[Clang][CodeGen] Promote in complex compound divassign #131453
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This stack of pull requests is managed by Graphite. Learn more about stacking. |
Maetveis
requested review from
zahiraam,
rjmccall,
AaronBallman,
jcranmer-intel and
efriedma-quic
llvmbot
added
clang
|
@llvm/pr-subscribers-clang-codegen @llvm/pr-subscribers-clang Author: Mészáros Gergely (Maetveis) ChangesWhen Fixes: #131129 Patch is 78.24 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/131453.diff 2 Files Affected:
diff --git a/clang/lib/CodeGen/CGExprComplex.cpp b/clang/lib/CodeGen/CGExprComplex.cpp
index a8a65a2f956f8..dc1a34ee82805 100644
--- a/clang/lib/CodeGen/CGExprComplex.cpp
+++ b/clang/lib/CodeGen/CGExprComplex.cpp
@@ -1212,19 +1212,24 @@ EmitCompoundAssignLValue(const CompoundAssignOperator *E,
OpInfo.FPFeatures = E->getFPFeaturesInEffect(CGF.getLangOpts());
CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, OpInfo.FPFeatures);
+ const bool IsComplexDivisor = E->getOpcode() == BO_DivAssign &&
+ E->getRHS()->getType()->isAnyComplexType();
+
// Load the RHS and LHS operands.
// __block variables need to have the rhs evaluated first, plus this should
// improve codegen a little.
QualType PromotionTypeCR;
- PromotionTypeCR = getPromotionType(E->getStoredFPFeaturesOrDefault(),
- E->getComputationResultType());
+ PromotionTypeCR =
+ getPromotionType(E->getStoredFPFeaturesOrDefault(),
+ E->getComputationResultType(), IsComplexDivisor);
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);
// The RHS should have been converted to the computation type.
if (E->getRHS()->getType()->isRealFloatingType()) {
@@ -1252,8 +1257,9 @@ 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);
if (LHSTy->isAnyComplexType()) {
ComplexPairTy LHSVal = EmitLoadOfLValue(LHS, Loc);
if (!PromotionTypeLHS.isNull())
diff --git a/clang/test/CodeGen/cx-complex-range.c b/clang/test/CodeGen/cx-complex-range.c
index 06a349fbc2a47..a724e1ca8cb6d 100644
--- a/clang/test/CodeGen/cx-complex-range.c
+++ b/clang/test/CodeGen/cx-complex-range.c
@@ -721,44 +721,32 @@ _Complex float divf(_Complex float a, _Complex float b) {
// PRMTD-NEXT: [[B_REAL:%.*]] = load float, ptr [[B_REALP]], align 4
// PRMTD-NEXT: [[B_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[B]], i32 0, i32 1
// PRMTD-NEXT: [[B_IMAG:%.*]] = load float, ptr [[B_IMAGP]], align 4
+// PRMTD-NEXT: [[EXT:%.*]] = fpext float [[B_REAL]] to double
+// PRMTD-NEXT: [[EXT1:%.*]] = fpext float [[B_IMAG]] to double
