Skip to content

[Clang][X86] Replace F16C vcvtph2ps/256 intrinsics with (convert|shuffle)vector builtins #152911

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 5 commits into from
Aug 12, 2025
+14 −39
Merged

[Clang][X86] Replace F16C vcvtph2ps/256 intrinsics with (convert|shuffle)vector builtins #152911

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
5 commits
Select commit Hold shift + click to select a range
ea2135f
[Clang][X86] Replace F16C `vcvtph2ps/256` intrinsics with __builtin_c…
moorabbit Aug 10, 2025
98baf57
RKSimon's comment
moorabbit Aug 11, 2025
b50d611
RKSimon's comment #2
moorabbit Aug 12, 2025
77ce810
clang-format
moorabbit Aug 12, 2025
fb8532a
Merge branch 'main' into replace-ia32-vcvtph2ps-256-intrinsics
moorabbit Aug 12, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
8 changes: 0 additions & 8 deletions clang/include/clang/Basic/BuiltinsX86.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -757,14 +757,6 @@ let Features = "f16c", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] i
def vcvtps2ph256 : X86Builtin<"_Vector<8, short>(_Vector<8, float>, _Constant int)">;
}

let Features = "f16c", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
def vcvtph2ps : X86Builtin<"_Vector<4, float>(_Vector<8, short>)">;
}

let Features = "f16c", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
def vcvtph2ps256 : X86Builtin<"_Vector<8, float>(_Vector<8, short>)">;
}

let Features = "rdrnd", Attributes = [NoThrow] in {
def rdrand16_step : X86Builtin<"unsigned int(unsigned short *)">;
def rdrand32_step : X86Builtin<"unsigned int(unsigned int *)">;
Expand Down
2 changes: 0 additions & 2 deletions clang/lib/CodeGen/TargetBuiltins/X86.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2841,8 +2841,6 @@ Value *CodeGenFunction::EmitX86BuiltinExpr(unsigned BuiltinID,
return getCmpIntrinsicCall(Intrinsic::x86_sse2_cmp_sd, 7);

// f16c half2float intrinsics
case X86::BI__builtin_ia32_vcvtph2ps:
case X86::BI__builtin_ia32_vcvtph2ps256:
case X86::BI__builtin_ia32_vcvtph2ps_mask:
case X86::BI__builtin_ia32_vcvtph2ps256_mask:
case X86::BI__builtin_ia32_vcvtph2ps512_mask: {
Expand Down
13 changes: 8 additions & 5 deletions clang/lib/Headers/f16cintrin.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,9 +38,7 @@
static __inline float __DEFAULT_FN_ATTRS128
_cvtsh_ss(unsigned short __a)
{
__v8hi __v = {(short)__a, 0, 0, 0, 0, 0, 0, 0};
__v4sf __r = __builtin_ia32_vcvtph2ps(__v);
return __r[0];
return (float)__builtin_bit_cast(__fp16, __a);
}

/// Converts a 32-bit single-precision float value to a 16-bit
Expand Down Expand Up @@ -109,7 +107,10 @@ _cvtsh_ss(unsigned short __a)
static __inline __m128 __DEFAULT_FN_ATTRS128
_mm_cvtph_ps(__m128i __a)
{
return (__m128)__builtin_ia32_vcvtph2ps((__v8hi)__a);
typedef __fp16 __v4fp16 __attribute__((__vector_size__(8)));

__v4hi __v = __builtin_shufflevector((__v8hi)__a, (__v8hi)__a, 0, 1, 2, 3);
return (__m128) __builtin_convertvector((__v4fp16)__v, __v4sf);
}

/// Converts a 256-bit vector of [8 x float] into a 128-bit vector
Expand Down Expand Up @@ -153,7 +154,9 @@ _mm_cvtph_ps(__m128i __a)
static __inline __m256 __DEFAULT_FN_ATTRS256
_mm256_cvtph_ps(__m128i __a)
{
return (__m256)__builtin_ia32_vcvtph2ps256((__v8hi)__a);
typedef __fp16 __v8fp16 __attribute__((__vector_size__(16), __aligned__(16)));

return (__m256) __builtin_convertvector((__v8fp16)__a, __v8sf);
}

#undef __DEFAULT_FN_ATTRS128
Expand Down
15 changes: 3 additions & 12 deletions clang/test/CodeGen/X86/f16c-builtins-constrained.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,17 +8,8 @@

float test_cvtsh_ss(unsigned short a) {
// CHECK-LABEL: test_cvtsh_ss
// CHECK: insertelement <8 x i16> poison, i16 %{{.*}}, i32 0
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 1
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 2
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 3
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 4
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 5
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 6
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 7
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
// CHECK: call <4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half> %{{.*}}, metadata !"fpexcept.strict")
// CHECK: extractelement <4 x float> %{{.*}}, i32 0
// CHECK: [[CONV:%.*]] = call {{.*}}float @llvm.experimental.constrained.fpext.f32.f16(half %{{.*}}, metadata !"fpexcept.strict")
// CHECK: ret float [[CONV]]
return _cvtsh_ss(a);
}

Expand All @@ -38,7 +29,7 @@ unsigned short test_cvtss_sh(float a) {

__m128 test_mm_cvtph_ps(__m128i a) {
// CHECK-LABEL: test_mm_cvtph_ps
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
// CHECK: call {{.*}}<4 x float> @llvm.experimental.constrained.fpext.v4f32.v4f16(<4 x half> %{{.*}}, metadata !"fpexcept.strict")
return _mm_cvtph_ps(a);
}
Expand Down
15 changes: 3 additions & 12 deletions clang/test/CodeGen/X86/f16c-builtins.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8,17 +8,8 @@

float test_cvtsh_ss(unsigned short a) {
// CHECK-LABEL: test_cvtsh_ss
// CHECK: insertelement <8 x i16> poison, i16 %{{.*}}, i32 0
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 1
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 2
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 3
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 4
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 5
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 6
// CHECK: insertelement <8 x i16> %{{.*}}, i16 0, i32 7
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
// CHECK: fpext <4 x half> %{{.*}} to <4 x float>
// CHECK: extractelement <4 x float> %{{.*}}, i32 0
// CHECK: [[CONV:%.*]] = fpext half %{{.*}} to float
// CHECK: ret float [[CONV]]
return _cvtsh_ss(a);
}

Expand All @@ -35,7 +26,7 @@ unsigned short test_cvtss_sh(float a) {

__m128 test_mm_cvtph_ps(__m128i a) {
// CHECK-LABEL: test_mm_cvtph_ps
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
// CHECK: shufflevector <8 x i16> %{{.*}}, <8 x i16> %{{.*}}, <4 x i32> <i32 0, i32 1, i32 2, i32 3>
// CHECK: fpext <4 x half> %{{.*}} to <4 x float>
return _mm_cvtph_ps(a);
}
Expand Down