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.
40197 Commits
32 Branches
896MB
Tree: f46bb608d9
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'f46bb608d9'
${ noResults }
FFmpeg/libavcodec/dsputil.c

955 lines
32KB
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. static int sse4_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  39.                   int line_size, int h)

  40. {

  41.     int s = 0, i;

  42.     uint32_t *sq = ff_square_tab + 256;

  43. 

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

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

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

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

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

  49.         pix1 += line_size;

  50.         pix2 += line_size;

  51.     }

  52.     return s;

  53. }

  54. 

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

  56.                   int line_size, int h)

  57. {

  58.     int s = 0, i;

  59.     uint32_t *sq = ff_square_tab + 256;

  60. 

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

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

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

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

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

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

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

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

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

  70.         pix1 += line_size;

  71.         pix2 += line_size;

  72.     }

  73.     return s;

  74. }

  75. 

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

  77.                    int line_size, int h)

  78. {

  79.     int s = 0, i;

  80.     uint32_t *sq = ff_square_tab + 256;

  81. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  99. 

  100.         pix1 += line_size;

  101.         pix2 += line_size;

  102.     }

  103.     return s;

  104. }

  105. 

  106. static int sum_abs_dctelem_c(int16_t *block)

  107. {

  108.     int sum = 0, i;

  109. 

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

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

  112.     return sum;

  113. }

  114. 

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

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

  117. 

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

  119.                               int line_size, int h)

  120. {

  121.     int s = 0, i;

  122. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  140.         pix1 += line_size;

  141.         pix2 += line_size;

  142.     }

  143.     return s;

  144. }

  145. 

  146. static int pix_abs16_x2_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]  - avg2(pix2[0],  pix2[1]));

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  168.         pix1 += line_size;

  169.         pix2 += line_size;

  170.     }

  171.     return s;

  172. }

  173. 

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

  175.                           int line_size, int h)

  176. {

  177.     int s = 0, i;

  178.     uint8_t *pix3 = pix2 + line_size;

  179. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  197.         pix1 += line_size;

  198.         pix2 += line_size;

  199.         pix3 += line_size;

  200.     }

  201.     return s;

  202. }

  203. 

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

  205.                            int line_size, int h)

  206. {

  207.     int s = 0, i;

  208.     uint8_t *pix3 = pix2 + line_size;

  209. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  227.         pix1 += line_size;

  228.         pix2 += line_size;

  229.         pix3 += line_size;

  230.     }

  231.     return s;

  232. }

  233. 

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

  235.                              int line_size, int h)

  236. {

  237.     int s = 0, i;

  238. 

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

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

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

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

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

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

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

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

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

  248.         pix1 += line_size;

  249.         pix2 += line_size;

  250.     }

  251.     return s;

  252. }

  253. 

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

  255.                          int line_size, int h)

  256. {

  257.     int s = 0, i;

  258. 

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

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

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

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

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

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

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

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

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

  268.         pix1 += line_size;

  269.         pix2 += line_size;

  270.     }

  271.     return s;

  272. }

  273. 

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

  275.                          int line_size, int h)

  276. {

  277.     int s = 0, i;

  278.     uint8_t *pix3 = pix2 + line_size;

  279. 

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

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

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

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

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

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

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

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

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

  289.         pix1 += line_size;

  290.         pix2 += line_size;

  291.         pix3 += line_size;

  292.     }

  293.     return s;

  294. }

  295. 

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

  297.                           int line_size, int h)

  298. {

  299.     int s = 0, i;

  300.     uint8_t *pix3 = pix2 + line_size;

  301. 

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

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

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

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

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

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

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

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

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

  311.         pix1 += line_size;

  312.         pix2 += line_size;

  313.         pix3 += line_size;

  314.     }

  315.     return s;

  316. }

  317. 

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

  319. {

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

  321. 

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

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

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

  325.         if (y + 1 < h) {

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

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

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

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

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

  331.         }

  332.         s1 += stride;

  333.         s2 += stride;

  334.     }

  335. 

  336.     if (c)

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

  338.     else

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

  340. }

  341. 

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

  343. {

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

  345. 

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

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

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

  349.         if (y + 1 < h) {

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

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

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

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

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

  355.         }

  356.         s1 += stride;

  357.         s2 += stride;

  358.     }

  359. 

  360.     if (c)

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

  362.     else

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

  364. }

  365. 

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

  367.                     int stride, int h)

  368. {

  369.     return 0;

  370. }

  371. 

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

  373. {

  374.     int i;

  375. 

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

  377. 

