1(initializerListVectorConstruction>()); + CALL_SUBTEST_1(initializerListVectorConstruction>()); + + CALL_SUBTEST_2(initializerListMatrixConstruction()); + CALL_SUBTEST_2(initializerListMatrixConstruction()); + CALL_SUBTEST_2(initializerListMatrixConstruction()); + CALL_SUBTEST_2(initializerListMatrixConstruction()); + CALL_SUBTEST_2(initializerListMatrixConstruction()); + CALL_SUBTEST_2(initializerListMatrixConstruction()); + CALL_SUBTEST_2(initializerListMatrixConstruction>()); + CALL_SUBTEST_2(initializerListMatrixConstruction>()); + + CALL_SUBTEST_3(initializerListArrayConstruction()); + CALL_SUBTEST_3(initializerListArrayConstruction()); + CALL_SUBTEST_3(initializerListArrayConstruction()); + CALL_SUBTEST_3(initializerListArrayConstruction()); + CALL_SUBTEST_3(initializerListArrayConstruction()); + CALL_SUBTEST_3(initializerListArrayConstruction()); + CALL_SUBTEST_3(initializerListArrayConstruction>()); + CALL_SUBTEST_3(initializerListArrayConstruction>()); + + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction>()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction>()); + CALL_SUBTEST_4(fixedsizeVariadicVectorConstruction3<0>()); + + CALL_SUBTEST_5(TestMethodDispatching::run()); + CALL_SUBTEST_5(TestMethodDispatching::run()); + + CALL_SUBTEST_6(dynamicVectorConstruction()); + CALL_SUBTEST_6(dynamicVectorConstruction()); + CALL_SUBTEST_6(dynamicVectorConstruction()); + CALL_SUBTEST_6(dynamicVectorConstruction()); + CALL_SUBTEST_6(dynamicVectorConstruction()); + CALL_SUBTEST_6(dynamicVectorConstruction()); + CALL_SUBTEST_6(dynamicVectorConstruction>()); + CALL_SUBTEST_6(dynamicVectorConstruction>()); +} diff --git a/thirdparty/eigen/test/inplace_decomposition.cpp b/thirdparty/eigen/test/inplace_decomposition.cpp new file mode 100644 index 00000000..e3aa9957 --- /dev/null +++ b/thirdparty/eigen/test/inplace_decomposition.cpp @@ -0,0 +1,110 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2016 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" +#include +#include +#include + +// This file test inplace decomposition through Ref<>, as supported by Cholesky, LU, and QR decompositions. + +template void inplace(bool square = false, bool SPD = false) +{ + typedef typename MatrixType::Scalar Scalar; + typedef Matrix RhsType; + typedef Matrix ResType; + + Index rows = MatrixType::RowsAtCompileTime==Dynamic ? internal::random(2,EIGEN_TEST_MAX_SIZE/2) : Index(MatrixType::RowsAtCompileTime); + Index cols = MatrixType::ColsAtCompileTime==Dynamic ? (square?rows:internal::random(2,rows)) : Index(MatrixType::ColsAtCompileTime); + + MatrixType A = MatrixType::Random(rows,cols); + RhsType b = RhsType::Random(rows); + ResType x(cols); + + if(SPD) + { + assert(square); + A.topRows(cols) = A.topRows(cols).adjoint() * A.topRows(cols); + A.diagonal().array() += 1e-3; + } + + MatrixType A0 = A; + MatrixType A1 = A; + + DecType dec(A); + + // Check that the content of A has been modified + VERIFY_IS_NOT_APPROX( A, A0 ); + + // Check that the decomposition is correct: + if(rows==cols) + { + VERIFY_IS_APPROX( A0 * (x = dec.solve(b)), b ); + } + else + { + VERIFY_IS_APPROX( A0.transpose() * A0 * (x = dec.solve(b)), A0.transpose() * b ); + } + + // Check that modifying A breaks the current dec: + A.setRandom(); + if(rows==cols) + { + VERIFY_IS_NOT_APPROX( A0 * (x = dec.solve(b)), b ); + } + else + { + VERIFY_IS_NOT_APPROX( A0.transpose() * A0 * (x = dec.solve(b)), A0.transpose() * b ); + } + + // Check that calling compute(A1) does not modify A1: + A = A0; + dec.compute(A1); + VERIFY_IS_EQUAL(A0,A1); + VERIFY_IS_NOT_APPROX( A, A0 ); + if(rows==cols) + { + VERIFY_IS_APPROX( A0 * (x = dec.solve(b)), b ); + } + else + { + VERIFY_IS_APPROX( A0.transpose() * A0 * (x = dec.solve(b)), A0.transpose() * b ); + } +} + + +EIGEN_DECLARE_TEST(inplace_decomposition) +{ + EIGEN_UNUSED typedef Matrix Matrix43d; + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1(( inplace >, MatrixXd>(true,true) )); + CALL_SUBTEST_1(( inplace >, Matrix4d>(true,true) )); + + CALL_SUBTEST_2(( inplace >, MatrixXd>(true,true) )); + CALL_SUBTEST_2(( inplace >, Matrix4d>(true,true) )); + + CALL_SUBTEST_3(( inplace >, MatrixXd>(true,false) )); + CALL_SUBTEST_3(( inplace >, Matrix4d>(true,false) )); + + CALL_SUBTEST_4(( inplace >, MatrixXd>(true,false) )); + CALL_SUBTEST_4(( inplace >, Matrix4d>(true,false) )); + + CALL_SUBTEST_5(( inplace >, MatrixXd>(false,false) )); + CALL_SUBTEST_5(( inplace >, Matrix43d>(false,false) )); + + CALL_SUBTEST_6(( inplace >, MatrixXd>(false,false) )); + CALL_SUBTEST_6(( inplace >, Matrix43d>(false,false) )); + + CALL_SUBTEST_7(( inplace >, MatrixXd>(false,false) )); + CALL_SUBTEST_7(( inplace >, Matrix43d>(false,false) )); + + CALL_SUBTEST_8(( inplace >, MatrixXd>(false,false) )); + CALL_SUBTEST_8(( inplace >, Matrix43d>(false,false) )); + } +} diff --git a/thirdparty/eigen/test/integer_types.cpp b/thirdparty/eigen/test/integer_types.cpp index 950f8e9b..31f4100c 100644 --- a/thirdparty/eigen/test/integer_types.cpp +++ b/thirdparty/eigen/test/integer_types.cpp @@ -18,7 +18,6 @@ template void signed_integer_type_tests(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; enum { is_signed = (Scalar(-1) > Scalar(0)) ? 0 : 1 }; @@ -49,7 +48,6 @@ template void signed_integer_type_tests(const MatrixType& m template void integer_type_tests(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; VERIFY(NumTraits::IsInteger); @@ -133,7 +131,18 @@ template void integer_type_tests(const MatrixType& m) VERIFY_IS_APPROX((m1 * m2.transpose()) * m1, m1 * (m2.transpose() * m1)); } -void test_integer_types() +template +void integer_types_extra() +{ + VERIFY_IS_EQUAL(int(internal::scalar_div_cost::value), 8); + VERIFY_IS_EQUAL(int(internal::scalar_div_cost::value), 8); + if(sizeof(long)>sizeof(int)) { + VERIFY(int(internal::scalar_div_cost::value) > int(internal::scalar_div_cost::value)); + VERIFY(int(internal::scalar_div_cost::value) > int(internal::scalar_div_cost::value)); + } +} + +EIGEN_DECLARE_TEST(integer_types) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( integer_type_tests(Matrix()) ); @@ -153,9 +162,12 @@ void test_integer_types() CALL_SUBTEST_6( integer_type_tests(Matrix()) ); +#if EIGEN_HAS_CXX11 CALL_SUBTEST_7( integer_type_tests(Matrix()) ); CALL_SUBTEST_7( signed_integer_type_tests(Matrix()) ); CALL_SUBTEST_8( integer_type_tests(Matrix(1, 5)) ); +#endif } + CALL_SUBTEST_9( integer_types_extra<0>() ); } diff --git a/thirdparty/eigen/test/inverse.cpp b/thirdparty/eigen/test/inverse.cpp index 8187b088..9cedfa1e 100644 --- a/thirdparty/eigen/test/inverse.cpp +++ b/thirdparty/eigen/test/inverse.cpp @@ -11,10 +11,59 @@ #include "main.h" #include -template void inverse(const MatrixType& m) +template +void inverse_for_fixed_size(const MatrixType&, typename internal::enable_if::type* = 0) +{ +} + +template +void inverse_for_fixed_size(const MatrixType& m1, typename internal::enable_if::type* = 0) { using std::abs; - typedef typename MatrixType::Index Index; + + MatrixType m2, identity = MatrixType::Identity(); + + typedef typename MatrixType::Scalar Scalar; + typedef typename NumTraits::Real RealScalar; + typedef Matrix VectorType; + + //computeInverseAndDetWithCheck tests + //First: an invertible matrix + bool invertible; + Scalar det; + + m2.setZero(); + m1.computeInverseAndDetWithCheck(m2, det, invertible); + VERIFY(invertible); + VERIFY_IS_APPROX(identity, m1*m2); + VERIFY_IS_APPROX(det, m1.determinant()); + + m2.setZero(); + m1.computeInverseWithCheck(m2, invertible); + VERIFY(invertible); + VERIFY_IS_APPROX(identity, m1*m2); + + //Second: a rank one matrix (not invertible, except for 1x1 matrices) + VectorType v3 = VectorType::Random(); + MatrixType m3 = v3*v3.transpose(), m4; + m3.computeInverseAndDetWithCheck(m4, det, invertible); + VERIFY( m1.rows()==1 ? invertible : !invertible ); + VERIFY_IS_MUCH_SMALLER_THAN(abs(det-m3.determinant()), RealScalar(1)); + m3.computeInverseWithCheck(m4, invertible); + VERIFY( m1.rows()==1 ? invertible : !invertible ); + + // check with submatrices + { + Matrix m5; + m5.setRandom(); + m5.topLeftCorner(m1.rows(),m1.rows()) = m1; + m2 = m5.template topLeftCorner().inverse(); + VERIFY_IS_APPROX( (m5.template topLeftCorner()), m2.inverse() ); + } +} + +template void inverse(const MatrixType& m) +{ /* this test covers the following files: Inverse.h */ @@ -40,35 +89,7 @@ template void inverse(const MatrixType& m) // since for the general case we implement separately row-major and col-major, test that VERIFY_IS_APPROX(MatrixType(m1.transpose().inverse()), MatrixType(m1.inverse().transpose())); -#if !defined(EIGEN_TEST_PART_5) && !defined(EIGEN_TEST_PART_6) - typedef typename NumTraits::Real RealScalar; - typedef Matrix VectorType; - - //computeInverseAndDetWithCheck tests - //First: an invertible matrix - bool invertible; - RealScalar det; - - m2.setZero(); - m1.computeInverseAndDetWithCheck(m2, det, invertible); - VERIFY(invertible); - VERIFY_IS_APPROX(identity, m1*m2); - VERIFY_IS_APPROX(det, m1.determinant()); - - m2.setZero(); - m1.computeInverseWithCheck(m2, invertible); - VERIFY(invertible); - VERIFY_IS_APPROX(identity, m1*m2); - - //Second: a rank one matrix (not invertible, except for 1x1 matrices) - VectorType v3 = VectorType::Random(rows); - MatrixType m3 = v3*v3.transpose(), m4(rows,cols); - m3.computeInverseAndDetWithCheck(m4, det, invertible); - VERIFY( rows==1 ? invertible : !invertible ); - VERIFY_IS_MUCH_SMALLER_THAN(abs(det-m3.determinant()), RealScalar(1)); - m3.computeInverseWithCheck(m4, invertible); - VERIFY( rows==1 ? invertible : !invertible ); -#endif + inverse_for_fixed_size(m1); // check in-place inversion if(MatrixType::RowsAtCompileTime>=2 && MatrixType::RowsAtCompileTime<=4) @@ -84,7 +105,23 @@ template void inverse(const MatrixType& m) } } -void test_inverse() +template +void inverse_zerosized() +{ + Matrix A(0,0); + { + Matrix b, x; + x = A.inverse() * b; + } + { + Matrix b(0,1), x; + x = A.inverse() * b; + VERIFY_IS_EQUAL(x.rows(), 0); + VERIFY_IS_EQUAL(x.cols(), 1); + } +} + +EIGEN_DECLARE_TEST(inverse) { int s = 0; for(int i = 0; i < g_repeat; i++) { @@ -93,12 +130,21 @@ void test_inverse() CALL_SUBTEST_3( inverse(Matrix3f()) ); CALL_SUBTEST_4( inverse(Matrix4f()) ); CALL_SUBTEST_4( inverse(Matrix()) ); + s = internal::random(50,320); CALL_SUBTEST_5( inverse(MatrixXf(s,s)) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) + CALL_SUBTEST_5( inverse_zerosized() ); + CALL_SUBTEST_5( inverse(MatrixXf(0, 0)) ); + CALL_SUBTEST_5( inverse(MatrixXf(1, 1)) ); + s = internal::random(25,100); CALL_SUBTEST_6( inverse(MatrixXcd(s,s)) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) + CALL_SUBTEST_7( inverse(Matrix4d()) ); CALL_SUBTEST_7( inverse(Matrix()) ); + + CALL_SUBTEST_8( inverse(Matrix4cd()) ); } - TEST_SET_BUT_UNUSED_VARIABLE(s) } diff --git a/thirdparty/eigen/test/io.cpp b/thirdparty/eigen/test/io.cpp new file mode 100644 index 00000000..aa14e76e --- /dev/null +++ b/thirdparty/eigen/test/io.cpp @@ -0,0 +1,71 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2019 Joel Holdsworth +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include + +#include "main.h" + +template +struct check_ostream_impl +{ + static void run() + { + const Array array(123); + std::ostringstream ss; + ss << array; + VERIFY(ss.str() == "123"); + + check_ostream_impl< std::complex >::run(); + }; +}; + +template<> +struct check_ostream_impl +{ + static void run() + { + const Array array(1, 0); + std::ostringstream ss; + ss << array; + VERIFY(ss.str() == "1 0"); + }; +}; + +template +struct check_ostream_impl< std::complex > +{ + static void run() + { + const Array,1,1> array(std::complex(12, 34)); + std::ostringstream ss; + ss << array; + VERIFY(ss.str() == "(12,34)"); + }; +}; + +template +static void check_ostream() +{ + check_ostream_impl::run(); +} + +EIGEN_DECLARE_TEST(rand) +{ + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); + CALL_SUBTEST(check_ostream()); +} diff --git a/thirdparty/eigen/test/is_same_dense.cpp b/thirdparty/eigen/test/is_same_dense.cpp new file mode 100644 index 00000000..23dd806e --- /dev/null +++ b/thirdparty/eigen/test/is_same_dense.cpp @@ -0,0 +1,41 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2015 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" + +using internal::is_same_dense; + +EIGEN_DECLARE_TEST(is_same_dense) +{ + typedef Matrix ColMatrixXd; + typedef Matrix,Dynamic,Dynamic,ColMajor> ColMatrixXcd; + ColMatrixXd m1(10,10); + ColMatrixXcd m2(10,10); + Ref ref_m1(m1); + Ref > ref_m2_real(m2.real()); + Ref const_ref_m1(m1); + + VERIFY(is_same_dense(m1,m1)); + VERIFY(is_same_dense(m1,ref_m1)); + VERIFY(is_same_dense(const_ref_m1,m1)); + VERIFY(is_same_dense(const_ref_m1,ref_m1)); + + VERIFY(is_same_dense(m1.block(0,0,m1.rows(),m1.cols()),m1)); + VERIFY(!is_same_dense(m1.row(0),m1.col(0))); + + Ref const_ref_m1_row(m1.row(1)); + VERIFY(!is_same_dense(m1.row(1),const_ref_m1_row)); + + Ref const_ref_m1_col(m1.col(1)); + VERIFY(is_same_dense(m1.col(1),const_ref_m1_col)); + + + VERIFY(!is_same_dense(m1, ref_m2_real)); + VERIFY(!is_same_dense(m2, ref_m2_real)); +} diff --git a/thirdparty/eigen/test/jacobi.cpp b/thirdparty/eigen/test/jacobi.cpp index 7ccd4124..5604797f 100644 --- a/thirdparty/eigen/test/jacobi.cpp +++ b/thirdparty/eigen/test/jacobi.cpp @@ -14,7 +14,6 @@ template void jacobi(const MatrixType& m = MatrixType()) { - typedef typename MatrixType::Index Index; Index rows = m.rows(); Index cols = m.cols(); @@ -58,7 +57,7 @@ void jacobi(const MatrixType& m = MatrixType()) } } -void test_jacobi() +EIGEN_DECLARE_TEST(jacobi) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1(( jacobi() )); diff --git a/thirdparty/eigen/test/jacobisvd.cpp b/thirdparty/eigen/test/jacobisvd.cpp index 12c556e5..c8818c88 100644 --- a/thirdparty/eigen/test/jacobisvd.cpp +++ b/thirdparty/eigen/test/jacobisvd.cpp @@ -1,292 +1,71 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2008 Gael Guennebaud +// Copyright (C) 2008-2014 Gael Guennebaud // Copyright (C) 2009 Benoit Jacob // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +// We explicitly disable deprecated declarations for this set of tests +// because we purposely verify assertions for the deprecated SVD runtime +// option behavior. +#if defined(__GNUC__) +#pragma GCC diagnostic ignored "-Wdeprecated-declarations" +#elif defined(_MSC_VER) +#pragma warning( disable : 4996 ) +#endif + // discard stack allocation as that too bypasses malloc #define EIGEN_STACK_ALLOCATION_LIMIT 0 #define EIGEN_RUNTIME_NO_MALLOC #include "main.h" #include -template -void jacobisvd_check_full(const MatrixType& m, const JacobiSVD& svd) -{ - typedef typename MatrixType::Index Index; - Index rows = m.rows(); - Index cols = m.cols(); - - enum { - RowsAtCompileTime = MatrixType::RowsAtCompileTime, - ColsAtCompileTime = MatrixType::ColsAtCompileTime - }; - - typedef typename MatrixType::Scalar Scalar; - typedef Matrix MatrixUType; - typedef Matrix MatrixVType; - - MatrixType sigma = MatrixType::Zero(rows,cols); - sigma.diagonal() = svd.singularValues().template cast(); - MatrixUType u = svd.matrixU(); - MatrixVType v = svd.matrixV(); - - VERIFY_IS_APPROX(m, u * sigma * v.adjoint()); - VERIFY_IS_UNITARY(u); - VERIFY_IS_UNITARY(v); -} - -template -void jacobisvd_compare_to_full(const MatrixType& m, - unsigned int computationOptions, - const