projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 4204361) | patch
[mlir][GPUToNVVM] Fix bug in mma elementwise lowering
authorThomas Raoux <thomasraoux@google.com>
Wed, 15 Jun 2022 17:16:56 +0000 (17:16 +0000)
committerThomas Raoux <thomasraoux@google.com>
Wed, 15 Jun 2022 17:23:17 +0000 (17:23 +0000)
commita6f2c2291ede82777fb3b92c6f2ea78d56e97aca
tree2bed031e2db99b2b49038c191c61b0aa8c54492atree | snapshot
parent4204361fed2daffbd71ac2ffdc9387ade1a8f88fcommit | diff
[mlir][GPUToNVVM] Fix bug in mma elementwise lowering

The maxf implementation of wmma elementwise op was incorrect as the
operands of the select to check for Nan were swapped.

Differential Revision: https://reviews.llvm.org/D127879
mlir/lib/Conversion/GPUToNVVM/WmmaOpsToNvvm.cpp diff | blob | history
mlir/test/Conversion/GPUToNVVM/wmma-ops-to-nvvm.mlir diff | blob | history
Domain: System / Toolchain;