Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on Nov 5, 2024

• Merge pull request #7377 from Multicorewareinc:master

  Show description for 2772bf1
  xnnpack-bot

  committed
  2772bf1

  Copy full SHA for 2772bf1

• X32-packw AVX and Scalar GIO layout microkernels

  Show description for 8d65f60
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  8d65f60

  Copy full SHA for 8d65f60

Commits on Nov 4, 2024

• F32-GEMM avx512 fix remainder handling when nc > 16

  Show description for 96717b4
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  96717b4

  Copy full SHA for 96717b4

• x32_packw_bench add attention parameters for GIO benchmarks

  Show description for fb229be
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  fb229be

  Copy full SHA for fb229be

• Since the reduction axes will be sorted anyway, remove the requirement that they are already sorted.

  Show description for d62f1eb
  

  gonnet

  authored and

  xnnpack-bot

  committed
  d62f1eb

  Copy full SHA for d62f1eb

• Merge branch 'google:master' into master

  shalinimcw

  authored
  5753c40

  Copy full SHA for 5753c40

• Add x32 avx512 gemm microkernels. These are way faster

  Show description for 1fed338
  

  alankelly

  authored and

  xnnpack-bot

  committed
  1fed338

  Copy full SHA for 1fed338

• Allow for negative `reduction_axes` parameters in the `Reduce` and `StaticReduce` ops.

  Show description for 9dc7da7
  

  gonnet

  authored and

  xnnpack-bot

  committed
  9dc7da7

  Copy full SHA for 9dc7da7

• Binary element ops should process data in blocks of multiples of the element tile, only one block should do remainder handling.

  Show description for 824c6c6
  

  alankelly

  authored and

  xnnpack-bot

  committed
  824c6c6

  Copy full SHA for 824c6c6

• Propagate rank through XNNPack's subgraph

  Show description for ddfdc77
  

  alankelly

  authored and

  xnnpack-bot

  committed
  ddfdc77

  Copy full SHA for ddfdc77

Commits on Nov 3, 2024

• X32-packw for avx NR=8 with GIO layout

  Show description for 97ced0e
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  97ced0e

  Copy full SHA for 97ced0e

Commits on Nov 1, 2024

• On x86 Clang/GCC, only use _Float16 if __F16C__ is defined (i.e.: we are being compiled with `--copt=-mf16c` or some other flag which implies the f16c feature is available, `--cpu=haswell`); we nev…

  Show description for ebd4532
  xnnpack-bot

  committed
  ebd4532

  Copy full SHA for ebd4532

• Use a lookup table for any op we don't have a config for

  Show description for 24a8252
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  24a8252

  Copy full SHA for 24a8252

• Merge pull request #7367 from mcr229:kleidi_fix

  Show description for c88c850
  xnnpack-bot

  committed
  c88c850

  Copy full SHA for c88c850

• Remove unused qs16 to qs8 cvt kernels

  Show description for 4bf6a3d
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  4bf6a3d

  Copy full SHA for 4bf6a3d

• Optimize f16/bf16 abs, sign, negate reference kernels

  Show description for d6d28c6
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  d6d28c6

  Copy full SHA for d6d28c6

• Remove ops we can rely on reference kernels for instead

  Show description for 60927ca
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  60927ca

  Copy full SHA for 60927ca

• Modify Slinky usage in XNNPACK to always link in Slinky when building on Linux-x86; in this configuration, you can enable Slinky via a runtime flag rather than a compile-time flag. (On non-Linux-x8…

  Show description for 8c3d6e5
  xnnpack-bot

  committed
  8c3d6e5

  Copy full SHA for 8c3d6e5

Commits on Oct 31, 2024

• X8-PACKW AVX for F32_QC8W GEMM microkernels

  Show description for 40ed254
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  40ed254

  Copy full SHA for 40ed254

• move src/packing.cc into src/reference/packing

  Show description for f232442
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  f232442

  Copy full SHA for f232442

• Don't generate atan2 for integer datatypes

  Show description for dd91665
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  dd91665

  Copy full SHA for dd91665

• Modify Slinky usage in XNNPACK to allow enabling via a runtime flag instead of (well, in actually addition to) a compile-time flag.

  Show description for b0cc65e
  xnnpack-bot

  committed
  b0cc65e

  Copy full SHA for b0cc65e

• Move XNN_FLAG_SLINKY_ENABLED from subgraph.h into xnnpack.h since that's the "public" interface that (eg) TFLite and friends use.

  Show description for 6bee884
  xnnpack-bot

  committed
  6bee884

  Copy full SHA for 6bee884

• Add experimental reference kernels for all StableHLO ops

  Show description for 3926e26
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  3926e26

  Copy full SHA for 3926e26

Commits on Oct 30, 2024

• Move XNN_FLAG_SLINKY_ENABLED from subgraph.h into xnnpack.h since that's the "public" interface that (eg) TFLite and friends use.

  Show description for 1846008
  xnnpack-bot

  committed
  1846008

  Copy full SHA for 1846008

• Modify Slinky usage in XNNPACK to allow enabling via a runtime flag instead of (well, in actually addition to) a compile-time flag.

  Show description for 49a1929
  xnnpack-bot

  committed
  49a1929

  Copy full SHA for 49a1929

• Add implementation of BFloat16 unary and binary elementwise ops

  Show description for 92b6d87
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  92b6d87

  Copy full SHA for 92b6d87

• Fix generate scripts when run in open source repos

  Show description for 847248f
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  847248f

  Copy full SHA for 847248f

• Enable QS8-PACKW AVX for QD8 VNNI GEMM microkernels

  Show description for 8bcbc32
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  8bcbc32

  Copy full SHA for 8bcbc32

• Remove compute_type in favor of just looking at datatypes

  Show description for dbb48fa
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  dbb48fa

  Copy full SHA for dbb48fa

• Add BFloat16 datatype support for unary and binary ops.

  Show description for 7aeb369
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  7aeb369

  Copy full SHA for 7aeb369

• Fix divide-by-zero in `BatchMatrixMultiply` operator if any of the input dimensions are zero.

  Show description for 1e8b450
  

  gonnet

  authored and

  xnnpack-bot

  committed
  1e8b450

  Copy full SHA for 1e8b450

• Fix syntax errors

  shalinimcw

  committed
  f53ab0e

  Copy full SHA for f53ab0e

• Fix warnings when building with CMake

  Show description for bba8796
  

  dsharletg

  authored and

  xnnpack-bot

  committed
  bba8796

  Copy full SHA for bba8796

• Add gnr256 and future sde flags

  shalinimcw

  committed
  55afdc9

  Copy full SHA for 55afdc9

Pagination

Previous

Next