[PATCH] D22195: AMDGPU: Move SIWholeQuadMode pass to after machine scheduling

[Nicolai Hähnle via llvm-commits]

nhaehnle created this revision.
nhaehnle added reviewers: arsenm, tstellarAMD, mareko.
nhaehnle added a subscriber: llvm-commits.
Herald added subscribers: kzhuravl, arsenm.

Most of this change is about adjusting the analysis, which can now no longer
rely on SSA form and must use LiveIntervals instead.

WQM-related instructions manipulate EXEC and therefore act as a scheduling
barrier for most instructions. With this change, shaders with a mix of WQM
and exact instructions should be scheduled much better because there is no
EXEC interference. This opens the way for additional improvements.

http://reviews.llvm.org/D22195

Files:
  lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
  lib/Target/AMDGPU/SIWholeQuadMode.cpp
  test/CodeGen/AMDGPU/llvm.amdgcn.ps.live.ll
  test/CodeGen/AMDGPU/wqm.ll

-------------- next part --------------
A non-text attachment was scrubbed...
Name: D22195.63420.patch
Type: text/x-patch
Size: 27371 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20160710/152681b0/attachment.bin>