Skip to content

[Clang][CodeGen] Do not promote if complex divisor is real #131451

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 1 commit into from
Mar 19, 2025
Merged

[Clang][CodeGen] Do not promote if complex divisor is real #131451

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
36 changes: 20 additions & 16 deletions clang/lib/CodeGen/CGExprComplex.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -286,8 +286,7 @@ class ComplexExprEmitter
ComplexPairTy EmitComplexBinOpLibCall(StringRef LibCallName,
const BinOpInfo &Op);

QualType HigherPrecisionTypeForComplexArithmetic(QualType ElementType,
bool IsDivOpCode) {
QualType HigherPrecisionTypeForComplexArithmetic(QualType ElementType) {
ASTContext &Ctx = CGF.getContext();
const QualType HigherElementType =
Ctx.GetHigherPrecisionFPType(ElementType);
Expand All @@ -314,7 +313,7 @@ class ComplexExprEmitter
}

QualType getPromotionType(FPOptionsOverride Features, QualType Ty,
bool IsDivOpCode = false) {
bool IsComplexDivisor) {
if (auto *CT = Ty->getAs<ComplexType>()) {
QualType ElementType = CT->getElementType();
bool IsFloatingType = ElementType->isFloatingType();
Expand All @@ -325,10 +324,9 @@ class ComplexExprEmitter
Features.getComplexRangeOverride() ==
CGF.getLangOpts().getComplexRange();

if (IsDivOpCode && IsFloatingType && IsComplexRangePromoted &&
if (IsComplexDivisor && IsFloatingType && IsComplexRangePromoted &&
(HasNoComplexRangeOverride || HasMatchingComplexRange))
return HigherPrecisionTypeForComplexArithmetic(ElementType,
IsDivOpCode);
return HigherPrecisionTypeForComplexArithmetic(ElementType);
if (ElementType.UseExcessPrecision(CGF.getContext()))
return CGF.getContext().getComplexType(CGF.getContext().FloatTy);
}
Expand All @@ -339,9 +337,10 @@ class ComplexExprEmitter

#define HANDLEBINOP(OP) \
ComplexPairTy VisitBin##OP(const BinaryOperator *E) { \
QualType promotionTy = getPromotionType( \
E->getStoredFPFeaturesOrDefault(), E->getType(), \
(E->getOpcode() == BinaryOperatorKind::BO_Div) ? true : false); \
QualType promotionTy = \
getPromotionType(E->getStoredFPFeaturesOrDefault(), E->getType(), \
(E->getOpcode() == BinaryOperatorKind::BO_Div && \
E->getRHS()->getType()->isAnyComplexType())); \
ComplexPairTy result = EmitBin##OP(EmitBinOps(E, promotionTy)); \
if (!promotionTy.isNull()) \
result = CGF.EmitUnPromotedValue(result, E->getType()); \
Expand Down Expand Up @@ -639,7 +638,8 @@ ComplexPairTy ComplexExprEmitter::VisitUnaryPlus(const UnaryOperator *E,
QualType promotionTy =
PromotionType.isNull()
? getPromotionType(E->getStoredFPFeaturesOrDefault(),
E->getSubExpr()->getType())
E->getSubExpr()->getType(),
/*IsComplexDivisor=*/false)
: PromotionType;
ComplexPairTy result = VisitPlus(E, promotionTy);
if (!promotionTy.isNull())
Expand All @@ -661,7 +661,8 @@ ComplexPairTy ComplexExprEmitter::VisitUnaryMinus(const UnaryOperator *E,
QualType promotionTy =
PromotionType.isNull()
? getPromotionType(E->getStoredFPFeaturesOrDefault(),
E->getSubExpr()->getType())
E->getSubExpr()->getType(),
/*IsComplexDivisor=*/false)
: PromotionType;
ComplexPairTy result = VisitMinus(E, promotionTy);
if (!promotionTy.isNull())
Expand Down Expand Up @@ -1218,14 +1219,16 @@ EmitCompoundAssignLValue(const CompoundAssignOperator *E,
// improve codegen a little.
QualType PromotionTypeCR;
PromotionTypeCR = getPromotionType(E->getStoredFPFeaturesOrDefault(),
E->getComputationResultType());
E->getComputationResultType(),
/*IsComplexDivisor=*/false);
if (PromotionTypeCR.isNull())
PromotionTypeCR = E->getComputationResultType();
OpInfo.Ty = PromotionTypeCR;
QualType ComplexElementTy =
OpInfo.Ty->castAs<ComplexType>()->getElementType();
QualType PromotionTypeRHS = getPromotionType(
E->getStoredFPFeaturesOrDefault(), E->getRHS()->getType());
QualType PromotionTypeRHS =
getPromotionType(E->getStoredFPFeaturesOrDefault(),
E->getRHS()->getType(), /*IsComplexDivisor=*/false);