// PRMTD-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// PRMTD-NEXT: [[DOTREALP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 0
// PRMTD-NEXT: [[DOTREAL:%.*]] = load float, ptr [[DOTREALP]], align 4
// PRMTD-NEXT: [[DOTIMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 1
// PRMTD-NEXT: [[DOTIMAG:%.*]] = load float, ptr [[DOTIMAGP]], align 4
-// PRMTD-NEXT: [[TMP1:%.*]] = call float @llvm.fabs.f32(float [[B_REAL]])
-// PRMTD-NEXT: [[TMP2:%.*]] = call float @llvm.fabs.f32(float [[B_IMAG]])
-// PRMTD-NEXT: [[ABS_CMP:%.*]] = fcmp ugt float [[TMP1]], [[TMP2]]
-// PRMTD-NEXT: br i1 [[ABS_CMP]], label [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI:%.*]], label [[ABS_RHSR_LESS_THAN_ABS_RHSI:%.*]]
-// PRMTD: abs_rhsr_greater_or_equal_abs_rhsi:
-// PRMTD-NEXT: [[TMP3:%.*]] = fdiv float [[B_IMAG]], [[B_REAL]]
-// PRMTD-NEXT: [[TMP4:%.*]] = fmul float [[TMP3]], [[B_IMAG]]
-// PRMTD-NEXT: [[TMP5:%.*]] = fadd float [[B_REAL]], [[TMP4]]
-// PRMTD-NEXT: [[TMP6:%.*]] = fmul float [[DOTIMAG]], [[TMP3]]
-// PRMTD-NEXT: [[TMP7:%.*]] = fadd float [[DOTREAL]], [[TMP6]]
-// PRMTD-NEXT: [[TMP8:%.*]] = fdiv float [[TMP7]], [[TMP5]]
-// PRMTD-NEXT: [[TMP9:%.*]] = fmul float [[DOTREAL]], [[TMP3]]
-// PRMTD-NEXT: [[TMP10:%.*]] = fsub float [[DOTIMAG]], [[TMP9]]
-// PRMTD-NEXT: [[TMP11:%.*]] = fdiv float [[TMP10]], [[TMP5]]
-// PRMTD-NEXT: br label [[COMPLEX_DIV:%.*]]
-// PRMTD: abs_rhsr_less_than_abs_rhsi:
-// PRMTD-NEXT: [[TMP12:%.*]] = fdiv float [[B_REAL]], [[B_IMAG]]
-// PRMTD-NEXT: [[TMP13:%.*]] = fmul float [[TMP12]], [[B_REAL]]
-// PRMTD-NEXT: [[TMP14:%.*]] = fadd float [[B_IMAG]], [[TMP13]]
-// PRMTD-NEXT: [[TMP15:%.*]] = fmul float [[DOTREAL]], [[TMP12]]
-// PRMTD-NEXT: [[TMP16:%.*]] = fadd float [[TMP15]], [[DOTIMAG]]
-// PRMTD-NEXT: [[TMP17:%.*]] = fdiv float [[TMP16]], [[TMP14]]
-// PRMTD-NEXT: [[TMP18:%.*]] = fmul float [[DOTIMAG]], [[TMP12]]
-// PRMTD-NEXT: [[TMP19:%.*]] = fsub float [[TMP18]], [[DOTREAL]]
-// PRMTD-NEXT: [[TMP20:%.*]] = fdiv float [[TMP19]], [[TMP14]]
-// PRMTD-NEXT: br label [[COMPLEX_DIV]]
-// PRMTD: complex_div:
-// PRMTD-NEXT: [[TMP21:%.*]] = phi float [ [[TMP8]], [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI]] ], [ [[TMP17]], [[ABS_RHSR_LESS_THAN_ABS_RHSI]] ]
-// PRMTD-NEXT: [[TMP22:%.*]] = phi float [ [[TMP11]], [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI]] ], [ [[TMP20]], [[ABS_RHSR_LESS_THAN_ABS_RHSI]] ]
-// PRMTD-NEXT: [[DOTREALP1:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 0
-// PRMTD-NEXT: [[DOTIMAGP2:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 1
-// PRMTD-NEXT: store float [[TMP21]], ptr [[DOTREALP1]], align 4
