This is an archive of the discontinued LLVM Phabricator instance.

Differential D133061

[mlir][amdgpu] Fix signed/unsigned comparison for abid/cbsz comparison
ClosedPublic

Authored by rsuderman on Aug 31 2022, 3:17 PM.

Details

Reviewers
herhut
krzysz00
stella.stamenova
Commits
rG7e52e0fc723e: [mlir][amdgpu] Fix signed/unsigned comparison for abid/cbsz comparison
Summary

Unsigned/signed comparison failure due to implicit signed value.

Diff Detail

Event Timeline

rsuderman created this revision.Aug 31 2022, 3:17 PM
Herald added a project: Restricted Project. · View Herald TranscriptAug 31 2022, 3:17 PM
Herald added subscribers: bzcheeseman, kosarev, sdasgup3 and 28 others. · View Herald Transcript
rsuderman requested review of this revision.Aug 31 2022, 3:17 PM
Herald added a reviewer: herhut. · View Herald TranscriptAug 31 2022, 3:17 PM
Herald added a project: Restricted Project. · View Herald Transcript
Herald added subscribers: stephenneuendorffer, nicolasvasilache, wdng. · View Herald Transcript
rsuderman added a reviewer: krzysz00.Aug 31 2022, 3:18 PM
rsuderman added a reviewer: stella.stamenova.Aug 31 2022, 3:24 PM
stella.stamenova accepted this revision.Aug 31 2022, 3:27 PM

Thanks for taking care of this!

This revision is now accepted and ready to land.Aug 31 2022, 3:27 PM
This revision was landed with ongoing or failed builds.Aug 31 2022, 3:30 PM
Closed by commit rG7e52e0fc723e: [mlir][amdgpu] Fix signed/unsigned comparison for abid/cbsz comparison (authored by rsuderman). · Explain Why
This revision was automatically updated to reflect the committed changes.
rsuderman added a commit: rG7e52e0fc723e: [mlir][amdgpu] Fix signed/unsigned comparison for abid/cbsz comparison.
Harbormaster completed remote builds in B184462: Diff 457108.Aug 31 2022, 3:36 PM
krzysz00 added a comment.Aug 31 2022, 3:57 PM

Thanks for catching the bug

Revision Contents

PathSize
mlir/
lib/
Dialect/
AMDGPU/
IR/
2 lines

Diff 457113

mlir/lib/Dialect/AMDGPU/IR/AMDGPUDialect.cpp

Show First 20 Lines • Show All 105 Lines • ▼ Show 20 Lines return emitOpError("expected " + Twine(numDestElems) +
Twine(destLen)); Twine(destLen));
if (destElem.isF64() && getBlgp() != MFMAPermB::none) if (destElem.isF64() && getBlgp() != MFMAPermB::none)
return emitOpError( return emitOpError(
"double-precision ops do not support permuting lanes of B"); "double-precision ops do not support permuting lanes of B");
if (destElem.isF64() && getCbsz() != 0) if (destElem.isF64() && getCbsz() != 0)
return emitOpError( return emitOpError(
"double-precision ops do not support permuting lanes of A"); "double-precision ops do not support permuting lanes of A");
if (getAbid() >= (1 << getCbsz())) if (getAbid() >= (1u << getCbsz()))
return emitOpError( return emitOpError(
"block ID for permuting A (abid) must be below 2 ** cbsz"); "block ID for permuting A (abid) must be below 2 ** cbsz");
if ((getNegateA() || getNegateB() || getNegateC()) && !destElem.isF64()) if ((getNegateA() || getNegateB() || getNegateC()) && !destElem.isF64())
return emitOpError( return emitOpError(
"negation flags only available for double-precision operations"); "negation flags only available for double-precision operations");
return success(); return success();
Show All 9 Lines