shylie/llvm-project
1
0
Fork 0
Code Issues Pull Requests Actions 6 Packages Projects Releases Wiki Activity

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

Browse Source 
Fixes #86164
This commit is contained in:
Marc Auberer 2024-03-24 16:14:56 +01:00 committed by GitHub
parent e8d5223ce4
commit b3fe27f2be
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
4 changed files with 35 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

11
llvm/include/llvm/IR/InstrTypes.h
View File

@ -1058,6 +1058,17 @@ public:
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());

10
llvm/lib/IR/Instructions.cpp
View File

@ -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();

4
llvm/lib/Transforms/InstCombine/InstCombineVectorOps.cpp
View File

@ -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)) {

11
llvm/test/Transforms/InstCombine/scalarization.ll
View File

@ -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(