[all-commits] [llvm/llvm-project] 792253: [mlir][linalg] Add patterns to convert matmul to t...

[Cullen Rhodes via All-commits]

  Branch: refs/heads/main
  Home:   https://github.com/llvm/llvm-project
  Commit: 79225349748bb556fd027cc0bfeb73b1e9a632f4
      https://github.com/llvm/llvm-project/commit/79225349748bb556fd027cc0bfeb73b1e9a632f4
  Author: Cullen Rhodes <cullen.rhodes at arm.com>
  Date:   2024-04-23 (Tue, 23 Apr 2024)

  Changed paths:
    M mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformEnums.td
    M mlir/include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td
    M mlir/include/mlir/Dialect/Linalg/Transforms/Transforms.h
    M mlir/lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp
    M mlir/lib/Dialect/Linalg/Transforms/CMakeLists.txt
    A mlir/lib/Dialect/Linalg/Transforms/TransposeMatmul.cpp
    A mlir/test/Dialect/Linalg/transpose-matmul-a.mlir
    A mlir/test/Dialect/Linalg/transpose-matmul-b.mlir
    A mlir/test/Dialect/Linalg/transpose-matmul.mlir

  Log Message:
  -----------
  [mlir][linalg] Add patterns to convert matmul to transposed variants (#89075)

This adds patterns to convert from the Linalg matmul and batch_matmul
ops to the transposed variants. By default the LHS matrix is transposed.

Our work enabling a lowering path from linalg.matmul to ArmSME has
revealed the current lowering results in non-contiguous memory accesses
for the A matrix and very poor performance.

These patterns provide a simple option to fix this.



To unsubscribe from these emails, change your notification settings at https://github.com/llvm/llvm-project/settings/notifications