You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

150 lines
6.6KB

  1. /*
  2. * This file is part of Libav.
  3. *
  4. * Libav is free software; you can redistribute it and/or
  5. * modify it under the terms of the GNU Lesser General Public
  6. * License as published by the Free Software Foundation; either
  7. * version 2.1 of the License, or (at your option) any later version.
  8. *
  9. * Libav is distributed in the hope that it will be useful,
  10. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  11. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  12. * Lesser General Public License for more details.
  13. *
  14. * You should have received a copy of the GNU Lesser General Public
  15. * License along with Libav; if not, write to the Free Software
  16. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  17. */
  18. .macro transpose_8x8B r0, r1, r2, r3, r4, r5, r6, r7, r8, r9
  19. trn1 \r8\().8B, \r0\().8B, \r1\().8B
  20. trn2 \r9\().8B, \r0\().8B, \r1\().8B
  21. trn1 \r1\().8B, \r2\().8B, \r3\().8B
  22. trn2 \r3\().8B, \r2\().8B, \r3\().8B
  23. trn1 \r0\().8B, \r4\().8B, \r5\().8B
  24. trn2 \r5\().8B, \r4\().8B, \r5\().8B
  25. trn1 \r2\().8B, \r6\().8B, \r7\().8B
  26. trn2 \r7\().8B, \r6\().8B, \r7\().8B
  27. trn1 \r4\().4H, \r0\().4H, \r2\().4H
  28. trn2 \r2\().4H, \r0\().4H, \r2\().4H
  29. trn1 \r6\().4H, \r5\().4H, \r7\().4H
  30. trn2 \r7\().4H, \r5\().4H, \r7\().4H
  31. trn1 \r5\().4H, \r9\().4H, \r3\().4H
  32. trn2 \r9\().4H, \r9\().4H, \r3\().4H
  33. trn1 \r3\().4H, \r8\().4H, \r1\().4H
  34. trn2 \r8\().4H, \r8\().4H, \r1\().4H
  35. trn1 \r0\().2S, \r3\().2S, \r4\().2S
  36. trn2 \r4\().2S, \r3\().2S, \r4\().2S
  37. trn1 \r1\().2S, \r5\().2S, \r6\().2S
  38. trn2 \r5\().2S, \r5\().2S, \r6\().2S
  39. trn2 \r6\().2S, \r8\().2S, \r2\().2S
  40. trn1 \r2\().2S, \r8\().2S, \r2\().2S
  41. trn1 \r3\().2S, \r9\().2S, \r7\().2S
  42. trn2 \r7\().2S, \r9\().2S, \r7\().2S
  43. .endm
  44. .macro transpose_8x16B r0, r1, r2, r3, r4, r5, r6, r7, t0, t1
  45. trn1 \t0\().16B, \r0\().16B, \r1\().16B
  46. trn2 \t1\().16B, \r0\().16B, \r1\().16B
  47. trn1 \r1\().16B, \r2\().16B, \r3\().16B
  48. trn2 \r3\().16B, \r2\().16B, \r3\().16B
  49. trn1 \r0\().16B, \r4\().16B, \r5\().16B
  50. trn2 \r5\().16B, \r4\().16B, \r5\().16B
  51. trn1 \r2\().16B, \r6\().16B, \r7\().16B
  52. trn2 \r7\().16B, \r6\().16B, \r7\().16B
  53. trn1 \r4\().8H, \r0\().8H, \r2\().8H
  54. trn2 \r2\().8H, \r0\().8H, \r2\().8H
  55. trn1 \r6\().8H, \r5\().8H, \r7\().8H
  56. trn2 \r7\().8H, \r5\().8H, \r7\().8H
  57. trn1 \r5\().8H, \t1\().8H, \r3\().8H
  58. trn2 \t1\().8H, \t1\().8H, \r3\().8H
  59. trn1 \r3\().8H, \t0\().8H, \r1\().8H
  60. trn2 \t0\().8H, \t0\().8H, \r1\().8H
