This is an archive of the discontinued LLVM Phabricator instance.

[AMDGPU] Preliminary patch for divergence driven instruction selection. Fold immediate SMRD offset.
ClosedPublic

Authored by alex-t on Sep 4 2018, 1:21 AM.

Details

Summary

This patch handles specific processing of the buffer read instruction. It allows to fold immediate offset in case the COPY still retains between the move and read.

Diff Detail

Repository
rL LLVM

Event Timeline

alex-t created this revision.Sep 4 2018, 1:21 AM

Please attach full context diff.

arsenm added inline comments.Sep 4 2018, 11:23 AM
lib/Target/AMDGPU/SIInstrInfo.cpp
4094 ↗(On Diff #163760)

This isn't needed, copy only supports registers

alex-t updated this revision to Diff 164186.Sep 6 2018, 4:15 AM

Unnecessary "isReg()" check removed.
Full context unified diff.

alex-t marked an inline comment as done.Sep 6 2018, 4:15 AM
This revision is now accepted and ready to land.Sep 6 2018, 11:52 AM
This revision was automatically updated to reflect the committed changes.