fmaintrin.h 6.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234
  1. /*===---- fma4intrin.h - FMA4 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 <fmaintrin.h> directly; include <immintrin.h> instead."
  25. #endif
  26. #ifndef __FMAINTRIN_H
  27. #define __FMAINTRIN_H
  28. #ifndef __FMA__
  29. # error "FMA instruction set is not enabled"
  30. #else
  31. /* Define the default attributes for the functions in this file. */
  32. #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__))
  33. static __inline__ __m128 __DEFAULT_FN_ATTRS
  34. _mm_fmadd_ps(__m128 __A, __m128 __B, __m128 __C)
  35. {
  36. return (__m128)__builtin_ia32_vfmaddps(__A, __B, __C);
  37. }
  38. static __inline__ __m128d __DEFAULT_FN_ATTRS
  39. _mm_fmadd_pd(__m128d __A, __m128d __B, __m128d __C)
  40. {
  41. return (__m128d)__builtin_ia32_vfmaddpd(__A, __B, __C);
  42. }
  43. static __inline__ __m128 __DEFAULT_FN_ATTRS
  44. _mm_fmadd_ss(__m128 __A, __m128 __B, __m128 __C)
  45. {
  46. return (__m128)__builtin_ia32_vfmaddss(__A, __B, __C);
  47. }
  48. static __inline__ __m128d __DEFAULT_FN_ATTRS
  49. _mm_fmadd_sd(__m128d __A, __m128d __B, __m128d __C)
  50. {
  51. return (__m128d)__builtin_ia32_vfmaddsd(__A, __B, __C);
  52. }
  53. static __inline__ __m128 __DEFAULT_FN_ATTRS
  54. _mm_fmsub_ps(__m128 __A, __m128 __B, __m128 __C)
  55. {
  56. return (__m128)__builtin_ia32_vfmsubps(__A, __B, __C);
  57. }
  58. static __inline__ __m128d __DEFAULT_FN_ATTRS
  59. _mm_fmsub_pd(__m128d __A, __m128d __B, __m128d __C)
  60. {
  61. return (__m128d)__builtin_ia32_vfmsubpd(__A, __B, __C);
  62. }
  63. static __inline__ __m128 __DEFAULT_FN_ATTRS
  64. _mm_fmsub_ss(__m128 __A, __m128 __B, __m128 __C)
  65. {
  66. return (__m128)__builtin_ia32_vfmsubss(__A, __B, __C);
  67. }
  68. static __inline__ __m128d __DEFAULT_FN_ATTRS
  69. _mm_fmsub_sd(__m128d __A, __m128d __B, __m128d __C)
  70. {
  71. return (__m128d)__builtin_ia32_vfmsubsd(__A, __B, __C);
  72. }
  73. static __inline__ __m128 __DEFAULT_FN_ATTRS
  74. _mm_fnmadd_ps(__m128 __A, __m128 __B, __m128 __C)
  75. {
  76. return (__m128)__builtin_ia32_vfnmaddps(__A, __B, __C);
  77. }
  78. static __inline__ __m128d __DEFAULT_FN_ATTRS
  79. _mm_fnmadd_pd(__m128d __A, __m128d __B, __m128d __C)
  80. {
  81. return (__m128d)__builtin_ia32_vfnmaddpd(__A, __B, __C);
  82. }
  83. static __inline__ __m128 __DEFAULT_FN_ATTRS
  84. _mm_fnmadd_ss(__m128 __A, __m128 __B, __m128 __C)
  85. {
  86. return (__m128)__builtin_ia32_vfnmaddss(__A, __B, __C);
  87. }
  88. static __inline__ __m128d __DEFAULT_FN_ATTRS
  89. _mm_fnmadd_sd(__m128d __A, __m128d __B, __m128d __C)
  90. {
  91. return (__m128d)__builtin_ia32_vfnmaddsd(__A, __B, __C);
  92. }
  93. static __inline__ __m128 __DEFAULT_FN_ATTRS
  94. _mm_fnmsub_ps(__m128 __A, __m128 __B, __m128 __C)
  95. {
  96. return (__m128)__builtin_ia32_vfnmsubps(__A, __B, __C);
  97. }
  98. static __inline__ __m128d __DEFAULT_FN_ATTRS
  99. _mm_fnmsub_pd(__m128d __A, __m128d __B, __m128d __C)
  100. {
  101. return (__m128d)__builtin_ia32_vfnmsubpd(__A, __B, __C);
  102. }
  103. static __inline__ __m128 __DEFAULT_FN_ATTRS
  104. _mm_fnmsub_ss(__m128 __A, __m128 __B, __m128 __C)
  105. {
  106. return (__m128)__builtin_ia32_vfnmsubss(__A, __B, __C);
  107. }
  108. static __inline__ __m128d __DEFAULT_FN_ATTRS