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

  379.         switch (type & 0xFF) {

  380.         case FF_CMP_SAD:

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

  382.             break;

  383.         case FF_CMP_SATD:

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

  385.             break;

  386.         case FF_CMP_SSE:

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

  388.             break;

  389.         case FF_CMP_DCT:

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

  391.             break;

  392.         case FF_CMP_DCT264:

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

  394.             break;

  395.         case FF_CMP_DCTMAX:

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

  397.             break;

  398.         case FF_CMP_PSNR:

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

  400.             break;

  401.         case FF_CMP_BIT:

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

  403.             break;

  404.         case FF_CMP_RD:

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

  406.             break;

  407.         case FF_CMP_VSAD:

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

  409.             break;

  410.         case FF_CMP_VSSE:

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

  412.             break;

  413.         case FF_CMP_ZERO:

  414.             cmp[i] = zero_cmp;

  415.             break;

  416.         case FF_CMP_NSSE:

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

  418.             break;

  419.         default:

  420.             av_log(NULL, AV_LOG_ERROR,

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

  422.         }

  423.     }

  424. }

  425. 

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

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

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

  429. 

  430. #define BUTTERFLY1(x, y)                        \

  431.     {                                           \

  432.         int a, b;                               \

  433.         a = x;                                  \

  434.         b = y;                                  \

  435.         x = a + b;                              \

  436.         y = a - b;                              \

  437.     }

  438. 

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

  440. 

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

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

  443. {

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

  445. 

  446.     assert(h == 8);

  447. 

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

  449.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  462. 

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

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

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

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

  467. 

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

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

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

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

  472.     }

  473. 

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

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

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

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

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

  479. 

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

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

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

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

  484. 

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

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

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

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

  489.     }

  490.     return sum;

  491. }

  492. 

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

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

  495. {

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

  497. 

  498.     assert(h == 8);

  499. 

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

  501.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  510. 

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

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

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

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

  515. 

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

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

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

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

  520.     }

  521. 

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

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

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

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

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

  527. 

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

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

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

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

  532. 

  533.         sum +=

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

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

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

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

  538.     }

  539. 

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

  541. 

  542.     return sum;

  543. }

  544. 

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

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

  547. {

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

  549. 

  550.     assert(h == 8);

  551. 

  552.     s->pdsp.diff_pixels(temp, src1, src2, stride);

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

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

  555. }

  556. 

  557. #if CONFIG_GPL

  558. #define DCT8_1D                                         \

  559.     {                                                   \

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

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

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

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

  564.         const int a0  = s07 + s34;                      \

  565.         const int a1  = s16 + s25;                      \

  566.         const int a2  = s07 - s34;                      \

  567.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  584.     }

  585. 

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

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

  588. {

  589.     int16_t dct[8][8];

  590.     int i, sum = 0;

  591. 

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

  593. 

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

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

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

  597.         DCT8_1D

  598. #undef SRC

  599. #undef DST

  600. 

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

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

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

  604.             DCT8_1D

  605. #undef SRC

  606. #undef DST

  607.             return sum;

  608. }

  609. #endif

  610. 

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

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

  613. {

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

  615.     int sum = 0, i;

  616. 

  617.     assert(h == 8);

  618. 

  619.     s->pdsp.diff_pixels(temp, src1, src2, stride);

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

  621. 

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

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

  624. 

  625.     return sum;

  626. }

  627. 

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

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

  630. {

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

  632.     int16_t *const bak = temp + 64;

  633.     int sum = 0, i;

  634. 

  635.     assert(h == 8);

  636.     s->mb_intra = 0;

  637. 

  638.     s->pdsp.diff_pixels(temp, src1, src2, stride);

  639. 

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

  641. 

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

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

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

  645.     ff_simple_idct_8(temp); // FIXME

  646. 

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

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

  649. 

  650.     return sum;

  651. }

  652. 

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

  654.                    int stride, int h)

  655. {

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

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

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

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

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

  661.     const int esc_length = s->ac_esc_length;

  662.     uint8_t *length, *last_length;

  663. 

  664.     assert(h == 8);

  665. 

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

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

  668. 

  669.     s->pdsp.diff_pixels(temp, lsrc1, lsrc2, 8);

  670. 

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

  672.     last                               =

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

  674. 

  675.     bits = 0;

  676. 

  677.     if (s->mb_intra) {

  678.         start_i     = 1;

  679.         length      = s->intra_ac_vlc_length;

  680.         last_length = s->intra_ac_vlc_last_length;

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

  682.     } else {

  683.         start_i     = 0;

  684.         length      = s->inter_ac_vlc_length;

  685.         last_length = s->inter_ac_vlc_last_length;

  686.     }

  687. 

  688.     if (last >= start_i) {

  689.         run = 0;

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

  691.             int j = scantable[i];

  692.             level = temp[j];

  693. 

  694.             if (level) {

  695.                 level += 64;

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

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

  698.                 else

  699.                     bits += esc_length;

  700.                 run = 0;

  701.             } else

  702.                 run++;

  703.         }

  704.         i = scantable[last];

  705. 

  706.         level = temp[i] + 64;

  707. 

  708.         assert(level - 64);

  709. 

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

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

  712.         } else

  713.             bits += esc_length;

  714.     }

  715. 

  716.     if (last >= 0) {

  717.         if (s->mb_intra)

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

  719.         else

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

  721.     }

  722. 

