Add support for clang-cl on Windows

[anthony-linaro]

Thir PR adds support for clang-cl (clang, but pretending to be MSVC) to sse2neon on Windows ARM64 platforms. Done as part of some blender work, as using clang-cl gives a ~20-40% speedup compared to MSVC.

This is a WIP, as several tests are still failing in Release - Debug seems to be fine - if anyone has any ideas, please let me know, otherwise I'll keep looking at them through a debugger.

Compiled with the command line (via a VS2022 Native ARM64 Tools CMD window):

msbuild sse2neon.vcxproj /p:Configuration=Release /p:CLToolExe=clang-cl.exe /p:CLToolPath=C:\Program Files\LLVM\bin\

Failing+skipped tests:

Test mm_set_flush_zero_mode         skipped
Test mm_set_rounding_mode           failed
Test mm_setcsr                      failed
Test mm_storeu_si16                 failed
Test mm_storeu_si64                 failed
Test mm_cvtpd_epi32                 failed
Test mm_cvtpd_pi32                  failed
Test mm_cvttpd_epi32                failed
Test mm_cvttpd_pi32                 failed
Test mm_cvttsd_si64x                skipped
Test mm_storeu_si32                 failed
Test mm_alignr_epi8                 skipped
Test mm_alignr_pi8                  skipped
Test mm_set_denormals_zero_mode     failed
Test rdtsc                          failed
SSE2NEONTest Complete!
Passed:  508
Failed:  11
Ignored: 4
Coverage rate: 97.14%

I can provide access to a machine via VPN (wireguard) for the purposes of this PR, if a maintainer without a relevant machine wishes to have a go.

[anthony-linaro]

Taking a deeper look, it seems to me that anything that touches FPCR or uses the msr and mrs asm instruction seems to be failing (plus a few other, non-asm related tests).

[anthony-linaro]

Now only mm_cvttpd_epi32 and mm_cvttpd_pi32 fail, which according to #635, will be fixed in #638

I guess this is ready for review then

[jserv]

Now only mm_cvttpd_epi32 and mm_cvttpd_pi32 fail, which according to #635, will be fixed in #638

Pull request #638 has been merged. Could you rebase and confirm?

[anthony-linaro]

@jserv I am now seeing 2 failures:

Test mm_cvttpd_epi32                failed
Test rdtsc                          failed

[anthony-linaro]

@jserv Are you able to review? Are the test fialures still down to issues with the test harness itself, rather than sse2neon?

I'd like to be able to move blender to using clang-cl for Windows ARM64 platforms, as it has a nearly 40% perf improvement in places, but I would need this to be merged to go ahead.

[jserv]

Thank @anthony-linaro for contributing!