tmmintrin.h 6.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230
  1. /*===---- tmmintrin.h - SSSE3 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 __TMMINTRIN_H
  24. #define __TMMINTRIN_H
  25. #ifndef __SSSE3__
  26. #error "SSSE3 instruction set not enabled"
  27. #else
  28. #include <pmmintrin.h>
  29. /* Define the default attributes for the functions in this file. */
  30. #define __DEFAULT_FN_ATTRS __attribute__((__always_inline__, __nodebug__))
  31. static __inline__ __m64 __DEFAULT_FN_ATTRS
  32. _mm_abs_pi8(__m64 __a)
  33. {
  34. return (__m64)__builtin_ia32_pabsb((__v8qi)__a);
  35. }
  36. static __inline__ __m128i __DEFAULT_FN_ATTRS
  37. _mm_abs_epi8(__m128i __a)
  38. {
  39. return (__m128i)__builtin_ia32_pabsb128((__v16qi)__a);
  40. }
  41. static __inline__ __m64 __DEFAULT_FN_ATTRS
  42. _mm_abs_pi16(__m64 __a)
  43. {
  44. return (__m64)__builtin_ia32_pabsw((__v4hi)__a);
  45. }
  46. static __inline__ __m128i __DEFAULT_FN_ATTRS
  47. _mm_abs_epi16(__m128i __a)
  48. {
  49. return (__m128i)__builtin_ia32_pabsw128((__v8hi)__a);
  50. }
  51. static __inline__ __m64 __DEFAULT_FN_ATTRS
  52. _mm_abs_pi32(__m64 __a)
  53. {
  54. return (__m64)__builtin_ia32_pabsd((__v2si)__a);
  55. }
  56. static __inline__ __m128i __DEFAULT_FN_ATTRS
  57. _mm_abs_epi32(__m128i __a)
  58. {
  59. return (__m128i)__builtin_ia32_pabsd128((__v4si)__a);
  60. }
  61. #define _mm_alignr_epi8(a, b, n) __extension__ ({ \
  62. __m128i __a = (a); \
  63. __m128i __b = (b); \
  64. (__m128i)__builtin_ia32_palignr128((__v16qi)__a, (__v16qi)__b, (n)); })
  65. #define _mm_alignr_pi8(a, b, n) __extension__ ({ \
  66. __m64 __a = (a); \
  67. __m64 __b = (b); \
  68. (__m64)__builtin_ia32_palignr((__v8qi)__a, (__v8qi)__b, (n)); })
  69. static __inline__ __m128i __DEFAULT_FN_ATTRS
  70. _mm_hadd_epi16(__m128i __a, __m128i __b)
  71. {
  72. return (__m128i)__builtin_ia32_phaddw128((__v8hi)__a, (__v8hi)__b);
  73. }
  74. static __inline__ __m128i __DEFAULT_FN_ATTRS
  75. _mm_hadd_epi32(__m128i __a, __m128i __b)
  76. {
  77. return (__m128i)__builtin_ia32_phaddd128((__v4si)__a, (__v4si)__b);
  78. }
  79. static __inline__ __m64 __DEFAULT_FN_ATTRS
  80. _mm_hadd_pi16(__m64 __a, __m64 __b)
  81. {
  82. return (__m64)__builtin_ia32_phaddw((__v4hi)__a, (__v4hi)__b);
  83. }
  84. static __inline__ __m64 __DEFAULT_FN_ATTRS
  85. _mm_hadd_pi32(__m64 __a, __m64 __b)
  86. {
  87. return (__m64)__builtin_ia32_phaddd((__v2si)__a, (__v2si)__b);
  88. }
  89. static __inline__ __m128i __DEFAULT_FN_ATTRS
  90. _mm_hadds_epi16(__m128i __a, __m128i __b)
  91. {
  92. return (__m128i)__builtin_ia32_phaddsw128((__v8hi)__a, (__v8hi)__b);
  93. }
  94. static __inline__ __m64 __DEFAULT_FN_ATTRS
  95. _mm_hadds_pi16(__m64 __a, __m64 __b)
  96. {
  97. return (__m64)__builtin_ia32_phaddsw((__v4hi)__a, (__v4hi)__b);
  98. }
  99. static __inline__ __m128i __DEFAULT_FN_ATTRS
  100. _mm_hsub_epi16(__m128i __a, __m128i __b)
  101. {
  102. return (__m128i)__builtin_ia32_phsubw128((__v8hi)__a, (__v8hi)__b);
  103. }
  104. static __inline__ __m128i __DEFAULT_FN_ATTRS
  105. _mm_hsub_epi32(__m128i __a, __m128i __b)
  106. {
