investigate gcc 16-20251214 arm64 `-ffast-math -march=native` test failures

[mr-c]

I don't remember if we've run -ffast-math on arm64 before, so these might not be unique to gcc-16

Compiler version

gcc 16.0.0 "gcc (Ubuntu 16-20251214-0ubuntu1~0mistercppa2) 16.0.0 20251214 (experimental) [trunk r16-6108-g427ddf95eef]")

• Obvious differences, like x86/avx512/fpclass (../test/x86/avx512/fpclass.c:46)
• sign flips, like "-452.779999 ~= 452.779999" (../test/x86/avx512/range.c:168)
  • all the test/x86/avx512/range{,_round} failures
• Unclassified, need to extract full logs:
  • x86/svml/native/* (but not emul)
  • x86/sse{,2}/*
  • x86/avx512/xorsign/*

Log extract

test/x86/avx512/fpclass-native-c

not ok 1 fpclass/mm256_fpclass_ps_mask
ok 2 fpclass/mm512_fpclass_ph_mask
not ok 3 fpclass/mm512_fpclass_pd_mask
--- stderr ---
../test/x86/avx512/fpclass.c:46: assertion failed: simde_mm256_fpclass_ps_mask(a, 0x04) == 32 (0 == 32)
../test/x86/avx512/fpclass.c:182: assertion failed: simde_mm512_fpclass_pd_mask(a, 0x04) == 32 (0 == 32)

test/x86/avx512/range-native-c

not ok 1 range/mm_range_ps
not ok 2 range/mm_mask_range_ps
not ok 3 range/mm_maskz_range_ps
not ok 4 range/mm256_range_ps
not ok 5 range/mm256_mask_range_ps
not ok 6 range/mm256_maskz_range_ps
not ok 7 range/mm512_range_ps
not ok 8 range/mm512_mask_range_ps
not ok 9 range/mm512_maskz_range_ps
not ok 10 range/mm_range_pd
not ok 11 range/mm_mask_range_pd
not ok 12 range/mm_maskz_range_pd
not ok 13 range/mm256_range_pd
not ok 14 range/mm256_mask_range_pd
not ok 15 range/mm256_maskz_range_pd
not ok 16 range/mm512_range_pd
not ok 17 range/mm512_mask_range_pd
not ok 18 range/mm512_maskz_range_pd
not ok 19 range/x_mm_range_ss
not ok 20 range/mm_mask_range_ss
ok 21 range/mm_maskz_range_ss
not ok 22 range/x_mm_range_sd
not ok 23 range/mm_mask_range_sd
not ok 24 range/mm_maskz_range_sd
--- stderr ---
../test/x86/avx512/range.c:47: assertion failed: r[0] ~= simde_mm_loadu_ps(test_vec[0].r)[0] (-837.489990 ~= 837.489990)
../test/x86/avx512/range.c:168: assertion failed: r[1] ~= e[1] (-452.779999 ~= 452.779999)
../test/x86/avx512/range.c:251: assertion failed: r[1] ~= e[1] (467.920013 ~= -467.920013)
../test/x86/avx512/range.c:379: assertion failed: r[3] ~= simde_mm256_loadu_ps(test_vec[0].r)[3] (-546.739990 ~= 546.739990)
../test/x86/avx512/range.c:504: assertion failed: r[6] ~= e[6] (-734.270020 ~= 734.270020)
../test/x86/avx512/range.c:618: assertion failed: r[5] ~= e[5] (-848.080017 ~= 848.080017)
../test/x86/avx512/range.c:815: assertion failed: r[1] ~= simde_mm512_loadu_ps(test_vec[0].r)[1] (-647.159973 ~= 647.159973)
../test/x86/avx512/range.c:948: assertion failed: r[3] ~= e[3] (964.989990 ~= -964.989990)
../test/x86/avx512/range.c:1124: assertion failed: r[0] ~= e[0] (439.739990 ~= -439.739990)
../test/x86/avx512/range.c:1291: assertion failed: r[1] ~= simde_mm_loadu_pd(test_vec[0].r)[1] (-609.240000 ~= 609.240000)
../test/x86/avx512/range.c:1433: assertion failed: r[1] ~= e[1] (444.770000 ~= -444.770000)
../test/x86/avx512/range.c:1507: assertion failed: r[0] ~= e[0] (-104.570000 ~= 104.570000)
../test/x86/avx512/range.c:1599: assertion failed: r[1] ~= simde_mm256_loadu_pd(test_vec[0].r)[1] (-758.340000 ~= 758.340000)
../test/x86/avx512/range.c:1720: assertion failed: r[3] ~= e[3] (913.200000 ~= -913.200000)
../test/x86/avx512/range.c:1803: assertion failed: r[2] ~= e[2] (951.080000 ~= -951.080000)
../test/x86/avx512/range.c:1931: assertion failed: r[0] ~= simde_mm512_loadu_pd(test_vec[0].r)[0] (-577.370000 ~= 577.370000)
../test/x86/avx512/range.c:2056: assertion failed: r[3] ~= e[3] (-31.500000 ~= 31.500000)
../test/x86/avx512/range.c:2188: assertion failed: r[5] ~= e[5] (-322.110000 ~= 322.110000)
../test/x86/avx512/range.c:2314: assertion failed: r[0] ~= simde_mm_loadu_ps(test_vec[2].r)[0] (-585.239990 ~= 585.239990)
../test/x86/avx512/range.c:2433: assertion failed: r[0] ~= simde_mm_loadu_ps(test_vec[1].r)[0] (-285.399994 ~= 285.399994)
../test/x86/avx512/range.c:2662: assertion failed: r[0] ~= simde_mm_loadu_pd(test_vec[1].r)[0] (-660.600000 ~= 660.600000)
../test/x86/avx512/range.c:2816: assertion failed: r[0] ~= simde_mm_loadu_pd(test_vec[6].r)[0] (-471.290000 ~= 471.290000)
../test/x86/avx512/range.c:2914: assertion failed: r[0] ~= simde_mm_loadu_pd(test_vec[2].r)[0] (-653.590000 ~= 653.590000)