-// PRMTD-NEXT: store float [[TMP22]], ptr [[DOTIMAGP2]], align 4
+// PRMTD-NEXT: [[CONV:%.*]] = fpext float [[DOTREAL]] to double
+// PRMTD-NEXT: [[CONV2:%.*]] = fpext float [[DOTIMAG]] to double
+// PRMTD-NEXT: [[TMP1:%.*]] = fmul double [[CONV]], [[EXT]]
+// PRMTD-NEXT: [[TMP2:%.*]] = fmul double [[CONV2]], [[EXT1]]
+// PRMTD-NEXT: [[TMP3:%.*]] = fadd double [[TMP1]], [[TMP2]]
+// PRMTD-NEXT: [[TMP4:%.*]] = fmul double [[EXT]], [[EXT]]
+// PRMTD-NEXT: [[TMP5:%.*]] = fmul double [[EXT1]], [[EXT1]]
+// PRMTD-NEXT: [[TMP6:%.*]] = fadd double [[TMP4]], [[TMP5]]
+// PRMTD-NEXT: [[TMP7:%.*]] = fmul double [[CONV2]], [[EXT]]
+// PRMTD-NEXT: [[TMP8:%.*]] = fmul double [[CONV]], [[EXT1]]
+// PRMTD-NEXT: [[TMP9:%.*]] = fsub double [[TMP7]], [[TMP8]]
+// PRMTD-NEXT: [[TMP10:%.*]] = fdiv double [[TMP3]], [[TMP6]]
+// PRMTD-NEXT: [[TMP11:%.*]] = fdiv double [[TMP9]], [[TMP6]]
+// PRMTD-NEXT: [[CONV3:%.*]] = fptrunc double [[TMP10]] to float
+// PRMTD-NEXT: [[CONV4:%.*]] = fptrunc double [[TMP11]] to float
+// PRMTD-NEXT: [[DOTREALP5:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 0
+// PRMTD-NEXT: [[DOTIMAGP6:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 1
+// PRMTD-NEXT: store float [[CONV3]], ptr [[DOTREALP5]], align 4
+// PRMTD-NEXT: store float [[CONV4]], ptr [[DOTIMAGP6]], align 4
// PRMTD-NEXT: ret void
//
// X86WINPRMTD-LABEL: define dso_local void @divassignf(
@@ -772,44 +760,32 @@ _Complex float divf(_Complex float a, _Complex float b) {
// X86WINPRMTD-NEXT: [[B_REAL:%.*]] = load float, ptr [[B_REALP]], align 4
// X86WINPRMTD-NEXT: [[B_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[B]], i32 0, i32 1
// X86WINPRMTD-NEXT: [[B_IMAG:%.*]] = load float, ptr [[B_IMAGP]], align 4
+// X86WINPRMTD-NEXT: [[EXT:%.*]] = fpext float [[B_REAL]] to double
+// X86WINPRMTD-NEXT: [[EXT1:%.*]] = fpext float [[B_IMAG]] to double
// X86WINPRMTD-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// X86WINPRMTD-NEXT: [[DOTREALP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 0
// X86WINPRMTD-NEXT: [[DOTREAL:%.*]] = load float, ptr [[DOTREALP]], align 4
// X86WINPRMTD-NEXT: [[DOTIMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 1
// X86WINPRMTD-NEXT: [[DOTIMAG:%.*]] = load float, ptr [[DOTIMAGP]], align 4
-// X86WINPRMTD-NEXT: [[TMP1:%.*]] = call float @llvm.fabs.f32(float [[B_REAL]])
-// X86WINPRMTD-NEXT: [[TMP2:%.*]] = call float @llvm.fabs.f32(float [[B_IMAG]])
-// X86WINPRMTD-NEXT: [[ABS_CMP:%.*]] = fcmp ugt float [[TMP1]], [[TMP2]]
