[VectorCombine] Remove requirement for Instructions in shuffleToIdentity

[davemgreen]

This removes the check that both operands of the original shuffle are instructions, which is a relic from a previous version that held more variables as Instructions.

[llvmbot]

@llvm/pr-subscribers-llvm-transforms

Author: David Green (davemgreen)

Changes

This removes the check that both operands of the original shuffle are instructions, which is a relic from a previous version that held more variables as Instructions.

---

Full diff: https://github.com/llvm/llvm-project/pull/93543.diff

2 Files Affected:

• (modified) llvm/lib/Transforms/Vectorize/VectorCombine.cpp (+1-2)
• (modified) llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll (+2-5)

diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index 056f0d6b3ee6c..ffed73cdb18c0 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -1757,8 +1757,7 @@ static Value *generateNewInstTree(ArrayRef<InstLane> Item, FixedVectorType *Ty,
 // do so.
 bool VectorCombine::foldShuffleToIdentity(Instruction &I) {
   auto *Ty = dyn_cast<FixedVectorType>(I.getType());
-  if (!Ty || !isa<Instruction>(I.getOperand(0)) ||
-      !isa<Instruction>(I.getOperand(1)))
+  if (!Ty)
     return false;
 
   SmallVector<InstLane> Start(Ty->getNumElements());
diff --git a/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll b/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
index 5cbda8a1e112e..71c49636a1735 100644
--- a/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
+++ b/llvm/test/Transforms/VectorCombine/AArch64/shuffletoidentity.ll
@@ -15,9 +15,7 @@ define <8 x i8> @trivial(<8 x i8> %a) {
 
 define <4 x i32> @add_same_operands(<4 x i32> %x) {
 ; CHECK-LABEL: @add_same_operands(
-; CHECK-NEXT:    [[SHUF:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[SHUF]], [[SHUF]]
-; CHECK-NEXT:    [[REVSHUF:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[REVSHUF:%.*]] = add <4 x i32> [[X:%.*]], [[X]]
 ; CHECK-NEXT:    ret <4 x i32> [[REVSHUF]]
 ;
   %shuf = shufflevector <4 x i32> %x, <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
@@ -364,8 +362,7 @@ define <8 x i8> @inner_shuffle(<8 x i8> %a, <8 x i8> %b, <8 x i8> %c) {
 define <4 x i32> @extrause_add_same_operands(<4 x i32> %x) {
 ; CHECK-LABEL: @extrause_add_same_operands(
 ; CHECK-NEXT:    [[SHUF:%.*]] = shufflevector <4 x i32> [[X:%.*]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
-; CHECK-NEXT:    [[ADD:%.*]] = add <4 x i32> [[SHUF]], [[SHUF]]
-; CHECK-NEXT:    [[REVSHUF:%.*]] = shufflevector <4 x i32> [[ADD]], <4 x i32> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>
+; CHECK-NEXT:    [[REVSHUF:%.*]] = add <4 x i32> [[X]], [[X]]
 ; CHECK-NEXT:    [[ADD2:%.*]] = add <4 x i32> [[SHUF]], [[REVSHUF]]
 ; CHECK-NEXT:    ret <4 x i32> [[ADD2]]
 ;

[RKSimon]

LGTM

[ronlieb]

@davemgreen we noticed a regression in two of our hip tests that seemed to be caused by this patch.
dont have a reproducer as of now. reporting it now. we will work on a test case , might take a day or two on our end.

[davemgreen]

Hi - yeah I can take a look if you have a reproducer. The regression is in the performance, not correctness? There is a quick cost-model added in #93937, in case that helps your problem.