[clang] b15bf30 - Reapply "clang: Use new frexp intrinsic for builtins and add f16 version"
Matt Arsenault via cfe-commits
cfe-commits at lists.llvm.org
Fri Jun 30 06:07:37 PDT 2023
Author: Matt Arsenault
Date: 2023-06-30T09:07:23-04:00
New Revision: b15bf305ca3e9ce63aaef7247d32fb3a75174531
URL: https://github.com/llvm/llvm-project/commit/b15bf305ca3e9ce63aaef7247d32fb3a75174531
DIFF: https://github.com/llvm/llvm-project/commit/b15bf305ca3e9ce63aaef7247d32fb3a75174531.diff
LOG: Reapply "clang: Use new frexp intrinsic for builtins and add f16 version"
This reverts commit 0c545a441285a73e00b859dd52f1a85cb9eeeefc.
ARM libcall expansion was fixed in 160d7227e043cd8f491cb706916da953cabfd905
Added:
Modified:
clang/include/clang/Basic/Builtins.def
clang/lib/CodeGen/CGBuiltin.cpp
clang/lib/Headers/__clang_hip_math.h
clang/test/CodeGen/aix-builtin-mapping.c
clang/test/CodeGen/builtin-attributes.c
clang/test/CodeGen/math-builtins-long.c
clang/test/CodeGen/math-builtins.c
clang/test/CodeGenOpenCL/builtins-generic-amdgcn.cl
clang/test/Headers/__clang_hip_math.hip
Removed:
################################################################################
diff --git a/clang/include/clang/Basic/Builtins.def b/clang/include/clang/Basic/Builtins.def
index c0e045865e2c20..fe00a2f69922a1 100644
--- a/clang/include/clang/Basic/Builtins.def
+++ b/clang/include/clang/Basic/Builtins.def
@@ -143,6 +143,7 @@ BUILTIN(__builtin_frexp , "ddi*" , "Fn")
BUILTIN(__builtin_frexpf, "ffi*" , "Fn")
BUILTIN(__builtin_frexpl, "LdLdi*", "Fn")
BUILTIN(__builtin_frexpf128, "LLdLLdi*", "Fn")
+BUILTIN(__builtin_frexpf16, "hhi*" , "Fn")
BUILTIN(__builtin_huge_val, "d", "ncE")
BUILTIN(__builtin_huge_valf, "f", "ncE")
BUILTIN(__builtin_huge_vall, "Ld", "ncE")
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 3a0588ad752b84..bb59d2ea90f554 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -652,6 +652,24 @@ emitMaybeConstrainedFPToIntRoundBuiltin(CodeGenFunction &CGF, const CallExpr *E,
}
}
+static Value *emitFrexpBuiltin(CodeGenFunction &CGF, const CallExpr *E,
+ llvm::Intrinsic::ID IntrinsicID) {
+ llvm::Value *Src0 = CGF.EmitScalarExpr(E->getArg(0));
+ llvm::Value *Src1 = CGF.EmitScalarExpr(E->getArg(1));
+
+ QualType IntPtrTy = E->getArg(1)->getType()->getPointeeType();
+ llvm::Type *IntTy = CGF.ConvertType(IntPtrTy);
+ llvm::Function *F =
+ CGF.CGM.getIntrinsic(IntrinsicID, {Src0->getType(), IntTy});
+ llvm::Value *Call = CGF.Builder.CreateCall(F, Src0);
+
+ llvm::Value *Exp = CGF.Builder.CreateExtractValue(Call, 1);
+ LValue LV = CGF.MakeNaturalAlignAddrLValue(Src1, IntPtrTy);
+ CGF.EmitStoreOfScalar(Exp, LV);
+
+ return CGF.Builder.CreateExtractValue(Call, 0);
+}
+
/// EmitFAbs - Emit a call to @llvm.fabs().
