https://github.com/wzssyqa updated https://github.com/llvm/llvm-project/pull/113133
>From 34654e8872814090664dd5ac476e801d83fa6cfc Mon Sep 17 00:00:00 2001 From: YunQiang Su <yunqi...@isrc.iscas.ac.cn> Date: Mon, 21 Oct 2024 15:18:38 +0800 Subject: [PATCH 1/7] Clang: emit llvm.minnum and llvm.maxnum with nsz always See: https://github.com/llvm/llvm-project/pull/112852 We will define llvm.minnum and llvm.maxnum with +0.0>-0.0, by default, while libc doesn't require it. --- clang/lib/CodeGen/CGBuiltin.cpp | 33 ++++++++++++-------- clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c | 33 ++++++++++++++++++++ llvm/include/llvm/IR/IRBuilder.h | 16 ++++++---- llvm/lib/IR/IRBuilder.cpp | 4 +-- 4 files changed, 65 insertions(+), 21 deletions(-) create mode 100644 clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp index 65fac01d58362..f8a13101d586b 100644 --- a/clang/lib/CodeGen/CGBuiltin.cpp +++ b/clang/lib/CodeGen/CGBuiltin.cpp @@ -689,19 +689,20 @@ static Value *emitUnaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, // Emit an intrinsic that has 2 operands of the same type as its result. // Depending on mode, this may be a constrained floating-point intrinsic. -static Value *emitBinaryMaybeConstrainedFPBuiltin(CodeGenFunction &CGF, - const CallExpr *E, unsigned IntrinsicID, - unsigned ConstrainedIntrinsicID) { +static Value *emitBinaryMaybeConstrainedFPBuiltin( + CodeGenFunction &CGF, const CallExpr *E, unsigned IntrinsicID, + unsigned ConstrainedIntrinsicID, llvm::FastMathFlags *FMF = nullptr) { llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0)); llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1)); CodeGenFunction::CGFPOptionsRAII FPOptsRAII(CGF, E); if (CGF.Builder.getIsFPConstrained()) { Function *F = CGF.CGM.getIntrinsic(ConstrainedIntrinsicID, Src0->getType()); - return CGF.Builder.CreateConstrainedFPCall(F, { Src0, Src1 }); + return CGF.Builder.CreateConstrainedFPCall(F, {Src0, Src1}, "", + std::nullopt, std::nullopt, FMF); } else { Function *F = CGF.CGM.getIntrinsic(IntrinsicID, Src0->getType()); - return CGF.Builder.CreateCall(F, { Src0, Src1 }); + return CGF.Builder.CreateCall(F, {Src0, Src1}, "", nullptr, FMF); } } @@ -3203,10 +3204,13 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, case Builtin::BI__builtin_fmaxf: case Builtin::BI__builtin_fmaxf16: case Builtin::BI__builtin_fmaxl: - case Builtin::BI__builtin_fmaxf128: - return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, - Intrinsic::maxnum, - Intrinsic::experimental_constrained_maxnum)); + case Builtin::BI__builtin_fmaxf128: { + llvm::FastMathFlags FMF; + FMF.setNoSignedZeros(); + return RValue::get(emitBinaryMaybeConstrainedFPBuiltin( + *this, E, Intrinsic::maxnum, + Intrinsic::experimental_constrained_maxnum, &FMF)); + } case Builtin::BIfmin: case Builtin::BIfminf: @@ -3215,10 +3219,13 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID, case Builtin::BI__builtin_fminf: case Builtin::BI__builtin_fminf16: case Builtin::BI__builtin_fminl: - case Builtin::BI__builtin_fminf128: - return RValue::get(emitBinaryMaybeConstrainedFPBuiltin(*this, E, - Intrinsic::minnum, - Intrinsic::experimental_constrained_minnum)); + case Builtin::BI__builtin_fminf128: { + llvm::FastMathFlags FMF; + FMF.setNoSignedZeros(); + return RValue::get(emitBinaryMaybeConstrainedFPBuiltin( + *this, E, Intrinsic::minnum, + Intrinsic::experimental_constrained_minnum, &FMF)); + } case Builtin::BIfmaximum_num: case Builtin::BIfmaximum_numf: diff --git a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c new file mode 100644 index 0000000000000..9798baf0432fe --- /dev/null +++ b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c @@ -0,0 +1,33 @@ +// RUN: %clang_cc1 -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK + +float fminf (float, float); +double fmin (double, double); +long double fminl (long double, long double); +float fmaxf (float, float); +double fmax (double, double); +long double fmaxl (long double, long double); + +// CHECK: call nsz float @llvm.minnum.f32 +float fmin1(float a, float b) { + return fminf(a, b); +} +// CHECK: call nsz double @llvm.minnum.f64 +float fmin2(double a, double b) { + return fmin(a, b); +} +// CHECK: call nsz x86_fp80 @llvm.minnum.f80 +float fmin3(long double a, long double b) { + return fminl(a, b); +} +// CHECK: call nsz float @llvm.maxnum.f32 +float fmax1(float a, float b) { + return fmaxf(a, b); +} +// CHECK: call nsz double @llvm.maxnum.f64 +float fmax2(double a, double b) { + return fmax(a, b); +} +// CHECK: call nsz x86_fp80 @llvm.maxnum.f80 +float fmax3(long double a, long double b) { + return fmaxl(a, b); +} diff --git a/llvm/include/llvm/IR/IRBuilder.h b/llvm/include/llvm/IR/IRBuilder.h index 933dbb306d1fc..a310e023e77e6 100644 --- a/llvm/include/llvm/IR/IRBuilder.h +++ b/llvm/include/llvm/IR/IRBuilder.h @@ -2448,12 +2448,14 @@ class IRBuilderBase { public: CallInst *CreateCall(FunctionType *FTy, Value *Callee, ArrayRef<Value *> Args = {}, const Twine &Name = "", - MDNode *FPMathTag = nullptr) { + MDNode *FPMathTag = nullptr, + FastMathFlags *uFMF = nullptr) { CallInst *CI = CallInst::Create(FTy, Callee, Args, DefaultOperandBundles); if (IsFPConstrained) setConstrainedFPCallAttr(CI); - if (isa<FPMathOperator>(CI)) - setFPAttrs(CI, FPMathTag, FMF); + if (isa<FPMathOperator>(CI)) { + setFPAttrs(CI, FPMathTag, uFMF ? (FMF | *uFMF) : FMF); + } return Insert(CI, Name); } @@ -2469,9 +2471,10 @@ class IRBuilderBase { } CallInst *CreateCall(FunctionCallee Callee, ArrayRef<Value *> Args = {}, - const Twine &Name = "", MDNode *FPMathTag = nullptr) { + const Twine &Name = "", MDNode *FPMathTag = nullptr, + FastMathFlags *uFMF = nullptr) { return CreateCall(Callee.getFunctionType(), Callee.getCallee(), Args, Name, - FPMathTag); + FPMathTag, uFMF); } CallInst *CreateCall(FunctionCallee Callee, ArrayRef<Value *> Args, @@ -2484,7 +2487,8 @@ class IRBuilderBase { CallInst *CreateConstrainedFPCall( Function *Callee, ArrayRef<Value *> Args, const Twine &Name = "", std::optional<RoundingMode> Rounding = std::nullopt, - std::optional<fp::ExceptionBehavior> Except = std::nullopt); + std::optional<fp::ExceptionBehavior> Except = std::nullopt, + FastMathFlags *FMF = nullptr); Value *CreateSelect(Value *C, Value *True, Value *False, const Twine &Name = "", Instruction *MDFrom = nullptr); diff --git a/llvm/lib/IR/IRBuilder.cpp b/llvm/lib/IR/IRBuilder.cpp index 134459265cecb..f6f6c6bd6e6be 100644 --- a/llvm/lib/IR/IRBuilder.cpp +++ b/llvm/lib/IR/IRBuilder.cpp @@ -1036,7 +1036,7 @@ CallInst *IRBuilderBase::CreateConstrainedFPCmp( CallInst *IRBuilderBase::CreateConstrainedFPCall( Function *Callee, ArrayRef<Value *> Args, const Twine &Name, std::optional<RoundingMode> Rounding, - std::optional<fp::ExceptionBehavior> Except) { + std::optional<fp::ExceptionBehavior> Except, FastMathFlags *FMF) { llvm::SmallVector<Value *, 6> UseArgs; append_range(UseArgs, Args); @@ -1045,7 +1045,7 @@ CallInst *IRBuilderBase::CreateConstrainedFPCall( UseArgs.push_back(getConstrainedFPRounding(Rounding)); UseArgs.push_back(getConstrainedFPExcept(Except)); - CallInst *C = CreateCall(Callee, UseArgs, Name); + CallInst *C = CreateCall(Callee, UseArgs, Name, nullptr, FMF); setConstrainedFPCallAttr(C); return C; } >From 38c6bd7236188b80809d7fbd731582b83b9b7d24 Mon Sep 17 00:00:00 2001 From: YunQiang Su <yunqi...