avx512vlcdintrin.h 8.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249
  1. /*===---- avx512vlcdintrin.h - AVX512VL and AVX512CD 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. #ifndef __IMMINTRIN_H
  24. #error "Never use <avx512vlcdintrin.h> directly; include <immintrin.h> instead."
  25. #endif
  26. #ifndef __AVX512VLCDINTRIN_H
  27. #define __AVX512VLCDINTRIN_H
  28. /* Define the default attributes for the functions in this file. */
  29. #define __DEFAULT_FN_ATTRS128 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512cd"), __min_vector_width__(128)))
  30. #define __DEFAULT_FN_ATTRS256 __attribute__((__always_inline__, __nodebug__, __target__("avx512vl,avx512cd"), __min_vector_width__(256)))
  31. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  32. _mm_broadcastmb_epi64 (__mmask8 __A)
  33. {
  34. return (__m128i) _mm_set1_epi64x((long long) __A);
  35. }
  36. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  37. _mm256_broadcastmb_epi64 (__mmask8 __A)
  38. {
  39. return (__m256i) _mm256_set1_epi64x((long long)__A);
  40. }
  41. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  42. _mm_broadcastmw_epi32 (__mmask16 __A)
  43. {
  44. return (__m128i) _mm_set1_epi32((int)__A);
  45. }
  46. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  47. _mm256_broadcastmw_epi32 (__mmask16 __A)
  48. {
  49. return (__m256i) _mm256_set1_epi32((int)__A);
  50. }
  51. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  52. _mm_conflict_epi64 (__m128i __A)
  53. {
  54. return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
  55. (__v2di) _mm_undefined_si128 (),
  56. (__mmask8) -1);
  57. }
  58. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  59. _mm_mask_conflict_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
  60. {
  61. return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
  62. (__v2di) __W,
  63. (__mmask8) __U);
  64. }
  65. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  66. _mm_maskz_conflict_epi64 (__mmask8 __U, __m128i __A)
  67. {
  68. return (__m128i) __builtin_ia32_vpconflictdi_128_mask ((__v2di) __A,
  69. (__v2di)
  70. _mm_setzero_si128 (),
  71. (__mmask8) __U);
  72. }
  73. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  74. _mm256_conflict_epi64 (__m256i __A)
  75. {
  76. return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
  77. (__v4di) _mm256_undefined_si256 (),
  78. (__mmask8) -1);
  79. }
  80. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  81. _mm256_mask_conflict_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
  82. {
  83. return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
  84. (__v4di) __W,
  85. (__mmask8) __U);
  86. }
  87. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  88. _mm256_maskz_conflict_epi64 (__mmask8 __U, __m256i __A)
  89. {
  90. return (__m256i) __builtin_ia32_vpconflictdi_256_mask ((__v4di) __A,
  91. (__v4di) _mm256_setzero_si256 (),
  92. (__mmask8) __U);
  93. }
  94. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  95. _mm_conflict_epi32 (__m128i __A)
  96. {
  97. return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
  98. (__v4si) _mm_undefined_si128 (),
  99. (__mmask8) -1);
  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_vpconflictsi_128_mask ((__v4si) __A,
  105. (__v4si) __W,
  106. (__mmask8) __U);
  107. }
  108. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  109. _mm_maskz_conflict_epi32 (__mmask8 __U, __m128i __A)
  110. {
  111. return (__m128i) __builtin_ia32_vpconflictsi_128_mask ((__v4si) __A,
  112. (__v4si) _mm_setzero_si128 (),
  113. (__mmask8) __U);
  114. }
  115. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  116. _mm256_conflict_epi32 (__m256i __A)
  117. {
