Merge pull request #379 from oscarbenjamin/pr_windows_arm

oscarbenjamin · web-flow · commit 63a3480fc9dc · 2026-03-23T16:13:35.000Z
Add Windows on ARM wheels.
diff --git a/.github/workflows/buildwheel.yml b/.github/workflows/buildwheel.yml
@@ -10,7 +10,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        os: [ubuntu-22.04, ubuntu-22.04-arm, windows-2022, macos-15-intel, macos-14]
+        os: [ubuntu-22.04, ubuntu-22.04-arm, windows-2022, windows-11-arm, macos-15-intel, macos-14]
 
     steps:
       - uses: actions/checkout@v6.0.2
@@ -22,7 +22,14 @@ jobs:
       - uses: msys2/setup-msys2@v2.30.0
         with:
           msystem: ucrt64
-        if: ${{ startsWith( matrix.os , 'windows' ) }}
+          update: true
+        if: ${{ matrix.os == 'windows-2022' }}
+
+      - uses: msys2/setup-msys2@v2.30.0
+        with:
+          msystem: clangarm64
+          update: true
+        if: ${{ matrix.os == 'windows-11-arm' }}
 
         # Install pkgconfig on Windows from choco rather than from msys and
         # avoid using the Strawberry one.
@@ -41,7 +48,7 @@ jobs:
         uses: pypa/cibuildwheel@298ed2fb2c105540f5ed055e8a6ad78d82dd3a7e # v3.3.1
         env:
           # override setting in pyproject.toml to use msys2 instead of msys64 bash
-          CIBW_BEFORE_ALL_WINDOWS: msys2 -c bin/cibw_before_all_windows.sh
+          CIBW_BEFORE_ALL_WINDOWS: ${{ matrix.os == 'windows-11-arm' && 'msys2 -c bin/cibw_before_all_windows_arm64.sh' || 'msys2 -c bin/cibw_before_all_windows_amd64.sh' }}
 
       - uses: actions/upload-artifact@v7
         with:
@@ -81,12 +88,16 @@ jobs:
           ubuntu-24.04-arm,
           windows-2022,
           windows-2025,
+          windows-11-arm,
           macos-15-intel,
           macos-14,
           macos-15,
           ]
         # This list to be kept in sync with python-requires in pyproject.toml.
         python-version: ['3.11', '3.12', '3.13', '3.13t', '3.14', '3.14t', 'pypy3.11']
+        exclude:
+          - os: windows-11-arm
+            python-version: pypy3.11
 
     steps:
       - uses: actions/setup-python@v6
diff --git a/bin/build_dependencies_unix.sh b/bin/build_dependencies_unix.sh
@@ -18,6 +18,12 @@ set -o errexit
 
 SKIP_GMP=no
 SKIP_MPFR=no
+PATCH_GMP_C23=no
+PATCH_LDD=no
+PATCH_IMMINTRIN=no
+GMP_FAT_ARG="--enable-fat"
+GMP_ASSEMBLY_ARG=
+HOST_ARG=
 
 USE_GMP=gmp
 PATCH_GMP_ARM64=no
@@ -38,6 +44,9 @@ do
       echo "  --host <HOST>     - set the host (target) for GMP build"
       echo "  --skip-gmp        - skip building GMP"
       echo "  --skip-mpfr       - skip building MPFR"
+      echo "  --disable-assembly - disable GMP assembly routines"
+      echo "  --patch-ldd       - patch flint shared linking for mingw on arm64"
+      echo "  --patch-immintrin - patch flint arm64 msvc header to avoid immintrin.h"
       echo
       echo "Legacy options:"
       echo "  --gmp gmp         - build based on GMP (default)"
@@ -84,6 +93,12 @@ do
       SKIP_MPFR=yes
       shift
     ;;
+    --disable-assembly)
+      # GMP does not allow --enable-fat together with --disable-assembly.
+      GMP_FAT_ARG=
+      GMP_ASSEMBLY_ARG="--disable-assembly"
+      shift
+    ;;
     --patch-gmp-arm64)
       # Needed only for GMP 6.2.1 on OSX arm64 (Apple M1) hardware
       # As of GMP 6.3.0 this patch is no longer needed
@@ -95,6 +110,16 @@ do
       PATCH_GMP_C23=yes
       shift
     ;;
+    --patch-ldd)
+      # Needed only for the FLINT shared build on mingw arm64.
+      PATCH_LDD=yes
+      shift
+    ;;
+    --patch-immintrin)
+      # Needed only for the FLINT headers consumed by MSVC on Windows arm64.
+      PATCH_IMMINTRIN=yes
+      shift
+    ;;
     --use-gmp-github-mirror)
       USE_GMP_GITHUB_MIRROR=yes
       shift
@@ -191,7 +216,8 @@ if [ "$USE_GMP" = "gmp" ]; then
       ./configfsf.guess
 
