Code

[JS/WebGPU] Optimize MatMulNBits#19852

Code

Merged

satyajandhyala merged 12 commits intomainmicrosoft/onnxruntime:mainfrom sajandhy/webgpu_optimize_matmulnbitsmicrosoft/onnxruntime:sajandhy/webgpu_optimize_matmulnbitsCopy head branch name to clipboardMar 13, 2024

+194-71

Commits

Commits on Mar 10, 2024

• Temporarily remove uniforms to debug.

  satyajandhyala

  committed
  d114d9a

  Copy full SHA for d114d9a

Commits on Mar 11, 2024

• Vectorize MatMulNBits.

  satyajandhyala

  committed
  f186004

  Copy full SHA for f186004

• Use uppercase letters for M, N and K.

  satyajandhyala

  committed
  a825fbb

  Copy full SHA for a825fbb

• Restore uniforms.

  satyajandhyala

  committed
  daa3730

  Copy full SHA for daa3730

• Added a testcase with 8x8 output.

  satyajandhyala

  committed
  171362b

  Copy full SHA for 171362b

Commits on Mar 12, 2024

• Added output values vectorization.

  satyajandhyala

  committed
  2361183

  Copy full SHA for 2361183

• Revert "Use uppercase letters for M, N and K."

  Show description for f8131e7
  satyajandhyala

  committed
  f8131e7

  Copy full SHA for f8131e7

• Formating changes.

  satyajandhyala

  committed
  602f07f

  Copy full SHA for 602f07f

• Formatting

  satyajandhyala

  committed
  e65f050

  Copy full SHA for e65f050

• typo

  satyajandhyala

  committed
  143faa7

  Copy full SHA for 143faa7

• Use mat4x2 and mat2x4 instead of array when aComponent is 2 or 4.

  satyajandhyala

  committed
  e99fd21

  Copy full SHA for e99fd21

• Fix lint errors.

  satyajandhyala

  committed
  357a3a7

  Copy full SHA for 357a3a7