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.
40175 Commits
32 Branches
570MB
Tree: a9aee08d90
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a9aee08d90'
${ noResults }
FFmpeg/libavcodec/dsputil.c

995 lines
33KB
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 "avcodec.h"

  30. #include "copy_block.h"

  31. #include "dsputil.h"

  32. #include "simple_idct.h"

  33. #include "mpegvideo.h"

  34. #include "config.h"

  35. 

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

  37. 

  38. #define BIT_DEPTH 16

  39. #include "dsputilenc_template.c"

  40. #undef BIT_DEPTH

  41. 

  42. #define BIT_DEPTH 8

  43. #include "dsputilenc_template.c"

  44. 

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

  46.                   int line_size, int h)

  47. {

  48.     int s = 0, i;

  49.     uint32_t *sq = ff_square_tab + 256;

  50. 

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

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

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

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

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

  56.         pix1 += line_size;

  57.         pix2 += line_size;

  58.     }

  59.     return s;

  60. }

  61. 

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

  63.                   int line_size, int h)

  64. {

  65.     int s = 0, i;

  66.     uint32_t *sq = ff_square_tab + 256;

  67. 

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

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

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

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

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

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

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

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

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

  77.         pix1 += line_size;

  78.         pix2 += line_size;

  79.     }

  80.     return s;

  81. }

  82. 

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

  84.                    int line_size, int h)

  85. {

  86.     int s = 0, i;

  87.     uint32_t *sq = ff_square_tab + 256;

  88. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  106. 

  107.         pix1 += line_size;

  108.         pix2 += line_size;

  109.     }

  110.     return s;

  111. }

  112. 

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

  114.                           const uint8_t *s2, int stride)

  115. {

  116.     int i;

  117. 

  118.     /* read the pixels */

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

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

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

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

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

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

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

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

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

  128.         s1      += stride;

  129.         s2      += stride;

  130.         block   += 8;

  131.     }

  132. }

  133. 

  134. static int sum_abs_dctelem_c(int16_t *block)

  135. {

  136.     int sum = 0, i;

  137. 

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

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

  140.     return sum;

  141. }

  142. 

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

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

  145. 

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

  147.                               int line_size, int h)

  148. {

  149.     int s = 0, i;

  150. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  168.         pix1 += line_size;

  169.         pix2 += line_size;

  170.     }

  171.     return s;

  172. }

  173. 

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

  175.                           int line_size, int h)

  176. {

  177.     int s = 0, i;

  178. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  196.         pix1 += line_size;

  197.         pix2 += line_size;

  198.     }

  199.     return s;

  200. }

  201. 

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

  203.                           int line_size, int h)

  204. {

  205.     int s = 0, i;

  206.     uint8_t *pix3 = pix2 + line_size;

  207. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  225.         pix1 += line_size;

  226.         pix2 += line_size;

  227.         pix3 += line_size;

  228.     }

  229.     return s;

  230. }

  231. 

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

  233.                            int line_size, int h)

  234. {

  235.     int s = 0, i;

  236.     uint8_t *pix3 = pix2 + line_size;

  237. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  255.         pix1 += line_size;

  256.         pix2 += line_size;

  257.         pix3 += line_size;

  258.     }

  259.     return s;

  260. }

  261. 

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

  263.                              int line_size, int h)

  264. {

  265.     int s = 0, i;

  266. 

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

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

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

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

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

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

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

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

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

  276.         pix1 += line_size;

  277.         pix2 += line_size;

  278.     }

  279.     return s;

  280. }

  281. 

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

  283.                          int line_size, int h)

  284. {

  285.     int s = 0, i;

  286. 

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

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

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

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

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

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

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

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

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

  296.         pix1 += line_size;

  297.         pix2 += line_size;

  298.     }

  299.     return s;

  300. }

  301. 