       ./configure --prefix=$PREFIX\
-        --enable-fat\
+        $GMP_FAT_ARG\
+        $GMP_ASSEMBLY_ARG\
         --enable-shared=yes\
         --enable-static=no\
         --host=$HOST_ARG
@@ -310,6 +336,20 @@ echo
 curl -O -L https://github.com/flintlib/flint/releases/download/v$FLINTVER/flint-$FLINTVER.tar.gz
 tar xf flint-$FLINTVER.tar.gz
 cd flint-$FLINTVER
+  if [ "$PATCH_LDD" = "yes" ]; then
+    echo
+    echo --------------------------------------------
+    echo "           patching FLINT"
+    echo --------------------------------------------
+    patch -N -Z -p1 < ../../../bin/patch-flint-windows-arm64-link.diff
+  fi
+  if [ "$PATCH_IMMINTRIN" = "yes" ]; then
+    echo
+    echo --------------------------------------------
+    echo "           patching FLINT"
+    echo --------------------------------------------
+    patch -N -Z -p1 < ../../../bin/patch-flint-windows-arm64-immintrin.diff
+  fi
   ./bootstrap.sh
   ./configure --prefix=$PREFIX\
     --host=$HOST_ARG\
diff --git a/bin/cibw_before_all_windows.sh b/bin/cibw_before_all_windows.sh
@@ -2,52 +2,26 @@
 
 set -o errexit
 
-pacman -S --noconfirm \
-    mingw-w64-ucrt-x86_64-gcc\
-    mingw-w64-ucrt-x86_64-tools-git\
-    m4\
-    make\
-    base-devel\
-    autoconf-wrapper\
-    automake-wrapper\
-    libtool\
-    git\
-    #
-
-bin/build_dependencies_unix.sh \
-    --use-gmp-github-mirror\
-    --patch-C23\
-    #
+if [ $# -lt 2 ]; then
+    echo "usage: $0 <llvm-machine> <build-dependencies-args>..."
+    exit 1
+fi
 
-# Assumes the standard GitHub Actions Windows 2022 runner layout.
-PATH="$PATH:$(find "/c/Program Files/Microsoft Visual Studio/2022/" -name "Hostx86")/x64/"
+llvm_machine="$1"
+shift
 
-if [ "${RUNNER_ARCH}" = "ARM64" ]
-then
-  msvc_machine=arm64
-else
-  msvc_machine=x64
-fi
+bin/build_dependencies_unix.sh "$@"
 
 mkdir -p .local/lib
 cd .local/bin
 for dll_file in libgmp-*.dll libmpfr-*.dll libflint*.dll
 do
-  lib_name=$(basename -s .dll ${dll_file})
-  exports_file=${lib_name}-exports.txt
+  lib_name=$(basename -s .dll "${dll_file}")
   def_file=${lib_name}.def
-  lib_file=${lib_name}.lib
-  name=$(echo ${lib_name}|sed 's/^lib//;s/[-.][0-9].*$//')
-
-  dumpbin //exports ${dll_file} > ${exports_file}
-
-  echo LIBRARY ${lib_name} > ${def_file}
-  echo EXPORTS >> ${def_file}
-  awk 'NR>19 && $4 != "" {print $4 " @"$1}' ${exports_file} >> ${def_file}
-  sed -i 's/$/\r/' ${def_file}
+  name=$(echo "${lib_name}" | sed 's/^lib//;s/[-.][0-9].*$//')
 
