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.

134 lines
7.1KB

  1. /*
  2. * Copyright (c) 2010 Mans Rullgard <mans@mansr.com>
  3. *
  4. * This file is part of FFmpeg.
  5. *
  6. * FFmpeg is free software; you can redistribute it and/or
  7. * modify it under the terms of the GNU Lesser General Public
  8. * License as published by the Free Software Foundation; either
  9. * version 2.1 of the License, or (at your option) any later version.
  10. *
  11. * FFmpeg is distributed in the hope that it will be useful,
  12. * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13. * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
  14. * Lesser General Public License for more details.
  15. *
  16. * You should have received a copy of the GNU Lesser General Public
  17. * License along with FFmpeg; if not, write to the Free Software
  18. * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19. */
  20. #include <stdint.h>
  21. #include "libavcodec/dsputil.h"
  22. #include "libavcodec/h264dsp.h"
  23. void ff_h264_v_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
  24. int beta, int8_t *tc0);
  25. void ff_h264_h_loop_filter_luma_neon(uint8_t *pix, int stride, int alpha,
  26. int beta, int8_t *tc0);
  27. void ff_h264_v_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
  28. int beta, int8_t *tc0);
  29. void ff_h264_h_loop_filter_chroma_neon(uint8_t *pix, int stride, int alpha,
  30. int beta, int8_t *tc0);
  31. void ff_weight_h264_pixels_16x16_neon(uint8_t *ds, int stride, int log2_den,
  32. int weight, int offset);
  33. void ff_weight_h264_pixels_16x8_neon(uint8_t *ds, int stride, int log2_den,
  34. int weight, int offset);
  35. void ff_weight_h264_pixels_8x16_neon(uint8_t *ds, int stride, int log2_den,
  36. int weight, int offset);
  37. void ff_weight_h264_pixels_8x8_neon(uint8_t *ds, int stride, int log2_den,
  38. int weight, int offset);
  39. void ff_weight_h264_pixels_8x4_neon(uint8_t *ds, int stride, int log2_den,
  40. int weight, int offset);
  41. void ff_weight_h264_pixels_4x8_neon(uint8_t *ds, int stride, int log2_den,
  42. int weight, int offset);
  43. void ff_weight_h264_pixels_4x4_neon(uint8_t *ds, int stride, int log2_den,
  44. int weight, int offset);
  45. void ff_weight_h264_pixels_4x2_neon(uint8_t *ds, int stride, int log2_den,
  46. int weight, int offset);
  47. void ff_biweight_h264_pixels_16x16_neon(uint8_t *dst, uint8_t *src, int stride,
  48. int log2_den, int weightd, int weights,
  49. int offset);
  50. void ff_biweight_h264_pixels_16x8_neon(uint8_t *dst, uint8_t *src, int stride,
  51. int log2_den, int weightd, int weights,
  52. int offset);
  53. void ff_biweight_h264_pixels_8x16_neon(uint8_t *dst, uint8_t *src, int stride,
  54. int log2_den, int weightd, int weights,
  55. int offset);
  56. void ff_biweight_h264_pixels_8x8_neon(uint8_t *dst, uint8_t *src, int stride,
  57. int log2_den, int weightd, int weights,
  58. int offset);
  59. void ff_biweight_h264_pixels_8x4_neon(uint8_t *dst, uint8_t *src, int stride,
  60. int log2_den, int weightd, int weights,
  61. int offset);
  62. void ff_biweight_h264_pixels_4x8_neon(uint8_t *dst, uint8_t *src, int stride,
  63. int log2_den, int weightd, int weights,