// The RHS should have been converted to the computation type.
if (E->getRHS()->getType()->isRealFloatingType()) {
Expand Down Expand Up @@ -1253,8 +1256,9 @@ EmitCompoundAssignLValue(const CompoundAssignOperator *E,

// Load from the l-value and convert it.
SourceLocation Loc = E->getExprLoc();
QualType PromotionTypeLHS = getPromotionType(
E->getStoredFPFeaturesOrDefault(), E->getComputationLHSType());
QualType PromotionTypeLHS =
getPromotionType(E->getStoredFPFeaturesOrDefault(),
E->getComputationLHSType(), /*IsComplexDivisor=*/false);
if (LHSTy->isAnyComplexType()) {
ComplexPairTy LHSVal = EmitLoadOfLValue(LHS, Loc);
if (!PromotionTypeLHS.isNull())
Expand Down
52 changes: 16 additions & 36 deletions clang/test/CodeGen/cx-complex-range-real.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -340,18 +340,13 @@ _Complex float mulbf(float a, _Complex float b) {
// PRMTD-NEXT: [[A_REAL:%.*]] = load float, ptr [[A_REALP]], align 4
// PRMTD-NEXT: [[A_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[A]], i32 0, i32 1
// PRMTD-NEXT: [[A_IMAG:%.*]] = load float, ptr [[A_IMAGP]], align 4
// PRMTD-NEXT: [[EXT:%.*]] = fpext float [[A_REAL]] to double
// PRMTD-NEXT: [[EXT1:%.*]] = fpext float [[A_IMAG]] to double
// PRMTD-NEXT: [[TMP0:%.*]] = load float, ptr [[B_ADDR]], align 4
// PRMTD-NEXT: [[EXT2:%.*]] = fpext float [[TMP0]] to double
// PRMTD-NEXT: [[TMP1:%.*]] = fdiv double [[EXT]], [[EXT2]]
// PRMTD-NEXT: [[TMP2:%.*]] = fdiv double [[EXT1]], [[EXT2]]
// PRMTD-NEXT: [[UNPROMOTION:%.*]] = fptrunc double [[TMP1]] to float
// PRMTD-NEXT: [[UNPROMOTION3:%.*]] = fptrunc double [[TMP2]] to float
// PRMTD-NEXT: [[TMP1:%.*]] = fdiv float [[A_REAL]], [[TMP0]]
// PRMTD-NEXT: [[TMP2:%.*]] = fdiv float [[A_IMAG]], [[TMP0]]
// PRMTD-NEXT: [[RETVAL_REALP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[RETVAL]], i32 0, i32 0
// PRMTD-NEXT: [[RETVAL_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[RETVAL]], i32 0, i32 1
// PRMTD-NEXT: store float [[UNPROMOTION]], ptr [[RETVAL_REALP]], align 4
// PRMTD-NEXT: store float [[UNPROMOTION3]], ptr [[RETVAL_IMAGP]], align 4
// PRMTD-NEXT: store float [[TMP1]], ptr [[RETVAL_REALP]], align 4
// PRMTD-NEXT: store float [[TMP2]], ptr [[RETVAL_IMAGP]], align 4
// PRMTD-NEXT: [[TMP3:%.*]] = load <2 x float>, ptr [[RETVAL]], align 4
// PRMTD-NEXT: ret <2 x float> [[TMP3]]
//
Expand All @@ -367,18 +362,13 @@ _Complex float mulbf(float a, _Complex float b) {
// PRMTD_STRICT-NEXT: [[A_REAL:%.*]] = load float, ptr [[A_REALP]], align 4
// PRMTD_STRICT-NEXT: [[A_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[A]], i32 0, i32 1
// PRMTD_STRICT-NEXT: [[A_IMAG:%.*]] = load float, ptr [[A_IMAGP]], align 4
// PRMTD_STRICT-NEXT: [[EXT:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[A_REAL]], metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[EXT1:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[A_IMAG]], metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP0:%.*]] = load float, ptr [[B_ADDR]], align 4