static Value *EmitFAbs(CodeGenFunction &CGF, Value *V) {
Function *F = CGF.CGM.getIntrinsic(Intrinsic::fabs, V->getType());
@@ -3062,6 +3080,12 @@ RValue CodeGenFunction::EmitBuiltinExpr(const GlobalDecl GD, unsigned BuiltinID,
{ Src0->getType(), Src1->getType() });
return RValue::get(Builder.CreateCall(F, { Src0, Src1 }));
}
+ case Builtin::BI__builtin_frexp:
+ case Builtin::BI__builtin_frexpf:
+ case Builtin::BI__builtin_frexpl:
+ case Builtin::BI__builtin_frexpf128:
+ case Builtin::BI__builtin_frexpf16:
+ return RValue::get(emitFrexpBuiltin(*this, E, Intrinsic::frexp));
case Builtin::BI__builtin_isgreater:
case Builtin::BI__builtin_isgreaterequal:
case Builtin::BI__builtin_isless:
diff --git a/clang/lib/Headers/__clang_hip_math.h b/clang/lib/Headers/__clang_hip_math.h
index c67959673f936a..a914496cb7b14a 100644
--- a/clang/lib/Headers/__clang_hip_math.h
+++ b/clang/lib/Headers/__clang_hip_math.h
@@ -257,8 +257,7 @@ float fmodf(float __x, float __y) { return __ocml_fmod_f32(__x, __y); }
__DEVICE__
float frexpf(float __x, int *__nptr) {
- *__nptr = __builtin_amdgcn_frexp_expf(__x);
- return __builtin_amdgcn_frexp_mantf(__x);
+ return __builtin_frexpf(__x, __nptr);
}
__DEVICE__
@@ -806,8 +805,7 @@ double fmod(double __x, double __y) { return __ocml_fmod_f64(__x, __y); }
__DEVICE__
double frexp(double __x, int *__nptr) {
- *__nptr = __builtin_amdgcn_frexp_exp(__x);
- return __builtin_amdgcn_frexp_mant(__x);
+ return __builtin_frexp(__x, __nptr);
}
__DEVICE__
diff --git a/clang/test/CodeGen/aix-builtin-mapping.c b/clang/test/CodeGen/aix-builtin-mapping.c
index 98fcfd4a3a6fce..a79218c6f1d8b9 100644
--- a/clang/test/CodeGen/aix-builtin-mapping.c
+++ b/clang/test/CodeGen/aix-builtin-mapping.c
@@ -18,5 +18,5 @@ int main()
}
// CHECK: %call = call double @modf(double noundef 1.000000e+00, ptr noundef %DummyLongDouble) #3
-// CHECK: %call1 = call double @frexp(double noundef 0.000000e+00, ptr noundef %DummyInt) #3
+// CHECK: %{{.+}} = call { double, i32 } @llvm.frexp.f64.i32(double 0.000000e+00)
// CHECK: %{{.+}} = call double @llvm.ldexp.f64.i32(double 1.000000e+00, i32 1)
diff --git a/clang/test/CodeGen/builtin-attributes.c b/clang/test/CodeGen/builtin-attributes.c
index fd1e107a41d583..ec184e3a7b0488 100644
--- a/clang/test/CodeGen/builtin-attributes.c
+++ b/clang/test/CodeGen/builtin-attributes.c
@@ -4,6 +4,10 @@
int printf(const char *, ...);
void exit(int);
+float frexpf(float, int*);
+double frexp(double, int*);
+long double frexpl(long double, int*);
+
// CHECK: declare i32 @printf(ptr noundef, ...)
void f0() {
printf("a\n");
@@ -49,9 +53,9 @@ char* f2(char* a, char* b) {
// CHECK: ret
int f3(double x) {
int e;
- __builtin_frexp(x, &e);
- __builtin_frexpf(x, &e);
- __builtin_frexpl(x, &e);
+ frexp(x, &e);
+ frexpf(x, &e);
+ frexpl(x, &e);
__builtin_modf(x, &e);
__builtin_modff(x, &e);
__builtin_modfl(x, &e);
diff --git a/clang/test/CodeGen/math-builtins-long.c b/clang/test/CodeGen/math-builtins-long.c
index e32fc99480ead5..f3c328dcbfcd7c 100644
--- a/clang/test/CodeGen/math-builtins-long.c
+++ b/clang/test/CodeGen/math-builtins-long.c
@@ -34,10 +34,10 @@ void foo(long double f, long double *l, int *i, const char *c) {
// PPCF128: call fp128 @llvm.fabs.f128(fp128 %{{.+}})
