[PATCH] D150571: [AMDGPU] Avoid RegScavenger::forward in copyPhysReg/indirectCopyToAGPR

[Jay Foad via Phabricator via llvm-commits]

foad created this revision.
foad added reviewers: AMDGPU, arsenm, RKSimon, bogner.
Herald added subscribers: StephenFan, kerbowa, hiraditya, tpr, dstuttard, yaxunl, jvesely, kzhuravl.
Herald added a project: All.
foad requested review of this revision.
Herald added subscribers: llvm-commits, wdng.
Herald added a project: LLVM.

RegScavenger::backward is preferred because it does not rely on accurate
kill flags.


Repository:
  rG LLVM Github Monorepo

https://reviews.llvm.org/D150571

Files:
  llvm/lib/Target/AMDGPU/SIInstrInfo.cpp


Index: llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
===================================================================
--- llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
+++ llvm/lib/Target/AMDGPU/SIInstrInfo.cpp
@@ -618,8 +618,8 @@
     }
   }
 
-  RS.enterBasicBlock(MBB);
-  RS.forward(MI);
+  RS.enterBasicBlockEnd(MBB);
+  RS.backward(MI);
 
   // Ideally we want to have three registers for a long reg_sequence copy
   // to hide 2 waitstates between v_mov_b32 and accvgpr_write.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D150571.522169.patch
Type: text/x-patch
Size: 473 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20230515/8a17d80c/attachment.bin>