-// X86WINPRMTD-NEXT: br i1 [[ABS_CMP]], label [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI:%.*]], label [[ABS_RHSR_LESS_THAN_ABS_RHSI:%.*]]
-// X86WINPRMTD: abs_rhsr_greater_or_equal_abs_rhsi:
-// X86WINPRMTD-NEXT: [[TMP3:%.*]] = fdiv float [[B_IMAG]], [[B_REAL]]
-// X86WINPRMTD-NEXT: [[TMP4:%.*]] = fmul float [[TMP3]], [[B_IMAG]]
-// X86WINPRMTD-NEXT: [[TMP5:%.*]] = fadd float [[B_REAL]], [[TMP4]]
-// X86WINPRMTD-NEXT: [[TMP6:%.*]] = fmul float [[DOTIMAG]], [[TMP3]]
-// X86WINPRMTD-NEXT: [[TMP7:%.*]] = fadd float [[DOTREAL]], [[TMP6]]
-// X86WINPRMTD-NEXT: [[TMP8:%.*]] = fdiv float [[TMP7]], [[TMP5]]
-// X86WINPRMTD-NEXT: [[TMP9:%.*]] = fmul float [[DOTREAL]], [[TMP3]]
-// X86WINPRMTD-NEXT: [[TMP10:%.*]] = fsub float [[DOTIMAG]], [[TMP9]]
-// X86WINPRMTD-NEXT: [[TMP11:%.*]] = fdiv float [[TMP10]], [[TMP5]]
-// X86WINPRMTD-NEXT: br label [[COMPLEX_DIV:%.*]]
-// X86WINPRMTD: abs_rhsr_less_than_abs_rhsi:
-// X86WINPRMTD-NEXT: [[TMP12:%.*]] = fdiv float [[B_REAL]], [[B_IMAG]]
-// X86WINPRMTD-NEXT: [[TMP13:%.*]] = fmul float [[TMP12]], [[B_REAL]]
-// X86WINPRMTD-NEXT: [[TMP14:%.*]] = fadd float [[B_IMAG]], [[TMP13]]
-// X86WINPRMTD-NEXT: [[TMP15:%.*]] = fmul float [[DOTREAL]], [[TMP12]]
-// X86WINPRMTD-NEXT: [[TMP16:%.*]] = fadd float [[TMP15]], [[DOTIMAG]]
-// X86WINPRMTD-NEXT: [[TMP17:%.*]] = fdiv float [[TMP16]], [[TMP14]]
-// X86WINPRMTD-NEXT: [[TMP18:%.*]] = fmul float [[DOTIMAG]], [[TMP12]]
-// X86WINPRMTD-NEXT: [[TMP19:%.*]] = fsub float [[TMP18]], [[DOTREAL]]
-// X86WINPRMTD-NEXT: [[TMP20:%.*]] = fdiv float [[TMP19]], [[TMP14]]
-// X86WINPRMTD-NEXT: br label [[COMPLEX_DIV]]
-// X86WINPRMTD: complex_div:
-// X86WINPRMTD-NEXT: [[TMP21:%.*]] = phi float [ [[TMP8]], [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI]] ], [ [[TMP17]], [[ABS_RHSR_LESS_THAN_ABS_RHSI]] ]
-// X86WINPRMTD-NEXT: [[TMP22:%.*]] = phi float [ [[TMP11]], [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI]] ], [ [[TMP20]], [[ABS_RHSR_LESS_THAN_ABS_RHSI]] ]
-// X86WINPRMTD-NEXT: [[DOTREALP1:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 0
-// X86WINPRMTD-NEXT: [[DOTIMAGP2:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 1
-// X86WINPRMTD-NEXT: store float [[TMP21]], ptr [[DOTREALP1]], align 4
-// X86WINPRMTD-NEXT: store float [[TMP22]], ptr [[DOTIMAGP2]], align 4
+// X86WINPRMTD-NEXT: [[CONV:%.*]] = fpext float [[DOTREAL]] to double
+// X86WINPRMTD-NEXT: [[CONV2:%.*]] = fpext float [[DOTIMAG]] to double
+// X86WINPRMTD-NEXT: [[TMP1:%.*]] = fmul double [[CONV]], [[EXT]]
+// X86WINPRMTD-NEXT: [[TMP2:%.*]] = fmul double [[CONV2]], [[EXT1]]
