Craig Topper | 934f86a | 2018-05-04 21:45:25 +0000 | [diff] [blame] | 1 | /*===---- avx512vlcdintrin.h - AVX512VL and AVX512CD intrinsics ------------=== |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 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 | #ifndef __IMMINTRIN_H |
| 24 | #error "Never use <avx512vlcdintrin.h> directly; include <immintrin.h> instead." |
| 25 | #endif |
| 26 | |
| 27 | #ifndef __AVX512VLCDINTRIN_H |
| 28 | #define __AVX512VLCDINTRIN_H |
| 29 | |
| 30 | /* Define the default attributes for the functions in this file. */ |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 31 | #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512cd"), __min_vector_width__(128))) |
| 32 | #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512cd"), __min_vector_width__(256))) |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 33 | |
| 34 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 35 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 36 | _mm_broadcastmb_epi64 (__mmask8 __A) |
Fangrui Song | 6907ce2 | 2018-07-30 19:24:48 +0000 | [diff] [blame] | 37 | { |
Jina Nahias | 48e298b | 2017-11-06 07:04:12 +0000 | [diff] [blame] | 38 | return (__m128i) _mm_set1_epi64x((long long) __A); |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 39 | } |
| 40 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 41 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 42 | _mm256_broadcastmb_epi64 (__mmask8 __A) |
| 43 | { |
Jina Nahias | 48e298b | 2017-11-06 07:04:12 +0000 | [diff] [blame] | 44 | return (__m256i) _mm256_set1_epi64x((long long)__A); |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 45 | } |
| 46 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 47 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 48 | _mm_broadcastmw_epi32 (__mmask16 __A) |
| 49 | { |
Jina Nahias | 48e298b | 2017-11-06 07:04:12 +0000 | [diff] [blame] | 50 | return (__m128i) _mm_set1_epi32((int)__A); |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 51 | } |
| 52 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 53 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 54 | _mm256_broadcastmw_epi32 (__mmask16 __A) |
| 55 | { |
Jina Nahias | 48e298b | 2017-11-06 07:04:12 +0000 | [diff] [blame] | 56 | return (__m256i) _mm256_set1_epi32((int)__A); |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 57 | } |
| 58 | |
| 59 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 60 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 61 | _mm_conflict_epi64 (__m128i __A) |
| 62 | { |
| 63 | return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A, |
| 64 | (__v2di) _mm_undefined_si128 (), |
| 65 | (__mmask8) -1); |
| 66 | } |
| 67 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 68 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 69 | _mm_mask_conflict_epi64 (__m128i __W, __mmask8 __U, __m128i __A) |
| 70 | { |
| 71 | return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A, |
| 72 | (__v2di) __W, |
| 73 | (__mmask8) __U); |
| 74 | } |
| 75 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 76 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 77 | _mm_maskz_conflict_epi64 (__mmask8 __U, __m128i __A) |
| 78 | { |
| 79 | return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A, |
| 80 | (__v2di) |
Craig Topper | dff5b31 | 2018-05-30 18:02:11 +0000 | [diff] [blame] | 81 | _mm_setzero_si128 (), |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 82 | (__mmask8) __U); |
| 83 | } |
| 84 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 85 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 86 | _mm256_conflict_epi64 (__m256i __A) |
| 87 | { |
| 88 | return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A, |
| 89 | (__v4di) _mm256_undefined_si256 (), |
| 90 | (__mmask8) -1); |
| 91 | } |
| 92 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 93 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 94 | _mm256_mask_conflict_epi64 (__m256i __W, __mmask8 __U, __m256i __A) |
| 95 | { |
| 96 | return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A, |
| 97 | (__v4di) __W, |
| 98 | (__mmask8) __U); |
| 99 | } |
| 100 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 101 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 102 | _mm256_maskz_conflict_epi64 (__mmask8 __U, __m256i __A) |
| 103 | { |
| 104 | return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A, |
| 105 | (__v4di) _mm256_setzero_si256 (), |
| 106 | (__mmask8) __U); |
| 107 | } |
| 108 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 109 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 110 | _mm_conflict_epi32 (__m128i __A) |
| 111 | { |
| 112 | return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A, |
| 113 | (__v4si) _mm_undefined_si128 (), |
| 114 | (__mmask8) -1); |
| 115 | } |
| 116 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 117 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 118 | _mm_mask_conflict_epi32 (__m128i __W, __mmask8 __U, __m128i __A) |
| 119 | { |
| 120 | return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A, |
| 121 | (__v4si) __W, |
| 122 | (__mmask8) __U); |
| 123 | } |
| 124 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 125 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 126 | _mm_maskz_conflict_epi32 (__mmask8 __U, __m128i __A) |
| 127 | { |
| 128 | return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A, |
| 129 | (__v4si) _mm_setzero_si128 (), |
| 130 | (__mmask8) __U); |
| 131 | } |
| 132 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 133 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 134 | _mm256_conflict_epi32 (__m256i __A) |
| 135 | { |
| 136 | return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A, |
| 137 | (__v8si) _mm256_undefined_si256 (), |
| 138 | (__mmask8) -1); |
| 139 | } |
| 140 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 141 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 142 | _mm256_mask_conflict_epi32 (__m256i __W, __mmask8 __U, __m256i __A) |
| 143 | { |
