Skip to content

[InstCombine] Fix constant swap case of fcmp + fadd + sel xfrm #119419

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 1 commit into from
Jan 2, 2025
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
44 changes: 26 additions & 18 deletions llvm/lib/Transforms/InstCombine/InstCombineSelect.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -3769,22 +3769,9 @@ static Value *foldSelectIntoAddConstant(SelectInst &SI,
if (!SIFOp || !SIFOp->hasNoSignedZeros() || !SIFOp->hasNoNaNs())
return nullptr;

// select((fcmp Pred, X, 0), (fadd X, C), C)
// => fadd((select (fcmp Pred, X, 0), X, 0), C)
//
// Pred := OGT, OGE, OLT, OLE, UGT, UGE, ULT, and ULE
Instruction *FAdd;
Constant *C;
Value *X, *Z;
CmpPredicate Pred;

// Note: OneUse check for `Cmp` is necessary because it makes sure that other
// InstCombine folds don't undo this transformation and cause an infinite
// loop. Furthermore, it could also increase the operation count.
if (match(&SI, m_Select(m_OneUse(m_FCmp(Pred, m_Value(X), m_Value(Z))),
m_OneUse(m_Instruction(FAdd)), m_Constant(C))) ||
match(&SI, m_Select(m_OneUse(m_FCmp(Pred, m_Value(X), m_Value(Z))),
m_Constant(C), m_OneUse(m_Instruction(FAdd))))) {
auto TryFoldIntoAddConstant =
[&Builder, &SI](CmpInst::Predicate Pred, Value *X, Value *Z,
Instruction *FAdd, Constant *C, bool Swapped) -> Value * {
// Only these relational predicates can be transformed into maxnum/minnum
// intrinsic.
if (!CmpInst::isRelational(Pred) || !match(Z, m_AnyZeroFP()))
Expand All @@ -3793,7 +3780,8 @@ static Value *foldSelectIntoAddConstant(SelectInst &SI,
if (!match(FAdd, m_FAdd(m_Specific(X), m_Specific(C))))
return nullptr;

Value *NewSelect = Builder.CreateSelect(SI.getCondition(), X, Z, "", &SI);
Value *NewSelect = Builder.CreateSelect(SI.getCondition(), Swapped ? Z : X,
Swapped ? X : Z, "", &SI);
NewSelect->takeName(&SI);

Value *NewFAdd = Builder.CreateFAdd(NewSelect, C);
Expand All @@ -3808,7 +3796,27 @@ static Value *foldSelectIntoAddConstant(SelectInst &SI,
cast<Instruction>(NewSelect)->setFastMathFlags(NewFMF);

return NewFAdd;
}
};

// select((fcmp Pred, X, 0), (fadd X, C), C)
// => fadd((select (fcmp Pred, X, 0), X, 0), C)
//
// Pred := OGT, OGE, OLT, OLE, UGT, UGE, ULT, and ULE
Instruction *FAdd;
Constant *C;
Value *X, *Z;
CmpPredicate Pred;

// Note: OneUse check for `Cmp` is necessary because it makes sure that other
// InstCombine folds don't undo this transformation and cause an infinite
// loop. Furthermore, it could also increase the operation count.
if (match(&SI, m_Select(m_OneUse(m_FCmp(Pred, m_Value(X), m_Value(Z))),
m_OneUse(m_Instruction(FAdd)), m_Constant(C))))
return TryFoldIntoAddConstant(Pred, X, Z, FAdd, C, /*Swapped=*/false);

if (match(&SI, m_Select(m_OneUse(m_FCmp(Pred, m_Value(X), m_Value(Z))),
m_Constant(C), m_OneUse(m_Instruction(FAdd)))))
return TryFoldIntoAddConstant(Pred, X, Z, FAdd, C, /*Swapped=*/true);

return nullptr;
}
Expand Down
16 changes: 8 additions & 8 deletions llvm/test/Transforms/InstCombine/fcmp-fadd-select.ll
Original file line number Diff line number Diff line change
Expand Up @@ -19,7 +19,7 @@ define float @test_fcmp_ogt_fadd_select_constant(float %in) {
define float @test_fcmp_ogt_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_ogt_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.maxnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.minnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down Expand Up @@ -87,7 +87,7 @@ define float @test_fcmp_olt_fadd_select_constant(float %in) {
define float @test_fcmp_olt_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_olt_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.minnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.maxnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down Expand Up @@ -155,7 +155,7 @@ define float @test_fcmp_oge_fadd_select_constant(float %in) {
define float @test_fcmp_oge_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_oge_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.maxnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.minnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down Expand Up @@ -223,7 +223,7 @@ define float @test_fcmp_ole_fadd_select_constant(float %in) {
define float @test_fcmp_ole_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_ole_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.minnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[SEL_NEW:%.*]] = call nsz float @llvm.maxnum.f32(float [[IN]], float 0.000000e+00)
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down Expand Up @@ -293,7 +293,7 @@ define float @test_fcmp_ugt_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_ugt_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[CMP1_INV:%.*]] = fcmp ole float [[IN]], 0.000000e+00
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float 0.000000e+00, float [[IN]]
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float [[IN]], float 0.000000e+00
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down Expand Up @@ -366,7 +366,7 @@ define float @test_fcmp_uge_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_uge_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[CMP1_INV:%.*]] = fcmp olt float [[IN]], 0.000000e+00
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float 0.000000e+00, float [[IN]]
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float [[IN]], float 0.000000e+00
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down Expand Up @@ -439,7 +439,7 @@ define float @test_fcmp_ult_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_ult_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[CMP1_INV:%.*]] = fcmp oge float [[IN]], 0.000000e+00
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float 0.000000e+00, float [[IN]]
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float [[IN]], float 0.000000e+00
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down Expand Up @@ -512,7 +512,7 @@ define float @test_fcmp_ule_fadd_select_constant_swapped(float %in) {
; CHECK-LABEL: define float @test_fcmp_ule_fadd_select_constant_swapped(
; CHECK-SAME: float [[IN:%.*]]) {
; CHECK-NEXT: [[CMP1_INV:%.*]] = fcmp ogt float [[IN]], 0.000000e+00
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float 0.000000e+00, float [[IN]]
; CHECK-NEXT: [[SEL_NEW:%.*]] = select i1 [[CMP1_INV]], float [[IN]], float 0.000000e+00
; CHECK-NEXT: [[ADD_NEW:%.*]] = fadd nnan nsz float [[SEL_NEW]], 1.000000e+00
; CHECK-NEXT: ret float [[ADD_NEW]]
;
Expand Down
Loading