Add implementation of WhereAmI() to support NO_AFFINITY=0 on ARM64

[martin-frbg]

fixes #4623

[NickelWenzel]

@martin-frbg this change might have broken windows arm64 builds, see the failed azure pipelines in this PR which prevents from bumping the version to 0.3.28 in the vcpkg port.

At least that is what i see in the attached log.

Change Dir: 'D:/b/openblas/arm64-windows-dbg'

Run Build Command(s): "C:/Program Files/Microsoft Visual Studio/2022/Enterprise/Common7/IDE/CommonExtensions/Microsoft/CMake/Ninja/ninja.exe" -v -v -j33 install
[1/1740] C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1438~1.331\bin\Hostx64\arm64\cl.exe    -ID:\b\openblas\src\v0.3.28-8b4f07f037.clean -ID:\b\openblas\arm64-windows-dbg /nologo /DWIN32 /D_WINDOWS /utf-8 /MP   -DGEMM_GEMV_FORWARD -DNO_LAPACK -DNO_LAPACKE -DNO_WARMUP -DMAX_CPU_NUMBER=0 -DMAX_PARALLEL_NUMBER=1 -DNO_AFFINITY -DVERSION="\"0.3.28\"" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 /D_DEBUG /MDd /Z7 /Ob0 /Od /RTC1 /showIncludes /Fointerface\CMakeFiles\interface.dir\CMakeFiles\cnrm2.c.obj /Fdinterface\CMakeFiles\interface.dir\ /FS -c D:\b\openblas\arm64-windows-dbg\interface\CMakeFiles\cnrm2.c
FAILED: interface/CMakeFiles/interface.dir/CMakeFiles/cnrm2.c.obj 
C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1438~1.331\bin\Hostx64\arm64\cl.exe    -ID:\b\openblas\src\v0.3.28-8b4f07f037.clean -ID:\b\openblas\arm64-windows-dbg /nologo /DWIN32 /D_WINDOWS /utf-8 /MP   -DGEMM_GEMV_FORWARD -DNO_LAPACK -DNO_LAPACKE -DNO_WARMUP -DMAX_CPU_NUMBER=0 -DMAX_PARALLEL_NUMBER=1 -DNO_AFFINITY -DVERSION="\"0.3.28\"" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 /D_DEBUG /MDd /Z7 /Ob0 /Od /RTC1 /showIncludes /Fointerface\CMakeFiles\interface.dir\CMakeFiles\cnrm2.c.obj /Fdinterface\CMakeFiles\interface.dir\ /FS -c D:\b\openblas\arm64-windows-dbg\interface\CMakeFiles\cnrm2.c
D:\b\openblas\src\v0.3.28-8b4f07f037.clean\common_arm64.h(60): error C2065: '__asm__': undeclared identifier
D:\b\openblas\src\v0.3.28-8b4f07f037.clean\common_arm64.h(60): error C2143: syntax error: missing ';' before 'volatile'
[2/1740] C:\PROGRA~1\MICROS~1\2022\ENTERP~1\VC\Tools\MSVC\1438~1.331\bin\Hostx64\arm64\cl.exe    -ID:\b\openblas\src\v0.3.28-8b4f07f037.clean -ID:\b\openblas\arm64-windows-dbg /nologo /DWIN32 /D_WINDOWS /utf-8 /MP   -DGEMM_GEMV_FORWARD -DNO_LAPACK -DNO_LAPACKE -DNO_WARMUP -DMAX_CPU_NUMBER=0 -DMAX_PARALLEL_NUMBER=1 -DNO_AFFINITY -DVERSION="\"0.3.28\"" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 /D_DEBUG /MDd /Z7 /Ob0 /Od /RTC1 /showIncludes /Fointerface\CMakeFiles\interface.dir\CMakeFiles\sasum.c.obj /Fdinterface\CMakeFiles\interface.dir\ /FS -c D:\b\openblas\arm64-windows-dbg\interface\CMakeFiles\sasum.c
FAILED: interface/CMakeFiles/interface.dir/CMakeFiles/sasum.c.obj 

install-arm64-windows-dbg-out.log
install-arm64-uwp-dbg-out.log

@JAicewizard FYI

[martin-frbg]

Ah, that addition probably needs to be conditional on #ifndef MSVC as Visual Studio does not handle the inline assembly (nor its GNU/AT&T-style syntax). Unfortunately I do not have WoA hardware so I have no idea if there are further obstacles to expect

[NickelWenzel]

The canonical way looking at the other common_xxx.h headers seems to be to surround WhereAmI by #ifndef NO_AFFINITY.

I opened a PR for this. Feel free to have a look and as this is my first contribution (to open source ever) don't hesitate if I didn't set sth up correctly.

[JAicewizard]

Oh yeah I still had that PR open, I had no idea how to debug the faillure. Thanks for noticing the issue and being able to diagnose it!

[martin-frbg]

That's slightly different - when "NO_AFFINITY" is not set, the WhereAmI() function is simply not needed. In your case, it is that particular compiler that does not like the way WhereAmI() is implemented (e.g. LLVM would probably compile it without complaining) - but cpu affinity handling is not implemented in the Windows version of the blas_server code anyway, so in practice either ifndef should work.

[martin-frbg]

C_MSVC actually - sorry. Now hopefully fixed in PR #4927

[martin-frbg]

@NickelWenzel sorry, had not realized that you opened the PR against OpenBLAS rather than the vpkpg port. Let's go with your version of the fix then.