Code

Avx512 perf improvements#996

Code

Open

cchudant wants to merge 7 commits intosonos:mainsonos/tract:mainfrom cchudant:avx512-perfcchudant/tract:avx512-perfCopy head branch name to clipboard

+819-2,532

Commits

Commits on Mar 12, 2023

• kernel generation

  cchudant

  committed
  d5f38b8

  Copy full SHA for d5f38b8

• fix kernel gen, throughput benchmarks

  cchudant

  committed
  5109d97

  Copy full SHA for 5109d97

• unroll on k

  cchudant

  committed
  38b1791

  Copy full SHA for 38b1791

• better scatter/gather, feature for compiling all kernels, better kernel selection

  cchudant

  committed
  c2304c1

  Copy full SHA for c2304c1

• fix CI issues

  cchudant

  committed
  9fa9601

  Copy full SHA for 9fa9601

• fix compile issue

  cchudant

  committed
  8ed3477

  Copy full SHA for 8ed3477

• fix kernel definition on rust side

  cchudant

  committed
  e8f82bc

  Copy full SHA for e8f82bc