Get ARM64 TARGET information from the registry on Windows

[martin-frbg]

No description provided.

[dnoan]

@martin-frbg , it's great to see support for Snapdragon X Elite on WoA. However, has anyone successfully built and tested the code? I decided to give it a try with clang-cl 19.1.7, but I'm encountering these two LLVM bugs when compiling the NEOVERSEN1 kernels:
llvm/llvm-project#66912
llvm/llvm-project#47432

[8900/12616] Building C object kernel\CMakeFiles\kernel_NEOVERSEN1.dir\CMakeFiles\sasum_k_NEOVERSEN1.c.obj
FAILED: kernel/CMakeFiles/kernel_NEOVERSEN1.dir/CMakeFiles/sasum_k_NEOVERSEN1.c.obj
C:\install\llvm-19.7.0\bin\clang-cl.exe  /nologo  -IC:\src\OpenBLAS\lapack-netlib\LAPACKE\include -IC:\src\OpenBLAS -IC:\src\OpenBLAS\build\kernel_config\NEOVERSEN1 -MD -Wno-everything  -DHAVE_C11 -DMS_ABI -Xclang -fopenmp -DUSE_OPENMP -Wall -DF_INTERFACE_GFORT -DGEMM_GEMV_FORWARD -DSMALL_MATRIX_OPT -DDYNAMIC_ARCH -DSMP_SERVER -DNO_WARMUP -DMAX_CPU_NUMBER=256 -DMAX_PARALLEL_NUMBER=1 -DBUFFERSIZE=22 -DNO_AFFINITY -DVERSION="\"0.3.29\"" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 /O2 /Ob2 /DNDEBUG /Zi -DBUILD_KERNEL -DTABLE_NAME=gotoblas_NEOVERSEN1  -DTS=_NEOVERSEN1 /showIncludes /Fokernel\CMakeFiles\kernel_NEOVERSEN1.dir\CMakeFiles\sasum_k_NEOVERSEN1.c.obj /Fdkernel\CMakeFiles\kernel_NEOVERSEN1.dir\ -c -- C:\src\OpenBLAS\build\kernel\CMakeFiles\sasum_k_NEOVERSEN1.c
fatal error: error in backend: Failed to evaluate function length in SEH unwind info
PLEASE submit a bug report to https://github.com/llvm/llvm-project/issues/ and include the crash backtrace, preprocessed source, and associated run script.
Stack dump:
0.      Program arguments: C:\\install\\llvm-19.7.0\\bin\\clang-cl.exe /nologo -IC:\\src\\OpenBLAS\\lapack-netlib\\LAPACKE\\include -IC:\\src\\OpenBLAS -IC:\\src\\OpenBLAS\\build\\kernel_config\\NEOVERSEN1 -MD -Wno-everything -DHAVE_C11 -DMS_ABI -Xclang -fopenmp -DUSE_OPENMP -Wall -DF_INTERFACE_GFORT -DGEMM_GEMV_FORWARD -DSMALL_MATRIX_OPT -DDYNAMIC_ARCH -DSMP_SERVER -DNO_WARMUP -DMAX_CPU_NUMBER=256 -DMAX_PARALLEL_NUMBER=1 -DBUFFERSIZE=22 -DNO_AFFINITY -DVERSION=\"0.3.29\" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 /O2 /Ob2 /DNDEBUG /Zi -DBUILD_KERNEL -DTABLE_NAME=gotoblas_NEOVERSEN1 -DTS=_NEOVERSEN1 /showIncludes /Fokernel\\CMakeFiles\\kernel_NEOVERSEN1.dir\\CMakeFiles\\sasum_k_NEOVERSEN1.c.obj /Fdkernel\\CMakeFiles\\kernel_NEOVERSEN1.dir\\ -c -- C:\\src\\OpenBLAS\\build\\kernel\\CMakeFiles\\sasum_k_NEOVERSEN1.c
1.      <eof> parser at end of file
2.      Code generation
3.      Running pass 'Function Pass Manager' on module 'C:\src\OpenBLAS\build\kernel\CMakeFiles\sasum_k_NEOVERSEN1.c'.
4.      Running pass 'AArch64 Assembly Printer' on function '@sasum_k_NEOVERSEN1'
Exception Code: 0xE0000046
 #0 0x00007ff931696248 (C:\WINDOWS\System32\KERNELBASE.dll+0xb6248)
 #1 0x00007ff7a9e18788 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xbb8788)
 #2 0x00007ff7a92f50d4 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x950d4)
 #3 0x00007ff7a9e1f2fc (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xbbf2fc)
 #4 0x00007ff7a9e1f48c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xbbf48c)
 #5 0x00007ff7a9ccdb28 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xa6db28)
 #6 0x00007ff7a9ccc150 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xa6c150)
 #7 0x00007ff7a9ccd948 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xa6d948)
 #8 0x00007ff7a9cb525c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xa5525c)
 #9 0x00007ff7aa8c1e24 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x1661e24)
#10 0x00007ff7aa8c1e9c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x1661e9c)
#11 0x00007ff7aa88006c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x162006c)
#12 0x00007ff7a931666c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xb666c)
#13 0x00007ff7a9814cd4 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x5b4cd4)
#14 0x00007ff7a9a8028c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x82028c)
#15 0x00007ff7a9a80484 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x820484)
#16 0x00007ff7a9a8067c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x82067c)
#17 0x00007ff7a9a80018 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x820018)
#18 0x00007ff7aa06a688 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xe0a688)
#19 0x00007ff7aa069a18 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xe09a18)
#20 0x00007ff7aa06a018 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xe0a018)
#21 0x00007ff7aa51029c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x12b029c)
#22 0x00007ff7ab2d2248 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x2072248)
#23 0x00007ff7aa6ef354 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x148f354)
#24 0x00007ff7aa50f308 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x12af308)
#25 0x00007ff7aa6ef210 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x148f210)
#26 0x00007ff7aa6daaf8 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x147aaf8)
#27 0x00007ff7aa76f344 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x150f344)
#28 0x00007ff7a92f6bd8 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x96bd8)
#29 0x00007ff7a92f1210 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x91210)
#30 0x00007ff7aa5ead6c (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x138ad6c)
#31 0x00007ff7a9e185b0 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0xbb85b0)
#32 0x00007ff7aa5eb4bc (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x138b4bc)
#33 0x00007ff7aa54de08 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x12ede08)
#34 0x00007ff7aa54e044 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x12ee044)
#35 0x00007ff7aa5335dc (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x12d35dc)
#36 0x00007ff7a92f3068 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x93068)
#37 0x00007ff7a92ffb90 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x9fb90)
#38 0x00007ff7ac34a624 (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x30ea624)
#39 0x00007ff7ac34a6bc (C:\install\llvm-19.7.0\bin\clang-cl.exe+0x30ea6bc)
#40 0x00007ff932d18740 (C:\WINDOWS\System32\KERNEL32.DLL+0x8740)
#41 0x00007ff9359b1084 (C:\WINDOWS\SYSTEM32\ntdll.dll+0xd1084)
clang-cl: error: clang frontend command failed with exit code 70 (use -v to see invocation)
clang version 19.1.7
Target: aarch64-pc-windows-msvc
Thread model: posix
InstalledDir: C:\install\llvm-19.7.0\bin
clang-cl: note: diagnostic msg:
********************