__builtin_fabsl(f);
- // F80: call x86_fp80 @frexpl(x86_fp80 noundef %{{.+}}, ptr noundef %{{.+}})
- // PPC: call ppc_fp128 @frexpl(ppc_fp128 noundef %{{.+}}, ptr noundef %{{.+}})
- // X86F128: call fp128 @frexpl(fp128 noundef %{{.+}}, ptr noundef %{{.+}})
- // PPCF128: call fp128 @frexpf128(fp128 noundef %{{.+}}, ptr noundef %{{.+}})
+ // F80: call { x86_fp80, i32 } @llvm.frexp.f80.i32(x86_fp80 %{{.+}})
+ // PPC: call { ppc_fp128, i32 } @llvm.frexp.ppcf128.i32(ppc_fp128 %{{.+}})
+ // X86F128: call { fp128, i32 } @llvm.frexp.f128.i32(fp128 %{{.+}})
+ // PPCF128: call { fp128, i32 } @llvm.frexp.f128.i32(fp128 %{{.+}})
__builtin_frexpl(f,i);
// F80: store x86_fp80 0xK7FFF8000000000000000, ptr
diff --git a/clang/test/CodeGen/math-builtins.c b/clang/test/CodeGen/math-builtins.c
index 62335019379206..257da44b5888a1 100644
--- a/clang/test/CodeGen/math-builtins.c
+++ b/clang/test/CodeGen/math-builtins.c
@@ -12,6 +12,30 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
// NO__ERRNO: frem float
// NO__ERRNO: frem x86_fp80
// NO__ERRNO: frem fp128
+
+// NO__ERRNO: [[FREXP_F64:%.+]] = call { double, i32 } @llvm.frexp.f64.i32(double %{{.+}})
+// NO__ERRNO-NEXT: [[FREXP_F64_1:%.+]] = extractvalue { double, i32 } [[FREXP_F64]], 1
+// NO__ERRNO-NEXT: store i32 [[FREXP_F64_1]], ptr %{{.+}}, align 4
+// NO__ERRNO-NEXT: [[FREXP_F64_0:%.+]] = extractvalue { double, i32 } [[FREXP_F64]], 0
+
+// NO__ERRNO: [[FREXP_F32:%.+]] = call { float, i32 } @llvm.frexp.f32.i32(float %{{.+}})
+// NO__ERRNO-NEXT: [[FREXP_F32_1:%.+]] = extractvalue { float, i32 } [[FREXP_F32]], 1
+// NO__ERRNO-NEXT: store i32 [[FREXP_F32_1]], ptr %{{.+}}, align 4
+// NO__ERRNO-NEXT: [[FREXP_F32_0:%.+]] = extractvalue { float, i32 } [[FREXP_F32]], 0
+
+
+// NO__ERRNO: [[FREXP_F80:%.+]] = call { x86_fp80, i32 } @llvm.frexp.f80.i32(x86_fp80 %{{.+}})
+// NO__ERRNO-NEXT: [[FREXP_F80_1:%.+]] = extractvalue { x86_fp80, i32 } [[FREXP_F80]], 1
+// NO__ERRNO-NEXT: store i32 [[FREXP_F80_1]], ptr %{{.+}}, align 4
+// NO__ERRNO-NEXT: [[FREXP_F80_0:%.+]] = extractvalue { x86_fp80, i32 } [[FREXP_F80]], 0
+
+
+// NO__ERRNO: [[FREXP_F128:%.+]] = call { fp128, i32 } @llvm.frexp.f128.i32(fp128 %{{.+}})
+// NO__ERRNO-NEXT: [[FREXP_F128_1:%.+]] = extractvalue { fp128, i32 } [[FREXP_F128]], 1
+// NO__ERRNO-NEXT: store i32 [[FREXP_F128_1]], ptr %{{.+}}, align 4
+// NO__ERRNO-NEXT: [[FREXP_F128_0:%.+]] = extractvalue { fp128, i32 } [[FREXP_F128]], 0
+
+
// HAS_ERRNO: declare double @fmod(double noundef, double noundef) [[NOT_READNONE:#[0-9]+]]
// HAS_ERRNO: declare float @fmodf(float noundef, float noundef) [[NOT_READNONE]]
// HAS_ERRNO: declare x86_fp80 @fmodl(x86_fp80 noundef, x86_fp80 noundef) [[NOT_READNONE]]
@@ -52,14 +76,14 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
__builtin_frexp(f,i); __builtin_frexpf(f,i); __builtin_frexpl(f,i); __builtin_frexpf128(f,i);
-// NO__ERRNO: declare double @frexp(double noundef, ptr noundef) [[NOT_READNONE:#[0-9]+]]
-// NO__ERRNO: declare float @frexpf(float noundef, ptr noundef) [[NOT_READNONE]]
-// NO__ERRNO: declare x86_fp80 @frexpl(x86_fp80 noundef, ptr noundef) [[NOT_READNONE]]
-// NO__ERRNO: declare fp128 @frexpf128(fp128 noundef, ptr noundef) [[NOT_READNONE]]
