Mercurial > hg > CbC > CbC_llvm

comparison clang/lib/Headers/avx512vlvnniintrin.h @ 236:c4bab56944e8 llvm-original

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
LLVM 16
author kono
date Wed, 09 Nov 2022 17:45:10 +0900
parents 79ff65ed7e25
children
comparison
equal deleted inserted replaced
232:70dce7da266c 236:c4bab56944e8
23 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer 23 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
24 /// in \a S, and store the packed 32-bit results in DST. 24 /// in \a S, and store the packed 32-bit results in DST.
25 /// 25 ///
26 /// This intrinsic corresponds to the <c> VPDPBUSD </c> instructions. 26 /// This intrinsic corresponds to the <c> VPDPBUSD </c> instructions.
27 /// 27 ///
28 /// \operation 28 /// \code{.operation}
29 /// FOR j := 0 to 7 29 /// FOR j := 0 to 7
30 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j])) 30 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
31 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1])) 31 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
32 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2])) 32 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
33 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3])) 33 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
34 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4 34 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4
35 /// ENDFOR 35 /// ENDFOR
36 /// DST[MAX:256] := 0 36 /// DST[MAX:256] := 0
37 /// \endoperation 37 /// \endcode
38 #define _mm256_dpbusd_epi32(S, A, B) \ 38 #define _mm256_dpbusd_epi32(S, A, B) \
39 (__m256i)__builtin_ia32_vpdpbusd256((__v8si)(S), (__v8si)(A), (__v8si)(B)) 39 ((__m256i)__builtin_ia32_vpdpbusd256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
40 40
41 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with 41 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with
42 /// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed 42 /// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed
43 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer 43 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
44 /// in \a S using signed saturation, and store the packed 32-bit results in DST. 44 /// in \a S using signed saturation, and store the packed 32-bit results in DST.
45 /// 45 ///
46 /// This intrinsic corresponds to the <c> VPDPBUSDS </c> instructions. 46 /// This intrinsic corresponds to the <c> VPDPBUSDS </c> instructions.
47 /// 47 ///
48 /// \operation 48 /// \code{.operation}
49 /// FOR j := 0 to 7 49 /// FOR j := 0 to 7
50 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j])) 50 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
51 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1])) 51 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
52 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2])) 52 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
53 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3])) 53 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
54 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4) 54 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4)
55 /// ENDFOR 55 /// ENDFOR
56 /// DST[MAX:256] := 0 56 /// DST[MAX:256] := 0
57 /// \endoperation 57 /// \endcode
58 #define _mm256_dpbusds_epi32(S, A, B) \ 58 #define _mm256_dpbusds_epi32(S, A, B) \
59 (__m256i)__builtin_ia32_vpdpbusds256((__v8si)(S), (__v8si)(A), (__v8si)(B)) 59 ((__m256i)__builtin_ia32_vpdpbusds256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
60 60
61 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with 61 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
62 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit 62 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
63 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S, 63 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S,
64 /// and store the packed 32-bit results in DST. 64 /// and store the packed 32-bit results in DST.
65 /// 65 ///
66 /// This intrinsic corresponds to the <c> VPDPWSSD </c> instructions. 66 /// This intrinsic corresponds to the <c> VPDPWSSD </c> instructions.
67 /// 67 ///
68 /// \operation 68 /// \code{.operation}
69 /// FOR j := 0 to 7 69 /// FOR j := 0 to 7
70 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j]) 70 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
71 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1]) 71 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
72 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2 72 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2
73 /// ENDFOR 73 /// ENDFOR
74 /// DST[MAX:256] := 0 74 /// DST[MAX:256] := 0
75 /// \endoperation 75 /// \endcode
76 #define _mm256_dpwssd_epi32(S, A, B) \ 76 #define _mm256_dpwssd_epi32(S, A, B) \
77 (__m256i)__builtin_ia32_vpdpwssd256((__v8si)(S), (__v8si)(A), (__v8si)(B)) 77 ((__m256i)__builtin_ia32_vpdpwssd256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
78 78
79 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with 79 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
80 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit 80 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
81 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S 81 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S
82 /// using signed saturation, and store the packed 32-bit results in DST. 82 /// using signed saturation, and store the packed 32-bit results in DST.
