Home Explore Help
Register Sign In
third
/
srs
mirror of https://github.com/ossrs/srs
2
0
Fork 0
Files
Tree: c2d75ca395
Branches Tags
srs
/
trunk
/
3rdparty
/
ffmpeg-4-fit
/
libavcodec
/
aarch64
/
h264qpel_init_aarch64.c

h264qpel_init_aarch64.c 11 KB
History Raw

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172 
  1. /*
    2. 

  2.  * ARM NEON optimised DSP functions
    3. 

  3.  * Copyright (c) 2008 Mans Rullgard <mans@mansr.com>
    4. 

  4.  *
    5. 

  5.  * This file is part of FFmpeg.
    6. 

  6.  *
    7. 

  7.  * FFmpeg is free software; you can redistribute it and/or
    8. 

  8.  * modify it under the terms of the GNU Lesser General Public
    9. 

  9.  * License as published by the Free Software Foundation; either
    10. 

  10.  * version 2.1 of the License, or (at your option) any later version.
    11. 

  11.  *
    12. 

  12.  * FFmpeg is distributed in the hope that it will be useful,
    13. 

  13.  * but WITHOUT ANY WARRANTY; without even the implied warranty of
    14. 

  14.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
    15. 

  15.  * Lesser General Public License for more details.
    16. 

  16.  *
    17. 

  17.  * You should have received a copy of the GNU Lesser General Public
    18. 

  18.  * License along with FFmpeg; if not, write to the Free Software
    19. 

  19.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
    20. 

  20.  */
    21. 

  21. 

  22. #include <stdint.h>
    23. 

  23. 

  24. #include "config.h"
    25. 

  25. #include "libavutil/attributes.h"
    26. 

  26. #include "libavutil/cpu.h"
    27. 

  27. #include "libavutil/aarch64/cpu.h"
    28. 

  28. #include "libavcodec/h264qpel.h"
    29. 

  29. 

  30. void ff_put_h264_qpel16_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    31. 

  31. void ff_put_h264_qpel16_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    32. 

  32. void ff_put_h264_qpel16_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    33. 

  33. void ff_put_h264_qpel16_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    34. 

  34. void ff_put_h264_qpel16_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    35. 

  35. void ff_put_h264_qpel16_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    36. 

  36. void ff_put_h264_qpel16_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    37. 

  37. void ff_put_h264_qpel16_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    38. 

  38. void ff_put_h264_qpel16_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    39. 

  39. void ff_put_h264_qpel16_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    40. 

  40. void ff_put_h264_qpel16_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    41. 

  41. void ff_put_h264_qpel16_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    42. 

  42. void ff_put_h264_qpel16_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    43. 

  43. void ff_put_h264_qpel16_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    44. 

  44. void ff_put_h264_qpel16_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    45. 

  45. void ff_put_h264_qpel16_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    46. 

  46. 

  47. void ff_put_h264_qpel8_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    48. 

  48. void ff_put_h264_qpel8_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    49. 

  49. void ff_put_h264_qpel8_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    50. 

  50. void ff_put_h264_qpel8_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    51. 

  51. void ff_put_h264_qpel8_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    52. 

  52. void ff_put_h264_qpel8_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    53. 

  53. void ff_put_h264_qpel8_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    54. 

  54. void ff_put_h264_qpel8_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    55. 

  55. void ff_put_h264_qpel8_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    56. 

  56. void ff_put_h264_qpel8_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    57. 

  57. void ff_put_h264_qpel8_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    58. 

  58. void ff_put_h264_qpel8_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    59. 

  59. void ff_put_h264_qpel8_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    60. 

  60. void ff_put_h264_qpel8_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    61. 

  61. void ff_put_h264_qpel8_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    62. 

  62. void ff_put_h264_qpel8_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    63. 

  63. 

  64. void ff_avg_h264_qpel16_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    65. 

  65. void ff_avg_h264_qpel16_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    66. 

  66. void ff_avg_h264_qpel16_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    67. 

  67. void ff_avg_h264_qpel16_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    68. 

  68. void ff_avg_h264_qpel16_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    69. 

  69. void ff_avg_h264_qpel16_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    70. 

  70. void ff_avg_h264_qpel16_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    71. 

  71. void ff_avg_h264_qpel16_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    72. 

  72. void ff_avg_h264_qpel16_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    73. 

  73. void ff_avg_h264_qpel16_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    74. 

  74. void ff_avg_h264_qpel16_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    75. 

  75. void ff_avg_h264_qpel16_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    76. 

  76. void ff_avg_h264_qpel16_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    77. 

  77. void ff_avg_h264_qpel16_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    78. 

  78. void ff_avg_h264_qpel16_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    79. 

  79. void ff_avg_h264_qpel16_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    80. 

  80. 

  81. void ff_avg_h264_qpel8_mc00_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    82. 

  82. void ff_avg_h264_qpel8_mc10_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    83. 

  83. void ff_avg_h264_qpel8_mc20_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    84. 

  84. void ff_avg_h264_qpel8_mc30_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    85. 

  85. void ff_avg_h264_qpel8_mc01_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    86. 