-// HAS_ERRNO: declare double @frexp(double noundef, ptr noundef) [[NOT_READNONE]]
-// HAS_ERRNO: declare float @frexpf(float noundef, ptr noundef) [[NOT_READNONE]]
-// HAS_ERRNO: declare x86_fp80 @frexpl(x86_fp80 noundef, ptr noundef) [[NOT_READNONE]]
-// HAS_ERRNO: declare fp128 @frexpf128(fp128 noundef, ptr noundef) [[NOT_READNONE]]
+// NO__ERRNO: declare { double, i32 } @llvm.frexp.f64.i32(double) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare { float, i32 } @llvm.frexp.f32.i32(float) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare { x86_fp80, i32 } @llvm.frexp.f80.i32(x86_fp80) [[READNONE_INTRINSIC]]
+// NO__ERRNO: declare { fp128, i32 } @llvm.frexp.f128.i32(fp128) [[READNONE_INTRINSIC]]
+// HAS_ERRNO: declare { double, i32 } @llvm.frexp.f64.i32(double) [[READNONE_INTRINSIC]]
+// HAS_ERRNO: declare { float, i32 } @llvm.frexp.f32.i32(float) [[READNONE_INTRINSIC]]
+// HAS_ERRNO: declare { x86_fp80, i32 } @llvm.frexp.f80.i32(x86_fp80) [[READNONE_INTRINSIC]]
+// HAS_ERRNO: declare { fp128, i32 } @llvm.frexp.f128.i32(fp128) [[READNONE_INTRINSIC]]
__builtin_huge_val(); __builtin_huge_valf(); __builtin_huge_vall(); __builtin_huge_valf128();
@@ -88,7 +112,7 @@ void foo(double *d, float f, float *fp, long double *l, int *i, const char *c) {
__builtin_modf(f,d); __builtin_modff(f,fp); __builtin_modfl(f,l); __builtin_modff128(f,l);
-// NO__ERRNO: declare double @modf(double noundef, ptr noundef) [[NOT_READNONE]]
+// NO__ERRNO: declare double @modf(double noundef, ptr noundef) [[NOT_READNONE:#[0-9]+]]
// NO__ERRNO: declare float @modff(float noundef, ptr noundef) [[NOT_READNONE]]
// NO__ERRNO: declare x86_fp80 @modfl(x86_fp80 noundef, ptr noundef) [[NOT_READNONE]]
// NO__ERRNO: declare fp128 @modff128(fp128 noundef, ptr noundef) [[NOT_READNONE]]
diff --git a/clang/test/CodeGenOpenCL/builtins-generic-amdgcn.cl b/clang/test/CodeGenOpenCL/builtins-generic-amdgcn.cl
index aa8f13310b5fd8..b093fcbf7d9894 100644
--- a/clang/test/CodeGenOpenCL/builtins-generic-amdgcn.cl
+++ b/clang/test/CodeGenOpenCL/builtins-generic-amdgcn.cl
@@ -39,3 +39,33 @@ float test_builtin_ldexpf(float v, int e) {
double test_builtin_ldexp(double v, int e) {
return __builtin_ldexp(v, e);
}
+
+// CHECK-LABEL: @test_builtin_frexpf16(
+// CHECK: [[VAL:%.+]] = tail call { half, i32 } @llvm.frexp.f16.i32(half %v)
+// CHECK: [[EXTRACT_1:%.+]] = extractvalue { half, i32 } [[VAL]], 1
+// CHECK: store i32 [[EXTRACT_1]], ptr addrspace(5)
+// CHECK: [[EXTRACT_0:%.+]] = extractvalue { half, i32 } [[VAL]], 0
+// CHECK: ret half [[EXTRACT_0]]
+half test_builtin_frexpf16(half v, int* e) {
+ return __builtin_frexpf16(v, e);
+}
+
+// CHECK-LABEL: @test_builtin_frexpf(
+// CHECK: [[VAL:%.+]] = tail call { float, i32 } @llvm.frexp.f32.i32(float %v)
+// CHECK: [[EXTRACT_1:%.+]] = extractvalue { float, i32 } [[VAL]], 1
+// CHECK: store i32 [[EXTRACT_1]], ptr addrspace(5)
+// CHECK: [[EXTRACT_0:%.+]] = extractvalue { float, i32 } [[VAL]], 0
+// CHECK: ret float [[EXTRACT_0]]
+float test_builtin_frexpf(float v, int* e) {
+ return __builtin_frexpf(v, e);
+}
+
+// CHECK-LABEL: @test_builtin_frexp(
+// CHECK: [[VAL:%.+]] = tail call { double, i32 } @llvm.frexp.f64.i32(double %v)
