projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: d850b34) | patch
[mlir][VectorToGPU] Support transposed+broadcasted 2D MMA load
authorLei Zhang <antiagainst@google.com>
Thu, 15 Dec 2022 19:34:14 +0000 (19:34 +0000)
committerLei Zhang <antiagainst@google.com>
Thu, 15 Dec 2022 19:34:32 +0000 (19:34 +0000)
commitf1db4aec3083e9388e2b8f38263a5a2f04a9bc02
tree60b982d71ee79aa3c25f10cb731c77c4ed40f5betree | snapshot
parentd850b340a9dd91fdb3c4dddc8328524c721576becommit | diff
[mlir][VectorToGPU] Support transposed+broadcasted 2D MMA load

This is loading from 2-D memref, in addition to D139655 where we
load from 1-D memref cases.

Reviewed By: ThomasRaoux

Differential Revision: https://reviews.llvm.org/D140136
mlir/lib/Conversion/VectorToGPU/VectorToGPU.cpp diff | blob | history
mlir/test/Conversion/VectorToGPU/vector-to-mma-ops.mlir diff | blob | history
Domain: System / Toolchain;