falkTX
/
FFmpeg
mirror of https://github.com/falkTX/FFmpeg.git
1
0
Fork 0
Code Issues 0 Releases 338 Wiki Activity
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.
40140 Commits
32 Branches
325MB
Tree: e3fcb14347
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e3fcb14347'
${ noResults }
FFmpeg/libavcodec/dsputil.c

1156 lines
37KB
Raw Blame History 

  1. /*

  2.  * DSP utils

  3.  * Copyright (c) 2000, 2001 Fabrice Bellard

  4.  * Copyright (c) 2002-2004 Michael Niedermayer <michaelni@gmx.at>

  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. /**

  24.  * @file

  25.  * DSP utils

  26.  */

  27. 

  28. #include "libavutil/attributes.h"

  29. #include "libavutil/imgutils.h"

  30. #include "avcodec.h"

  31. #include "copy_block.h"

  32. #include "dct.h"

  33. #include "dsputil.h"

  34. #include "simple_idct.h"

  35. #include "faandct.h"

  36. #include "imgconvert.h"

  37. #include "mathops.h"

  38. #include "mpegvideo.h"

  39. #include "config.h"

  40. 

  41. uint32_t ff_square_tab[512] = { 0, };

  42. 

  43. #define BIT_DEPTH 16

  44. #include "dsputilenc_template.c"

  45. #undef BIT_DEPTH

  46. 

  47. #define BIT_DEPTH 8

  48. #include "dsputilenc_template.c"

  49. 

  50. static int pix_sum_c(uint8_t *pix, int line_size)

  51. {

  52.     int s = 0, i, j;

  53. 

  54.     for (i = 0; i < 16; i++) {

  55.         for (j = 0; j < 16; j += 8) {

  56.             s   += pix[0];

  57.             s   += pix[1];

  58.             s   += pix[2];

  59.             s   += pix[3];

  60.             s   += pix[4];

  61.             s   += pix[5];

  62.             s   += pix[6];

  63.             s   += pix[7];

  64.             pix += 8;

  65.         }

  66.         pix += line_size - 16;

  67.     }

  68.     return s;

  69. }

  70. 

  71. static int pix_norm1_c(uint8_t *pix, int line_size)

  72. {

  73.     int s = 0, i, j;

  74.     uint32_t *sq = ff_square_tab + 256;

  75. 

  76.     for (i = 0; i < 16; i++) {

  77.         for (j = 0; j < 16; j += 8) {

  78. #if 0

  79.             s += sq[pix[0]];

  80.             s += sq[pix[1]];

  81.             s += sq[pix[2]];

  82.             s += sq[pix[3]];

  83.             s += sq[pix[4]];

  84.             s += sq[pix[5]];

  85.             s += sq[pix[6]];

  86.             s += sq[pix[7]];

  87. #else

  88. #if HAVE_FAST_64BIT

  89.             register uint64_t x = *(uint64_t *) pix;

  90.             s += sq[x         & 0xff];

  91.             s += sq[(x >>  8) & 0xff];

  92.             s += sq[(x >> 16) & 0xff];

  93.             s += sq[(x >> 24) & 0xff];

  94.             s += sq[(x >> 32) & 0xff];

  95.             s += sq[(x >> 40) & 0xff];

  96.             s += sq[(x >> 48) & 0xff];

  97.             s += sq[(x >> 56) & 0xff];

  98. #else

  99.             register uint32_t x = *(uint32_t *) pix;

  100.             s += sq[x         & 0xff];

  101.             s += sq[(x >>  8) & 0xff];

  102.             s += sq[(x >> 16) & 0xff];

  103.             s += sq[(x >> 24) & 0xff];

  104.             x  = *(uint32_t *) (pix + 4);

  105.             s += sq[x         & 0xff];

  106.             s += sq[(x >>  8) & 0xff];

  107.             s += sq[(x >> 16) & 0xff];

  108.             s += sq[(x >> 24) & 0xff];

  109. #endif

  110. #endif

  111.             pix += 8;

  112.         }

  113.         pix += line_size - 16;

  114.     }

  115.     return s;

  116. }

  117. 

  118. static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  119.                   int line_size, int h)

  120. {

  121.     int s = 0, i;

  122.     uint32_t *sq = ff_square_tab + 256;

  123. 

  124.     for (i = 0; i < h; i++) {

  125.         s    += sq[pix1[0] - pix2[0]];

  126.         s    += sq[pix1[1] - pix2[1]];

  127.         s    += sq[pix1[2] - pix2[2]];

  128.         s    += sq[pix1[3] - pix2[3]];

  129.         pix1 += line_size;

  130.         pix2 += line_size;

  131.     }

  132.     return s;

  133. }

  134. 

  135. static int sse8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  136.                   int line_size, int h)

  137. {

  138.     int s = 0, i;

  139.     uint32_t *sq = ff_square_tab + 256;

  140. 

  141.     for (i = 0; i < h; i++) {

  142.         s    += sq[pix1[0] - pix2[0]];

  143.         s    += sq[pix1[1] - pix2[1]];

  144.         s    += sq[pix1[2] - pix2[2]];

  145.         s    += sq[pix1[3] - pix2[3]];

  146.         s    += sq[pix1[4] - pix2[4]];

  147.         s    += sq[pix1[5] - pix2[5]];

  148.         s    += sq[pix1[6] - pix2[6]];

  149.         s    += sq[pix1[7] - pix2[7]];

  150.         pix1 += line_size;

  151.         pix2 += line_size;

  152.     }

  153.     return s;

  154. }

  155. 

  156. static int sse16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  157.                    int line_size, int h)

  158. {

  159.     int s = 0, i;

  160.     uint32_t *sq = ff_square_tab + 256;

  161. 

