Page MenuHomePhabricator

[MLIR] Remove all-reduce lowering from GPU to NVVM. Use in-dialect lowering instead.
ClosedPublic

Authored by csigg on Jan 31 2020, 10:40 AM.

Diff Detail

Event Timeline

csigg created this revision.Jan 31 2020, 10:40 AM

Unit tests: fail. 62321 tests passed, 1 failed and 838 were skipped.

failed: MLIR.Conversion/GPUToNVVM/gpu-to-nvvm.mlir

clang-tidy: pass.

clang-format: pass.

Build artifacts: diff.json, clang-tidy.txt, clang-format.patch, CMakeCache.txt, console-log.txt, test-results.xml

Pre-merge checks is in beta. Report issue. Please join beta or enable it for your project.

rriddle retitled this revision from Remove all-reduce lowering from GPU to NVVM. Use in-dialect lowering instead. to [MLIR] Remove all-reduce lowering from GPU to NVVM. Use in-dialect lowering instead..Jan 31 2020, 12:22 PM
herhut added a comment.Feb 4 2020, 2:05 AM

This seems good to me but the tests are failing. From just looking at the code not sure why.

csigg updated this revision to Diff 249084.Mon, Mar 9, 6:44 AM

Avoid DCE in tests by returning values from functions.

csigg updated this revision to Diff 249086.Mon, Mar 9, 6:50 AM

Remove --dump-input-on-failure.

Harbormaster completed remote builds in B48539: Diff 249084.
herhut accepted this revision.Tue, Mar 10, 8:36 AM

Thanks for cleaning this up.

This revision is now accepted and ready to land.Tue, Mar 10, 8:36 AM
This revision was automatically updated to reflect the committed changes.