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

  724. 

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

  726. 

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

  728. }

  729. 

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

  731.                     int stride, int h)

  732. {

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

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

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

  736.     const int esc_length = s->ac_esc_length;

  737.     uint8_t *length, *last_length;

  738. 

  739.     assert(h == 8);

  740. 

  741.     s->pdsp.diff_pixels(temp, src1, src2, stride);

  742. 

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

  744.     last                               =

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

  746. 

  747.     bits = 0;

  748. 

  749.     if (s->mb_intra) {

  750.         start_i     = 1;

  751.         length      = s->intra_ac_vlc_length;

  752.         last_length = s->intra_ac_vlc_last_length;

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

  754.     } else {

  755.         start_i     = 0;

  756.         length      = s->inter_ac_vlc_length;

  757.         last_length = s->inter_ac_vlc_last_length;

  758.     }

  759. 

  760.     if (last >= start_i) {

  761.         run = 0;

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

  763.             int j = scantable[i];

  764.             level = temp[j];

  765. 

  766.             if (level) {

  767.                 level += 64;

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

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

  770.                 else

  771.                     bits += esc_length;

  772.                 run = 0;

  773.             } else

  774.                 run++;

  775.         }

  776.         i = scantable[last];

  777. 

  778.         level = temp[i] + 64;

  779. 

  780.         assert(level - 64);

  781. 

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

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

  784.         else

  785.             bits += esc_length;

  786.     }

  787. 

  788.     return bits;

  789. }

  790. 

  791. #define VSAD_INTRA(size)                                                \

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

  793.                                     uint8_t *s, uint8_t *dummy,         \

  794.                                     int stride, int h)                  \

  795. {                                                                       \

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

  797.                                                                         \

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

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

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

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

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

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

  804.         }                                                               \

  805.         s += stride;                                                    \

  806.     }                                                                   \

  807.                                                                         \

  808.     return score;                                                       \

  809. }

  810. VSAD_INTRA(8)

  811. VSAD_INTRA(16)

  812. 

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

  814.                     int stride, int h)

  815. {

  816.     int score = 0, x, y;

  817. 

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

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

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

  821.         s1 += stride;

  822.         s2 += stride;

  823.     }

  824. 

  825.     return score;

  826. }

  827. 

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

  829. #define VSSE_INTRA(size)                                                \

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

  831.                                     uint8_t *s, uint8_t *dummy,         \

  832.                                     int stride, int h)                  \

  833. {                                                                       \

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

  835.                                                                         \

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

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

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

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

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

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

  842.         }                                                               \

  843.         s += stride;                                                    \

  844.     }                                                                   \

  845.                                                                         \

  846.     return score;                                                       \

  847. }

  848. VSSE_INTRA(8)

  849. VSSE_INTRA(16)

  850. 

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

  852.                     int stride, int h)

  853. {

  854.     int score = 0, x, y;

  855. 

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

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

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

  859.         s1 += stride;

  860.         s2 += stride;

  861.     }

  862. 

  863.     return score;

  864. }

  865. 

  866. #define WRAPPER8_16_SQ(name8, name16)                                   \

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

  868.                   int stride, int h)                                    \

  869. {                                                                       \

  870.     int score = 0;                                                      \

  871.                                                                         \

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

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

  874.     if (h == 16) {                                                      \

  875.         dst   += 8 * stride;                                            \

  876.         src   += 8 * stride;                                            \

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

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

  879.     }                                                                   \

  880.     return score;                                                       \

  881. }

  882. 

  883. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  884. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  885. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  886. #if CONFIG_GPL

  887. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  888. #endif

  889. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  890. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  891. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  892. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  893. 

  894. /* init static data */

  895. av_cold void ff_dsputil_static_init(void)

  896. {

  897.     int i;

  898. 

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

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

  901. }

  902. 

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

  904. {

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

  906. 

  907.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  908. 

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

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

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

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

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

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

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

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

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

  918. 

  919. #define SET_CMP_FUNC(name)                      \

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

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

  922. 

  923.     SET_CMP_FUNC(hadamard8_diff)

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

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

  926.     SET_CMP_FUNC(dct_sad)

  927.     SET_CMP_FUNC(dct_max)

  928. #if CONFIG_GPL

  929.     SET_CMP_FUNC(dct264_sad)

  930. #endif

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

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

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

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

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

  936.     SET_CMP_FUNC(quant_psnr)

  937.     SET_CMP_FUNC(rd)

  938.     SET_CMP_FUNC(bit)

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

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

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

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

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

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

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

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

  947. 

  948.     if (ARCH_ARM)

  949.         ff_dsputil_init_arm(c, avctx, high_bit_depth);

  950.     if (ARCH_PPC)

  951.         ff_dsputil_init_ppc(c, avctx, high_bit_depth);

  952.     if (ARCH_X86)

  953.         ff_dsputil_init_x86(c, avctx, high_bit_depth);

  954. }