Commits

Branch selector

 master

User selector

Aelphy

Datepicker

All time

Commit History

Commits on Dec 19, 2025

• Fixed full byte requirement in 2-bit Fully Connected test for qs8 version.

  Show description for f52fbff
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  f52fbff

  Copy full SHA for f52fbff

Commits on Dec 18, 2025

• Added qs8_qc2w FullyConnected.

  Show description for 91bc5c2
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  91bc5c2

  Copy full SHA for 91bc5c2

Commits on Dec 17, 2025

• Improved k1=1 reductions. Most of the cases have significant speed-up except scalar kernels, sum bf16_f32 for f16c and sse2, and x8_s32 for avx2.

  Show description for 103c631
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  103c631

  Copy full SHA for 103c631

• Allowed not to pass channel wise zero point to accommodate asymmetric (zero point is assumed 0) 2-bit case.

  Show description for adf70ec
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  adf70ec

  Copy full SHA for adf70ec

Commits on Dec 16, 2025

• Fixed n=0 case that left uninitialized vectors.

  Show description for 30cd310
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  30cd310

  Copy full SHA for 30cd310

• Attepmted to fix 2-bit fully-connected operator flakyness.

  Show description for d9ad8b4
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  d9ad8b4

  Copy full SHA for d9ad8b4

Commits on Dec 15, 2025

• Added scalar kernels to qc2w gemm-config.

  Show description for f51a395
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  f51a395

  Copy full SHA for f51a395

Commits on Dec 12, 2025

• Added simd extract tests.

  Show description for 7746844
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  7746844

  Copy full SHA for 7746844

• Added convert<bf16x16, f32x16> and convert<f16x16, f32x16> to x86_avx512f to decrease intrinsics usage in reduce kernels.

  Show description for 421648d
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  421648d

  Copy full SHA for 421648d

Commits on Dec 11, 2025

• Added convert<bf16x8, f32x8> to x86_avx2 to decrease intrinsics usage in reduce kernels.

  Show description for 1e2d4d0
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  1e2d4d0

  Copy full SHA for 1e2d4d0

• Fixed int2 fully connected setting producers in the beginning of row_sum rewrite.

  Show description for be53d66
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  be53d66

  Copy full SHA for be53d66

• Templated neondot int2 GEMMs.

  Show description for b666baa
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  b666baa

  Copy full SHA for b666baa

• Added scalar kernels for int2 gemm.

  Show description for dfaf2ae
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  dfaf2ae

  Copy full SHA for dfaf2ae

Commits on Dec 8, 2025

• Added bf16 sum and sum_squared to sse2.

  Show description for 3e588fd
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  3e588fd

  Copy full SHA for 3e588fd

• Added bf16 sum and sum_squared to avx512.

  Show description for 8ff7ddb
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  8ff7ddb

  Copy full SHA for 8ff7ddb

• Added bf16 sum and sum_squared to avx2.

  Show description for 4d0cba7
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  4d0cba7

  Copy full SHA for 4d0cba7

Commits on Dec 6, 2025

• Added bf16 sum and sum_squared to arm neon.

  Show description for 995241f
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  995241f

  Copy full SHA for 995241f

Commits on Dec 5, 2025

• Added sum_squared reductions to arm.

  Show description for 0399c7d
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  0399c7d

  Copy full SHA for 0399c7d

Commits on Dec 4, 2025

• Added sum_squared reductions for more AVX variants.

  Show description for 149835f
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  149835f

  Copy full SHA for 149835f

• Added sum_squared for AVX512BW.

  Show description for 4a84acb
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  4a84acb

  Copy full SHA for 4a84acb

• Added sum_squared for avx512bf16.

  Show description for 2a8e706
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  2a8e706

  Copy full SHA for 2a8e706

• Fixed 2-bit gemm benchmark. Number of elements in packed weights didn't account for row_sum.

  Show description for 549cca9
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  549cca9

  Copy full SHA for 549cca9

• Added Map to ynnpack reductions.

  Show description for 7f398de
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  7f398de

  Copy full SHA for 7f398de

Commits on Nov 25, 2025

• Changing to uint2 computations in kernels.

  Show description for 7b0b9f7
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  7b0b9f7

  Copy full SHA for 7b0b9f7

Commits on Nov 14, 2025

• Fixed xnn_datatype_qcint4 valid data types on subgraph level broken with cl/832274417.

  Show description for 55a55b7
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  55a55b7

  Copy full SHA for 55a55b7

• Fixed static_reduce benchmark. There is no need to change input dim.

  Show description for f091b37
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  f091b37

  Copy full SHA for f091b37

• Added 2-bit qc2w variant for FullyConnected.

  Show description for c36a183
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  c36a183

  Copy full SHA for c36a183

Commits on Nov 6, 2025

• Added subgraph level static_reduce benchmark.

  Show description for dc983b6
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  dc983b6

  Copy full SHA for dc983b6

Commits on Nov 4, 2025

• Fixed weights cache tests inside fully_connected operator.

  Show description for b69a4cf
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  b69a4cf

  Copy full SHA for b69a4cf

• Simplified fully_connected operator usage of nr.

  Show description for 5047c42
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  5047c42

  Copy full SHA for 5047c42

Commits on Oct 21, 2025

• Changed K1_1 strategy for f32 to go with single accumulator and maximally

  Show description for e5cb8c0
  

  Aelphy

  authored and

  JonathanC-ARM

  committed
  e5cb8c0

  Copy full SHA for e5cb8c0

• Added multiple_of to handle all multiples in reductions simply.

  Show description for 7eb618d
  

  Aelphy

  authored and

  JonathanC-ARM

  committed
  7eb618d

  Copy full SHA for 7eb618d

Commits on Oct 20, 2025

• Changed K1_1 strategy for f32 to go with single accumulator and maximally

  Show description for c882917
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  c882917

  Copy full SHA for c882917

• Added multiple_of to handle all multiples in reductions simply.

  Show description for 27f9411
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  27f9411

  Copy full SHA for 27f9411

Commits on Oct 16, 2025

• Added xf16_f32 reductions to neon.

  Show description for 133d109
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  133d109

  Copy full SHA for 133d109

Pagination

Previous

Next