  86. void ff_avg_h264_qpel8_mc11_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    87. 

  87. void ff_avg_h264_qpel8_mc21_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    88. 

  88. void ff_avg_h264_qpel8_mc31_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    89. 

  89. void ff_avg_h264_qpel8_mc02_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    90. 

  90. void ff_avg_h264_qpel8_mc12_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    91. 

  91. void ff_avg_h264_qpel8_mc22_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    92. 

  92. void ff_avg_h264_qpel8_mc32_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    93. 

  93. void ff_avg_h264_qpel8_mc03_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    94. 

  94. void ff_avg_h264_qpel8_mc13_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    95. 

  95. void ff_avg_h264_qpel8_mc23_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    96. 

  96. void ff_avg_h264_qpel8_mc33_neon(uint8_t *dst, const uint8_t *src, ptrdiff_t stride);
    97. 

  97. 

  98. av_cold void ff_h264qpel_init_aarch64(H264QpelContext *c, int bit_depth)
    99. 

  99. {
    100. 

  100.     const int high_bit_depth = bit_depth > 8;
    101. 

  101.     int cpu_flags = av_get_cpu_flags();
    102. 

  102. 

  103.     if (have_neon(cpu_flags) && !high_bit_depth) {
    104. 

  104.         c->put_h264_qpel_pixels_tab[0][ 0] = ff_put_h264_qpel16_mc00_neon;
    105. 

  105.         c->put_h264_qpel_pixels_tab[0][ 1] = ff_put_h264_qpel16_mc10_neon;
    106. 

  106.         c->put_h264_qpel_pixels_tab[0][ 2] = ff_put_h264_qpel16_mc20_neon;
    107. 

  107.         c->put_h264_qpel_pixels_tab[0][ 3] = ff_put_h264_qpel16_mc30_neon;
    108. 

  108.         c->put_h264_qpel_pixels_tab[0][ 4] = ff_put_h264_qpel16_mc01_neon;
    109. 

  109.         c->put_h264_qpel_pixels_tab[0][ 5] = ff_put_h264_qpel16_mc11_neon;
    110. 

  110.         c->put_h264_qpel_pixels_tab[0][ 6] = ff_put_h264_qpel16_mc21_neon;
    111. 

  111.         c->put_h264_qpel_pixels_tab[0][ 7] = ff_put_h264_qpel16_mc31_neon;
    112. 

  112.         c->put_h264_qpel_pixels_tab[0][ 8] = ff_put_h264_qpel16_mc02_neon;
    113. 

  113.         c->put_h264_qpel_pixels_tab[0][ 9] = ff_put_h264_qpel16_mc12_neon;
    114. 

  114.         c->put_h264_qpel_pixels_tab[0][10] = ff_put_h264_qpel16_mc22_neon;
    115. 

  115.         c->put_h264_qpel_pixels_tab[0][11] = ff_put_h264_qpel16_mc32_neon;
    116. 

  116.         c->put_h264_qpel_pixels_tab[0][12] = ff_put_h264_qpel16_mc03_neon;
    117. 

  117.         c->put_h264_qpel_pixels_tab[0][13] = ff_put_h264_qpel16_mc13_neon;
    118. 

  118.         c->put_h264_qpel_pixels_tab[0][14] = ff_put_h264_qpel16_mc23_neon;
    119. 

  119.         c->put_h264_qpel_pixels_tab[0][15] = ff_put_h264_qpel16_mc33_neon;
    120. 

  120. 

  121.         c->put_h264_qpel_pixels_tab[1][ 0] = ff_put_h264_qpel8_mc00_neon;
    122. 

  122.         c->put_h264_qpel_pixels_tab[1][ 1] = ff_put_h264_qpel8_mc10_neon;
    123. 

  123.         c->put_h264_qpel_pixels_tab[1][ 2] = ff_put_h264_qpel8_mc20_neon;
    124. 

  124.         c->put_h264_qpel_pixels_tab[1][ 3] = ff_put_h264_qpel8_mc30_neon;
    125. 

  125.         c->put_h264_qpel_pixels_tab[1][ 4] = ff_put_h264_qpel8_mc01_neon;
    126. 

  126.         c->put_h264_qpel_pixels_tab[1][ 5] = ff_put_h264_qpel8_mc11_neon;
    127. 

  127.         c->put_h264_qpel_pixels_tab[1][ 6] = ff_put_h264_qpel8_mc21_neon;
    128. 

  128.         c->put_h264_qpel_pixels_tab[1][ 7] = ff_put_h264_qpel8_mc31_neon;
    129. 

  129.         c->put_h264_qpel_pixels_tab[1][ 8] = ff_put_h264_qpel8_mc02_neon;
    130. 

  130.         c->put_h264_qpel_pixels_tab[1][ 9] = ff_put_h264_qpel8_mc12_neon;
    131. 

  131.         c->put_h264_qpel_pixels_tab[1][10] = ff_put_h264_qpel8_mc22_neon;
    132. 

  132.         c->put_h264_qpel_pixels_tab[1][11] = ff_put_h264_qpel8_mc32_neon;
    133. 