@isrc.iscas.ac.cn> Date: Tue, 22 Oct 2024 09:56:06 +0800 Subject: [PATCH 2/7] fix testcases --- clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c | 36 ++++++++++++--- clang/test/Headers/__clang_hip_math.hip | 48 ++++++++++---------- 2 files changed, 54 insertions(+), 30 deletions(-) diff --git a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c index 9798baf0432fe..a708d5cd3ac9e 100644 --- a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c +++ b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c @@ -8,26 +8,50 @@ double fmax (double, double); long double fmaxl (long double, long double); // CHECK: call nsz float @llvm.minnum.f32 -float fmin1(float a, float b) { +float fmin32(float a, float b) { return fminf(a, b); } +// CHECK: call nsz float @llvm.minnum.f32 +float fmin32b(float a, float b) { + return __builtin_fminf(a, b); +} // CHECK: call nsz double @llvm.minnum.f64 -float fmin2(double a, double b) { +float fmin64(double a, double b) { return fmin(a, b); } +// CHECK: call nsz double @llvm.minnum.f64 +float fmin64b(double a, double b) { + return __builtin_fmin(a, b); +} // CHECK: call nsz x86_fp80 @llvm.minnum.f80 -float fmin3(long double a, long double b) { +float fmin80(long double a, long double b) { return fminl(a, b); } +// CHECK: call nsz x86_fp80 @llvm.minnum.f80 +float fmin80b(long double a, long double b) { + return __builtin_fminl(a, b); +} // CHECK: call nsz float @llvm.maxnum.f32 -float fmax1(float a, float b) { +float fmax32(float a, float b) { return fmaxf(a, b); } +// CHECK: call nsz float @llvm.maxnum.f32 +float fmax32b(float a, float b) { + return __builtin_fmaxf(a, b); +} // CHECK: call nsz double @llvm.maxnum.f64 -float fmax2(double a, double b) { +float fmax64(double a, double b) { return fmax(a, b); } +// CHECK: call nsz double @llvm.maxnum.f64 +float fmax64b(double a, double b) { + return __builtin_fmax(a, b); +} // CHECK: call nsz x86_fp80 @llvm.maxnum.f80 -float fmax3(long double a, long double b) { +float fmax80(long double a, long double b) { return fmaxl(a, b); } +// CHECK: call nsz x86_fp80 @llvm.maxnum.f80 +float fmax80b(long double a, long double b) { + return __builtin_fmaxl(a, b); +} diff --git a/clang/test/Headers/__clang_hip_math.hip b/clang/test/Headers/__clang_hip_math.hip index e4254d1e64bec..b37f11e16fbd4 100644 --- a/clang/test/Headers/__clang_hip_math.hip +++ b/clang/test/Headers/__clang_hip_math.hip @@ -1248,17 +1248,17 @@ extern "C" __device__ double test_fma_rn(double x, double y, double z) { // DEFAULT-LABEL: @test_fmaxf( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) // DEFAULT-NEXT: ret float [[TMP0]] // // FINITEONLY-LABEL: @test_fmaxf( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) // FINITEONLY-NEXT: ret float [[TMP0]] // // APPROX-LABEL: @test_fmaxf( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) // APPROX-NEXT: ret float [[TMP0]] // extern "C" __device__ float test_fmaxf(float x, float y) { @@ -1267,17 +1267,17 @@ extern "C" __device__ float test_fmaxf(float x, float y) { // DEFAULT-LABEL: @test_fmax( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) // DEFAULT-NEXT: ret double [[TMP0]] // // FINITEONLY-LABEL: @test_fmax( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) // FINITEONLY-NEXT: ret double [[TMP0]] // // APPROX-LABEL: @test_fmax( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) // APPROX-NEXT: ret double [[TMP0]] // extern "C" __device__ double test_fmax(double x, double y) { @@ -1286,17 +1286,17 @@ extern "C" __device__ double test_fmax(double x, double y) { // DEFAULT-LABEL: @test_fminf( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) // DEFAULT-NEXT: ret float [[TMP0]] // // FINITEONLY-LABEL: @test_fminf( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) // FINITEONLY-NEXT: ret float [[TMP0]] // // APPROX-LABEL: @test_fminf( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) // APPROX-NEXT: ret float [[TMP0]] // extern "C" __device__ float test_fminf(float x, float y) { @@ -1305,17 +1305,17 @@ extern "C" __device__ float test_fminf(float x, float y) { // DEFAULT-LABEL: @test_fmin( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) // DEFAULT-NEXT: ret double [[TMP0]] // // FINITEONLY-LABEL: @test_fmin( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) // FINITEONLY-NEXT: ret double [[TMP0]] // // APPROX-LABEL: @test_fmin( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) // APPROX-NEXT: ret double [[TMP0]] // extern "C" __device__ double test_fmin(double x, double y) { @@ -5114,17 +5114,17 @@ extern "C" __device__ double test__fma_rn(double x, double y, double z) { // DEFAULT-LABEL: @test_float_min( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) // DEFAULT-NEXT: ret float [[TMP0]] // // FINITEONLY-LABEL: @test_float_min( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) // FINITEONLY-NEXT: ret float [[TMP0]] // // APPROX-LABEL: @test_float_min( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.minnum.f32(float [[X:%.*]], float [[Y:%.*]]) // APPROX-NEXT: ret float [[TMP0]] // extern "C" __device__ float test_float_min(float x, float y) { @@ -5133,17 +5133,17 @@ extern "C" __device__ float test_float_min(float x, float y) { // DEFAULT-LABEL: @test_float_max( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) // DEFAULT-NEXT: ret float [[TMP0]] // // FINITEONLY-LABEL: @test_float_max( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) // FINITEONLY-NEXT: ret float [[TMP0]] // // APPROX-LABEL: @test_float_max( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef float @llvm.maxnum.f32(float [[X:%.*]], float [[Y:%.*]]) // APPROX-NEXT: ret float [[TMP0]] // extern "C" __device__ float test_float_max(float x, float y) { @@ -5152,17 +5152,17 @@ extern "C" __device__ float test_float_max(float x, float y) { // DEFAULT-LABEL: @test_double_min( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) // DEFAULT-NEXT: ret double [[TMP0]] // // FINITEONLY-LABEL: @test_double_min( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) // FINITEONLY-NEXT: ret double [[TMP0]] // // APPROX-LABEL: @test_double_min( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.minnum.f64(double [[X:%.*]], double [[Y:%.*]]) // APPROX-NEXT: ret double [[TMP0]] // extern "C" __device__ double test_double_min(double x, double y) { @@ -5171,17 +5171,17 @@ extern "C" __device__ double test_double_min(double x, double y) { // DEFAULT-LABEL: @test_double_max( // DEFAULT-NEXT: entry: -// DEFAULT-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// DEFAULT-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) // DEFAULT-NEXT: ret double [[TMP0]] // // FINITEONLY-LABEL: @test_double_max( // FINITEONLY-NEXT: entry: -// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call nnan ninf nsz contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) // FINITEONLY-NEXT: ret double [[TMP0]] // // APPROX-LABEL: @test_double_max( // APPROX-NEXT: entry: -// APPROX-NEXT: [[TMP0:%.*]] = tail call contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) +// APPROX-NEXT: [[TMP0:%.*]] = tail call nsz contract noundef double @llvm.maxnum.f64(double [[X:%.*]], double [[Y:%.*]]) // APPROX-NEXT: ret double [[TMP0]] // extern "C" __device__ double test_double_max(double x, double y) { >From 74a7574e7d42c9120439e79eec9e20c367287578 Mon Sep 17 00:00:00 2001 From: YunQiang Su <yunqi...@isrc.iscas.ac.cn> Date: Tue, 22 Oct 2024 10:21:35 +0800 Subject: [PATCH 3/7] -ffp-exception-behavior=strict --- clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c | 45 +++++++------------- 1 file changed, 15 insertions(+), 30 deletions(-) diff --git a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c index a708d5cd3ac9e..f2ae4a49f6303 100644 --- a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c +++ b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c @@ -1,4 +1,5 @@ // RUN: %clang_cc1 -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clang_cc1 -ffp-exception-behavior=strict -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK-STRICT float fminf (float, float); double fmin (double, double); @@ -8,50 +9,34 @@ double fmax (double, double); long double fmaxl (long double, long double); // CHECK: call nsz float @llvm.minnum.f32 -float fmin32(float a, float b) { +// CHECK-STRICT: call nsz float @llvm.experimental.constrained.minnum.f32{{.*}} #2 +float fmin1(float a, float b) { return fminf(a, b); } -// CHECK: call nsz float @llvm.minnum.f32 -float fmin32b(float a, float b) { - return __builtin_fminf(a, b); -} // CHECK: call nsz double @llvm.minnum.f64 -float fmin64(double a, double b) { +// CHECK-STRICT: call nsz double @llvm.experimental.constrained.minnum.f64{{.*}} #2 +float fmin2(double a, double b) { return fmin(a, b); } -// CHECK: call nsz double @llvm.minnum.f64 -float fmin64b(double a, double b) { - return __builtin_fmin(a, b); -} // CHECK: call nsz x86_fp80 @llvm.minnum.f80 -float fmin80(long double a, long double b) { +// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80{{.*}} #2 +float fmin3(long double a, long double b) { return fminl(a, b); } -// CHECK: call nsz x86_fp80 @llvm.minnum.f80 -float fmin80b(long double a, long double b) { - return __builtin_fminl(a, b); -} // CHECK: call nsz float @llvm.maxnum.f32 -float fmax32(float a, float b) { +// CHECK-STRICT: call nsz float @llvm.experimental.constrained.maxnum.f32{{.*}} #2 +float fmax1(float a, float b) { return fmaxf(a, b); } -// CHECK: call nsz float @llvm.maxnum.f32 -float fmax32b(float a, float b) { - return __builtin_fmaxf(a, b); -} // CHECK: call nsz double @llvm.maxnum.f64 -float fmax64(double a, double b) { +// CHECK-STRICT: call nsz double @llvm.experimental.constrained.maxnum.f64{{.