  162.     for (i = 0; i < h; i++) {

  163.         s += sq[pix1[0]  - pix2[0]];

  164.         s += sq[pix1[1]  - pix2[1]];

  165.         s += sq[pix1[2]  - pix2[2]];

  166.         s += sq[pix1[3]  - pix2[3]];

  167.         s += sq[pix1[4]  - pix2[4]];

  168.         s += sq[pix1[5]  - pix2[5]];

  169.         s += sq[pix1[6]  - pix2[6]];

  170.         s += sq[pix1[7]  - pix2[7]];

  171.         s += sq[pix1[8]  - pix2[8]];

  172.         s += sq[pix1[9]  - pix2[9]];

  173.         s += sq[pix1[10] - pix2[10]];

  174.         s += sq[pix1[11] - pix2[11]];

  175.         s += sq[pix1[12] - pix2[12]];

  176.         s += sq[pix1[13] - pix2[13]];

  177.         s += sq[pix1[14] - pix2[14]];

  178.         s += sq[pix1[15] - pix2[15]];

  179. 

  180.         pix1 += line_size;

  181.         pix2 += line_size;

  182.     }

  183.     return s;

  184. }

  185. 

  186. static void diff_pixels_c(int16_t *restrict block, const uint8_t *s1,

  187.                           const uint8_t *s2, int stride)

  188. {

  189.     int i;

  190. 

  191.     /* read the pixels */

  192.     for (i = 0; i < 8; i++) {

  193.         block[0] = s1[0] - s2[0];

  194.         block[1] = s1[1] - s2[1];

  195.         block[2] = s1[2] - s2[2];

  196.         block[3] = s1[3] - s2[3];

  197.         block[4] = s1[4] - s2[4];

  198.         block[5] = s1[5] - s2[5];

  199.         block[6] = s1[6] - s2[6];

  200.         block[7] = s1[7] - s2[7];

  201.         s1      += stride;

  202.         s2      += stride;

  203.         block   += 8;

  204.     }

  205. }

  206. 

  207. static int sum_abs_dctelem_c(int16_t *block)

  208. {

  209.     int sum = 0, i;

  210. 

  211.     for (i = 0; i < 64; i++)

  212.         sum += FFABS(block[i]);

  213.     return sum;

  214. }

  215. 

  216. #define avg2(a, b) ((a + b + 1) >> 1)

  217. #define avg4(a, b, c, d) ((a + b + c + d + 2) >> 2)

  218. 

  219. static inline int pix_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  220.                               int line_size, int h)

  221. {

  222.     int s = 0, i;

  223. 

  224.     for (i = 0; i < h; i++) {

  225.         s    += abs(pix1[0]  - pix2[0]);

  226.         s    += abs(pix1[1]  - pix2[1]);

  227.         s    += abs(pix1[2]  - pix2[2]);

  228.         s    += abs(pix1[3]  - pix2[3]);

  229.         s    += abs(pix1[4]  - pix2[4]);

  230.         s    += abs(pix1[5]  - pix2[5]);

  231.         s    += abs(pix1[6]  - pix2[6]);

  232.         s    += abs(pix1[7]  - pix2[7]);

  233.         s    += abs(pix1[8]  - pix2[8]);

  234.         s    += abs(pix1[9]  - pix2[9]);

  235.         s    += abs(pix1[10] - pix2[10]);

  236.         s    += abs(pix1[11] - pix2[11]);

  237.         s    += abs(pix1[12] - pix2[12]);

  238.         s    += abs(pix1[13] - pix2[13]);

  239.         s    += abs(pix1[14] - pix2[14]);

  240.         s    += abs(pix1[15] - pix2[15]);

  241.         pix1 += line_size;

  242.         pix2 += line_size;

  243.     }

  244.     return s;

  245. }

  246. 

  247. static int pix_abs16_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  248.                           int line_size, int h)

  249. {

  250.     int s = 0, i;

  251. 

  252.     for (i = 0; i < h; i++) {

  253.         s    += abs(pix1[0]  - avg2(pix2[0],  pix2[1]));

  254.         s    += abs(pix1[1]  - avg2(pix2[1],  pix2[2]));

  255.         s    += abs(pix1[2]  - avg2(pix2[2],  pix2[3]));

  256.         s    += abs(pix1[3]  - avg2(pix2[3],  pix2[4]));

  257.         s    += abs(pix1[4]  - avg2(pix2[4],  pix2[5]));

  258.         s    += abs(pix1[5]  - avg2(pix2[5],  pix2[6]));

  259.         s    += abs(pix1[6]  - avg2(pix2[6],  pix2[7]));

  260.         s    += abs(pix1[7]  - avg2(pix2[7],  pix2[8]));

  261.         s    += abs(pix1[8]  - avg2(pix2[8],  pix2[9]));

  262.         s    += abs(pix1[9]  - avg2(pix2[9],  pix2[10]));

  263.         s    += abs(pix1[10] - avg2(pix2[10], pix2[11]));

  264.         s    += abs(pix1[11] - avg2(pix2[11], pix2[12]));

  265.         s    += abs(pix1[12] - avg2(pix2[12], pix2[13]));

  266.         s    += abs(pix1[13] - avg2(pix2[13], pix2[14]));

  267.         s    += abs(pix1[14] - avg2(pix2[14], pix2[15]));

  268.         s    += abs(pix1[15] - avg2(pix2[15], pix2[16]));

  269.         pix1 += line_size;

  270.         pix2 += line_size;

  271.     }

  272.     return s;

  273. }

  274. 

  275. static int pix_abs16_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  276.                           int line_size, int h)

  277. {

  278.     int s = 0, i;

  279.     uint8_t *pix3 = pix2 + line_size;

  280. 

