Use the flag NUM_THREADS & USE_THREADS while building OpenBLAS for Windows on ARM64.

[Akash9824]

I have observed that when I install NumPy using pip install numpy and run a np.dot() workload, it only utilizes 4 cores (4 threads), even though my Windows on ARM64 device has 12 cores.

I suspect that since we are not using NUM_THREADS while building for ARM64 in this script, it ends up using the number of cores available on the build machine as the value for NUM_THREADS.

To avoid this dependency on the build machine's core count, can we use a flag during the OpenBLAS build similar to what we do for x64 to make the number of threads configurable at runtime?

[mattip]

Nice catch. The macro should be set here or when setting %interface_flags%. Want to send a PR?

openblas-libs/tools/build_steps_win_arm64.bat

Lines 96 to 100 in 970cb75

	:: Run CMake and Ninja build
	cmake .. -G Ninja -DCMAKE_BUILD_TYPE=Release -DTARGET=ARMV8 -DBUILD_SHARED_LIBS=ON -DARCH=arm64 ^
	-DBINARY=%build_bits% -DCMAKE_SYSTEM_PROCESSOR=ARM64 -DCMAKE_C_COMPILER=clang-cl ^
	-DCMAKE_Fortran_COMPILER=flang-new -DSYMBOLPREFIX="scipy_" -DLIBNAMEPREFIX="scipy_" %interface_flags%
	if errorlevel 1 exit /b 1

[Akash9824]

I think we can add these flags here, anyway since the number of threads will be overwritten by the current core count, whether it is 64 or 32.

openblas-libs/tools/build_steps_win_arm64.bat

Lines 96 to 100 in 970cb75

:: Run CMake and Ninja build
cmake .. -G Ninja -DCMAKE_BUILD_TYPE=Release -DTARGET=ARMV8 -DBUILD_SHARED_LIBS=ON -DARCH=arm64 ^
-DBINARY=%build_bits% -DCMAKE_SYSTEM_PROCESSOR=ARM64 -DCMAKE_C_COMPILER=clang-cl ^
-DCMAKE_Fortran_COMPILER=flang-new -DSYMBOLPREFIX="scipy_" -DLIBNAMEPREFIX="scipy_" %interface_flags%
if errorlevel 1 exit /b 1

[mattip]

Closed by #214