Warning: This file is not a C or C++ file. It does not have highlighting.
1 | /*===---- avx512vlcdintrin.h - AVX512VL and AVX512CD intrinsics ------------=== |
---|---|
2 | * |
3 | * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions. |
4 | * See https://llvm.org/LICENSE.txt for license information. |
5 | * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception |
6 | * |
7 | *===-----------------------------------------------------------------------=== |
8 | */ |
9 | #ifndef __IMMINTRIN_H |
10 | #error "Never use <avx512vlcdintrin.h> directly; include <immintrin.h> instead." |
11 | #endif |
12 | |
13 | #ifndef __AVX512VLCDINTRIN_H |
14 | #define __AVX512VLCDINTRIN_H |
15 | |
16 | /* Define the default attributes for the functions in this file. */ |
17 | #define __DEFAULT_FN_ATTRS128 \ |
18 | __attribute__((__always_inline__, __nodebug__, \ |
19 | __target__("avx512vl,avx512cd,no-evex512"), \ |
20 | __min_vector_width__(128))) |
21 | #define __DEFAULT_FN_ATTRS256 \ |
22 | __attribute__((__always_inline__, __nodebug__, \ |
23 | __target__("avx512vl,avx512cd,no-evex512"), \ |
24 | __min_vector_width__(256))) |
25 | |
26 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
27 | _mm_broadcastmb_epi64 (__mmask8 __A) |
28 | { |
29 | return (__m128i) _mm_set1_epi64x((long long) __A); |
30 | } |
31 | |
32 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
33 | _mm256_broadcastmb_epi64 (__mmask8 __A) |
34 | { |
35 | return (__m256i) _mm256_set1_epi64x((long long)__A); |
36 | } |
37 | |
38 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
39 | _mm_broadcastmw_epi32 (__mmask16 __A) |
40 | { |
41 | return (__m128i) _mm_set1_epi32((int)__A); |
42 | } |
43 | |
44 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
45 | _mm256_broadcastmw_epi32 (__mmask16 __A) |
46 | { |
47 | return (__m256i) _mm256_set1_epi32((int)__A); |
48 | } |
49 | |
50 | |
51 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
52 | _mm_conflict_epi64 (__m128i __A) |
53 | { |
54 | return (__m128i) __builtin_ia32_vpconflictdi_128 ((__v2di) __A); |
55 | } |
56 | |
57 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
58 | _mm_mask_conflict_epi64 (__m128i __W, __mmask8 __U, __m128i __A) |
59 | { |
60 | return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, |
61 | (__v2di)_mm_conflict_epi64(__A), |
62 | (__v2di)__W); |
63 | } |
64 | |
65 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
66 | _mm_maskz_conflict_epi64 (__mmask8 __U, __m128i __A) |
67 | { |
68 | return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, |
69 | (__v2di)_mm_conflict_epi64(__A), |
70 | (__v2di)_mm_setzero_si128()); |
71 | } |
72 | |
73 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
74 | _mm256_conflict_epi64 (__m256i __A) |
75 | { |
76 | return (__m256i) __builtin_ia32_vpconflictdi_256 ((__v4di) __A); |
77 | } |
78 | |
79 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
80 | _mm256_mask_conflict_epi64 (__m256i __W, __mmask8 __U, __m256i __A) |
81 | { |
82 | return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, |
83 | (__v4di)_mm256_conflict_epi64(__A), |
84 | (__v4di)__W); |
85 | } |
86 | |
87 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
88 | _mm256_maskz_conflict_epi64 (__mmask8 __U, __m256i __A) |
89 | { |
90 | return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, |
91 | (__v4di)_mm256_conflict_epi64(__A), |
92 | (__v4di)_mm256_setzero_si256()); |
93 | } |
94 | |
95 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
96 | _mm_conflict_epi32 (__m128i __A) |
97 | { |
98 | return (__m128i) __builtin_ia32_vpconflictsi_128 ((__v4si) __A); |
99 | } |
100 | |
101 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
102 | _mm_mask_conflict_epi32 (__m128i __W, __mmask8 __U, __m128i __A) |
103 | { |
104 | return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, |
105 | (__v4si)_mm_conflict_epi32(__A), |
106 | (__v4si)__W); |
107 | } |
108 | |
109 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
110 | _mm_maskz_conflict_epi32 (__mmask8 __U, __m128i __A) |
111 | { |
112 | return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, |
113 | (__v4si)_mm_conflict_epi32(__A), |
