1 /*===---- avx512dqintrin.h - AVX512DQ intrinsics ---------------------------===
2 *
3 * Permission is hereby granted, free of charge, to any person obtaining a copy
4 * of this software and associated documentation files (the "Software"), to deal
5 * in the Software without restriction, including without limitation the rights
6 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
7 * copies of the Software, and to permit persons to whom the Software is
8 * furnished to do so, subject to the following conditions:
9 *
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
18 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
19 * THE SOFTWARE.
20 *
21 *===-----------------------------------------------------------------------===
22 */
23
24 #ifndef __IMMINTRIN_H
25 #error "Never use <avx512dqintrin.h> directly; include <immintrin.h> instead."
26 #endif
27
28 #ifndef __AVX512DQINTRIN_H
29 #define __AVX512DQINTRIN_H
30
31 /* Define the default attributes for the functions in this file. */
32 #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__))
33
34 static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mullo_epi64(__m512i __A,__m512i __B)35 _mm512_mullo_epi64 (__m512i __A, __m512i __B) {
36 return (__m512i) ((__v8di) __A * (__v8di) __B);
37 }
38
39 static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_mask_mullo_epi64(__m512i __W,__mmask8 __U,__m512i __A,__m512i __B)40 _mm512_mask_mullo_epi64 (__m512i __W, __mmask8 __U, __m512i __A, __m512i __B) {
41 return (__m512i) __builtin_ia32_pmullq512_mask ((__v8di) __A,
42 (__v8di) __B,
43 (__v8di) __W,
44 (__mmask8) __U);
45 }
46
47 static __inline__ __m512i __DEFAULT_FN_ATTRS
_mm512_maskz_mullo_epi64(__mmask8 __U,__m512i __A,__m512i __B)48 _mm512_maskz_mullo_epi64 (__mmask8 __U, __m512i __A, __m512i __B) {
49 return (__m512i) __builtin_ia32_pmullq512_mask ((__v8di) __A,
50 (__v8di) __B,
51 (__v8di)
52 _mm512_setzero_si512 (),
53 (__mmask8) __U);
54 }
55
56 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_xor_pd(__m512d __A,__m512d __B)57 _mm512_xor_pd (__m512d __A, __m512d __B) {
58 return (__m512d) ((__v8di) __A ^ (__v8di) __B);
59 }
60
61 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_xor_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)62 _mm512_mask_xor_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
63 return (__m512d) __builtin_ia32_xorpd512_mask ((__v8df) __A,
64 (__v8df) __B,
65 (__v8df) __W,
66 (__mmask8) __U);
67 }
68
69 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_xor_pd(__mmask8 __U,__m512d __A,__m512d __B)70 _mm512_maskz_xor_pd (__mmask8 __U, __m512d __A, __m512d __B) {
71 return (__m512d) __builtin_ia32_xorpd512_mask ((__v8df) __A,
72 (__v8df) __B,
73 (__v8df)
74 _mm512_setzero_pd (),
75 (__mmask8) __U);
76 }
77
78 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_xor_ps(__m512 __A,__m512 __B)79 _mm512_xor_ps (__m512 __A, __m512 __B) {
80 return (__m512) ((__v16si) __A ^ (__v16si) __B);
81 }
82
83 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_xor_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)84 _mm512_mask_xor_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
85 return (__m512) __builtin_ia32_xorps512_mask ((__v16sf) __A,
86 (__v16sf) __B,
87 (__v16sf) __W,
88 (__mmask16) __U);
89 }
90
91 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_xor_ps(__mmask16 __U,__m512 __A,__m512 __B)92 _mm512_maskz_xor_ps (__mmask16 __U, __m512 __A, __m512 __B) {
93 return (__m512) __builtin_ia32_xorps512_mask ((__v16sf) __A,
94 (__v16sf) __B,
95 (__v16sf)
96 _mm512_setzero_ps (),
97 (__mmask16) __U);
98 }
99
100 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_or_pd(__m512d __A,__m512d __B)101 _mm512_or_pd (__m512d __A, __m512d __B) {
102 return (__m512d) ((__v8di) __A | (__v8di) __B);
103 }
104
105 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_or_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)106 _mm512_mask_or_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
107 return (__m512d) __builtin_ia32_orpd512_mask ((__v8df) __A,
108 (__v8df) __B,
109 (__v8df) __W,
110 (__mmask8) __U);
111 }
112
113 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_or_pd(__mmask8 __U,__m512d __A,__m512d __B)114 _mm512_maskz_or_pd (__mmask8 __U, __m512d __A, __m512d __B) {
115 return (__m512d) __builtin_ia32_orpd512_mask ((__v8df) __A,
116 (__v8df) __B,
117 (__v8df)
118 _mm512_setzero_pd (),
119 (__mmask8) __U);
120 }
121
