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.
40315 Commits
32 Branches
570MB
Tree: a0ce85ac7d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a0ce85ac7d'
${ noResults }
FFmpeg/libavcodec/me_cmp.c

943 lines
31KB
Raw Blame History 

  1. /*

  2.  * This file is part of Libav.

  3.  *

  4.  * Libav is free software; you can redistribute it and/or

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

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

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

  8.  *

  9.  * Libav is distributed in the hope that it will be useful,

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

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

  12.  * Lesser General Public License for more details.

  13.  *

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

  15.  * License along with Libav; if not, write to the Free Software

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

  17.  */

  18. 

  19. #include "libavutil/attributes.h"

  20. #include "avcodec.h"

  21. #include "copy_block.h"

  22. #include "simple_idct.h"

  23. #include "me_cmp.h"

  24. #include "mpegvideo.h"

  25. #include "config.h"

  26. 

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

  28. 

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

  30.                   int line_size, int h)

  31. {

  32.     int s = 0, i;

  33.     uint32_t *sq = ff_square_tab + 256;

  34. 

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

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

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

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

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

  40.         pix1 += line_size;

  41.         pix2 += line_size;

  42.     }

  43.     return s;

  44. }

  45. 

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

  47.                   int line_size, int h)

  48. {

  49.     int s = 0, i;

  50.     uint32_t *sq = ff_square_tab + 256;

  51. 

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

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

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

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

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

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

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

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

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

  61.         pix1 += line_size;

  62.         pix2 += line_size;

  63.     }

  64.     return s;

  65. }

  66. 

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

  68.                    int line_size, int h)

  69. {

  70.     int s = 0, i;

  71.     uint32_t *sq = ff_square_tab + 256;

  72. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  90. 

  91.         pix1 += line_size;

  92.         pix2 += line_size;

  93.     }

  94.     return s;

  95. }

  96. 

  97. static int sum_abs_dctelem_c(int16_t *block)

  98. {

  99.     int sum = 0, i;

  100. 

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

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

  103.     return sum;

  104. }

  105. 

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

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

  108. 

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

  110.                               int line_size, int h)

  111. {

  112.     int s = 0, i;

  113. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  131.         pix1 += line_size;

  132.         pix2 += line_size;

  133.     }

  134.     return s;

  135. }

  136. 

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

  138.                           int line_size, int h)

  139. {

  140.     int s = 0, i;

  141. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  159.         pix1 += line_size;

  160.         pix2 += line_size;

  161.     }

  162.     return s;

  163. }

  164. 

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

  166.                           int line_size, int h)

  167. {

  168.     int s = 0, i;

  169.     uint8_t *pix3 = pix2 + line_size;

  170. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  188.         pix1 += line_size;

  189.         pix2 += line_size;

  190.         pix3 += line_size;

  191.     }

  192.     return s;

  193. }

  194. 

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

  196.                            int line_size, int h)

  197. {

  198.     int s = 0, i;

  199.     uint8_t *pix3 = pix2 + line_size;

  200. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  218.         pix1 += line_size;

  219.         pix2 += line_size;

  220.         pix3 += line_size;

  221.     }

  222.     return s;

  223. }

  224. 

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

  226.                              int line_size, int h)

  227. {

  228.     int s = 0, i;

  229. 

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

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

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

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

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

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

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

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

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

  239.         pix1 += line_size;

  240.         pix2 += line_size;

  241.     }

  242.     return s;

  243. }

  244. 

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

  246.                          int line_size, int h)

  247. {

  248.     int s = 0, i;

  249. 

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

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

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

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

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

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

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

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

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

  259.         pix1 += line_size;

  260.         pix2 += line_size;

  261.     }

  262.     return s;

  263. }

  264. 

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

  266.                          int line_size, int h)

  267. {

  268.     int s = 0, i;

  269.     uint8_t *pix3 = pix2 + line_size;

  270. 

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

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

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

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

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

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

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

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

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

  280.         pix1 += line_size;

  281.         pix2 += line_size;

  282.         pix3 += line_size;

  283.     }

  284.     return s;

  285. }

  286. 

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

  288.                           int line_size, int h)

  289. {

  290.     int s = 0, i;

  291.     uint8_t *pix3 = pix2 + line_size;

  292. 

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

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

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

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

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

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

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

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

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

  302.         pix1 += line_size;

  303.         pix2 += line_size;

  304.         pix3 += line_size;

  305.     }

  306.     return s;

  307. }

  308. 

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

  310. {

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

  312. 

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

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

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

  316.         if (y + 1 < h) {

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

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

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

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

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

  322.         }

  323.         s1 += stride;

  324.         s2 += stride;

  325.     }

  326. 

  327.     if (c)

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

  329.     else

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

  331. }

  332. 

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

  334. {

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

  336. 

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

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

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

  340.         if (y + 1 < h) {

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

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

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

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

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

  346.         }

  347.         s1 += stride;

  348.         s2 += stride;

  349.     }

  350. 

  351.     if (c)

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

  353.     else

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

  355. }

  356. 

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

  358.                     int stride, int h)

  359. {

  360.     return 0;

  361. }

  362. 