// PRMTD_STRICT-NEXT: [[EXT2:%.*]] = call double @llvm.experimental.constrained.fpext.f64.f32(float [[TMP0]], metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP1:%.*]] = call double @llvm.experimental.constrained.fdiv.f64(double [[EXT]], double [[EXT2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP2:%.*]] = call double @llvm.experimental.constrained.fdiv.f64(double [[EXT1]], double [[EXT2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[UNPROMOTION:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP1]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[UNPROMOTION3:%.*]] = call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP1:%.*]] = call float @llvm.experimental.constrained.fdiv.f32(float [[A_REAL]], float [[TMP0]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP2:%.*]] = call float @llvm.experimental.constrained.fdiv.f32(float [[A_IMAG]], float [[TMP0]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[RETVAL_REALP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[RETVAL]], i32 0, i32 0
// PRMTD_STRICT-NEXT: [[RETVAL_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[RETVAL]], i32 0, i32 1
// PRMTD_STRICT-NEXT: store float [[UNPROMOTION]], ptr [[RETVAL_REALP]], align 4
// PRMTD_STRICT-NEXT: store float [[UNPROMOTION3]], ptr [[RETVAL_IMAGP]], align 4
// PRMTD_STRICT-NEXT: store float [[TMP1]], ptr [[RETVAL_REALP]], align 4
// PRMTD_STRICT-NEXT: store float [[TMP2]], ptr [[RETVAL_IMAGP]], align 4
// PRMTD_STRICT-NEXT: [[TMP3:%.*]] = load <2 x float>, ptr [[RETVAL]], align 4
// PRMTD_STRICT-NEXT: ret <2 x float> [[TMP3]]
//
Expand Down Expand Up @@ -539,18 +529,13 @@ void divassignf(_Complex float *a, float b) {
// PRMTD-NEXT: [[A_REAL:%.*]] = load double, ptr [[A_REALP]], align 8
// PRMTD-NEXT: [[A_IMAGP:%.*]] = getelementptr inbounds nuw { double, double }, ptr [[A]], i32 0, i32 1
// PRMTD-NEXT: [[A_IMAG:%.*]] = load double, ptr [[A_IMAGP]], align 8
// PRMTD-NEXT: [[EXT:%.*]] = fpext double [[A_REAL]] to x86_fp80
// PRMTD-NEXT: [[EXT1:%.*]] = fpext double [[A_IMAG]] to x86_fp80
// PRMTD-NEXT: [[TMP2:%.*]] = load double, ptr [[B_ADDR]], align 8
// PRMTD-NEXT: [[EXT2:%.*]] = fpext double [[TMP2]] to x86_fp80
// PRMTD-NEXT: [[TMP3:%.*]] = fdiv x86_fp80 [[EXT]], [[EXT2]]
// PRMTD-NEXT: [[TMP4:%.*]] = fdiv x86_fp80 [[EXT1]], [[EXT2]]
// PRMTD-NEXT: [[UNPROMOTION:%.*]] = fptrunc x86_fp80 [[TMP3]] to double
// PRMTD-NEXT: [[UNPROMOTION3:%.*]] = fptrunc x86_fp80 [[TMP4]] to double
// PRMTD-NEXT: [[TMP3:%.*]] = fdiv double [[A_REAL]], [[TMP2]]
// PRMTD-NEXT: [[TMP4:%.*]] = fdiv double [[A_IMAG]], [[TMP2]]
// PRMTD-NEXT: [[RETVAL_REALP:%.*]] = getelementptr inbounds nuw { double, double }, ptr [[RETVAL]], i32 0, i32 0
// PRMTD-NEXT: [[RETVAL_IMAGP:%.*]] = getelementptr inbounds nuw { double, double }, ptr [[RETVAL]], i32 0, i32 1