  281.     for (i = 0; i < h; i++) {

  282.         s    += abs(pix1[0]  - avg2(pix2[0],  pix3[0]));

  283.         s    += abs(pix1[1]  - avg2(pix2[1],  pix3[1]));

  284.         s    += abs(pix1[2]  - avg2(pix2[2],  pix3[2]));

  285.         s    += abs(pix1[3]  - avg2(pix2[3],  pix3[3]));

  286.         s    += abs(pix1[4]  - avg2(pix2[4],  pix3[4]));

  287.         s    += abs(pix1[5]  - avg2(pix2[5],  pix3[5]));

  288.         s    += abs(pix1[6]  - avg2(pix2[6],  pix3[6]));

  289.         s    += abs(pix1[7]  - avg2(pix2[7],  pix3[7]));

  290.         s    += abs(pix1[8]  - avg2(pix2[8],  pix3[8]));

  291.         s    += abs(pix1[9]  - avg2(pix2[9],  pix3[9]));

  292.         s    += abs(pix1[10] - avg2(pix2[10], pix3[10]));

  293.         s    += abs(pix1[11] - avg2(pix2[11], pix3[11]));

  294.         s    += abs(pix1[12] - avg2(pix2[12], pix3[12]));

  295.         s    += abs(pix1[13] - avg2(pix2[13], pix3[13]));

  296.         s    += abs(pix1[14] - avg2(pix2[14], pix3[14]));

  297.         s    += abs(pix1[15] - avg2(pix2[15], pix3[15]));

  298.         pix1 += line_size;

  299.         pix2 += line_size;

  300.         pix3 += line_size;

  301.     }

  302.     return s;

  303. }

  304. 

  305. static int pix_abs16_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  306.                            int line_size, int h)

  307. {

  308.     int s = 0, i;

  309.     uint8_t *pix3 = pix2 + line_size;

  310. 

  311.     for (i = 0; i < h; i++) {

  312.         s    += abs(pix1[0]  - avg4(pix2[0],  pix2[1],  pix3[0],  pix3[1]));

  313.         s    += abs(pix1[1]  - avg4(pix2[1],  pix2[2],  pix3[1],  pix3[2]));

  314.         s    += abs(pix1[2]  - avg4(pix2[2],  pix2[3],  pix3[2],  pix3[3]));

  315.         s    += abs(pix1[3]  - avg4(pix2[3],  pix2[4],  pix3[3],  pix3[4]));

  316.         s    += abs(pix1[4]  - avg4(pix2[4],  pix2[5],  pix3[4],  pix3[5]));

  317.         s    += abs(pix1[5]  - avg4(pix2[5],  pix2[6],  pix3[5],  pix3[6]));

  318.         s    += abs(pix1[6]  - avg4(pix2[6],  pix2[7],  pix3[6],  pix3[7]));

  319.         s    += abs(pix1[7]  - avg4(pix2[7],  pix2[8],  pix3[7],  pix3[8]));

  320.         s    += abs(pix1[8]  - avg4(pix2[8],  pix2[9],  pix3[8],  pix3[9]));

  321.         s    += abs(pix1[9]  - avg4(pix2[9],  pix2[10], pix3[9],  pix3[10]));

  322.         s    += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));

  323.         s    += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));

  324.         s    += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));

  325.         s    += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));

  326.         s    += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));

  327.         s    += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));

  328.         pix1 += line_size;

  329.         pix2 += line_size;

  330.         pix3 += line_size;

  331.     }

  332.     return s;

  333. }

  334. 

  335. static inline int pix_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  336.                              int line_size, int h)

  337. {

  338.     int s = 0, i;

  339. 

  340.     for (i = 0; i < h; i++) {

  341.         s    += abs(pix1[0] - pix2[0]);

  342.         s    += abs(pix1[1] - pix2[1]);

  343.         s    += abs(pix1[2] - pix2[2]);

  344.         s    += abs(pix1[3] - pix2[3]);

  345.         s    += abs(pix1[4] - pix2[4]);

  346.         s    += abs(pix1[5] - pix2[5]);

  347.         s    += abs(pix1[6] - pix2[6]);

  348.         s    += abs(pix1[7] - pix2[7]);

  349.         pix1 += line_size;

  350.         pix2 += line_size;

  351.     }

  352.     return s;

  353. }

  354. 

  355. static int pix_abs8_x2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  356.                          int line_size, int h)

  357. {

  358.     int s = 0, i;

  359. 

  360.     for (i = 0; i < h; i++) {

  361.         s    += abs(pix1[0] - avg2(pix2[0], pix2[1]));

  362.         s    += abs(pix1[1] - avg2(pix2[1], pix2[2]));

  363.         s    += abs(pix1[2] - avg2(pix2[2], pix2[3]));

  364.         s    += abs(pix1[3] - avg2(pix2[3], pix2[4]));

  365.         s    += abs(pix1[4] - avg2(pix2[4], pix2[5]));

  366.         s    += abs(pix1[5] - avg2(pix2[5], pix2[6]));

  367.         s    += abs(pix1[6] - avg2(pix2[6], pix2[7]));

  368.         s    += abs(pix1[7] - avg2(pix2[7], pix2[8]));

  369.         pix1 += line_size;

  370.         pix2 += line_size;

  371.     }

  372.     return s;

  373. }

  374. 

  375. static int pix_abs8_y2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  376.                          int line_size, int h)

  377. {

  378.     int s = 0, i;

  379.     uint8_t *pix3 = pix2 + line_size;

  380. 

  381.     for (i = 0; i < h; i++) {

  382.         s    += abs(pix1[0] - avg2(pix2[0], pix3[0]));

  383.         s    += abs(pix1[1] - avg2(pix2[1], pix3[1]));

  384.         s    += abs(pix1[2] - avg2(pix2[2], pix3[2]));

  385.         s    += abs(pix1[3] - avg2(pix2[3], pix3[3]));

  386.         s    += abs(pix1[4] - avg2(pix2[4], pix3[4]));

  387.         s    += abs(pix1[5] - avg2(pix2[5], pix3[5]));

  388.         s    += abs(pix1[6] - avg2(pix2[6], pix3[6]));

  389.         s    += abs(pix1[7] - avg2(pix2[7], pix3[7]));

  390.         pix1 += line_size;

  391.         pix2 += line_size;

  392.         pix3 += line_size;

  393.     }

  394.     return s;

  395. }

  396. 

  397. static int pix_abs8_xy2_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  398.                           int line_size, int h)

  399. {

  400.     int s = 0, i;

  401.     uint8_t *pix3 = pix2 + line_size;

  402. 

