Code

[GEMM codegen] Add e2e pipeline#316

Code

Open

Xinyu302 wants to merge 49 commits intobytedance:mainbytedance/byteir:mainfrom Xinyu302:gemm-e2eXinyu302/byteir:gemm-e2eCopy head branch name to clipboard

+2,131-223

Commits

Commits on Jun 23, 2024

• add gpu-to-vector and nvvm lowering and handle dynamic shared memory

  

  yangxinyu

  committed
  0c7ed3e

  Copy full SHA for 0c7ed3e

• remove gpu pipelining

  Show description for 2f1dfc0
  

  yangxinyu

  committed
  2f1dfc0

  Copy full SHA for 2f1dfc0

• clang-format

  

  yangxinyu

  committed
  973466a

  Copy full SHA for 973466a

• dynamic shared mem support in runtime

  

  yangxinyu

  committed
  d534ccb

  Copy full SHA for d534ccb

• modify test file

  

  yangxinyu

  committed
  17fbbb2

  Copy full SHA for 17fbbb2

• handle kernel outline and add barrier at linalg promotion

  

  yangxinyu

  committed
  d60f90c

  Copy full SHA for d60f90c

• add isLinalgOpMatmul

  

  yangxinyu

  committed
  1644860

  Copy full SHA for 1644860

• fix bug of GPUPackSharedMemory

  

  yangxinyu

  committed
  c4d3b35

  Copy full SHA for c4d3b35

Commits on Jun 24, 2024

• fix bug for single iteration

  Show description for 9dcf5d7
  

  yangxinyu

  committed
  9dcf5d7

  Copy full SHA for 9dcf5d7

• add shared memory swizzle

  

  yangxinyu

  committed
  0b256ec

  Copy full SHA for 0b256ec

• support fp16 fp16 fp32

  

  yangxinyu

  committed
  b976885

  Copy full SHA for b976885

• judge fp16 ext matmul

  

  yangxinyu

  committed
  fee4551

  Copy full SHA for fee4551

• fix bug for multibuffer

  

  yangxinyu

  committed
  770a99c

  Copy full SHA for 770a99c

• fix bug for pack shared memory, don't know why yet

  

  yangxinyu

  committed
  98f8f42

  Copy full SHA for 98f8f42

• add whole pipeline

  

  yangxinyu

  committed
  c43cd7b

  Copy full SHA for c43cd7b

Commits on Jun 25, 2024

• add support for bmm

  

  yangxinyu

  committed
  e991f04

  Copy full SHA for e991f04

• add epilogue fusion support at LinalgTensorOp level

  Show description for c17477d
  

  yangxinyu

  committed
  c17477d

  Copy full SHA for c17477d

• bug fix: GPUDistributeWarp tilesize calculation

  

  yangxinyu

  committed
  b8d7471

  Copy full SHA for b8d7471

Commits on Jun 26, 2024

• support TF32x1

  

  yangxinyu

  committed
  0f7c2a7

  Copy full SHA for 0f7c2a7

• disable createGPUBlockSwizzlePass temporarily

  

  yangxinyu

  committed
  bce0e27

  Copy full SHA for bce0e27

Commits on Jul 2, 2024

• Merge remote-tracking branch 'upstream/main' into gemm-e2e

  

  yangxinyu

  committed
  f1b8d2e

  Copy full SHA for f1b8d2e

• fix test failure

  

  yangxinyu

  committed
  24a307d

  Copy full SHA for 24a307d

• add enable_gemm_codegen option

  

  yangxinyu

  committed
  18ae7bc

  Copy full SHA for 18ae7bc

• small fix

  

  yangxinyu

  committed
  e8e0279

  Copy full SHA for e8e0279

• fix small bug

  

  yangxinyu

  committed
  0da6bd1

  Copy full SHA for 0da6bd1

• rename a utils and add support for threadblock swizzle in bmm

  

  yangxinyu

  committed
  5999ee7

  Copy full SHA for 5999ee7

• clang-format

  

  yangxinyu

  committed
  c6b0e59

  Copy full SHA for c6b0e59

• add numerical test for gemm codegen

  

  yangxinyu

  committed
  93e0780

  Copy full SHA for 93e0780

• Merge remote-tracking branch 'upstream/main' into gemm-e2e

  

  yangxinyu

  committed
  b5f4c1a

  Copy full SHA for b5f4c1a

• bug fix

  

  yangxinyu

  committed
  9924fa8

  Copy full SHA for 9924fa8

Commits on Jul 3, 2024

• fix filecheck

  

  yangxinyu

  committed
  b490893

  Copy full SHA for b490893

• fix

  

  yangxinyu

  committed
  36b10ac

  Copy full SHA for 36b10ac

• fix according to xg's opinion

  

  yangxinyu

  committed
  04b2050

  Copy full SHA for 04b2050

• support epilogue fusion started from linalg-memref-opt

  

  yangxinyu

  committed
  3e488e3

  Copy full SHA for 3e488e3

Commits on Jul 4, 2024

• Merge remote-tracking branch 'upstream/main' into gemm-e2e

  

  yangxinyu

  committed
  39671f6

  Copy full SHA for 39671f6

• shape judgement

  

  yangxinyu

  committed
  8e8b2e6

  Copy full SHA for 8e8b2e6

Commits on Jul 5, 2024

• fix tensor shape in matmulEpilogueFusion.mlir

  Xinyu302

  committed
  a372c78

  Copy full SHA for a372c78

• Merge remote-tracking branch 'upstream/main' into gemm-e2e

  Xinyu302

  committed
  9987ca3

  Copy full SHA for 9987ca3

• Merge branch 'main' into gemm-e2e

  Xinyu302

  authored
  94fce00

  Copy full SHA for 94fce00

• support epilogue fussion

  

  yangxinyu

  committed
  d7d714d

  Copy full SHA for d7d714d

• delete useless print

  

  yangxinyu

  committed
  9d53a1e

  Copy full SHA for 9d53a1e

• Merge branch 'main' into gemm-e2e

  

  Xinyu Yang

  authored
  d93c6be

  Copy full SHA for d93c6be

Commits on Jul 16, 2024

• Merge remote-tracking branch 'upstream/main' into gemm-e2e

  

  yangxinyu

  committed
  ee2ab65

  Copy full SHA for ee2ab65

• add test case for matmul epilogue canonicalize

  

  yangxinyu

  committed
  976eff8

  Copy full SHA for 976eff8

Commits on Jul 19, 2024

• to pass test

  

  yangxinyu

  committed
  30f4ef8

  Copy full SHA for 30f4ef8

Commits on Aug 12, 2024

• Merge remote-tracking branch 'upstream/main' into gemm-e2e

  Xinyu302

  committed
  e929786

  Copy full SHA for e929786

Commits on Aug 15, 2024

• fix bug

  Xinyu302

  committed
  aa396a0

  Copy full SHA for aa396a0

• renew test case

  Xinyu302

  committed
  a9e8f3b

  Copy full SHA for a9e8f3b

• add gemm relu test

  Xinyu302

  committed
  d78c3d3

  Copy full SHA for d78c3d3