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.
44689 Commits
32 Branches
824MB
Tree: c5714b51aa
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c5714b51aa'
${ noResults }
FFmpeg/libavcodec/me_cmp.c

945 lines
32KB
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.                   ptrdiff_t stride, 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 += stride;

  41.         pix2 += stride;

  42.     }

  43.     return s;

  44. }

  45. 

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

  47.                   ptrdiff_t stride, 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 += stride;

  62.         pix2 += stride;

  63.     }

  64.     return s;

  65. }

  66. 

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

  68.                    ptrdiff_t stride, 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 += stride;

  92.         pix2 += stride;

  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.                               ptrdiff_t stride, 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 += stride;

  132.         pix2 += stride;

  133.     }

  134.     return s;

  135. }

  136. 

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

  138.                           ptrdiff_t stride, 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 += stride;

  160.         pix2 += stride;

  161.     }

  162.     return s;

  163. }

  164. 

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

  166.                           ptrdiff_t stride, int h)

  167. {

  168.     int s = 0, i;

  169.     uint8_t *pix3 = pix2 + stride;

  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 += stride;

  189.         pix2 += stride;

  190.         pix3 += stride;

  191.     }

  192.     return s;

  193. }

  194. 

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

  196.                            ptrdiff_t stride, int h)

  197. {

  198.     int s = 0, i;

  199.     uint8_t *pix3 = pix2 + stride;

  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 += stride;

  219.         pix2 += stride;

  220.         pix3 += stride;

  221.     }

  222.     return s;

  223. }

  224. 

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

  226.                              ptrdiff_t stride, 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 += stride;

  240.         pix2 += stride;

  241.     }

  242.     return s;

  243. }

  244. 

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

  246.                          ptrdiff_t stride, 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 += stride;

  260.         pix2 += stride;

  261.     }

  262.     return s;

  263. }

  264. 

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

  266.                          ptrdiff_t stride, int h)

  267. {

  268.     int s = 0, i;

  269.     uint8_t *pix3 = pix2 + stride;

  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 += stride;

  281.         pix2 += stride;

  282.         pix3 += stride;

  283.     }

  284.     return s;

  285. }

  286. 

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

  288.                           ptrdiff_t stride, int h)

  289. {

  290.     int s = 0, i;

  291.     uint8_t *pix3 = pix2 + stride;

  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 += stride;

  303.         pix2 += stride;

  304.         pix3 += stride;

  305.     }

  306.     return s;

  307. }

  308. 

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

  310.                     ptrdiff_t stride, int h)

  311. {

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

  313. 

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

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

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

  317.         if (y + 1 < h) {

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

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

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

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

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

  323.         }

  324.         s1 += stride;

  325.         s2 += stride;

  326.     }

  327. 

  328.     if (c)

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

  330.     else

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

  332. }

  333. 

  334. static int nsse8_c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,

  335.                    ptrdiff_t stride, int h)

  336. {

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

  338. 

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

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

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

  342.         if (y + 1 < h) {

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

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

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

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

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

  348.         }

  349.         s1 += stride;

  350.         s2 += stride;

  351.     }

  352. 

  353.     if (c)

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

  355.     else

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

  357. }

  358. 

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

  360.                     ptrdiff_t stride, int h)

  361. {

  362.     return 0;

  363. }

  364. 

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

  366. {

  367.     int i;

  368. 

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

  370. 

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

  372.         switch (type & 0xFF) {

  373.         case FF_CMP_SAD:

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

  375.             break;

  376.         case FF_CMP_SATD:

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

  378.             break;

  379.         case FF_CMP_SSE:

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

  381.             break;

  382.         case FF_CMP_DCT:

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

  384.             break;

  385.         case FF_CMP_DCT264:

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

  387.             break;

  388.         case FF_CMP_DCTMAX:

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

  390.             break;

  391.         case FF_CMP_PSNR:

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

  393.             break;

  394.         case FF_CMP_BIT:

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

  396.             break;

  397.         case FF_CMP_RD:

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

  399.             break;

  400.         case FF_CMP_VSAD:

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

  402.             break;

  403.         case FF_CMP_VSSE:

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

  405.             break;

  406.         case FF_CMP_ZERO:

  407.             cmp[i] = zero_cmp;

  408.             break;

  409.         case FF_CMP_NSSE:

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

  411.             break;

  412.         default:

  413.             av_log(NULL, AV_LOG_ERROR,

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

  415.         }

  416.     }

  417. }

  418. 

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

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

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

  422. 