  403.     for (i = 0; i < h; i++) {

  404.         s    += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));

  405.         s    += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));

  406.         s    += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));

  407.         s    += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));

  408.         s    += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));

  409.         s    += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));

  410.         s    += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));

  411.         s    += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));

  412.         pix1 += line_size;

  413.         pix2 += line_size;

  414.         pix3 += line_size;

  415.     }

  416.     return s;

  417. }

  418. 

  419. static int nsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)

  420. {

  421.     int score1 = 0, score2 = 0, x, y;

  422. 

  423.     for (y = 0; y < h; y++) {

  424.         for (x = 0; x < 16; x++)

  425.             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);

  426.         if (y + 1 < h) {

  427.             for (x = 0; x < 15; x++)

  428.                 score2 += FFABS(s1[x]     - s1[x + stride] -

  429.                                 s1[x + 1] + s1[x + stride + 1]) -

  430.                           FFABS(s2[x]     - s2[x + stride] -

  431.                                 s2[x + 1] + s2[x + stride + 1]);

  432.         }

  433.         s1 += stride;

  434.         s2 += stride;

  435.     }

  436. 

  437.     if (c)

  438.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  439.     else

  440.         return score1 + FFABS(score2) * 8;

  441. }

  442. 

  443. static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2, int stride, int h)

  444. {

  445.     int score1 = 0, score2 = 0, x, y;

  446. 

  447.     for (y = 0; y < h; y++) {

  448.         for (x = 0; x < 8; x++)

  449.             score1 += (s1[x] - s2[x]) * (s1[x] - s2[x]);

  450.         if (y + 1 < h) {

  451.             for (x = 0; x < 7; x++)

  452.                 score2 += FFABS(s1[x]     - s1[x + stride] -

  453.                                 s1[x + 1] + s1[x + stride + 1]) -

  454.                           FFABS(s2[x]     - s2[x + stride] -

  455.                                 s2[x + 1] + s2[x + stride + 1]);

  456.         }

  457.         s1 += stride;

  458.         s2 += stride;

  459.     }

  460. 

  461.     if (c)

  462.         return score1 + FFABS(score2) * c->avctx->nsse_weight;

  463.     else

  464.         return score1 + FFABS(score2) * 8;

  465. }

  466. 

  467. static int try_8x8basis_c(int16_t rem[64], int16_t weight[64],

  468.                           int16_t basis[64], int scale)

  469. {

  470.     int i;

  471.     unsigned int sum = 0;

  472. 

  473.     for (i = 0; i < 8 * 8; i++) {

  474.         int b = rem[i] + ((basis[i] * scale +

  475.                            (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  476.                           (BASIS_SHIFT - RECON_SHIFT));

  477.         int w = weight[i];

  478.         b >>= RECON_SHIFT;

  479.         assert(-512 < b && b < 512);

  480. 

  481.         sum += (w * b) * (w * b) >> 4;

  482.     }

  483.     return sum >> 2;

  484. }

  485. 

  486. static void add_8x8basis_c(int16_t rem[64], int16_t basis[64], int scale)

  487. {

  488.     int i;

  489. 

  490.     for (i = 0; i < 8 * 8; i++)

  491.         rem[i] += (basis[i] * scale +

  492.                    (1 << (BASIS_SHIFT - RECON_SHIFT - 1))) >>

  493.                   (BASIS_SHIFT - RECON_SHIFT);

  494. }

  495. 

  496. static int zero_cmp(MpegEncContext *s, uint8_t *a, uint8_t *b,

  497.                     int stride, int h)

  498. {

  499.     return 0;

  500. }

  501. 

  502. void ff_set_cmp(DSPContext *c, me_cmp_func *cmp, int type)

  503. {

  504.     int i;

  505. 

  506.     memset(cmp, 0, sizeof(void *) * 6);

  507. 

  508.     for (i = 0; i < 6; i++) {

  509.         switch (type & 0xFF) {

  510.         case FF_CMP_SAD:

  511.             cmp[i] = c->sad[i];

  512.             break;

  513.         case FF_CMP_SATD:

  514.             cmp[i] = c->hadamard8_diff[i];

  515.             break;

  516.         case FF_CMP_SSE:

  517.             cmp[i] = c->sse[i];

  518.             break;

  519.         case FF_CMP_DCT:

  520.             cmp[i] = c->dct_sad[i];

  521.             break;

  522.         case FF_CMP_DCT264:

  523.             cmp[i] = c->dct264_sad[i];

  524.             break;

  525.         case FF_CMP_DCTMAX:

  526.             cmp[i] = c->dct_max[i];

  527.             break;

  528.         case FF_CMP_PSNR:

  529.             cmp[i] = c->quant_psnr[i];

  530.             break;

  531.         case FF_CMP_BIT:

  532.             cmp[i] = c->bit[i];

  533.             break;

  534.         case FF_CMP_RD:

  535.             cmp[i] = c->rd[i];

  536.             break;

  537.         case FF_CMP_VSAD:

  538.             cmp[i] = c->vsad[i];

  539.             break;

  540.         case FF_CMP_VSSE:

  541.             cmp[i] = c->vsse[i];

  542.             break;

  543.         case FF_CMP_ZERO:

  544.             cmp[i] = zero_cmp;

  545.             break;

  546.         case FF_CMP_NSSE:

  547.             cmp[i] = c->nsse[i];

  548.             break;

  549.         default:

  550.             av_log(NULL, AV_LOG_ERROR,

  551.                    "internal error in cmp function selection\n");

  552.         }

  553.     }

  554. }

  555. 

  556. #define BUTTERFLY2(o1, o2, i1, i2)              \

  557.     o1 = (i1) + (i2);                           \

  558.     o2 = (i1) - (i2);

  559. 

  560. #define BUTTERFLY1(x, y)                        \

  561.     {                                           \

  562.         int a, b;                               \

  563.         a = x;                                  \

  564.         b = y;                                  \

  565.         x = a + b;                              \

  566.         y = a - b;                              \

  567.     }

  568. 

  569. #define BUTTERFLYA(x, y) (FFABS((x) + (y)) + FFABS((x) - (y)))

  570. 