-  lib //def:${def_file} //out:${lib_file} //machine:${msvc_machine}
-  rm ${exports_file} ${def_file} ${lib_name}.exp
-  mv ${lib_file} ../lib/${name}.lib
+  gendef "${dll_file}"
+  llvm-lib /def:"${def_file}" /out:"../lib/${name}.lib" /machine:${llvm_machine} /nologo
+  rm "${def_file}"
 done
 cd ../..
diff --git a/bin/cibw_before_all_windows_amd64.sh b/bin/cibw_before_all_windows_amd64.sh
@@ -0,0 +1,22 @@
+#!/bin/bash
+
+set -o errexit
+
+pacman -S --noconfirm \
+    mingw-w64-ucrt-x86_64-gcc \
+    mingw-w64-ucrt-x86_64-llvm-tools \
+    mingw-w64-ucrt-x86_64-tools-git \
+    m4 \
+    make \
+    base-devel \
+    autoconf-wrapper \
+    automake-wrapper \
+    libtool \
+    git \
+    #
+
+bin/cibw_before_all_windows.sh \
+    x64 \
+    --use-gmp-github-mirror \
+    --patch-C23 \
+    #
diff --git a/bin/cibw_before_all_windows_arm64.sh b/bin/cibw_before_all_windows_arm64.sh
@@ -0,0 +1,26 @@
+#!/bin/bash
+
+set -o errexit
+
+pacman -S --noconfirm \
+    mingw-w64-clang-aarch64-toolchain \
+    mingw-w64-clang-aarch64-llvm-tools \
+    mingw-w64-clang-aarch64-tools-git \
+    m4 \
+    make \
+    base-devel \
+    autoconf-wrapper \
+    automake-wrapper \
+    libtool \
+    git \
+    #
+
+bin/cibw_before_all_windows.sh \
+    arm64 \
+    --use-gmp-github-mirror \
+    --disable-assembly \
+    --host aarch64-w64-mingw32 \
+    --patch-C23 \
+    --patch-ldd \
+    --patch-immintrin \
+    #
diff --git a/bin/patch-flint-windows-arm64-immintrin.diff b/bin/patch-flint-windows-arm64-immintrin.diff
@@ -0,0 +1,12 @@
+diff --git a/src/longlong_msc_arm64.h b/src/longlong_msc_arm64.h
+index 44d96f03f..6ade48289 100644
+--- a/src/longlong_msc_arm64.h
++++ b/src/longlong_msc_arm64.h
+@@ -14,7 +14,6 @@
+ 
+ #include <stdlib.h>
+ #include <intrin.h>
+-#include <immintrin.h>
+ 
+ /* Trailing and leading zeros */
+ # define flint_clz _CountLeadingZeros64
diff --git a/bin/patch-flint-windows-arm64-link.diff b/bin/patch-flint-windows-arm64-link.diff
@@ -0,0 +1,27 @@
+diff --git a/Makefile.in b/Makefile.in
+index 0f25aa2b5..f6f46b7d6 100644
+--- a/Makefile.in
++++ b/Makefile.in
+@@ -416,18 +416,12 @@ endif
+ ifneq ($(SHARED), 0)
+ shared: $(FLINT_DIR)/$(FLINT_LIB_FULL)
+ 
+-# The following is to avoid reaching the maximum length of command line
+-# arguments, mainly present on MinGW.
+-define xxx_merged_lobj_rule
+-$(BUILD_DIR)/$(1)_merged.lo: $($(1)_LOBJS) | $(BUILD_DIR)
+-	@$(LD) -r $($(1)_LOBJS) -o $(BUILD_DIR)/$(1)_merged.lo
+-endef
+-$(foreach dir, $(DIRS), $(eval $(call xxx_merged_lobj_rule,$(dir))))
+-MERGED_LOBJS:=$(foreach dir, $(DIRS),$(BUILD_DIR)/$(dir)_merged.lo)
++SHARED_LINK_RSP := $(BUILD_DIR)/libflint-shared.rsp
+ 
+-$(FLINT_DIR)/$(FLINT_LIB_FULL): $(MERGED_LOBJS)
++$(FLINT_DIR)/$(FLINT_LIB_FULL): $(LOBJS) | $(BUILD_DIR)
+ 	@echo "Building $(FLINT_LIB_FULL)"
+-	@$(CC) $(CFLAGS) -shared $(EXTRA_SHARED_FLAGS) $(MERGED_LOBJS) -o $(FLINT_LIB_FULL) $(LDFLAGS) $(LIBS)
++	@: $(file >$(SHARED_LINK_RSP))$(foreach obj,$(LOBJS),$(file >>$(SHARED_LINK_RSP),$(obj)))
++	@$(CC) $(CFLAGS) -shared $(EXTRA_SHARED_FLAGS) @$(SHARED_LINK_RSP) -o $(FLINT_LIB_FULL) $(LDFLAGS) $(LIBS)
+ 	@$(RM_F) $(FLINT_LIB)
+ 	@$(RM_F) $(FLINT_LIB_MAJOR)
+ 	@$(LN_S) $(FLINT_LIB_FULL) $(FLINT_LIB)
diff --git a/doc/source/build.rst b/doc/source/build.rst
@@ -352,21 +352,30 @@ builds on Windows using MSVC.
 