  109. _mm_fnmsub_sd(__m128d __A, __m128d __B, __m128d __C)
  110. {
  111. return (__m128d)__builtin_ia32_vfnmsubsd(__A, __B, __C);
  112. }
  113. static __inline__ __m128 __DEFAULT_FN_ATTRS
  114. _mm_fmaddsub_ps(__m128 __A, __m128 __B, __m128 __C)
  115. {
  116. return (__m128)__builtin_ia32_vfmaddsubps(__A, __B, __C);
  117. }
  118. static __inline__ __m128d __DEFAULT_FN_ATTRS
  119. _mm_fmaddsub_pd(__m128d __A, __m128d __B, __m128d __C)
  120. {
  121. return (__m128d)__builtin_ia32_vfmaddsubpd(__A, __B, __C);
  122. }
  123. static __inline__ __m128 __DEFAULT_FN_ATTRS
  124. _mm_fmsubadd_ps(__m128 __A, __m128 __B, __m128 __C)
  125. {
  126. return (__m128)__builtin_ia32_vfmsubaddps(__A, __B, __C);
  127. }
  128. static __inline__ __m128d __DEFAULT_FN_ATTRS
  129. _mm_fmsubadd_pd(__m128d __A, __m128d __B, __m128d __C)
  130. {
  131. return (__m128d)__builtin_ia32_vfmsubaddpd(__A, __B, __C);
  132. }
  133. static __inline__ __m256 __DEFAULT_FN_ATTRS
  134. _mm256_fmadd_ps(__m256 __A, __m256 __B, __m256 __C)
  135. {
  136. return (__m256)__builtin_ia32_vfmaddps256(__A, __B, __C);
  137. }
  138. static __inline__ __m256d __DEFAULT_FN_ATTRS
  139. _mm256_fmadd_pd(__m256d __A, __m256d __B, __m256d __C)
  140. {
  141. return (__m256d)__builtin_ia32_vfmaddpd256(__A, __B, __C);
  142. }
  143. static __inline__ __m256 __DEFAULT_FN_ATTRS
  144. _mm256_fmsub_ps(__m256 __A, __m256 __B, __m256 __C)
  145. {
  146. return (__m256)__builtin_ia32_vfmsubps256(__A, __B, __C);
  147. }
  148. static __inline__ __m256d __DEFAULT_FN_ATTRS
  149. _mm256_fmsub_pd(__m256d __A, __m256d __B, __m256d __C)
  150. {
  151. return (__m256d)__builtin_ia32_vfmsubpd256(__A, __B, __C);
  152. }
  153. static __inline__ __m256 __DEFAULT_FN_ATTRS
  154. _mm256_fnmadd_ps(__m256 __A, __m256 __B, __m256 __C)
  155. {
  156. return (__m256)__builtin_ia32_vfnmaddps256(__A, __B, __C);
  157. }
  158. static __inline__ __m256d __DEFAULT_FN_ATTRS
  159. _mm256_fnmadd_pd(__m256d __A, __m256d __B, __m256d __C)
  160. {
  161. return (__m256d)__builtin_ia32_vfnmaddpd256(__A, __B, __C);
  162. }
  163. static __inline__ __m256 __DEFAULT_FN_ATTRS
  164. _mm256_fnmsub_ps(__m256 __A, __m256 __B, __m256 __C)
  165. {
  166. return (__m256)__builtin_ia32_vfnmsubps256(__A, __B, __C);
  167. }
  168. static __inline__ __m256d __DEFAULT_FN_ATTRS
  169. _mm256_fnmsub_pd(__m256d __A, __m256d __B, __m256d __C)
  170. {
  171. return (__m256d)__builtin_ia32_vfnmsubpd256(__A, __B, __C);
  172. }
  173. static __inline__ __m256 __DEFAULT_FN_ATTRS
  174. _mm256_fmaddsub_ps(__m256 __A, __m256 __B, __m256 __C)
  175. {
  176. return (__m256)__builtin_ia32_vfmaddsubps256(__A, __B, __C);
  177. }
  178. static __inline__ __m256d __DEFAULT_FN_ATTRS
  179. _mm256_fmaddsub_pd(__m256d __A, __m256d __B, __m256d __C)
  180. {
  181. return (__m256d)__builtin_ia32_vfmaddsubpd256(__A, __B, __C);
  182. }
  183. static __inline__ __m256 __DEFAULT_FN_ATTRS
  184. _mm256_fmsubadd_ps(__m256 __A, __m256 __B, __m256 __C)
  185. {
  186. return (__m256)__builtin_ia32_vfmsubaddps256(__A, __B, __C);
  187. }
  188. static __inline__ __m256d __DEFAULT_FN_ATTRS
  189. _mm256_fmsubadd_pd(__m256d __A, __m256d __B, __m256d __C)
  190. {
  191. return (__m256d)__builtin_ia32_vfmsubaddpd256(__A, __B, __C);
  192. }
  193. #undef __DEFAULT_FN_ATTRS
  194. #endif /* __FMA__ */
  195. #endif /* __FMAINTRIN_H */