  571. static int hadamard8_diff8x8_c(MpegEncContext *s, uint8_t *dst,

  572.                                uint8_t *src, int stride, int h)

  573. {

  574.     int i, temp[64], sum = 0;

  575. 

  576.     assert(h == 8);

  577. 

  578.     for (i = 0; i < 8; i++) {

  579.         // FIXME: try pointer walks

  580.         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],

  581.                    src[stride * i + 0] - dst[stride * i + 0],

  582.                    src[stride * i + 1] - dst[stride * i + 1]);

  583.         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],

  584.                    src[stride * i + 2] - dst[stride * i + 2],

  585.                    src[stride * i + 3] - dst[stride * i + 3]);

  586.         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],

  587.                    src[stride * i + 4] - dst[stride * i + 4],

  588.                    src[stride * i + 5] - dst[stride * i + 5]);

  589.         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],

  590.                    src[stride * i + 6] - dst[stride * i + 6],

  591.                    src[stride * i + 7] - dst[stride * i + 7]);

  592. 

  593.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);

  594.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);

  595.         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);

  596.         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);

  597. 

  598.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);

  599.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);

  600.         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);

  601.         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);

  602.     }

  603. 

  604.     for (i = 0; i < 8; i++) {

  605.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);

  606.         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);

  607.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);

  608.         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);

  609. 

  610.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);

  611.         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);

  612.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);

  613.         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);

  614. 

  615.         sum += BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i]) +

  616.                BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i]) +

  617.                BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i]) +

  618.                BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);

  619.     }

  620.     return sum;

  621. }

  622. 

  623. static int hadamard8_intra8x8_c(MpegEncContext *s, uint8_t *src,

  624.                                 uint8_t *dummy, int stride, int h)

  625. {

  626.     int i, temp[64], sum = 0;

  627. 

  628.     assert(h == 8);

  629. 

  630.     for (i = 0; i < 8; i++) {

  631.         // FIXME: try pointer walks

  632.         BUTTERFLY2(temp[8 * i + 0], temp[8 * i + 1],

  633.                    src[stride * i + 0], src[stride * i + 1]);

  634.         BUTTERFLY2(temp[8 * i + 2], temp[8 * i + 3],

  635.                    src[stride * i + 2], src[stride * i + 3]);

  636.         BUTTERFLY2(temp[8 * i + 4], temp[8 * i + 5],

  637.                    src[stride * i + 4], src[stride * i + 5]);

  638.         BUTTERFLY2(temp[8 * i + 6], temp[8 * i + 7],

  639.                    src[stride * i + 6], src[stride * i + 7]);

  640. 

  641.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 2]);

  642.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 3]);

  643.         BUTTERFLY1(temp[8 * i + 4], temp[8 * i + 6]);

  644.         BUTTERFLY1(temp[8 * i + 5], temp[8 * i + 7]);

  645. 

  646.         BUTTERFLY1(temp[8 * i + 0], temp[8 * i + 4]);

  647.         BUTTERFLY1(temp[8 * i + 1], temp[8 * i + 5]);

  648.         BUTTERFLY1(temp[8 * i + 2], temp[8 * i + 6]);

  649.         BUTTERFLY1(temp[8 * i + 3], temp[8 * i + 7]);

  650.     }

  651. 

  652.     for (i = 0; i < 8; i++) {

  653.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 1 + i]);

  654.         BUTTERFLY1(temp[8 * 2 + i], temp[8 * 3 + i]);

  655.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 5 + i]);

  656.         BUTTERFLY1(temp[8 * 6 + i], temp[8 * 7 + i]);

  657. 

  658.         BUTTERFLY1(temp[8 * 0 + i], temp[8 * 2 + i]);

  659.         BUTTERFLY1(temp[8 * 1 + i], temp[8 * 3 + i]);

  660.         BUTTERFLY1(temp[8 * 4 + i], temp[8 * 6 + i]);

  661.         BUTTERFLY1(temp[8 * 5 + i], temp[8 * 7 + i]);

  662. 

  663.         sum +=

  664.             BUTTERFLYA(temp[8 * 0 + i], temp[8 * 4 + i])

  665.             + BUTTERFLYA(temp[8 * 1 + i], temp[8 * 5 + i])

  666.             + BUTTERFLYA(temp[8 * 2 + i], temp[8 * 6 + i])

  667.             + BUTTERFLYA(temp[8 * 3 + i], temp[8 * 7 + i]);

  668.     }

  669. 

  670.     sum -= FFABS(temp[8 * 0] + temp[8 * 4]); // -mean

  671. 

  672.     return sum;

  673. }

  674. 

  675. static int dct_sad8x8_c(MpegEncContext *s, uint8_t *src1,

  676.                         uint8_t *src2, int stride, int h)

  677. {

  678.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  679. 

  680.     assert(h == 8);

  681. 

  682.     s->dsp.diff_pixels(temp, src1, src2, stride);

  683.     s->dsp.fdct(temp);

  684.     return s->dsp.sum_abs_dctelem(temp);

  685. }

  686. 

  687. #if CONFIG_GPL

  688. #define DCT8_1D                                         \

  689.     {                                                   \

  690.         const int s07 = SRC(0) + SRC(7);                \

  691.         const int s16 = SRC(1) + SRC(6);                \

  692.         const int s25 = SRC(2) + SRC(5);                \

  693.         const int s34 = SRC(3) + SRC(4);                \

  694.         const int a0  = s07 + s34;                      \

  695.         const int a1  = s16 + s25;                      \

  696.         const int a2  = s07 - s34;                      \

  697.         const int a3  = s16 - s25;                      \

  698.         const int d07 = SRC(0) - SRC(7);                \

  699.         const int d16 = SRC(1) - SRC(6);                \

  700.         const int d25 = SRC(2) - SRC(5);                \

  701.         const int d34 = SRC(3) - SRC(4);                \

  702.         const int a4  = d16 + d25 + (d07 + (d07 >> 1)); \

  703.         const int a5  = d07 - d34 - (d25 + (d25 >> 1)); \

  704.         const int a6  = d07 + d34 - (d16 + (d16 >> 1)); \

  705.         const int a7  = d16 - d25 + (d34 + (d34 >> 1)); \

  706.         DST(0, a0 + a1);                                \

  707.         DST(1, a4 + (a7 >> 2));                         \

  708.         DST(2, a2 + (a3 >> 1));                         \

  709.         DST(3, a5 + (a6 >> 2));                         \

  710.         DST(4, a0 - a1);                                \

  711.         DST(5, a6 - (a5 >> 2));                         \

  712.         DST(6, (a2 >> 1) - a3);                         \

  713.         DST(7, (a4 >> 2) - a7);                         \

  714.     }

  715. 

