Build fails on AMD Ryzen 7730U

[DoctorDro]

Dear all,
I am on Fedora WSL, I cloned the repository and did the following steps:

cmake -DBLIS_CONFIG_FAMILY=auto ..
make

I got

In file included from /home/DoctorDro/blis/kernels/zen/3/sup/bli_gemmsup_rv_zen_asm_s6x16m.c:39:
/home/DoctorDro/blis/kernels/zen/3/sup/bli_gemmsup_rv_zen_asm_s6x16m.c: In function ‘bli_sgemmsup_rv_zen_asm_6x8m’:
/home/DoctorDro/blis/frame/include/bli_x86_asm_macros.h:102:21: error: ‘asm’ operand has impossible constraints or there are not enough registers
  102 | #define BEGIN_ASM() __asm__ volatile (
      |                     ^~~~~~~
/home/DoctorDro/blis/frame/include/bli_x86_asm_macros.h:154:21: note: in expansion of macro ‘BEGIN_ASM’
  154 | #define begin_asm() BEGIN_ASM()
      |                     ^~~~~~~~~
/home/DoctorDro/blis/kernels/zen/3/sup/bli_gemmsup_rv_zen_asm_s6x16m.c:1146:5: note: in expansion of macro ‘begin_asm’
 1146 |     begin_asm()
      |     ^~~~~~~~~
compilation terminated due to -Wfatal-errors.
make[2]: *** [kernels/CMakeFiles/zen_KERNELS.dir/build.make:751: kernels/CMakeFiles/zen_KERNELS.dir/zen/3/sup/bli_gemmsup_rv_zen_asm_s6x16m.c.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:1318: kernels/CMakeFiles/zen_KERNELS.dir/all] Error 2

I successfully built with clang, clang++, using the configure script. But when I switch to gcc, g++ I got the error above.
./configure --prefix="/opt/amd/aocl" --enable-threading=openmp --complex-return=intel CC=gcc CXX=g++ auto

[kvaragan]

Hi can you try with dev branch. That contains all the latest fixes.
I think its the same issue as #34

[DoctorDro]

I can verify that they both compile fine. I hope that this inte-complex or complex-intel flag is automatically deactivated when you compile with gcc ? Right? I did not specify it to be on the safe side. But could you please suggest a configure command which is the best according to your experience (in terms of options ---> performance max) for AMD 7730U?

By the way linking and running DGEMM using MKL the program exits doing nothing. I think in MKL they introduced a flag that when they identify AMD architecture they just exit.

[chandrkr]

Hi @DoctorDro,

Could you please share the source code you used, along with the versions of the compiler and assembler you're using to build it? We're currently unable to reproduce the issue on our end, and having these details would greatly help us in debugging.
Thanks!

[chandrkr]

Hi @DoctorDro,

I can verify that they both compile fine. I hope that this inte-complex or complex-intel flag is automatically deactivated when you compile with gcc ? Right? I did not specify it to be on the safe side. But could you please suggest a configure command which is the best according to your experience (in terms of options ---> performance max) for AMD 7730U?

We are by default using the GNU type for complex return, so there is no need to explicitly set intel-complex or complex-intel when compiling with GCC — these flags are automatically inactive in that case.
For best performance on AMD 7730U, the recommended CMake configure command is:

cmake -S <blis path> -B <build_dir> \  
-DCMAKE_CONFIGURATION_TYPES=Release \  
-DBLIS_CONFIG_FAMILY=auto \  
-DENABLE_CBLAS=ON \  
-DENABLE_ADDON=aocl_gemm \  
-DENABLE_THREADING=openmp \  
-DBLAS_INT_SIZE=32 \ 
 -DCMAKE_BUILD_TYPE=Release \  
-DCMAKE_INSTALL_PREFIX=<install_dir>

Please let us know if you need more information.

Thanks!