avx512pfintrin.h 9.8 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264
  1. /* Copyright (C) 2013-2019 Free Software Foundation, Inc.
  2. This file is part of GCC.
  3. GCC is free software; you can redistribute it and/or modify
  4. it under the terms of the GNU General Public License as published by
  5. the Free Software Foundation; either version 3, or (at your option)
  6. any later version.
  7. GCC is distributed in the hope that it will be useful,
  8. but WITHOUT ANY WARRANTY; without even the implied warranty of
  9. MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
  10. GNU General Public License for more details.
  11. Under Section 7 of GPL version 3, you are granted additional
  12. permissions described in the GCC Runtime Library Exception, version
  13. 3.1, as published by the Free Software Foundation.
  14. You should have received a copy of the GNU General Public License and
  15. a copy of the GCC Runtime Library Exception along with this program;
  16. see the files COPYING3 and COPYING.RUNTIME respectively. If not, see
  17. <http://www.gnu.org/licenses/>. */
  18. #ifndef _IMMINTRIN_H_INCLUDED
  19. #error "Never use <avx512pfintrin.h> directly; include <immintrin.h> instead."
  20. #endif
  21. #ifndef _AVX512PFINTRIN_H_INCLUDED
  22. #define _AVX512PFINTRIN_H_INCLUDED
  23. #ifndef __AVX512PF__
  24. #pragma GCC push_options
  25. #pragma GCC target("avx512pf")
  26. #define __DISABLE_AVX512PF__
  27. #endif /* __AVX512PF__ */
  28. /* Internal data types for implementing the intrinsics. */
  29. typedef long long __v8di __attribute__ ((__vector_size__ (64)));
  30. typedef int __v16si __attribute__ ((__vector_size__ (64)));
  31. /* The Intel API is flexible enough that we must allow aliasing with other
  32. vector types, and their scalar components. */
  33. typedef long long __m512i __attribute__ ((__vector_size__ (64), __may_alias__));
  34. typedef unsigned char __mmask8;
  35. typedef unsigned short __mmask16;
  36. #ifdef __OPTIMIZE__
  37. extern __inline void
  38. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  39. _mm512_prefetch_i32gather_pd (__m256i __index, void const *__addr,
  40. int __scale, int __hint)
  41. {
  42. __builtin_ia32_gatherpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
  43. __scale, __hint);
  44. }
  45. extern __inline void
  46. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  47. _mm512_prefetch_i32gather_ps (__m512i __index, void const *__addr,
  48. int __scale, int __hint)
  49. {
  50. __builtin_ia32_gatherpfdps ((__mmask16) 0xFFFF, (__v16si) __index, __addr,
  51. __scale, __hint);
  52. }
  53. extern __inline void
  54. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  55. _mm512_mask_prefetch_i32gather_pd (__m256i __index, __mmask8 __mask,
  56. void const *__addr, int __scale, int __hint)
  57. {
  58. __builtin_ia32_gatherpfdpd (__mask, (__v8si) __index, __addr, __scale,
  59. __hint);
  60. }
  61. extern __inline void
  62. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  63. _mm512_mask_prefetch_i32gather_ps (__m512i __index, __mmask16 __mask,
  64. void const *__addr, int __scale, int __hint)
  65. {
  66. __builtin_ia32_gatherpfdps (__mask, (__v16si) __index, __addr, __scale,
  67. __hint);
  68. }
  69. extern __inline void
  70. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  71. _mm512_prefetch_i64gather_pd (__m512i __index, void const *__addr,
  72. int __scale, int __hint)
  73. {
  74. __builtin_ia32_gatherpfqpd ((__mmask8) 0xFF, (__v8di) __index, __addr,
