This is an archive of the discontinued LLVM Phabricator instance.

[AMDGPU] Fold V_CNDMASK with identical source operands
ClosedPublic

Authored by rampitec on Mar 23 2017, 6:25 PM.

Details

Diff Detail

Repository
rL LLVM

Event Timeline

rampitec created this revision.Mar 23 2017, 6:25 PM
arsenm added inline comments.Mar 24 2017, 10:55 AM
lib/Target/AMDGPU/SIFoldOperands.cpp
599 ↗(On Diff #92894)

Probably can drop handling the _e32 version

600–601 ↗(On Diff #92894)

You do not need to check the machine encoded versions

test/CodeGen/AMDGPU/fold-cndmask.mir
29 ↗(On Diff #92894)

Need tests for the e32 version or to drop handling them

rampitec updated this revision to Diff 92982.Mar 24 2017, 11:20 AM

Removed target encoded opcodes.
Added test for _e32 and fixed its handling.

rampitec marked 2 inline comments as done.Mar 24 2017, 11:20 AM
rampitec updated this revision to Diff 92985.Mar 24 2017, 11:25 AM

Fixed gcc warning about enum and non-enum in ?: operator.

arsenm accepted this revision.Mar 24 2017, 11:32 AM

LGTM

This revision is now accepted and ready to land.Mar 24 2017, 11:32 AM
This revision was automatically updated to reflect the committed changes.