  118. return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
  119. (__v8si) _mm256_undefined_si256 (),
  120. (__mmask8) -1);
  121. }
  122. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  123. _mm256_mask_conflict_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
  124. {
  125. return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
  126. (__v8si) __W,
  127. (__mmask8) __U);
  128. }
  129. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  130. _mm256_maskz_conflict_epi32 (__mmask8 __U, __m256i __A)
  131. {
  132. return (__m256i) __builtin_ia32_vpconflictsi_256_mask ((__v8si) __A,
  133. (__v8si)
  134. _mm256_setzero_si256 (),
  135. (__mmask8) __U);
  136. }
  137. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  138. _mm_lzcnt_epi32 (__m128i __A)
  139. {
  140. return (__m128i) __builtin_ia32_vplzcntd_128 ((__v4si) __A);
  141. }
  142. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  143. _mm_mask_lzcnt_epi32 (__m128i __W, __mmask8 __U, __m128i __A)
  144. {
  145. return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U,
  146. (__v4si)_mm_lzcnt_epi32(__A),
  147. (__v4si)__W);
  148. }
  149. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  150. _mm_maskz_lzcnt_epi32 (__mmask8 __U, __m128i __A)
  151. {
  152. return (__m128i)__builtin_ia32_selectd_128((__mmask8)__U,
  153. (__v4si)_mm_lzcnt_epi32(__A),
  154. (__v4si)_mm_setzero_si128());
  155. }
  156. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  157. _mm256_lzcnt_epi32 (__m256i __A)
  158. {
  159. return (__m256i) __builtin_ia32_vplzcntd_256 ((__v8si) __A);
  160. }
  161. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  162. _mm256_mask_lzcnt_epi32 (__m256i __W, __mmask8 __U, __m256i __A)
  163. {
  164. return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U,
  165. (__v8si)_mm256_lzcnt_epi32(__A),
  166. (__v8si)__W);
  167. }
  168. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  169. _mm256_maskz_lzcnt_epi32 (__mmask8 __U, __m256i __A)
  170. {
  171. return (__m256i)__builtin_ia32_selectd_256((__mmask8)__U,
  172. (__v8si)_mm256_lzcnt_epi32(__A),
  173. (__v8si)_mm256_setzero_si256());
  174. }
  175. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  176. _mm_lzcnt_epi64 (__m128i __A)
  177. {
  178. return (__m128i) __builtin_ia32_vplzcntq_128 ((__v2di) __A);
  179. }
  180. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  181. _mm_mask_lzcnt_epi64 (__m128i __W, __mmask8 __U, __m128i __A)
  182. {
  183. return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U,
  184. (__v2di)_mm_lzcnt_epi64(__A),
  185. (__v2di)__W);
  186. }
  187. static __inline__ __m128i __DEFAULT_FN_ATTRS128
  188. _mm_maskz_lzcnt_epi64 (__mmask8 __U, __m128i __A)
  189. {
  190. return (__m128i)__builtin_ia32_selectq_128((__mmask8)__U,
  191. (__v2di)_mm_lzcnt_epi64(__A),
  192. (__v2di)_mm_setzero_si128());
  193. }
  194. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  195. _mm256_lzcnt_epi64 (__m256i __A)
  196. {
  197. return (__m256i) __builtin_ia32_vplzcntq_256 ((__v4di) __A);
  198. }
  199. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  200. _mm256_mask_lzcnt_epi64 (__m256i __W, __mmask8 __U, __m256i __A)
  201. {
  202. return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U,
  203. (__v4di)_mm256_lzcnt_epi64(__A),
  204. (__v4di)__W);
  205. }
  206. static __inline__ __m256i __DEFAULT_FN_ATTRS256
  207. _mm256_maskz_lzcnt_epi64 (__mmask8 __U, __m256i __A)
  208. {
  209. return (__m256i)__builtin_ia32_selectq_256((__mmask8)__U,
  210. (__v4di)_mm256_lzcnt_epi64(__A),
  211. (__v4di)_mm256_setzero_si256());
  212. }
  213. #undef __DEFAULT_FN_ATTRS128
  214. #undef __DEFAULT_FN_ATTRS256
  215. #endif /* __AVX512VLCDINTRIN_H */