JacobiSVD& referenceSvd) -{ - typedef typename MatrixType::Index Index; - Index rows = m.rows(); - Index cols = m.cols(); - Index diagSize = (std::min)(rows, cols); - - JacobiSVD svd(m, computationOptions); - - VERIFY_IS_APPROX(svd.singularValues(), referenceSvd.singularValues()); - if(computationOptions & ComputeFullU) - VERIFY_IS_APPROX(svd.matrixU(), referenceSvd.matrixU()); - if(computationOptions & ComputeThinU) - VERIFY_IS_APPROX(svd.matrixU(), referenceSvd.matrixU().leftCols(diagSize)); - if(computationOptions & ComputeFullV) - VERIFY_IS_APPROX(svd.matrixV(), referenceSvd.matrixV()); - if(computationOptions & ComputeThinV) - VERIFY_IS_APPROX(svd.matrixV(), referenceSvd.matrixV().leftCols(diagSize)); -} - -template -void jacobisvd_solve(const MatrixType& m, unsigned int computationOptions) -{ - typedef typename MatrixType::Scalar Scalar; - typedef typename MatrixType::RealScalar RealScalar; - typedef typename MatrixType::Index Index; - Index rows = m.rows(); - Index cols = m.cols(); - - enum { - RowsAtCompileTime = MatrixType::RowsAtCompileTime, - ColsAtCompileTime = MatrixType::ColsAtCompileTime - }; - - typedef Matrix RhsType; - typedef Matrix SolutionType; - - RhsType rhs = RhsType::Random(rows, internal::random(1, cols)); - JacobiSVD svd(m, computationOptions); - - if(internal::is_same::value) svd.setThreshold(1e-8); - else if(internal::is_same::value) svd.setThreshold(1e-4); - - SolutionType x = svd.solve(rhs); - - RealScalar residual = (m*x-rhs).norm(); - // Check that there is no significantly better solution in the neighborhood of x - if(!test_isMuchSmallerThan(residual,rhs.norm())) - { - // If the residual is very small, then we have an exact solution, so we are already good. - for(int k=0;k::epsilon(); - RealScalar residual_y = (m*y-rhs).norm(); - VERIFY( test_isApprox(residual_y,residual) || residual < residual_y ); - - y.row(k) = x.row(k).array() - 2*NumTraits::epsilon(); - residual_y = (m*y-rhs).norm(); - VERIFY( test_isApprox(residual_y,residual) || residual < residual_y ); - } - } - - // evaluate normal equation which works also for least-squares solutions - if(internal::is_same::value) - { - // This test is not stable with single precision. - // This is probably because squaring m signicantly affects the precision. - VERIFY_IS_APPROX(m.adjoint()*m*x,m.adjoint()*rhs); - } - - // check minimal norm solutions - { - // generate a full-rank m x n problem with m MatrixType2; - typedef Matrix RhsType2; - typedef Matrix MatrixType2T; - Index rank = RankAtCompileTime2==Dynamic ? internal::random(1,cols) : Index(RankAtCompileTime2); - MatrixType2 m2(rank,cols); - int guard = 0; - do { - m2.setRandom(); - } while(m2.jacobiSvd().setThreshold(test_precision()).rank()!=rank && (++guard)<10); - VERIFY(guard<10); - RhsType2 rhs2 = RhsType2::Random(rank); - // use QR to find a reference minimal norm solution - HouseholderQR qr(m2.adjoint()); - Matrix tmp = qr.matrixQR().topLeftCorner(rank,rank).template triangularView().adjoint().solve(rhs2); - tmp.conservativeResize(cols); - tmp.tail(cols-rank).setZero(); - SolutionType x21 = qr.householderQ() * tmp; - // now check with SVD - JacobiSVD svd2(m2, computationOptions); - SolutionType x22 = svd2.solve(rhs2); - VERIFY_IS_APPROX(m2*x21, rhs2); - VERIFY_IS_APPROX(m2*x22, rhs2); - VERIFY_IS_APPROX(x21, x22); - - // Now check with a rank deficient matrix - typedef Matrix MatrixType3; - typedef Matrix RhsType3; - Index rows3 = RowsAtCompileTime3==Dynamic ? internal::random(rank+1,2*cols) : Index(RowsAtCompileTime3); - Matrix C = Matrix::Random(rows3,rank); - MatrixType3 m3 = C * m2; - RhsType3 rhs3 = C * rhs2; - JacobiSVD svd3(m3, computationOptions); - SolutionType x3 = svd3.solve(rhs3); - if(svd3.rank()!=rank) { - std::cout << m3 << "\n\n"; - std::cout << svd3.singularValues().transpose() << "\n"; - std::cout << svd3.rank() << " == " << rank << "\n"; - std::cout << x21.norm() << " == " << x3.norm() << "\n"; - } -// VERIFY_IS_APPROX(m3*x3, rhs3); - VERIFY_IS_APPROX(m3*x21, rhs3); - VERIFY_IS_APPROX(m2*x3, rhs2); - - VERIFY_IS_APPROX(x21, x3); - } -} - -template -void jacobisvd_test_all_computation_options(const MatrixType& m) -{ - if (QRPreconditioner == NoQRPreconditioner && m.rows() != m.cols()) - return; - JacobiSVD fullSvd(m, ComputeFullU|ComputeFullV); - CALL_SUBTEST(( jacobisvd_check_full(m, fullSvd) )); - CALL_SUBTEST(( jacobisvd_solve(m, ComputeFullU | ComputeFullV) )); - - #if defined __INTEL_COMPILER - // remark #111: statement is unreachable - #pragma warning disable 111 - #endif - if(QRPreconditioner == FullPivHouseholderQRPreconditioner) - return; - - CALL_SUBTEST(( jacobisvd_compare_to_full(m, ComputeFullU, fullSvd) )); - CALL_SUBTEST(( jacobisvd_compare_to_full(m, ComputeFullV, fullSvd) )); - CALL_SUBTEST(( jacobisvd_compare_to_full(m, 0, fullSvd) )); - - if (MatrixType::ColsAtCompileTime == Dynamic) { - // thin U/V are only available with dynamic number of columns - CALL_SUBTEST(( jacobisvd_compare_to_full(m, ComputeFullU|ComputeThinV, fullSvd) )); - CALL_SUBTEST(( jacobisvd_compare_to_full(m, ComputeThinV, fullSvd) )); - CALL_SUBTEST(( jacobisvd_compare_to_full(m, ComputeThinU|ComputeFullV, fullSvd) )); - CALL_SUBTEST(( jacobisvd_compare_to_full(m, ComputeThinU , fullSvd) )); - CALL_SUBTEST(( jacobisvd_compare_to_full(m, ComputeThinU|ComputeThinV, fullSvd) )); - CALL_SUBTEST(( jacobisvd_solve(m, ComputeFullU | ComputeThinV) )); - CALL_SUBTEST(( jacobisvd_solve(m, ComputeThinU | ComputeFullV) )); - CALL_SUBTEST(( jacobisvd_solve(m, ComputeThinU | ComputeThinV) )); - - // test reconstruction - typedef typename MatrixType::Index Index; - Index diagSize = (std::min)(m.rows(), m.cols()); - JacobiSVD svd(m, ComputeThinU | ComputeThinV); - VERIFY_IS_APPROX(m, svd.matrixU().leftCols(diagSize) * svd.singularValues().asDiagonal() * svd.matrixV().leftCols(diagSize).adjoint()); - } -} +#define SVD_DEFAULT(M) JacobiSVD +#define SVD_FOR_MIN_NORM(M) JacobiSVD +#include "svd_common.h" +// Check all variants of JacobiSVD template void jacobisvd(const MatrixType& a = MatrixType(), bool pickrandom = true) { MatrixType m = a; if(pickrandom) - { - typedef typename MatrixType::Scalar Scalar; - typedef typename MatrixType::RealScalar RealScalar; - typedef typename MatrixType::Index Index; - Index diagSize = (std::min)(a.rows(), a.cols()); - RealScalar s = std::numeric_limits::max_exponent10/4; - s = internal::random(1,s); - Matrix d = Matrix::Random(diagSize); - for(Index k=0; k(-s,s)); - m = Matrix::Random(a.rows(),diagSize) * d.asDiagonal() * Matrix::Random(diagSize,a.cols()); - // cancel some coeffs - Index n = internal::random(0,m.size()-1); - for(Index i=0; i(0,m.rows()-1), internal::random(0,m.cols()-1)) = Scalar(0); - } + svd_fill_random(m); - CALL_SUBTEST(( jacobisvd_test_all_computation_options(m) )); - CALL_SUBTEST(( jacobisvd_test_all_computation_options(m) )); - CALL_SUBTEST(( jacobisvd_test_all_computation_options(m) )); - CALL_SUBTEST(( jacobisvd_test_all_computation_options(m) )); + CALL_SUBTEST(( svd_test_all_computation_options >(m, true) )); // check full only + CALL_SUBTEST(( svd_test_all_computation_options >(m, false) )); + CALL_SUBTEST(( svd_test_all_computation_options >(m, false) )); + if(m.rows()==m.cols()) + CALL_SUBTEST(( svd_test_all_computation_options >(m, false) )); } template void jacobisvd_verify_assert(const MatrixType& m) { - typedef typename MatrixType::Scalar Scalar; - typedef typename MatrixType::Index Index; + svd_verify_assert >(m); + svd_verify_assert >(m, true); + svd_verify_assert >(m); + svd_verify_assert >(m); Index rows = m.rows(); Index cols = m.cols(); enum { - RowsAtCompileTime = MatrixType::RowsAtCompileTime, ColsAtCompileTime = MatrixType::ColsAtCompileTime }; - typedef Matrix RhsType; - - RhsType rhs(rows); - - JacobiSVD svd; - VERIFY_RAISES_ASSERT(svd.matrixU()) - VERIFY_RAISES_ASSERT(svd.singularValues()) - VERIFY_RAISES_ASSERT(svd.matrixV()) - VERIFY_RAISES_ASSERT(svd.solve(rhs)) MatrixType a = MatrixType::Zero(rows, cols); a.setZero(); - svd.compute(a, 0); - VERIFY_RAISES_ASSERT(svd.matrixU()) - VERIFY_RAISES_ASSERT(svd.matrixV()) - svd.singularValues(); - VERIFY_RAISES_ASSERT(svd.solve(rhs)) if (ColsAtCompileTime == Dynamic) { - svd.compute(a, ComputeThinU); - svd.matrixU(); - VERIFY_RAISES_ASSERT(svd.matrixV()) - VERIFY_RAISES_ASSERT(svd.solve(rhs)) - - svd.compute(a, ComputeThinV); - svd.matrixV(); - VERIFY_RAISES_ASSERT(svd.matrixU()) - VERIFY_RAISES_ASSERT(svd.solve(rhs)) - JacobiSVD svd_fullqr; VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeFullU|ComputeThinV)) VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeThinU|ComputeThinV)) VERIFY_RAISES_ASSERT(svd_fullqr.compute(a, ComputeThinU|ComputeFullV)) } - else - { - VERIFY_RAISES_ASSERT(svd.compute(a, ComputeThinU)) - VERIFY_RAISES_ASSERT(svd.compute(a, ComputeThinV)) - } } template @@ -300,128 +79,36 @@ void jacobisvd_method() VERIFY_RAISES_ASSERT(m.jacobiSvd().matrixU()); VERIFY_RAISES_ASSERT(m.jacobiSvd().matrixV()); VERIFY_IS_APPROX(m.jacobiSvd(ComputeFullU|ComputeFullV).solve(m), m); + VERIFY_IS_APPROX(m.jacobiSvd(ComputeFullU|ComputeFullV).transpose().solve(m), m); + VERIFY_IS_APPROX(m.jacobiSvd(ComputeFullU|ComputeFullV).adjoint().solve(m), m); } -// work around stupid msvc error when constructing at compile time an expression that involves -// a division by zero, even if the numeric type has floating point -template -EIGEN_DONT_INLINE Scalar zero() { return Scalar(0); } - -// workaround aggressive optimization in ICC -template EIGEN_DONT_INLINE T sub(T a, T b) { return a - b; } - -template -void jacobisvd_inf_nan() -{ - // all this function does is verify we don't iterate infinitely on nan/inf values - - JacobiSVD svd; - typedef typename MatrixType::Scalar Scalar; - Scalar some_inf = Scalar(1) / zero(); - VERIFY(sub(some_inf, some_inf) != sub(some_inf, some_inf)); - svd.compute(MatrixType::Constant(10,10,some_inf), ComputeFullU | ComputeFullV); - - Scalar nan = std::numeric_limits::quiet_NaN(); - VERIFY(nan != nan); - svd.compute(MatrixType::Constant(10,10,nan), ComputeFullU | ComputeFullV); - - MatrixType m = MatrixType::Zero(10,10); - m(internal::random(0,9), internal::random(0,9)) = some_inf; - svd.compute(m, ComputeFullU | ComputeFullV); - - m = MatrixType::Zero(10,10); - m(internal::random(0,9), internal::random(0,9)) = nan; - svd.compute(m, ComputeFullU | ComputeFullV); - - // regression test for bug 791 - m.resize(3,3); - m << 0, 2*NumTraits::epsilon(), 0.5, - 0, -0.5, 0, - nan, 0, 0; - svd.compute(m, ComputeFullU | ComputeFullV); -} - -// Regression test for bug 286: JacobiSVD loops indefinitely with some -// matrices containing denormal numbers. -void jacobisvd_bug286() -{ -#if defined __INTEL_COMPILER -// shut up warning #239: floating point underflow -#pragma warning push -#pragma warning disable 239 -#endif - Matrix2d M; - M << -7.90884e-313, -4.94e-324, - 0, 5.60844e-313; -#if defined __INTEL_COMPILER -#pragma warning pop -#endif - JacobiSVD svd; - svd.compute(M); // just check we don't loop indefinitely +namespace Foo { +// older compiler require a default constructor for Bar +// cf: https://stackoverflow.com/questions/7411515/ +class Bar {public: Bar() {}}; +bool operator<(const Bar&, const Bar&) { return true; } } - -void jacobisvd_preallocate() +// regression test for a very strange MSVC issue for which simply +// including SVDBase.h messes up with std::max and custom scalar type +void msvc_workaround() { - Vector3f v(3.f, 2.f, 1.f); - MatrixXf m = v.asDiagonal(); - - internal::set_is_malloc_allowed(false); - VERIFY_RAISES_ASSERT(VectorXf tmp(10);) - JacobiSVD svd; - internal::set_is_malloc_allowed(true); - svd.compute(m); - VERIFY_IS_APPROX(svd.singularValues(), v); - - JacobiSVD svd2(3,3); - internal::set_is_malloc_allowed(false); - svd2.compute(m); - internal::set_is_malloc_allowed(true); - VERIFY_IS_APPROX(svd2.singularValues(), v); - VERIFY_RAISES_ASSERT(svd2.matrixU()); - VERIFY_RAISES_ASSERT(svd2.matrixV()); - svd2.compute(m, ComputeFullU | ComputeFullV); - VERIFY_IS_APPROX(svd2.matrixU(), Matrix3f::Identity()); - VERIFY_IS_APPROX(svd2.matrixV(), Matrix3f::Identity()); - internal::set_is_malloc_allowed(false); - svd2.compute(m); - internal::set_is_malloc_allowed(true); - - JacobiSVD svd3(3,3,ComputeFullU|ComputeFullV); - internal::set_is_malloc_allowed(false); - svd2.compute(m); - internal::set_is_malloc_allowed(true); - VERIFY_IS_APPROX(svd2.singularValues(), v); - VERIFY_IS_APPROX(svd2.matrixU(), Matrix3f::Identity()); - VERIFY_IS_APPROX(svd2.matrixV(), Matrix3f::Identity()); - internal::set_is_malloc_allowed(false); - svd2.compute(m, ComputeFullU|ComputeFullV); - internal::set_is_malloc_allowed(true); + const Foo::Bar a; + const Foo::Bar b; + std::max EIGEN_NOT_A_MACRO (a,b); } -void test_jacobisvd() +EIGEN_DECLARE_TEST(jacobisvd) { CALL_SUBTEST_3(( jacobisvd_verify_assert(Matrix3f()) )); CALL_SUBTEST_4(( jacobisvd_verify_assert(Matrix4d()) )); CALL_SUBTEST_7(( jacobisvd_verify_assert(MatrixXf(10,12)) )); CALL_SUBTEST_8(( jacobisvd_verify_assert(MatrixXcd(7,5)) )); + + CALL_SUBTEST_11(svd_all_trivial_2x2(jacobisvd)); + CALL_SUBTEST_12(svd_all_trivial_2x2(jacobisvd)); for(int i = 0; i < g_repeat; i++) { - Matrix2cd m; - m << 0, 1, - 0, 1; - CALL_SUBTEST_1(( jacobisvd(m, false) )); - m << 1, 0, - 1, 0; - CALL_SUBTEST_1(( jacobisvd(m, false) )); - - Matrix2d n; - n << 0, 0, - 0, 0; - CALL_SUBTEST_2(( jacobisvd(n, false) )); - n << 0, 0, - 0, 1; - CALL_SUBTEST_2(( jacobisvd(n, false) )); - CALL_SUBTEST_3(( jacobisvd() )); CALL_SUBTEST_4(( jacobisvd() )); CALL_SUBTEST_5(( jacobisvd >() )); @@ -440,8 +127,14 @@ void test_jacobisvd() (void) c; // Test on inf/nan matrix - CALL_SUBTEST_7( jacobisvd_inf_nan() ); - CALL_SUBTEST_10( jacobisvd_inf_nan() ); + CALL_SUBTEST_7( (svd_inf_nan, MatrixXf>()) ); + CALL_SUBTEST_10( (svd_inf_nan, MatrixXd>()) ); + + // bug1395 test compile-time vectors as input + CALL_SUBTEST_13(( jacobisvd_verify_assert(Matrix()) )); + CALL_SUBTEST_13(( jacobisvd_verify_assert(Matrix()) )); + CALL_SUBTEST_13(( jacobisvd_verify_assert(Matrix(r)) )); + CALL_SUBTEST_13(( jacobisvd_verify_assert(Matrix(c)) )); } CALL_SUBTEST_7(( jacobisvd(MatrixXf(internal::random(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/2), internal::random(EIGEN_TEST_MAX_SIZE/4, EIGEN_TEST_MAX_SIZE/2))) )); @@ -455,8 +148,9 @@ void test_jacobisvd() CALL_SUBTEST_7( JacobiSVD(10,10) ); // Check that preallocation avoids subsequent mallocs - CALL_SUBTEST_9( jacobisvd_preallocate() ); + CALL_SUBTEST_9( svd_preallocate() ); + + CALL_SUBTEST_2( svd_underoverflow() ); - // Regression check for bug 286 - CALL_SUBTEST_2( jacobisvd_bug286() ); + msvc_workaround(); } diff --git a/thirdparty/eigen/test/klu_support.cpp b/thirdparty/eigen/test/klu_support.cpp new file mode 100644 index 00000000..f806ad50 --- /dev/null +++ b/thirdparty/eigen/test/klu_support.cpp @@ -0,0 +1,32 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2011 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#define EIGEN_NO_DEBUG_SMALL_PRODUCT_BLOCKS +#include "sparse_solver.h" + +#include + +template void test_klu_support_T() +{ + KLU > klu_colmajor; + KLU > klu_rowmajor; + + check_sparse_square_solving(klu_colmajor); + check_sparse_square_solving(klu_rowmajor); + + //check_sparse_square_determinant(umfpack_colmajor); + //check_sparse_square_determinant(umfpack_rowmajor); +} + +EIGEN_DECLARE_TEST(klu_support) +{ + CALL_SUBTEST_1(test_klu_support_T()); + CALL_SUBTEST_2(test_klu_support_T >()); +} + diff --git a/thirdparty/eigen/test/linearstructure.cpp b/thirdparty/eigen/test/linearstructure.cpp index 618984d5..46ee5162 100644 --- a/thirdparty/eigen/test/linearstructure.cpp +++ b/thirdparty/eigen/test/linearstructure.cpp @@ -2,11 +2,15 @@ // for linear algebra. // // Copyright (C) 2006-2008 Benoit Jacob +// Copyright (C) 2014 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +static bool g_called; +#define EIGEN_SCALAR_BINARY_OP_PLUGIN { g_called |= (!internal::is_same::value); } + #include "main.h" template void linearStructure(const MatrixType& m) @@ -15,8 +19,8 @@ template void linearStructure(const MatrixType& m) /* this test covers the following files: CwiseUnaryOp.h, CwiseBinaryOp.h, SelfCwiseBinaryOp.h */ - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; + typedef typename MatrixType::RealScalar RealScalar; Index rows = m.rows(); Index cols = m.cols(); @@ -28,7 +32,7 @@ template void linearStructure(const MatrixType& m) m3(rows, cols); Scalar s1 = internal::random(); - while (abs(s1)<1e-3) s1 = internal::random(); + while (abs(s1)(); Index r = internal::random(0, rows-1), c = internal::random(0, cols-1); @@ -68,8 +72,61 @@ template void linearStructure(const MatrixType& m) VERIFY_IS_APPROX(m1.block(0,0,rows,cols) * s1, m1 * s1); } -void test_linearstructure() +// Make sure that complex * real and real * complex are properly optimized +template void real_complex(DenseIndex rows = MatrixType::RowsAtCompileTime, DenseIndex cols = MatrixType::ColsAtCompileTime) +{ + typedef typename MatrixType::Scalar Scalar; + typedef typename MatrixType::RealScalar RealScalar; + + RealScalar s = internal::random(); + MatrixType m1 = MatrixType::Random(rows, cols); + + g_called = false; + VERIFY_IS_APPROX(s*m1, Scalar(s)*m1); + VERIFY(g_called && "real * matrix not properly optimized"); + + g_called = false; + VERIFY_IS_APPROX(m1*s, m1*Scalar(s)); + VERIFY(g_called && "matrix * real not properly optimized"); + + g_called = false; + VERIFY_IS_APPROX(m1/s, m1/Scalar(s)); + VERIFY(g_called && "matrix / real not properly optimized"); + + g_called = false; + VERIFY_IS_APPROX(s+m1.array(), Scalar(s)+m1.array()); + VERIFY(g_called && "real + matrix not properly optimized"); + + g_called = false; + VERIFY_IS_APPROX(m1.array()+s, m1.array()+Scalar(s)); + VERIFY(g_called && "matrix + real not properly optimized"); + + g_called = false; + VERIFY_IS_APPROX(s-m1.array(), Scalar(s)-m1.array()); + VERIFY(g_called && "real - matrix not properly optimized"); + + g_called = false; + VERIFY_IS_APPROX(m1.array()-s, m1.array()-Scalar(s)); + VERIFY(g_called && "matrix - real not properly optimized"); +} + +template +void linearstructure_overflow() +{ + // make sure that /=scalar and /scalar do not overflow + // rational: 1.0/4.94e-320 overflow, but m/4.94e-320 should not + Matrix4d m2, m3; + m3 = m2 = Matrix4d::Random()*1e-20; + m2 = m2 / 4.9e-320; + VERIFY_IS_APPROX(m2.cwiseQuotient(m2), Matrix4d::Ones()); + m3 /= 4.9e-320; + VERIFY_IS_APPROX(m3.cwiseQuotient(m3), Matrix4d::Ones()); +} + +EIGEN_DECLARE_TEST(linearstructure) { + g_called = true; + VERIFY(g_called); // avoid `unneeded-internal-declaration` warning. for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( linearStructure(Matrix()) ); CALL_SUBTEST_2( linearStructure(Matrix2f()) ); @@ -80,5 +137,11 @@ void test_linearstructure() CALL_SUBTEST_7( linearStructure(MatrixXi (internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); CALL_SUBTEST_8( linearStructure(MatrixXcd(internal::random(1,EIGEN_TEST_MAX_SIZE/2), internal::random(1,EIGEN_TEST_MAX_SIZE/2))) ); CALL_SUBTEST_9( linearStructure(ArrayXXf (internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + CALL_SUBTEST_10( linearStructure(ArrayXXcf (internal::random(1,EIGEN_TEST_MAX_SIZE), internal::random(1,EIGEN_TEST_MAX_SIZE))) ); + + CALL_SUBTEST_11( real_complex() ); + CALL_SUBTEST_11( real_complex(10,10) ); + CALL_SUBTEST_11( real_complex(10,10) ); } + CALL_SUBTEST_4( linearstructure_overflow<0>() ); } diff --git a/thirdparty/eigen/test/lscg.cpp b/thirdparty/eigen/test/lscg.cpp new file mode 100644 index 00000000..feb2347a --- /dev/null +++ b/thirdparty/eigen/test/lscg.cpp @@ -0,0 +1,37 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2011 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "sparse_solver.h" +#include + +template void test_lscg_T() +{ + LeastSquaresConjugateGradient > lscg_colmajor_diag; + LeastSquaresConjugateGradient, IdentityPreconditioner> lscg_colmajor_I; + LeastSquaresConjugateGradient > lscg_rowmajor_diag; + LeastSquaresConjugateGradient, IdentityPreconditioner> lscg_rowmajor_I; + + CALL_SUBTEST( check_sparse_square_solving(lscg_colmajor_diag) ); + CALL_SUBTEST( check_sparse_square_solving(lscg_colmajor_I) ); + + CALL_SUBTEST( check_sparse_leastsquare_solving(lscg_colmajor_diag) ); + CALL_SUBTEST( check_sparse_leastsquare_solving(lscg_colmajor_I) ); + + CALL_SUBTEST( check_sparse_square_solving(lscg_rowmajor_diag) ); + CALL_SUBTEST( check_sparse_square_solving(lscg_rowmajor_I) ); + + CALL_SUBTEST( check_sparse_leastsquare_solving(lscg_rowmajor_diag) ); + CALL_SUBTEST( check_sparse_leastsquare_solving(lscg_rowmajor_I) ); +} + +EIGEN_DECLARE_TEST(lscg) +{ + CALL_SUBTEST_1(test_lscg_T()); + CALL_SUBTEST_2(test_lscg_T >()); +} diff --git a/thirdparty/eigen/test/lu.cpp b/thirdparty/eigen/test/lu.cpp index 37465269..1bbadcbf 100644 --- a/thirdparty/eigen/test/lu.cpp +++ b/thirdparty/eigen/test/lu.cpp @@ -9,11 +9,18 @@ #include "main.h" #include +#include "solverbase.h" using namespace std; +template +typename MatrixType::RealScalar matrix_l1_norm(const MatrixType& m) { + return m.cwiseAbs().colwise().sum().maxCoeff(); +} + template void lu_non_invertible() { - typedef typename MatrixType::Index Index; + STATIC_CHECK(( internal::is_same::StorageIndex,int>::value )); + typedef typename MatrixType::RealScalar RealScalar; /* this test covers the following files: LU.h @@ -53,6 +60,10 @@ template void lu_non_invertible() // The image of the zero matrix should consist of a single (zero) column vector VERIFY((MatrixType::Zero(rows,cols).fullPivLu().image(MatrixType::Zero(rows,cols)).cols() == 1)); + // The kernel of the zero matrix is the entire space, and thus is an invertible matrix of dimensions cols. + KernelMatrixType kernel = MatrixType::Zero(rows,cols).fullPivLu().kernel(); + VERIFY((kernel.fullPivLu().isInvertible())); + MatrixType m1(rows, cols), m3(rows, cols2); CMatrixType m2(cols, cols2); createRandomPIMatrixOfRank(rank, rows, cols, m1); @@ -82,10 +93,12 @@ template void lu_non_invertible() VERIFY(!lu.isInjective()); VERIFY(!lu.isInvertible()); VERIFY(!lu.isSurjective()); - VERIFY((m1 * m1kernel).isMuchSmallerThan(m1)); + VERIFY_IS_MUCH_SMALLER_THAN((m1 * m1kernel), m1); VERIFY(m1image.fullPivLu().rank() == rank); VERIFY_IS_APPROX(m1 * m1.adjoint() * m1image, m1image); + check_solverbase(m1, lu, rows, cols, cols2); + m2 = CMatrixType::Random(cols,cols2); m3 = m1*m2; m2 = CMatrixType::Random(cols,cols2); @@ -97,12 +110,12 @@ template void lu_non_invertible() template void lu_invertible() { /* this test covers the following files: - LU.h + FullPivLU.h */ typedef typename NumTraits::Real RealScalar; - DenseIndex size = MatrixType::RowsAtCompileTime; + Index size = MatrixType::RowsAtCompileTime; if( size==Dynamic) - size = internal::random(1,EIGEN_TEST_MAX_SIZE); + size = internal::random(1,EIGEN_TEST_MAX_SIZE); MatrixType m1(size, size), m2(size, size), m3(size, size); FullPivLU lu; @@ -120,30 +133,51 @@ template void lu_invertible() VERIFY(lu.isSurjective()); VERIFY(lu.isInvertible()); VERIFY(lu.image(m1).fullPivLu().isInvertible()); + + check_solverbase(m1, lu, size, size, size); + + MatrixType m1_inverse = lu.inverse(); m3 = MatrixType::Random(size,size); m2 = lu.solve(m3); - VERIFY_IS_APPROX(m3, m1*m2); - VERIFY_IS_APPROX(m2, lu.inverse()*m3); + VERIFY_IS_APPROX(m2, m1_inverse*m3); + + RealScalar rcond = (RealScalar(1) / matrix_l1_norm(m1)) / matrix_l1_norm(m1_inverse); + const RealScalar rcond_est = lu.rcond(); + // Verify that the estimated condition number is within a factor of 10 of the + // truth. + VERIFY(rcond_est > rcond / 10 && rcond_est < rcond * 10); // Regression test for Bug 302 MatrixType m4 = MatrixType::Random(size,size); VERIFY_IS_APPROX(lu.solve(m3*m4), lu.solve(m3)*m4); } -template void lu_partial_piv() +template void lu_partial_piv(Index size = MatrixType::ColsAtCompileTime) { /* this test covers the following files: PartialPivLU.h */ - typedef typename MatrixType::Index Index; - Index rows = internal::random(1,4); - Index cols = rows; + typedef typename NumTraits::Real RealScalar; - MatrixType m1(cols, rows); + MatrixType m1(size, size), m2(size, size), m3(size, size); m1.setRandom(); PartialPivLU plu(m1); + STATIC_CHECK(( internal::is_same::StorageIndex,int>::value )); + VERIFY_IS_APPROX(m1, plu.reconstructedMatrix()); + + check_solverbase(m1, plu, size, size, size); + + MatrixType m1_inverse = plu.inverse(); + m3 = MatrixType::Random(size,size); + m2 = plu.solve(m3); + VERIFY_IS_APPROX(m2, m1_inverse*m3); + + RealScalar rcond = (RealScalar(1) / matrix_l1_norm(m1)) / matrix_l1_norm(m1_inverse); + const RealScalar rcond_est = plu.rcond(); + // Verify that the estimate is within a factor of 10 of the truth. + VERIFY(rcond_est > rcond / 10 && rcond_est < rcond * 10); } template void lu_verify_assert() @@ -157,6 +191,8 @@ template void lu_verify_assert() VERIFY_RAISES_ASSERT(lu.kernel()) VERIFY_RAISES_ASSERT(lu.image(tmp)) VERIFY_RAISES_ASSERT(lu.solve(tmp)) + VERIFY_RAISES_ASSERT(lu.transpose().solve(tmp)) + VERIFY_RAISES_ASSERT(lu.adjoint().solve(tmp)) VERIFY_RAISES_ASSERT(lu.determinant()) VERIFY_RAISES_ASSERT(lu.rank()) VERIFY_RAISES_ASSERT(lu.dimensionOfKernel()) @@ -169,19 +205,25 @@ template void lu_verify_assert() VERIFY_RAISES_ASSERT(plu.matrixLU()) VERIFY_RAISES_ASSERT(plu.permutationP()) VERIFY_RAISES_ASSERT(plu.solve(tmp)) + VERIFY_RAISES_ASSERT(plu.transpose().solve(tmp)) + VERIFY_RAISES_ASSERT(plu.adjoint().solve(tmp)) VERIFY_RAISES_ASSERT(plu.determinant()) VERIFY_RAISES_ASSERT(plu.inverse()) } -void test_lu() +EIGEN_DECLARE_TEST(lu) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( lu_non_invertible() ); CALL_SUBTEST_1( lu_invertible() ); CALL_SUBTEST_1( lu_verify_assert() ); + CALL_SUBTEST_1( lu_partial_piv() ); CALL_SUBTEST_2( (lu_non_invertible >()) ); CALL_SUBTEST_2( (lu_verify_assert >()) ); + CALL_SUBTEST_2( lu_partial_piv() ); + CALL_SUBTEST_2( lu_partial_piv() ); + CALL_SUBTEST_2( (lu_partial_piv >()) ); CALL_SUBTEST_3( lu_non_invertible() ); CALL_SUBTEST_3( lu_invertible() ); @@ -189,7 +231,7 @@ void test_lu() CALL_SUBTEST_4( lu_non_invertible() ); CALL_SUBTEST_4( lu_invertible() ); - CALL_SUBTEST_4( lu_partial_piv() ); + CALL_SUBTEST_4( lu_partial_piv(internal::random(1,EIGEN_TEST_MAX_SIZE)) ); CALL_SUBTEST_4( lu_verify_assert() ); CALL_SUBTEST_5( lu_non_invertible() ); @@ -198,7 +240,7 @@ void test_lu() CALL_SUBTEST_6( lu_non_invertible() ); CALL_SUBTEST_6( lu_invertible() ); - CALL_SUBTEST_6( lu_partial_piv() ); + CALL_SUBTEST_6( lu_partial_piv(internal::random(1,EIGEN_TEST_MAX_SIZE)) ); CALL_SUBTEST_6( lu_verify_assert() ); CALL_SUBTEST_7(( lu_non_invertible >() )); diff --git a/thirdparty/eigen/test/main.h b/thirdparty/eigen/test/main.h index 66420486..19bbf1b8 100644 --- a/thirdparty/eigen/test/main.h +++ b/thirdparty/eigen/test/main.h @@ -1,3 +1,4 @@ + // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // @@ -17,6 +18,7 @@ #include #include #include +#include // The following includes of STL headers have to be done _before_ the // definition of macros min() and max(). The reason is that many STL @@ -38,27 +40,111 @@ // definitions. #include #include +// Disable ICC's std::complex operator specializations so we can use our own. +#define _OVERRIDE_COMPLEX_SPECIALIZATION_ 1 #include #include #include +#include #include +#if __cplusplus >= 201103L || (defined(_MSVC_LANG) && _MSVC_LANG >= 201103L) +#include +#include +#ifdef EIGEN_USE_THREADS +#include +#endif +#endif + +// Same for cuda_fp16.h +#if defined(__CUDACC__) && !defined(EIGEN_NO_CUDA) + // Means the compiler is either nvcc or clang with CUDA enabled + #define EIGEN_CUDACC __CUDACC__ +#endif +#if defined(EIGEN_CUDACC) +#include + #define EIGEN_CUDA_SDK_VER (CUDA_VERSION * 10) +#else + #define EIGEN_CUDA_SDK_VER 0 +#endif +#if EIGEN_CUDA_SDK_VER >= 70500 +#include +#endif // To test that all calls from Eigen code to std::min() and std::max() are // protected by parenthesis against macro expansion, the min()/max() macros // are defined here and any not-parenthesized min/max call will cause a // compiler error. -#define min(A,B) please_protect_your_min_with_parentheses -#define max(A,B) please_protect_your_max_with_parentheses +#if !defined(__HIPCC__) && !defined(EIGEN_USE_SYCL) + // + // HIP header files include the following files + // + // + // + // which seem to contain not-parenthesized calls to "max"/"min", triggering the following check and causing the compile to fail + // + // Including those header files before the following macro definition for "min" / "max", only partially resolves the issue + // This is because other HIP header files also define "isnan" / "isinf" / "isfinite" functions, which are needed in other + // headers. + // + // So instead choosing to simply disable this check for HIP + // + #define min(A,B) please_protect_your_min_with_parentheses + #define max(A,B) please_protect_your_max_with_parentheses + #define isnan(X) please_protect_your_isnan_with_parentheses + #define isinf(X) please_protect_your_isinf_with_parentheses + #define isfinite(X) please_protect_your_isfinite_with_parentheses +#endif + + +// test possible conflicts +struct real {}; +struct imag {}; + +#ifdef M_PI +#undef M_PI +#endif +#define M_PI please_use_EIGEN_PI_instead_of_M_PI #define FORBIDDEN_IDENTIFIER (this_identifier_is_forbidden_to_avoid_clashes) this_identifier_is_forbidden_to_avoid_clashes // B0 is defined in POSIX header termios.h #define B0 FORBIDDEN_IDENTIFIER +// `I` may be defined by complex.h: +#define I FORBIDDEN_IDENTIFIER +// _res is defined by resolv.h +#define _res FORBIDDEN_IDENTIFIER + +// Unit tests calling Eigen's blas library must preserve the default blocking size +// to avoid troubles. +#ifndef EIGEN_NO_DEBUG_SMALL_PRODUCT_BLOCKS +#define EIGEN_DEBUG_SMALL_PRODUCT_BLOCKS +#endif // shuts down ICC's remark #593: variable "XXX" was set but never used -#define TEST_SET_BUT_UNUSED_VARIABLE(X) X = X + 0; +#define TEST_SET_BUT_UNUSED_VARIABLE(X) EIGEN_UNUSED_VARIABLE(X) + +#ifdef TEST_ENABLE_TEMPORARY_TRACKING + +static long int nb_temporaries; +static long int nb_temporaries_on_assert = -1; + +inline void on_temporary_creation(long int size) { + // here's a great place to set a breakpoint when debugging failures in this test! + if(size!