Commits

Branch selector

 master

User selector

All users

All time

Commit History

Commits on Apr 23, 2026

• Add definition int2x4 packed type and reference 4-bit and 2-bit dequantization kernels.

  Show description for 5b531e8
  

  vksnk

  authored and

  xnnpack-bot

  committed
  5b531e8

  Copy full SHA for 5b531e8

• Refactor `make_schedule` to not require a buffer

  Show description for c14e464
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  c14e464

  Copy full SHA for c14e464

• Add flag to disable cache wiping in benchmarks

  Show description for d013dc9
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  d013dc9

  Copy full SHA for d013dc9

• Add squared difference kernels

  Show description for e935697
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  e935697

  Copy full SHA for e935697

• Add more static-reduce benchmarks

  Show description for b92be20
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  b92be20

  Copy full SHA for b92be20

• Remove unused variable

  Show description for 4c2c302
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  4c2c302

  Copy full SHA for 4c2c302

• Fix coefficient in approxgelu test

  Show description for d71a227
  

  veblush

  authored and

  xnnpack-bot

  committed
  d71a227

  Copy full SHA for d71a227

• RVV x32-transposec microkernels updated in bazel

  Show description for 5f0cbd8
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  5f0cbd8

  Copy full SHA for 5f0cbd8

• Fix f32_simd_hvx_test on real HVX hardware

  Show description for 4cf3bfd
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  4cf3bfd

  Copy full SHA for 4cf3bfd

Commits on Apr 22, 2026

• Improve tile sizes for binary x86_avx2 kernels. Tuned with AI agents.

  Show description for 83e0a1b
  

  mariecwhite

  authored and

  xnnpack-bot

  committed
  83e0a1b

  Copy full SHA for 83e0a1b

• Fix CMake build break for f16-gemm on ARM.

  Show description for 597b457
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  597b457

  Copy full SHA for 597b457

• Remove quantized tensors from YNNPACK public API

  Show description for cd29842
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  cd29842

  Copy full SHA for cd29842

• Disable filter_zero_point_must_be_zero test for YNNPACK

  Show description for 7db8197
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  7db8197

  Copy full SHA for 7db8197

• Ninja and CMake build invocations removed the 2* multiplier

  Show description for 8b7cb2a
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  8b7cb2a

  Copy full SHA for 8b7cb2a

• Add and fix expand_dims and broadcast related rewrites

  Show description for 6646492
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  6646492

  Copy full SHA for 6646492

• Improve tile sizes of x86_avx kernels. Tuned with AI agents.

  Show description for fe3c235
  

  mariecwhite

  authored and

  xnnpack-bot

  committed
  fe3c235

  Copy full SHA for fe3c235

• Merge pull request #10018 from KavinSai-synthara:fix/fc-qs8-filter-zero-point-validation

  Show description for 980b868
  xnnpack-bot

  committed
  980b868

  Copy full SHA for 980b868

• Introduce experimental graph builder.

  Show description for 8ed22b4
  

  qukhan

  authored and

  xnnpack-bot

  committed
  8ed22b4

  Copy full SHA for 8ed22b4

• Implement `default`ed `operator==` functions for compatibility with C++17.

  Show description for 6022c49
  

  qukhan

  authored and

  xnnpack-bot

  committed
  6022c49

  Copy full SHA for 6022c49

• Fix unused variable error.

  Show description for fedcadb
  

  qukhan

  authored and

  xnnpack-bot

  committed
  fedcadb

  Copy full SHA for fedcadb

• Clean up dependencies in Tensor API BUILD files.

  Show description for 12451e8
  

  qukhan

  authored and

  xnnpack-bot

  committed
  12451e8

  Copy full SHA for 12451e8

• Polyfill `source_location` for older compilers and standards.

  Show description for 1cba69d
  

  qukhan

  authored and

  xnnpack-bot

  committed
  1cba69d

  Copy full SHA for 1cba69d

• Improve tile sizes of unary x86_avx2_fma3 kernels. Tuned with AI agents.

  Show description for 2ad25fc
  

  mariecwhite

  authored and

  xnnpack-bot

  committed
  2ad25fc

  Copy full SHA for 2ad25fc

• Improve tile sizes of x86_avx2 kernels. Tuned with AI agents.

  Show description for 748fc6f
  

  mariecwhite

  authored and

  xnnpack-bot

  committed
  748fc6f

  Copy full SHA for 748fc6f

• Merge pull request #10024 from GregoryComer:bf16-minmax-neon

  Show description for 8339714
  xnnpack-bot

  committed
  8339714

  Copy full SHA for 8339714

• Fix unsimplified extents from padding ops.

  Show description for ec4a71a
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  ec4a71a

  Copy full SHA for ec4a71a

• Improve tile sizes of x86_sse2 kernels. Tuned with AI agents.

  Show description for 0a0df7d
  

  mariecwhite

  authored and

  xnnpack-bot

  committed
  0a0df7d

  Copy full SHA for 0a0df7d

Commits on Apr 21, 2026

• Add Neon bf16 minmax kernels

  GregoryComer

  committed
  a28b56b

  Copy full SHA for a28b56b

• Improve tile sizes of x86_sse41 kernels. Tuned with AI agents.

  Show description for 7a3b6ee
  

  mariecwhite

  authored and

  xnnpack-bot

  committed
  7a3b6ee

  Copy full SHA for 7a3b6ee

• Update f16_igemm_bench to include WAsm Relaxed SIMD microkernels

  Show description for f40a338
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  f40a338

  Copy full SHA for f40a338

• F16-IGEMM for WAsm Relaxed SIMD

  Show description for ee5e30f
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  ee5e30f

  Copy full SHA for ee5e30f

• Re-enable bfmmla kernel when msan is enabled

  Show description for adccb40
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  adccb40

  Copy full SHA for adccb40

Commits on Apr 20, 2026

• Merge pull request #9996 from keaganchern:optimized-transpose

  Show description for 6af4ee5
  xnnpack-bot

  committed
  6af4ee5

  Copy full SHA for 6af4ee5

• Add fusion rules for reduce and elementwise arithmetic

  Show description for bfe081d
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  bfe081d

  Copy full SHA for bfe081d

• Avoid constant folding cheap ops or ops that make the data bigger

  Show description for afa0984
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  afa0984

  Copy full SHA for afa0984