// PRMTD-NEXT: store double [[UNPROMOTION]], ptr [[RETVAL_REALP]], align 8
// PRMTD-NEXT: store double [[UNPROMOTION3]], ptr [[RETVAL_IMAGP]], align 8
// PRMTD-NEXT: store double [[TMP3]], ptr [[RETVAL_REALP]], align 8
// PRMTD-NEXT: store double [[TMP4]], ptr [[RETVAL_IMAGP]], align 8
// PRMTD-NEXT: [[TMP5:%.*]] = load { double, double }, ptr [[RETVAL]], align 8
// PRMTD-NEXT: ret { double, double } [[TMP5]]
//
Expand All @@ -569,18 +554,13 @@ void divassignf(_Complex float *a, float b) {
// PRMTD_STRICT-NEXT: [[A_REAL:%.*]] = load double, ptr [[A_REALP]], align 8
// PRMTD_STRICT-NEXT: [[A_IMAGP:%.*]] = getelementptr inbounds nuw { double, double }, ptr [[A]], i32 0, i32 1
// PRMTD_STRICT-NEXT: [[A_IMAG:%.*]] = load double, ptr [[A_IMAGP]], align 8
// PRMTD_STRICT-NEXT: [[EXT:%.*]] = call x86_fp80 @llvm.experimental.constrained.fpext.f80.f64(double [[A_REAL]], metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[EXT1:%.*]] = call x86_fp80 @llvm.experimental.constrained.fpext.f80.f64(double [[A_IMAG]], metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP2:%.*]] = load double, ptr [[B_ADDR]], align 8
// PRMTD_STRICT-NEXT: [[EXT2:%.*]] = call x86_fp80 @llvm.experimental.constrained.fpext.f80.f64(double [[TMP2]], metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP3:%.*]] = call x86_fp80 @llvm.experimental.constrained.fdiv.f80(x86_fp80 [[EXT]], x86_fp80 [[EXT2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP4:%.*]] = call x86_fp80 @llvm.experimental.constrained.fdiv.f80(x86_fp80 [[EXT1]], x86_fp80 [[EXT2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[UNPROMOTION:%.*]] = call double @llvm.experimental.constrained.fptrunc.f64.f80(x86_fp80 [[TMP3]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[UNPROMOTION3:%.*]] = call double @llvm.experimental.constrained.fptrunc.f64.f80(x86_fp80 [[TMP4]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP3:%.*]] = call double @llvm.experimental.constrained.fdiv.f64(double [[A_REAL]], double [[TMP2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[TMP4:%.*]] = call double @llvm.experimental.constrained.fdiv.f64(double [[A_IMAG]], double [[TMP2]], metadata !"round.dynamic", metadata !"fpexcept.strict") #[[ATTR3]]
// PRMTD_STRICT-NEXT: [[RETVAL_REALP:%.*]] = getelementptr inbounds nuw { double, double }, ptr [[RETVAL]], i32 0, i32 0
// PRMTD_STRICT-NEXT: [[RETVAL_IMAGP:%.*]] = getelementptr inbounds nuw { double, double }, ptr [[RETVAL]], i32 0, i32 1
// PRMTD_STRICT-NEXT: store double [[UNPROMOTION]], ptr [[RETVAL_REALP]], align 8
// PRMTD_STRICT-NEXT: store double [[UNPROMOTION3]], ptr [[RETVAL_IMAGP]], align 8
// PRMTD_STRICT-NEXT: store double [[TMP3]], ptr [[RETVAL_REALP]], align 8
// PRMTD_STRICT-NEXT: store double [[TMP4]], ptr [[RETVAL_IMAGP]], align 8
// PRMTD_STRICT-NEXT: [[TMP5:%.*]] = load { double, double }, ptr [[RETVAL]], align 8
// PRMTD_STRICT-NEXT: ret { double, double } [[TMP5]]
//
Expand Down
Loading