+// X86WINPRMTD-NEXT: [[TMP3:%.*]] = fadd double [[TMP1]], [[TMP2]]
+// X86WINPRMTD-NEXT: [[TMP4:%.*]] = fmul double [[EXT]], [[EXT]]
+// X86WINPRMTD-NEXT: [[TMP5:%.*]] = fmul double [[EXT1]], [[EXT1]]
+// X86WINPRMTD-NEXT: [[TMP6:%.*]] = fadd double [[TMP4]], [[TMP5]]
+// X86WINPRMTD-NEXT: [[TMP7:%.*]] = fmul double [[CONV2]], [[EXT]]
+// X86WINPRMTD-NEXT: [[TMP8:%.*]] = fmul double [[CONV]], [[EXT1]]
+// X86WINPRMTD-NEXT: [[TMP9:%.*]] = fsub double [[TMP7]], [[TMP8]]
+// X86WINPRMTD-NEXT: [[TMP10:%.*]] = fdiv double [[TMP3]], [[TMP6]]
+// X86WINPRMTD-NEXT: [[TMP11:%.*]] = fdiv double [[TMP9]], [[TMP6]]
+// X86WINPRMTD-NEXT: [[CONV3:%.*]] = fptrunc double [[TMP10]] to float
+// X86WINPRMTD-NEXT: [[CONV4:%.*]] = fptrunc double [[TMP11]] to float
+// X86WINPRMTD-NEXT: [[DOTREALP5:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 0
+// X86WINPRMTD-NEXT: [[DOTIMAGP6:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 1
+// X86WINPRMTD-NEXT: store float [[CONV3]], ptr [[DOTREALP5]], align 4
+// X86WINPRMTD-NEXT: store float [[CONV4]], ptr [[DOTIMAGP6]], align 4
// X86WINPRMTD-NEXT: ret void
//
// AVRFP32-LABEL: define dso_local void @divassignf(
@@ -880,44 +856,32 @@ _Complex float divf(_Complex float a, _Complex float b) {
// AVRFP64-NEXT: [[B_REAL:%.*]] = load float, ptr [[B_REALP]], align 1
// AVRFP64-NEXT: [[B_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[B]], i32 0, i32 1
// AVRFP64-NEXT: [[B_IMAG:%.*]] = load float, ptr [[B_IMAGP]], align 1
+// AVRFP64-NEXT: [[EXT:%.*]] = fpext float [[B_REAL]] to double
+// AVRFP64-NEXT: [[EXT1:%.*]] = fpext float [[B_IMAG]] to double
// AVRFP64-NEXT: [[TMP2:%.*]] = load ptr, ptr [[A_ADDR]], align 1
// AVRFP64-NEXT: [[DOTREALP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP2]], i32 0, i32 0
// AVRFP64-NEXT: [[DOTREAL:%.*]] = load float, ptr [[DOTREALP]], align 1
// AVRFP64-NEXT: [[DOTIMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP2]], i32 0, i32 1
// AVRFP64-NEXT: [[DOTIMAG:%.*]] = load float, ptr [[DOTIMAGP]], align 1
-// AVRFP64-NEXT: [[TMP3:%.*]] = call addrspace(1) float @llvm.fabs.f32(float [[B_REAL]])
-// AVRFP64-NEXT: [[TMP4:%.*]] = call addrspace(1) float @llvm.fabs.f32(float [[B_IMAG]])
-// AVRFP64-NEXT: [[ABS_CMP:%.*]] = fcmp ugt float [[TMP3]], [[TMP4]]
-// AVRFP64-NEXT: br i1 [[ABS_CMP]], label [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI:%.*]], label [[ABS_RHSR_LESS_THAN_ABS_RHSI:%.*]]
-// AVRFP64: abs_rhsr_greater_or_equal_abs_rhsi:
-// AVRFP64-NEXT: [[TMP5:%.*]] = fdiv float [[B_IMAG]], [[B_REAL]]
-// AVRFP64-NEXT: [[TMP6:%.*]] = fmul float [[TMP5]], [[B_IMAG]]
