Code

rCk working truth#1622

Code

Draft

alugorey wants to merge 15 commits intoROCm:rocm_gemm_ckROCm/pytorch:rocm_gemm_ckfrom alugorey:ck_working_truthalugorey/pytorch:ck_working_truthCopy head branch name to clipboard

+7,142-628

Commits

Commits on Jun 26, 2024

• add ck blas backend selector

  jeffdaily

  committed
  b7a21fb

  Copy full SHA for b7a21fb

Commits on Jun 27, 2024

• add composable_kernel submodule

  jeffdaily

  committed
  9534797

  Copy full SHA for 9534797

• copy bfloat16 gemm from fbgemm

  jeffdaily

  committed
  851d4b5

  Copy full SHA for 851d4b5

Commits on Jun 28, 2024

• CK gemm header (ROCm#1445)

  Show description for 9cbbb40
  alugorey

  authored
  9cbbb40

  Copy full SHA for 9cbbb40

Commits on Jul 1, 2024

• use BLAS arg types for ck gemm kernel

  jeffdaily

  committed
  ae6a64b

  Copy full SHA for ae6a64b

Commits on Jul 2, 2024

• swap bf16 for float

  jeffdaily

  committed
  9420e57

  Copy full SHA for 9420e57

• Ck template (ROCm#1447)

  Show description for 6cdf163
  alugorey

  authored
  6cdf163

  Copy full SHA for 6cdf163

Commits on Jul 17, 2024

• Float CK GEMM backend - Initial prototype

  alugorey

  authored
  442ccd7

  Copy full SHA for 442ccd7

Commits on Jul 18, 2024

• bfloat16 and half prototype added

  jeffdaily

  committed
  de05526

  Copy full SHA for de05526

• lint

  jeffdaily

  committed
  dc29f54

  Copy full SHA for dc29f54

• split ck_gemm.hip into 3 files for faster compilation

  jeffdaily

  committed
  6dddb73

  Copy full SHA for 6dddb73

Commits on Aug 1, 2024

• support all layouts

  alugorey

  authored
  248fb31

  Copy full SHA for 248fb31

Commits on Aug 16, 2024

• Only run CK tests on ROCm (ROCm#1527)

  Show description for fb7957d
  alugorey

  authored
  fb7957d

  Copy full SHA for fb7957d

Commits on Aug 20, 2024

• Fix layouts (ROCm#1548)

  alugorey

  authored
  9c2a52c

  Copy full SHA for 9c2a52c

Commits on Oct 7, 2024

• CK SDPA FA working, needs clean up

  alugorey

  committed
  8b8a8e5

  Copy full SHA for 8b8a8e5