 The `MSYS2 <https://www.msys2.org/>`_ project provides a Unix-like environment
 for Windows and a package manager that can be used to install the dependencies.
-The git repo for ``python-flint`` has a script `bin/cibw_before_all_windows.sh
-<https://github.com/flintlib/python-flint/blob/master/bin/cibw_before_all_windows.sh>`_
-that installs the dependencies under MSYS2 and builds ``GMP``, ``MPFR``,
-``FLINT``. This script is used for building the Windows binaries for PyPI. We
-use the ``UCRT64`` (``mingw-w64-ucrt-x86_64``) toolchain under MSYS2 to build
-``GMP``, ``MPFR`` and ``FLINT`` because it makes it possible to have a fat
-build of ``GMP``
-(``--enable-fat``) which bundles micro-architecture specific optimisations for
-``x86_64`` in a redistributable shared library. This is important for
-performance on modern ``x86_64`` CPUs and is not possible if building ``GMP``
-with MSVC. The Python extension modules themselves are then built with MSVC via
+The git repo for ``python-flint`` has scripts
+`bin/cibw_before_all_windows_amd64.sh
+<https://github.com/flintlib/python-flint/blob/master/bin/cibw_before_all_windows_amd64.sh>`_
+and `bin/cibw_before_all_windows_arm64.sh
+<https://github.com/flintlib/python-flint/blob/master/bin/cibw_before_all_windows_arm64.sh>`_
+that install the dependencies under MSYS2 and build ``GMP``, ``MPFR``,
+``FLINT``. These scripts are used for building the Windows binaries for PyPI.
+
+For ``x86_64`` wheels we use the ``UCRT64`` (``mingw-w64-ucrt-x86_64``)
+toolchain under MSYS2 to build ``GMP``, ``MPFR`` and ``FLINT`` because it
+makes it possible to have a fat build of ``GMP`` (``--enable-fat``) which
+bundles micro-architecture specific optimisations for ``x86_64`` in a
+redistributable shared library. This is important for performance on modern
+``x86_64`` CPUs and is not possible if building ``GMP`` with MSVC.
+
+For Windows ``arm64`` wheels we use the ``CLANGARM64`` MSYS2 toolchain instead.
+The ``GMP`` build there does not use ``--enable-fat`` and instead uses the
+generic build that works with that toolchain.
+
+The Python extension modules themselves are then built with MSVC via
 ``meson --vsenv`` while linking against the MSYS2-built ``GMP``, ``MPFR`` and
 ``FLINT`` libraries through ``pkg-config``. This mixed-toolchain arrangement
-keeps the ``GMP`` fat build while using the standard Windows compiler for the
-extension modules.
+keeps the MSYS2 dependency builds while using the standard Windows compiler for
+the extension modules on both ``x86_64`` and ``arm64``.
 
 
 .. _non_standard_location:
diff --git a/pyproject.toml b/pyproject.toml
@@ -94,7 +94,7 @@ files = ["src"]
 addopts = "--import-mode=importlib"
 
 [tool.cibuildwheel]
-skip = "*-win32 *-manylinux_i686 *-manylinux_armv7l *-musllinux_*"
+skip = "*-win32 pp*-win_arm64 *-manylinux_i686 *-manylinux_armv7l *-musllinux_*"
 
 enable = [
     # Uncomment this to test beta versions of CPython in CI (but comment out
@@ -154,7 +154,6 @@ repair-wheel-command = [
 ]
 
 [tool.cibuildwheel.windows]
-before-all = "C:\\msys64\\usr\\bin\\bash bin/cibw_before_all_windows.sh"
 before-build = "pip install wheel delvewheel"
 config-settings = {setup-args = ["--vsenv"], build-dir = "build"}
 repair-wheel-command = [