*}} #2 +float fmax2(double a, double b) { return fmax(a, b); } -// CHECK: call nsz double @llvm.maxnum.f64 -float fmax64b(double a, double b) { - return __builtin_fmax(a, b); -} // CHECK: call nsz x86_fp80 @llvm.maxnum.f80 -float fmax80(long double a, long double b) { +// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80{{.*}} #2 +float fmax3(long double a, long double b) { return fmaxl(a, b); } -// CHECK: call nsz x86_fp80 @llvm.maxnum.f80 -float fmax80b(long double a, long double b) { - return __builtin_fmaxl(a, b); -} + +//CHECK-STRICT: attributes #2 = { strictfp } >From f2f07bdf9c3aafd0da9e63315d3f9f2ad8e5d06d Mon Sep 17 00:00:00 2001 From: YunQiang Su <yunqi...@isrc.iscas.ac.cn> Date: Tue, 22 Oct 2024 10:36:46 +0800 Subject: [PATCH 4/7] add missing builtin test --- clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c | 40 +++++++++++++++++--- 1 file changed, 35 insertions(+), 5 deletions(-) diff --git a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c index f2ae4a49f6303..afe626d921125 100644 --- a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c +++ b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c @@ -10,33 +10,63 @@ long double fmaxl (long double, long double); // CHECK: call nsz float @llvm.minnum.f32 // CHECK-STRICT: call nsz float @llvm.experimental.constrained.minnum.f32{{.*}} #2 -float fmin1(float a, float b) { +float fmin32(float a, float b) { return fminf(a, b); } +// CHECK: call nsz float @llvm.minnum.f32 +// CHECK-STRICT: call nsz float @llvm.experimental.constrained.minnum.f32{{.*}} #2 +float fmin32b(float a, float b) { + return __builtin_fminf(a, b); +} // CHECK: call nsz double @llvm.minnum.f64 // CHECK-STRICT: call nsz double @llvm.experimental.constrained.minnum.f64{{.*}} #2 -float fmin2(double a, double b) { +float fmin64(double a, double b) { return fmin(a, b); } +// CHECK: call nsz double @llvm.minnum.f64 +// CHECK-STRICT: call nsz double @llvm.experimental.constrained.minnum.f64{{.*}} #2 +float fmin64b(double a, double b) { + return __builtin_fmin(a, b); +} // CHECK: call nsz x86_fp80 @llvm.minnum.f80 // CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80{{.*}} #2 -float fmin3(long double a, long double b) { +float fmin80(long double a, long double b) { return fminl(a, b); } +// CHECK: call nsz x86_fp80 @llvm.minnum.f80 +// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80{{.*}} #2 +float fmin80b(long double a, long double b) { + return __builtin_fminl(a, b); +} // CHECK: call nsz float @llvm.maxnum.f32 // CHECK-STRICT: call nsz float @llvm.experimental.constrained.maxnum.f32{{.*}} #2 -float fmax1(float a, float b) { +float fmax32(float a, float b) { return fmaxf(a, b); } +// CHECK: call nsz float @llvm.maxnum.f32 +// CHECK-STRICT: call nsz float @llvm.experimental.constrained.maxnum.f32{{.*}} #2 +float fmax32b(float a, float b) { + return __builtin_fmaxf(a, b); +} // CHECK: call nsz double @llvm.maxnum.f64 // CHECK-STRICT: call nsz double @llvm.experimental.constrained.maxnum.f64{{.*}} #2 -float fmax2(double a, double b) { +float fmax64(double a, double b) { return fmax(a, b); } +// CHECK: call nsz double @llvm.maxnum.f64 +// CHECK-STRICT: call nsz double @llvm.experimental.constrained.maxnum.f64{{.*}} #2 +float fmax64b(double a, double b) { + return __builtin_fmax(a, b); +} // CHECK: call nsz x86_fp80 @llvm.maxnum.f80 // CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80{{.*}} #2 float fmax3(long double a, long double b) { return fmaxl(a, b); } +// CHECK: call nsz x86_fp80 @llvm.maxnum.f80 +// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80{{.*}} #2 +float fmax80b(long double a, long double b) { + return __builtin_fmaxl(a, b); +} //CHECK-STRICT: attributes #2 = { strictfp } >From 055a68180d13365a7657dbc807961248ab5e111d Mon Sep 17 00:00:00 2001 From: YunQiang Su <yunqi...@isrc.iscas.ac.cn> Date: Tue, 22 Oct 2024 11:03:09 +0800 Subject: [PATCH 5/7] test auto vectorize --- clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c | 60 +++++++++++++++++++- 1 file changed, 58 insertions(+), 2 deletions(-) diff --git a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c index afe626d921125..4729fd6e27896 100644 --- a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c +++ b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c @@ -1,5 +1,5 @@ -// RUN: %clang_cc1 -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK -// RUN: %clang_cc1 -ffp-exception-behavior=strict -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK-STRICT +// RUN: %clang_cc1 -vectorize-loops -vectorize-slp -O3 -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK +// RUN: %clang_cc1 -vectorize-loops -vectorize-slp -O3 -ffp-exception-behavior=strict -DENSTRICT=1 -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK-STRICT float fminf (float, float); double fmin (double, double); @@ -18,6 +18,20 @@ float fmin32(float a, float b) { float fmin32b(float a, float b) { return __builtin_fminf(a, b); } +#if !defined(ENSTRICT) +// CHECK: call nsz <4 x float> @llvm.minnum.v4f32(<4 x float> %0, <4 x float> %1) +float *pfmin32(float* a, float* b, float* restrict c) { + for (int i=0; i<4; i++) + c[i] = fminf(a[i], b[i]); + return c; +} +// CHECK: call nsz <4 x float> @llvm.minnum.v4f32(<4 x float> %0, <4 x float> %1) +float *pfmin32b(float* a, float* b, float* restrict c) { + for (int i=0; i<4; i++) + c[i] = __builtin_fminf(a[i], b[i]); + return c; +} +#endif // CHECK: call nsz double @llvm.minnum.f64 // CHECK-STRICT: call nsz double @llvm.experimental.constrained.minnum.f64{{.*}} #2 float fmin64(double a, double b) { @@ -28,6 +42,20 @@ float fmin64(double a, double b) { float fmin64b(double a, double b) { return __builtin_fmin(a, b); } +#if !defined(ENSTRICT) +// CHECK: call nsz <2 x double> @llvm.minnum.v2f64(<2 x double> %0, <2 x double> %1) +double *pfmin64(double* a, double* b, double* restrict c) { + for (int i=0; i<2; i++) + c[i] = fmin(a[i], b[i]); + return c; +} +// CHECK: call nsz <2 x double> @llvm.minnum.v2f64(<2 x double> %0, <2 x double> %1) +double *pfmin64b(double* a, double* b, double* restrict c) { + for (int i=0; i<2; i++) + c[i] = __builtin_fmin(a[i], b[i]); + return c; +} +#endif // CHECK: call nsz x86_fp80 @llvm.minnum.f80 // CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80{{.*}} #2 float fmin80(long double a, long double b) { @@ -48,6 +76,20 @@ float fmax32(float a, float b) { float fmax32b(float a, float b) { return __builtin_fmaxf(a, b); } +#if !defined(ENSTRICT) +// CHECK: call nsz <4 x float> @llvm.maxnum.v4f32(<4 x float> %0, <4 x float> %1) +float *pfmax32(float* a, float* b, float* restrict c) { + for (int i=0; i<4; i++) + c[i] = fmaxf(a[i], b[i]); + return c; +} +// CHECK: call nsz <4 x float> @llvm.maxnum.v4f32(<4 x float> %0, <4 x float> %1) +float *pfmax32b(float* a, float* b, float* restrict c) { + for (int i=0; i<4; i++) + c[i] = __builtin_fmaxf(a[i], b[i]); + return c; +} +#endif // CHECK: call nsz double @llvm.maxnum.f64 // CHECK-STRICT: call nsz double @llvm.experimental.constrained.maxnum.f64{{.*}} #2 float fmax64(double a, double b) { @@ -63,6 +105,20 @@ float fmax64b(double a, double b) { float fmax3(long double a, long double b) { return fmaxl(a, b); } +#if !defined(ENSTRICT) +// CHECK: call nsz <2 x double> @llvm.maxnum.v2f64(<2 x double> %0, <2 x double> %1) +double *pfmax64(double* a, double* b, double* restrict c) { + for (int i=0; i<2; i++) + c[i] = fmax(a[i], b[i]); + return c; +} +// CHECK: call nsz <2 x double> @llvm.maxnum.v2f64(<2 x double> %0, <2 x double> %1) +double *pfmax64b(double* a, double* b, double* restrict c) { + for (int i=0; i<2; i++) + c[i] = __builtin_fmax(a[i], b[i]); + return c; +} +#endif // CHECK: call nsz x86_fp80 @llvm.maxnum.f80 // CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80{{.*}} #2 float fmax80b(long double a, long double b) { >From 93e987657cbd08d035593e1cdb8796295d68b5a8 Mon Sep 17 00:00:00 2001 From: YunQiang Su <yunqi...@isrc.iscas.ac.cn> Date: Wed, 23 Oct 2024 11:10:52 +0800 Subject: [PATCH 6/7] fix test cases --- clang/test/CodeGen/RISCV/math-builtins.c | 24 +++++++++---------- clang/test/CodeGen/builtins.c | 12 +++++----- .../test/CodeGen/constrained-math-builtins.c | 16 ++++++------- clang/test/CodeGen/math-builtins-long.c | 16 ++++++------- clang/test/CodeGenOpenCL/builtins-f16.cl | 4 ++-- .../amdgcn_openmp_device_math_constexpr.cpp | 8 +++---- 6 files changed, 40 insertions(+), 40 deletions(-) diff --git a/clang/test/CodeGen/RISCV/math-builtins.c b/clang/test/CodeGen/RISCV/math-builtins.c index 9630d62f0f482..e4f456c8eabd7 100644 --- a/clang/test/CodeGen/RISCV/math-builtins.c +++ b/clang/test/CodeGen/RISCV/math-builtins.c @@ -134,22 +134,22 @@ long double truncl(long double); // RV32-NEXT: [[TMP44:%.