  75. __scale, __hint);
  76. }
  77. extern __inline void
  78. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  79. _mm512_prefetch_i64gather_ps (__m512i __index, void const *__addr,
  80. int __scale, int __hint)
  81. {
  82. __builtin_ia32_gatherpfqps ((__mmask8) 0xFF, (__v8di) __index, __addr,
  83. __scale, __hint);
  84. }
  85. extern __inline void
  86. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  87. _mm512_mask_prefetch_i64gather_pd (__m512i __index, __mmask8 __mask,
  88. void const *__addr, int __scale, int __hint)
  89. {
  90. __builtin_ia32_gatherpfqpd (__mask, (__v8di) __index, __addr, __scale,
  91. __hint);
  92. }
  93. extern __inline void
  94. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  95. _mm512_mask_prefetch_i64gather_ps (__m512i __index, __mmask8 __mask,
  96. void const *__addr, int __scale, int __hint)
  97. {
  98. __builtin_ia32_gatherpfqps (__mask, (__v8di) __index, __addr, __scale,
  99. __hint);
  100. }
  101. extern __inline void
  102. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  103. _mm512_prefetch_i32scatter_pd (void *__addr, __m256i __index, int __scale,
  104. int __hint)
  105. {
  106. __builtin_ia32_scatterpfdpd ((__mmask8) 0xFF, (__v8si) __index, __addr,
  107. __scale, __hint);
  108. }
  109. extern __inline void
  110. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  111. _mm512_prefetch_i32scatter_ps (void *__addr, __m512i __index, int __scale,
  112. int __hint)
  113. {
  114. __builtin_ia32_scatterpfdps ((__mmask16) 0xFFFF, (__v16si) __index, __addr,
  115. __scale, __hint);
  116. }
  117. extern __inline void
  118. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  119. _mm512_mask_prefetch_i32scatter_pd (void *__addr, __mmask8 __mask,
  120. __m256i __index, int __scale, int __hint)
  121. {
  122. __builtin_ia32_scatterpfdpd (__mask, (__v8si) __index, __addr, __scale,
  123. __hint);
  124. }
  125. extern __inline void
  126. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  127. _mm512_mask_prefetch_i32scatter_ps (void *__addr, __mmask16 __mask,
  128. __m512i __index, int __scale, int __hint)
  129. {
  130. __builtin_ia32_scatterpfdps (__mask, (__v16si) __index, __addr, __scale,
  131. __hint);
  132. }
  133. extern __inline void
  134. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  135. _mm512_prefetch_i64scatter_pd (void *__addr, __m512i __index, int __scale,
  136. int __hint)
  137. {
  138. __builtin_ia32_scatterpfqpd ((__mmask8) 0xFF, (__v8di) __index,__addr,
  139. __scale, __hint);
  140. }
  141. extern __inline void
  142. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  143. _mm512_prefetch_i64scatter_ps (void *__addr, __m512i __index, int __scale,
  144. int __hint)
  145. {
  146. __builtin_ia32_scatterpfqps ((__mmask8) 0xFF, (__v8di) __index, __addr,
  147. __scale, __hint);
  148. }
  149. extern __inline void
  150. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  151. _mm512_mask_prefetch_i64scatter_pd (void *__addr, __mmask8 __mask,
  152. __m512i __index, int __scale, int __hint)
  153. {
  154. __builtin_ia32_scatterpfqpd (__mask, (__v8di) __index, __addr, __scale,
  155. __hint);
  156. }
  157. extern __inline void
  158. __attribute__ ((__gnu_inline__, __always_inline__, __artificial__))
  159. _mm512_mask_prefetch_i64scatter_ps (void *__addr, __mmask8 __mask,
  160. __m512i __index, int __scale, int __hint)
  161. {
  162. __builtin_ia32_scatterpfqps (__mask, (__v8di) __index, __addr, __scale,
