avx512vbmi2intrin.h 14 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397
  1. /*===------------- avx512vbmi2intrin.h - VBMI2 intrinsics ------------------===
  2. *
  3. *
  4. * Permission is hereby granted, free of charge, to any person obtaining a copy
  5. * of this software and associated documentation files (the "Software"), to deal
  6. * in the Software without restriction, including without limitation the rights
  7. * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
  8. * copies of the Software, and to permit persons to whom the Software is
  9. * furnished to do so, subject to the following conditions:
  10. *
  11. * The above copyright notice and this permission notice shall be included in
  12. * all copies or substantial portions of the Software.
  13. *
  14. * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  15. * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  16. * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
  17. * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  18. * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
  19. * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
  20. * THE SOFTWARE.
  21. *
  22. *===-----------------------------------------------------------------------===
  23. */
  24. #ifndef __IMMINTRIN_H
  25. #error "Never use <avx512vbmi2intrin.h> directly; include <immintrin.h> instead."
  26. #endif
  27. #ifndef __AVX512VBMI2INTRIN_H
  28. #define __AVX512VBMI2INTRIN_H
  29. /* Define the default attributes for the functions in this file. */
  30. #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__, __target__("avx512vbmi2"), __min_vector_width__(512)))
  31. static __inline__ __m512i __DEFAULT_FN_ATTRS
  32. _mm512_mask_compress_epi16(__m512i __S, __mmask32 __U, __m512i __D)
  33. {
  34. return (__m512i) __builtin_ia32_compresshi512_mask ((__v32hi) __D,
  35. (__v32hi) __S,
  36. __U);
  37. }
  38. static __inline__ __m512i __DEFAULT_FN_ATTRS
  39. _mm512_maskz_compress_epi16(__mmask32 __U, __m512i __D)
  40. {
  41. return (__m512i) __builtin_ia32_compresshi512_mask ((__v32hi) __D,
  42. (__v32hi) _mm512_setzero_si512(),
  43. __U);
  44. }
  45. static __inline__ __m512i __DEFAULT_FN_ATTRS
  46. _mm512_mask_compress_epi8(__m512i __S, __mmask64 __U, __m512i __D)
  47. {
  48. return (__m512i) __builtin_ia32_compressqi512_mask ((__v64qi) __D,
  49. (__v64qi) __S,
  50. __U);
  51. }
  52. static __inline__ __m512i __DEFAULT_FN_ATTRS
  53. _mm512_maskz_compress_epi8(__mmask64 __U, __m512i __D)
  54. {
  55. return (__m512i) __builtin_ia32_compressqi512_mask ((__v64qi) __D,
  56. (__v64qi) _mm512_setzero_si512(),
  57. __U);
  58. }
  59. static __inline__ void __DEFAULT_FN_ATTRS
  60. _mm512_mask_compressstoreu_epi16(void *__P, __mmask32 __U, __m512i __D)
  61. {
  62. __builtin_ia32_compressstorehi512_mask ((__v32hi *) __P, (__v32hi) __D,
  63. __U);
  64. }
  65. static __inline__ void __DEFAULT_FN_ATTRS
  66. _mm512_mask_compressstoreu_epi8(void *__P, __mmask64 __U, __m512i __D)
  67. {
  68. __builtin_ia32_compressstoreqi512_mask ((__v64qi *) __P, (__v64qi) __D,
  69. __U);
  70. }
  71. static __inline__ __m512i __DEFAULT_FN_ATTRS
  72. _mm512_mask_expand_epi16(__m512i __S, __mmask32 __U, __m512i __D)
  73. {
  74. return (__m512i) __builtin_ia32_expandhi512_mask ((__v32hi) __D,
  75. (__v32hi) __S,
  76. __U);
  77. }
  78. static __inline__ __m512i __DEFAULT_FN_ATTRS
