[LoopVectorizer] Propagate underlying instruction to the cloned instances of VPPartialReductionRecipes

[NickGuy-Arm]

No description provided.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

@llvm/pr-subscribers-vectorizers

Author: Nicholas Guy (NickGuy-Arm)

Changes

---

Full diff: https://github.com/llvm/llvm-project/pull/123638.diff

1 Files Affected:

• (modified) llvm/lib/Transforms/Vectorize/VPlan.h (+2-1)

diff --git a/llvm/lib/Transforms/Vectorize/VPlan.h b/llvm/lib/Transforms/Vectorize/VPlan.h
index 87f87bf1437196..9d8136a0ebf57d 100644
--- a/llvm/lib/Transforms/Vectorize/VPlan.h
+++ b/llvm/lib/Transforms/Vectorize/VPlan.h
@@ -2459,7 +2459,8 @@ class VPPartialReductionRecipe : public VPSingleDefRecipe {
   ~VPPartialReductionRecipe() override = default;
 
   VPPartialReductionRecipe *clone() override {
-    return new VPPartialReductionRecipe(Opcode, getOperand(0), getOperand(1));
+    return new VPPartialReductionRecipe(Opcode, getOperand(0), getOperand(1),
+       getUnderlyingInstr());
   }
 
   VP_CLASSOF_IMPL(VPDef::VPPartialReductionSC)

[github-actions]

✅ With the latest revision this PR passed the C/C++ code formatter.

[fhahn]

IIUC this should only change VPlan printing or is the underlying instr used by other parts of the recipe? If it's the former, I guess to could be tested by check the printed VPlan during epilogue vectorization?

[fhahn]

please use patterns to match the vp<> value numbers.

[NickGuy-Arm]

Done

[fhahn]

Thanks for adding the test. Do we need a separate test file or could we just enable epilogue vectorization in the existing llvm/test/Transforms/LoopVectorize/AArch64/vplan-printing.ll?

[NickGuy-Arm]

Done, I was a bit skeptical at first about merging them, but they went together with no issues.

[fhahn]

LGTM thanks