PLEASE ATTACH THE FOLLOWING FILES TO THE BUG REPORT:
Preprocessed source(s) and associated run script(s) are located at:
clang-cl: note: diagnostic msg: C:\Users\buildusr\AppData\Local\Temp\sasum_k_NEOVERSEN1-6937e0.c
clang-cl: note: diagnostic msg: C:\Users\buildusr\AppData\Local\Temp\sasum_k_NEOVERSEN1-6937e0.sh
clang-cl: note: diagnostic msg:

********************

[martin-frbg]

are you building 0.3.29 or the develop branch ? 0.3.29 will only build with the ARMV8 target on WoA, the workarounds for the unsupported parts of the Neoverse kernels are only in the develop branch ( in the PR before this one)

[dnoan]

That was 0.3.29 with some modifications. I just tried the develop branch and I got these errors:

FAILED: kernel/CMakeFiles/kernel_NEOVERSEN1.dir/CMakeFiles/cdotc_k_NEOVERSEN1.c.obj
C:\install\llvm-19.7.0\bin\clang-cl.exe  /nologo  -IC:\src\OpenBLAS\lapack-netlib\LAPACKE\include -IC:\src\OpenBLAS -IC:\src\OpenBLAS\build\kernel_config\NEOVERSEN1 -MD -Wno-everything  -DHAVE_C11 -DMS_ABI -Xclang -fopenmp -DUSE_OPENMP -Wall -DF_INTERFACE_GFORT -DGEMM_GEMV_FORWARD -DSMALL_MATRIX_OPT -DDYNAMIC_ARCH -DSMP_SERVER -DNO_WARMUP -DMAX_CPU_NUMBER=256 -DMAX_PARALLEL_NUMBER=1 -DBUFFERSIZE=22 -DNO_AFFINITY -DVERSION="\"0.3.29.dev\"" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 /O2 /Ob2 /DNDEBUG /Zi -DBUILD_KERNEL -DTABLE_NAME=gotoblas_NEOVERSEN1  -DTS=_NEOVERSEN1 /showIncludes /Fokernel\CMakeFiles\kernel_NEOVERSEN1.dir\CMakeFiles\cdotc_k_NEOVERSEN1.c.obj /Fdkernel\CMakeFiles\kernel_NEOVERSEN1.dir\ -c -- C:\src\OpenBLAS\build\kernel\CMakeFiles\cdotc_k_NEOVERSEN1.c
In file included from C:\src\OpenBLAS\build\kernel\CMakeFiles\cdotc_k_NEOVERSEN1.c:9:
C:/src/OpenBLAS/kernel/arm64/zdot_thunderx2t99.c(301,5): error: expected expression
  301 |         cf=OPENBLAS_MAKE_COMPLEX_FLOAT(dotr, doti);
      |            ^
C:\src\OpenBLAS\common.h(622,42): note: expanded from macro 'OPENBLAS_MAKE_COMPLEX_FLOAT'
  622 | #define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_float(r,i)
      |                                          ^
C:\src\OpenBLAS\common.h(579,54): note: expanded from macro 'openblas_make_complex_float'
  579 |   #define openblas_make_complex_float(real, imag)    {(real), (imag)}
      |                                                      ^
In file included from C:\src\OpenBLAS\build\kernel\CMakeFiles\cdotc_k_NEOVERSEN1.c:9:
C:/src/OpenBLAS/kernel/arm64/zdot_thunderx2t99.c(350,11): error: expected expression
  350 |                         zdot = OPENBLAS_MAKE_COMPLEX_FLOAT (CREAL(zdot) + CREAL(*ptr), CIMAG(zdot) + CIMAG(*ptr));
      |                                ^
C:\src\OpenBLAS\common.h(622,42): note: expanded from macro 'OPENBLAS_MAKE_COMPLEX_FLOAT'
  622 | #define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_float(r,i)
      |                                          ^
C:\src\OpenBLAS\common.h(579,54): note: expanded from macro 'openblas_make_complex_float'
  579 |   #define openblas_make_complex_float(real, imag)    {(real), (imag)}
      |                                                      ^

