Skip to content

[InstCombine] Preserve the sign bit of NaN in SimplifyDemandedUseFPClass #137287

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.

Sign up for GitHub

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 3 commits into from
Apr 28, 2025
Merged

[InstCombine] Preserve the sign bit of NaN in SimplifyDemandedUseFPClass #137287

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e94b6c3
[InstCombine] Preserve the sign bit of NaN in `SimplifyDemandedUseFPC…
dtcxzyw Apr 25, 2025
6c3097d
[InstCombine] Handle FMF in `SimplifyDemandedUseFPClass`
dtcxzyw Apr 25, 2025
81c9a92
Add more tests. NFC.
dtcxzyw Apr 25, 2025
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
19 changes: 13 additions & 6 deletions llvm/lib/Transforms/InstCombine/InstCombineSimplifyDemanded.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1959,9 +1959,11 @@ static Constant *getFPClassConstant(Type *Ty, FPClassTest Mask) {
}
}

Value *InstCombinerImpl::SimplifyDemandedUseFPClass(
Value *V, const FPClassTest DemandedMask, KnownFPClass &Known,
unsigned Depth, Instruction *CxtI) {
Value *InstCombinerImpl::SimplifyDemandedUseFPClass(Value *V,
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not sure why this changed, the signature didn't?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is needed by DemandedMask &= ~fcNan;.

FPClassTest DemandedMask,
KnownFPClass &Known,
unsigned Depth,
Instruction *CxtI) {
assert(Depth <= MaxAnalysisRecursionDepth && "Limit Search Depth");
Type *VTy = V->getType();

Expand All @@ -1985,7 +1987,12 @@ Value *InstCombinerImpl::SimplifyDemandedUseFPClass(
if (!I->hasOneUse())
return nullptr;

// TODO: Should account for nofpclass/FastMathFlags on current instruction
if (auto *FPOp = dyn_cast<FPMathOperator>(I)) {
if (FPOp->hasNoNaNs())
DemandedMask &= ~fcNan;
if (FPOp->hasNoInfs())
DemandedMask &= ~fcInf;
}
switch (I->getOpcode()) {
case Instruction::FNeg: {
if (SimplifyDemandedFPClass(I, 0, llvm::fneg(DemandedMask), Known,
Expand Down Expand Up @@ -2013,13 +2020,13 @@ Value *InstCombinerImpl::SimplifyDemandedUseFPClass(
if (SimplifyDemandedFPClass(I, 0, DemandedMaskAnySign, Known, Depth + 1))
return I;

if ((DemandedMask & fcPositive) == fcNone) {
if ((DemandedMask & fcNegative) == DemandedMask) {
// Roundabout way of replacing with fneg(fabs)
I->setOperand(1, ConstantFP::get(VTy, -1.0));
return I;
}

if ((DemandedMask & fcNegative) == fcNone) {
if ((DemandedMask & fcPositive) == DemandedMask) {
// Roundabout way of replacing with fabs
I->setOperand(1, ConstantFP::getZero(VTy));
return I;
Expand Down
35 changes: 22 additions & 13 deletions llvm/test/Transforms/InstCombine/simplify-demanded-fpclass.ll
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -678,12 +678,11 @@ define nofpclass(inf) float @ret_nofpclass_inf__copysign_negative_select_pinf_rh
ret float %copysign
}

; can fold to fneg(fabs(x))
; can't fold to fneg(fabs(x))
define nofpclass(pinf pnorm psub pzero) float @ret_nofpclass_nopositives_copysign(float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(pinf pzero psub pnorm) float @ret_nofpclass_nopositives_copysign
; CHECK-SAME: (float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[COPYSIGN:%.*]] = fneg float [[TMP1]]
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.copysign.f32(float [[X]], float [[UNKNOWN_SIGN]])
; CHECK-NEXT: ret float [[COPYSIGN]]
;
%copysign = call float @llvm.copysign.f32(float %x, float %unknown.sign)
Expand All @@ -703,6 +702,18 @@ define nofpclass(pinf pnorm psub pzero) float @ret_nofpclass_nopositives_copysig
}

; can fold to fneg(fabs(x))
define nofpclass(pnorm psub pzero) float @ret_nofpclass_nopositive_finites_copysign_nnan_ninf_flag(float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(pzero psub pnorm) float @ret_nofpclass_nopositive_finites_copysign_nnan_ninf_flag
; CHECK-SAME: (float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call nnan ninf float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[COPYSIGN:%.*]] = fneg nnan ninf float [[TMP1]]
; CHECK-NEXT: ret float [[COPYSIGN]]
;
%copysign = call nnan ninf float @llvm.copysign.f32(float %x, float %unknown.sign)
ret float %copysign
}

