aboutsummaryrefslogtreecommitdiff
path: root/contrib/llvm-project/clang/lib/Headers/avx512vlvnniintrin.h
diff options
context:
space:
mode:
Diffstat (limited to 'contrib/llvm-project/clang/lib/Headers/avx512vlvnniintrin.h')
-rw-r--r--contrib/llvm-project/clang/lib/Headers/avx512vlvnniintrin.h205
1 files changed, 150 insertions, 55 deletions
diff --git a/contrib/llvm-project/clang/lib/Headers/avx512vlvnniintrin.h b/contrib/llvm-project/clang/lib/Headers/avx512vlvnniintrin.h
index b7c8fa08c653..71ac1b4370d4 100644
--- a/contrib/llvm-project/clang/lib/Headers/avx512vlvnniintrin.h
+++ b/contrib/llvm-project/clang/lib/Headers/avx512vlvnniintrin.h
@@ -18,13 +18,157 @@
#define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512vnni"), __min_vector_width__(128)))
#define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512vnni"), __min_vector_width__(256)))
+/// 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
+/// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
+/// in \a S, and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPBUSD </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 7
+/// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
+/// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
+/// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
+/// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
+/// DST.dword[j] := S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4
+/// ENDFOR
+/// DST[MAX:256] := 0
+/// \endoperation
+#define _mm256_dpbusd_epi32(S, A, B) \
+ (__m256i)__builtin_ia32_vpdpbusd256((__v8si)(S), (__v8si)(A), (__v8si)(B))
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_dpbusd_epi32(__m256i __S, __m256i __A, __m256i __B)
-{
- return (__m256i)__builtin_ia32_vpdpbusd256((__v8si)__S, (__v8si)__A,
- (__v8si)__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
+/// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
+/// in \a S using signed saturation, and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPBUSDS </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 7
+/// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
+/// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
+/// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
+/// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
+/// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4)
+/// ENDFOR
+/// DST[MAX:256] := 0
+/// \endoperation
+#define _mm256_dpbusds_epi32(S, A, B) \
+ (__m256i)__builtin_ia32_vpdpbusds256((__v8si)(S), (__v8si)(A), (__v8si)(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
+/// results. Sum these 2 results with the corresponding 32-bit integer in \a S,
+/// and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPWSSD </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 7
+/// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
+/// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
+/// DST.dword[j] := S.dword[j] + tmp1 + tmp2
+/// ENDFOR
+/// DST[MAX:256] := 0
+/// \endoperation
+#define _mm256_dpwssd_epi32(S, A, B) \
+ (__m256i)__builtin_ia32_vpdpwssd256((__v8si)(S), (__v8si)(A), (__v8si)(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
+/// results. Sum these 2 results with the corresponding 32-bit integer in \a S
+/// using signed saturation, and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPWSSDS </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 7
+/// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
+/// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
+/// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2)
+/// ENDFOR
+/// DST[MAX:256] := 0
+/// \endoperation
+#define _mm256_dpwssds_epi32(S, A, B) \
+ (__m256i)__builtin_ia32_vpdpwssds256((__v8si)(S), (__v8si)(A), (__v8si)(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
+/// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
+/// in \a S, and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPBUSD </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 3
+/// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
+/// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
+/// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
+/// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
+/// DST.dword[j] := S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4
+/// ENDFOR
+/// DST[MAX:128] := 0
+/// \endoperation
+#define _mm_dpbusd_epi32(S, A, B) \
+ (__m128i)__builtin_ia32_vpdpbusd128((__v4si)(S), (__v4si)(A), (__v4si)(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
+/// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
+/// in \a S using signed saturation, and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPBUSDS </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 3
+/// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
+/// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
+/// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
+/// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
+/// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4)
+/// ENDFOR
+/// DST[MAX:128] := 0
+/// \endoperation
+#define _mm_dpbusds_epi32(S, A, B) \
+ (__m128i)__builtin_ia32_vpdpbusds128((__v4si)(S), (__v4si)(A), (__v4si)(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
+/// results. Sum these 2 results with the corresponding 32-bit integer in \a S,
+/// and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPWSSD </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 3
+/// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
+/// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
+/// DST.dword[j] := S.dword[j] + tmp1 + tmp2
+/// ENDFOR
+/// DST[MAX:128] := 0
+/// \endoperation
+#define _mm_dpwssd_epi32(S, A, B) \
+ (__m128i)__builtin_ia32_vpdpwssd128((__v4si)(S), (__v4si)(A), (__v4si)(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
+/// results. Sum these 2 results with the corresponding 32-bit integer in \a S
+/// using signed saturation, and store the packed 32-bit results in DST.