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

  303.                          int line_size, int h)

  304. {

  305.     int s = 0, i;

  306.     uint8_t *pix3 = pix2 + line_size;

  307. 

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

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

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

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

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

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

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

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

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

  317.         pix1 += line_size;

  318.         pix2 += line_size;

  319.         pix3 += line_size;

  320.     }

  321.     return s;

  322. }

  323. 

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

  325.                           int line_size, int h)

  326. {

  327.     int s = 0, i;

  328.     uint8_t *pix3 = pix2 + line_size;

  329. 

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

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

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

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

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

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

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

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

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

  339.         pix1 += line_size;

  340.         pix2 += line_size;

  341.         pix3 += line_size;

  342.     }

  343.     return s;

  344. }

  345. 

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

  347. {

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

  349. 

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

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

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

  353.         if (y + 1 < h) {

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

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

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

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

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

  359.         }

  360.         s1 += stride;

  361.         s2 += stride;

  362.     }

  363. 

  364.     if (c)

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

  366.     else

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

  368. }

  369. 

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

  371. {

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

  373. 

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

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

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

  377.         if (y + 1 < h) {

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

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

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

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

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

  383.         }

  384.         s1 += stride;

  385.         s2 += stride;

  386.     }

  387. 

  388.     if (c)

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

  390.     else

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

  392. }

  393. 

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

  395.                     int stride, int h)

  396. {

  397.     return 0;

  398. }

  399. 

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

  401. {

  402.     int i;

  403. 

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

  405. 

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

  407.         switch (type & 0xFF) {

  408.         case FF_CMP_SAD:

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

  410.             break;

  411.         case FF_CMP_SATD:

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

  413.             break;

  414.         case FF_CMP_SSE:

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

  416.             break;

  417.         case FF_CMP_DCT:

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

  419.             break;

  420.         case FF_CMP_DCT264:

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

  422.             break;

  423.         case FF_CMP_DCTMAX:

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

  425.             break;

  426.         case FF_CMP_PSNR:

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

  428.             break;

  429.         case FF_CMP_BIT:

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

  431.             break;

  432.         case FF_CMP_RD:

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

  434.             break;

  435.         case FF_CMP_VSAD:

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

  437.             break;

  438.         case FF_CMP_VSSE:

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

  440.             break;

  441.         case FF_CMP_ZERO:

  442.             cmp[i] = zero_cmp;

  443.             break;

  444.         case FF_CMP_NSSE:

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

  446.             break;

  447.         default:

  448.             av_log(NULL, AV_LOG_ERROR,

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

  450.         }

  451.     }

  452. }

  453. 

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

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

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

  457. 

  458. #define BUTTERFLY1(x, y)                        \

  459.     {                                           \

  460.         int a, b;                               \

  461.         a = x;                                  \

  462.         b = y;                                  \

  463.         x = a + b;                              \

  464.         y = a - b;                              \

  465.     }

  466. 

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

  468. 

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

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

  471. {

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

  473. 

  474.     assert(h == 8);

  475. 

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

  477.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  490. 

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

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

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

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

  495. 

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

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

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

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

  500.     }

  501. 

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

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

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

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

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

  507. 

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

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

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

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

  512. 

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

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

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

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

  517.     }

  518.     return sum;

  519. }

  520. 

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

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

  523. {

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

  525. 

  526.     assert(h == 8);

  527. 

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

  529.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  538. 

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

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

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

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

  543. 

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

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

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

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

  548.     }

  549. 

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

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

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

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

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

  555. 

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

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

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

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

  560. 

  561.         sum +=

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

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

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

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

  566.     }

  567. 

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

  569. 

  570.     return sum;

  571. }

  572. 

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

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

  575. {

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

  577. 

  578.     assert(h == 8);

  579. 

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

  581.     s->fdsp.fdct(temp);

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

  583. }

  584. 

