[llvm] [AMDGPU] LRO: allow same-BB non-lookthrough users for PHI (PR #160909)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: None (michaelselehov)

<details>
<summary>Changes</summary>

Loop headers frequently consume the loop-carried value in the header block via non-lookthrough ops (e.g. byte-wise vector binops). LiveRegOptimizer’s same-BB filter currently prunes these users, so the loop-carried PHI is not coerced to i32 and the intended packed form is lost.

Relax the filter: when the def is a PHI, allow same-BB non-lookthrough users. Also fix the check to look at the user (CII) rather than the def (II) so the walk does not terminate prematurely.

---
Full diff: https://github.com/llvm/llvm-project/pull/160909.diff


1 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp (+4-1) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp b/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp
index 38718c43a61dd..7504f1a8cea09 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPULateCodeGenPrepare.cpp
@@ -150,7 +150,10 @@ class LiveRegOptimizer {
       if (!CVisited.insert(CII).second)
         continue;
 
-      if (CII->getParent() == II->getParent() && !IsLookThru(II))
+      // Same-BB filter must look at the *user*; and allow non-lookthrough
+      // users when the def is a PHI (loop-header pattern).
+      if (CII->getParent() == II->getParent() && !IsLookThru(CII) &&
+          !isa<PHINode>(II))
         continue;
 
       if (isOpLegal(CII))

``````````

</details>


https://github.com/llvm/llvm-project/pull/160909