  716. static int dct264_sad8x8_c(MpegEncContext *s, uint8_t *src1,

  717.                            uint8_t *src2, int stride, int h)

  718. {

  719.     int16_t dct[8][8];

  720.     int i, sum = 0;

  721. 

  722.     s->dsp.diff_pixels(dct[0], src1, src2, stride);

  723. 

  724. #define SRC(x) dct[i][x]

  725. #define DST(x, v) dct[i][x] = v

  726.     for (i = 0; i < 8; i++)

  727.         DCT8_1D

  728. #undef SRC

  729. #undef DST

  730. 

  731. #define SRC(x) dct[x][i]

  732. #define DST(x, v) sum += FFABS(v)

  733.         for (i = 0; i < 8; i++)

  734.             DCT8_1D

  735. #undef SRC

  736. #undef DST

  737.             return sum;

  738. }

  739. #endif

  740. 

  741. static int dct_max8x8_c(MpegEncContext *s, uint8_t *src1,

  742.                         uint8_t *src2, int stride, int h)

  743. {

  744.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  745.     int sum = 0, i;

  746. 

  747.     assert(h == 8);

  748. 

  749.     s->dsp.diff_pixels(temp, src1, src2, stride);

  750.     s->dsp.fdct(temp);

  751. 

  752.     for (i = 0; i < 64; i++)

  753.         sum = FFMAX(sum, FFABS(temp[i]));

  754. 

  755.     return sum;

  756. }

  757. 

  758. static int quant_psnr8x8_c(MpegEncContext *s, uint8_t *src1,

  759.                            uint8_t *src2, int stride, int h)

  760. {

  761.     LOCAL_ALIGNED_16(int16_t, temp, [64 * 2]);

  762.     int16_t *const bak = temp + 64;

  763.     int sum = 0, i;

  764. 

  765.     assert(h == 8);

  766.     s->mb_intra = 0;

  767. 

  768.     s->dsp.diff_pixels(temp, src1, src2, stride);

  769. 

  770.     memcpy(bak, temp, 64 * sizeof(int16_t));

  771. 

  772.     s->block_last_index[0 /* FIXME */] =

  773.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  774.     s->dct_unquantize_inter(s, temp, 0, s->qscale);

  775.     ff_simple_idct_8(temp); // FIXME

  776. 

  777.     for (i = 0; i < 64; i++)

  778.         sum += (temp[i] - bak[i]) * (temp[i] - bak[i]);

  779. 

  780.     return sum;

  781. }

  782. 

  783. static int rd8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  784.                    int stride, int h)

  785. {

  786.     const uint8_t *scantable = s->intra_scantable.permutated;

  787.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  788.     LOCAL_ALIGNED_16(uint8_t, lsrc1, [64]);

  789.     LOCAL_ALIGNED_16(uint8_t, lsrc2, [64]);

  790.     int i, last, run, bits, level, distortion, start_i;

  791.     const int esc_length = s->ac_esc_length;

  792.     uint8_t *length, *last_length;

  793. 

  794.     assert(h == 8);

  795. 

  796.     copy_block8(lsrc1, src1, 8, stride, 8);

  797.     copy_block8(lsrc2, src2, 8, stride, 8);

  798. 

  799.     s->dsp.diff_pixels(temp, lsrc1, lsrc2, 8);

  800. 

  801.     s->block_last_index[0 /* FIXME */] =

  802.     last                               =

  803.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  804. 

  805.     bits = 0;

  806. 

  807.     if (s->mb_intra) {

  808.         start_i     = 1;

  809.         length      = s->intra_ac_vlc_length;

  810.         last_length = s->intra_ac_vlc_last_length;

  811.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  812.     } else {

  813.         start_i     = 0;

  814.         length      = s->inter_ac_vlc_length;

  815.         last_length = s->inter_ac_vlc_last_length;

  816.     }

  817. 

  818.     if (last >= start_i) {

  819.         run = 0;

  820.         for (i = start_i; i < last; i++) {

  821.             int j = scantable[i];

  822.             level = temp[j];

  823. 

  824.             if (level) {

  825.                 level += 64;

  826.                 if ((level & (~127)) == 0)

  827.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  828.                 else

  829.                     bits += esc_length;

  830.                 run = 0;

  831.             } else

  832.                 run++;

  833.         }

  834.         i = scantable[last];

  835. 

  836.         level = temp[i] + 64;

  837. 

  838.         assert(level - 64);

  839. 

  840.         if ((level & (~127)) == 0) {

  841.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  842.         } else

  843.             bits += esc_length;

  844.     }

  845. 

  846.     if (last >= 0) {

  847.         if (s->mb_intra)

  848.             s->dct_unquantize_intra(s, temp, 0, s->qscale);

  849.         else

  850.             s->dct_unquantize_inter(s, temp, 0, s->qscale);

  851.     }

  852. 

  853.     s->idsp.idct_add(lsrc2, 8, temp);

  854. 

  855.     distortion = s->dsp.sse[1](NULL, lsrc2, lsrc1, 8, 8);

  856. 

  857.     return distortion + ((bits * s->qscale * s->qscale * 109 + 64) >> 7);

  858. }

  859. 

  860. static int bit8x8_c(MpegEncContext *s, uint8_t *src1, uint8_t *src2,

  861.                     int stride, int h)

  862. {

  863.     const uint8_t *scantable = s->intra_scantable.permutated;

  864.     LOCAL_ALIGNED_16(int16_t, temp, [64]);

  865.     int i, last, run, bits, level, start_i;

  866.     const int esc_length = s->ac_esc_length;

  867.     uint8_t *length, *last_length;

  868. 

