OSDN Git Service

AMDGPU/InsertWaitcnts: Update VGPR/SGPR bounds when brackets are merged
authorCarl Ritson <carl.ritson@amd.com>
Wed, 19 Dec 2018 10:17:49 +0000 (10:17 +0000)
committerCarl Ritson <carl.ritson@amd.com>
Wed, 19 Dec 2018 10:17:49 +0000 (10:17 +0000)
commit8cea59c941e5b2cbbc3b56c5e92ad8e3e91f83e4
tree8f4d3beec74b225d9a5669d5fa3f27c8d8377476
parent331ffd31b3dd49b3f02a27556938b836b679f564
AMDGPU/InsertWaitcnts: Update VGPR/SGPR bounds when brackets are merged

Summary:
Fix an issue where VGPR/SGPR bounds are not properly extended when brackets are merged.
This manifests as missing waitcnt insertions when multiple brackets are forwarded to a successor block and the first forward has lower VGPR/SGPR bounds.

Irreducible loop test has been extended based on a CTS failure detected for GFX9.

Reviewers: nhaehnle

Reviewed By: nhaehnle

Subscribers: arsenm, kzhuravl, jvesely, wdng, yaxunl, dstuttard, tpr, t-tye, jfb, llvm-commits

Differential Revision: https://reviews.llvm.org/D55602

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@349611 91177308-0d34-0410-b5e6-96231b3b80d8
lib/Target/AMDGPU/SIInsertWaitcnts.cpp
test/CodeGen/AMDGPU/waitcnt-loop-irreducible.mir