[llvm] [AMDGPU][NewPM] Make the pass flow consistent with the legacy pipeline. (PR #136551)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Christudasan Devadasan (cdevadas)

<details>
<summary>Changes</summary>



---
Full diff: https://github.com/llvm/llvm-project/pull/136551.diff


1 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp (+3-5) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 90e3489ced923..b6cc5137d711a 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -2081,13 +2081,11 @@ void AMDGPUCodeGenPassBuilder::addCodeGenPrepare(AddIRPass &addPass) const {
 
 void AMDGPUCodeGenPassBuilder::addPreISel(AddIRPass &addPass) const {
 
-  if (TM.getOptLevel() > CodeGenOptLevel::None)
+  if (TM.getOptLevel() > CodeGenOptLevel::None) {
     addPass(FlattenCFGPass());
-
-  if (TM.getOptLevel() > CodeGenOptLevel::None)
     addPass(SinkingPass());
-
-  addPass(AMDGPULateCodeGenPreparePass(TM));
+    addPass(AMDGPULateCodeGenPreparePass(TM));
+  }
 
   // Merge divergent exit nodes. StructurizeCFG won't recognize the multi-exit
   // regions formed by them.

``````````

</details>


https://github.com/llvm/llvm-project/pull/136551