  79. _mm512_maskz_expand_epi16(__mmask32 __U, __m512i __D)
  80. {
  81. return (__m512i) __builtin_ia32_expandhi512_mask ((__v32hi) __D,
  82. (__v32hi) _mm512_setzero_si512(),
  83. __U);
  84. }
  85. static __inline__ __m512i __DEFAULT_FN_ATTRS
  86. _mm512_mask_expand_epi8(__m512i __S, __mmask64 __U, __m512i __D)
  87. {
  88. return (__m512i) __builtin_ia32_expandqi512_mask ((__v64qi) __D,
  89. (__v64qi) __S,
  90. __U);
  91. }
  92. static __inline__ __m512i __DEFAULT_FN_ATTRS
  93. _mm512_maskz_expand_epi8(__mmask64 __U, __m512i __D)
  94. {
  95. return (__m512i) __builtin_ia32_expandqi512_mask ((__v64qi) __D,
  96. (__v64qi) _mm512_setzero_si512(),
  97. __U);
  98. }
  99. static __inline__ __m512i __DEFAULT_FN_ATTRS
  100. _mm512_mask_expandloadu_epi16(__m512i __S, __mmask32 __U, void const *__P)
  101. {
  102. return (__m512i) __builtin_ia32_expandloadhi512_mask ((const __v32hi *)__P,
  103. (__v32hi) __S,
  104. __U);
  105. }
  106. static __inline__ __m512i __DEFAULT_FN_ATTRS
  107. _mm512_maskz_expandloadu_epi16(__mmask32 __U, void const *__P)
  108. {
  109. return (__m512i) __builtin_ia32_expandloadhi512_mask ((const __v32hi *)__P,
  110. (__v32hi) _mm512_setzero_si512(),
  111. __U);
  112. }
  113. static __inline__ __m512i __DEFAULT_FN_ATTRS
  114. _mm512_mask_expandloadu_epi8(__m512i __S, __mmask64 __U, void const *__P)
  115. {
  116. return (__m512i) __builtin_ia32_expandloadqi512_mask ((const __v64qi *)__P,
  117. (__v64qi) __S,
  118. __U);
  119. }
  120. static __inline__ __m512i __DEFAULT_FN_ATTRS
  121. _mm512_maskz_expandloadu_epi8(__mmask64 __U, void const *__P)
  122. {
  123. return (__m512i) __builtin_ia32_expandloadqi512_mask ((const __v64qi *)__P,
  124. (__v64qi) _mm512_setzero_si512(),
  125. __U);
  126. }
  127. #define _mm512_shldi_epi64(A, B, I) \
  128. (__m512i)__builtin_ia32_vpshldq512((__v8di)(__m512i)(A), \
  129. (__v8di)(__m512i)(B), (int)(I))
  130. #define _mm512_mask_shldi_epi64(S, U, A, B, I) \
  131. (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
  132. (__v8di)_mm512_shldi_epi64((A), (B), (I)), \
  133. (__v8di)(__m512i)(S))
  134. #define _mm512_maskz_shldi_epi64(U, A, B, I) \
  135. (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
  136. (__v8di)_mm512_shldi_epi64((A), (B), (I)), \
  137. (__v8di)_mm512_setzero_si512())
  138. #define _mm512_shldi_epi32(A, B, I) \
  139. (__m512i)__builtin_ia32_vpshldd512((__v16si)(__m512i)(A), \
  140. (__v16si)(__m512i)(B), (int)(I))
  141. #define _mm512_mask_shldi_epi32(S, U, A, B, I) \
  142. (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
  143. (__v16si)_mm512_shldi_epi32((A), (B), (I)), \
  144. (__v16si)(__m512i)(S))
  145. #define _mm512_maskz_shldi_epi32(U, A, B, I) \
  146. (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
  147. (__v16si)_mm512_shldi_epi32((A), (B), (I)), \
  148. (__v16si)_mm512_setzero_si512())
  149. #define _mm512_shldi_epi16(A, B, I) \
  150. (__m512i)__builtin_ia32_vpshldw512((__v32hi)(__m512i)(A), \
  151. (__v32hi)(__m512i)(B), (int)(I))
  152. #define _mm512_mask_shldi_epi16(S, U, A, B, I) \
  153. (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
  154. (__v32hi)_mm512_shldi_epi16((A), (B), (I)), \
  155. (__v32hi)(__m512i)(S))
  156. #define _mm512_maskz_shldi_epi16(U, A, B, I) \
  157. (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
  158. (__v32hi)_mm512_shldi_epi16((A), (B), (I)), \