FAILED: kernel/CMakeFiles/kernel_NEOVERSEN1.dir/CMakeFiles/cdotu_k_NEOVERSEN1.c.obj
C:\install\llvm-19.7.0\bin\clang-cl.exe  /nologo  -IC:\src\OpenBLAS\lapack-netlib\LAPACKE\include -IC:\src\OpenBLAS -IC:\src\OpenBLAS\build\kernel_config\NEOVERSEN1 -MD -Wno-everything  -DHAVE_C11 -DMS_ABI -Xclang -fopenmp -DUSE_OPENMP -Wall -DF_INTERFACE_GFORT -DGEMM_GEMV_FORWARD -DSMALL_MATRIX_OPT -DDYNAMIC_ARCH -DSMP_SERVER -DNO_WARMUP -DMAX_CPU_NUMBER=256 -DMAX_PARALLEL_NUMBER=1 -DBUFFERSIZE=22 -DNO_AFFINITY -DVERSION="\"0.3.29.dev\"" -DBUILD_SINGLE -DBUILD_DOUBLE -DBUILD_COMPLEX -DBUILD_COMPLEX16 /O2 /Ob2 /DNDEBUG /Zi -DBUILD_KERNEL -DTABLE_NAME=gotoblas_NEOVERSEN1  -DTS=_NEOVERSEN1 /showIncludes /Fokernel\CMakeFiles\kernel_NEOVERSEN1.dir\CMakeFiles\cdotu_k_NEOVERSEN1.c.obj /Fdkernel\CMakeFiles\kernel_NEOVERSEN1.dir\ -c -- C:\src\OpenBLAS\build\kernel\CMakeFiles\cdotu_k_NEOVERSEN1.c
In file included from C:\src\OpenBLAS\build\kernel\CMakeFiles\cdotu_k_NEOVERSEN1.c:8:
C:/src/OpenBLAS/kernel/arm64/zdot_thunderx2t99.c(301,5): error: expected expression
  301 |         cf=OPENBLAS_MAKE_COMPLEX_FLOAT(dotr, doti);
      |            ^
C:\src\OpenBLAS\common.h(622,42): note: expanded from macro 'OPENBLAS_MAKE_COMPLEX_FLOAT'
  622 | #define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_float(r,i)
      |                                          ^
C:\src\OpenBLAS\common.h(579,54): note: expanded from macro 'openblas_make_complex_float'
  579 |   #define openblas_make_complex_float(real, imag)    {(real), (imag)}
      |                                                      ^
In file included from C:\src\OpenBLAS\build\kernel\CMakeFiles\cdotu_k_NEOVERSEN1.c:8:
C:/src/OpenBLAS/kernel/arm64/zdot_thunderx2t99.c(350,11): error: expected expression
  350 |                         zdot = OPENBLAS_MAKE_COMPLEX_FLOAT (CREAL(zdot) + CREAL(*ptr), CIMAG(zdot) + CIMAG(*ptr));
      |                                ^
C:\src\OpenBLAS\common.h(622,42): note: expanded from macro 'OPENBLAS_MAKE_COMPLEX_FLOAT'
  622 | #define OPENBLAS_MAKE_COMPLEX_FLOAT(r,i) openblas_make_complex_float(r,i)
      |                                          ^
C:\src\OpenBLAS\common.h(579,54): note: expanded from macro 'openblas_make_complex_float'
  579 |   #define openblas_make_complex_float(real, imag)    {(real), (imag)}
      |    
                                                  ^
2 errors generated.

[martin-frbg]

ok, I need to check that - zdot_thunderx2t99.c should not get compiled on WoA due to these, but my original patch of the KERNEL file used bad syntax and I did not get around to retesting on WoA after fixing that on Unix