Skip to content

[x86][AVX-VNNI] Fix VPDPBUSD Argument Types #155194

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
RKSimon merged 10 commits into from
Sep 10, 2025
Merged

[x86][AVX-VNNI] Fix VPDPBUSD Argument Types #155194

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
944873d
[x86][AVX-VNNI] Fix VPDPBUSD Argument Types
BaiXilin Aug 24, 2025
7868ed7
Fixed memory sanitizer assertion failures
BaiXilin Aug 26, 2025
67e46df
Fix variable type style
BaiXilin Aug 29, 2025
48b398b
Fixed builtins second parameter type: should be unsigned char
BaiXilin Sep 5, 2025
4c31af8
Merge branch 'main' into new-fix-vnni-types
BaiXilin Sep 10, 2025
5f10ec6
Merge branch 'main' into new-fix-vnni-types
BaiXilin Sep 10, 2025
685ef89
Merge branch 'new-fix-vnni-types' of github.com:BaiXilin/llvm-fix-vnn…
BaiXilin Sep 10, 2025
97f8231
Add auto-upgrade to support intrinsics with old signatures. Added
BaiXilin Sep 10, 2025
e259ce4
Resolve code review suggestions
BaiXilin Sep 10, 2025
fde5d76
Merge branch 'main' into new-fix-vnni-types
RKSimon Sep 10, 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
12 changes: 6 additions & 6 deletions clang/include/clang/Basic/BuiltinsX86.td
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1057,27 +1057,27 @@ let Features = "avx512f", Attributes = [NoThrow, Const, RequiredVectorWidth<512>
}

let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
def vpdpbusd128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
def vpdpbusd128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<16, unsigned char>, _Vector<16, char>)">;
}

let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
def vpdpbusd256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
def vpdpbusd256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<32, unsigned char>, _Vector<32, char>)">;
}

let Features = "avx512vnni", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
def vpdpbusd512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
def vpdpbusd512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<64, unsigned char>, _Vector<64, char>)">;
}

let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
def vpdpbusds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<4, int>, _Vector<4, int>)">;
def vpdpbusds128 : X86Builtin<"_Vector<4, int>(_Vector<4, int>, _Vector<16, unsigned char>, _Vector<16, char>)">;
}

let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<256>] in {
def vpdpbusds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<8, int>, _Vector<8, int>)">;
def vpdpbusds256 : X86Builtin<"_Vector<8, int>(_Vector<8, int>, _Vector<32, unsigned char>, _Vector<32, char>)">;
}

let Features = "avx512vnni", Attributes = [NoThrow, Const, RequiredVectorWidth<512>] in {
def vpdpbusds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<16, int>, _Vector<16, int>)">;
def vpdpbusds512 : X86Builtin<"_Vector<16, int>(_Vector<16, int>, _Vector<64, unsigned char>, _Vector<64, char>)">;
}

let Features = "avx512vl,avx512vnni|avxvnni", Attributes = [NoThrow, Const, RequiredVectorWidth<128>] in {
Expand Down
18 changes: 10 additions & 8 deletions clang/lib/Headers/avx512vlvnniintrin.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,8 +41,8 @@
/// ENDFOR
/// DST[MAX:256] := 0
/// \endcode
#define _mm256_dpbusd_epi32(S, A, B) \
((__m256i)__builtin_ia32_vpdpbusd256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
#define _mm256_dpbusd_epi32(S, A, B) \
((__m256i)__builtin_ia32_vpdpbusd256((__v8si)(S), (__v32qu)(A), (__v32qi)(B)))

/// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with
/// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed
Expand All @@ -61,8 +61,9 @@
/// ENDFOR
/// DST[MAX:256] := 0
/// \endcode
#define _mm256_dpbusds_epi32(S, A, B) \
((__m256i)__builtin_ia32_vpdpbusds256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
#define _mm256_dpbusds_epi32(S, A, B) \
((__m256i)__builtin_ia32_vpdpbusds256((__v8si)(S), (__v32qu)(A), \
(__v32qi)(B)))

/// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
/// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
Expand Down Expand Up @@ -117,8 +118,8 @@
/// ENDFOR
/// DST[MAX:128] := 0
/// \endcode
#define _mm_dpbusd_epi32(S, A, B) \
((__m128i)__builtin_ia32_vpdpbusd128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
#define _mm_dpbusd_epi32(S, A, B) \
((__m128i)__builtin_ia32_vpdpbusd128((__v4si)(S), (__v16qu)(A), (__v16qi)(B)))

/// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with
/// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed
Expand All @@ -137,8 +138,9 @@
/// ENDFOR
/// DST[MAX:128] := 0
/// \endcode
#define _mm_dpbusds_epi32(S, A, B) \
((__m128i)__builtin_ia32_vpdpbusds128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
#define _mm_dpbusds_epi32(S, A, B) \
((__m128i)__builtin_ia32_vpdpbusds128((__v4si)(S), (__v16qu)(A), \
(__v16qi)(B)))

/// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
/// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
Expand Down
8 changes: 4 additions & 4 deletions clang/lib/Headers/avx512vnniintrin.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -22,8 +22,8 @@
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_dpbusd_epi32(__m512i __S, __m512i __A, __m512i __B)
{
return (__m512i)__builtin_ia32_vpdpbusd512((__v16si)__S, (__v16si)__A,
(__v16si)__B);
return (__m512i)__builtin_ia32_vpdpbusd512((__v16si)__S, (__v64qu)__A,
(__v64qi)__B);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
Expand All @@ -45,8 +45,8 @@ _mm512_maskz_dpbusd_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B)
static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_dpbusds_epi32(__m512i __S, __m512i __A, __m512i __B)
{
return (__m512i)__builtin_ia32_vpdpbusds512((__v16si)__S, (__v16si)__A,
(__v16si)__B);
return (__m512i)__builtin_ia32_vpdpbusds512((__v16si)__S, (__v64qu)__A,
(__v64qi)__B);
}

static __inline__ __m512i __DEFAULT_FN_ATTRS
Expand Down
12 changes: 8 additions & 4 deletions clang/lib/Headers/avxvnniintrin.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -63,7 +63,8 @@
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_dpbusd_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
{
return (__m256i)__builtin_ia32_vpdpbusd256((__v8si)__S, (__v8si)__A, (__v8si)__B);
return (__m256i)__builtin_ia32_vpdpbusd256((__v8si)__S, (__v32qu)__A,
(__v32qi)__B);
}

/// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a __A with
Expand All @@ -86,7 +87,8 @@ _mm256_dpbusd_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_dpbusds_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
{
return (__m256i)__builtin_ia32_vpdpbusds256((__v8si)__S, (__v8si)__A, (__v8si)__B);
return (__m256i)__builtin_ia32_vpdpbusds256((__v8si)__S, (__v32qu)__A,
(__v32qi)__B);
}

/// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
Expand Down Expand Up @@ -151,7 +153,8 @@ _mm256_dpwssds_avx_epi32(__m256i __S, __m256i __A, __m256i __B)
static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_dpbusd_avx_epi32(__m128i __S, __m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_vpdpbusd128((__v4si)__S, (__v4si)__A, (__v4si)__B);
return (__m128i)__builtin_ia32_vpdpbusd128((__v4si)__S, (__v16qu)__A,
(__v16qi)__B);
}

/// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a __A with
Expand All @@ -174,7 +177,8 @@ _mm_dpbusd_avx_epi32(__m128i __S, __m128i __A, __m128i __B)
static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_dpbusds_avx_epi32(__m128i __S, __m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_vpdpbusds128((__v4si)__S, (__v4si)__A, (__v4si)__B);
return (__m128i)__builtin_ia32_vpdpbusds128((__v4si)__S, (__v16qu)__A,
(__v16qi)__B);
}

/// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a __A with
Expand Down
24 changes: 12 additions & 12 deletions clang/test/CodeGen/X86/avx512vlvnni-builtins.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,41 +7,41 @@

__m256i test_mm256_mask_dpbusd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_mask_dpbusd_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
// CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
return _mm256_mask_dpbusd_epi32(__S, __U, __A, __B);
}

__m256i test_mm256_maskz_dpbusd_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_maskz_dpbusd_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
// CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
return _mm256_maskz_dpbusd_epi32(__U, __S, __A, __B);
}

__m256i test_mm256_dpbusd_epi32(__m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_dpbusd_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
return _mm256_dpbusd_epi32(__S, __A, __B);
}

__m256i test_mm256_mask_dpbusds_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_mask_dpbusds_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
// CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
return _mm256_mask_dpbusds_epi32(__S, __U, __A, __B);
}

__m256i test_mm256_maskz_dpbusds_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_maskz_dpbusds_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
// CHECK: select <8 x i1> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}}
return _mm256_maskz_dpbusds_epi32(__U, __S, __A, __B);
}

__m256i test_mm256_dpbusds_epi32(__m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_dpbusds_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
return _mm256_dpbusds_epi32(__S, __A, __B);
}

Expand Down Expand Up @@ -87,41 +87,41 @@ __m256i test_mm256_dpwssds_epi32(__m256i __S, __m256i __A, __m256i __B) {

__m128i test_mm_mask_dpbusd_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_mask_dpbusd_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
// CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
return _mm_mask_dpbusd_epi32(__S, __U, __A, __B);
}

__m128i test_mm_maskz_dpbusd_epi32(__mmask8 __U, __m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_maskz_dpbusd_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
// CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
return _mm_maskz_dpbusd_epi32(__U, __S, __A, __B);
}

__m128i test_mm_dpbusd_epi32(__m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_dpbusd_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
return _mm_dpbusd_epi32(__S, __A, __B);
}

__m128i test_mm_mask_dpbusds_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_mask_dpbusds_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
// CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
return _mm_mask_dpbusds_epi32(__S, __U, __A, __B);
}

