Fixes for LoongArch LSX + fast math#1391

Merged

mr-c merged 15 commits into

simd-everywhere:masterfrom

mr-c:fixes-for-loongarch-lsx-fast-math2

Feb 22, 2026

Commits on Feb 22, 2026

x86/sse2: Fix typo in LSX implementation for simde_mm_cvttpd_epi32

iv-m
authored and
mr-c
committed
x86/sse2 mm_cvttps_epi32 : Fix fast math LSX codepath

iv-m
authored and
mr-c
committed
x86/sse2 loong64: Fix fast math codepath for simde_mm_cvtps_epi32

iv-m
authored and
mr-c
committed
x86/sse: Fix fast math LSX codepath for simde_mm_cmpunord_ps

iv-m
authored and
mr-c
committed
x86/sse2: Fix typo about SIMDE_LOONGARCH_LSX_NATIVE

HecaiYuan
authored and
mr-c
committed
gh-actions gcc-qemu: only add extra repository for gcc-15 on Ubuntu 24.04
mr-c
committed
x86/avx: Workaround for GCC ICE on loongarch64

iv-m
authored and
mr-c
committed
x86/avx, x86/sse: workaround for -Werror=maybe-uninitialized on loongarch64

iv-m
authored and
mr-c
committed
arm neon ext: small adjustment to reduce risk of -Werror=maybe-uninitialized
mr-c
committed
arm/neon: Disable (maybe) uninitialized variable warnings on loongarch64

iv-m
and
mr-c
committed
gcc loong64: work around the vec_perm_const bug in the LoongArch backend

HecaiYuan
and
mr-c
committed
diagnostics: fix typo
mr-c
committed
x86/avx512/fixupimm: work around GCC LoongArch bug 121064

HecaiYuan
authored and
mr-c
committed
test x86 avx512 roundscale{,_round}: skip infinity under SIMDE_FAST_MATH

HecaiYuan
authored and
mr-c
committed
gh-actions gcc-16: armel seems to be no longer available
mr-c
committed