commit	70dd3b8ff958c4cc268d0c324a56cfabe26fcdcf	[log] [tgz]
author	Cullen Rhodes <cullen.rhodes@arm.com>	Tue Apr 23 07:21:06 2024 +0100
committer	Copybara-Service <copybara-worker@google.com>	Mon Apr 22 23:27:48 2024 -0700
tree	d006cc2ed1d15d08463321b1447f167670426477
parent	761715c4268441545f29b157aa8565284a6c4e1e [diff]

[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.

GitOrigin-RevId: 79225349748bb556fd027cc0bfeb73b1e9a632f4

include/mlir/Dialect/Linalg/TransformOps/LinalgTransformEnums.td[diff]
include/mlir/Dialect/Linalg/TransformOps/LinalgTransformOps.td[diff]
include/mlir/Dialect/Linalg/Transforms/Transforms.h[diff]
lib/Dialect/Linalg/TransformOps/LinalgTransformOps.cpp[diff]
lib/Dialect/Linalg/Transforms/CMakeLists.txt[diff]
lib/Dialect/Linalg/Transforms/TransposeMatmul.cpp[Added - diff]
test/Dialect/Linalg/transpose-matmul-a.mlir[Added - diff]
test/Dialect/Linalg/transpose-matmul-b.mlir[Added - diff]
test/Dialect/Linalg/transpose-matmul.mlir[Added - diff]

9 files changed

tree: d006cc2ed1d15d08463321b1447f167670426477

benchmark/
cmake/
docs/
examples/
include/
lib/
python/
test/
tools/
unittests/
utils/
.clang-format
.clang-tidy
CMakeLists.txt
LICENSE.TXT
README.md

README.md

Multi-Level Intermediate Representation

See https://mlir.llvm.org/ for more information.