=0) nb_temporaries++; + if(nb_temporaries_on_assert>0) assert(nb_temporaries g_test_stack; - static int g_repeat; - static unsigned int g_seed; - static bool g_has_set_repeat, g_has_set_seed; + // level == 0 <=> abort if test fail + // level >= 1 <=> warning message to std::cerr if test fail + static int g_test_level = 0; + static int g_repeat = 1; + static unsigned int g_seed = 0; + static bool g_has_set_repeat = false, g_has_set_seed = false; + + class EigenTest + { + public: + EigenTest() : m_func(0) {} + EigenTest(const char* a_name, void (*func)(void)) + : m_name(a_name), m_func(func) + { + get_registered_tests().push_back(this); + } + const std::string& name() const { return m_name; } + void operator()() const { m_func(); } + + static const std::vector& all() { return get_registered_tests(); } + protected: + static std::vector& get_registered_tests() + { + static std::vector* ms_registered_tests = new std::vector(); + return *ms_registered_tests; + } + std::string m_name; + void (*m_func)(void); + }; + + // Declare and register a test, e.g.: + // EIGEN_DECLARE_TEST(mytest) { ... } + // will create a function: + // void test_mytest() { ... } + // that will be automatically called. + #define EIGEN_DECLARE_TEST(X) \ + void EIGEN_CAT(test_,X) (); \ + static EigenTest EIGEN_CAT(test_handler_,X) (EIGEN_MAKESTRING(X), & EIGEN_CAT(test_,X)); \ + void EIGEN_CAT(test_,X) () } -#define EI_PP_MAKE_STRING2(S) #S -#define EI_PP_MAKE_STRING(S) EI_PP_MAKE_STRING2(S) +#define TRACK std::cerr << __FILE__ << " " << __LINE__ << std::endl +// #define TRACK while() #define EIGEN_DEFAULT_IO_FORMAT IOFormat(4, 0, " ", "\n", "", "", "", "") +#if (defined(_CPPUNWIND) || defined(__EXCEPTIONS)) && !defined(__CUDA_ARCH__) && !defined(__HIP_DEVICE_COMPILE__) && !defined(__SYCL_DEVICE_ONLY__) + #define EIGEN_EXCEPTIONS +#endif + #ifndef EIGEN_NO_ASSERTION_CHECKING namespace Eigen @@ -111,9 +233,15 @@ namespace Eigen eigen_assert_exception(void) {} ~eigen_assert_exception() { Eigen::no_more_assert = false; } }; + + struct eigen_static_assert_exception + { + eigen_static_assert_exception(void) {} + ~eigen_static_assert_exception() { Eigen::no_more_assert = false; } + }; } // If EIGEN_DEBUG_ASSERTS is defined and if no assertion is triggered while - // one should have been, then the list of excecuted assertions is printed out. + // one should have been, then the list of executed assertions is printed out. // // EIGEN_DEBUG_ASSERTS is not enabled by default as it // significantly increases the compilation time @@ -135,33 +263,35 @@ namespace Eigen if(report_on_cerr_on_assert_failure) \ std::cerr << #a << " " __FILE__ << "(" << __LINE__ << ")\n"; \ Eigen::no_more_assert = true; \ - throw Eigen::eigen_assert_exception(); \ + EIGEN_THROW_X(Eigen::eigen_assert_exception()); \ } \ else if (Eigen::internal::push_assert) \ { \ - eigen_assert_list.push_back(std::string(EI_PP_MAKE_STRING(__FILE__) " (" EI_PP_MAKE_STRING(__LINE__) ") : " #a) ); \ + eigen_assert_list.push_back(std::string(EIGEN_MAKESTRING(__FILE__) " (" EIGEN_MAKESTRING(__LINE__) ") : " #a) ); \ } + #ifdef EIGEN_EXCEPTIONS #define VERIFY_RAISES_ASSERT(a) \ { \ Eigen::no_more_assert = false; \ - Eigen::eigen_assert_list.clear(); \ - Eigen::internal::push_assert = true; \ + Eigen::eigen_assert_list.clear(); \ + Eigen::internal::push_assert = true; \ Eigen::report_on_cerr_on_assert_failure = false; \ try { \ a; \ std::cerr << "One of the following asserts should have been triggered:\n"; \ - for (uint ai=0 ; ai // required for createRandomPIMatrixOfRank @@ -201,6 +366,8 @@ inline void verify_impl(bool condition, const char *testname, const char *file, { if (!condition) { + if(Eigen::g_test_level>0) + std::cerr << "WARNING: "; std::cerr << "Test " << testname << " failed in " << file << " (" << line << ")" << std::endl << " " << condition_as_string << std::endl; std::cerr << "Stack:\n"; @@ -208,24 +375,34 @@ inline void verify_impl(bool condition, const char *testname, const char *file, for(int i=test_stack_size-1; i>=0; --i) std::cerr << " - " << Eigen::g_test_stack[i] << "\n"; std::cerr << "\n"; - abort(); + if(Eigen::g_test_level==0) + abort(); } } -#define VERIFY(a) ::verify_impl(a, g_test_stack.back().c_str(), __FILE__, __LINE__, EI_PP_MAKE_STRING(a)) +#define VERIFY(a) ::verify_impl(a, g_test_stack.back().c_str(), __FILE__, __LINE__, EIGEN_MAKESTRING(a)) + +#define VERIFY_GE(a, b) ::verify_impl(a >= b, g_test_stack.back().c_str(), __FILE__, __LINE__, EIGEN_MAKESTRING(a >= b)) +#define VERIFY_LE(a, b) ::verify_impl(a <= b, g_test_stack.back().c_str(), __FILE__, __LINE__, EIGEN_MAKESTRING(a <= b)) + -#define VERIFY_IS_EQUAL(a, b) VERIFY(test_is_equal(a, b)) -#define VERIFY_IS_APPROX(a, b) VERIFY(test_isApprox(a, b)) +#define VERIFY_IS_EQUAL(a, b) VERIFY(test_is_equal(a, b, true)) +#define VERIFY_IS_NOT_EQUAL(a, b) VERIFY(test_is_equal(a, b, false)) +#define VERIFY_IS_APPROX(a, b) VERIFY(verifyIsApprox(a, b)) #define VERIFY_IS_NOT_APPROX(a, b) VERIFY(!test_isApprox(a, b)) #define VERIFY_IS_MUCH_SMALLER_THAN(a, b) VERIFY(test_isMuchSmallerThan(a, b)) #define VERIFY_IS_NOT_MUCH_SMALLER_THAN(a, b) VERIFY(!test_isMuchSmallerThan(a, b)) #define VERIFY_IS_APPROX_OR_LESS_THAN(a, b) VERIFY(test_isApproxOrLessThan(a, b)) #define VERIFY_IS_NOT_APPROX_OR_LESS_THAN(a, b) VERIFY(!test_isApproxOrLessThan(a, b)) +#define VERIFY_IS_CWISE_EQUAL(a, b) VERIFY(verifyIsCwiseApprox(a, b, true)) +#define VERIFY_IS_CWISE_APPROX(a, b) VERIFY(verifyIsCwiseApprox(a, b, false)) #define VERIFY_IS_UNITARY(a) VERIFY(test_isUnitary(a)) +#define STATIC_CHECK(COND) EIGEN_STATIC_ASSERT( (COND) , EIGEN_INTERNAL_ERROR_PLEASE_FILE_A_BUG_REPORT ) + #define CALL_SUBTEST(FUNC) do { \ - g_test_stack.push_back(EI_PP_MAKE_STRING(FUNC)); \ + g_test_stack.push_back(EIGEN_MAKESTRING(FUNC)); \ FUNC; \ g_test_stack.pop_back(); \ } while (0) @@ -233,34 +410,53 @@ inline void verify_impl(bool condition, const char *testname, const char *file, namespace Eigen { +template +typename internal::enable_if::value,bool>::type +is_same_type(const T1&, const T2&) +{ + return true; +} + template inline typename NumTraits::Real test_precision() { return NumTraits::dummy_precision(); } template<> inline float test_precision() { return 1e-3f; } template<> inline double test_precision() { return 1e-6; } +template<> inline long double test_precision() { return 1e-6l; } template<> inline float test_precision >() { return test_precision(); } template<> inline double test_precision >() { return test_precision(); } -template<> inline long double test_precision() { return 1e-6; } - -inline bool test_isApprox(const int& a, const int& b) -{ return internal::isApprox(a, b, test_precision()); } -inline bool test_isMuchSmallerThan(const int& a, const int& b) -{ return internal::isMuchSmallerThan(a, b, test_precision()); } -inline bool test_isApproxOrLessThan(const int& a, const int& b) -{ return internal::isApproxOrLessThan(a, b, test_precision()); } - -inline bool test_isApprox(const float& a, const float& b) -{ return internal::isApprox(a, b, test_precision()); } -inline bool test_isMuchSmallerThan(const float& a, const float& b) -{ return internal::isMuchSmallerThan(a, b, test_precision()); } -inline bool test_isApproxOrLessThan(const float& a, const float& b) -{ return internal::isApproxOrLessThan(a, b, test_precision()); } -inline bool test_isApprox(const double& a, const double& b) -{ return internal::isApprox(a, b, test_precision()); } - -inline bool test_isMuchSmallerThan(const double& a, const double& b) -{ return internal::isMuchSmallerThan(a, b, test_precision()); } -inline bool test_isApproxOrLessThan(const double& a, const double& b) -{ return internal::isApproxOrLessThan(a, b, test_precision()); } +template<> inline long double test_precision >() { return test_precision(); } + +#define EIGEN_TEST_SCALAR_TEST_OVERLOAD(TYPE) \ + inline bool test_isApprox(TYPE a, TYPE b) \ + { return numext::equal_strict(a, b) || \ + ((numext::isnan)(a) && (numext::isnan)(b)) || \ + (internal::isApprox(a, b, test_precision())); } \ + inline bool test_isCwiseApprox(TYPE a, TYPE b, bool exact) \ + { return numext::equal_strict(a, b) || \ + ((numext::isnan)(a) && (numext::isnan)(b)) || \ + (!exact && internal::isApprox(a, b, test_precision())); } \ + inline bool test_isMuchSmallerThan(TYPE a, TYPE b) \ + { return internal::isMuchSmallerThan(a, b, test_precision()); } \ + inline bool test_isApproxOrLessThan(TYPE a, TYPE b) \ + { return internal::isApproxOrLessThan(a, b, test_precision()); } + +EIGEN_TEST_SCALAR_TEST_OVERLOAD(short) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(unsigned short) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(int) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(unsigned int) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(long) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(unsigned long) +#if EIGEN_HAS_CXX11 +EIGEN_TEST_SCALAR_TEST_OVERLOAD(long long) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(unsigned long long) +#endif +EIGEN_TEST_SCALAR_TEST_OVERLOAD(float) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(double) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(half) +EIGEN_TEST_SCALAR_TEST_OVERLOAD(bfloat16) + +#undef EIGEN_TEST_SCALAR_TEST_OVERLOAD +#ifndef EIGEN_TEST_NO_COMPLEX inline bool test_isApprox(const std::complex& a, const std::complex& b) { return internal::isApprox(a, b, test_precision >()); } inline bool test_isMuchSmallerThan(const std::complex& a, const std::complex& b) @@ -271,6 +467,15 @@ inline bool test_isApprox(const std::complex& a, const std::complex& a, const std::complex& b) { return internal::isMuchSmallerThan(a, b, test_precision >()); } +#ifndef EIGEN_TEST_NO_LONGDOUBLE +inline bool test_isApprox(const std::complex& a, const std::complex& b) +{ return internal::isApprox(a, b, test_precision >()); } +inline bool test_isMuchSmallerThan(const std::complex& a, const std::complex& b) +{ return internal::isMuchSmallerThan(a, b, test_precision >()); } +#endif +#endif + +#ifndef EIGEN_TEST_NO_LONGDOUBLE inline bool test_isApprox(const long double& a, const long double& b) { bool ret = internal::isApprox(a, b, test_precision()); @@ -284,13 +489,136 @@ inline bool test_isMuchSmallerThan(const long double& a, const long double& b) { return internal::isMuchSmallerThan(a, b, test_precision()); } inline bool test_isApproxOrLessThan(const long double& a, const long double& b) { return internal::isApproxOrLessThan(a, b, test_precision()); } +#endif // EIGEN_TEST_NO_LONGDOUBLE + +// test_relative_error returns the relative difference between a and b as a real scalar as used in isApprox. +template +typename NumTraits::NonInteger test_relative_error(const EigenBase &a, const EigenBase &b) +{ + using std::sqrt; + typedef typename NumTraits::NonInteger RealScalar; + typename internal::nested_eval::type ea(a.derived()); + typename internal::nested_eval::type eb(b.derived()); + return sqrt(RealScalar((ea-eb).cwiseAbs2().sum()) / RealScalar((std::min)(eb.cwiseAbs2().sum(),ea.cwiseAbs2().sum()))); +} + +template +typename T1::RealScalar test_relative_error(const T1 &a, const T2 &b, const typename T1::Coefficients* = 0) +{ + return test_relative_error(a.coeffs(), b.coeffs()); +} + +template +typename T1::Scalar test_relative_error(const T1 &a, const T2 &b, const typename T1::MatrixType* = 0) +{ + return test_relative_error(a.matrix(), b.matrix()); +} + +template +S test_relative_error(const Translation &a, const Translation &b) +{ + return test_relative_error(a.vector(), b.vector()); +} + +template +S test_relative_error(const ParametrizedLine &a, const ParametrizedLine &b) +{ + return (std::max)(test_relative_error(a.origin(), b.origin()), test_relative_error(a.origin(), b.origin())); +} + +template +S test_relative_error(const AlignedBox &a, const AlignedBox &b) +{ + return (std::max)(test_relative_error((a.min)(), (b.min)()), test_relative_error((a.max)(), (b.max)())); +} + +template class SparseMatrixBase; +template +typename T1::RealScalar test_relative_error(const MatrixBase &a, const SparseMatrixBase &b) +{ + return test_relative_error(a,b.toDense()); +} + +template class SparseMatrixBase; +template +typename T1::RealScalar test_relative_error(const SparseMatrixBase &a, const MatrixBase &b) +{ + return test_relative_error(a.toDense(),b); +} + +template class SparseMatrixBase; +template +typename T1::RealScalar test_relative_error(const SparseMatrixBase &a, const SparseMatrixBase &b) +{ + return test_relative_error(a.toDense(),b.toDense()); +} + +template +typename NumTraits::Real>::NonInteger test_relative_error(const T1 &a, const T2 &b, typename internal::enable_if::Real>::value, T1>::type* = 0) +{ + typedef typename NumTraits::Real>::NonInteger RealScalar; + return numext::sqrt(RealScalar(numext::abs2(a-b))/(numext::mini)(RealScalar(numext::abs2(a)),RealScalar(numext::abs2(b)))); +} + +template +T test_relative_error(const Rotation2D &a, const Rotation2D &b) +{ + return test_relative_error(a.angle(), b.angle()); +} + +template +T test_relative_error(const AngleAxis &a, const AngleAxis &b) +{ + return (std::max)(test_relative_error(a.angle(), b.angle()), test_relative_error(a.axis(), b.axis())); +} template -inline bool test_isApprox(const Type1& a, const Type2& b) +inline bool test_isApprox(const Type1& a, const Type2& b, typename Type1::Scalar* = 0) // Enabled for Eigen's type only { return a.isApprox(b, test_precision()); } +// get_test_precision is a small wrapper to test_precision allowing to return the scalar precision for either scalars or expressions +template +typename NumTraits::Real get_test_precision(const T&, const typename T::Scalar* = 0) +{ + return test_precision::Real>(); +} + +template +typename NumTraits::Real get_test_precision(const T&,typename internal::enable_if::Real>::value, T>::type* = 0) +{ + return test_precision::Real>(); +} + +// verifyIsApprox is a wrapper to test_isApprox that outputs the relative difference magnitude if the test fails. +template +inline bool verifyIsApprox(const Type1& a, const Type2& b) +{ + bool ret = test_isApprox(a,b); + if(!ret) + { + std::cerr << "Difference too large wrt tolerance " << get_test_precision(a) << ", relative error is: " << test_relative_error(a,b) << std::endl; + } + return ret; +} + +// verifyIsCwiseApprox is a wrapper to test_isCwiseApprox that outputs the relative difference magnitude if the test fails. +template +inline bool verifyIsCwiseApprox(const Type1& a, const Type2& b, bool exact) +{ + bool ret = test_isCwiseApprox(a,b,exact); + if(!ret) { + if (exact) { + std::cerr << "Values are not an exact match"; + } else { + std::cerr << "Difference too large wrt tolerance " << get_test_precision(a); + } + std::cerr << ", relative error is: " << test_relative_error(a,b) << std::endl; + } + return ret; +} + // The idea behind this function is to compare the two scalars a and b where // the scalar ref is a hint about the expected order of magnitude of a and b. // WARNING: the scalar a and b must be positive @@ -326,17 +654,17 @@ inline bool test_isUnitary(const MatrixBase& m) // Forward declaration to avoid ICC warning template -bool test_is_equal(const T& actual, const U& expected); +bool test_is_equal(const T& actual, const U& expected, bool expect_equal=true); template -bool test_is_equal(const T& actual, const U& expected) +bool test_is_equal(const T& actual, const U& expected, bool expect_equal) { - if (actual==expected) + if ((actual==expected) == expect_equal) return true; // false: std::cerr - << std::endl << " actual = " << actual - << std::endl << " expected = " << expected << std::endl << std::endl; + << "\n actual = " << actual + << "\n expected " << (expect_equal ? "= " : "!