  159. (__v32hi)_mm512_setzero_si512())
  160. #define _mm512_shrdi_epi64(A, B, I) \
  161. (__m512i)__builtin_ia32_vpshrdq512((__v8di)(__m512i)(A), \
  162. (__v8di)(__m512i)(B), (int)(I))
  163. #define _mm512_mask_shrdi_epi64(S, U, A, B, I) \
  164. (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
  165. (__v8di)_mm512_shrdi_epi64((A), (B), (I)), \
  166. (__v8di)(__m512i)(S))
  167. #define _mm512_maskz_shrdi_epi64(U, A, B, I) \
  168. (__m512i)__builtin_ia32_selectq_512((__mmask8)(U), \
  169. (__v8di)_mm512_shrdi_epi64((A), (B), (I)), \
  170. (__v8di)_mm512_setzero_si512())
  171. #define _mm512_shrdi_epi32(A, B, I) \
  172. (__m512i)__builtin_ia32_vpshrdd512((__v16si)(__m512i)(A), \
  173. (__v16si)(__m512i)(B), (int)(I))
  174. #define _mm512_mask_shrdi_epi32(S, U, A, B, I) \
  175. (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
  176. (__v16si)_mm512_shrdi_epi32((A), (B), (I)), \
  177. (__v16si)(__m512i)(S))
  178. #define _mm512_maskz_shrdi_epi32(U, A, B, I) \
  179. (__m512i)__builtin_ia32_selectd_512((__mmask16)(U), \
  180. (__v16si)_mm512_shrdi_epi32((A), (B), (I)), \
  181. (__v16si)_mm512_setzero_si512())
  182. #define _mm512_shrdi_epi16(A, B, I) \
  183. (__m512i)__builtin_ia32_vpshrdw512((__v32hi)(__m512i)(A), \
  184. (__v32hi)(__m512i)(B), (int)(I))
  185. #define _mm512_mask_shrdi_epi16(S, U, A, B, I) \
  186. (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
  187. (__v32hi)_mm512_shrdi_epi16((A), (B), (I)), \
  188. (__v32hi)(__m512i)(S))
  189. #define _mm512_maskz_shrdi_epi16(U, A, B, I) \
  190. (__m512i)__builtin_ia32_selectw_512((__mmask32)(U), \
  191. (__v32hi)_mm512_shrdi_epi16((A), (B), (I)), \
  192. (__v32hi)_mm512_setzero_si512())
  193. static __inline__ __m512i __DEFAULT_FN_ATTRS
  194. _mm512_mask_shldv_epi64(__m512i __S, __mmask8 __U, __m512i __A, __m512i __B)
  195. {
  196. return (__m512i) __builtin_ia32_vpshldvq512_mask ((__v8di) __S,
  197. (__v8di) __A,
  198. (__v8di) __B,
  199. __U);
  200. }
  201. static __inline__ __m512i __DEFAULT_FN_ATTRS
  202. _mm512_maskz_shldv_epi64(__mmask8 __U, __m512i __S, __m512i __A, __m512i __B)
  203. {
  204. return (__m512i) __builtin_ia32_vpshldvq512_maskz ((__v8di) __S,
  205. (__v8di) __A,
  206. (__v8di) __B,
  207. __U);
  208. }
  209. static __inline__ __m512i __DEFAULT_FN_ATTRS
  210. _mm512_shldv_epi64(__m512i __S, __m512i __A, __m512i __B)
  211. {
  212. return (__m512i) __builtin_ia32_vpshldvq512_mask ((__v8di) __S,
  213. (__v8di) __A,
  214. (__v8di) __B,
  215. (__mmask8) -1);
  216. }
  217. static __inline__ __m512i __DEFAULT_FN_ATTRS
  218. _mm512_mask_shldv_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B)
  219. {
  220. return (__m512i) __builtin_ia32_vpshldvd512_mask ((__v16si) __S,
  221. (__v16si) __A,
  222. (__v16si) __B,
  223. __U);
  224. }
  225. static __inline__ __m512i __DEFAULT_FN_ATTRS
  226. _mm512_maskz_shldv_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B)
  227. {
  228. return (__m512i) __builtin_ia32_vpshldvd512_maskz ((__v16si) __S,
  229. (__v16si) __A,
  230. (__v16si) __B,
  231. __U);
  232. }
  233. static __inline__ __m512i __DEFAULT_FN_ATTRS
  234. _mm512_shldv_epi32(__m512i __S, __m512i __A, __m512i __B)
  235. {
  236. return (__m512i) __builtin_ia32_vpshldvd512_mask ((__v16si) __S,
  237. (__v16si) __A,
  238. (__v16si) __B,
  239. (__mmask16) -1);
  240. }