  363. void ff_set_cmp(MECmpContext *c, me_cmp_func *cmp, int type)

  364. {

  365.     int i;

  366. 

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

  368. 

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

  370.         switch (type & 0xFF) {

  371.         case FF_CMP_SAD:

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

  373.             break;

  374.         case FF_CMP_SATD:

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

  376.             break;

  377.         case FF_CMP_SSE:

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

  379.             break;

  380.         case FF_CMP_DCT:

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

  382.             break;

  383.         case FF_CMP_DCT264:

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

  385.             break;

  386.         case FF_CMP_DCTMAX:

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

  388.             break;

  389.         case FF_CMP_PSNR:

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

  391.             break;

  392.         case FF_CMP_BIT:

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

  394.             break;

  395.         case FF_CMP_RD:

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

  397.             break;

  398.         case FF_CMP_VSAD:

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

  400.             break;

  401.         case FF_CMP_VSSE:

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

  403.             break;

  404.         case FF_CMP_ZERO:

  405.             cmp[i] = zero_cmp;

  406.             break;

  407.         case FF_CMP_NSSE:

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

  409.             break;

  410.         default:

  411.             av_log(NULL, AV_LOG_ERROR,

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

  413.         }

  414.     }

  415. }

  416. 

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

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

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

  420. 

  421. #define BUTTERFLY1(x, y)                        \

  422.     {                                           \

  423.         int a, b;                               \

  424.         a = x;                                  \

  425.         b = y;                                  \

  426.         x = a + b;                              \

  427.         y = a - b;                              \

  428.     }

  429. 

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

  431. 

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

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

  434. {

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

  436. 

  437.     assert(h == 8);

  438. 

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

  440.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  453. 

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

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

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

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

  458. 

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

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

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

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

  463.     }

  464. 

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

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

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

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

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

  470. 

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

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

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

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

  475. 

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

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

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

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

  480.     }

  481.     return sum;

  482. }

  483. 

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

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

  486. {

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

  488. 

  489.     assert(h == 8);

  490. 

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

  492.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  501. 

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

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

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

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

  506. 

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

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

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

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

  511.     }

  512. 

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

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

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

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

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

  518. 

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

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

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

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

  523. 

  524.         sum +=

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

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

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

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

  529.     }

  530. 

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

  532. 

  533.     return sum;

  534. }

  535. 

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

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

  538. {

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

  540. 

  541.     assert(h == 8);

  542. 

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

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

  545.     return s->mecc.sum_abs_dctelem(temp);

  546. }

  547. 

  548. #if CONFIG_GPL

  549. #define DCT8_1D                                         \

  550.     {                                                   \

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

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

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

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

  555.         const int a0  = s07 + s34;                      \

  556.         const int a1  = s16 + s25;                      \

  557.         const int a2  = s07 - s34;                      \

  558.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  575.     }

  576. 

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

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

  579. {

  580.     int16_t dct[8][8];

  581.     int i, sum = 0;

  582. 

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

  584. 

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

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

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

  588.         DCT8_1D

  589. #undef SRC

  590. #undef DST

  591. 

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

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

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

  595.             DCT8_1D

  596. #undef SRC

  597. #undef DST

  598.             return sum;

  599. }

  600. #endif

  601. 

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

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

  604. {

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

  606.     int sum = 0, i;

  607. 

  608.     assert(h == 8);

  609. 

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

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

  612. 

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

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

  615. 

  616.     return sum;

  617. }

  618. 

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

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

  621. {

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

  623.     int16_t *const bak = temp + 64;

  624.     int sum = 0, i;

  625. 

  626.     assert(h == 8);

  627.     s->mb_intra = 0;

  628. 

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

  630. 

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

  632. 

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

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

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

  636.     ff_simple_idct_8(temp); // FIXME

  637. 

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

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

  640. 

  641.     return sum;

  642. }

  643. 

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

  645.                    int stride, int h)

  646. {

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

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

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

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

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

  652.     const int esc_length = s->ac_esc_length;

  653.     uint8_t *length, *last_length;

  654. 

  655.     assert(h == 8);

  656. 

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

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

  659. 

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

  661. 

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

  663.     last                               =

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

  665. 

  666.     bits = 0;

  667. 

  668.     if (s->mb_intra) {

  669.         start_i     = 1;

  670.         length      = s->intra_ac_vlc_length;

  671.         last_length = s->intra_ac_vlc_last_length;

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

  673.     } else {

  674.         start_i     = 0;

  675.         length      = s->inter_ac_vlc_length;

  676.         last_length = s->inter_ac_vlc_last_length;

  677.     }

  678. 

  679.     if (last >= start_i) {

  680.         run = 0;

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

  682.             int j = scantable[i];

  683.             level = temp[j];

  684. 

  685.             if (level) {

  686.                 level += 64;

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

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

  689.                 else

  690.                     bits += esc_length;

  691.                 run = 0;

  692.             } else

  693.                 run++;

  694.         }

  695.         i = scantable[last];

  696. 

  697.         level = temp[i] + 64;

  698. 