| 144 | return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A, |
| 145 | (__v8si) __W, |
| 146 | (__mmask8) __U); |
| 147 | } |
| 148 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 149 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 514f055 | 2016-04-27 15:35:13 +0000 | [diff] [blame] | 150 | _mm256_maskz_conflict_epi32 (__mmask8 __U, __m256i __A) |
| 151 | { |
| 152 | return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A, |
| 153 | (__v8si) |
| 154 | _mm256_setzero_si256 (), |
| 155 | (__mmask8) __U); |
| 156 | } |
| 157 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 158 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 159 | _mm_lzcnt_epi32 (__m128i __A) |
| 160 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 161 | return (__m128i) __builtin_ia32_vplzcntd_128 ((__v4si) __A); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 162 | } |
| 163 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 164 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 165 | _mm_mask_lzcnt_epi32 (__m128i __W, __mmask8 __U, __m128i __A) |
| 166 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 167 | return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, |
| 168 | (__v4si)_mm_lzcnt_epi32(__A), |
| 169 | (__v4si)__W); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 170 | } |
| 171 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 172 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 173 | _mm_maskz_lzcnt_epi32 (__mmask8 __U, __m128i __A) |
| 174 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 175 | return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U, |
| 176 | (__v4si)_mm_lzcnt_epi32(__A), |
| 177 | (__v4si)_mm_setzero_si128()); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 178 | } |
| 179 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 180 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 181 | _mm256_lzcnt_epi32 (__m256i __A) |
| 182 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 183 | return (__m256i) __builtin_ia32_vplzcntd_256 ((__v8si) __A); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 184 | } |
| 185 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 186 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 187 | _mm256_mask_lzcnt_epi32 (__m256i __W, __mmask8 __U, __m256i __A) |
| 188 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 189 | return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U, |
| 190 | (__v8si)_mm256_lzcnt_epi32(__A), |
| 191 | (__v8si)__W); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 192 | } |
| 193 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 194 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 195 | _mm256_maskz_lzcnt_epi32 (__mmask8 __U, __m256i __A) |
| 196 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 197 | return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U, |
| 198 | (__v8si)_mm256_lzcnt_epi32(__A), |
| 199 | (__v8si)_mm256_setzero_si256()); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 200 | } |
| 201 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 202 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 203 | _mm_lzcnt_epi64 (__m128i __A) |
| 204 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 205 | return (__m128i) __builtin_ia32_vplzcntq_128 ((__v2di) __A); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 206 | } |
| 207 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 208 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 209 | _mm_mask_lzcnt_epi64 (__m128i __W, __mmask8 __U, __m128i __A) |
| 210 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 211 | return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, |
| 212 | (__v2di)_mm_lzcnt_epi64(__A), |
| 213 | (__v2di)__W); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 214 | } |
| 215 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 216 | static __inline__ __m128i __DEFAULT_FN_ATTRS128 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 217 | _mm_maskz_lzcnt_epi64 (__mmask8 __U, __m128i __A) |
| 218 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 219 | return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U, |
| 220 | (__v2di)_mm_lzcnt_epi64(__A), |
Craig Topper | dff5b31 | 2018-05-30 18:02:11 +0000 | [diff] [blame] | 221 | (__v2di)_mm_setzero_si128()); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 222 | } |
| 223 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 224 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 225 | _mm256_lzcnt_epi64 (__m256i __A) |
| 226 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 227 | return (__m256i) __builtin_ia32_vplzcntq_256 ((__v4di) __A); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 228 | } |
| 229 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 230 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 231 | _mm256_mask_lzcnt_epi64 (__m256i __W, __mmask8 __U, __m256i __A) |
| 232 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 233 | return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, |
| 234 | (__v4di)_mm256_lzcnt_epi64(__A), |
| 235 | (__v4di)__W); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 236 | } |
| 237 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 238 | static __inline__ __m256i __DEFAULT_FN_ATTRS256 |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 239 | _mm256_maskz_lzcnt_epi64 (__mmask8 __U, __m256i __A) |
| 240 | { |
Craig Topper | 8e3689c | 2018-05-22 20:48:24 +0000 | [diff] [blame] | 241 | return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U, |
| 242 | (__v4di)_mm256_lzcnt_epi64(__A), |
| 243 | (__v4di)_mm256_setzero_si256()); |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 244 | } |
| 245 | |
Craig Topper | 74c10e3 | 2018-07-09 19:00:16 +0000 | [diff] [blame] | 246 | #undef __DEFAULT_FN_ATTRS128 |
| 247 | #undef __DEFAULT_FN_ATTRS256 |
Michael Zuckerman | 8c2900f | 2016-04-27 11:43:14 +0000 | [diff] [blame] | 248 | |
Craig Topper | 879b097 | 2016-06-21 06:53:58 +0000 | [diff] [blame] | 249 | #endif /* __AVX512VLCDINTRIN_H */ |