  241. static __inline__ __m512i __DEFAULT_FN_ATTRS
  242. _mm512_mask_shldv_epi16(__m512i __S, __mmask32 __U, __m512i __A, __m512i __B)
  243. {
  244. return (__m512i) __builtin_ia32_vpshldvw512_mask ((__v32hi) __S,
  245. (__v32hi) __A,
  246. (__v32hi) __B,
  247. __U);
  248. }
  249. static __inline__ __m512i __DEFAULT_FN_ATTRS
  250. _mm512_maskz_shldv_epi16(__mmask32 __U, __m512i __S, __m512i __A, __m512i __B)
  251. {
  252. return (__m512i) __builtin_ia32_vpshldvw512_maskz ((__v32hi) __S,
  253. (__v32hi) __A,
  254. (__v32hi) __B,
  255. __U);
  256. }
  257. static __inline__ __m512i __DEFAULT_FN_ATTRS
  258. _mm512_shldv_epi16(__m512i __S, __m512i __A, __m512i __B)
  259. {
  260. return (__m512i) __builtin_ia32_vpshldvw512_mask ((__v32hi) __S,
  261. (__v32hi) __A,
  262. (__v32hi) __B,
  263. (__mmask32) -1);
  264. }
  265. static __inline__ __m512i __DEFAULT_FN_ATTRS
  266. _mm512_mask_shrdv_epi64(__m512i __S, __mmask8 __U, __m512i __A, __m512i __B)
  267. {
  268. return (__m512i) __builtin_ia32_vpshrdvq512_mask ((__v8di) __S,
  269. (__v8di) __A,
  270. (__v8di) __B,
  271. __U);
  272. }
  273. static __inline__ __m512i __DEFAULT_FN_ATTRS
  274. _mm512_maskz_shrdv_epi64(__mmask8 __U, __m512i __S, __m512i __A, __m512i __B)
  275. {
  276. return (__m512i) __builtin_ia32_vpshrdvq512_maskz ((__v8di) __S,
  277. (__v8di) __A,
  278. (__v8di) __B,
  279. __U);
  280. }
  281. static __inline__ __m512i __DEFAULT_FN_ATTRS
  282. _mm512_shrdv_epi64(__m512i __S, __m512i __A, __m512i __B)
  283. {
  284. return (__m512i) __builtin_ia32_vpshrdvq512_mask ((__v8di) __S,
  285. (__v8di) __A,
  286. (__v8di) __B,
  287. (__mmask8) -1);
  288. }
  289. static __inline__ __m512i __DEFAULT_FN_ATTRS
  290. _mm512_mask_shrdv_epi32(__m512i __S, __mmask16 __U, __m512i __A, __m512i __B)
  291. {
  292. return (__m512i) __builtin_ia32_vpshrdvd512_mask ((__v16si) __S,
  293. (__v16si) __A,
  294. (__v16si) __B,
  295. __U);
  296. }
  297. static __inline__ __m512i __DEFAULT_FN_ATTRS
  298. _mm512_maskz_shrdv_epi32(__mmask16 __U, __m512i __S, __m512i __A, __m512i __B)
  299. {
  300. return (__m512i) __builtin_ia32_vpshrdvd512_maskz ((__v16si) __S,
  301. (__v16si) __A,
  302. (__v16si) __B,
  303. __U);
  304. }
  305. static __inline__ __m512i __DEFAULT_FN_ATTRS
  306. _mm512_shrdv_epi32(__m512i __S, __m512i __A, __m512i __B)
  307. {
  308. return (__m512i) __builtin_ia32_vpshrdvd512_mask ((__v16si) __S,
  309. (__v16si) __A,
  310. (__v16si) __B,
  311. (__mmask16) -1);
  312. }
  313. static __inline__ __m512i __DEFAULT_FN_ATTRS
  314. _mm512_mask_shrdv_epi16(__m512i __S, __mmask32 __U, __m512i __A, __m512i __B)
  315. {
  316. return (__m512i) __builtin_ia32_vpshrdvw512_mask ((__v32hi) __S,
  317. (__v32hi) __A,
  318. (__v32hi) __B,
  319. __U);
  320. }
  321. static __inline__ __m512i __DEFAULT_FN_ATTRS
  322. _mm512_maskz_shrdv_epi16(__mmask32 __U, __m512i __S, __m512i __A, __m512i __B)
  323. {
  324. return (__m512i) __builtin_ia32_vpshrdvw512_maskz ((__v32hi) __S,
  325. (__v32hi) __A,
  326. (__v32hi) __B,
  327. __U);
  328. }
  329. static __inline__ __m512i __DEFAULT_FN_ATTRS
  330. _mm512_shrdv_epi16(__m512i __S, __m512i __A, __m512i __B)
  331. {
  332. return (__m512i) __builtin_ia32_vpshrdvw512_mask ((__v32hi) __S,
  333. (__v32hi) __A,
  334. (__v32hi) __B,
  335. (__mmask32) -1);
  336. }
  337. #undef __DEFAULT_FN_ATTRS
  338. #endif