83 /// 83 ///
84 /// This intrinsic corresponds to the <c> VPDPWSSDS </c> instructions. 84 /// This intrinsic corresponds to the <c> VPDPWSSDS </c> instructions.
85 /// 85 ///
86 /// \operation 86 /// \code{.operation}
87 /// FOR j := 0 to 7 87 /// FOR j := 0 to 7
88 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j]) 88 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
89 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1]) 89 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
90 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2) 90 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2)
91 /// ENDFOR 91 /// ENDFOR
92 /// DST[MAX:256] := 0 92 /// DST[MAX:256] := 0
93 /// \endoperation 93 /// \endcode
94 #define _mm256_dpwssds_epi32(S, A, B) \ 94 #define _mm256_dpwssds_epi32(S, A, B) \
95 (__m256i)__builtin_ia32_vpdpwssds256((__v8si)(S), (__v8si)(A), (__v8si)(B)) 95 ((__m256i)__builtin_ia32_vpdpwssds256((__v8si)(S), (__v8si)(A), (__v8si)(B)))
96 96
97 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with 97 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with
98 /// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed 98 /// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed
99 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer 99 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
100 /// in \a S, and store the packed 32-bit results in DST. 100 /// in \a S, and store the packed 32-bit results in DST.
101 /// 101 ///
102 /// This intrinsic corresponds to the <c> VPDPBUSD </c> instructions. 102 /// This intrinsic corresponds to the <c> VPDPBUSD </c> instructions.
103 /// 103 ///
104 /// \operation 104 /// \code{.operation}
105 /// FOR j := 0 to 3 105 /// FOR j := 0 to 3
106 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j])) 106 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
107 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1])) 107 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
108 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2])) 108 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
109 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3])) 109 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
110 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4 110 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4
111 /// ENDFOR 111 /// ENDFOR
112 /// DST[MAX:128] := 0 112 /// DST[MAX:128] := 0
113 /// \endoperation 113 /// \endcode
114 #define _mm_dpbusd_epi32(S, A, B) \ 114 #define _mm_dpbusd_epi32(S, A, B) \
115 (__m128i)__builtin_ia32_vpdpbusd128((__v4si)(S), (__v4si)(A), (__v4si)(B)) 115 ((__m128i)__builtin_ia32_vpdpbusd128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
116 116
117 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with 117 /// Multiply groups of 4 adjacent pairs of unsigned 8-bit integers in \a A with
118 /// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed 118 /// corresponding signed 8-bit integers in \a B, producing 4 intermediate signed
119 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer 119 /// 16-bit results. Sum these 4 results with the corresponding 32-bit integer
120 /// in \a S using signed saturation, and store the packed 32-bit results in DST. 120 /// in \a S using signed saturation, and store the packed 32-bit results in DST.
121 /// 121 ///
122 /// This intrinsic corresponds to the <c> VPDPBUSDS </c> instructions. 122 /// This intrinsic corresponds to the <c> VPDPBUSDS </c> instructions.
123 /// 123 ///
124 /// \operation 124 /// \code{.operation}
125 /// FOR j := 0 to 3 125 /// FOR j := 0 to 3
126 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j])) 126 /// tmp1.word := Signed(ZeroExtend16(A.byte[4*j]) * SignExtend16(B.byte[4*j]))
127 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1])) 127 /// tmp2.word := Signed(ZeroExtend16(A.byte[4*j+1]) * SignExtend16(B.byte[4*j+1]))
128 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2])) 128 /// tmp3.word := Signed(ZeroExtend16(A.byte[4*j+2]) * SignExtend16(B.byte[4*j+2]))
129 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3])) 129 /// tmp4.word := Signed(ZeroExtend16(A.byte[4*j+3]) * SignExtend16(B.byte[4*j+3]))
130 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4) 130 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2 + tmp3 + tmp4)
131 /// ENDFOR 131 /// ENDFOR
132 /// DST[MAX:128] := 0 132 /// DST[MAX:128] := 0
133 /// \endoperation 133 /// \endcode
134 #define _mm_dpbusds_epi32(S, A, B) \ 134 #define _mm_dpbusds_epi32(S, A, B) \
135 (__m128i)__builtin_ia32_vpdpbusds128((__v4si)(S), (__v4si)(A), (__v4si)(B)) 135 ((__m128i)__builtin_ia32_vpdpbusds128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
136 136
137 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with 137 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
138 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit 138 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
139 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S, 139 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S,
140 /// and store the packed 32-bit results in DST. 140 /// and store the packed 32-bit results in DST.
