v0.8.4-rc2

What's Changed since v0.8.4-rc1

• prefer NATIVE implementations over SIMDE_VECTOR_SUBSCRIPT_OPS by @mr-c in 4f11a25
• Fix the portable memory functions and test them in CI by @mr-c in #1311
• psabi warnings: always silence by @mr-c in #1332

Clang

• detect-clang: detect version 20 by @mr-c in 7c7031e
• clang: mark two bugs as being fixed in Clang 19 by @mr-c in d6b282d

GCC

• arm64 gcc FRINT: skip native call on GCC by @mr-c in #1299
• gcc 13 + 15: two bugs fixed, skip their workarounds if possible by @mr-c in #1300
• silence some warnings when in -pedantic mode by @mr-c in #1302

MSCV

• x86 avx: fix test for MSVC where CONSTIFY doesn't work by @mr-c in a5f746b

loongarch implementations

• x86 sse2 for loongarch: fix GCC build failure by @jinboson in #1287
• x86 sse2: _mm_div_pd remove broken LSX implementation by @mr-c in 239188b
• x86 sse2: LSX fix for simde_x_mm_select_pd by @mr-c in c05bb81

PowerPC implementations

• x86 sse2: add powerpc*-darwin support for simde_mm_pause by @barracuda156 in #1319

X86

• sse2,sse3, avx: silence some false-positive warnings about uninitialized structs by @mr-c in 437b4f2

MMX

• sll,srl,sra}: clamp and test out-of-range variable shift amounts by @aqrit in 606ceb3 caf9394

SSE*

• {sll,srl,sra}: clamp and test out-of-range variable shift amounts by @aqrit in cf4ec2d 11ac5d8

AVX*

• use memcpy instead of HEDLEY_REINTERPRET_CAST by @mr-c in 40f8f99
• avx: fix native implementation of simde_mm256_permute2f128_si256 by @ethomag in #1338
• {sll,srl,sra}: clamp and test out-of-range variable shift amounts by @aqrit in 7bf1589 edb6dfd
• avx2: make simde_mm256_blend_epi16(a, b, imm8) map to native intrinsic by @ethomag in #1342
• avx2: _mm256_permute4x64_epi64: Avoid "may be used uninitialized" warning by @ethomag in 9898205

AVX512

• misc fixes for AVX512{F,VL}_NATIVE by @mr-c in #1279
• srli: add mask{,z} varients by @mr-c in #1310
• expand{,loadu}: implement more intrinsics by @mr-c in #1312
• cmpeq: finish the rest of the intrinsics by @mr-c in #1314
• slli & srli: add missing intrinsics by @mr-c in #1317
• loadu: almost all of the remaining intrinsics by @mr-c in #1320
• __mmask8 native by @mr-c in #1323
• sub: add mask{,z} varients by @michael-chuh in #1327
• storeu: add _mm256_mask_storeu_epi{8,32,64} by @michael-chuh in #1328
• storeu: fix mask storeu zeroing out original memory by @michael-chuh in #1331
• storeu: implement remaining instrinsics by @mr-c in #1334
• AVX512CD: implement remaining instrinsics by @mr-c in #1336
• cmpneq: finish the implementations; synch with cmpeq by @mr-c in 83141fd

RISC-V

• arm neon cge: riscv64 remove broken vcge_s32 impl by @mr-c in 942df2b
• [Neon][RVV] Enable RVV when there are zve64d and zvl128b flags. by @Ruhung in #1284
• [Neon][RVV] Enable RVV segment load/store only when we have __riscv_zvlsseg flag. by @Ruhung in #1285
• [Neon][RVV] add min.h and max.h RVV implementations. by @Ruhung in #1283
• [Neon][RVV] Fallback to autovec without mrvv-vector-bits flag. by @Ruhung in #1282
• [NEON] Fix mvn.h to correctly handle RVV instructions. by @Ruhung in 1d74b5b

wasm

• remove use of non-standard _div instructions by @mr-c in 36e30c5
• __wasm_fp16__ is required for Float16 with emscripten or wasi by @mr-c in #1333

ARM

• arm fp16 misc cleanups by @mr-c in #1335

NEON

• ld1{,q_x[234]}: speedups on SSE[32] & WASM by @mr-c in #1329
• Avoid undefined behaviour with signed integer multiplication by @rathann in #1296
• Add NEON float16 multi-vectors to native aliases by @stellar-aria in bb7c56f

CI / testing

• build(deps): bump hendrikmuhs/ccache-action from 1.2.16 to 1.2.17 by @dependabot[bot] in #1278
• gh-action: test on clang 19 & 20 by @mr-c in #1303
• build(deps): bump actions/checkout from 4 to 5 by @dependabot[bot] in #1305
• build(deps): bump actions/setup-python from 5 to 6 by @dependabot[bot] in #1316
• build(deps): bump actions/setup-dotnet from 4 to 5 by @dependabot[bot] in #1315
• gh-actions: x86 upgrade to GCC-14 by @mr-c in 1995014
• emscripten64 testing: the "--experimental-wasm-memory64" flag is no longer needed by @mr-c in #1280
• gh-actions: run actionlint automatically by @mr-c in 7da3fb1
• gh-actions: make actionlint/shellcheck clean by @mr-c in 0f2aecf
• gh-actions: simplify gcc loongarch64 testing by @mr-c in c9ee1e6
• gh-actions: macos 16, undef _LIBCPP_ENABLE_ASSERTIONS by @mr-c in 51743e7
• arm neon: extend native alias testing to the vector types by @mr-c in 53f8c08
• circleci: drop loongson test, already covered better in gh-actions by @mr-c in 1dd80a3
• test.h: define __STDC_FORMAT_MACROS if not defined by @barracuda156 in #1318
• Add i686 and clang packit config by @rathann in #1297
• build(deps): bump ad-m/github-push-action from 0.8.0 to 1.0.0 by @dependabot[bot] in #1325
• ccache upgrade by @mr-c in #1324
• gh-actions emscripten: silence several warnings by @mr-c in 47513c1
• Revert "gh-actions: pin emsdk to earlier version until llvm/llvm-project#117200 is fixed and released" by @mr-c in ced7f1e
• x86 avx512: skip testing NANs with fpclass if -ffast-math or equivalent by @mr-c in 0efee69

New Contributors

• @rathann made their first contribution in #1296
• @stellar-aria made their first contribution in #1291
• @aqrit made their first contribution in #1309
• @barracuda156 made their first contribution in #1319
• @michael-chuh made their first contribution in #1327
• @ethomag made their first contribution in #1338

Full Changelog: v0.8.4-rc1...v0.8.4-rc2