This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU/R600: Enable Load combine
ClosedPublic

Authored by jvesely on Aug 25 2016, 6:03 PM.

Diff Detail

Repository
rL LLVM

Event Timeline

jvesely updated this revision to Diff 69307.Aug 25 2016, 6:03 PM
jvesely retitled this revision from to AMDGPU/R600: Enable Load combine.
jvesely updated this object.
jvesely set the repository for this revision to rL LLVM.
jvesely updated this revision to Diff 69374.Aug 26 2016, 8:13 AM
jvesely edited edge metadata.

don't disable amdgcn tests

arsenm accepted this revision.Aug 26 2016, 3:25 PM
arsenm added a reviewer: arsenm.

LGTM

This revision is now accepted and ready to land.Aug 26 2016, 3:25 PM
arsenm added inline comments.Aug 26 2016, 3:26 PM
test/CodeGen/AMDGPU/load-local-i8.ll
853–876 ↗(On Diff #69374)

-DAG checks do not work as expected when you repeat them. This won't count the number of BFEs

jvesely added inline comments.Aug 26 2016, 3:57 PM
test/CodeGen/AMDGPU/load-local-i8.ll
853–876 ↗(On Diff #69374)

is there a correct way to match required number of BFEs? they can be arbitrarily interleaved with ASHR. Even the position of LDS_READ/WRITE is not guaranteed

arsenm added inline comments.Aug 26 2016, 4:59 PM
test/CodeGen/AMDGPU/load-local-i8.ll
853–876 ↗(On Diff #69374)

I don't think so, this problem comes up a lot

jvesely marked 3 inline comments as done.Aug 27 2016, 12:15 PM
jvesely added inline comments.
test/CodeGen/AMDGPU/load-local-i8.ll
853–876 ↗(On Diff #69374)

OK, I'll leave it as is. the test shows intent if it ever gets fixed.

This revision was automatically updated to reflect the committed changes.
jvesely marked an inline comment as done.