[llvm] b3fe27f - [InstCombine] Copy flags of extractelement for extelt -> icmp combine (#86366)

[via llvm-commits]

Author: Marc Auberer
Date: 2024-03-24T16:14:56+01:00
New Revision: b3fe27f2be0585d5d2ad46f96956ccfd76ca003e

URL: https://github.com/llvm/llvm-project/commit/b3fe27f2be0585d5d2ad46f96956ccfd76ca003e
DIFF: https://github.com/llvm/llvm-project/commit/b3fe27f2be0585d5d2ad46f96956ccfd76ca003e.diff

LOG: [InstCombine] Copy flags of extractelement for extelt -> icmp combine (#86366)

Fixes #86164

Added: 
    

Modified: 
    llvm/include/llvm/IR/InstrTypes.h
    llvm/lib/IR/Instructions.cpp
    llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
    llvm/test/Transforms/InstCombine/scalarization.ll

Removed: 
    


################################################################################
diff  --git a/llvm/include/llvm/IR/InstrTypes.h b/llvm/include/llvm/IR/InstrTypes.h
index e8c2cba8418dc8..d2b33fe9c651a1 100644
--- a/llvm/include/llvm/IR/InstrTypes.h
+++ b/llvm/include/llvm/IR/InstrTypes.h
@@ -1058,6 +1058,17 @@ class CmpInst : public Instruction {
   static CmpInst *Create(OtherOps Op, Predicate predicate, Value *S1,
                          Value *S2, const Twine &Name, BasicBlock *InsertAtEnd);
 
+  /// Construct a compare instruction, given the opcode, the predicate,
+  /// the two operands and the instruction to copy the flags from. Optionally
+  /// (if InstBefore is specified) insert the instruction into a BasicBlock
+  /// right before the specified instruction. The specified Instruction is
+  /// allowed to be a dereferenced end iterator. Create a CmpInst
+  static CmpInst *CreateWithCopiedFlags(OtherOps Op, Predicate Pred, Value *S1,
+                                        Value *S2,
+                                        const Instruction *FlagsSource,
+                                        const Twine &Name = "",
+                                        Instruction *InsertBefore = nullptr);
+
   /// Get the opcode casted to the right type
   OtherOps getOpcode() const {
     return static_cast<OtherOps>(Instruction::getOpcode());

diff  --git a/llvm/lib/IR/Instructions.cpp b/llvm/lib/IR/Instructions.cpp
index 494d50f89e374c..8ca211e6e2e72f 100644
--- a/llvm/lib/IR/Instructions.cpp
+++ b/llvm/lib/IR/Instructions.cpp
@@ -4623,6 +4623,16 @@ CmpInst::Create(OtherOps Op, Predicate predicate, Value *S1, Value *S2,
                       S1, S2, Name);
 }
 
+CmpInst *CmpInst::CreateWithCopiedFlags(OtherOps Op, Predicate Pred, Value *S1,
+                                        Value *S2,
+                                        const Instruction *FlagsSource,
+                                        const Twine &Name,
+                                        Instruction *InsertBefore) {
+  CmpInst *Inst = Create(Op, Pred, S1, S2, Name, InsertBefore);
+  Inst->copyIRFlags(FlagsSource);
+  return Inst;
+}
+
 void CmpInst::swapOperands() {
   if (ICmpInst *IC = dyn_cast<ICmpInst>(this))
     IC->swapOperands();

diff  --git a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
index c7f4fb17648c87..99f1f8eb34bb5a 100644
--- a/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
+++ b/llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
@@ -487,7 +487,9 @@ Instruction *InstCombinerImpl::visitExtractElementInst(ExtractElementInst &EI) {
     // extelt (cmp X, Y), Index --> cmp (extelt X, Index), (extelt Y, Index)
     Value *E0 = Builder.CreateExtractElement(X, Index);
     Value *E1 = Builder.CreateExtractElement(Y, Index);
-    return CmpInst::Create(cast<CmpInst>(SrcVec)->getOpcode(), Pred, E0, E1);
+    CmpInst *SrcCmpInst = cast<CmpInst>(SrcVec);
+    return CmpInst::CreateWithCopiedFlags(SrcCmpInst->getOpcode(), Pred, E0, E1,
+                                          SrcCmpInst);
   }
 
   if (auto *I = dyn_cast<Instruction>(SrcVec)) {

diff  --git a/llvm/test/Transforms/InstCombine/scalarization.ll b/llvm/test/Transforms/InstCombine/scalarization.ll
index fe6dc526bd50ee..7e645ef7e883e4 100644
--- a/llvm/test/Transforms/InstCombine/scalarization.ll
+++ b/llvm/test/Transforms/InstCombine/scalarization.ll
@@ -341,6 +341,17 @@ define i1 @extractelt_vector_fcmp_constrhs_dynidx(<2 x float> %arg, i32 %idx) {
   ret i1 %ext
 }
 
+define i1 @extractelt_vector_fcmp_copy_flags(<4 x float> %x) {
+; CHECK-LABEL: @extractelt_vector_fcmp_copy_flags(
+; CHECK-NEXT:    [[TMP1:%.*]] = extractelement <4 x float> [[X:%.*]], i64 2
+; CHECK-NEXT:    [[TMP2:%.*]] = fcmp nsz arcp oeq float [[TMP1]], 0.000000e+00
+; CHECK-NEXT:    ret i1 [[TMP2]]
+;
+  %cmp = fcmp nsz arcp oeq <4 x float> %x, zeroinitializer
+  %r = extractelement <4 x i1> %cmp, i32 2
+  ret i1 %r
+}
+
 define i1 @extractelt_vector_fcmp_not_cheap_to_scalarize_multi_use(<2 x float> %arg0, <2 x float> %arg1, <2 x float> %arg2, i32 %idx) {
 ;
 ; CHECK-LABEL: @extractelt_vector_fcmp_not_cheap_to_scalarize_multi_use(