projects / platform / upstream / llvm.git / commit
? search:
summary | shortlog | log | commit | commitdiff | tree
(parent: 3be2b0d) | patch
[mlir] Use affine dim instead of symbol in SCFToGPU lowering.
authorTres Popp <tpopp@google.com>
Tue, 20 Oct 2020 09:32:48 +0000 (11:32 +0200)
committerTres Popp <tpopp@google.com>
Tue, 20 Oct 2020 09:56:34 +0000 (11:56 +0200)
commit72d5ac90b9282cd17c8608a756ebb49c19ad4e04
tree26a7c5f30b20453795b2f11a3fb85844e2a24d81tree | snapshot
parent3be2b0d1a1e0b94e8586f695e8174f139f7a84d8commit | diff
[mlir] Use affine dim instead of symbol in SCFToGPU lowering.

This still satisfies the constraints required by the affine dialect and
gives more flexibility in what iteration bounds can be used when
loewring to the GPU dialect.

Differential Revision: https://reviews.llvm.org/D89782
mlir/lib/Conversion/SCFToGPU/SCFToGPU.cpp diff | blob | history
mlir/test/Conversion/SCFToGPU/parallel_loop.mlir diff | blob | history
Domain: System / Toolchain;