[AMDGPU] Fix a potential wrong return value indicating whether a pass modifies a function

[shiltian]

When the alloca is too big for vectorization, the function could have already
been modified in previous iteration of the for loop.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Shilei Tian (shiltian)

Changes

When the alloca is too big for vectorization, the function could have already
been modified in previous iteration of the for loop.

---

Full diff: https://github.com/llvm/llvm-project/pull/88197.diff

1 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp (+1-1)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
index 6f3cdf54dceec7..c0846b123d1870 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp
@@ -336,7 +336,7 @@ bool AMDGPUPromoteAllocaImpl::run(Function &F, bool PromoteToLDS) {
     if (AllocaCost > VectorizationBudget) {
       LLVM_DEBUG(dbgs() << "  Alloca too big for vectorization: " << *AI
                         << "\n");
-      return false;
+      return Changed;
     }
 
     if (tryPromoteAllocaToVector(*AI)) {

[arsenm]

We should also break out of the above loop on the first non-alloca inst

[shiltian]

We should also break out of the above loop on the first non-alloca inst

I'll do it in a follow-up patch.