-// AVRFP64-NEXT: [[TMP7:%.*]] = fadd float [[B_REAL]], [[TMP6]]
-// AVRFP64-NEXT: [[TMP8:%.*]] = fmul float [[DOTIMAG]], [[TMP5]]
-// AVRFP64-NEXT: [[TMP9:%.*]] = fadd float [[DOTREAL]], [[TMP8]]
-// AVRFP64-NEXT: [[TMP10:%.*]] = fdiv float [[TMP9]], [[TMP7]]
-// AVRFP64-NEXT: [[TMP11:%.*]] = fmul float [[DOTREAL]], [[TMP5]]
-// AVRFP64-NEXT: [[TMP12:%.*]] = fsub float [[DOTIMAG]], [[TMP11]]
-// AVRFP64-NEXT: [[TMP13:%.*]] = fdiv float [[TMP12]], [[TMP7]]
-// AVRFP64-NEXT: br label [[COMPLEX_DIV:%.*]]
-// AVRFP64: abs_rhsr_less_than_abs_rhsi:
-// AVRFP64-NEXT: [[TMP14:%.*]] = fdiv float [[B_REAL]], [[B_IMAG]]
-// AVRFP64-NEXT: [[TMP15:%.*]] = fmul float [[TMP14]], [[B_REAL]]
-// AVRFP64-NEXT: [[TMP16:%.*]] = fadd float [[B_IMAG]], [[TMP15]]
-// AVRFP64-NEXT: [[TMP17:%.*]] = fmul float [[DOTREAL]], [[TMP14]]
-// AVRFP64-NEXT: [[TMP18:%.*]] = fadd float [[TMP17]], [[DOTIMAG]]
-// AVRFP64-NEXT: [[TMP19:%.*]] = fdiv float [[TMP18]], [[TMP16]]
-// AVRFP64-NEXT: [[TMP20:%.*]] = fmul float [[DOTIMAG]], [[TMP14]]
-// AVRFP64-NEXT: [[TMP21:%.*]] = fsub float [[TMP20]], [[DOTREAL]]
-// AVRFP64-NEXT: [[TMP22:%.*]] = fdiv float [[TMP21]], [[TMP16]]
-// AVRFP64-NEXT: br label [[COMPLEX_DIV]]
-// AVRFP64: complex_div:
-// AVRFP64-NEXT: [[TMP23:%.*]] = phi float [ [[TMP10]], [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI]] ], [ [[TMP19]], [[ABS_RHSR_LESS_THAN_ABS_RHSI]] ]
-// AVRFP64-NEXT: [[TMP24:%.*]] = phi float [ [[TMP13]], [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI]] ], [ [[TMP22]], [[ABS_RHSR_LESS_THAN_ABS_RHSI]] ]
-// AVRFP64-NEXT: [[DOTREALP1:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP2]], i32 0, i32 0
-// AVRFP64-NEXT: [[DOTIMAGP2:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP2]], i32 0, i32 1
-// AVRFP64-NEXT: store float [[TMP23]], ptr [[DOTREALP1]], align 1
-// AVRFP64-NEXT: store float [[TMP24]], ptr [[DOTIMAGP2]], align 1
+// AVRFP64-NEXT: [[CONV:%.*]] = fpext float [[DOTREAL]] to double
+// AVRFP64-NEXT: [[CONV2:%.*]] = fpext float [[DOTIMAG]] to double
+// AVRFP64-NEXT: [[TMP3:%.*]] = fmul double [[CONV]], [[EXT]]
+// AVRFP64-NEXT: [[TMP4:%.*]] = fmul double [[CONV2]], [[EXT1]]
+// AVRFP64-NEXT: [[TMP5:%.*]] = fadd double [[TMP3]], [[TMP4]]
+// AVRFP64-NEXT: [[TMP6:%.*]] = fmul double [[EXT]], [[EXT]]
+// AVRFP64-NEXT: [[TMP7:%.*]] = fmul double [[EXT1]], [[EXT1]]
+// AVRFP64-NEXT: [[TMP8:%.*]] = fadd double [[TMP6]], [[TMP7]]
+// AVRFP64-NEXT: [[TMP9:%.*]] = fmul double [[CONV2]], [[EXT]]
+// AVRFP64-NEXT: [[TMP10:%.*]] = fmul double [[CONV]], [[EXT1]]