+// CHECK: [[EXTRACT_1:%.+]] = extractvalue { double, i32 } [[VAL]], 1
+// CHECK: store i32 [[EXTRACT_1]], ptr addrspace(5)
+// CHECK: [[EXTRACT_0:%.+]] = extractvalue { double, i32 } [[VAL]], 0
+// CHECK: ret double [[EXTRACT_0]]
+double test_builtin_frexp(double v, int* e) {
+ return __builtin_frexp(v, e);
+}
diff --git a/clang/test/Headers/__clang_hip_math.hip b/clang/test/Headers/__clang_hip_math.hip
index c4a848c9bc6211..8e5201a89ceca2 100644
--- a/clang/test/Headers/__clang_hip_math.hip
+++ b/clang/test/Headers/__clang_hip_math.hip
@@ -1061,37 +1061,25 @@ extern "C" __device__ double test_fmod(double x, double y) {
return fmod(x, y);
}
-// DEFAULT-LABEL: @test_frexpf(
-// DEFAULT-NEXT: entry:
-// DEFAULT-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[X:%.*]])
-// DEFAULT-NEXT: store i32 [[TMP0]], ptr [[Y:%.*]], align 4, !tbaa [[TBAA12:![0-9]+]]
-// DEFAULT-NEXT: [[TMP1:%.*]] = tail call contract float @llvm.amdgcn.frexp.mant.f32(float [[X]])
-// DEFAULT-NEXT: ret float [[TMP1]]
-//
-// FINITEONLY-LABEL: @test_frexpf(
-// FINITEONLY-NEXT: entry:
-// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.frexp.exp.i32.f32(float [[X:%.*]])
-// FINITEONLY-NEXT: store i32 [[TMP0]], ptr [[Y:%.*]], align 4, !tbaa [[TBAA12:![0-9]+]]
-// FINITEONLY-NEXT: [[TMP1:%.*]] = tail call nnan ninf contract float @llvm.amdgcn.frexp.mant.f32(float [[X]])
-// FINITEONLY-NEXT: ret float [[TMP1]]
+// CHECK-LABEL: @test_frexpf(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call { float, i32 } @llvm.frexp.f32.i32(float [[X:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { float, i32 } [[TMP0]], 1
+// CHECK-NEXT: store i32 [[TMP1]], ptr [[Y:%.*]], align 4, !tbaa [[TBAA12:![0-9]+]]
+// CHECK-NEXT: [[TMP2:%.*]] = extractvalue { float, i32 } [[TMP0]], 0
+// CHECK-NEXT: ret float [[TMP2]]
//
extern "C" __device__ float test_frexpf(float x, int* y) {
return frexpf(x, y);
}
-// DEFAULT-LABEL: @test_frexp(
-// DEFAULT-NEXT: entry:
-// DEFAULT-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.frexp.exp.i32.f64(double [[X:%.*]])
-// DEFAULT-NEXT: store i32 [[TMP0]], ptr [[Y:%.*]], align 4, !tbaa [[TBAA12]]
-// DEFAULT-NEXT: [[TMP1:%.*]] = tail call contract double @llvm.amdgcn.frexp.mant.f64(double [[X]])
-// DEFAULT-NEXT: ret double [[TMP1]]
-//
-// FINITEONLY-LABEL: @test_frexp(
-// FINITEONLY-NEXT: entry:
-// FINITEONLY-NEXT: [[TMP0:%.*]] = tail call i32 @llvm.amdgcn.frexp.exp.i32.f64(double [[X:%.*]])
-// FINITEONLY-NEXT: store i32 [[TMP0]], ptr [[Y:%.*]], align 4, !tbaa [[TBAA12]]
-// FINITEONLY-NEXT: [[TMP1:%.*]] = tail call nnan ninf contract double @llvm.amdgcn.frexp.mant.f64(double [[X]])
-// FINITEONLY-NEXT: ret double [[TMP1]]
+// CHECK-LABEL: @test_frexp(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call { double, i32 } @llvm.frexp.f64.i32(double [[X:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = extractvalue { double, i32 } [[TMP0]], 1
+// CHECK-NEXT: store i32 [[TMP1]], ptr [[Y:%.*]], align 4, !tbaa [[TBAA12]]
+// CHECK-NEXT: [[TMP2:%.*]] = extractvalue { double, i32 } [[TMP0]], 0
+// CHECK-NEXT: ret double [[TMP2]]
//
extern "C" __device__ double test_frexp(double x, int* y) {
return frexp(x, y);
More information about the cfe-commits
mailing list