*]] = call fp128 @llvm.floor.f128(fp128 [[TMP43]]) // RV32-NEXT: [[TMP45:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV32-NEXT: [[TMP46:%.*]] = load float, ptr [[FARG_ADDR]], align 4 -// RV32-NEXT: [[TMP47:%.*]] = call float @llvm.maxnum.f32(float [[TMP45]], float [[TMP46]]) +// RV32-NEXT: [[TMP47:%.*]] = call nsz float @llvm.maxnum.f32(float [[TMP45]], float [[TMP46]]) // RV32-NEXT: [[TMP48:%.*]] = load double, ptr [[DARG_ADDR]], align 8 // RV32-NEXT: [[TMP49:%.*]] = load double, ptr [[DARG_ADDR]], align 8 -// RV32-NEXT: [[TMP50:%.*]] = call double @llvm.maxnum.f64(double [[TMP48]], double [[TMP49]]) +// RV32-NEXT: [[TMP50:%.*]] = call nsz double @llvm.maxnum.f64(double [[TMP48]], double [[TMP49]]) // RV32-NEXT: [[TMP51:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 // RV32-NEXT: [[TMP52:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 -// RV32-NEXT: [[TMP53:%.*]] = call fp128 @llvm.maxnum.f128(fp128 [[TMP51]], fp128 [[TMP52]]) +// RV32-NEXT: [[TMP53:%.*]] = call nsz fp128 @llvm.maxnum.f128(fp128 [[TMP51]], fp128 [[TMP52]]) // RV32-NEXT: [[TMP54:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV32-NEXT: [[TMP55:%.*]] = load float, ptr [[FARG_ADDR]], align 4 -// RV32-NEXT: [[TMP56:%.*]] = call float @llvm.minnum.f32(float [[TMP54]], float [[TMP55]]) +// RV32-NEXT: [[TMP56:%.*]] = call nsz float @llvm.minnum.f32(float [[TMP54]], float [[TMP55]]) // RV32-NEXT: [[TMP57:%.*]] = load double, ptr [[DARG_ADDR]], align 8 // RV32-NEXT: [[TMP58:%.*]] = load double, ptr [[DARG_ADDR]], align 8 -// RV32-NEXT: [[TMP59:%.*]] = call double @llvm.minnum.f64(double [[TMP57]], double [[TMP58]]) +// RV32-NEXT: [[TMP59:%.*]] = call nsz double @llvm.minnum.f64(double [[TMP57]], double [[TMP58]]) // RV32-NEXT: [[TMP60:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 // RV32-NEXT: [[TMP61:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 -// RV32-NEXT: [[TMP62:%.*]] = call fp128 @llvm.minnum.f128(fp128 [[TMP60]], fp128 [[TMP61]]) +// RV32-NEXT: [[TMP62:%.*]] = call nsz fp128 @llvm.minnum.f128(fp128 [[TMP60]], fp128 [[TMP61]]) // RV32-NEXT: [[TMP63:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV32-NEXT: [[TMP64:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV32-NEXT: [[FMOD:%.*]] = frem float [[TMP63]], [[TMP64]] @@ -310,22 +310,22 @@ long double truncl(long double); // RV64-NEXT: [[TMP44:%.*]] = call fp128 @llvm.floor.f128(fp128 [[TMP43]]) // RV64-NEXT: [[TMP45:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV64-NEXT: [[TMP46:%.*]] = load float, ptr [[FARG_ADDR]], align 4 -// RV64-NEXT: [[TMP47:%.*]] = call float @llvm.maxnum.f32(float [[TMP45]], float [[TMP46]]) +// RV64-NEXT: [[TMP47:%.*]] = call nsz float @llvm.maxnum.f32(float [[TMP45]], float [[TMP46]]) // RV64-NEXT: [[TMP48:%.*]] = load double, ptr [[DARG_ADDR]], align 8 // RV64-NEXT: [[TMP49:%.*]] = load double, ptr [[DARG_ADDR]], align 8 -// RV64-NEXT: [[TMP50:%.*]] = call double @llvm.maxnum.f64(double [[TMP48]], double [[TMP49]]) +// RV64-NEXT: [[TMP50:%.*]] = call nsz double @llvm.maxnum.f64(double [[TMP48]], double [[TMP49]]) // RV64-NEXT: [[TMP51:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 // RV64-NEXT: [[TMP52:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 -// RV64-NEXT: [[TMP53:%.*]] = call fp128 @llvm.maxnum.f128(fp128 [[TMP51]], fp128 [[TMP52]]) +// RV64-NEXT: [[TMP53:%.*]] = call nsz fp128 @llvm.maxnum.f128(fp128 [[TMP51]], fp128 [[TMP52]]) // RV64-NEXT: [[TMP54:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV64-NEXT: [[TMP55:%.*]] = load float, ptr [[FARG_ADDR]], align 4 -// RV64-NEXT: [[TMP56:%.*]] = call float @llvm.minnum.f32(float [[TMP54]], float [[TMP55]]) +// RV64-NEXT: [[TMP56:%.*]] = call nsz float @llvm.minnum.f32(float [[TMP54]], float [[TMP55]]) // RV64-NEXT: [[TMP57:%.*]] = load double, ptr [[DARG_ADDR]], align 8 // RV64-NEXT: [[TMP58:%.*]] = load double, ptr [[DARG_ADDR]], align 8 -// RV64-NEXT: [[TMP59:%.*]] = call double @llvm.minnum.f64(double [[TMP57]], double [[TMP58]]) +// RV64-NEXT: [[TMP59:%.*]] = call nsz double @llvm.minnum.f64(double [[TMP57]], double [[TMP58]]) // RV64-NEXT: [[TMP60:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 // RV64-NEXT: [[TMP61:%.*]] = load fp128, ptr [[LDARG_ADDR]], align 16 -// RV64-NEXT: [[TMP62:%.*]] = call fp128 @llvm.minnum.f128(fp128 [[TMP60]], fp128 [[TMP61]]) +// RV64-NEXT: [[TMP62:%.*]] = call nsz fp128 @llvm.minnum.f128(fp128 [[TMP60]], fp128 [[TMP61]]) // RV64-NEXT: [[TMP63:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV64-NEXT: [[TMP64:%.*]] = load float, ptr [[FARG_ADDR]], align 4 // RV64-NEXT: [[FMOD:%.*]] = frem float [[TMP63]], [[TMP64]] diff --git a/clang/test/CodeGen/builtins.c b/clang/test/CodeGen/builtins.c index eda6c67fdad00..72e195e82d424 100644 --- a/clang/test/CodeGen/builtins.c +++ b/clang/test/CodeGen/builtins.c @@ -343,22 +343,22 @@ void test_float_builtin_ops(float F, double D, long double LD, int I) { // CHECK: call x86_fp80 @llvm.canonicalize.f80(x86_fp80 resf = __builtin_fminf(F, F); - // CHECK: call float @llvm.minnum.f32 + // CHECK: call nsz float @llvm.minnum.f32 resd = __builtin_fmin(D, D); - // CHECK: call double @llvm.minnum.f64 + // CHECK: call nsz double @llvm.minnum.f64 resld = __builtin_fminl(LD, LD); - // CHECK: call x86_fp80 @llvm.minnum.f80 + // CHECK: call nsz x86_fp80 @llvm.minnum.f80 resf = __builtin_fmaxf(F, F); - // CHECK: call float @llvm.maxnum.f32 + // CHECK: call nsz float @llvm.maxnum.f32 resd = __builtin_fmax(D, D); - // CHECK: call double @llvm.maxnum.f64 + // CHECK: call nsz double @llvm.maxnum.f64 resld = __builtin_fmaxl(LD, LD); - // CHECK: call x86_fp80 @llvm.maxnum.f80 + // CHECK: call nsz x86_fp80 @llvm.maxnum.f80 resf = __builtin_fminimum_numf(F, F); // CHECK: call float @llvm.minimumnum.f32 diff --git a/clang/test/CodeGen/constrained-math-builtins.c b/clang/test/CodeGen/constrained-math-builtins.c index 68b9e75283c54..ae07cce05b307 100644 --- a/clang/test/CodeGen/constrained-math-builtins.c +++ b/clang/test/CodeGen/constrained-math-builtins.c @@ -123,17 +123,17 @@ __builtin_atan2(f,f); __builtin_atan2f(f,f); __builtin_atan2l(f,f); __builtin_fmax(f,f); __builtin_fmaxf(f,f); __builtin_fmaxl(f,f); __builtin_fmaxf128(f,f); -// CHECK: call double @llvm.experimental.constrained.maxnum.f64(double %{{.*}}, double %{{.*}}, metadata !"fpexcept.strict") -// CHECK: call float @llvm.experimental.constrained.maxnum.f32(float %{{.*}}, float %{{.*}}, metadata !"fpexcept.strict") -// CHECK: call x86_fp80 @llvm.experimental.constrained.maxnum.f80(x86_fp80 %{{.*}}, x86_fp80 %{{.*}}, metadata !"fpexcept.strict") -// CHECK: call fp128 @llvm.experimental.constrained.maxnum.f128(fp128 %{{.*}}, fp128 %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz double @llvm.experimental.constrained.maxnum.f64(double %{{.*}}, double %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz float @llvm.experimental.constrained.maxnum.f32(float %{{.*}}, float %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80(x86_fp80 %{{.*}}, x86_fp80 %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz fp128 @llvm.experimental.constrained.maxnum.f128(fp128 %{{.*}}, fp128 %{{.*}}, metadata !"fpexcept.strict") __builtin_fmin(f,f); __builtin_fminf(f,f); __builtin_fminl(f,f); __builtin_fminf128(f,f); -// CHECK: call double @llvm.experimental.constrained.minnum.f64(double %{{.*}}, double %{{.*}}, metadata !"fpexcept.strict") -// CHECK: call float @llvm.experimental.constrained.minnum.f32(float %{{.*}}, float %{{.*}}, metadata !"fpexcept.strict") -// CHECK: call x86_fp80 @llvm.experimental.constrained.minnum.f80(x86_fp80 %{{.*}}, x86_fp80 %{{.*}}, metadata !"fpexcept.strict") -// CHECK: call fp128 @llvm.experimental.constrained.minnum.f128(fp128 %{{.*}}, fp128 %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz double @llvm.experimental.constrained.minnum.f64(double %{{.*}}, double %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz float @llvm.experimental.constrained.minnum.f32(float %{{.*}}, float %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80(x86_fp80 %{{.*}}, x86_fp80 %{{.*}}, metadata !"fpexcept.strict") +// CHECK: call nsz fp128 @llvm.experimental.constrained.minnum.f128(fp128 %{{.*}}, fp128 %{{.*}}, metadata !"fpexcept.strict") __builtin_llrint(f); __builtin_llrintf(f); __builtin_llrintl(f); __builtin_llrintf128(f); diff --git a/clang/test/CodeGen/math-builtins-long.c b/clang/test/CodeGen/math-builtins-long.c index 183349e0f0173..f1e4988e37871 100644 --- a/clang/test/CodeGen/math-builtins-long.c +++ b/clang/test/CodeGen/math-builtins-long.c @@ -148,16 +148,16 @@ void foo(long double f, long double *l, int *i, const char *c) { // PPCF128: call fp128 @llvm.floor.f128(fp128 %{{.