=") << expected << "\n\n"; return false; } @@ -347,11 +675,10 @@ bool test_is_equal(const T& actual, const U& expected) */ // Forward declaration to avoid ICC warning template -void createRandomPIMatrixOfRank(typename MatrixType::Index desired_rank, typename MatrixType::Index rows, typename MatrixType::Index cols, MatrixType& m); +void createRandomPIMatrixOfRank(Index desired_rank, Index rows, Index cols, MatrixType& m); template -void createRandomPIMatrixOfRank(typename MatrixType::Index desired_rank, typename MatrixType::Index rows, typename MatrixType::Index cols, MatrixType& m) +void createRandomPIMatrixOfRank(Index desired_rank, Index rows, Index cols, MatrixType& m) { - typedef typename internal::traits::Index Index; typedef typename internal::traits::Scalar Scalar; enum { Rows = MatrixType::RowsAtCompileTime, Cols = MatrixType::ColsAtCompileTime }; @@ -388,11 +715,10 @@ void createRandomPIMatrixOfRank(typename MatrixType::Index desired_rank, typenam // Forward declaration to avoid ICC warning template -void randomPermutationVector(PermutationVectorType& v, typename PermutationVectorType::Index size); +void randomPermutationVector(PermutationVectorType& v, Index size); template -void randomPermutationVector(PermutationVectorType& v, typename PermutationVectorType::Index size) +void randomPermutationVector(PermutationVectorType& v, Index size) { - typedef typename PermutationVectorType::Index Index; typedef typename PermutationVectorType::Scalar Scalar; v.resize(size); for(Index i = 0; i < size; ++i) v(i) = Scalar(i); @@ -411,12 +737,7 @@ template bool isNotNaN(const T& x) return x==x; } -template bool isNaN(const T& x) -{ - return x!=x; -} - -template bool isInf(const T& x) +template bool isPlusInf(const T& x) { return x > NumTraits::highest(); } @@ -437,16 +758,15 @@ template struct GetDifferentType > // Forward declaration to avoid ICC warning template std::string type_name(); -template std::string type_name() { return "other"; } -template<> std::string type_name() { return "float"; } -template<> std::string type_name() { return "double"; } -template<> std::string type_name() { return "int"; } -template<> std::string type_name >() { return "complex"; } -template<> std::string type_name >() { return "complex"; } -template<> std::string type_name >() { return "complex"; } - -// forward declaration of the main test function -void EIGEN_CAT(test_,EIGEN_TEST_FUNC)(); +template std::string type_name() { return "other"; } +template<> std::string type_name() { return "float"; } +template<> std::string type_name() { return "double"; } +template<> std::string type_name() { return "long double"; } +template<> std::string type_name() { return "int"; } +template<> std::string type_name >() { return "complex"; } +template<> std::string type_name >() { return "complex"; } +template<> std::string type_name >() { return "complex"; } +template<> std::string type_name >() { return "complex"; } using namespace Eigen; @@ -534,9 +854,16 @@ int main(int argc, char *argv[]) srand(g_seed); std::cout << "Repeating each test " << g_repeat << " times" << std::endl; - Eigen::g_test_stack.push_back(std::string(EI_PP_MAKE_STRING(EIGEN_TEST_FUNC))); + VERIFY(EigenTest::all().size()>0); + + for(std::size_t i=0; i void map_class_vector(const VectorType& m) { - typedef typename VectorType::Index Index; typedef typename VectorType::Scalar Scalar; Index size = m.size(); - // test Map.h Scalar* array1 = internal::aligned_new(size); Scalar* array2 = internal::aligned_new(size); Scalar* array3 = new Scalar[size+1]; - Scalar* array3unaligned = size_t(array3)%16 == 0 ? array3+1 : array3; + Scalar* array3unaligned = (internal::UIntPtr(array3)%EIGEN_MAX_ALIGN_BYTES) == 0 ? array3+1 : array3; + Scalar array4[EIGEN_TESTMAP_MAX_SIZE]; - Map(array1, size) = VectorType::Random(size); - Map(array2, size) = Map(array1, size); + Map(array1, size) = VectorType::Random(size); + Map(array2, size) = Map(array1, size); Map(array3unaligned, size) = Map(array1, size); - VectorType ma1 = Map(array1, size); - VectorType ma2 = Map(array2, size); + Map(array4, size) = Map(array1, size); + VectorType ma1 = Map(array1, size); + VectorType ma2 = Map(array2, size); VectorType ma3 = Map(array3unaligned, size); + VectorType ma4 = Map(array4, size); VERIFY_IS_EQUAL(ma1, ma2); VERIFY_IS_EQUAL(ma1, ma3); + VERIFY_IS_EQUAL(ma1, ma4); #ifdef EIGEN_VECTORIZE - if(internal::packet_traits::Vectorizable) - VERIFY_RAISES_ASSERT((Map(array3unaligned, size))) + if(internal::packet_traits::Vectorizable && size>=AlignedMax) + VERIFY_RAISES_ASSERT((Map(array3unaligned, size))) #endif internal::aligned_delete(array1, size); @@ -46,27 +50,68 @@ template void map_class_vector(const VectorType& m) template void map_class_matrix(const MatrixType& m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; Index rows = m.rows(), cols = m.cols(), size = rows*cols; + Scalar s1 = internal::random(); - // test Map.h + // array1 and array2 -> aligned heap allocation Scalar* array1 = internal::aligned_new(size); for(int i = 0; i < size; i++) array1[i] = Scalar(1); Scalar* array2 = internal::aligned_new(size); for(int i = 0; i < size; i++) array2[i] = Scalar(1); + // array3unaligned -> unaligned pointer to heap Scalar* array3 = new Scalar[size+1]; - for(int i = 0; i < size+1; i++) array3[i] = Scalar(1); - Scalar* array3unaligned = size_t(array3)%16 == 0 ? array3+1 : array3; - Map(array1, rows, cols) = MatrixType::Ones(rows,cols); - Map(array2, rows, cols) = Map(array1, rows, cols); - Map(array3unaligned, rows, cols) = Map(array1, rows, cols); - MatrixType ma1 = Map(array1, rows, cols); - MatrixType ma2 = Map(array2, rows, cols); + Index sizep1 = size + 1; // <- without this temporary MSVC 2103 generates bad code + for(Index i = 0; i < sizep1; i++) array3[i] = Scalar(1); + Scalar* array3unaligned = (internal::UIntPtr(array3)%EIGEN_MAX_ALIGN_BYTES) == 0 ? array3+1 : array3; + Scalar array4[256]; + if(size<=256) + for(int i = 0; i < size; i++) array4[i] = Scalar(1); + + Map map1(array1, rows, cols); + Map map2(array2, rows, cols); + Map map3(array3unaligned, rows, cols); + Map map4(array4, rows, cols); + + VERIFY_IS_EQUAL(map1, MatrixType::Ones(rows,cols)); + VERIFY_IS_EQUAL(map2, MatrixType::Ones(rows,cols)); + VERIFY_IS_EQUAL(map3, MatrixType::Ones(rows,cols)); + map1 = MatrixType::Random(rows,cols); + map2 = map1; + map3 = map1; + MatrixType ma1 = map1; + MatrixType ma2 = map2; + MatrixType ma3 = map3; + VERIFY_IS_EQUAL(map1, map2); + VERIFY_IS_EQUAL(map1, map3); VERIFY_IS_EQUAL(ma1, ma2); - MatrixType ma3 = Map(array3unaligned, rows, cols); VERIFY_IS_EQUAL(ma1, ma3); + VERIFY_IS_EQUAL(ma1, map3); + + VERIFY_IS_APPROX(s1*map1, s1*map2); + VERIFY_IS_APPROX(s1*ma1, s1*ma2); + VERIFY_IS_EQUAL(s1*ma1, s1*ma3); + VERIFY_IS_APPROX(s1*map1, s1*map3); + + map2 *= s1; + map3 *= s1; + VERIFY_IS_APPROX(s1*map1, map2); + VERIFY_IS_APPROX(s1*map1, map3); + + if(size<=256) + { + VERIFY_IS_EQUAL(map4, MatrixType::Ones(rows,cols)); + map4 = map1; + MatrixType ma4 = map4; + VERIFY_IS_EQUAL(map1, map4); + VERIFY_IS_EQUAL(ma1, map4); + VERIFY_IS_EQUAL(ma1, ma4); + VERIFY_IS_APPROX(s1*map1, s1*map4); + + map4 *= s1; + VERIFY_IS_APPROX(s1*map1, map4); + } internal::aligned_delete(array1, size); internal::aligned_delete(array2, size); @@ -75,16 +120,14 @@ template void map_class_matrix(const MatrixType& m) template void map_static_methods(const VectorType& m) { - typedef typename VectorType::Index Index; typedef typename VectorType::Scalar Scalar; Index size = m.size(); - // test Map.h Scalar* array1 = internal::aligned_new(size); Scalar* array2 = internal::aligned_new(size); Scalar* array3 = new Scalar[size+1]; - Scalar* array3unaligned = size_t(array3)%16 == 0 ? array3+1 : array3; + Scalar* array3unaligned = internal::UIntPtr(array3)%EIGEN_MAX_ALIGN_BYTES == 0 ? array3+1 : array3; VectorType::MapAligned(array1, size) = VectorType::Random(size); VectorType::Map(array2, size) = VectorType::Map(array1, size); @@ -109,16 +152,15 @@ template void check_const_correctness(const PlainObjec // verify that map-to-const don't have LvalueBit typedef typename internal::add_const::type ConstPlainObjectType; VERIFY( !(internal::traits >::Flags & LvalueBit) ); - VERIFY( !(internal::traits >::Flags & LvalueBit) ); + VERIFY( !(internal::traits >::Flags & LvalueBit) ); VERIFY( !(Map::Flags & LvalueBit) ); - VERIFY( !(Map::Flags & LvalueBit) ); + VERIFY( !(Map::Flags & LvalueBit) ); } template void map_not_aligned_on_scalar() { typedef Matrix MatrixType; - typedef typename MatrixType::Index Index; Index size = 11; Scalar* array1 = internal::aligned_new((size+1)*(size+1)+1); Scalar* array2 = reinterpret_cast(sizeof(Scalar)/2+std::size_t(array1)); @@ -136,12 +178,13 @@ void map_not_aligned_on_scalar() internal::aligned_delete(array1, (size+1)*(size+1)+1); } -void test_mapped_matrix() +EIGEN_DECLARE_TEST(mapped_matrix) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( map_class_vector(Matrix()) ); CALL_SUBTEST_1( check_const_correctness(Matrix()) ); CALL_SUBTEST_2( map_class_vector(Vector4d()) ); + CALL_SUBTEST_2( map_class_vector(VectorXd(13)) ); CALL_SUBTEST_2( check_const_correctness(Matrix4d()) ); CALL_SUBTEST_3( map_class_vector(RowVector4f()) ); CALL_SUBTEST_4( map_class_vector(VectorXcf(8)) ); @@ -159,7 +202,6 @@ void test_mapped_matrix() CALL_SUBTEST_8( map_static_methods(RowVector3d()) ); CALL_SUBTEST_9( map_static_methods(VectorXcd(8)) ); CALL_SUBTEST_10( map_static_methods(VectorXf(12)) ); - CALL_SUBTEST_11( map_not_aligned_on_scalar() ); } } diff --git a/thirdparty/eigen/test/mapstaticmethods.cpp b/thirdparty/eigen/test/mapstaticmethods.cpp index 5b512bde..d0128ba9 100644 --- a/thirdparty/eigen/test/mapstaticmethods.cpp +++ b/thirdparty/eigen/test/mapstaticmethods.cpp @@ -9,8 +9,12 @@ #include "main.h" +// GCC<=4.8 has spurious shadow warnings, because `ptr` re-appears inside template instantiations +// workaround: put these in an anonymous namespace +namespace { float *ptr; const float *const_ptr; +} template { static void run(const PlainObjectType& m) { - int rows = m.rows(), cols = m.cols(); + Index rows = m.rows(), cols = m.cols(); int i = internal::random(2,5), j = internal::random(2,5); @@ -115,7 +119,7 @@ struct mapstaticmethods_impl { static void run(const PlainObjectType& v) { - int size = v.size(); + Index size = v.size(); int i = internal::random(2,5); @@ -143,7 +147,7 @@ void mapstaticmethods(const PlainObjectType& m) VERIFY(true); // just to avoid 'unused function' warning } -void test_mapstaticmethods() +EIGEN_DECLARE_TEST(mapstaticmethods) { ptr = internal::aligned_new(1000); for(int i = 0; i < 1000; i++) ptr[i] = float(i); diff --git a/thirdparty/eigen/test/mapstride.cpp b/thirdparty/eigen/test/mapstride.cpp index b1dc9de2..fde73f2e 100644 --- a/thirdparty/eigen/test/mapstride.cpp +++ b/thirdparty/eigen/test/mapstride.cpp @@ -11,7 +11,6 @@ template void map_class_vector(const VectorType& m) { - typedef typename VectorType::Index Index; typedef typename VectorType::Scalar Scalar; Index size = m.size(); @@ -23,7 +22,7 @@ template void map_class_vector(const VectorTy Scalar* a_array = internal::aligned_new(arraysize+1); Scalar* array = a_array; if(Alignment!=Aligned) - array = (Scalar*)(ptrdiff_t(a_array) + (internal::packet_traits::AlignedOnScalar?sizeof(Scalar):sizeof(typename NumTraits::Real))); + array = (Scalar*)(internal::IntPtr(a_array) + (internal::packet_traits::AlignedOnScalar?sizeof(Scalar):sizeof(typename NumTraits::Real))); { Map > map(array, size); @@ -50,22 +49,35 @@ template void map_class_vector(const VectorTy template void map_class_matrix(const MatrixType& _m) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; Index rows = _m.rows(), cols = _m.cols(); MatrixType m = MatrixType::Random(rows,cols); + Scalar s1 = internal::random(); - Index arraysize = 2*(rows+4)*(cols+4); + Index arraysize = 4*(rows+4)*(cols+4); - Scalar* a_array = internal::aligned_new(arraysize+1); - Scalar* array = a_array; + Scalar* a_array1 = internal::aligned_new(arraysize+1); + Scalar* array1 = a_array1; if(Alignment!=Aligned) - array = (Scalar*)(ptrdiff_t(a_array) + (internal::packet_traits::AlignedOnScalar?sizeof(Scalar):sizeof(typename NumTraits::Real))); + array1 = (Scalar*)(internal::IntPtr(a_array1) + (internal::packet_traits::AlignedOnScalar?sizeof(Scalar):sizeof(typename NumTraits::Real))); + Scalar a_array2[256]; + Scalar* array2 = a_array2; + if(Alignment!=Aligned) + array2 = (Scalar*)(internal::IntPtr(a_array2) + (internal::packet_traits::AlignedOnScalar?sizeof(Scalar):sizeof(typename NumTraits::Real))); + else + array2 = (Scalar*)(((internal::UIntPtr(a_array2)+EIGEN_MAX_ALIGN_BYTES-1)/EIGEN_MAX_ALIGN_BYTES)*EIGEN_MAX_ALIGN_BYTES); + Index maxsize2 = a_array2 - array2 + 256; + // test no inner stride and some dynamic outer stride + for(int k=0; k<2; ++k) { + if(k==1 && (m.innerSize()+1)*m.outerSize() > maxsize2) + break; + Scalar* array = (k==0 ? array1 : array2); + Map > map(array, rows, cols, OuterStride(m.innerSize()+1)); map = m; VERIFY(map.outerStride() == map.innerSize()+1); @@ -75,11 +87,19 @@ template void map_class_matrix(const MatrixTy VERIFY(array[map.outerStride()*i+j] == m.coeffByOuterInner(i,j)); VERIFY(map.coeffByOuterInner(i,j) == m.coeffByOuterInner(i,j)); } + VERIFY_IS_APPROX(s1*map,s1*m); + map *= s1; + VERIFY_IS_APPROX(map,s1*m); } // test no inner stride and an outer stride of +4. This is quite important as for fixed-size matrices, // this allows to hit the special case where it's vectorizable. + for(int k=0; k<2; ++k) { + if(k==1 && (m.innerSize()+4)*m.outerSize() > maxsize2) + break; + Scalar* array = (k==0 ? array1 : array2); + enum { InnerSize = MatrixType::InnerSizeAtCompileTime, OuterStrideAtCompileTime = InnerSize==Dynamic ? Dynamic : InnerSize+4 @@ -94,10 +114,18 @@ template void map_class_matrix(const MatrixTy VERIFY(array[map.outerStride()*i+j] == m.coeffByOuterInner(i,j)); VERIFY(map.coeffByOuterInner(i,j) == m.coeffByOuterInner(i,j)); } + VERIFY_IS_APPROX(s1*map,s1*m); + map *= s1; + VERIFY_IS_APPROX(map,s1*m); } // test both inner stride and outer stride + for(int k=0; k<2; ++k) { + if(k==1 && (2*m.innerSize()+1)*(m.outerSize()*2) > maxsize2) + break; + Scalar* array = (k==0 ? array1 : array2); + Map > map(array, rows, cols, Stride(2*m.innerSize()+1, 2)); map = m; VERIFY(map.outerStride() == 2*map.innerSize()+1); @@ -108,15 +136,97 @@ template void map_class_matrix(const MatrixTy VERIFY(array[map.outerStride()*i+map.innerStride()*j] == m.coeffByOuterInner(i,j)); VERIFY(map.coeffByOuterInner(i,j) == m.coeffByOuterInner(i,j)); } + VERIFY_IS_APPROX(s1*map,s1*m); + map *= s1; + VERIFY_IS_APPROX(map,s1*m); } - internal::aligned_delete(a_array, arraysize+1); + // test inner stride and no outer stride + for(int k=0; k<2; ++k) + { + if(k==1 && (m.innerSize()*2)*m.outerSize() > maxsize2) + break; + Scalar* array = (k==0 ? array1 : array2); + + Map > map(array, rows, cols, InnerStride(2)); + map = m; + VERIFY(map.outerStride() == map.innerSize()*2); + for(int i = 0; i < m.outerSize(); ++i) + for(int j = 0; j < m.innerSize(); ++j) + { + VERIFY(array[map.innerSize()*i*2+j*2] == m.coeffByOuterInner(i,j)); + VERIFY(map.coeffByOuterInner(i,j) == m.coeffByOuterInner(i,j)); + } + VERIFY_IS_APPROX(s1*map,s1*m); + map *= s1; + VERIFY_IS_APPROX(map,s1*m); + } + + // test negative strides + { + Matrix::Map(a_array1, arraysize+1).setRandom(); + Index