114 | (__v4si)_mm_setzero_si128()); |
115 | } |
116 | |
117 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
118 | _mm256_conflict_epi32 (__m256i __A) |
119 | { |
120 | return (__m256i) __builtin_ia32_vpconflictsi_256 ((__v8si) __A); |
121 | } |
122 | |
123 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
124 | _mm256_mask_conflict_epi32 (__m256i __W, __mmask8 __U, __m256i __A) |
125 | { |
126 | return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U, |
127 | (__v8si)_mm256_conflict_epi32(__A), |
128 | (__v8si)__W); |
129 | } |
130 | |
131 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
132 | _mm256_maskz_conflict_epi32 (__mmask8 __U, __m256i __A) |
133 | { |
134 | return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U, |
135 | (__v8si)_mm256_conflict_epi32(__A), |
136 | (__v8si)_mm256_setzero_si256()); |
137 | } |
138 | |
139 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
140 | _mm_lzcnt_epi32 (__m128i __A) |
141 | { |
142 | return (__m128i) __builtin_ia32_vplzcntd_128 ((__v4si) __A); |
143 | } |
144 | |
145 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
146 | _mm_mask_lzcnt_epi32 (__m128i __W, __mmask8 __U, __m128i __A) |
147 | { |
148 | return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, |
149 | (__v4si)_mm_lzcnt_epi32(__A), |
150 | (__v4si)__W); |
151 | } |
152 | |
153 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
154 | _mm_maskz_lzcnt_epi32 (__mmask8 __U, __m128i __A) |
155 | { |
156 | return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, |
157 | (__v4si)_mm_lzcnt_epi32(__A), |
158 | (__v4si)_mm_setzero_si128()); |
159 | } |
160 | |
161 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
162 | _mm256_lzcnt_epi32 (__m256i __A) |
163 | { |
164 | return (__m256i) __builtin_ia32_vplzcntd_256 ((__v8si) __A); |
165 | } |
166 | |
167 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
168 | _mm256_mask_lzcnt_epi32 (__m256i __W, __mmask8 __U, __m256i __A) |
169 | { |
170 | return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U, |
171 | (__v8si)_mm256_lzcnt_epi32(__A), |
172 | (__v8si)__W); |
173 | } |
174 | |
175 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
176 | _mm256_maskz_lzcnt_epi32 (__mmask8 __U, __m256i __A) |
177 | { |
178 | return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U, |
179 | (__v8si)_mm256_lzcnt_epi32(__A), |
180 | (__v8si)_mm256_setzero_si256()); |
181 | } |
182 | |
183 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
184 | _mm_lzcnt_epi64 (__m128i __A) |
185 | { |
186 | return (__m128i) __builtin_ia32_vplzcntq_128 ((__v2di) __A); |
187 | } |
188 | |
189 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
190 | _mm_mask_lzcnt_epi64 (__m128i __W, __mmask8 __U, __m128i __A) |
191 | { |
192 | return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, |
193 | (__v2di)_mm_lzcnt_epi64(__A), |
194 | (__v2di)__W); |
195 | } |
196 | |
197 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
198 | _mm_maskz_lzcnt_epi64 (__mmask8 __U, __m128i __A) |
199 | { |
200 | return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, |
201 | (__v2di)_mm_lzcnt_epi64(__A), |
202 | (__v2di)_mm_setzero_si128()); |
203 | } |
204 | |
205 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
206 | _mm256_lzcnt_epi64 (__m256i __A) |
207 | { |
208 | return (__m256i) __builtin_ia32_vplzcntq_256 ((__v4di) __A); |
209 | } |
210 | |
211 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
212 | _mm256_mask_lzcnt_epi64 (__m256i __W, __mmask8 __U, __m256i __A) |
213 | { |
214 | return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, |
215 | (__v4di)_mm256_lzcnt_epi64(__A), |
216 | (__v4di)__W); |
217 | } |
218 | |
219 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
220 | _mm256_maskz_lzcnt_epi64 (__mmask8 __U, __m256i __A) |
221 | { |
222 | return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, |
223 | (__v4di)_mm256_lzcnt_epi64(__A), |
224 | (__v4di)_mm256_setzero_si256()); |
225 | } |
226 | |
227 | #undef __DEFAULT_FN_ATTRS128 |
228 | #undef __DEFAULT_FN_ATTRS256 |
229 | |
230 | #endif /* __AVX512VLCDINTRIN_H */ |
231 |
Warning: This file is not a C or C++ file. It does not have highlighting.