[llvm] VectorCombine: fix logical error after m_Trunc match (PR #91201)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-llvm-transforms

Author: Ramkumar Ramachandra (artagnon)

<details>
<summary>Changes</summary>

The matcher m_Trunc() matches an Operator with a given Opcode, which could either be an Instruction or ConstExpr. VectorCombine::foldTruncFromReductions() incorrectly assumes that the pattern matched is always an Instruction, and attempts a cast. Fix this.

Fixes #<!-- -->88796.

-- 8< --
Based on #<!-- -->91200.

---
Full diff: https://github.com/llvm/llvm-project/pull/91201.diff


2 Files Affected:

- (modified) llvm/lib/Transforms/Vectorize/VectorCombine.cpp (+1-2) 
- (added) llvm/test/Transforms/VectorCombine/pr88796.ll (+16) 


``````````diff
diff --git a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
index bbb70134870ab6..50a8209b465168 100644
--- a/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
+++ b/llvm/lib/Transforms/Vectorize/VectorCombine.cpp
@@ -1961,7 +1961,6 @@ bool VectorCombine::foldTruncFromReductions(Instruction &I) {
   if (!match(ReductionSrc, m_OneUse(m_Trunc(m_Value(TruncSrc)))))
     return false;
 
-  auto *Trunc = cast<CastInst>(ReductionSrc);
   auto *TruncSrcTy = cast<VectorType>(TruncSrc->getType());
   auto *ReductionSrcTy = cast<VectorType>(ReductionSrc->getType());
   Type *ResultTy = I.getType();
@@ -1969,7 +1968,7 @@ bool VectorCombine::foldTruncFromReductions(Instruction &I) {
   TTI::TargetCostKind CostKind = TTI::TCK_RecipThroughput;
   InstructionCost OldCost =
       TTI.getCastInstrCost(Instruction::Trunc, ReductionSrcTy, TruncSrcTy,
-                           TTI::CastContextHint::None, CostKind, Trunc) +
+                           TTI::CastContextHint::None, CostKind) +
       TTI.getArithmeticReductionCost(ReductionOpc, ReductionSrcTy, std::nullopt,
                                      CostKind);
   InstructionCost NewCost =
diff --git a/llvm/test/Transforms/VectorCombine/pr88796.ll b/llvm/test/Transforms/VectorCombine/pr88796.ll
new file mode 100644
index 00000000000000..4f26f5dcbb928d
--- /dev/null
+++ b/llvm/test/Transforms/VectorCombine/pr88796.ll
@@ -0,0 +1,16 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -passes=vector-combine -S %s | FileCheck %s
+
+define i32 @test() {
+; CHECK-LABEL: define i32 @test() {
+; CHECK-NEXT:  entry:
+; CHECK-NEXT:    [[TMP0:%.*]] = tail call i16 @llvm.vector.reduce.and.nxv8i16(<vscale x 8 x i16> trunc (<vscale x 8 x i32> shufflevector (<vscale x 8 x i32> insertelement (<vscale x 8 x i32> poison, i32 268435456, i64 0), <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer) to <vscale x 8 x i16>))
+; CHECK-NEXT:    ret i32 0
+;
+entry:
+  %0 = tail call i16 @llvm.vector.reduce.and.nxv8i16(<vscale x 8 x i16> trunc (<vscale x 8 x i32> shufflevector (<vscale x 8 x i32> insertelement (<vscale x 8 x i32> poison, i32 268435456, i64 0), <vscale x 8 x i32> poison, <vscale x 8 x i32> zeroinitializer) to <vscale x 8 x i16>))
+  ret i32 0
+}
+
+declare i16 @llvm.vector.reduce.and.nxv8i16(<vscale x 8 x i16>)
+

``````````

</details>


https://github.com/llvm/llvm-project/pull/91201