  163. __hint);
  164. }
  165. #else
  166. #define _mm512_prefetch_i32gather_pd(INDEX, ADDR, SCALE, HINT) \
  167. __builtin_ia32_gatherpfdpd ((__mmask8)0xFF, (__v8si)(__m256i)INDEX, \
  168. (void const *)ADDR, (int)SCALE, (int)HINT)
  169. #define _mm512_prefetch_i32gather_ps(INDEX, ADDR, SCALE, HINT) \
  170. __builtin_ia32_gatherpfdps ((__mmask16)0xFFFF, (__v16si)(__m512i)INDEX, \
  171. (void const *)ADDR, (int)SCALE, (int)HINT)
  172. #define _mm512_mask_prefetch_i32gather_pd(INDEX, MASK, ADDR, SCALE, HINT) \
  173. __builtin_ia32_gatherpfdpd ((__mmask8)MASK, (__v8si)(__m256i)INDEX, \
  174. (void const *)ADDR, (int)SCALE, (int)HINT)
  175. #define _mm512_mask_prefetch_i32gather_ps(INDEX, MASK, ADDR, SCALE, HINT) \
  176. __builtin_ia32_gatherpfdps ((__mmask16)MASK, (__v16si)(__m512i)INDEX, \
  177. (void const *)ADDR, (int)SCALE, (int)HINT)
  178. #define _mm512_prefetch_i64gather_pd(INDEX, ADDR, SCALE, HINT) \
  179. __builtin_ia32_gatherpfqpd ((__mmask8)0xFF, (__v8di)(__m512i)INDEX, \
  180. (void *)ADDR, (int)SCALE, (int)HINT)
  181. #define _mm512_prefetch_i64gather_ps(INDEX, ADDR, SCALE, HINT) \
  182. __builtin_ia32_gatherpfqps ((__mmask8)0xFF, (__v8di)(__m512i)INDEX, \
  183. (void *)ADDR, (int)SCALE, (int)HINT)
  184. #define _mm512_mask_prefetch_i64gather_pd(INDEX, MASK, ADDR, SCALE, HINT) \
  185. __builtin_ia32_gatherpfqpd ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
  186. (void *)ADDR, (int)SCALE, (int)HINT)
  187. #define _mm512_mask_prefetch_i64gather_ps(INDEX, MASK, ADDR, SCALE, HINT) \
  188. __builtin_ia32_gatherpfqps ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
  189. (void *)ADDR, (int)SCALE, (int)HINT)
  190. #define _mm512_prefetch_i32scatter_pd(ADDR, INDEX, SCALE, HINT) \
  191. __builtin_ia32_scatterpfdpd ((__mmask8)0xFF, (__v8si)(__m256i)INDEX, \
  192. (void *)ADDR, (int)SCALE, (int)HINT)
  193. #define _mm512_prefetch_i32scatter_ps(ADDR, INDEX, SCALE, HINT) \
  194. __builtin_ia32_scatterpfdps ((__mmask16)0xFFFF, (__v16si)(__m512i)INDEX, \
  195. (void *)ADDR, (int)SCALE, (int)HINT)
  196. #define _mm512_mask_prefetch_i32scatter_pd(ADDR, MASK, INDEX, SCALE, HINT) \
  197. __builtin_ia32_scatterpfdpd ((__mmask8)MASK, (__v8si)(__m256i)INDEX, \
  198. (void *)ADDR, (int)SCALE, (int)HINT)
  199. #define _mm512_mask_prefetch_i32scatter_ps(ADDR, MASK, INDEX, SCALE, HINT) \
  200. __builtin_ia32_scatterpfdps ((__mmask16)MASK, (__v16si)(__m512i)INDEX, \
  201. (void *)ADDR, (int)SCALE, (int)HINT)
  202. #define _mm512_prefetch_i64scatter_pd(ADDR, INDEX, SCALE, HINT) \
  203. __builtin_ia32_scatterpfqpd ((__mmask8)0xFF, (__v8di)(__m512i)INDEX, \
  204. (void *)ADDR, (int)SCALE, (int)HINT)
  205. #define _mm512_prefetch_i64scatter_ps(ADDR, INDEX, SCALE, HINT) \
  206. __builtin_ia32_scatterpfqps ((__mmask8)0xFF, (__v8di)(__m512i)INDEX, \
  207. (void *)ADDR, (int)SCALE, (int)HINT)
  208. #define _mm512_mask_prefetch_i64scatter_pd(ADDR, MASK, INDEX, SCALE, HINT) \
  209. __builtin_ia32_scatterpfqpd ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
  210. (void *)ADDR, (int)SCALE, (int)HINT)
  211. #define _mm512_mask_prefetch_i64scatter_ps(ADDR, MASK, INDEX, SCALE, HINT) \
  212. __builtin_ia32_scatterpfqps ((__mmask8)MASK, (__v8di)(__m512i)INDEX, \
  213. (void *)ADDR, (int)SCALE, (int)HINT)
  214. #endif
  215. #ifdef __DISABLE_AVX512PF__
  216. #undef __DISABLE_AVX512PF__
  217. #pragma GCC pop_options
  218. #endif /* __DISABLE_AVX512PF__ */
  219. #endif /* _AVX512PFINTRIN_H_INCLUDED */