test/x86/avx512/range_round-native-c

not ok 1 range_round/mm512_range_round_ps
not ok 2 range_round/mm512_mask_range_round_ps
not ok 3 range_round/mm512_maskz_range_round_ps
not ok 4 range_round/mm512_range_round_pd
not ok 5 range_round/mm512_mask_range_round_pd
not ok 6 range_round/mm512_maskz_range_round_pd
not ok 7 range_round/mm_range_round_ss
ok 8 range_round/mm_mask_range_round_ss
not ok 9 range_round/mm_maskz_range_round_ss
not ok 10 range_round/mm_range_round_sd
ok 11 range_round/mm_mask_range_round_sd
ok 12 range_round/mm_maskz_range_round_sd
--- stderr ---
../test/x86/avx512/range_round.c:137: assertion failed: r[1] ~= simde_mm512_loadu_ps(test_vec[0].r)[1] (-357.559998 ~= 357.559998)
../test/x86/avx512/range_round.c:379: assertion failed: r[1] ~= simde_mm512_loadu_ps(test_vec[2].r)[1] (-13.520000 ~= 13.520000)
../test/x86/avx512/range_round.c:574: assertion failed: r[1] ~= simde_mm512_loadu_ps(test_vec[0].r)[1] (941.440002 ~= -941.440002)
../test/x86/avx512/range_round.c:720: assertion failed: r[0] ~= simde_mm512_loadu_pd(test_vec[1].r)[0] (-738.410000 ~= 738.410000)
../test/x86/avx512/range_round.c:881: assertion failed: r[0] ~= simde_mm512_loadu_pd(test_vec[0].r)[0] (-355.520000 ~= 355.520000)
../test/x86/avx512/range_round.c:1040: assertion failed: r[6] ~= simde_mm512_loadu_pd(test_vec[0].r)[6] (670.850000 ~= -670.850000)
../test/x86/avx512/range_round.c:1172: assertion failed: r[0] ~= simde_mm_loadu_ps(test_vec[3].r)[0] (561.609985 ~= -561.609985)
../test/x86/avx512/range_round.c:1461: assertion failed: r[0] ~= simde_mm_loadu_ps(test_vec[7].r)[0] (609.650024 ~= -609.650024)
../test/x86/avx512/range_round.c:1568: assertion failed: r[0] ~= simde_mm_loadu_pd(test_vec[5].r)[0] (-49.160000 ~= 49.160000)