Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 4, 2025

• Fix padding initialization bug

  Show description for 43fd721
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  43fd721

  Copy full SHA for 43fd721

• Unpoison AMX results for msan

  Show description for 933c427
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  933c427

  Copy full SHA for 933c427

• Fixed weights cache tests inside fully_connected operator.

  Show description for b69a4cf
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  b69a4cf

  Copy full SHA for b69a4cf

• Remove empty else for xnn_f32_avgpool_minmax_ukernel_9p__hvx_u32 in avgpool config

  Show description for f8ad4e1
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  f8ad4e1

  Copy full SHA for f8ad4e1

• Simplified fully_connected operator usage of nr.

  Show description for 5047c42
  

  Aelphy

  authored and

  xnnpack-bot

  committed
  5047c42

  Copy full SHA for 5047c42

• Refactor batch matrix multiply kernel create functions.

  Show description for b046689
  

  qukhan

  authored and

  xnnpack-bot

  committed
  b046689

  Copy full SHA for b046689

• docs: add some basic build documentation

  Show description for 19abccf
  xnnpack-bot

  committed
  19abccf

  Copy full SHA for 19abccf

• Regenerate microkernels for load_ps (aligned)

  Show description for 3cf4fec
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  3cf4fec

  Copy full SHA for 3cf4fec

• Fix build error for reduction when avx disabled

  Show description for 557ca73
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  557ca73

  Copy full SHA for 557ca73

• Configs add else fallback on scalar microkernels when AVX disabled

  Show description for ddcdf14
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  ddcdf14

  Copy full SHA for ddcdf14

• xnn_enable_cpuinfo for x86

  Show description for ead5968
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  ead5968

  Copy full SHA for ead5968

• Fix typo for x8-packw.inc use XNN_ENABLE_AVX256SKX not NN_ENABLE_AVX256VNNI

  Show description for 25d8538
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  25d8538

  Copy full SHA for 25d8538

• rdsum2 cleanup vmask uninitialized

  Show description for 0d595e4
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  0d595e4

  Copy full SHA for 0d595e4

• Fix for strict aliasing build error under bazel

  Show description for dcc91f9
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  dcc91f9

  Copy full SHA for dcc91f9

• rdsum2 build error: 'vmask' may be used uninitialized

  Show description for e25a596
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  e25a596

  Copy full SHA for e25a596

Commits on Nov 3, 2025

• Use the same logic for `nr2` kernels in both `fully-connected` and `dynamic-fully-connected` ops.

  Show description for 9e86c0f
  

  gonnet

  authored and

  xnnpack-bot

  committed
  9e86c0f

  Copy full SHA for 9e86c0f

• Remove slicing and expansion of quantization parameters based on k-dimensions.

  Show description for 3d7e902
  

  vksnk

  authored and

  xnnpack-bot

  committed
  3d7e902

  Copy full SHA for 3d7e902

• Use aligned `_mm.*_load_ps` for weights where appropriate.

  Show description for 41b3ce3
  

  gonnet

  authored and

  xnnpack-bot

  committed
  41b3ce3

  Copy full SHA for 41b3ce3

• Subgraph rewrites for trivial arithmetic operations:

  Show description for e7276ec
  

  gonnet

  authored and

  xnnpack-bot

  committed
  e7276ec

  Copy full SHA for e7276ec

• Expand dims of the zero_point and scale buffers to match the dims of the input for the qd8 case.

  Show description for 9560d7a
  

  vksnk

  authored and

  xnnpack-bot

  committed
  9560d7a

  Copy full SHA for 9560d7a

• Remove accidental overwrite of `igemm_compute->tile[0] = mr` for inline packed LHSs in `convolution-nhwc`.

  Show description for e0e024e
  

  gonnet

  authored and

  xnnpack-bot

  committed
  e0e024e

  Copy full SHA for e0e024e

Commits on Oct 31, 2025

• Also pass copts to the target independent library

  Show description for 48c1b36
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  48c1b36

  Copy full SHA for 48c1b36

• Add basic ARM bf16 kernels

  Show description for 4829ecf
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  4829ecf

  Copy full SHA for 4829ecf

• Fix build error for posix_memalign

  Show description for 3cc15b4
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  3cc15b4

  Copy full SHA for 3cc15b4

• Remove microkernel_lists_test and associated rules

  Show description for fa28ee5
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  fa28ee5

  Copy full SHA for fa28ee5

• Conditionally disable test logic related to 1x1 convolution rewrites when building with YNNPACK.

  Show description for cb9e847
  

  vksnk

  authored and

  xnnpack-bot

  committed
  cb9e847

  Copy full SHA for cb9e847

• Add a transpose kernel for block_n = 64 tile_k = 1 kernels

  Show description for 0eb475b
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  0eb475b

  Copy full SHA for 0eb475b

• Use MSVC style command line arguments for clang-cl

  Show description for 60b257a
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  60b257a

  Copy full SHA for 60b257a

• Update dependencies (slinky)

  Show description for ac8a153
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  ac8a153

  Copy full SHA for ac8a153

Commits on Oct 30, 2025

• If B is not constant, reorder the loop over n outside the loop over m

  Show description for 2c1d953
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  2c1d953

  Copy full SHA for 2c1d953

• Add interleave kernels for 16- and 32-bit data

  Show description for c27cbc5
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  c27cbc5

  Copy full SHA for c27cbc5

Commits on Oct 29, 2025

• Also split channelwise filter when the convolution has number of groups > 1.

  Show description for fa16bf0
  

  vksnk

  authored and

  xnnpack-bot

  committed
  fa16bf0

  Copy full SHA for fa16bf0

• Add -O2 to kernel cc_library rules

  Show description for 7cf2a94
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  7cf2a94

  Copy full SHA for 7cf2a94

• Subgraph rewrites for Binary `minimum`, `maximum`, and Unary `clamp` nodes:

  Show description for e9bc43c
  

  gonnet

  authored and

  xnnpack-bot

  committed
  e9bc43c

  Copy full SHA for e9bc43c

• Reformat string literals in all `.c`, `.cc`, and `.h` files.

  Show description for e2430de
  

  gonnet

  authored and

  xnnpack-bot

  committed
  e2430de

  Copy full SHA for e2430de

Pagination

Previous

Next