+// AVRFP64-NEXT: [[TMP11:%.*]] = fsub double [[TMP9]], [[TMP10]]
+// AVRFP64-NEXT: [[TMP12:%.*]] = fdiv double [[TMP5]], [[TMP8]]
+// AVRFP64-NEXT: [[TMP13:%.*]] = fdiv double [[TMP11]], [[TMP8]]
+// AVRFP64-NEXT: [[CONV3:%.*]] = fptrunc double [[TMP12]] to float
+// AVRFP64-NEXT: [[CONV4:%.*]] = fptrunc double [[TMP13]] to float
+// AVRFP64-NEXT: [[DOTREALP5:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP2]], i32 0, i32 0
+// AVRFP64-NEXT: [[DOTIMAGP6:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP2]], i32 0, i32 1
+// AVRFP64-NEXT: store float [[CONV3]], ptr [[DOTREALP5]], align 1
+// AVRFP64-NEXT: store float [[CONV4]], ptr [[DOTIMAGP6]], align 1
// AVRFP64-NEXT: ret void
//
// BASIC_FAST-LABEL: define dso_local void @divassignf(
@@ -1044,44 +1008,32 @@ _Complex float divf(_Complex float a, _Complex float b) {
// PRMTD_FAST-NEXT: [[B_REAL:%.*]] = load float, ptr [[B_REALP]], align 4
// PRMTD_FAST-NEXT: [[B_IMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[B]], i32 0, i32 1
// PRMTD_FAST-NEXT: [[B_IMAG:%.*]] = load float, ptr [[B_IMAGP]], align 4
+// PRMTD_FAST-NEXT: [[EXT:%.*]] = fpext reassoc nnan ninf nsz arcp afn float [[B_REAL]] to double
+// PRMTD_FAST-NEXT: [[EXT1:%.*]] = fpext reassoc nnan ninf nsz arcp afn float [[B_IMAG]] to double
// PRMTD_FAST-NEXT: [[TMP0:%.*]] = load ptr, ptr [[A_ADDR]], align 8
// PRMTD_FAST-NEXT: [[DOTREALP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 0
// PRMTD_FAST-NEXT: [[DOTREAL:%.*]] = load float, ptr [[DOTREALP]], align 4
// PRMTD_FAST-NEXT: [[DOTIMAGP:%.*]] = getelementptr inbounds nuw { float, float }, ptr [[TMP0]], i32 0, i32 1
// PRMTD_FAST-NEXT: [[DOTIMAG:%.*]] = load float, ptr [[DOTIMAGP]], align 4
-// PRMTD_FAST-NEXT: [[TMP1:%.*]] = call reassoc nnan ninf nsz arcp afn float @llvm.fabs.f32(float [[B_REAL]])
-// PRMTD_FAST-NEXT: [[TMP2:%.*]] = call reassoc nnan ninf nsz arcp afn float @llvm.fabs.f32(float [[B_IMAG]])
-// PRMTD_FAST-NEXT: [[ABS_CMP:%.*]] = fcmp reassoc nnan ninf nsz arcp afn ugt float [[TMP1]], [[TMP2]]
-// PRMTD_FAST-NEXT: br i1 [[ABS_CMP]], label [[ABS_RHSR_GREATER_OR_EQUAL_ABS_RHSI:%.*]], label [[ABS_RHSR_LESS_THAN_ABS_RHSI:%.*]]
-// PRMTD_FAST: abs_rhsr_greater_or_equal_abs_rhsi:
-// PRMTD_FAST-NEXT: [[TMP3:%.*]] = fdiv reassoc nnan ninf nsz arcp afn float [[B_IMAG]], [[B_REAL]]
-// PRMTD_FAST-NEXT: [[TMP4:%.*]] = fmul reassoc nnan ninf nsz arcp afn float [[TMP3]], [[B_IMAG]]
-// PRMTD_FAST-NEXT: [[TMP5:%.*]] = fadd reassoc nnan ninf nsz arcp afn float [[B_REAL]], [[TMP4]]
-// PRMTD_FAST-NEXT: [[TMP6:%.*]] = fmul reassoc nnan ninf nsz arcp afn float [[DOTIMAG]], [[TMP3]]