+}}) __builtin_floorl(f); - // F80: call x86_fp80 @llvm.maxnum.f80(x86_fp80 %{{.+}}, x86_fp80 %{{.+}}) - // PPC: call ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %{{.+}}, ppc_fp128 %{{.+}}) - // X86F128: call fp128 @llvm.maxnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) - // PPCF128: call fp128 @llvm.maxnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) + // F80: call nsz x86_fp80 @llvm.maxnum.f80(x86_fp80 %{{.+}}, x86_fp80 %{{.+}}) + // PPC: call nsz ppc_fp128 @llvm.maxnum.ppcf128(ppc_fp128 %{{.+}}, ppc_fp128 %{{.+}}) + // X86F128: call nsz fp128 @llvm.maxnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) + // PPCF128: call nsz fp128 @llvm.maxnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) __builtin_fmaxl(f,f); - // F80: call x86_fp80 @llvm.minnum.f80(x86_fp80 %{{.+}}, x86_fp80 %{{.+}}) - // PPC: call ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %{{.+}}, ppc_fp128 %{{.+}}) - // X86F128: call fp128 @llvm.minnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) - // PPCF128: call fp128 @llvm.minnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) + // F80: call nsz x86_fp80 @llvm.minnum.f80(x86_fp80 %{{.+}}, x86_fp80 %{{.+}}) + // PPC: call nsz ppc_fp128 @llvm.minnum.ppcf128(ppc_fp128 %{{.+}}, ppc_fp128 %{{.+}}) + // X86F128: call nsz fp128 @llvm.minnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) + // PPCF128: call nsz fp128 @llvm.minnum.f128(fp128 %{{.+}}, fp128 %{{.+}}) __builtin_fminl(f,f); // F80: call x86_fp80 @llvm.nearbyint.f80(x86_fp80 %{{.+}}) diff --git a/clang/test/CodeGenOpenCL/builtins-f16.cl b/clang/test/CodeGenOpenCL/builtins-f16.cl index e8b62fe0830cd..f30ed0a1944ff 100644 --- a/clang/test/CodeGenOpenCL/builtins-f16.cl +++ b/clang/test/CodeGenOpenCL/builtins-f16.cl @@ -48,10 +48,10 @@ void test_half_builtins(half h0, half h1, half h2, int i0) { // CHECK: call half @llvm.fma.f16(half %h0, half %h1, half %h2) res = __builtin_fmaf16(h0, h1 ,h2); - // CHECK: call half @llvm.maxnum.f16(half %h0, half %h1) + // CHECK: call nsz half @llvm.maxnum.f16(half %h0, half %h1) res = __builtin_fmaxf16(h0, h1); - // CHECK: call half @llvm.minnum.f16(half %h0, half %h1) + // CHECK: call nsz half @llvm.minnum.f16(half %h0, half %h1) res = __builtin_fminf16(h0, h1); // CHECK: frem half %h0, %h1 diff --git a/clang/test/Headers/amdgcn_openmp_device_math_constexpr.cpp b/clang/test/Headers/amdgcn_openmp_device_math_constexpr.cpp index 0fdc02edc1508..855333132c4de 100644 --- a/clang/test/Headers/amdgcn_openmp_device_math_constexpr.cpp +++ b/clang/test/Headers/amdgcn_openmp_device_math_constexpr.cpp @@ -208,7 +208,7 @@ const float constexpr_fmaxf_f32 = fmaxf(2.0f, -4.0f); // CHECK-NEXT: store float -4.000000e+00, ptr [[__Y_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I]], align 4 -// CHECK-NEXT: [[TMP2:%.*]] = call noundef float @llvm.minnum.f32(float [[TMP0]], float [[TMP1]]) +// CHECK-NEXT: [[TMP2:%.*]] = call nsz noundef float @llvm.minnum.f32(float [[TMP0]], float [[TMP1]]) // CHECK-NEXT: store float [[TMP2]], ptr addrspacecast (ptr addrspace(1) @_ZL17constexpr_min_f32 to ptr), align 4 // CHECK-NEXT: ret void // @@ -226,7 +226,7 @@ const float constexpr_fmaxf_f32 = fmaxf(2.0f, -4.0f); // CHECK-NEXT: store float -4.000000e+00, ptr [[__Y_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I]], align 4 -// CHECK-NEXT: [[TMP2:%.*]] = call noundef float @llvm.maxnum.f32(float [[TMP0]], float [[TMP1]]) +// CHECK-NEXT: [[TMP2:%.*]] = call nsz noundef float @llvm.maxnum.f32(float [[TMP0]], float [[TMP1]]) // CHECK-NEXT: store float [[TMP2]], ptr addrspacecast (ptr addrspace(1) @_ZL17constexpr_max_f32 to ptr), align 4 // CHECK-NEXT: ret void // @@ -260,7 +260,7 @@ const float constexpr_fmaxf_f32 = fmaxf(2.0f, -4.0f); // CHECK-NEXT: store float -4.000000e+00, ptr [[__Y_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I]], align 4 -// CHECK-NEXT: [[TMP2:%.*]] = call noundef float @llvm.minnum.f32(float [[TMP0]], float [[TMP1]]) +// CHECK-NEXT: [[TMP2:%.*]] = call nsz noundef float @llvm.minnum.f32(float [[TMP0]], float [[TMP1]]) // CHECK-NEXT: store float [[TMP2]], ptr addrspacecast (ptr addrspace(1) @_ZL19constexpr_fminf_f32 to ptr), align 4 // CHECK-NEXT: ret void // @@ -278,7 +278,7 @@ const float constexpr_fmaxf_f32 = fmaxf(2.0f, -4.0f); // CHECK-NEXT: store float -4.000000e+00, ptr [[__Y_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP0:%.*]] = load float, ptr [[__X_ADDR_ASCAST_I]], align 4 // CHECK-NEXT: [[TMP1:%.*]] = load float, ptr [[__Y_ADDR_ASCAST_I]], align 4 -// CHECK-NEXT: [[TMP2:%.*]] = call noundef float @llvm.maxnum.f32(float [[TMP0]], float [[TMP1]]) +// CHECK-NEXT: [[TMP2:%.*]] = call nsz noundef float @llvm.maxnum.f32(float [[TMP0]], float [[TMP1]]) // CHECK-NEXT: store float [[TMP2]], ptr addrspacecast (ptr addrspace(1) @_ZL19constexpr_fmaxf_f32 to ptr), align 4 // CHECK-NEXT: ret void // >From a3f35676d6bc929d6065e44b65f6aaa01087ca7b Mon Sep 17 00:00:00 2001 From: YunQiang Su <s...@debian.org> Date: Thu, 27 Feb 2025 18:05:43 +0800 Subject: [PATCH 7/7] update testcase --- clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c | 577 ++++++++++++++++--- 1 file changed, 504 insertions(+), 73 deletions(-) diff --git a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c index 4729fd6e27896..ea07e9bc2d416 100644 --- a/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c +++ b/clang/test/CodeGen/fmaxnum_fminnum_use_nsz.c @@ -1,5 +1,14 @@ -// RUN: %clang_cc1 -vectorize-loops -vectorize-slp -O3 -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK -// RUN: %clang_cc1 -vectorize-loops -vectorize-slp -O3 -ffp-exception-behavior=strict -DENSTRICT=1 -triple x86_64 %s -emit-llvm -o - 2>&1 | FileCheck %s --check-prefix=CHECK-STRICT +// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5 +// RUN: %clang_cc1 -O3 -triple x86_64 %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK +// RUN: %clang_cc1 -fno-fast-math -O3 -triple x86_64 %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK +// +// RUN: %clang_cc1 -fsigned-zeros -O3 -triple x86_64 %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK +// RUN: %clang_cc1 -menable-no-nans -O3 -triple x86_64 %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK-NO-NANS +// RUN: %clang_cc1 -menable-no-infs -O3 -triple x86_64 %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK-NO-INFS +/// FIXME: -ffinite-math-only +// RUN: %clang_cc1 -ffast-math -O3 -triple x86_64 %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK-FAST +// +// RUN: %clang_cc1 -O3 -ffp-exception-behavior=strict -DENSTRICT=1 -triple x86_64 %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK-STRICT float fminf (float, float); double fmin (double, double); @@ -8,121 +17,543 @@ float fmaxf (float, float); double fmax (double, double); long double fmaxl (long double, long double); -// CHECK: call nsz float @llvm.minnum.f32 -// CHECK-STRICT: call nsz float @llvm.experimental.constrained.minnum.f32{{.*}} #2 +typedef float float4 __attribute__((ext_vector_type(4))); +typedef double double2 __attribute__((ext_vector_type(2))); + +// CHECK-LABEL: define dso_local noundef float @fmin32( +// CHECK-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-NEXT: ret float [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmin32( +// CHECK-NO-NANS-SAME: float noundef nofpclass(nan) [[A:%.*]], float noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-NO-NANS-NEXT: ret float [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmin32( +// CHECK-NO-INFS-SAME: float noundef nofpclass(inf) [[A:%.*]], float noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-NO-INFS-NEXT: ret float [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmin32( +// CHECK-FAST-SAME: float noundef nofpclass(nan inf) [[A:%.*]], float noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-FAST-NEXT: ret float [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmin32( +// CHECK-STRICT-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.experimental.constrained.minnum.f32(float [[A]], float [[B]], metadata !"fpexcept.strict") #[[ATTR2:[0-9]+]] +// CHECK-STRICT-NEXT: ret float [[TMP0]] +// float fmin32(float a, float b) { return fminf(a, b); } -// CHECK: call nsz float @llvm.minnum.f32 -// CHECK-STRICT: call nsz float @llvm.experimental.constrained.minnum.f32{{.