+///
+/// This intrinsic corresponds to the <c> VPDPWSSDS </c> instructions.
+///
+/// \operation
+/// FOR j := 0 to 3
+/// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
+/// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
+/// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2)
+/// ENDFOR
+/// DST[MAX:128] := 0
+/// \endoperation
+#define _mm_dpwssds_epi32(S, A, B) \
+ (__m128i)__builtin_ia32_vpdpwssds128((__v4si)(S), (__v4si)(A), (__v4si)(B))
static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_mask_dpbusd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B)
@@ -43,13 +187,6 @@ _mm256_maskz_dpbusd_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B)
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_dpbusds_epi32(__m256i __S, __m256i __A, __m256i __B)
-{
- return (__m256i)__builtin_ia32_vpdpbusds256((__v8si)__S, (__v8si)__A,
- (__v8si)__B);
-}
-
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_mask_dpbusds_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B)
{
return (__m256i)__builtin_ia32_selectd_256(__U,
@@ -66,13 +203,6 @@ _mm256_maskz_dpbusds_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B)
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_dpwssd_epi32(__m256i __S, __m256i __A, __m256i __B)
-{
- return (__m256i)__builtin_ia32_vpdpwssd256((__v8si)__S, (__v8si)__A,
- (__v8si)__B);
-}
-
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_mask_dpwssd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B)
{
return (__m256i)__builtin_ia32_selectd_256(__U,
@@ -89,13 +219,6 @@ _mm256_maskz_dpwssd_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B)
}
static __inline__ __m256i __DEFAULT_FN_ATTRS256
-_mm256_dpwssds_epi32(__m256i __S, __m256i __A, __m256i __B)
-{
- return (__m256i)__builtin_ia32_vpdpwssds256((__v8si)__S, (__v8si)__A,
- (__v8si)__B);
-}
-
-static __inline__ __m256i __DEFAULT_FN_ATTRS256
_mm256_mask_dpwssds_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B)
{
return (__m256i)__builtin_ia32_selectd_256(__U,
@@ -112,13 +235,6 @@ _mm256_maskz_dpwssds_epi32(__mmask8 __U, __m256i __S, __m256i __A, __m256i __B)
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_dpbusd_epi32(__m128i __S, __m128i __A, __m128i __B)
-{
- return (__m128i)__builtin_ia32_vpdpbusd128((__v4si)__S, (__v4si)__A,
- (__v4si)__B);
-}
-
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_mask_dpbusd_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_selectd_128(__U,
@@ -135,13 +251,6 @@ _mm_maskz_dpbusd_epi32(__mmask8 __U, __m128i __S, __m128i __A, __m128i __B)
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_dpbusds_epi32(__m128i __S, __m128i __A, __m128i __B)
-{
- return (__m128i)__builtin_ia32_vpdpbusds128((__v4si)__S, (__v4si)__A,
- (__v4si)__B);
-}
-
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_mask_dpbusds_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_selectd_128(__U,
@@ -158,13 +267,6 @@ _mm_maskz_dpbusds_epi32(__mmask8 __U, __m128i __S, __m128i __A, __m128i __B)
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_dpwssd_epi32(__m128i __S, __m128i __A, __m128i __B)
-{
- return (__m128i)__builtin_ia32_vpdpwssd128((__v4si)__S, (__v4si)__A,
- (__v4si)__B);
-}
-
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_mask_dpwssd_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_selectd_128(__U,
@@ -181,13 +283,6 @@ _mm_maskz_dpwssd_epi32(__mmask8 __U, __m128i __S, __m128i __A, __m128i __B)
}
static __inline__ __m128i __DEFAULT_FN_ATTRS128
-_mm_dpwssds_epi32(__m128i __S, __m128i __A, __m128i __B)
-{
- return (__m128i)__builtin_ia32_vpdpwssds128((__v4si)__S, (__v4si)__A,
- (__v4si)__B);
-}
-
-static __inline__ __m128i __DEFAULT_FN_ATTRS128
_mm_mask_dpwssds_epi32(__m128i __S, __mmask8 __U, __m128i __A, __m128i __B)
{
return (__m128i)__builtin_ia32_selectd_128(__U,
generated by cgit v1.2.3 (git 2.25.1) at 2024-04-24 03:09:44 +0000