  61. trn1 \r0\().4S, \r3\().4S, \r4\().4S
  62. trn2 \r4\().4S, \r3\().4S, \r4\().4S
  63. trn1 \r1\().4S, \r5\().4S, \r6\().4S
  64. trn2 \r5\().4S, \r5\().4S, \r6\().4S
  65. trn2 \r6\().4S, \t0\().4S, \r2\().4S
  66. trn1 \r2\().4S, \t0\().4S, \r2\().4S
  67. trn1 \r3\().4S, \t1\().4S, \r7\().4S
  68. trn2 \r7\().4S, \t1\().4S, \r7\().4S
  69. .endm
  70. .macro transpose_4x16B r0, r1, r2, r3, t4, t5, t6, t7
  71. trn1 \t4\().16B, \r0\().16B, \r1\().16B
  72. trn2 \t5\().16B, \r0\().16B, \r1\().16B
  73. trn1 \t6\().16B, \r2\().16B, \r3\().16B
  74. trn2 \t7\().16B, \r2\().16B, \r3\().16B
  75. trn1 \r0\().8H, \t4\().8H, \t6\().8H
  76. trn2 \r2\().8H, \t4\().8H, \t6\().8H
  77. trn1 \r1\().8H, \t5\().8H, \t7\().8H
  78. trn2 \r3\().8H, \t5\().8H, \t7\().8H
  79. .endm
  80. .macro transpose_4x8B r0, r1, r2, r3, t4, t5, t6, t7
  81. trn1 \t4\().8B, \r0\().8B, \r1\().8B
  82. trn2 \t5\().8B, \r0\().8B, \r1\().8B
  83. trn1 \t6\().8B, \r2\().8B, \r3\().8B
  84. trn2 \t7\().8B, \r2\().8B, \r3\().8B
  85. trn1 \r0\().4H, \t4\().4H, \t6\().4H
  86. trn2 \r2\().4H, \t4\().4H, \t6\().4H
  87. trn1 \r1\().4H, \t5\().4H, \t7\().4H
  88. trn2 \r3\().4H, \t5\().4H, \t7\().4H
  89. .endm
  90. .macro transpose_4x4H r0, r1, r2, r3, r4, r5, r6, r7
  91. trn1 \r4\().4H, \r0\().4H, \r1\().4H
  92. trn2 \r5\().4H, \r0\().4H, \r1\().4H
  93. trn1 \r6\().4H, \r2\().4H, \r3\().4H
  94. trn2 \r7\().4H, \r2\().4H, \r3\().4H
  95. trn1 \r0\().2S, \r4\().2S, \r6\().2S
  96. trn2 \r2\().2S, \r4\().2S, \r6\().2S
  97. trn1 \r1\().2S, \r5\().2S, \r7\().2S
  98. trn2 \r3\().2S, \r5\().2S, \r7\().2S
  99. .endm
  100. .macro transpose_8x8H r0, r1, r2, r3, r4, r5, r6, r7, r8, r9
  101. trn1 \r8\().8H, \r0\().8H, \r1\().8H
  102. trn2 \r9\().8H, \r0\().8H, \r1\().8H
  103. trn1 \r1\().8H, \r2\().8H, \r3\().8H
  104. trn2 \r3\().8H, \r2\().8H, \r3\().8H
  105. trn1 \r0\().8H, \r4\().8H, \r5\().8H
  106. trn2 \r5\().8H, \r4\().8H, \r5\().8H
  107. trn1 \r2\().8H, \r6\().8H, \r7\().8H
  108. trn2 \r7\().8H, \r6\().8H, \r7\().8H
  109. trn1 \r4\().4S, \r0\().4S, \r2\().4S
  110. trn2 \r2\().4S, \r0\().4S, \r2\().4S
  111. trn1 \r6\().4S, \r5\().4S, \r7\().4S
  112. trn2 \r7\().4S, \r5\().4S, \r7\().4S
  113. trn1 \r5\().4S, \r9\().4S, \r3\().4S
  114. trn2 \r9\().4S, \r9\().4S, \r3\().4S
  115. trn1 \r3\().4S, \r8\().4S, \r1\().4S
  116. trn2 \r8\().4S, \r8\().4S, \r1\().4S
  117. trn1 \r0\().2D, \r3\().2D, \r4\().2D
  118. trn2 \r4\().2D, \r3\().2D, \r4\().2D
  119. trn1 \r1\().2D, \r5\().2D, \r6\().2D
  120. trn2 \r5\().2D, \r5\().2D, \r6\().2D
  121. trn2 \r6\().2D, \r8\().2D, \r2\().2D
  122. trn1 \r2\().2D, \r8\().2D, \r2\().2D
  123. trn1 \r3\().2D, \r9\().2D, \r7\().2D
  124. trn2 \r7\().2D, \r9\().2D, \r7\().2D
  125. .endm