*}} #2 +// CHECK-LABEL: define dso_local noundef float @fmin32b( +// CHECK-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-NEXT: ret float [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmin32b( +// CHECK-NO-NANS-SAME: float noundef nofpclass(nan) [[A:%.*]], float noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-NO-NANS-NEXT: ret float [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmin32b( +// CHECK-NO-INFS-SAME: float noundef nofpclass(inf) [[A:%.*]], float noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-NO-INFS-NEXT: ret float [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmin32b( +// CHECK-FAST-SAME: float noundef nofpclass(nan inf) [[A:%.*]], float noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.minnum.f32(float [[A]], float [[B]]) +// CHECK-FAST-NEXT: ret float [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmin32b( +// CHECK-STRICT-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.experimental.constrained.minnum.f32(float [[A]], float [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret float [[TMP0]] +// float fmin32b(float a, float b) { return __builtin_fminf(a, b); } #if !defined(ENSTRICT) -// CHECK: call nsz <4 x float> @llvm.minnum.v4f32(<4 x float> %0, <4 x float> %1) -float *pfmin32(float* a, float* b, float* restrict c) { - for (int i=0; i<4; i++) - c[i] = fminf(a[i], b[i]); - return c; -} -// CHECK: call nsz <4 x float> @llvm.minnum.v4f32(<4 x float> %0, <4 x float> %1) -float *pfmin32b(float* a, float* b, float* restrict c) { - for (int i=0; i<4; i++) - c[i] = __builtin_fminf(a[i], b[i]); +// CHECK-LABEL: define dso_local noundef <4 x float> @pfmin32( +// CHECK-SAME: <4 x float> noundef [[A:%.*]], <4 x float> noundef [[B:%.*]], <4 x float> noundef [[C:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[ELT_MIN:%.*]] = tail call <4 x float> @llvm.minnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-NEXT: ret <4 x float> [[ELT_MIN]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) <4 x float> @pfmin32( +// CHECK-NO-NANS-SAME: <4 x float> noundef nofpclass(nan) [[A:%.*]], <4 x float> noundef nofpclass(nan) [[B:%.*]], <4 x float> noundef nofpclass(nan) [[C:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[ELT_MIN:%.*]] = tail call nnan <4 x float> @llvm.minnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-NO-NANS-NEXT: ret <4 x float> [[ELT_MIN]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) <4 x float> @pfmin32( +// CHECK-NO-INFS-SAME: <4 x float> noundef nofpclass(inf) [[A:%.*]], <4 x float> noundef nofpclass(inf) [[B:%.*]], <4 x float> noundef nofpclass(inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[ELT_MIN:%.*]] = tail call ninf <4 x float> @llvm.minnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-NO-INFS-NEXT: ret <4 x float> [[ELT_MIN]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) <4 x float> @pfmin32( +// CHECK-FAST-SAME: <4 x float> noundef nofpclass(nan inf) [[A:%.*]], <4 x float> noundef nofpclass(nan inf) [[B:%.*]], <4 x float> noundef nofpclass(nan inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2:[0-9]+]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[ELT_MIN:%.*]] = tail call reassoc nnan ninf nsz arcp afn <4 x float> @llvm.minnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-FAST-NEXT: ret <4 x float> [[ELT_MIN]] +// +float4 pfmin32(float4 a, float4 b, float4 c) { + c = __builtin_elementwise_min(a, b); return c; } #endif -// CHECK: call nsz double @llvm.minnum.f64 -// CHECK-STRICT: call nsz double @llvm.experimental.constrained.minnum.f64{{.*}} #2 +// CHECK-LABEL: define dso_local noundef float @fmin64( +// CHECK-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-NEXT: [[CONV:%.*]] = fptrunc double [[TMP0]] to float +// CHECK-NEXT: ret float [[CONV]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmin64( +// CHECK-NO-NANS-SAME: double noundef nofpclass(nan) [[A:%.*]], double noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-NO-NANS-NEXT: [[CONV:%.*]] = fptrunc nnan double [[TMP0]] to float +// CHECK-NO-NANS-NEXT: ret float [[CONV]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmin64( +// CHECK-NO-INFS-SAME: double noundef nofpclass(inf) [[A:%.*]], double noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-NO-INFS-NEXT: [[CONV:%.*]] = fptrunc ninf double [[TMP0]] to float +// CHECK-NO-INFS-NEXT: ret float [[CONV]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmin64( +// CHECK-FAST-SAME: double noundef nofpclass(nan inf) [[A:%.*]], double noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-FAST-NEXT: [[CONV:%.*]] = fptrunc reassoc nnan ninf nsz arcp afn double [[TMP0]] to float +// CHECK-FAST-NEXT: ret float [[CONV]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmin64( +// CHECK-STRICT-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.experimental.constrained.minnum.f64(double [[A]], double [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: [[CONV:%.*]] = tail call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP0]], metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret float [[CONV]] +// float fmin64(double a, double b) { return fmin(a, b); } -// CHECK: call nsz double @llvm.minnum.f64 -// CHECK-STRICT: call nsz double @llvm.experimental.constrained.minnum.f64{{.*}} #2 +// CHECK-LABEL: define dso_local noundef float @fmin64b( +// CHECK-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-NEXT: [[CONV:%.*]] = fptrunc double [[TMP0]] to float +// CHECK-NEXT: ret float [[CONV]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmin64b( +// CHECK-NO-NANS-SAME: double noundef nofpclass(nan) [[A:%.*]], double noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-NO-NANS-NEXT: [[CONV:%.*]] = fptrunc nnan double [[TMP0]] to float +// CHECK-NO-NANS-NEXT: ret float [[CONV]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmin64b( +// CHECK-NO-INFS-SAME: double noundef nofpclass(inf) [[A:%.*]], double noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-NO-INFS-NEXT: [[CONV:%.*]] = fptrunc ninf double [[TMP0]] to float +// CHECK-NO-INFS-NEXT: ret float [[CONV]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmin64b( +// CHECK-FAST-SAME: double noundef nofpclass(nan inf) [[A:%.*]], double noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn double @llvm.minnum.f64(double [[A]], double [[B]]) +// CHECK-FAST-NEXT: [[CONV:%.*]] = fptrunc reassoc nnan ninf nsz arcp afn double [[TMP0]] to float +// CHECK-FAST-NEXT: ret float [[CONV]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmin64b( +// CHECK-STRICT-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.experimental.constrained.minnum.f64(double [[A]], double [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: [[CONV:%.*]] = tail call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP0]], metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret float [[CONV]] +// float fmin64b(double a, double b) { return __builtin_fmin(a, b); } #if !defined(ENSTRICT) -// CHECK: call nsz <2 x double> @llvm.minnum.v2f64(<2 x double> %0, <2 x double> %1) -double *pfmin64(double* a, double* b, double* restrict c) { - for (int i=0; i<2; i++) - c[i] = fmin(a[i], b[i]); - return c; -} -// CHECK: call nsz <2 x double> @llvm.minnum.v2f64(<2 x double> %0, <2 x double> %1) -double *pfmin64b(double* a, double* b, double* restrict c) { - for (int i=0; i<2; i++) - c[i] = __builtin_fmin(a[i], b[i]); +// CHECK-LABEL: define dso_local noundef <2 x double> @pfmin64( +// CHECK-SAME: <2 x double> noundef [[A:%.*]], <2 x double> noundef [[B:%.*]], <2 x double> noundef [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[ELT_MIN:%.*]] = tail call <2 x double> @llvm.minnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-NEXT: ret <2 x double> [[ELT_MIN]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) <2 x double> @pfmin64( +// CHECK-NO-NANS-SAME: <2 x double> noundef nofpclass(nan) [[A:%.*]], <2 x double> noundef nofpclass(nan) [[B:%.*]], <2 x double> noundef nofpclass(nan) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[ELT_MIN:%.*]] = tail call nnan <2 x double> @llvm.minnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-NO-NANS-NEXT: ret <2 x double> [[ELT_MIN]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) <2 x double> @pfmin64( +// CHECK-NO-INFS-SAME: <2 x double> noundef nofpclass(inf) [[A:%.*]], <2 x double> noundef nofpclass(inf) [[B:%.*]], <2 x double> noundef nofpclass(inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[ELT_MIN:%.*]] = tail call ninf <2 x double> @llvm.minnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-NO-INFS-NEXT: ret <2 x double> [[ELT_MIN]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) <2 x double> @pfmin64( +// CHECK-FAST-SAME: <2 x double> noundef nofpclass(nan inf) [[A:%.