  423. #define BUTTERFLY1(x, y)                        \

  424.     {                                           \

  425.         int a, b;                               \

  426.         a = x;                                  \

  427.         b = y;                                  \

  428.         x = a + b;                              \

  429.         y = a - b;                              \

  430.     }

  431. 

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

  433. 

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

  435.                                uint8_t *src, ptrdiff_t stride, int h)

  436. {

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

  438. 

  439.     assert(h == 8);

  440. 

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

  442.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  455. 

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

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

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

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

  460. 

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

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

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

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

  465.     }

  466. 

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

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

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

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

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

  472. 

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

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

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

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

  477. 

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

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

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

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

  482.     }

  483.     return sum;

  484. }

  485. 

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

  487.                                 uint8_t *dummy, ptrdiff_t stride, int h)

  488. {

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

  490. 

  491.     assert(h == 8);

  492. 

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

  494.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  503. 

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

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

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

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

  508. 

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

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

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

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

  513.     }

  514. 

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

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

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

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

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

  520. 

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

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

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

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

  525. 

  526.         sum +=

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

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

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

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

  531.     }

  532. 

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

  534. 

  535.     return sum;

  536. }

  537. 

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

  539.                         uint8_t *src2, ptrdiff_t stride, int h)

  540. {

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

  542. 

  543.     assert(h == 8);

  544. 

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

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

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

  548. }

  549. 

  550. #if CONFIG_GPL

  551. #define DCT8_1D                                         \

  552.     {                                                   \

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

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

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

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

  557.         const int a0  = s07 + s34;                      \

  558.         const int a1  = s16 + s25;                      \

  559.         const int a2  = s07 - s34;                      \

  560.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  577.     }

  578. 

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

  580.                            uint8_t *src2, ptrdiff_t stride, int h)

  581. {

  582.     int16_t dct[8][8];

  583.     int i, sum = 0;

  584. 

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

  586. 

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

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

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

  590.         DCT8_1D

  591. #undef SRC

  592. #undef DST

  593. 

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

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

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

  597.             DCT8_1D

  598. #undef SRC

  599. #undef DST

  600.             return sum;

  601. }

  602. #endif

  603. 

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

  605.                         uint8_t *src2, ptrdiff_t stride, int h)

  606. {

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

  608.     int sum = 0, i;

  609. 

  610.     assert(h == 8);

  611. 

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

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

  614. 

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

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

  617. 

  618.     return sum;

  619. }

  620. 

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

  622.                            uint8_t *src2, ptrdiff_t stride, int h)

  623. {

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

  625.     int16_t *const bak = temp + 64;

  626.     int sum = 0, i;

  627. 

  628.     assert(h == 8);

  629.     s->mb_intra = 0;

  630. 

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

  632. 

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

  634. 

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

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

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

  638.     ff_simple_idct_8(temp); // FIXME

  639. 

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

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

  642. 

  643.     return sum;

  644. }

  645. 

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

  647.                    ptrdiff_t stride, int h)

  648. {

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

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

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

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

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

  654.     const int esc_length = s->ac_esc_length;

  655.     uint8_t *length, *last_length;

  656. 

  657.     assert(h == 8);

  658. 

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

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

  661. 

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

  663. 

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

  665.     last                               =

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

  667. 

  668.     bits = 0;

  669. 

  670.     if (s->mb_intra) {

  671.         start_i     = 1;

  672.         length      = s->intra_ac_vlc_length;

  673.         last_length = s->intra_ac_vlc_last_length;

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

  675.     } else {

  676.         start_i     = 0;

  677.         length      = s->inter_ac_vlc_length;

  678.         last_length = s->inter_ac_vlc_last_length;

  679.     }

  680. 

  681.     if (last >= start_i) {

  682.         run = 0;

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

  684.             int j = scantable[i];

  685.             level = temp[j];

  686. 

  687.             if (level) {

  688.                 level += 64;

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

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

  691.                 else

  692.                     bits += esc_length;

  693.                 run = 0;

  694.             } else

  695.                 run++;

  696.         }

  697.         i = scantable[last];

  698. 

  699.         level = temp[i] + 64;

  700. 

  701.         assert(level - 64);

  702. 

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

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

  705.         } else

  706.             bits += esc_length;

  707.     }

  708. 

  709.     if (last >= 0) {

  710.         if (s->mb_intra)

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

  712.         else

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

  714.     }

  715. 

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

  717. 

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

  719. 

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

  721. }

  722. 

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

  724.                     ptrdiff_t stride, int h)

  725. {

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

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

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

  729.     const int esc_length = s->ac_esc_length;

  730.     uint8_t *length, *last_length;

  731. 