  585. #if CONFIG_GPL

  586. #define DCT8_1D                                         \

  587.     {                                                   \

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

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

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

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

  592.         const int a0  = s07 + s34;                      \

  593.         const int a1  = s16 + s25;                      \

  594.         const int a2  = s07 - s34;                      \

  595.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  612.     }

  613. 

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

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

  616. {

  617.     int16_t dct[8][8];

  618.     int i, sum = 0;

  619. 

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

  621. 

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

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

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

  625.         DCT8_1D

  626. #undef SRC

  627. #undef DST

  628. 

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

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

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

  632.             DCT8_1D

  633. #undef SRC

  634. #undef DST

  635.             return sum;

  636. }

  637. #endif

  638. 

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

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

  641. {

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

  643.     int sum = 0, i;

  644. 

  645.     assert(h == 8);

  646. 

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

  648.     s->fdsp.fdct(temp);

  649. 

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

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

  652. 

  653.     return sum;

  654. }

  655. 

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

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

  658. {

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

  660.     int16_t *const bak = temp + 64;

  661.     int sum = 0, i;

  662. 

  663.     assert(h == 8);

  664.     s->mb_intra = 0;

  665. 

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

  667. 

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

  669. 

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

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

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

  673.     ff_simple_idct_8(temp); // FIXME

  674. 

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

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

  677. 

  678.     return sum;

  679. }

  680. 

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

  682.                    int stride, int h)

  683. {

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

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

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

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

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

  689.     const int esc_length = s->ac_esc_length;

  690.     uint8_t *length, *last_length;

  691. 

  692.     assert(h == 8);

  693. 

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

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

  696. 

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

  698. 

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

  700.     last                               =

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

  702. 

  703.     bits = 0;

  704. 

  705.     if (s->mb_intra) {

  706.         start_i     = 1;

  707.         length      = s->intra_ac_vlc_length;

  708.         last_length = s->intra_ac_vlc_last_length;

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

  710.     } else {

  711.         start_i     = 0;

  712.         length      = s->inter_ac_vlc_length;

  713.         last_length = s->inter_ac_vlc_last_length;

  714.     }

  715. 

  716.     if (last >= start_i) {

  717.         run = 0;

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

  719.             int j = scantable[i];

  720.             level = temp[j];

  721. 

  722.             if (level) {

  723.                 level += 64;

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

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

  726.                 else

  727.                     bits += esc_length;

  728.                 run = 0;

  729.             } else

  730.                 run++;

  731.         }

  732.         i = scantable[last];

  733. 

  734.         level = temp[i] + 64;

  735. 

  736.         assert(level - 64);

  737. 

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

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

  740.         } else

  741.             bits += esc_length;

  742.     }

  743. 

  744.     if (last >= 0) {

  745.         if (s->mb_intra)

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

  747.         else

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

  749.     }

  750. 

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

  752. 

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

  754. 

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

  756. }

  757. 

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

  759.                     int stride, int h)

  760. {

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

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

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

  764.     const int esc_length = s->ac_esc_length;

  765.     uint8_t *length, *last_length;

  766. 

  767.     assert(h == 8);

  768. 

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

  770. 

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

  772.     last                               =

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

  774. 

  775.     bits = 0;

  776. 

  777.     if (s->mb_intra) {

  778.         start_i     = 1;

  779.         length      = s->intra_ac_vlc_length;

  780.         last_length = s->intra_ac_vlc_last_length;

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

  782.     } else {

  783.         start_i     = 0;

  784.         length      = s->inter_ac_vlc_length;

  785.         last_length = s->inter_ac_vlc_last_length;

  786.     }

  787. 

  788.     if (last >= start_i) {

  789.         run = 0;

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

  791.             int j = scantable[i];

  792.             level = temp[j];

  793. 

  794.             if (level) {

  795.                 level += 64;

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

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

  798.                 else

  799.                     bits += esc_length;

  800.                 run = 0;

  801.             } else

  802.                 run++;

  803.         }

  804.         i = scantable[last];

  805. 