outerstride = m.innerSize()+4; + Scalar* array = array1; + + { + Map > map1(array, rows, cols, OuterStride<>( outerstride)); + Map > map2(array+(m.outerSize()-1)*outerstride, rows, cols, OuterStride<>(-outerstride)); + if(MatrixType::IsRowMajor) VERIFY_IS_APPROX(map1.colwise().reverse(), map2); + else VERIFY_IS_APPROX(map1.rowwise().reverse(), map2); + } + + { + Map > map1(array, rows, cols, OuterStride<>( outerstride)); + Map > map2(array+(m.outerSize()-1)*outerstride+m.innerSize()-1, rows, cols, Stride(-outerstride,-1)); + VERIFY_IS_APPROX(map1.reverse(), map2); + } + + { + Map > map1(array, rows, cols, OuterStride<>( outerstride)); + Map > map2(array+(m.outerSize()-1)*outerstride+m.innerSize()-1, rows, cols, Stride(-outerstride,-1)); + VERIFY_IS_APPROX(map1.reverse(), map2); + } + } + + internal::aligned_delete(a_array1, arraysize+1); } -void test_mapstride() +// Additional tests for inner-stride but no outer-stride +template +void bug1453() +{ + const int data[] = {0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31}; + typedef Matrix RowMatrixXi; + typedef Matrix ColMatrix23i; + typedef Matrix ColMatrix32i; + typedef Matrix RowMatrix23i; + typedef Matrix RowMatrix32i; + + VERIFY_IS_APPROX(MatrixXi::Map(data, 2, 3, InnerStride<2>()), MatrixXi::Map(data, 2, 3, Stride<4,2>())); + VERIFY_IS_APPROX(MatrixXi::Map(data, 2, 3, InnerStride<>(2)), MatrixXi::Map(data, 2, 3, Stride<4,2>())); + VERIFY_IS_APPROX(MatrixXi::Map(data, 3, 2, InnerStride<2>()), MatrixXi::Map(data, 3, 2, Stride<6,2>())); + VERIFY_IS_APPROX(MatrixXi::Map(data, 3, 2, InnerStride<>(2)), MatrixXi::Map(data, 3, 2, Stride<6,2>())); + + VERIFY_IS_APPROX(RowMatrixXi::Map(data, 2, 3, InnerStride<2>()), RowMatrixXi::Map(data, 2, 3, Stride<6,2>())); + VERIFY_IS_APPROX(RowMatrixXi::Map(data, 2, 3, InnerStride<>(2)), RowMatrixXi::Map(data, 2, 3, Stride<6,2>())); + VERIFY_IS_APPROX(RowMatrixXi::Map(data, 3, 2, InnerStride<2>()), RowMatrixXi::Map(data, 3, 2, Stride<4,2>())); + VERIFY_IS_APPROX(RowMatrixXi::Map(data, 3, 2, InnerStride<>(2)), RowMatrixXi::Map(data, 3, 2, Stride<4,2>())); + + VERIFY_IS_APPROX(ColMatrix23i::Map(data, InnerStride<2>()), MatrixXi::Map(data, 2, 3, Stride<4,2>())); + VERIFY_IS_APPROX(ColMatrix23i::Map(data, InnerStride<>(2)), MatrixXi::Map(data, 2, 3, Stride<4,2>())); + VERIFY_IS_APPROX(ColMatrix32i::Map(data, InnerStride<2>()), MatrixXi::Map(data, 3, 2, Stride<6,2>())); + VERIFY_IS_APPROX(ColMatrix32i::Map(data, InnerStride<>(2)), MatrixXi::Map(data, 3, 2, Stride<6,2>())); + + VERIFY_IS_APPROX(RowMatrix23i::Map(data, InnerStride<2>()), RowMatrixXi::Map(data, 2, 3, Stride<6,2>())); + VERIFY_IS_APPROX(RowMatrix23i::Map(data, InnerStride<>(2)), RowMatrixXi::Map(data, 2, 3, Stride<6,2>())); + VERIFY_IS_APPROX(RowMatrix32i::Map(data, InnerStride<2>()), RowMatrixXi::Map(data, 3, 2, Stride<4,2>())); + VERIFY_IS_APPROX(RowMatrix32i::Map(data, InnerStride<>(2)), RowMatrixXi::Map(data, 3, 2, Stride<4,2>())); +} + +EIGEN_DECLARE_TEST(mapstride) { for(int i = 0; i < g_repeat; i++) { - int maxn = 30; + int maxn = 3; CALL_SUBTEST_1( map_class_vector(Matrix()) ); CALL_SUBTEST_1( map_class_vector(Matrix()) ); CALL_SUBTEST_2( map_class_vector(Vector4d()) ); @@ -142,6 +252,8 @@ void test_mapstride() CALL_SUBTEST_5( map_class_matrix(MatrixXi(internal::random(1,maxn),internal::random(1,maxn))) ); CALL_SUBTEST_6( map_class_matrix(MatrixXcd(internal::random(1,maxn),internal::random(1,maxn))) ); CALL_SUBTEST_6( map_class_matrix(MatrixXcd(internal::random(1,maxn),internal::random(1,maxn))) ); + + CALL_SUBTEST_5( bug1453<0>() ); TEST_SET_BUT_UNUSED_VARIABLE(maxn); } diff --git a/thirdparty/eigen/test/meta.cpp b/thirdparty/eigen/test/meta.cpp index 3302c588..03025eec 100644 --- a/thirdparty/eigen/test/meta.cpp +++ b/thirdparty/eigen/test/meta.cpp @@ -9,14 +9,32 @@ #include "main.h" -void test_meta() +template +bool check_is_convertible(const From&, const To&) +{ + return internal::is_convertible::value; +} + +struct FooReturnType { + typedef int ReturnType; +}; + +struct MyInterface { + virtual void func() = 0; + virtual ~MyInterface() {} +}; +struct MyImpl : public MyInterface { + void func() {} +}; + +EIGEN_DECLARE_TEST(meta) { VERIFY((internal::conditional<(3<4),internal::true_type, internal::false_type>::type::value)); VERIFY(( internal::is_same::value)); VERIFY((!internal::is_same::value)); VERIFY((!internal::is_same::value)); VERIFY((!internal::is_same::value)); - + VERIFY(( internal::is_same::type >::value)); VERIFY(( internal::is_same::type >::value)); VERIFY(( internal::is_same::type >::value)); @@ -45,13 +63,60 @@ void test_meta() VERIFY(( internal::is_same< internal::add_const_on_value_type::type, const float* const>::value)); VERIFY(( internal::is_same< internal::add_const_on_value_type::type, const float* const>::value)); - + VERIFY(( internal::is_same::type >::value)); VERIFY(( internal::is_same::type >::value)); VERIFY(( internal::is_same::type >::value)); VERIFY(( internal::is_same::type >::value)); VERIFY(( internal::is_same::type >::value)); - + + + // is_convertible + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible >::value )); + STATIC_CHECK((!internal::is_convertible,double>::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + STATIC_CHECK((!internal::is_convertible::value )); + STATIC_CHECK((!internal::is_convertible::value )); + STATIC_CHECK(!( internal::is_convertible::value )); + + STATIC_CHECK(!( internal::is_convertible::value )); + STATIC_CHECK(( internal::is_convertible::value )); + + //STATIC_CHECK((!internal::is_convertible::value )); //does not even compile because the conversion is prevented by a static assertion + STATIC_CHECK((!internal::is_convertible::value )); + STATIC_CHECK((!internal::is_convertible::value )); + { + float f = 0.0f; + MatrixXf A, B; + VectorXf a, b; + VERIFY(( check_is_convertible(a.dot(b), f) )); + VERIFY(( check_is_convertible(a.transpose()*b, f) )); + VERIFY((!check_is_convertible(A*B, f) )); + VERIFY(( check_is_convertible(A*B, A) )); + } + + { + int i = 0; + VERIFY(( check_is_convertible(fix<3>(), i) )); + VERIFY((!check_is_convertible(i, fix()) )); + } + + + VERIFY(( internal::has_ReturnType::value )); + VERIFY(( internal::has_ReturnType >::value )); + VERIFY(( !internal::has_ReturnType::value )); + VERIFY(( !internal::has_ReturnType::value )); + VERIFY(internal::meta_sqrt<1>::ret == 1); #define VERIFY_META_SQRT(X) VERIFY(internal::meta_sqrt::ret == int(std::sqrt(double(X)))) VERIFY_META_SQRT(2); diff --git a/thirdparty/eigen/test/metis_support.cpp b/thirdparty/eigen/test/metis_support.cpp index 932b0407..b490dacd 100644 --- a/thirdparty/eigen/test/metis_support.cpp +++ b/thirdparty/eigen/test/metis_support.cpp @@ -3,24 +3,10 @@ // // Copyright (C) 2012 Désiré Nuentsa-Wakam // -// Eigen is free software; you can redistribute it and/or -// modify it under the terms of the GNU Lesser General Public -// License as published by the Free Software Foundation; either -// version 3 of the License, or (at your option) any later version. -// -// Alternatively, you can redistribute it and/or -// modify it under the terms of the GNU General Public License as -// published by the Free Software Foundation; either version 2 of -// the License, or (at your option) any later version. -// -// Eigen is distributed in the hope that it will be useful, but WITHOUT ANY -// WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS -// FOR A PARTICULAR PURPOSE. See the GNU Lesser General Public License or the -// GNU General Public License for more details. -// -// You should have received a copy of the GNU Lesser General Public -// License and a copy of the GNU General Public License along with -// Eigen. If not, see . +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + #include "sparse_solver.h" #include #include @@ -33,7 +19,7 @@ template void test_metis_T() check_sparse_square_solving(sparselu_metis); } -void test_metis_support() +EIGEN_DECLARE_TEST(metis_support) { CALL_SUBTEST_1(test_metis_T()); } diff --git a/thirdparty/eigen/test/miscmatrices.cpp b/thirdparty/eigen/test/miscmatrices.cpp index ef20dc74..e71712f3 100644 --- a/thirdparty/eigen/test/miscmatrices.cpp +++ b/thirdparty/eigen/test/miscmatrices.cpp @@ -14,7 +14,6 @@ template void miscMatrices(const MatrixType& m) /* this test covers the following files: DiagonalMatrix.h Ones.h */ - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; typedef Matrix VectorType; Index rows = m.rows(); @@ -35,7 +34,7 @@ template void miscMatrices(const MatrixType& m) VERIFY_IS_APPROX(square, MatrixType::Identity(rows, rows)); } -void test_miscmatrices() +EIGEN_DECLARE_TEST(miscmatrices) { for(int i = 0; i < g_repeat; i++) { CALL_SUBTEST_1( miscMatrices(Matrix()) ); diff --git a/thirdparty/eigen/test/mixingtypes.cpp b/thirdparty/eigen/test/mixingtypes.cpp index 6c2f7487..2af7b888 100644 --- a/thirdparty/eigen/test/mixingtypes.cpp +++ b/thirdparty/eigen/test/mixingtypes.cpp @@ -1,28 +1,76 @@ // This file is part of Eigen, a lightweight C++ template library // for linear algebra. // -// Copyright (C) 2008 Gael Guennebaud +// Copyright (C) 2008-2015 Gael Guennebaud // Copyright (C) 2008 Benoit Jacob // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -// work around "uninitialized" warnings and give that option some testing -#define EIGEN_INITIALIZE_MATRICES_BY_ZERO +#if defined(EIGEN_TEST_PART_7) #ifndef EIGEN_NO_STATIC_ASSERT #define EIGEN_NO_STATIC_ASSERT // turn static asserts into runtime asserts in order to check them #endif -// #ifndef EIGEN_DONT_VECTORIZE -// #define EIGEN_DONT_VECTORIZE // SSE intrinsics aren't designed to allow mixing types -// #endif +// ignore double-promotion diagnostic for clang and gcc, if we check for static assertion anyway: +// TODO do the same for MSVC? +#if defined(__clang__) +# if (__clang_major__ * 100 + __clang_minor__) >= 308 +# pragma clang diagnostic ignored "-Wdouble-promotion" +# endif +#elif defined(__GNUC__) + // TODO is there a minimal GCC version for this? At least g++-4.7 seems to be fine with this. +# pragma GCC diagnostic ignored "-Wdouble-promotion" +#endif + +#endif + + + +#if defined(EIGEN_TEST_PART_1) || defined(EIGEN_TEST_PART_2) || defined(EIGEN_TEST_PART_3) + +#ifndef EIGEN_DONT_VECTORIZE +#define EIGEN_DONT_VECTORIZE +#endif + +#endif + +static bool g_called; +#define EIGEN_SCALAR_BINARY_OP_PLUGIN { g_called |= (!internal::is_same::value); } #include "main.h" using namespace std; +#define VERIFY_MIX_SCALAR(XPR,REF) \ + g_called = false; \ + VERIFY_IS_APPROX(XPR,REF); \ + VERIFY( g_called && #XPR" not properly optimized"); + +template +void raise_assertion(Index size = SizeAtCompileType) +{ + // VERIFY_RAISES_ASSERT(mf+md); // does not even compile + Matrix vf; vf.setRandom(size); + Matrix vd; vd.setRandom(size); + VERIFY_RAISES_ASSERT(vf=vd); + VERIFY_RAISES_ASSERT(vf+=vd); + VERIFY_RAISES_ASSERT(vf-=vd); + VERIFY_RAISES_ASSERT(vd=vf); + VERIFY_RAISES_ASSERT(vd+=vf); + VERIFY_RAISES_ASSERT(vd-=vf); + + // vd.asDiagonal() * mf; // does not even compile + // vcd.asDiagonal() * mf; // does not even compile + +#if 0 // we get other compilation errors here than just static asserts + VERIFY_RAISES_ASSERT(vd.dot(vf)); +#endif +} + + template void mixingtypes(int size = SizeAtCompileType) { typedef std::complex CF; @@ -38,8 +86,10 @@ template void mixingtypes(int size = SizeAtCompileType) Mat_f mf = Mat_f::Random(size,size); Mat_d md = mf.template cast(); + //Mat_d rd = md; Mat_cf mcf = Mat_cf::Random(size,size); Mat_cd mcd = mcf.template cast >(); + Mat_cd rcd = mcd; Vec_f vf = Vec_f::Random(size,1); Vec_d vd = vf.template cast(); Vec_cf vcf = Vec_cf::Random(size,1); @@ -49,25 +99,56 @@ template void mixingtypes(int size = SizeAtCompileType) complex scf = internal::random >(); complex scd = internal::random >(); - mf+mf; - VERIFY_RAISES_ASSERT(mf+md); - VERIFY_RAISES_ASSERT(mf+mcf); - VERIFY_RAISES_ASSERT(vf=vd); - VERIFY_RAISES_ASSERT(vf+=vd); - VERIFY_RAISES_ASSERT(mcd=md); + + float epsf = std::sqrt(std::numeric_limits ::min EIGEN_EMPTY ()); + double epsd = std::sqrt(std::numeric_limits::min EIGEN_EMPTY ()); + + while(std::abs(sf )(); + while(std::abs(sd )(); + while(std::abs(scf)(); + while(std::abs(scd)(); // check scalar products - VERIFY_IS_APPROX(vcf * sf , vcf * complex(sf)); - VERIFY_IS_APPROX(sd * vcd, complex(sd) * vcd); - VERIFY_IS_APPROX(vf * scf , vf.template cast >() * scf); - VERIFY_IS_APPROX(scd * vd, scd * vd.template cast >()); + VERIFY_MIX_SCALAR(vcf * sf , vcf * complex(sf)); + VERIFY_MIX_SCALAR(sd * vcd , complex(sd) * vcd); + VERIFY_MIX_SCALAR(vf * scf , vf.template cast >() * scf); + VERIFY_MIX_SCALAR(scd * vd , scd * vd.template cast >()); + + VERIFY_MIX_SCALAR(vcf * 2 , vcf * complex(2)); + VERIFY_MIX_SCALAR(vcf * 2.1 , vcf * complex(2.1)); + VERIFY_MIX_SCALAR(2 * vcf, vcf * complex(2)); + VERIFY_MIX_SCALAR(2.1 * vcf , vcf * complex(2.1)); + + // check scalar quotients + VERIFY_MIX_SCALAR(vcf / sf , vcf / complex(sf)); + VERIFY_MIX_SCALAR(vf / scf , vf.template cast >() / scf); + VERIFY_MIX_SCALAR(vf.array() / scf, vf.template cast >().array() / scf); + VERIFY_MIX_SCALAR(scd / vd.array() , scd / vd.template cast >().array()); + + // check scalar increment + VERIFY_MIX_SCALAR(vcf.array() + sf , vcf.array() + complex(sf)); + VERIFY_MIX_SCALAR(sd + vcd.array(), complex(sd) + vcd.array()); + VERIFY_MIX_SCALAR(vf.array() + scf, vf.template cast >().array() + scf); + VERIFY_MIX_SCALAR(scd + vd.array() , scd + vd.template cast >().array()); + + // check scalar subtractions + VERIFY_MIX_SCALAR(vcf.array() - sf , vcf.array() - complex(sf)); + VERIFY_MIX_SCALAR(sd - vcd.array(), complex(sd) - vcd.array()); + VERIFY_MIX_SCALAR(vf.array() - scf, vf.template cast >().array() - scf); + VERIFY_MIX_SCALAR(scd - vd.array() , scd - vd.template cast >().array()); + + // check scalar powers + // NOTE: scalar exponents use a unary op. + VERIFY_IS_APPROX( pow(vcf.array(), sf), Eigen::pow(vcf.array(), complex(sf)) ); + VERIFY_IS_APPROX( vcf.array().pow(sf) , Eigen::pow(vcf.array(), complex(sf)) ); + VERIFY_MIX_SCALAR( pow(sd, vcd.array()), Eigen::pow(complex(sd), vcd.array()) ); + VERIFY_IS_APPROX( Eigen::pow(vf.array(), scf), Eigen::pow(vf.template cast >().array(), scf) ); + VERIFY_IS_APPROX( vf.array().pow(scf) , Eigen::pow(vf.template cast >().array(), scf) ); + VERIFY_MIX_SCALAR( Eigen::pow(scd, vd.array()), Eigen::pow(scd, vd.template cast >().array()) ); // check dot product vf.dot(vf); -#if 0 // we get other compilation errors here than just static asserts - VERIFY_RAISES_ASSERT(vd.dot(vf)); -#endif VERIFY_IS_APPROX(vcf.dot(vf), vcf.dot(vf.template cast >())); // check diagonal product @@ -75,8 +156,6 @@ template void mixingtypes(int size = SizeAtCompileType) VERIFY_IS_APPROX(vcd.asDiagonal() * md, vcd.asDiagonal() * md.template cast >()); VERIFY_IS_APPROX(mcf * vf.asDiagonal(), mcf * vf.template cast >().asDiagonal()); VERIFY_IS_APPROX(md * vcd.asDiagonal(), md.template cast >() * vcd.asDiagonal()); -// vd.asDiagonal() * mf; // does not even compile -// vcd.asDiagonal() * mf; // does not even compile // check inner product VERIFY_IS_APPROX((vf.transpose() * vcf).value(), (vf.template cast >().transpose() * vcf).value()); @@ -92,7 +171,6 @@ template void mixingtypes(int size = SizeAtCompileType) VERIFY_IS_APPROX(mcd.array() *= md.array(), mcd2.array() *= md.array().template cast >()); // check