Google Highway Versions Save

Performance-portable, length-agnostic SIMD with runtime dispatch

0.16.0

2 years ago
  • Add contrib/sort (vectorized quicksort)
  • Add IfNegativeThenElse, IfVecThenElse
  • Add Reverse2,4,8, ReverseBlocks, DupEven/Odd, AESLastRound
  • Add OrAnd, Min128, Max128, Lt128, SumsOf8
  • Support capped/partial vectors on RVV/SVE, int64 in WASM
  • Support SVE2, shared library build
  • Remove deprecated overloads without the required d arg (UpperHalf etc.)

0.15.0

2 years ago
  • New ops: CompressBlendedStore, ConcatOdd/Even, IndicesFromVec
  • New ops: OddEvenBlocks, SwapAdjacentBlocks, Reverse, RotateRight
  • Add bf16, unsigned comparisons, more lane types for Reverse/TableLookupLanes
  • Contrib: add sort(ing network) and dot(product)
  • Targets: update RVV for LLVM, add experimental WASM2
  • Separate library hwy_test for test utils
  • Add non-macro Simd<> aliases
  • Fixes: const V& for GCC, AVX3 BZHI, POPCNT with AVX on MSVC, avoid %zu

0.14.2

2 years ago
  • Add MaskedLoad
  • Fix non-glibc PPC, Windows GCC, MSVC 19.14
  • Opt-in for -Werror; separate design_philosophy.md

0.14.1

2 years ago
  • Add LoadMaskBits, CompressBits[Store]
  • Fix CPU feature check (AES/F16C) and warnings
  • Improved DASSERT - disabled in optimized builds

0.14.0

2 years ago
  • Add SVE, S-SSE3, AVX3_DL targets
  • Support partial vectors in all ops
  • Add PopulationCount, FindFirstTrue, Ne, TableLookupBytesOr0, AESRound, CLMul, MulOdd, HWY_CAP_FLOAT16

0.12.2

2 years ago
  • fix scalar-only test and Windows macro conflict with Load/StoreFence
  • replace deprecated wasm intrinsics

0.12.1

2 years ago
  • doc updates, ARM GCC support, fix s390/ppc, enable AVX2 on MSVC
  • fix warnings, faster ARM div/sqrt, separate hwy_contrib library
  • add Abs(i64)/FirstN/Pause, full support for partial vectors

0.12.0

3 years ago
  • Add Shift*8, Compress16, emulated Scatter/Gather, StoreInterleaved3/4
  • Remove deprecated HWY_*_LANES, deprecate HWY_GATHER_LANES
  • Proper IEEE rounding, reduce libstdc++ usage, inlined math

0.11.1

3 years ago

Fix asan/msan for older clang, finish f16 conversions, fix warnings.

0.11.0

3 years ago
  • Add RVV port (some functions pending compiler update)
  • Allow Shl/ShiftLeftSame on all targets
  • Add mask logical ops
  • Remove BitsFromMask, SumsOfU8x8
  • Strict NaN and float->int
  • More math functions