  806.         level = temp[i] + 64;

  807. 

  808.         assert(level - 64);

  809. 

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

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

  812.         else

  813.             bits += esc_length;

  814.     }

  815. 

  816.     return bits;

  817. }

  818. 

  819. #define VSAD_INTRA(size)                                                \

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

  821.                                     uint8_t *s, uint8_t *dummy,         \

  822.                                     int stride, int h)                  \

  823. {                                                                       \

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

  825.                                                                         \

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

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

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

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

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

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

  832.         }                                                               \

  833.         s += stride;                                                    \

  834.     }                                                                   \

  835.                                                                         \

  836.     return score;                                                       \

  837. }

  838. VSAD_INTRA(8)

  839. VSAD_INTRA(16)

  840. 

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

  842.                     int stride, int h)

  843. {

  844.     int score = 0, x, y;

  845. 

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

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

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

  849.         s1 += stride;

  850.         s2 += stride;

  851.     }

  852. 

  853.     return score;

  854. }

  855. 

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

  857. #define VSSE_INTRA(size)                                                \

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

  859.                                     uint8_t *s, uint8_t *dummy,         \

  860.                                     int stride, int h)                  \

  861. {                                                                       \

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

  863.                                                                         \

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

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

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

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

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

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

  870.         }                                                               \

  871.         s += stride;                                                    \

  872.     }                                                                   \

  873.                                                                         \

  874.     return score;                                                       \

  875. }

  876. VSSE_INTRA(8)

  877. VSSE_INTRA(16)

  878. 

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

  880.                     int stride, int h)

  881. {

  882.     int score = 0, x, y;

  883. 

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

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

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

  887.         s1 += stride;

  888.         s2 += stride;

  889.     }

  890. 

  891.     return score;

  892. }

  893. 

  894. #define WRAPPER8_16_SQ(name8, name16)                                   \

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

  896.                   int stride, int h)                                    \

  897. {                                                                       \

  898.     int score = 0;                                                      \

  899.                                                                         \

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

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

  902.     if (h == 16) {                                                      \

  903.         dst   += 8 * stride;                                            \

  904.         src   += 8 * stride;                                            \

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

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

  907.     }                                                                   \

  908.     return score;                                                       \

  909. }

  910. 

  911. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  912. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  913. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  914. #if CONFIG_GPL

  915. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  916. #endif

  917. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  918. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  919. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  920. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  921. 

  922. /* init static data */

  923. av_cold void ff_dsputil_static_init(void)

  924. {

  925.     int i;

  926. 

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

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

  929. }

  930. 

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

  932. {

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

  934. 

  935.     c->diff_pixels = diff_pixels_c;

  936. 

  937.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  938. 

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

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

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

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

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

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

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

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

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

  948. 

  949. #define SET_CMP_FUNC(name)                      \

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

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

  952. 

  953.     SET_CMP_FUNC(hadamard8_diff)

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

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

  956.     SET_CMP_FUNC(dct_sad)

  957.     SET_CMP_FUNC(dct_max)

  958. #if CONFIG_GPL

  959.     SET_CMP_FUNC(dct264_sad)

  960. #endif

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

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

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

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

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

  966.     SET_CMP_FUNC(quant_psnr)

  967.     SET_CMP_FUNC(rd)

  968.     SET_CMP_FUNC(bit)

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

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

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

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

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

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

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

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

  977. 

  978.     switch (avctx->bits_per_raw_sample) {

  979.     case 9:

  980.     case 10:

  981.         c->get_pixels = get_pixels_16_c;

  982.         break;

  983.     default:

  984.         c->get_pixels = get_pixels_8_c;

  985.         break;

  986.     }

  987. 

  988.     if (ARCH_ARM)

  989.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  990.     if (ARCH_PPC)

  991.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  992.     if (ARCH_X86)

  993.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  994. }