; can't fold to fneg(fabs(x))
define nofpclass(nan pinf pnorm psub pzero) float @ret_nofpclass_nopositives_nonan_copysign(float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(nan pinf pzero psub pnorm) float @ret_nofpclass_nopositives_nonan_copysign
; CHECK-SAME: (float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
Expand All @@ -714,11 +725,11 @@ define nofpclass(nan pinf pnorm psub pzero) float @ret_nofpclass_nopositives_non
ret float %copysign
}

; can fold to fabs(x)
; can't fold to fabs(x)
define nofpclass(ninf nnorm nsub nzero) float @ret_nofpclass_nonegatives_copysign(float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_nofpclass_nonegatives_copysign
; CHECK-SAME: (float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.copysign.f32(float [[X]], float [[UNKNOWN_SIGN]])
; CHECK-NEXT: ret float [[COPYSIGN]]
;
%copysign = call float @llvm.copysign.f32(float %x, float %unknown.sign)
Expand Down Expand Up @@ -763,7 +774,7 @@ define nofpclass(pinf pnorm psub pzero) float @ret_nofpclass_nopositives__copysi
define nofpclass(inf nnorm nsub nzero) float @ret_nofpclass_no_negatives_noinf__copysign_unknown_select_pinf_rhs(i1 %cond, float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(inf nzero nsub nnorm) float @ret_nofpclass_no_negatives_noinf__copysign_unknown_select_pinf_rhs
; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.copysign.f32(float [[X]], float [[UNKNOWN_SIGN]])
; CHECK-NEXT: ret float [[COPYSIGN]]
;
%select = select i1 %cond, float %x, float 0x7FF0000000000000
Expand All @@ -775,8 +786,7 @@ define nofpclass(inf nnorm nsub nzero) float @ret_nofpclass_no_negatives_noinf__
define nofpclass(inf pnorm psub pzero) float @ret_nofpclass_no_positives_noinf__copysign_unknown_select_pinf_rhs(i1 %cond, float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(inf pzero psub pnorm) float @ret_nofpclass_no_positives_noinf__copysign_unknown_select_pinf_rhs
; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[COPYSIGN:%.*]] = fneg float [[TMP1]]
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.copysign.f32(float [[X]], float [[UNKNOWN_SIGN]])
; CHECK-NEXT: ret float [[COPYSIGN]]
;
%select = select i1 %cond, float %x, float 0x7FF0000000000000
Expand All @@ -788,8 +798,8 @@ define nofpclass(inf pnorm psub pzero) float @ret_nofpclass_no_positives_noinf__
define nofpclass(ninf nnorm nsub nzero) float @ret_nofpclass_no_negatives__copysign_unknown_select_pinf_rhs(i1 %cond, float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(ninf nzero nsub nnorm) float @ret_nofpclass_no_negatives__copysign_unknown_select_pinf_rhs
; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[COPYSIGN:%.*]] = select i1 [[COND]], float [[TMP1]], float 0x7FF0000000000000
; CHECK-NEXT: [[SELECT:%.*]] = select i1 [[COND]], float [[X]], float 0x7FF0000000000000
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.copysign.f32(float [[SELECT]], float [[UNKNOWN_SIGN]])
; CHECK-NEXT: ret float [[COPYSIGN]]
;
%select = select i1 %cond, float %x, float 0x7FF0000000000000
Expand All @@ -801,9 +811,8 @@ define nofpclass(ninf nnorm nsub nzero) float @ret_nofpclass_no_negatives__copys
define nofpclass(pinf pnorm psub pzero) float @ret_nofpclass_no_positives__copysign_unknown_select_pinf_rhs(i1 %cond, float %x, float %unknown.sign) {
; CHECK-LABEL: define nofpclass(pinf pzero psub pnorm) float @ret_nofpclass_no_positives__copysign_unknown_select_pinf_rhs
; CHECK-SAME: (i1 [[COND:%.*]], float [[X:%.*]], float [[UNKNOWN_SIGN:%.*]]) {
; CHECK-NEXT: [[TMP1:%.*]] = call float @llvm.fabs.f32(float [[X]])
; CHECK-NEXT: [[DOTNEG:%.*]] = fneg float [[TMP1]]
; CHECK-NEXT: [[COPYSIGN:%.*]] = select i1 [[COND]], float [[DOTNEG]], float 0xFFF0000000000000
; CHECK-NEXT: [[SELECT:%.*]] = select i1 [[COND]], float [[X]], float 0x7FF0000000000000
; CHECK-NEXT: [[COPYSIGN:%.*]] = call float @llvm.copysign.f32(float [[SELECT]], float [[UNKNOWN_SIGN]])
; CHECK-NEXT: ret float [[COPYSIGN]]
;
%select = select i1 %cond, float %x, float 0x7FF0000000000000
Expand Down
Loading