Commits

Branch selector

 master

User selector

All users

Datepicker

All time

Commit History

Commits on May 17, 2022

• Enable 4X8 IGEMM for Cortex A7/32/A35

  Show description for 1060c81
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  1060c81

  Copy full SHA for 1060c81

• 4X8 IGEMM for Cortex A7/32/A35

  Show description for ed4e203
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  ed4e203

  Copy full SHA for ed4e203

• Add scalar 24 bit transpose ukernels with tests and bench.

  Show description for ddeeed7
  

  alankelly

  authored and

  xnnpack-bot

  committed
  ddeeed7

  Copy full SHA for ddeeed7

Commits on May 16, 2022

• QU8 AArch32 4x8 GEMM microkernel remove stack padding to 16 bytes.

  Show description for 469e614
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  469e614

  Copy full SHA for 469e614

• Increase MAX MR to 8 for 8x16 microkernels

  Show description for e30b1b1
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  e30b1b1

  Copy full SHA for e30b1b1

• Use helper to check that input datatype matches output datatype

  Show description for e3b8f13
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  e3b8f13

  Copy full SHA for e3b8f13

• Do not create code cache in runtime if XNN_ENABLE_JIT is false

  Show description for 1b4fcd6
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  1b4fcd6

  Copy full SHA for 1b4fcd6

• Fix bug in Fully Connected operator when using SR>1 microkernels

  Show description for 540190a
  

  Maratyszcza

  authored and

  xnnpack-bot

  committed
  540190a

  Copy full SHA for 540190a

Commits on May 15, 2022

• Enable Q8 1x8 AArch32 GEMM for Cortex A55r0

  Show description for 7a44af2
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  7a44af2

  Copy full SHA for 7a44af2

Commits on May 13, 2022

• Avoid allocating an unused memory buffer when we grow the cache

  Show description for 10fd0b1
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  10fd0b1

  Copy full SHA for 10fd0b1

• Use helper to check that output_min and output_max are not NaN

  Show description for 71c5c26
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  71c5c26

  Copy full SHA for 71c5c26

• Add a compile time flag (xnn_enable_gemm_m_specialization) to enable selecting GEMM/IGEMM microkernels of different MR based on a heuristic (defaults to on)

  Show description for 11a8c70
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  11a8c70

  Copy full SHA for 11a8c70

• Fix selection of gemm2 microkernels

  Show description for bfc3670
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  bfc3670

  Copy full SHA for bfc3670

• Enable Q8 1x8 AArch32 GEMM for Cortex A53 / A57

  Show description for 3a59429
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  3a59429

  Copy full SHA for 3a59429

Commits on May 12, 2022

• Allow for small differences in clamp.

  Show description for f973bb3
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  f973bb3

  Copy full SHA for f973bb3

• Enable Q8 1x8 AArch32 GEMM for Cortex A7 / A35

  Show description for 5aede66
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  5aede66

  Copy full SHA for 5aede66

• Q8 1x8 AArch32 GEMM for Cortex A7 / A35

  Show description for 8a624cf
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  8a624cf

  Copy full SHA for 8a624cf

• Use heuristic MR to pick a smaller microkernel for deconvolution subconvolution path

  Show description for ec4fd92
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  ec4fd92

  Copy full SHA for ec4fd92

Commits on May 11, 2022

• Remove the assumption that MR1 kernel always exists by tweaking how we calculate heuristic MR

  Show description for 5bd02aa
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  5bd02aa

  Copy full SHA for 5bd02aa

• Add xx-transpose bench

  Show description for 99834b9
  

  alankelly

  authored and

  xnnpack-bot

  committed
  99834b9

  Copy full SHA for 99834b9

• Fix gemm2 microkernel indices in init.c

  Show description for db977bf
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  db977bf

  Copy full SHA for db977bf

Commits on May 10, 2022

• Fix comment for output_max in Q8 AArch32 GEMM/IGEMM JIT microkernels

  Show description for b1ac6e3
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  b1ac6e3

  Copy full SHA for b1ac6e3

• Fix comment for output_max in Q8 AArch32 GEMM/IGEMM microkernels

  Show description for f13d209
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  f13d209

  Copy full SHA for f13d209

• The assumption that element_size and log2_element_size can safely be in a union fails on an input_shape = [1]

  Show description for d34dc03
  

  alankelly

  authored and

  xnnpack-bot

  committed
  d34dc03

  Copy full SHA for d34dc03

• Add transpose and transpose normalization tests

  Show description for d7b03a3
  

  alankelly

  authored and

  xnnpack-bot

  committed
  d7b03a3

  Copy full SHA for d7b03a3

• Enable QC8 4x8 GEMM for Cortex A32

  Show description for 5876b2b
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  5876b2b

  Copy full SHA for 5876b2b

Commits on May 9, 2022

• QC8 4x8 GEMM for Cortex A35

  Show description for 79749cf
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  79749cf

  Copy full SHA for 79749cf

• Transpose normalization removes dimensions of size 1 before flattening redundant dimensions. Test added.

  Show description for f42a24e
  

  alankelly

  authored and

  xnnpack-bot

  committed
  f42a24e

  Copy full SHA for f42a24e

Commits on May 7, 2022

• Fix JIT QS8 benchmarks, need to call xnn_initialize first

  Show description for 9e2f581
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  9e2f581

  Copy full SHA for 9e2f581

Commits on May 6, 2022

• Allow xnn_ukernel_igemm to have multiple GEMMs

  Show description for b182904
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  b182904

  Copy full SHA for b182904

• Fix CMake build

  Show description for 95d3fb8
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  95d3fb8

  Copy full SHA for 95d3fb8

Commits on May 5, 2022

• f16-dwconv2d-chw/5x5s2p2-neonfp16arith stride 2 microkernels

  Show description for 3caa8b9
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  3caa8b9

  Copy full SHA for 3caa8b9

Commits on May 4, 2022

• Add initial heuristic MR selection and add 4x8 GEMM and IGEMM microkernel for A75

  Show description for 7d7cf97
  

  ngzhian

  authored and

  xnnpack-bot

  committed
  7d7cf97

  Copy full SHA for 7d7cf97

• Apply alphanumeric sort to test functions

  Show description for 145a33d
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  145a33d

  Copy full SHA for 145a33d

• Remove unused FMA parameter from f16-dwconv2d-chw templates

  Show description for b30c8ad
  

  fbarchard

  authored and

  xnnpack-bot

  committed
  b30c8ad

  Copy full SHA for b30c8ad

Pagination

Previous

Next