  133.         c->put_h264_qpel_pixels_tab[1][12] = ff_put_h264_qpel8_mc03_neon;
    134. 

  134.         c->put_h264_qpel_pixels_tab[1][13] = ff_put_h264_qpel8_mc13_neon;
    135. 

  135.         c->put_h264_qpel_pixels_tab[1][14] = ff_put_h264_qpel8_mc23_neon;
    136. 

  136.         c->put_h264_qpel_pixels_tab[1][15] = ff_put_h264_qpel8_mc33_neon;
    137. 

  137. 

  138.         c->avg_h264_qpel_pixels_tab[0][ 0] = ff_avg_h264_qpel16_mc00_neon;
    139. 

  139.         c->avg_h264_qpel_pixels_tab[0][ 1] = ff_avg_h264_qpel16_mc10_neon;
    140. 

  140.         c->avg_h264_qpel_pixels_tab[0][ 2] = ff_avg_h264_qpel16_mc20_neon;
    141. 

  141.         c->avg_h264_qpel_pixels_tab[0][ 3] = ff_avg_h264_qpel16_mc30_neon;
    142. 

  142.         c->avg_h264_qpel_pixels_tab[0][ 4] = ff_avg_h264_qpel16_mc01_neon;
    143. 

  143.         c->avg_h264_qpel_pixels_tab[0][ 5] = ff_avg_h264_qpel16_mc11_neon;
    144. 

  144.         c->avg_h264_qpel_pixels_tab[0][ 6] = ff_avg_h264_qpel16_mc21_neon;
    145. 

  145.         c->avg_h264_qpel_pixels_tab[0][ 7] = ff_avg_h264_qpel16_mc31_neon;
    146. 

  146.         c->avg_h264_qpel_pixels_tab[0][ 8] = ff_avg_h264_qpel16_mc02_neon;
    147. 

  147.         c->avg_h264_qpel_pixels_tab[0][ 9] = ff_avg_h264_qpel16_mc12_neon;
    148. 

  148.         c->avg_h264_qpel_pixels_tab[0][10] = ff_avg_h264_qpel16_mc22_neon;
    149. 

  149.         c->avg_h264_qpel_pixels_tab[0][11] = ff_avg_h264_qpel16_mc32_neon;
    150. 

  150.         c->avg_h264_qpel_pixels_tab[0][12] = ff_avg_h264_qpel16_mc03_neon;
    151. 

  151.         c->avg_h264_qpel_pixels_tab[0][13] = ff_avg_h264_qpel16_mc13_neon;
    152. 

  152.         c->avg_h264_qpel_pixels_tab[0][14] = ff_avg_h264_qpel16_mc23_neon;
    153. 

  153.         c->avg_h264_qpel_pixels_tab[0][15] = ff_avg_h264_qpel16_mc33_neon;
    154. 

  154. 

  155.         c->avg_h264_qpel_pixels_tab[1][ 0] = ff_avg_h264_qpel8_mc00_neon;
    156. 

  156.         c->avg_h264_qpel_pixels_tab[1][ 1] = ff_avg_h264_qpel8_mc10_neon;
    157. 

  157.         c->avg_h264_qpel_pixels_tab[1][ 2] = ff_avg_h264_qpel8_mc20_neon;
    158. 

  158.         c->avg_h264_qpel_pixels_tab[1][ 3] = ff_avg_h264_qpel8_mc30_neon;
    159. 

  159.         c->avg_h264_qpel_pixels_tab[1][ 4] = ff_avg_h264_qpel8_mc01_neon;
    160. 

  160.         c->avg_h264_qpel_pixels_tab[1][ 5] = ff_avg_h264_qpel8_mc11_neon;
    161. 

  161.         c->avg_h264_qpel_pixels_tab[1][ 6] = ff_avg_h264_qpel8_mc21_neon;
    162. 

  162.         c->avg_h264_qpel_pixels_tab[1][ 7] = ff_avg_h264_qpel8_mc31_neon;
    163. 

  163.         c->avg_h264_qpel_pixels_tab[1][ 8] = ff_avg_h264_qpel8_mc02_neon;
    164. 

  164.         c->avg_h264_qpel_pixels_tab[1][ 9] = ff_avg_h264_qpel8_mc12_neon;
    165. 

  165.         c->avg_h264_qpel_pixels_tab[1][10] = ff_avg_h264_qpel8_mc22_neon;
    166. 

  166.         c->avg_h264_qpel_pixels_tab[1][11] = ff_avg_h264_qpel8_mc32_neon;
    167. 

  167.         c->avg_h264_qpel_pixels_tab[1][12] = ff_avg_h264_qpel8_mc03_neon;
    168. 

  168.         c->avg_h264_qpel_pixels_tab[1][13] = ff_avg_h264_qpel8_mc13_neon;
    169. 

  169.         c->avg_h264_qpel_pixels_tab[1][14] = ff_avg_h264_qpel8_mc23_neon;
    170. 

  170.         c->avg_h264_qpel_pixels_tab[1][15] = ff_avg_h264_qpel8_mc33_neon;
    171. 

  171.     }
    172. 

  172. }
    173.