141 /// 141 ///
142 /// This intrinsic corresponds to the <c> VPDPWSSD </c> instructions. 142 /// This intrinsic corresponds to the <c> VPDPWSSD </c> instructions.
143 /// 143 ///
144 /// \operation 144 /// \code{.operation}
145 /// FOR j := 0 to 3 145 /// FOR j := 0 to 3
146 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j]) 146 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
147 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1]) 147 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
148 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2 148 /// DST.dword[j] := S.dword[j] + tmp1 + tmp2
149 /// ENDFOR 149 /// ENDFOR
150 /// DST[MAX:128] := 0 150 /// DST[MAX:128] := 0
151 /// \endoperation 151 /// \endcode
152 #define _mm_dpwssd_epi32(S, A, B) \ 152 #define _mm_dpwssd_epi32(S, A, B) \
153 (__m128i)__builtin_ia32_vpdpwssd128((__v4si)(S), (__v4si)(A), (__v4si)(B)) 153 ((__m128i)__builtin_ia32_vpdpwssd128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
154 154
155 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with 155 /// Multiply groups of 2 adjacent pairs of signed 16-bit integers in \a A with
156 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit 156 /// corresponding 16-bit integers in \a B, producing 2 intermediate signed 32-bit
157 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S 157 /// results. Sum these 2 results with the corresponding 32-bit integer in \a S
158 /// using signed saturation, and store the packed 32-bit results in DST. 158 /// using signed saturation, and store the packed 32-bit results in DST.
159 /// 159 ///
160 /// This intrinsic corresponds to the <c> VPDPWSSDS </c> instructions. 160 /// This intrinsic corresponds to the <c> VPDPWSSDS </c> instructions.
161 /// 161 ///
162 /// \operation 162 /// \code{.operation}
163 /// FOR j := 0 to 3 163 /// FOR j := 0 to 3
164 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j]) 164 /// tmp1.dword := SignExtend32(A.word[2*j]) * SignExtend32(B.word[2*j])
165 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1]) 165 /// tmp2.dword := SignExtend32(A.word[2*j+1]) * SignExtend32(B.word[2*j+1])
166 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2) 166 /// DST.dword[j] := Saturate32(S.dword[j] + tmp1 + tmp2)
167 /// ENDFOR 167 /// ENDFOR
168 /// DST[MAX:128] := 0 168 /// DST[MAX:128] := 0
169 /// \endoperation 169 /// \endcode
170 #define _mm_dpwssds_epi32(S, A, B) \ 170 #define _mm_dpwssds_epi32(S, A, B) \
171 (__m128i)__builtin_ia32_vpdpwssds128((__v4si)(S), (__v4si)(A), (__v4si)(B)) 171 ((__m128i)__builtin_ia32_vpdpwssds128((__v4si)(S), (__v4si)(A), (__v4si)(B)))
172 172
173 static __inline__ __m256i __DEFAULT_FN_ATTRS256 173 static __inline__ __m256i __DEFAULT_FN_ATTRS256
174 _mm256_mask_dpbusd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B) 174 _mm256_mask_dpbusd_epi32(__m256i __S, __mmask8 __U, __m256i __A, __m256i __B)
175 { 175 {
176 return (__m256i)__builtin_ia32_selectd_256(__U, 176 return (__m256i)__builtin_ia32_selectd_256(__U,