__m128i test_mm_maskz_dpbusds_epi32(__mmask8 __U, __m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_maskz_dpbusds_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
// CHECK: select <4 x i1> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}}
return _mm_maskz_dpbusds_epi32(__U, __S, __A, __B);
}

__m128i test_mm_dpbusds_epi32(__m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_dpbusds_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
return _mm_dpbusds_epi32(__S, __A, __B);
}

Expand Down
12 changes: 6 additions & 6 deletions clang/test/CodeGen/X86/avx512vnni-builtins.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,41 +7,41 @@

__m512i test_mm512_mask_dpbusd_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: test_mm512_mask_dpbusd_epi32
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})
// CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_dpbusd_epi32(__S, __U, __A, __B);
}

__m512i test_mm512_maskz_dpbusd_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) {
// CHECK-LABEL: test_mm512_maskz_dpbusd_epi32
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})
// CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_dpbusd_epi32(__U, __S, __A, __B);
}

__m512i test_mm512_dpbusd_epi32(__m512i __S, __m512i __A, __m512i __B) {
// CHECK-LABEL: test_mm512_dpbusd_epi32
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusd.512(<16 x i32> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})
return _mm512_dpbusd_epi32(__S, __A, __B);
}

__m512i test_mm512_mask_dpbusds_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B) {
// CHECK-LABEL: test_mm512_mask_dpbusds_epi32
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})
// CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_mask_dpbusds_epi32(__S, __U, __A, __B);
}

__m512i test_mm512_maskz_dpbusds_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B) {
// CHECK-LABEL: test_mm512_maskz_dpbusds_epi32
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})
// CHECK: select <16 x i1> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}}
return _mm512_maskz_dpbusds_epi32(__U, __S, __A, __B);
}

__m512i test_mm512_dpbusds_epi32(__m512i __S, __m512i __A, __m512i __B) {
// CHECK-LABEL: test_mm512_dpbusds_epi32
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %{{.*}}, <16 x i32> %{{.*}}, <16 x i32> %{{.*}})
// CHECK: call <16 x i32> @llvm.x86.avx512.vpdpbusds.512(<16 x i32> %{{.*}}, <64 x i8> %{{.*}}, <64 x i8> %{{.*}})
return _mm512_dpbusds_epi32(__S, __A, __B);
}

Expand Down
16 changes: 8 additions & 8 deletions clang/test/CodeGen/X86/avxvnni-builtins.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7,13 +7,13 @@

__m256i test_mm256_dpbusd_epi32(__m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_dpbusd_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
return _mm256_dpbusd_epi32(__S, __A, __B);
}

__m256i test_mm256_dpbusds_epi32(__m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_dpbusds_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
return _mm256_dpbusds_epi32(__S, __A, __B);
}

Expand All @@ -31,13 +31,13 @@ __m256i test_mm256_dpwssds_epi32(__m256i __S, __m256i __A, __m256i __B) {

__m128i test_mm_dpbusd_epi32(__m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_dpbusd_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
return _mm_dpbusd_epi32(__S, __A, __B);
}

__m128i test_mm_dpbusds_epi32(__m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_dpbusds_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
return _mm_dpbusds_epi32(__S, __A, __B);
}

Expand All @@ -55,13 +55,13 @@ __m128i test_mm_dpwssds_epi32(__m128i __S, __m128i __A, __m128i __B) {

__m256i test_mm256_dpbusd_avx_epi32(__m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_dpbusd_avx_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusd.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
return _mm256_dpbusd_avx_epi32(__S, __A, __B);
}

__m256i test_mm256_dpbusds_avx_epi32(__m256i __S, __m256i __A, __m256i __B) {
// CHECK-LABEL: test_mm256_dpbusds_avx_epi32
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <8 x i32> %{{.*}}, <8 x i32> %{{.*}})
// CHECK: call <8 x i32> @llvm.x86.avx512.vpdpbusds.256(<8 x i32> %{{.*}}, <32 x i8> %{{.*}}, <32 x i8> %{{.*}})
return _mm256_dpbusds_avx_epi32(__S, __A, __B);
}

Expand All @@ -79,13 +79,13 @@ __m256i test_mm256_dpwssds_avx_epi32(__m256i __S, __m256i __A, __m256i __B) {

__m128i test_mm_dpbusd_avx_epi32(__m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_dpbusd_avx_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusd.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
return _mm_dpbusd_avx_epi32(__S, __A, __B);
}

__m128i test_mm_dpbusds_avx_epi32(__m128i __S, __m128i __A, __m128i __B) {
// CHECK-LABEL: test_mm_dpbusds_avx_epi32
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <4 x i32> %{{.*}}, <4 x i32> %{{.*}})
// CHECK: call <4 x i32> @llvm.x86.avx512.vpdpbusds.128(<4 x i32> %{{.*}}, <16 x i8> %{{.*}}, <16 x i8> %{{.*}})
return _mm_dpbusds_avx_epi32(__S, __A, __B);
}

Expand Down
Loading