  732.     assert(h == 8);

  733. 

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

  735. 

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

  737.     last                               =

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

  739. 

  740.     bits = 0;

  741. 

  742.     if (s->mb_intra) {

  743.         start_i     = 1;

  744.         length      = s->intra_ac_vlc_length;

  745.         last_length = s->intra_ac_vlc_last_length;

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

  747.     } else {

  748.         start_i     = 0;

  749.         length      = s->inter_ac_vlc_length;

  750.         last_length = s->inter_ac_vlc_last_length;

  751.     }

  752. 

  753.     if (last >= start_i) {

  754.         run = 0;

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

  756.             int j = scantable[i];

  757.             level = temp[j];

  758. 

  759.             if (level) {

  760.                 level += 64;

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

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

  763.                 else

  764.                     bits += esc_length;

  765.                 run = 0;

  766.             } else

  767.                 run++;

  768.         }

  769.         i = scantable[last];

  770. 

  771.         level = temp[i] + 64;

  772. 

  773.         assert(level - 64);

  774. 

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

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

  777.         else

  778.             bits += esc_length;

  779.     }

  780. 

  781.     return bits;

  782. }

  783. 

  784. #define VSAD_INTRA(size)                                                \

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

  786.                                     uint8_t *s, uint8_t *dummy,         \

  787.                                     ptrdiff_t stride, int h)            \

  788. {                                                                       \

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

  790.                                                                         \

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

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

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

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

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

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

  797.         }                                                               \

  798.         s += stride;                                                    \

  799.     }                                                                   \

  800.                                                                         \

  801.     return score;                                                       \

  802. }

  803. VSAD_INTRA(8)

  804. VSAD_INTRA(16)

  805. 

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

  807.                     ptrdiff_t stride, int h)

  808. {

  809.     int score = 0, x, y;

  810. 

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

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

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

  814.         s1 += stride;

  815.         s2 += stride;

  816.     }

  817. 

  818.     return score;

  819. }

  820. 

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

  822. #define VSSE_INTRA(size)                                                \

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

  824.                                     uint8_t *s, uint8_t *dummy,         \

  825.                                     ptrdiff_t stride, int h)            \

  826. {                                                                       \

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

  828.                                                                         \

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

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

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

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

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

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

  835.         }                                                               \

  836.         s += stride;                                                    \

  837.     }                                                                   \

  838.                                                                         \

  839.     return score;                                                       \

  840. }

  841. VSSE_INTRA(8)

  842. VSSE_INTRA(16)

  843. 

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

  845.                     ptrdiff_t stride, int h)

  846. {

  847.     int score = 0, x, y;

  848. 

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

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

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

  852.         s1 += stride;

  853.         s2 += stride;

  854.     }

  855. 

  856.     return score;

  857. }

  858. 

  859. #define WRAPPER8_16_SQ(name8, name16)                                   \

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

  861.                   ptrdiff_t stride, int h)                              \

  862. {                                                                       \

  863.     int score = 0;                                                      \

  864.                                                                         \

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

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

  867.     if (h == 16) {                                                      \

  868.         dst   += 8 * stride;                                            \

  869.         src   += 8 * stride;                                            \

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

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

  872.     }                                                                   \

  873.     return score;                                                       \

  874. }

  875. 

  876. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  877. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  878. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  879. #if CONFIG_GPL

  880. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  881. #endif

  882. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  883. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  884. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  885. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  886. 

  887. av_cold void ff_me_cmp_init_static(void)

  888. {

  889.     int i;

  890. 

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

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

  893. }

  894. 

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

  896. {

  897.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  898. 

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

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

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

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

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

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

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

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

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

  908. 

  909. #define SET_CMP_FUNC(name)                      \

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

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

  912. 

  913.     SET_CMP_FUNC(hadamard8_diff)

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

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

  916.     SET_CMP_FUNC(dct_sad)

  917.     SET_CMP_FUNC(dct_max)

  918. #if CONFIG_GPL

  919.     SET_CMP_FUNC(dct264_sad)

  920. #endif

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

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

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

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

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

  926.     SET_CMP_FUNC(quant_psnr)

  927.     SET_CMP_FUNC(rd)

  928.     SET_CMP_FUNC(bit)

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

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

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

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

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

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

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

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

  937. 

  938.     if (ARCH_ARM)

  939.         ff_me_cmp_init_arm(c, avctx);

  940.     if (ARCH_PPC)

  941.         ff_me_cmp_init_ppc(c, avctx);

  942.     if (ARCH_X86)

  943.         ff_me_cmp_init_x86(c, avctx);

  944. }