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.

143 lines
4.0KB

  1. /*
  2. * RV30/40 decoder common dsp functions
  3. * Copyright (c) 2007 Mike Melanson, Konstantin Shishkov
  4. * Copyright (c) 2011 Janne Grunau
  5. *
  6. * This file is part of Libav.
  7. *
  8. * Libav is free software; you can redistribute it and/or
  9. * modify it under the terms of the GNU Lesser General Public
  10. * License as published by the Free Software Foundation; either
  11. * version 2.1 of the License, or (at your option) any later version.
  12. *
  13. * Libav is distributed in the hope that it will be useful,
  14. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  15. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  16. * Lesser General Public License for more details.
  17. *
  18. * You should have received a copy of the GNU Lesser General Public
  19. * License along with Libav; if not, write to the Free Software
  20. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  21. */
  22. /**
  23. * @file
  24. * RV30/40 decoder common dsp functions
  25. */
  26. #include "dsputil.h"
  27. #include "rv34dsp.h"
  28. #include "libavutil/common.h"
  29. /**
  30. * @name RV30/40 inverse transform functions
  31. * @{
  32. */
  33. static av_always_inline void rv34_row_transform(int temp[16], DCTELEM *block)
  34. {
  35. int i;
  36. for(i = 0; i < 4; i++){
  37. const int z0 = 13*(block[i+4*0] + block[i+4*2]);
  38. const int z1 = 13*(block[i+4*0] - block[i+4*2]);
  39. const int z2 = 7* block[i+4*1] - 17*block[i+4*3];
  40. const int z3 = 17* block[i+4*1] + 7*block[i+4*3];
  41. temp[4*i+0] = z0 + z3;
  42. temp[4*i+1] = z1 + z2;
  43. temp[4*i+2] = z1 - z2;
  44. temp[4*i+3] = z0 - z3;
  45. }
  46. }
  47. /**
  48. * Real Video 3.0/4.0 inverse transform + sample reconstruction
  49. * Code is almost the same as in SVQ3, only scaling is different.
  50. */
  51. static void rv34_idct_add_c(uint8_t *dst, ptrdiff_t stride, DCTELEM *block){
  52. int temp[16];
  53. int i;
  54. rv34_row_transform(temp, block);
  55. memset(block, 0, 16*sizeof(DCTELEM));
  56. for(i = 0; i < 4; i++){
  57. const int z0 = 13*(temp[4*0+i] + temp[4*2+i]) + 0x200;
  58. const int z1 = 13*(temp[4*0+i] - temp[4*2+i]) + 0x200;
  59. const int z2 = 7* temp[4*1+i] - 17*temp[4*3+i];
  60. const int z3 = 17* temp[4*1+i] + 7*temp[4*3+i];
  61. dst[0] = av_clip_uint8( dst[0] + ( (z0 + z3) >> 10 ) );
  62. dst[1] = av_clip_uint8( dst[1] + ( (z1 + z2) >> 10 ) );
  63. dst[2] = av_clip_uint8( dst[2] + ( (z1 - z2) >> 10 ) );
  64. dst[3] = av_clip_uint8( dst[3] + ( (z0 - z3) >> 10 ) );
  65. dst += stride;
  66. }
  67. }
  68. /**
  69. * RealVideo 3.0/4.0 inverse transform for DC block
  70. *
  71. * Code is almost the same as rv34_inv_transform()
  72. * but final coefficients are multiplied by 1.5 and have no rounding.
  73. */
  74. static void rv34_inv_transform_noround_c(DCTELEM *block){
  75. int temp[16];
  76. int i;
  77. rv34_row_transform(temp, block);
  78. for(i = 0; i < 4; i++){
  79. const int z0 = 39*(temp[4*0+i] + temp[4*2+i]);
  80. const int z1 = 39*(temp[4*0+i] - temp[4*2+i]);
  81. const int z2 = 21* temp[4*1+i] - 51*temp[4*3+i];
  82. const int z3 = 51* temp[4*1+i] + 21*temp[4*3+i];
  83. block[i*4+0] = (z0 + z3) >> 11;
  84. block[i*4+1] = (z1 + z2) >> 11;
  85. block[i*4+2] = (z1 - z2) >> 11;
  86. block[i*4+3] = (z0 - z3) >> 11;
  87. }
  88. }
  89. static void rv34_idct_dc_add_c(uint8_t *dst, ptrdiff_t stride, int dc)
  90. {
  91. int i, j;
  92. dc = (13*13*dc + 0x200) >> 10;
  93. for (i = 0; i < 4; i++)
  94. {
  95. for (j = 0; j < 4; j++)
  96. dst[j] = av_clip_uint8( dst[j] + dc );
  97. dst += stride;
  98. }
  99. }
  100. static void rv34_inv_transform_dc_noround_c(DCTELEM *block)
  101. {
  102. DCTELEM dc = (13 * 13 * 3 * block[0]) >> 11;
  103. int i, j;
  104. for (i = 0; i < 4; i++, block += 4)
  105. for (j = 0; j < 4; j++)
  106. block[j] = dc;
  107. }
  108. /** @} */ // transform
  109. av_cold void ff_rv34dsp_init(RV34DSPContext *c, DSPContext* dsp) {
  110. c->rv34_inv_transform = rv34_inv_transform_noround_c;
  111. c->rv34_inv_transform_dc = rv34_inv_transform_dc_noround_c;
  112. c->rv34_idct_add = rv34_idct_add_c;
  113. c->rv34_idct_dc_add = rv34_idct_dc_add_c;
  114. if (HAVE_NEON)
  115. ff_rv34dsp_init_neon(c, dsp);
  116. if (HAVE_MMX)
  117. ff_rv34dsp_init_x86(c, dsp);
  118. }