  869.     assert(h == 8);

  870. 

  871.     s->dsp.diff_pixels(temp, src1, src2, stride);

  872. 

  873.     s->block_last_index[0 /* FIXME */] =

  874.     last                               =

  875.         s->fast_dct_quantize(s, temp, 0 /* FIXME */, s->qscale, &i);

  876. 

  877.     bits = 0;

  878. 

  879.     if (s->mb_intra) {

  880.         start_i     = 1;

  881.         length      = s->intra_ac_vlc_length;

  882.         last_length = s->intra_ac_vlc_last_length;

  883.         bits       += s->luma_dc_vlc_length[temp[0] + 256]; // FIXME: chroma

  884.     } else {

  885.         start_i     = 0;

  886.         length      = s->inter_ac_vlc_length;

  887.         last_length = s->inter_ac_vlc_last_length;

  888.     }

  889. 

  890.     if (last >= start_i) {

  891.         run = 0;

  892.         for (i = start_i; i < last; i++) {

  893.             int j = scantable[i];

  894.             level = temp[j];

  895. 

  896.             if (level) {

  897.                 level += 64;

  898.                 if ((level & (~127)) == 0)

  899.                     bits += length[UNI_AC_ENC_INDEX(run, level)];

  900.                 else

  901.                     bits += esc_length;

  902.                 run = 0;

  903.             } else

  904.                 run++;

  905.         }

  906.         i = scantable[last];

  907. 

  908.         level = temp[i] + 64;

  909. 

  910.         assert(level - 64);

  911. 

  912.         if ((level & (~127)) == 0)

  913.             bits += last_length[UNI_AC_ENC_INDEX(run, level)];

  914.         else

  915.             bits += esc_length;

  916.     }

  917. 

  918.     return bits;

  919. }

  920. 

  921. #define VSAD_INTRA(size)                                                \

  922. static int vsad_intra ## size ## _c(MpegEncContext *c,                  \

  923.                                     uint8_t *s, uint8_t *dummy,         \

  924.                                     int stride, int h)                  \

  925. {                                                                       \

  926.     int score = 0, x, y;                                                \

  927.                                                                         \

  928.     for (y = 1; y < h; y++) {                                           \

  929.         for (x = 0; x < size; x += 4) {                                 \

  930.             score += FFABS(s[x]     - s[x + stride])     +              \

  931.                      FFABS(s[x + 1] - s[x + stride + 1]) +              \

  932.                      FFABS(s[x + 2] - s[x + 2 + stride]) +              \

  933.                      FFABS(s[x + 3] - s[x + 3 + stride]);               \

  934.         }                                                               \

  935.         s += stride;                                                    \

  936.     }                                                                   \

  937.                                                                         \

  938.     return score;                                                       \

  939. }

  940. VSAD_INTRA(8)

  941. VSAD_INTRA(16)

  942. 

  943. static int vsad16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  944.                     int stride, int h)

  945. {

  946.     int score = 0, x, y;

  947. 

  948.     for (y = 1; y < h; y++) {

  949.         for (x = 0; x < 16; x++)

  950.             score += FFABS(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  951.         s1 += stride;

  952.         s2 += stride;

  953.     }

  954. 

  955.     return score;

  956. }

  957. 

  958. #define SQ(a) ((a) * (a))

  959. #define VSSE_INTRA(size)                                                \

  960. static int vsse_intra ## size ## _c(MpegEncContext *c,                  \

  961.                                     uint8_t *s, uint8_t *dummy,         \

  962.                                     int stride, int h)                  \

  963. {                                                                       \

  964.     int score = 0, x, y;                                                \

  965.                                                                         \

  966.     for (y = 1; y < h; y++) {                                           \

  967.         for (x = 0; x < size; x += 4) {                                 \

  968.             score += SQ(s[x]     - s[x + stride]) +                     \

  969.                      SQ(s[x + 1] - s[x + stride + 1]) +                 \

  970.                      SQ(s[x + 2] - s[x + stride + 2]) +                 \

  971.                      SQ(s[x + 3] - s[x + stride + 3]);                  \

  972.         }                                                               \

  973.         s += stride;                                                    \

  974.     }                                                                   \

  975.                                                                         \

  976.     return score;                                                       \

  977. }

  978. VSSE_INTRA(8)

  979. VSSE_INTRA(16)

  980. 

  981. static int vsse16_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  982.                     int stride, int h)

  983. {

  984.     int score = 0, x, y;

  985. 

  986.     for (y = 1; y < h; y++) {

  987.         for (x = 0; x < 16; x++)

  988.             score += SQ(s1[x] - s2[x] - s1[x + stride] + s2[x + stride]);

  989.         s1 += stride;

  990.         s2 += stride;

  991.     }

  992. 

  993.     return score;

  994. }

  995. 

  996. #define WRAPPER8_16_SQ(name8, name16)                                   \

  997. static int name16(MpegEncContext *s, uint8_t *dst, uint8_t *src,        \

  998.                   int stride, int h)                                    \

  999. {                                                                       \

  1000.     int score = 0;                                                      \

  1001.                                                                         \

  1002.     score += name8(s, dst, src, stride, 8);                             \

  1003.     score += name8(s, dst + 8, src + 8, stride, 8);                     \

  1004.     if (h == 16) {                                                      \

  1005.         dst   += 8 * stride;                                            \

  1006.         src   += 8 * stride;                                            \

  1007.         score += name8(s, dst, src, stride, 8);                         \

  1008.         score += name8(s, dst + 8, src + 8, stride, 8);                 \

  1009.     }                                                                   \

  1010.     return score;                                                       \

  1011. }

  1012. 

  1013. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  1014. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  1015. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  1016. #if CONFIG_GPL

  1017. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  1018. #endif

  1019. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  1020. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  1021. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  1022. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  1023. 

  1024. /* draw the edges of width 'w' of an image of size width, height */

  1025. // FIXME: Check that this is OK for MPEG-4 interlaced.