  107. return (__m128i)__builtin_ia32_phsubd128((__v4si)__a, (__v4si)__b);
  108. }
  109. static __inline__ __m64 __DEFAULT_FN_ATTRS
  110. _mm_hsub_pi16(__m64 __a, __m64 __b)
  111. {
  112. return (__m64)__builtin_ia32_phsubw((__v4hi)__a, (__v4hi)__b);
  113. }
  114. static __inline__ __m64 __DEFAULT_FN_ATTRS
  115. _mm_hsub_pi32(__m64 __a, __m64 __b)
  116. {
  117. return (__m64)__builtin_ia32_phsubd((__v2si)__a, (__v2si)__b);
  118. }
  119. static __inline__ __m128i __DEFAULT_FN_ATTRS
  120. _mm_hsubs_epi16(__m128i __a, __m128i __b)
  121. {
  122. return (__m128i)__builtin_ia32_phsubsw128((__v8hi)__a, (__v8hi)__b);
  123. }
  124. static __inline__ __m64 __DEFAULT_FN_ATTRS
  125. _mm_hsubs_pi16(__m64 __a, __m64 __b)
  126. {
  127. return (__m64)__builtin_ia32_phsubsw((__v4hi)__a, (__v4hi)__b);
  128. }
  129. static __inline__ __m128i __DEFAULT_FN_ATTRS
  130. _mm_maddubs_epi16(__m128i __a, __m128i __b)
  131. {
  132. return (__m128i)__builtin_ia32_pmaddubsw128((__v16qi)__a, (__v16qi)__b);
  133. }
  134. static __inline__ __m64 __DEFAULT_FN_ATTRS
  135. _mm_maddubs_pi16(__m64 __a, __m64 __b)
  136. {
  137. return (__m64)__builtin_ia32_pmaddubsw((__v8qi)__a, (__v8qi)__b);
  138. }
  139. static __inline__ __m128i __DEFAULT_FN_ATTRS
  140. _mm_mulhrs_epi16(__m128i __a, __m128i __b)
  141. {
  142. return (__m128i)__builtin_ia32_pmulhrsw128((__v8hi)__a, (__v8hi)__b);
  143. }
  144. static __inline__ __m64 __DEFAULT_FN_ATTRS
  145. _mm_mulhrs_pi16(__m64 __a, __m64 __b)
  146. {
  147. return (__m64)__builtin_ia32_pmulhrsw((__v4hi)__a, (__v4hi)__b);
  148. }
  149. static __inline__ __m128i __DEFAULT_FN_ATTRS
  150. _mm_shuffle_epi8(__m128i __a, __m128i __b)
  151. {
  152. return (__m128i)__builtin_ia32_pshufb128((__v16qi)__a, (__v16qi)__b);
  153. }
  154. static __inline__ __m64 __DEFAULT_FN_ATTRS
  155. _mm_shuffle_pi8(__m64 __a, __m64 __b)
  156. {
  157. return (__m64)__builtin_ia32_pshufb((__v8qi)__a, (__v8qi)__b);
  158. }
  159. static __inline__ __m128i __DEFAULT_FN_ATTRS
  160. _mm_sign_epi8(__m128i __a, __m128i __b)
  161. {
  162. return (__m128i)__builtin_ia32_psignb128((__v16qi)__a, (__v16qi)__b);
  163. }
  164. static __inline__ __m128i __DEFAULT_FN_ATTRS
  165. _mm_sign_epi16(__m128i __a, __m128i __b)
  166. {
  167. return (__m128i)__builtin_ia32_psignw128((__v8hi)__a, (__v8hi)__b);
  168. }
  169. static __inline__ __m128i __DEFAULT_FN_ATTRS
  170. _mm_sign_epi32(__m128i __a, __m128i __b)
  171. {
  172. return (__m128i)__builtin_ia32_psignd128((__v4si)__a, (__v4si)__b);
  173. }
  174. static __inline__ __m64 __DEFAULT_FN_ATTRS
  175. _mm_sign_pi8(__m64 __a, __m64 __b)
  176. {
  177. return (__m64)__builtin_ia32_psignb((__v8qi)__a, (__v8qi)__b);
  178. }
  179. static __inline__ __m64 __DEFAULT_FN_ATTRS
  180. _mm_sign_pi16(__m64 __a, __m64 __b)
  181. {
  182. return (__m64)__builtin_ia32_psignw((__v4hi)__a, (__v4hi)__b);
  183. }
  184. static __inline__ __m64 __DEFAULT_FN_ATTRS
  185. _mm_sign_pi32(__m64 __a, __m64 __b)
  186. {
  187. return (__m64)__builtin_ia32_psignd((__v2si)__a, (__v2si)__b);
  188. }
  189. #undef __DEFAULT_FN_ATTRS
  190. #endif /* __SSSE3__ */
  191. #endif /* __TMMINTRIN_H */