*]], <2 x double> noundef nofpclass(nan inf) [[B:%.*]], <2 x double> noundef nofpclass(nan inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[ELT_MIN:%.*]] = tail call reassoc nnan ninf nsz arcp afn <2 x double> @llvm.minnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-FAST-NEXT: ret <2 x double> [[ELT_MIN]] +// +double2 pfmin64(double2 a, double2 b, double2 c) { + c = __builtin_elementwise_min(a, b); return c; } #endif -// CHECK: call nsz x86_fp80 @llvm.minnum.f80 -// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80{{.*}} #2 -float fmin80(long double a, long double b) { +// CHECK-LABEL: define dso_local noundef x86_fp80 @fmin80( +// CHECK-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) x86_fp80 @fmin80( +// CHECK-NO-NANS-SAME: x86_fp80 noundef nofpclass(nan) [[A:%.*]], x86_fp80 noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-NANS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) x86_fp80 @fmin80( +// CHECK-NO-INFS-SAME: x86_fp80 noundef nofpclass(inf) [[A:%.*]], x86_fp80 noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-INFS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) x86_fp80 @fmin80( +// CHECK-FAST-SAME: x86_fp80 noundef nofpclass(nan inf) [[A:%.*]], x86_fp80 noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-FAST-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local x86_fp80 @fmin80( +// CHECK-STRICT-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret x86_fp80 [[TMP0]] +// +long double fmin80(long double a, long double b) { return fminl(a, b); } -// CHECK: call nsz x86_fp80 @llvm.minnum.f80 -// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80{{.*}} #2 -float fmin80b(long double a, long double b) { +// CHECK-LABEL: define dso_local noundef x86_fp80 @fmin80b( +// CHECK-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) x86_fp80 @fmin80b( +// CHECK-NO-NANS-SAME: x86_fp80 noundef nofpclass(nan) [[A:%.*]], x86_fp80 noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-NANS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) x86_fp80 @fmin80b( +// CHECK-NO-INFS-SAME: x86_fp80 noundef nofpclass(inf) [[A:%.*]], x86_fp80 noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-INFS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) x86_fp80 @fmin80b( +// CHECK-FAST-SAME: x86_fp80 noundef nofpclass(nan inf) [[A:%.*]], x86_fp80 noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn x86_fp80 @llvm.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-FAST-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local x86_fp80 @fmin80b( +// CHECK-STRICT-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.experimental.constrained.minnum.f80(x86_fp80 [[A]], x86_fp80 [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret x86_fp80 [[TMP0]] +// +long double fmin80b(long double a, long double b) { return __builtin_fminl(a, b); } -// CHECK: call nsz float @llvm.maxnum.f32 -// CHECK-STRICT: call nsz float @llvm.experimental.constrained.maxnum.f32{{.*}} #2 +// CHECK-LABEL: define dso_local noundef float @fmax32( +// CHECK-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-NEXT: ret float [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmax32( +// CHECK-NO-NANS-SAME: float noundef nofpclass(nan) [[A:%.*]], float noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-NO-NANS-NEXT: ret float [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmax32( +// CHECK-NO-INFS-SAME: float noundef nofpclass(inf) [[A:%.*]], float noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-NO-INFS-NEXT: ret float [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmax32( +// CHECK-FAST-SAME: float noundef nofpclass(nan inf) [[A:%.*]], float noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-FAST-NEXT: ret float [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmax32( +// CHECK-STRICT-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.experimental.constrained.maxnum.f32(float [[A]], float [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret float [[TMP0]] +// float fmax32(float a, float b) { return fmaxf(a, b); } -// CHECK: call nsz float @llvm.maxnum.f32 -// CHECK-STRICT: call nsz float @llvm.experimental.constrained.maxnum.f32{{.*}} #2 +// CHECK-LABEL: define dso_local noundef float @fmax32b( +// CHECK-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-NEXT: ret float [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmax32b( +// CHECK-NO-NANS-SAME: float noundef nofpclass(nan) [[A:%.*]], float noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-NO-NANS-NEXT: ret float [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmax32b( +// CHECK-NO-INFS-SAME: float noundef nofpclass(inf) [[A:%.*]], float noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-NO-INFS-NEXT: ret float [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmax32b( +// CHECK-FAST-SAME: float noundef nofpclass(nan inf) [[A:%.*]], float noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn float @llvm.maxnum.f32(float [[A]], float [[B]]) +// CHECK-FAST-NEXT: ret float [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmax32b( +// CHECK-STRICT-SAME: float noundef [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz float @llvm.experimental.constrained.maxnum.f32(float [[A]], float [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret float [[TMP0]] +// float fmax32b(float a, float b) { return __builtin_fmaxf(a, b); } #if !defined(ENSTRICT) -// CHECK: call nsz <4 x float> @llvm.maxnum.v4f32(<4 x float> %0, <4 x float> %1) -float *pfmax32(float* a, float* b, float* restrict c) { - for (int i=0; i<4; i++) - c[i] = fmaxf(a[i], b[i]); - return c; -} -// CHECK: call nsz <4 x float> @llvm.maxnum.v4f32(<4 x float> %0, <4 x float> %1) -float *pfmax32b(float* a, float* b, float* restrict c) { - for (int i=0; i<4; i++) - c[i] = __builtin_fmaxf(a[i], b[i]); +// CHECK-LABEL: define dso_local noundef <4 x float> @pfmax32( +// CHECK-SAME: <4 x float> noundef [[A:%.*]], <4 x float> noundef [[B:%.*]], <4 x float> noundef [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[ELT_MAX:%.*]] = tail call <4 x float> @llvm.maxnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-NEXT: ret <4 x float> [[ELT_MAX]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) <4 x float> @pfmax32( +// CHECK-NO-NANS-SAME: <4 x float> noundef nofpclass(nan) [[A:%.*]], <4 x float> noundef nofpclass(nan) [[B:%.*]], <4 x float> noundef nofpclass(nan) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[ELT_MAX:%.*]] = tail call nnan <4 x float> @llvm.maxnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-NO-NANS-NEXT: ret <4 x float> [[ELT_MAX]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) <4 x float> @pfmax32( +// CHECK-NO-INFS-SAME: <4 x float> noundef nofpclass(inf) [[A:%.*]], <4 x float> noundef nofpclass(inf) [[B:%.*]], <4 x float> noundef nofpclass(inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[ELT_MAX:%.*]] = tail call ninf <4 x float> @llvm.maxnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-NO-INFS-NEXT: ret <4 x float> [[ELT_MAX]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) <4 x float> @pfmax32( +// CHECK-FAST-SAME: <4 x float> noundef nofpclass(nan inf) [[A:%.*]], <4 x float> noundef nofpclass(nan inf) [[B:%.*]], <4 x float> noundef nofpclass(nan inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[ELT_MAX:%.*]] = tail call reassoc nnan ninf nsz arcp afn <4 x float> @llvm.maxnum.v4f32(<4 x float> [[A]], <4 x float> [[B]]) +// CHECK-FAST-NEXT: ret <4 x float> [[ELT_MAX]] +// +float4 pfmax32(float4 a, float4 b, float4 c) { + c = __builtin_elementwise_max(a, b); return c; } #endif -// CHECK: call nsz double @llvm.maxnum.f64 -// CHECK-STRICT: call nsz double @llvm.experimental.constrained.maxnum.f64{{.*}} #2 +// CHECK-LABEL: define dso_local noundef float @fmax64( +// CHECK-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-NEXT: [[CONV:%.*]] = fptrunc double [[TMP0]] to float +// CHECK-NEXT: ret float [[CONV]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmax64( +// CHECK-NO-NANS-SAME: double noundef nofpclass(nan) [[A:%.*]], double noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-NO-NANS-NEXT: [[CONV:%.*]] = fptrunc nnan double [[TMP0]] to float +// CHECK-NO-NANS-NEXT: ret float [[CONV]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmax64( +// CHECK-NO-INFS-SAME: double noundef nofpclass(inf) [[A:%.*]], double noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-NO-INFS-NEXT: [[CONV:%.