matrix-matrix products - VERIFY_IS_APPROX(sd*md*mcd, (sd*md).template cast().eval()*mcd); VERIFY_IS_APPROX(sd*mcd*md, sd*mcd*md.template cast()); VERIFY_IS_APPROX(scd*md*mcd, scd*md.template cast().eval()*mcd); @@ -103,6 +181,20 @@ template void mixingtypes(int size = SizeAtCompileType) VERIFY_IS_APPROX(scf*mf*mcf, scf*mf.template cast()*mcf); VERIFY_IS_APPROX(scf*mcf*mf, scf*mcf*mf.template cast()); + VERIFY_IS_APPROX(sd*md.adjoint()*mcd, (sd*md).template cast().eval().adjoint()*mcd); + VERIFY_IS_APPROX(sd*mcd.adjoint()*md, sd*mcd.adjoint()*md.template cast()); + VERIFY_IS_APPROX(sd*md.adjoint()*mcd.adjoint(), (sd*md).template cast().eval().adjoint()*mcd.adjoint()); + VERIFY_IS_APPROX(sd*mcd.adjoint()*md.adjoint(), sd*mcd.adjoint()*md.template cast().adjoint()); + VERIFY_IS_APPROX(sd*md*mcd.adjoint(), (sd*md).template cast().eval()*mcd.adjoint()); + VERIFY_IS_APPROX(sd*mcd*md.adjoint(), sd*mcd*md.template cast().adjoint()); + + VERIFY_IS_APPROX(sf*mf.adjoint()*mcf, (sf*mf).template cast().eval().adjoint()*mcf); + VERIFY_IS_APPROX(sf*mcf.adjoint()*mf, sf*mcf.adjoint()*mf.template cast()); + VERIFY_IS_APPROX(sf*mf.adjoint()*mcf.adjoint(), (sf*mf).template cast().eval().adjoint()*mcf.adjoint()); + VERIFY_IS_APPROX(sf*mcf.adjoint()*mf.adjoint(), sf*mcf.adjoint()*mf.template cast().adjoint()); + VERIFY_IS_APPROX(sf*mf*mcf.adjoint(), (sf*mf).template cast().eval()*mcf.adjoint()); + VERIFY_IS_APPROX(sf*mcf*mf.adjoint(), sf*mcf*mf.template cast().adjoint()); + VERIFY_IS_APPROX(sf*mf*vcf, (sf*mf).template cast().eval()*vcf); VERIFY_IS_APPROX(scf*mf*vcf,(scf*mf.template cast()).eval()*vcf); VERIFY_IS_APPROX(sf*mcf*vf, sf*mcf*vf.template cast()); @@ -122,11 +214,117 @@ template void mixingtypes(int size = SizeAtCompileType) VERIFY_IS_APPROX(scd*vcd.adjoint()*md, scd*vcd.adjoint()*md.template cast().eval()); VERIFY_IS_APPROX(sd*vd.adjoint()*mcd, sd*vd.adjoint().template cast().eval()*mcd); VERIFY_IS_APPROX(scd*vd.adjoint()*mcd, scd*vd.adjoint().template cast().eval()*mcd); + + VERIFY_IS_APPROX( sd*vcd.adjoint()*md.template triangularView(), sd*vcd.adjoint()*md.template cast().eval().template triangularView()); + VERIFY_IS_APPROX(scd*vcd.adjoint()*md.template triangularView(), scd*vcd.adjoint()*md.template cast().eval().template triangularView()); + VERIFY_IS_APPROX( sd*vcd.adjoint()*md.transpose().template triangularView(), sd*vcd.adjoint()*md.transpose().template cast().eval().template triangularView()); + VERIFY_IS_APPROX(scd*vcd.adjoint()*md.transpose().template triangularView(), scd*vcd.adjoint()*md.transpose().template cast().eval().template triangularView()); + VERIFY_IS_APPROX( sd*vd.adjoint()*mcd.template triangularView(), sd*vd.adjoint().template cast().eval()*mcd.template triangularView()); + VERIFY_IS_APPROX(scd*vd.adjoint()*mcd.template triangularView(), scd*vd.adjoint().template cast().eval()*mcd.template triangularView()); + VERIFY_IS_APPROX( sd*vd.adjoint()*mcd.transpose().template triangularView(), sd*vd.adjoint().template cast().eval()*mcd.transpose().template triangularView()); + VERIFY_IS_APPROX(scd*vd.adjoint()*mcd.transpose().template triangularView(), scd*vd.adjoint().template cast().eval()*mcd.transpose().template triangularView()); + + // Not supported yet: trmm +// VERIFY_IS_APPROX(sd*mcd*md.template triangularView(), sd*mcd*md.template cast().eval().template triangularView()); +// VERIFY_IS_APPROX(scd*mcd*md.template triangularView(), scd*mcd*md.template cast().eval().template triangularView()); +// VERIFY_IS_APPROX(sd*md*mcd.template triangularView(), sd*md.template cast().eval()*mcd.template triangularView()); +// VERIFY_IS_APPROX(scd*md*mcd.template triangularView(), scd*md.template cast().eval()*mcd.template triangularView()); + + // Not supported yet: symv +// VERIFY_IS_APPROX(sd*vcd.adjoint()*md.template selfadjointView(), sd*vcd.adjoint()*md.template cast().eval().template selfadjointView()); +// VERIFY_IS_APPROX(scd*vcd.adjoint()*md.template selfadjointView(), scd*vcd.adjoint()*md.template cast().eval().template selfadjointView()); +// VERIFY_IS_APPROX(sd*vd.adjoint()*mcd.template selfadjointView(), sd*vd.adjoint().template cast().eval()*mcd.template selfadjointView()); +// VERIFY_IS_APPROX(scd*vd.adjoint()*mcd.template selfadjointView(), scd*vd.adjoint().template cast().eval()*mcd.template selfadjointView()); + + // Not supported yet: symm +// VERIFY_IS_APPROX(sd*vcd.adjoint()*md.template selfadjointView(), sd*vcd.adjoint()*md.template cast().eval().template selfadjointView()); +// VERIFY_IS_APPROX(scd*vcd.adjoint()*md.template selfadjointView(), scd*vcd.adjoint()*md.template cast().eval().template selfadjointView()); +// VERIFY_IS_APPROX(sd*vd.adjoint()*mcd.template selfadjointView(), sd*vd.adjoint().template cast().eval()*mcd.template selfadjointView()); +// VERIFY_IS_APPROX(scd*vd.adjoint()*mcd.template selfadjointView(), scd*vd.adjoint().template cast().eval()*mcd.template selfadjointView()); + + rcd.setZero(); + VERIFY_IS_APPROX(Mat_cd(rcd.template triangularView() = sd * mcd * md), + Mat_cd((sd * mcd * md.template cast().eval()).template triangularView())); + VERIFY_IS_APPROX(Mat_cd(rcd.template triangularView() = sd * md * mcd), + Mat_cd((sd * md.template cast().eval() * mcd).template triangularView())); + VERIFY_IS_APPROX(Mat_cd(rcd.template triangularView() = scd * mcd * md), + Mat_cd((scd * mcd * md.template cast().eval()).template triangularView())); + VERIFY_IS_APPROX(Mat_cd(rcd.template triangularView() = scd * md * mcd), + Mat_cd((scd * md.template cast().eval() * mcd).template triangularView())); + + + VERIFY_IS_APPROX( md.array() * mcd.array(), md.template cast().eval().array() * mcd.array() ); + VERIFY_IS_APPROX( mcd.array() * md.array(), mcd.array() * md.template cast().eval().array() ); + + VERIFY_IS_APPROX( md.array() + mcd.array(), md.template cast().eval().array() + mcd.array() ); + VERIFY_IS_APPROX( mcd.array() + md.array(), mcd.array() + md.template cast().eval().array() ); + + VERIFY_IS_APPROX( md.array() - mcd.array(), md.template cast().eval().array() - mcd.array() ); + VERIFY_IS_APPROX( mcd.array() - md.array(), mcd.array() - md.template cast().eval().array() ); + + if(mcd.array().abs().minCoeff()>epsd) + { + VERIFY_IS_APPROX( md.array() / mcd.array(), md.template cast().eval().array() / mcd.array() ); + } + if(md.array().abs().minCoeff()>epsd) + { + VERIFY_IS_APPROX( mcd.array() / md.array(), mcd.array() / md.template cast().eval().array() ); + } + + if(md.array().abs().minCoeff()>epsd || mcd.array().abs().minCoeff()>epsd) + { + VERIFY_IS_APPROX( md.array().pow(mcd.array()), md.template cast().eval().array().pow(mcd.array()) ); + VERIFY_IS_APPROX( mcd.array().pow(md.array()), mcd.array().pow(md.template cast().eval().array()) ); + + VERIFY_IS_APPROX( pow(md.array(),mcd.array()), md.template cast().eval().array().pow(mcd.array()) ); + VERIFY_IS_APPROX( pow(mcd.array(),md.array()), mcd.array().pow(md.template cast().eval().array()) ); + } + + rcd = mcd; + VERIFY_IS_APPROX( rcd = md, md.template cast().eval() ); + rcd = mcd; + VERIFY_IS_APPROX( rcd += md, mcd + md.template cast().eval() ); + rcd = mcd; + VERIFY_IS_APPROX( rcd -= md, mcd - md.template cast().eval() ); + rcd = mcd; + VERIFY_IS_APPROX( rcd.array() *= md.array(), mcd.array() * md.template cast().eval().array() ); + rcd = mcd; + if(md.array().abs().minCoeff()>epsd) + { + VERIFY_IS_APPROX( rcd.array() /= md.array(), mcd.array() / md.template cast().eval().array() ); + } + + rcd = mcd; + VERIFY_IS_APPROX( rcd.noalias() += md + mcd*md, mcd + (md.template cast().eval()) + mcd*(md.template cast().eval())); + + VERIFY_IS_APPROX( rcd.noalias() = md*md, ((md*md).eval().template cast()) ); + rcd = mcd; + VERIFY_IS_APPROX( rcd.noalias() += md*md, mcd + ((md*md).eval().template cast()) ); + rcd = mcd; + VERIFY_IS_APPROX( rcd.noalias() -= md*md, mcd - ((md*md).eval().template cast()) ); + + VERIFY_IS_APPROX( rcd.noalias() = mcd + md*md, mcd + ((md*md).eval().template cast()) ); + rcd = mcd; + VERIFY_IS_APPROX( rcd.noalias() += mcd + md*md, mcd + mcd + ((md*md).eval().template cast()) ); + rcd = mcd; + VERIFY_IS_APPROX( rcd.noalias() -= mcd + md*md, - ((md*md).eval().template cast()) ); } -void test_mixingtypes() +EIGEN_DECLARE_TEST(mixingtypes) { - CALL_SUBTEST_1(mixingtypes<3>()); - CALL_SUBTEST_2(mixingtypes<4>()); - CALL_SUBTEST_3(mixingtypes(internal::random(1,EIGEN_TEST_MAX_SIZE))); + g_called = false; // Silence -Wunneeded-internal-declaration. + for(int i = 0; i < g_repeat; i++) { + CALL_SUBTEST_1(mixingtypes<3>()); + CALL_SUBTEST_2(mixingtypes<4>()); + CALL_SUBTEST_3(mixingtypes(internal::random(1,EIGEN_TEST_MAX_SIZE))); + + CALL_SUBTEST_4(mixingtypes<3>()); + CALL_SUBTEST_5(mixingtypes<4>()); + CALL_SUBTEST_6(mixingtypes(internal::random(1,EIGEN_TEST_MAX_SIZE))); + CALL_SUBTEST_7(raise_assertion(internal::random(1,EIGEN_TEST_MAX_SIZE))); + } + CALL_SUBTEST_7(raise_assertion<0>()); + CALL_SUBTEST_7(raise_assertion<3>()); + CALL_SUBTEST_7(raise_assertion<4>()); + CALL_SUBTEST_7(raise_assertion(0)); } diff --git a/thirdparty/eigen/test/mpl2only.cpp b/thirdparty/eigen/test/mpl2only.cpp index 5ef0d2b2..296350d0 100644 --- a/thirdparty/eigen/test/mpl2only.cpp +++ b/thirdparty/eigen/test/mpl2only.cpp @@ -7,12 +7,16 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +#ifndef EIGEN_MPL2_ONLY #define EIGEN_MPL2_ONLY +#endif #include #include #include #include +#include #include +#include int main() { diff --git a/thirdparty/eigen/test/nestbyvalue.cpp b/thirdparty/eigen/test/nestbyvalue.cpp new file mode 100644 index 00000000..3a86bea5 --- /dev/null +++ b/thirdparty/eigen/test/nestbyvalue.cpp @@ -0,0 +1,37 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2019 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#define TEST_ENABLE_TEMPORARY_TRACKING + +#include "main.h" + +typedef NestByValue CpyMatrixXd; +typedef CwiseBinaryOp,const CpyMatrixXd,const CpyMatrixXd> XprType; + +XprType get_xpr_with_temps(const MatrixXd& a) +{ + MatrixXd t1 = a.rowwise().reverse(); + MatrixXd t2 = a+a; + return t1.nestByValue() + t2.nestByValue(); +} + +EIGEN_DECLARE_TEST(nestbyvalue) +{ + for(int i = 0; i < g_repeat; i++) { + Index rows = internal::random(1,EIGEN_TEST_MAX_SIZE); + Index cols = internal::random(1,EIGEN_TEST_MAX_SIZE); + MatrixXd a = MatrixXd::Random(rows,cols); + nb_temporaries = 0; + XprType x = get_xpr_with_temps(a); + VERIFY_IS_EQUAL(nb_temporaries,6); + MatrixXd b = x; + VERIFY_IS_EQUAL(nb_temporaries,6+1); + VERIFY_IS_APPROX(b, a.rowwise().reverse().eval() + (a+a).eval()); + } +} diff --git a/thirdparty/eigen/test/nesting_ops.cpp b/thirdparty/eigen/test/nesting_ops.cpp index 1e852328..4b5fc21f 100644 --- a/thirdparty/eigen/test/nesting_ops.cpp +++ b/thirdparty/eigen/test/nesting_ops.cpp @@ -2,16 +2,37 @@ // for linear algebra. // // Copyright (C) 2010 Hauke Heibel +// Copyright (C) 2015 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. +#define TEST_ENABLE_TEMPORARY_TRACKING + #include "main.h" -template void run_nesting_ops(const MatrixType& _m) +template +void use_n_times(const XprType &xpr) { - typename MatrixType::Nested m(_m); + typename internal::nested_eval::type mat(xpr); + typename XprType::PlainObject res(mat.rows(), mat.cols()); + nb_temporaries--; // remove res + res.setZero(); + for(int i=0; i +bool verify_eval_type(const XprType &, const ReferenceType&) +{ + typedef typename internal::nested_eval::type EvalType; + return internal::is_same::type, typename internal::remove_all::type>::value; +} + +template void run_nesting_ops_1(const MatrixType& _m) +{ + typename internal::nested_eval::type m(_m); // Make really sure that we are in debug mode! VERIFY_RAISES_ASSERT(eigen_assert(false)); @@ -24,10 +45,63 @@ template void run_nesting_ops(const MatrixType& _m) VERIFY_IS_APPROX( (m.transpose() * m).array().abs().sum(), (m.transpose() * m).array().abs().sum() ); } -void test_nesting_ops() +template void run_nesting_ops_2(const MatrixType& _m) { - CALL_SUBTEST_1(run_nesting_ops(MatrixXf::Random(25,25))); - CALL_SUBTEST_2(run_nesting_ops(MatrixXd::Random(25,25))); - CALL_SUBTEST_3(run_nesting_ops(Matrix4f::Random())); - CALL_SUBTEST_4(run_nesting_ops(Matrix4d::Random())); + typedef typename MatrixType::Scalar Scalar; + Index rows = _m.rows(); + Index cols = _m.cols(); + MatrixType m1 = MatrixType::Random(rows,cols); + Matrix m2; + + if((MatrixType::SizeAtCompileTime==Dynamic)) + { + VERIFY_EVALUATION_COUNT( use_n_times<1>(m1 + m1*m1), 1 ); + VERIFY_EVALUATION_COUNT( use_n_times<10>(m1 + m1*m1), 1 ); + + VERIFY_EVALUATION_COUNT( use_n_times<1>(m1.template triangularView().solve(m1.col(0))), 1 ); + VERIFY_EVALUATION_COUNT( use_n_times<10>(m1.template triangularView().solve(m1.col(0))), 1 ); + + VERIFY_EVALUATION_COUNT( use_n_times<1>(Scalar(2)*m1.template triangularView().solve(m1.col(0))), 2 ); // FIXME could be one by applying the scaling in-place on the solve result + VERIFY_EVALUATION_COUNT( use_n_times<1>(m1.col(0)+m1.template triangularView().solve(m1.col(0))), 2 ); // FIXME could be one by adding m1.col() inplace + VERIFY_EVALUATION_COUNT( use_n_times<10>(m1.col(0)+m1.template triangularView().solve(m1.col(0))), 2 ); + } + + { + VERIFY( verify_eval_type<10>(m1, m1) ); + if(!NumTraits::IsComplex) + { + VERIFY( verify_eval_type<3>(2*m1, 2*m1) ); + VERIFY( verify_eval_type<4>(2*m1, m1) ); + } + else + { + VERIFY( verify_eval_type<2>(2*m1, 2*m1) ); + VERIFY( verify_eval_type<3>(2*m1, m1) ); + } + VERIFY( verify_eval_type<2>(m1+m1, m1+m1) ); + VERIFY( verify_eval_type<3>(m1+m1, m1) ); + VERIFY( verify_eval_type<1>(m1*m1.transpose(), m2) ); + VERIFY( verify_eval_type<1>(m1*(m1+m1).transpose(), m2) ); + VERIFY( verify_eval_type<2>(m1*m1.transpose(), m2) ); + VERIFY( verify_eval_type<1>(m1+m1*m1, m1) ); + + VERIFY( verify_eval_type<1>(m1.template triangularView().solve(m1), m1) ); + VERIFY( verify_eval_type<1>(m1+m1.template triangularView().solve(m1), m1) ); + } +} + + +EIGEN_DECLARE_TEST(nesting_ops) +{ + CALL_SUBTEST_1(run_nesting_ops_1(MatrixXf::Random(25,25))); + CALL_SUBTEST_2(run_nesting_ops_1(MatrixXcd::Random(25,25))); + CALL_SUBTEST_3(run_nesting_ops_1(Matrix4f::Random())); + CALL_SUBTEST_4(run_nesting_ops_1(Matrix2d::Random())); + + Index s = internal::random(1,EIGEN_TEST_MAX_SIZE); + CALL_SUBTEST_1( run_nesting_ops_2(MatrixXf(s,s)) ); + CALL_SUBTEST_2( run_nesting_ops_2(MatrixXcd(s,s)) ); + CALL_SUBTEST_3( run_nesting_ops_2(Matrix4f()) ); + CALL_SUBTEST_4( run_nesting_ops_2(Matrix2d()) ); + TEST_SET_BUT_UNUSED_VARIABLE(s) } diff --git a/thirdparty/eigen/test/nomalloc.cpp b/thirdparty/eigen/test/nomalloc.cpp index 8e040235..cb4c073e 100644 --- a/thirdparty/eigen/test/nomalloc.cpp +++ b/thirdparty/eigen/test/nomalloc.cpp @@ -8,20 +8,10 @@ // Public License v. 2.0. If a copy of the MPL was not distributed // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. -// this hack is needed to make this file compiles with -pedantic (gcc) -#ifdef __GNUC__ -#define throw(X) -#endif - -#ifdef __INTEL_COMPILER - // disable "warning #76: argument to macro is empty" produced by the above hack - #pragma warning disable 76 -#endif - // discard stack allocation as that too bypasses malloc #define EIGEN_STACK_ALLOCATION_LIMIT 0 -// any heap allocation will raise an assert -#define EIGEN_NO_MALLOC +// heap allocation will raise an assert if enabled at runtime +#define EIGEN_RUNTIME_NO_MALLOC #include "main.h" #include @@ -34,7 +24,6 @@ template void nomalloc(const MatrixType& m) { /* this test check no dynamic memory allocation are issued with fixed-size matrices */ - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; Index rows = m.rows(); @@ -88,14 +77,15 @@ template void nomalloc(const MatrixType& m) VERIFY_IS_APPROX(m2,m2); m2.template selfadjointView().rankUpdate(m1.col(0),-1); - m2.template selfadjointView().rankUpdate(m1.row(0),-1); + m2.template selfadjointView().rankUpdate(m1.row(0),-1); + m2.template selfadjointView().rankUpdate(m1.col(0), m1.col(0)); // rank-2 // The following fancy matrix-matrix products are not safe yet regarding static allocation -// m1 += m1.template triangularView() * m2.col(; -// m1.template selfadjointView().rankUpdate(m2); -// m1 += m1.template triangularView() * m2; -// m1 += m1.template selfadjointView() * m2; -// VERIFY_IS_APPROX(m1,m1); + m2.template selfadjointView().rankUpdate(m1); + m2 += m2.template triangularView() * m1; + m2.template triangularView() = m2 * m2; + m1 += m1.template selfadjointView() * m2; + VERIFY_IS_APPROX(m2,m2); } template @@ -171,7 +161,7 @@ void test_zerosized() { Eigen::VectorXd v; // explicit zero-sized: Eigen::ArrayXXd A0(0,0); - Eigen::ArrayXd v0(std::ptrdiff_t(0)); // FIXME ArrayXd(0) is ambiguous + Eigen::ArrayXd v0(0); // assigning empty objects to each other: A=A0; @@ -182,10 +172,12 @@ template void test_reference(const MatrixType& m) { typedef typename MatrixType::Scalar Scalar; enum { Flag = MatrixType::IsRowMajor ? Eigen::RowMajor : Eigen::ColMajor}; enum { TransposeFlag = !MatrixType::IsRowMajor ? Eigen::RowMajor : Eigen::ColMajor}; - typename MatrixType::Index rows = m.rows(), cols=m.cols(); + Index rows = m.rows(), cols=m.cols(); + typedef Eigen::Matrix MatrixX; + typedef Eigen::Matrix MatrixXT; // Dynamic reference: - typedef Eigen::Ref > Ref; - typedef Eigen::Ref > RefT; + typedef Eigen::Ref Ref; + typedef Eigen::Ref RefT; Ref r1(m); Ref r2(m.block(rows/3, cols/4, rows/2, cols/2)); @@ -195,10 +187,30 @@ template void test_reference(const MatrixType& m) { VERIFY_RAISES_ASSERT(RefT r5(m)); VERIFY_RAISES_ASSERT(Ref r6(m.transpose())); VERIFY_RAISES_ASSERT(Ref r7(Scalar(2) * m)); + + // Copy constructors shall also never malloc + Ref r8 = r1; + RefT r9 = r3; + + // Initializing from a compatible Ref shall also never malloc + Eigen::Ref > r10=r8, r11=m; + + // Initializing from an incompatible Ref will malloc: + typedef Eigen::Ref RefAligned; + VERIFY_RAISES_ASSERT(RefAligned r12=r10); + VERIFY_RAISES_ASSERT(Ref r13=r10); // r10 has more dynamic strides + } -void test_nomalloc() +EIGEN_DECLARE_TEST(nomalloc) { + // create some dynamic objects + Eigen::MatrixXd M1 = MatrixXd::Random(3,3); + Ref R1 = 2.0*M1; // Ref requires temporary + + // from here on prohibit malloc: + Eigen::internal::set_is_malloc_allowed(false); + // check that our operator new is indeed called: VERIFY_RAISES_ASSERT(MatrixXd dummy(MatrixXd::Random(3,3))); CALL_SUBTEST_1(nomalloc(Matrix()) ); @@ -207,6 +219,10 @@ void test_nomalloc() // Check decomposition modules with dynamic matrices that have a known compile-time max size (ctms) CALL_SUBTEST_4(ctms_decompositions()); + CALL_SUBTEST_5(test_zerosized()); + CALL_SUBTEST_6(test_reference(Matrix())); + CALL_SUBTEST_7(test_reference(R1)); + CALL_SUBTEST_8(Ref R2 = M1.topRows<2>(); test_reference(R2)); } diff --git a/thirdparty/eigen/test/nullary.cpp b/thirdparty/eigen/test/nullary.cpp index fbc721a1..9b25ea4f 100644 --- a/thirdparty/eigen/test/nullary.cpp +++ b/thirdparty/eigen/test/nullary.cpp @@ -2,6 +2,7 @@ // for linear algebra. // // Copyright (C) 2010-2011 Jitse Niesen +// Copyright (C) 2016 Gael Guennebaud // // This Source Code Form is subject to the terms of the Mozilla // Public License v. 2.0. If a copy of the MPL was not distributed @@ -12,7 +13,6 @@ template bool equalsIdentity(const MatrixType& A) { - typedef typename MatrixType::Index Index; typedef typename MatrixType::Scalar Scalar; Scalar zero = static_cast(0); @@ -30,50 +30,104 @@ bool equalsIdentity(const MatrixType& A) bool diagOK = (A.diagonal().array() == 1).all(); return offDiagOK && diagOK; + +} + +template +void check_extremity_accuracy(const VectorType &v, const typename VectorType::Scalar &low, const typename VectorType::Scalar &high) +{ + typedef typename VectorType::Scalar Scalar; + typedef typename VectorType::RealScalar RealScalar; + + RealScalar prec = internal::is_same::value ? NumTraits::dummy_precision()*10 : NumTraits::dummy_precision()/10; + Index size = v.size(); + + if(size<20) + return; + + for (int i=0; isize-6) + { + Scalar ref = (low*RealScalar(size-i-1))/RealScalar(size-1) + (high*RealScalar(i))/RealScalar(size-1); + if(std::abs(ref)>1) + { + if(!internal::isApprox(v(i), ref, prec)) + std::cout << v(i) << " != " << ref << " ; relative error: " << std::abs((v(i)-ref)/ref) << " ; required precision: " << prec << " ; range: " << low << "," << high << " ; i: " << i << "\n"; + VERIFY(internal::isApprox(v(i), (low*RealScalar(size-i-1))/RealScalar(size-1) + (high*RealScalar(i))/RealScalar(size-1), prec)); + } + } + } } template void testVectorType(const VectorType& base) { - typedef typename internal::traits::Index Index; - typedef typename internal::traits::Scalar Scalar; + typedef typename VectorType::Scalar Scalar; + typedef typename VectorType::RealScalar RealScalar; const Index size = base.size(); Scalar high = internal::random(-500,500); Scalar low = (size == 1 ? high : internal::random(-500,500)); - if (low>high) std::swap(low,high); + if (numext::real(low)>numext::real(high)) std::swap(low,high); + + // check low==high + if(internal::random(0.f,1.f)<0.05f) + low = high; + // check abs(low) >> abs(high) + else if(size>2 && std::numeric_limits::max_exponent10>0 && internal::random(0.f,1.f)<0.1f) + low = -internal::random(1,2) * RealScalar(std::pow(RealScalar(10),std::numeric_limits::max_exponent10/2)); - const Scalar step = ((size == 1) ? 1 : (high-low)/(size-1)); + const Scalar step = ((size == 1) ? 1 : (high-low)/RealScalar(size-1)); // check whether the result yields what we expect it to do VectorType m(base); m.setLinSpaced(size,low,high); - VectorType n(size); - for (int i=0; i::IsInteger) + { + VectorType n(size); + for (int i=0; i::IsInteger) || (range_length>=size && (Index(range_length)%(size-1))==0) || (Index(range_length+1)::IsInteger) || (range_length>=size)) + for (int i=0; i::epsilon() ); + // random access version + m = VectorType::LinSpaced(size,low,high); + VERIFY_IS_APPROX(m,n); + VERIFY( internal::isApprox(m(m.size()-1),high) ); + VERIFY( size==1 || internal::isApprox(m(0),low) ); + VERIFY_IS_EQUAL(m(m.size()-1) , high); + if(!NumTraits::IsInteger) + CALL_SUBTEST( check_extremity_accuracy(m, low, high) ); + } - // These guys sometimes fail! This is not good. Any ideas how to fix them!? - //VERIFY( m(m.size()-1) == high ); - //VERIFY( m(0) == low ); + VERIFY( numext::real(m(m.size()-1)) <= numext::real(high) ); + VERIFY( (m.array().real() <= numext::real(high)).all() ); + VERIFY( (m.array().real() >= numext::real(low)).all() ); - // sequential access version - m = VectorType::LinSpaced(Sequential,size,low,high); - VERIFY_IS_APPROX(m,n); - // These guys sometimes fail! This is not good. Any ideas how to fix them!? - //VERIFY( m(m.size()-1) == high ); - //VERIFY( m(0) == low ); + VERIFY( numext::real(m(m.size()-1)) >= numext::real(low) ); + if(size>=1) + { + VERIFY( internal::isApprox(m(0),low) ); + VERIFY_IS_EQUAL(m(0) , low); + } // check whether everything works with row and col major vectors Matrix row_vector(size); @@ -82,7 +136,7 @@ void testVectorType(const VectorType& base) col_vector.setLinSpaced(size,low,high); // when using the extended precision (e.g., FPU) the relative error might exceed 1 bit // when computing the squared sum in isApprox, thus the 2x factor. - VERIFY( row_vector.isApprox(col_vector.transpose(), Scalar(2)*NumTraits::epsilon())); + VERIFY( row_vector.isApprox(col_vector.transpose(), RealScalar(2)*NumTraits::epsilon())); Matrix size_changer(size+50); size_changer.setLinSpaced(size,low,high); @@ -95,37 +149,193 @@ void testVectorType(const VectorType& base) VERIFY_IS_APPROX( ScalarMatrix::LinSpaced(1,low,high), ScalarMatrix::Constant(high) ); // regression test for bug 526 (linear vectorized transversal) - if (size > 1) { + if (size > 1 && (!NumTraits::IsInteger)) { m.tail(size-1).setLinSpaced(low, high); VERIFY_IS_APPROX(m(size-1), high); } + + // regression test for bug 1383 (LinSpaced with empty size/range) + { + Index n0 = VectorType::SizeAtCompileTime==Dynamic ? 0 : VectorType::SizeAtCompileTime; + low = internal::random(); + m = VectorType::LinSpaced(n0,low,low-RealScalar(1)); + VERIFY(m.size()==n0); + + if(VectorType::SizeAtCompileTime==Dynamic) + { + VERIFY_IS_EQUAL(VectorType::LinSpaced(n0,0,Scalar(n0-1)).sum(),Scalar(0)); + VERIFY_IS_EQUAL(VectorType::LinSpaced(n0,low,low-RealScalar(1)).sum(),Scalar(0)); + } + + m.setLinSpaced(n0,0,Scalar(n0-1)); + VERIFY(m.size()==n0); + m.setLinSpaced(n0,low,low-RealScalar(1)); + VERIFY(m.size()==n0); + + // empty range only: + VERIFY_IS_APPROX(VectorType::LinSpaced(size,low,low),VectorType::Constant(size,low)); + m.setLinSpaced(size,low,low); + VERIFY_IS_APPROX(m,VectorType::Constant(size,low)); + + if(NumTraits::IsInteger) + { + VERIFY_IS_APPROX( VectorType::LinSpaced(size,low,low+Scalar(size-1)), VectorType::LinSpaced(size,low+Scalar(size-1),low).reverse() ); + + if(VectorType::SizeAtCompileTime==Dynamic) + { + // Check negative multiplicator path: + for(Index k=1; k<5; ++k) + VERIFY_IS_APPROX( VectorType::LinSpaced(size,low,low+Scalar((size-1)*k)), VectorType::LinSpaced(size,low+Scalar((size-1)*k),low).reverse() ); + // Check negative divisor path: + for(Index k=1; k<5; ++k) + VERIFY_IS_APPROX( VectorType::LinSpaced(size*k,low,low+Scalar(size-1)), VectorType::LinSpaced(size*k,low+Scalar(size-1),low).reverse() ); + } + } + } + + // test setUnit() + if(m.size()>0) + { + for(Index k=0; k<10; ++k) + { + Index i = internal::random(0,m.size()-1); + m.setUnit(i); + VERIFY_IS_APPROX( m, VectorType::Unit(m.size(), i) ); + } + if(VectorType::SizeAtCompileTime==Dynamic) + { + Index i = internal::random(0,2*m.size()-1); + m.setUnit(2*m.size(),i); + VERIFY_IS_APPROX( m, VectorType::Unit(m.size(),i) ); + } + } + } template void testMatrixType(const MatrixType& m) { - typedef typename MatrixType::Index Index; + using std::abs; const Index rows = m.rows(); const Index cols = m.cols(); + typedef typename MatrixType::Scalar Scalar; + typedef typename MatrixType::RealScalar RealScalar; + + Scalar s1; + do { + s1 = internal::random(); + } while(abs(s1)::IsInteger)); MatrixType A; A.setIdentity(rows, cols); VERIFY(equalsIdentity(A)); VERIFY(equalsIdentity(MatrixType::Identity(rows, cols))); + + + A = MatrixType::Constant(rows,cols,s1); + Index i = internal::random(0,rows-1); + Index j = internal::random(0,cols-1); + VERIFY_IS_APPROX( MatrixType::Constant(rows,cols,s1)(i,j), s1 ); + VERIFY_IS_APPROX( MatrixType::Constant(rows,cols,s1).coeff(i,j), s1 ); + VERIFY_IS_APPROX( A(i,j), s1 ); } -void test_nullary() +template +void bug79() +{ + // Assignment of a RowVectorXd to a MatrixXd (regression test for bug #79). + VERIFY( (MatrixXd(RowVectorXd::LinSpaced(3, 0, 1)) - RowVector3d(0, 0.5, 1)).norm() < std::numeric_limits::epsilon() ); +} + +template +void bug1630() +{ + Array4d x4 = Array4d::LinSpaced(0.0, 1.0); + Array3d x3(Array4d::LinSpaced(0.0, 1.0).head(3)); + VERIFY_IS_APPROX(x4.head(3), x3); +} + +template +void nullary_overflow() +{ + // Check possible overflow issue + int n = 60000; + ArrayXi a1(n), a2(n); + a1.setLinSpaced(n, 0, n-1); + for(int i=0; i +void nullary_internal_logic() +{ + // check some internal logic + VERIFY(( internal::has_nullary_operator >::value )); + VERIFY(( !internal::has_unary_operator >::value )); + VERIFY(( !internal::has_binary_operator >::value )); + VERIFY(( internal::functor_has_linear_access >::ret )); + + VERIFY(( !internal::has_nullary_operator >::value )); + VERIFY(( !internal::has_unary_operator >::value )); + VERIFY(( internal::has_binary_operator >::value )); + VERIFY(( !internal::functor_has_linear_access >::ret )); + + VERIFY(( !internal::has_nullary_operator >::value )); + VERIFY(( internal::has_unary_operator >::value )); + VERIFY(( !internal::has_binary_operator >::value )); + VERIFY(( internal::functor_has_linear_access >::ret )); + + // Regression unit test for a weird MSVC bug. + // Search "nullary_wrapper_workaround_msvc" in CoreEvaluators.h for the details. + // See also traits::match. + { + MatrixXf A = MatrixXf::Random(3,3); + Ref R = 2.0*A; + VERIFY_IS_APPROX(R, A+A); + + Ref R1 = MatrixXf::Random(3,3)+A; + + VectorXi V = VectorXi::Random(3); + Ref R2 = VectorXi::LinSpaced(3,1,3)+V; + VERIFY_IS_APPROX(R2, V+Vector3i(1,2,3)); + + VERIFY(( internal::has_nullary_operator >::value )); + VERIFY(( !internal::has_unary_operator >::value )); + VERIFY(( !internal::has_binary_operator >::value )); + VERIFY(( internal::functor_has_linear_access >::ret )); + + VERIFY(( !internal::has_nullary_operator >::value )); + VERIFY(( internal::has_unary_operator >::value )); + VERIFY(( !internal::has_binary_operator >::value )); + VERIFY(( internal::functor_has_linear_access >::ret )); + } +} + +EIGEN_DECLARE_TEST(nullary) { CALL_SUBTEST_1( testMatrixType(Matrix2d()) ); CALL_SUBTEST_2( testMatrixType(MatrixXcf(internal::random(1,300),internal::random(1,300))) ); CALL_SUBTEST_3( testMatrixType(MatrixXf(internal::random(1,300),internal::random(1,300))) ); - for(int i = 0; i < g_repeat; i++) { - CALL_SUBTEST_4( testVectorType(VectorXd(internal::random(1,300))) ); + for(int i = 0; i < g_repeat*10; i++) { + CALL_SUBTEST_3( testVectorType(VectorXcd(internal::random(1,30000))) ); + CALL_SUBTEST_4( testVectorType(VectorXd(internal::random(1,30000))) ); CALL_SUBTEST_5( testVectorType(Vector4d()) ); // regression test for bug 232 CALL_SUBTEST_6( testVectorType(Vector3d()) ); - CALL_SUBTEST_7( testVectorType(VectorXf(internal::random(1,300))) ); + CALL_SUBTEST_7( testVectorType(VectorXf(internal::random(1,30000))) ); CALL_SUBTEST_8( testVectorType(Vector3f()) ); + CALL_SUBTEST_8( testVectorType(Vector4f()) ); + CALL_SUBTEST_8( testVectorType(Matrix()) ); CALL_SUBTEST_8( testVectorType(Matrix()) ); + + CALL_SUBTEST_9( testVectorType(VectorXi(internal::random(1,10))) ); + CALL_SUBTEST_9( testVectorType(VectorXi(internal::random(9,300))) ); + CALL_SUBTEST_9( testVectorType(Matrix()) ); } + + CALL_SUBTEST_6( bug79<0>() ); + CALL_SUBTEST_6( bug1630<0>() ); + CALL_SUBTEST_9( nullary_overflow<0>() ); + CALL_SUBTEST_10( nullary_internal_logic<0>() ); } diff --git a/thirdparty/eigen/test/num_dimensions.cpp b/thirdparty/eigen/test/num_dimensions.cpp new file mode 100644 index 00000000..7ad7ef69 --- /dev/null +++ b/thirdparty/eigen/test/num_dimensions.cpp @@ -0,0 +1,90 @@ +// This file is part of Eigen, a lightweight C++ template library +// for linear algebra. +// +// Copyright (C) 2018 Gael Guennebaud +// +// This Source Code Form is subject to the terms of the Mozilla +// Public License v. 2.0. If a copy of the MPL was not distributed +// with this file, You can obtain one at http://mozilla.org/MPL/2.0/. + +#include "main.h" +#include + +template +void check_dim(const Xpr& ) { + STATIC_CHECK( Xpr::NumDimensions == ExpectedDim ); +} + +#if EIGEN_HAS_CXX11 +template