arm64 improvements for x86 clmul & fma

[mr-c]

As inspired by @victorshoup in #1349 #1348

[victorshoup]

I tried using this PR on my macbook. The FMA code works perfectly now.
But I'm not getting the PCLMUL code to work: it seems SIMDE_DETECT_CLANG_VERSION_NOT(22,0,0) is true on my platform, so I'm not getting the good stuff...

I'm not sure if it's possible to update my compiler really, and I wouldn't expect others to do so either.
I've just recently updated developer tools on my macbook, so I don't think there is much to do.
Any thoughts?

$ gcc --version
Apple clang version 17.0.0 (clang-1700.3.19.1)
Target: arm64-apple-darwin24.6.0
Thread model: posix
InstalledDir: /Library/Developer/CommandLineTools/usr/bin

[victorshoup]

I suppose I could "brew install gcc"...I think I've done this in the past on another machine...I always get a bit nervous doing that...or do you think I'm generally better off this way?

[victorshoup]

More feedback... I installed gcc-15 using homebrew. My PCLMUL code now fails to compile:

/opt/homebrew/Cellar/gcc/15.2.0/lib/gcc/current/gcc/aarch64-apple-darwin24/15/include/arm_neon.h: In function 'void pclmul_mul1(long unsigned int*, long unsigned int, long unsigned int)':
/opt/homebrew/Cellar/gcc/15.2.0/lib/gcc/current/gcc/aarch64-apple-darwin24/15/include/arm_neon.h:7297:1: error: inlining failed in call to 'always_inline' 'poly128_t vmull_p64(poly64_t, poly64_t)': target specific option mismatch
7297 | vmull_p64 (poly64_t __a, poly64_t __b)
| ^~~~~~~~~
In file included from /Users/shoup/repos/simde/x86/avx512/../sse3.h:30,
from /Users/shoup/repos/simde/x86/avx512/../ssse3.h:30,
from /Users/shoup/repos/simde/x86/avx512/../sse4.1.h:31,
from /Users/shoup/repos/simde/x86/avx512/../sse4.2.h:31,
from /Users/shoup/repos/simde/x86/avx512/../avx.h:32:
/Users/shoup/repos/simde/x86/clmul.h:210:18: note: called from here
210 | vmull_p64(
| ~~~~~~~~~^~~
211 | vgetq_lane_p64(vreinterpretq_p64_u64(simde__m128i_to_neon_u64(a)), (imm8 ) & 1),
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
212 | vgetq_lane_p64(vreinterpretq_p64_u64(simde__m128i_to_neon_u64(b)), (imm8 >> 4) & 1)
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
213 | )
| ~

Note that the calling code is just this:

void
pclmul_mul1 (unsigned long *c, unsigned long a, unsigned long b)
{
   __m128i aa = _mm_setr_epi64( _mm_cvtsi64_m64(a), _mm_cvtsi64_m64(0));
   __m128i bb = _mm_setr_epi64( _mm_cvtsi64_m64(b), _mm_cvtsi64_m64(0));
   _mm_storeu_si128((__m128i*)c, _mm_clmulepi64_si128(aa, bb, 0));
}

Not sure what to make of this...
So for now, my asm hack is still the only working solution for me...

I'm not in a rush, though...I will be happy to help in any way I can...