*]] = fptrunc ninf double [[TMP0]] to float +// CHECK-NO-INFS-NEXT: ret float [[CONV]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmax64( +// CHECK-FAST-SAME: double noundef nofpclass(nan inf) [[A:%.*]], double noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-FAST-NEXT: [[CONV:%.*]] = fptrunc reassoc nnan ninf nsz arcp afn double [[TMP0]] to float +// CHECK-FAST-NEXT: ret float [[CONV]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmax64( +// CHECK-STRICT-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.experimental.constrained.maxnum.f64(double [[A]], double [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: [[CONV:%.*]] = tail call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP0]], metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret float [[CONV]] +// float fmax64(double a, double b) { return fmax(a, b); } -// CHECK: call nsz double @llvm.maxnum.f64 -// CHECK-STRICT: call nsz double @llvm.experimental.constrained.maxnum.f64{{.*}} #2 +// CHECK-LABEL: define dso_local noundef float @fmax64b( +// CHECK-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-NEXT: [[CONV:%.*]] = fptrunc double [[TMP0]] to float +// CHECK-NEXT: ret float [[CONV]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) float @fmax64b( +// CHECK-NO-NANS-SAME: double noundef nofpclass(nan) [[A:%.*]], double noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-NO-NANS-NEXT: [[CONV:%.*]] = fptrunc nnan double [[TMP0]] to float +// CHECK-NO-NANS-NEXT: ret float [[CONV]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) float @fmax64b( +// CHECK-NO-INFS-SAME: double noundef nofpclass(inf) [[A:%.*]], double noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-NO-INFS-NEXT: [[CONV:%.*]] = fptrunc ninf double [[TMP0]] to float +// CHECK-NO-INFS-NEXT: ret float [[CONV]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) float @fmax64b( +// CHECK-FAST-SAME: double noundef nofpclass(nan inf) [[A:%.*]], double noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn double @llvm.maxnum.f64(double [[A]], double [[B]]) +// CHECK-FAST-NEXT: [[CONV:%.*]] = fptrunc reassoc nnan ninf nsz arcp afn double [[TMP0]] to float +// CHECK-FAST-NEXT: ret float [[CONV]] +// +// CHECK-STRICT-LABEL: define dso_local float @fmax64b( +// CHECK-STRICT-SAME: double noundef [[A:%.*]], double noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz double @llvm.experimental.constrained.maxnum.f64(double [[A]], double [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: [[CONV:%.*]] = tail call float @llvm.experimental.constrained.fptrunc.f32.f64(double [[TMP0]], metadata !"round.tonearest", metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret float [[CONV]] +// float fmax64b(double a, double b) { return __builtin_fmax(a, b); } -// CHECK: call nsz x86_fp80 @llvm.maxnum.f80 -// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80{{.*}} #2 -float fmax3(long double a, long double b) { - return fmaxl(a, b); -} #if !defined(ENSTRICT) -// CHECK: call nsz <2 x double> @llvm.maxnum.v2f64(<2 x double> %0, <2 x double> %1) -double *pfmax64(double* a, double* b, double* restrict c) { - for (int i=0; i<2; i++) - c[i] = fmax(a[i], b[i]); - return c; -} -// CHECK: call nsz <2 x double> @llvm.maxnum.v2f64(<2 x double> %0, <2 x double> %1) -double *pfmax64b(double* a, double* b, double* restrict c) { - for (int i=0; i<2; i++) - c[i] = __builtin_fmax(a[i], b[i]); +// CHECK-LABEL: define dso_local noundef <2 x double> @pfmax64( +// CHECK-SAME: <2 x double> noundef [[A:%.*]], <2 x double> noundef [[B:%.*]], <2 x double> noundef [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[ELT_MAX:%.*]] = tail call <2 x double> @llvm.maxnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-NEXT: ret <2 x double> [[ELT_MAX]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) <2 x double> @pfmax64( +// CHECK-NO-NANS-SAME: <2 x double> noundef nofpclass(nan) [[A:%.*]], <2 x double> noundef nofpclass(nan) [[B:%.*]], <2 x double> noundef nofpclass(nan) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[ELT_MAX:%.*]] = tail call nnan <2 x double> @llvm.maxnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-NO-NANS-NEXT: ret <2 x double> [[ELT_MAX]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) <2 x double> @pfmax64( +// CHECK-NO-INFS-SAME: <2 x double> noundef nofpclass(inf) [[A:%.*]], <2 x double> noundef nofpclass(inf) [[B:%.*]], <2 x double> noundef nofpclass(inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[ELT_MAX:%.*]] = tail call ninf <2 x double> @llvm.maxnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-NO-INFS-NEXT: ret <2 x double> [[ELT_MAX]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) <2 x double> @pfmax64( +// CHECK-FAST-SAME: <2 x double> noundef nofpclass(nan inf) [[A:%.*]], <2 x double> noundef nofpclass(nan inf) [[B:%.*]], <2 x double> noundef nofpclass(nan inf) [[C:%.*]]) local_unnamed_addr #[[ATTR2]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[ELT_MAX:%.*]] = tail call reassoc nnan ninf nsz arcp afn <2 x double> @llvm.maxnum.v2f64(<2 x double> [[A]], <2 x double> [[B]]) +// CHECK-FAST-NEXT: ret <2 x double> [[ELT_MAX]] +// +double2 pfmax64(double2 a, double2 b, double2 c) { + c = __builtin_elementwise_max(a, b); return c; } #endif -// CHECK: call nsz x86_fp80 @llvm.maxnum.f80 -// CHECK-STRICT: call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80{{.*}} #2 -float fmax80b(long double a, long double b) { +// CHECK-LABEL: define dso_local noundef x86_fp80 @fmax80( +// CHECK-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) x86_fp80 @fmax80( +// CHECK-NO-NANS-SAME: x86_fp80 noundef nofpclass(nan) [[A:%.*]], x86_fp80 noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-NANS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) x86_fp80 @fmax80( +// CHECK-NO-INFS-SAME: x86_fp80 noundef nofpclass(inf) [[A:%.*]], x86_fp80 noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-INFS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) x86_fp80 @fmax80( +// CHECK-FAST-SAME: x86_fp80 noundef nofpclass(nan inf) [[A:%.*]], x86_fp80 noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-FAST-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local x86_fp80 @fmax80( +// CHECK-STRICT-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret x86_fp80 [[TMP0]] +// +long double fmax80(long double a, long double b) { + return fmaxl(a, b); +} +// CHECK-LABEL: define dso_local noundef x86_fp80 @fmax80b( +// CHECK-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NEXT: [[ENTRY:.*:]] +// CHECK-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-NANS-LABEL: define dso_local nofpclass(nan) x86_fp80 @fmax80b( +// CHECK-NO-NANS-SAME: x86_fp80 noundef nofpclass(nan) [[A:%.*]], x86_fp80 noundef nofpclass(nan) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-NANS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-NANS-NEXT: [[TMP0:%.*]] = tail call nnan nsz x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-NANS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-NO-INFS-LABEL: define dso_local nofpclass(inf) x86_fp80 @fmax80b( +// CHECK-NO-INFS-SAME: x86_fp80 noundef nofpclass(inf) [[A:%.*]], x86_fp80 noundef nofpclass(inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-NO-INFS-NEXT: [[ENTRY:.*:]] +// CHECK-NO-INFS-NEXT: [[TMP0:%.*]] = tail call ninf nsz x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-NO-INFS-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-FAST-LABEL: define dso_local nofpclass(nan inf) x86_fp80 @fmax80b( +// CHECK-FAST-SAME: x86_fp80 noundef nofpclass(nan inf) [[A:%.*]], x86_fp80 noundef nofpclass(nan inf) [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-FAST-NEXT: [[ENTRY:.*:]] +// CHECK-FAST-NEXT: [[TMP0:%.*]] = tail call reassoc nnan ninf nsz arcp afn x86_fp80 @llvm.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]]) +// CHECK-FAST-NEXT: ret x86_fp80 [[TMP0]] +// +// CHECK-STRICT-LABEL: define dso_local x86_fp80 @fmax80b( +// CHECK-STRICT-SAME: x86_fp80 noundef [[A:%.*]], x86_fp80 noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] { +// CHECK-STRICT-NEXT: [[ENTRY:.*:]] +// CHECK-STRICT-NEXT: [[TMP0:%.*]] = tail call nsz x86_fp80 @llvm.experimental.constrained.maxnum.f80(x86_fp80 [[A]], x86_fp80 [[B]], metadata !"fpexcept.strict") #[[ATTR2]] +// CHECK-STRICT-NEXT: ret x86_fp80 [[TMP0]] +// +long double fmax80b(long double a, long double b) { return __builtin_fmaxl(a, b); } -//CHECK-STRICT: attributes #2 = { strictfp } _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