  64. int offset);
  65. void ff_biweight_h264_pixels_4x4_neon(uint8_t *dst, uint8_t *src, int stride,
  66. int log2_den, int weightd, int weights,
  67. int offset);
  68. void ff_biweight_h264_pixels_4x2_neon(uint8_t *dst, uint8_t *src, int stride,
  69. int log2_den, int weightd, int weights,
  70. int offset);
  71. void ff_h264_idct_add_neon(uint8_t *dst, DCTELEM *block, int stride);
  72. void ff_h264_idct_dc_add_neon(uint8_t *dst, DCTELEM *block, int stride);
  73. void ff_h264_idct_add16_neon(uint8_t *dst, const int *block_offset,
  74. DCTELEM *block, int stride,
  75. const uint8_t nnzc[6*8]);
  76. void ff_h264_idct_add16intra_neon(uint8_t *dst, const int *block_offset,
  77. DCTELEM *block, int stride,
  78. const uint8_t nnzc[6*8]);
  79. void ff_h264_idct_add8_neon(uint8_t **dest, const int *block_offset,
  80. DCTELEM *block, int stride,
  81. const uint8_t nnzc[6*8]);
  82. void ff_h264_idct8_add_neon(uint8_t *dst, DCTELEM *block, int stride);
  83. void ff_h264_idct8_dc_add_neon(uint8_t *dst, DCTELEM *block, int stride);
  84. void ff_h264_idct8_add4_neon(uint8_t *dst, const int *block_offset,
  85. DCTELEM *block, int stride,
  86. const uint8_t nnzc[6*8]);
  87. static void ff_h264dsp_init_neon(H264DSPContext *c)
  88. {
  89. c->h264_v_loop_filter_luma = ff_h264_v_loop_filter_luma_neon;
  90. c->h264_h_loop_filter_luma = ff_h264_h_loop_filter_luma_neon;
  91. c->h264_v_loop_filter_chroma = ff_h264_v_loop_filter_chroma_neon;
  92. c->h264_h_loop_filter_chroma = ff_h264_h_loop_filter_chroma_neon;
  93. c->weight_h264_pixels_tab[0] = ff_weight_h264_pixels_16x16_neon;
  94. c->weight_h264_pixels_tab[1] = ff_weight_h264_pixels_16x8_neon;
  95. c->weight_h264_pixels_tab[2] = ff_weight_h264_pixels_8x16_neon;
  96. c->weight_h264_pixels_tab[3] = ff_weight_h264_pixels_8x8_neon;
  97. c->weight_h264_pixels_tab[4] = ff_weight_h264_pixels_8x4_neon;
  98. c->weight_h264_pixels_tab[5] = ff_weight_h264_pixels_4x8_neon;
  99. c->weight_h264_pixels_tab[6] = ff_weight_h264_pixels_4x4_neon;
  100. c->weight_h264_pixels_tab[7] = ff_weight_h264_pixels_4x2_neon;
  101. c->biweight_h264_pixels_tab[0] = ff_biweight_h264_pixels_16x16_neon;
  102. c->biweight_h264_pixels_tab[1] = ff_biweight_h264_pixels_16x8_neon;
  103. c->biweight_h264_pixels_tab[2] = ff_biweight_h264_pixels_8x16_neon;
  104. c->biweight_h264_pixels_tab[3] = ff_biweight_h264_pixels_8x8_neon;
  105. c->biweight_h264_pixels_tab[4] = ff_biweight_h264_pixels_8x4_neon;
  106. c->biweight_h264_pixels_tab[5] = ff_biweight_h264_pixels_4x8_neon;
  107. c->biweight_h264_pixels_tab[6] = ff_biweight_h264_pixels_4x4_neon;
  108. c->biweight_h264_pixels_tab[7] = ff_biweight_h264_pixels_4x2_neon;
  109. c->h264_idct_add = ff_h264_idct_add_neon;
  110. c->h264_idct_dc_add = ff_h264_idct_dc_add_neon;
  111. c->h264_idct_add16 = ff_h264_idct_add16_neon;
  112. c->h264_idct_add16intra = ff_h264_idct_add16intra_neon;
  113. c->h264_idct_add8 = ff_h264_idct_add8_neon;
  114. c->h264_idct8_add = ff_h264_idct8_add_neon;
  115. c->h264_idct8_dc_add = ff_h264_idct8_dc_add_neon;
  116. c->h264_idct8_add4 = ff_h264_idct8_add4_neon;
  117. }
  118. void ff_h264dsp_init_arm(H264DSPContext *c)
  119. {
  120. if (HAVE_NEON) ff_h264dsp_init_neon(c);
  121. }