  699.         assert(level - 64);

  700. 

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

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

  703.         } else

  704.             bits += esc_length;

  705.     }

  706. 

  707.     if (last >= 0) {

  708.         if (s->mb_intra)

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

  710.         else

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

  712.     }

  713. 

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

  715. 

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

  717. 

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

  719. }

  720. 

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

  722.                     int stride, int h)

  723. {

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

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

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

  727.     const int esc_length = s->ac_esc_length;

  728.     uint8_t *length, *last_length;

  729. 

  730.     assert(h == 8);

  731. 

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

  733. 

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

  735.     last                               =

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

  737. 

  738.     bits = 0;

  739. 

  740.     if (s->mb_intra) {

  741.         start_i     = 1;

  742.         length      = s->intra_ac_vlc_length;

  743.         last_length = s->intra_ac_vlc_last_length;

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

  745.     } else {

  746.         start_i     = 0;

  747.         length      = s->inter_ac_vlc_length;

  748.         last_length = s->inter_ac_vlc_last_length;

  749.     }

  750. 

  751.     if (last >= start_i) {

  752.         run = 0;

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

  754.             int j = scantable[i];

  755.             level = temp[j];

  756. 

  757.             if (level) {

  758.                 level += 64;

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

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

  761.                 else

  762.                     bits += esc_length;

  763.                 run = 0;

  764.             } else

  765.                 run++;

  766.         }

  767.         i = scantable[last];

  768. 

  769.         level = temp[i] + 64;

  770. 

  771.         assert(level - 64);

  772. 

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

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

  775.         else

  776.             bits += esc_length;

  777.     }

  778. 

  779.     return bits;

  780. }

  781. 

  782. #define VSAD_INTRA(size)                                                \

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

  784.                                     uint8_t *s, uint8_t *dummy,         \

  785.                                     int stride, int h)                  \

  786. {                                                                       \

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

  788.                                                                         \

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

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

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

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

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

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

  795.         }                                                               \

  796.         s += stride;                                                    \

  797.     }                                                                   \

  798.                                                                         \

  799.     return score;                                                       \

  800. }

  801. VSAD_INTRA(8)

  802. VSAD_INTRA(16)

  803. 

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

  805.                     int stride, int h)

  806. {

  807.     int score = 0, x, y;

  808. 

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

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

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

  812.         s1 += stride;

  813.         s2 += stride;

  814.     }

  815. 

  816.     return score;

  817. }

  818. 

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

  820. #define VSSE_INTRA(size)                                                \

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

  822.                                     uint8_t *s, uint8_t *dummy,         \

  823.                                     int stride, int h)                  \

  824. {                                                                       \

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

  826.                                                                         \

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

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

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

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

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

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

  833.         }                                                               \

  834.         s += stride;                                                    \

  835.     }                                                                   \

  836.                                                                         \

  837.     return score;                                                       \

  838. }

  839. VSSE_INTRA(8)

  840. VSSE_INTRA(16)

  841. 

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

  843.                     int stride, int h)

  844. {

  845.     int score = 0, x, y;

  846. 

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

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

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

  850.         s1 += stride;

  851.         s2 += stride;

  852.     }

  853. 

  854.     return score;

  855. }

  856. 

  857. #define WRAPPER8_16_SQ(name8, name16)                                   \

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

  859.                   int stride, int h)                                    \

  860. {                                                                       \

  861.     int score = 0;                                                      \

  862.                                                                         \

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

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

  865.     if (h == 16) {                                                      \

  866.         dst   += 8 * stride;                                            \

  867.         src   += 8 * stride;                                            \

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

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

  870.     }                                                                   \

  871.     return score;                                                       \

  872. }

  873. 

  874. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  875. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  876. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  877. #if CONFIG_GPL

  878. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  879. #endif

  880. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  881. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  882. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  883. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  884. 

  885. av_cold void ff_me_cmp_init_static(void)

  886. {

  887.     int i;

  888. 

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

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

  891. }

  892. 

  893. av_cold void ff_me_cmp_init(MECmpContext *c, AVCodecContext *avctx)

  894. {

  895.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  896. 

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

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

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

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

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

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

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

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

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

  906. 

  907. #define SET_CMP_FUNC(name)                      \

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

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

  910. 

  911.     SET_CMP_FUNC(hadamard8_diff)

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

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

  914.     SET_CMP_FUNC(dct_sad)

  915.     SET_CMP_FUNC(dct_max)

  916. #if CONFIG_GPL

  917.     SET_CMP_FUNC(dct264_sad)

  918. #endif

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

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

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

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

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

  924.     SET_CMP_FUNC(quant_psnr)

  925.     SET_CMP_FUNC(rd)

  926.     SET_CMP_FUNC(bit)

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

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

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

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

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

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

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

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

  935. 

  936.     if (ARCH_ARM)

  937.         ff_me_cmp_init_arm(c, avctx);

  938.     if (ARCH_PPC)

  939.         ff_me_cmp_init_ppc(c, avctx);

  940.     if (ARCH_X86)

  941.         ff_me_cmp_init_x86(c, avctx);

  942. }