-// PRMTD_FAST-NEXT: [[TMP7:%.*]] = fadd reassoc nnan ninf nsz arcp afn float [[DOTREAL]], [[TMP6]]
-// PRMTD_FAST-NEXT: [[TMP8:%.*]] = fdiv reassoc nnan ninf nsz arcp afn float [[TMP7]], [[TMP5]]
-// PRMTD_FAST-NEXT: [[TMP9:%.*]] = fmul reassoc nnan ninf nsz arcp afn float [[DOTREAL]], [[TMP3]]
-// PRMTD_FAST-NEXT: [[TMP10:%.*]] = fsub reassoc nnan ninf nsz arcp afn float [[DOTIMAG]], [[TMP9]]
-// PRMTD_FAST-NEXT: [[TMP11:%.*]] = fdiv reassoc nnan ninf nsz arcp afn float [[TMP10]], [[TMP5]]
-// PRMTD_FAST-NEXT: br label [[COMPLEX_DIV:%.*]]
-// PRMTD_FAST: abs_rhsr_less_than_abs_rhsi:
-// PRMTD_FAST-NEXT: [[TMP12:%.*]] = fdiv reassoc nnan ninf nsz arcp afn float [[B_REAL]], [[B_IMAG]]
-// PRMTD_FAST-NEXT: [[TMP13:%.*]] = fmul reassoc nnan ninf nsz arcp afn float [[TMP12]], [[B_REAL]]
-// PRMTD_FAST-NEXT: [[TMP14:%.*]] = fadd reassoc nnan ninf nsz arcp afn float [[B_IMAG]], [[TMP13]]
-// PRMTD_FAST-NEXT: [[TMP15:%.*]] = fmul reassoc nnan ninf nsz arcp afn float [[DOTREAL]], [[TMP12]]
-// PRMTD_FAST-NEXT: [[TMP16:%.*]] = fadd reassoc nnan ninf nsz arcp afn float [[TMP15]], [[DOTIMAG]]
-// PRMTD_FAST-NEXT: [[TM...
[truncated]
|
Maetveis
force-pushed
the
users/Maetveis/complex-promoted-fix-assign
branch
from
de63899 to
9d50aa0
Compare
Maetveis
force-pushed
the
users/Maetveis/complex-promoted-fix-real
branch
2 times, most recently
from
c6e05d7 to
549f55f
Compare
Maetveis
force-pushed
the
users/Maetveis/complex-promoted-fix-assign
branch
from
9d50aa0 to
f02c8a6
Compare
Maetveis
force-pushed
the
users/Maetveis/complex-promoted-fix-real
branch
from
549f55f to
631db26
Compare
Maetveis
force-pushed
the
users/Maetveis/complex-promoted-fix-assign
branch
from
f02c8a6 to
57b488a
Compare
Maetveis
force-pushed
the
users/Maetveis/complex-promoted-fix-real
branch
from
631db26 to
5767414
Compare
Base automatically changed from
users/Maetveis/complex-promoted-fix-real
to
main
March 19, 2025 06:26
When `-fcomplex-arithmetic=promoted` is set complex divassign `/=` should promote to a wider type the same way division (without assignment) does. Prior to this change, Smith's algorithm would be used for divassign. Fixes: #131129
Maetveis
force-pushed
the
users/Maetveis/complex-promoted-fix-assign
branch
from
57b488a to
f2a2bf5
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
When
-fcomplex-arithmetic=promotedis set complex divassign/=shouldpromote to a wider type the same way division (without assignment) does.
Prior to this change, Smith's algorithm would be used for divassign.
Fixes: #131129