122 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_or_ps(__m512 __A,__m512 __B)123 _mm512_or_ps (__m512 __A, __m512 __B) {
124 return (__m512) ((__v16si) __A | (__v16si) __B);
125 }
126
127 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_or_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)128 _mm512_mask_or_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
129 return (__m512) __builtin_ia32_orps512_mask ((__v16sf) __A,
130 (__v16sf) __B,
131 (__v16sf) __W,
132 (__mmask16) __U);
133 }
134
135 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_or_ps(__mmask16 __U,__m512 __A,__m512 __B)136 _mm512_maskz_or_ps (__mmask16 __U, __m512 __A, __m512 __B) {
137 return (__m512) __builtin_ia32_orps512_mask ((__v16sf) __A,
138 (__v16sf) __B,
139 (__v16sf)
140 _mm512_setzero_ps (),
141 (__mmask16) __U);
142 }
143
144 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_and_pd(__m512d __A,__m512d __B)145 _mm512_and_pd (__m512d __A, __m512d __B) {
146 return (__m512d) ((__v8di) __A & (__v8di) __B);
147 }
148
149 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_and_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)150 _mm512_mask_and_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
151 return (__m512d) __builtin_ia32_andpd512_mask ((__v8df) __A,
152 (__v8df) __B,
153 (__v8df) __W,
154 (__mmask8) __U);
155 }
156
157 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_and_pd(__mmask8 __U,__m512d __A,__m512d __B)158 _mm512_maskz_and_pd (__mmask8 __U, __m512d __A, __m512d __B) {
159 return (__m512d) __builtin_ia32_andpd512_mask ((__v8df) __A,
160 (__v8df) __B,
161 (__v8df)
162 _mm512_setzero_pd (),
163 (__mmask8) __U);
164 }
165
166 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_and_ps(__m512 __A,__m512 __B)167 _mm512_and_ps (__m512 __A, __m512 __B) {
168 return (__m512) ((__v16si) __A & (__v16si) __B);
169 }
170
171 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_and_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)172 _mm512_mask_and_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
173 return (__m512) __builtin_ia32_andps512_mask ((__v16sf) __A,
174 (__v16sf) __B,
175 (__v16sf) __W,
176 (__mmask16) __U);
177 }
178
179 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_and_ps(__mmask16 __U,__m512 __A,__m512 __B)180 _mm512_maskz_and_ps (__mmask16 __U, __m512 __A, __m512 __B) {
181 return (__m512) __builtin_ia32_andps512_mask ((__v16sf) __A,
182 (__v16sf) __B,
183 (__v16sf)
184 _mm512_setzero_ps (),
185 (__mmask16) __U);
186 }
187
188 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_andnot_pd(__m512d __A,__m512d __B)189 _mm512_andnot_pd (__m512d __A, __m512d __B) {
190 return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
191 (__v8df) __B,
192 (__v8df)
193 _mm512_setzero_pd (),
194 (__mmask8) -1);
195 }
196
197 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_mask_andnot_pd(__m512d __W,__mmask8 __U,__m512d __A,__m512d __B)198 _mm512_mask_andnot_pd (__m512d __W, __mmask8 __U, __m512d __A, __m512d __B) {
199 return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
200 (__v8df) __B,
201 (__v8df) __W,
202 (__mmask8) __U);
203 }
204
205 static __inline__ __m512d __DEFAULT_FN_ATTRS
_mm512_maskz_andnot_pd(__mmask8 __U,__m512d __A,__m512d __B)206 _mm512_maskz_andnot_pd (__mmask8 __U, __m512d __A, __m512d __B) {
207 return (__m512d) __builtin_ia32_andnpd512_mask ((__v8df) __A,
208 (__v8df) __B,
209 (__v8df)
210 _mm512_setzero_pd (),
211 (__mmask8) __U);
212 }
213
214 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_andnot_ps(__m512 __A,__m512 __B)215 _mm512_andnot_ps (__m512 __A, __m512 __B) {
216 return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
217 (__v16sf) __B,
218 (__v16sf)
219 _mm512_setzero_ps (),
220 (__mmask16) -1);
221 }
222
223 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_mask_andnot_ps(__m512 __W,__mmask16 __U,__m512 __A,__m512 __B)224 _mm512_mask_andnot_ps (__m512 __W, __mmask16 __U, __m512 __A, __m512 __B) {
225 return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
226 (__v16sf) __B,
227 (__v16sf) __W,
228 (__mmask16) __U);
229 }
230
231 static __inline__ __m512 __DEFAULT_FN_ATTRS
_mm512_maskz_andnot_ps(__mmask16 __U,__m512 __A,__m512 __B)232 _mm512_maskz_andnot_ps (__mmask16 __U, __m512 __A, __m512 __B) {
233 return (__m512) __builtin_ia32_andnps512_mask ((__v16sf) __A,
234 (__v16sf) __B,
235 (__v16sf)
236 _mm512_setzero_ps (),
237 (__mmask16) __U);
238 }
239
240 #undef __DEFAULT_FN_ATTRS
241
242 #endif
243