This is an archive of the discontinued LLVM Phabricator instance.

AMDGPU: waitcnt operand fixes
ClosedPublic

Authored by nhaustov on Jan 19 2016, 4:51 AM.

Details

Summary

Allow lgkmcnt up to 0xF (hardware allows that).
Fix mask for ExpCnt in AMDGPUInstPrinter.

Diff Detail

Repository
rL LLVM

Event Timeline

nhaustov updated this revision to Diff 45243.Jan 19 2016, 4:51 AM
nhaustov retitled this revision from to AMDGPU: waitcnt operand fixes.
nhaustov updated this object.
nhaustov added reviewers: tstellarAMD, arsenm.
nhaustov set the repository for this revision to rL LLVM.
arsenm edited edge metadata.Jan 19 2016, 10:45 AM

This is a 5 bit field so isn't the max 0x1f?

This is a 5 bit field so isn't the max 0x1f?

That's what the doc says, however it might be a typo/mistake. AMD internal code has 4, including headers with bitfield definitions. So only 4 is safe for now.

arsenm accepted this revision.Jan 19 2016, 11:12 AM
arsenm edited edge metadata.

LGTM

This revision is now accepted and ready to land.Jan 19 2016, 11:12 AM