  1026. static void draw_edges_8_c(uint8_t *buf, int wrap, int width, int height,

  1027.                            int w, int h, int sides)

  1028. {

  1029.     uint8_t *ptr = buf, *last_line;

  1030.     int i;

  1031. 

  1032.     /* left and right */

  1033.     for (i = 0; i < height; i++) {

  1034.         memset(ptr - w, ptr[0], w);

  1035.         memset(ptr + width, ptr[width - 1], w);

  1036.         ptr += wrap;

  1037.     }

  1038. 

  1039.     /* top and bottom + corners */

  1040.     buf -= w;

  1041.     last_line = buf + (height - 1) * wrap;

  1042.     if (sides & EDGE_TOP)

  1043.         for (i = 0; i < h; i++)

  1044.             // top

  1045.             memcpy(buf - (i + 1) * wrap, buf, width + w + w);

  1046.     if (sides & EDGE_BOTTOM)

  1047.         for (i = 0; i < h; i++)

  1048.             // bottom

  1049.             memcpy(last_line + (i + 1) * wrap, last_line, width + w + w);

  1050. }

  1051. 

  1052. /* init static data */

  1053. av_cold void ff_dsputil_static_init(void)

  1054. {

  1055.     int i;

  1056. 

  1057.     for (i = 0; i < 512; i++)

  1058.         ff_square_tab[i] = (i - 256) * (i - 256);

  1059. }

  1060. 

  1061. av_cold void ff_dsputil_init(DSPContext *c, AVCodecContext *avctx)

  1062. {

  1063.     const unsigned high_bit_depth = avctx->bits_per_raw_sample > 8;

  1064. 

  1065. #if CONFIG_ENCODERS

  1066.     if (avctx->bits_per_raw_sample == 10) {

  1067.         c->fdct    = ff_jpeg_fdct_islow_10;

  1068.         c->fdct248 = ff_fdct248_islow_10;

  1069.     } else {

  1070.         if (avctx->dct_algo == FF_DCT_FASTINT) {

  1071.             c->fdct    = ff_fdct_ifast;

  1072.             c->fdct248 = ff_fdct_ifast248;

  1073.         } else if (avctx->dct_algo == FF_DCT_FAAN) {

  1074.             c->fdct    = ff_faandct;

  1075.             c->fdct248 = ff_faandct248;

  1076.         } else {

  1077.             c->fdct    = ff_jpeg_fdct_islow_8; // slow/accurate/default

  1078.             c->fdct248 = ff_fdct248_islow_8;

  1079.         }

  1080.     }

  1081. #endif /* CONFIG_ENCODERS */

  1082. 

  1083.     c->diff_pixels = diff_pixels_c;

  1084. 

  1085.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  1086. 

  1087.     c->pix_sum   = pix_sum_c;

  1088.     c->pix_norm1 = pix_norm1_c;

  1089. 

  1090.     /* TODO [0] 16  [1] 8 */

  1091.     c->pix_abs[0][0] = pix_abs16_c;

  1092.     c->pix_abs[0][1] = pix_abs16_x2_c;

  1093.     c->pix_abs[0][2] = pix_abs16_y2_c;

  1094.     c->pix_abs[0][3] = pix_abs16_xy2_c;

  1095.     c->pix_abs[1][0] = pix_abs8_c;

  1096.     c->pix_abs[1][1] = pix_abs8_x2_c;

  1097.     c->pix_abs[1][2] = pix_abs8_y2_c;

  1098.     c->pix_abs[1][3] = pix_abs8_xy2_c;

  1099. 

  1100. #define SET_CMP_FUNC(name)                      \

  1101.     c->name[0] = name ## 16_c;                  \

  1102.     c->name[1] = name ## 8x8_c;

  1103. 

  1104.     SET_CMP_FUNC(hadamard8_diff)

  1105.     c->hadamard8_diff[4] = hadamard8_intra16_c;

  1106.     c->hadamard8_diff[5] = hadamard8_intra8x8_c;

  1107.     SET_CMP_FUNC(dct_sad)

  1108.     SET_CMP_FUNC(dct_max)

  1109. #if CONFIG_GPL

  1110.     SET_CMP_FUNC(dct264_sad)

  1111. #endif

  1112.     c->sad[0] = pix_abs16_c;

  1113.     c->sad[1] = pix_abs8_c;

  1114.     c->sse[0] = sse16_c;

  1115.     c->sse[1] = sse8_c;

  1116.     c->sse[2] = sse4_c;

  1117.     SET_CMP_FUNC(quant_psnr)

  1118.     SET_CMP_FUNC(rd)

  1119.     SET_CMP_FUNC(bit)

  1120.     c->vsad[0] = vsad16_c;

  1121.     c->vsad[4] = vsad_intra16_c;

  1122.     c->vsad[5] = vsad_intra8_c;

  1123.     c->vsse[0] = vsse16_c;

  1124.     c->vsse[4] = vsse_intra16_c;

  1125.     c->vsse[5] = vsse_intra8_c;

  1126.     c->nsse[0] = nsse16_c;

  1127.     c->nsse[1] = nsse8_c;

  1128. 

  1129.     c->try_8x8basis = try_8x8basis_c;

  1130.     c->add_8x8basis = add_8x8basis_c;

  1131. 

  1132.     c->shrink[0] = av_image_copy_plane;

  1133.     c->shrink[1] = ff_shrink22;

  1134.     c->shrink[2] = ff_shrink44;

  1135.     c->shrink[3] = ff_shrink88;

  1136. 

  1137.     c->draw_edges = draw_edges_8_c;

  1138. 

  1139.     switch (avctx->bits_per_raw_sample) {

  1140.     case 9:

  1141.     case 10:

  1142.         c->get_pixels = get_pixels_16_c;

  1143.         break;

  1144.     default:

  1145.         c->get_pixels = get_pixels_8_c;

  1146.         break;

  1147.     }

  1148. 

  1149.     if (ARCH_ARM)

  1150.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  1151.     if (ARCH_PPC)

  1152.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  1153.     if (ARCH_X86)

  1154.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  1155. }