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.
100531 Commits
32 Branches
424MB
Tree: c8c2dfbc37
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c8c2dfbc37'
${ noResults }
FFmpeg/libavcodec/me_cmp.c

1100 lines
40KB
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 FFmpeg.

  7.  *

  8.  * FFmpeg 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.  * FFmpeg 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 FFmpeg; if not, write to the Free Software

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

  21.  */

  22. 

  23. #include "libavutil/attributes.h"

  24. #include "libavutil/internal.h"

  25. #include "libavutil/mem_internal.h"

  26. #include "avcodec.h"

  27. #include "copy_block.h"

  28. #include "simple_idct.h"

  29. #include "me_cmp.h"

  30. #include "mpegvideo.h"

  31. #include "config.h"

  32. 

  33. /* (i - 256) * (i - 256) */

  34. const uint32_t ff_square_tab[512] = {

  35.     65536, 65025, 64516, 64009, 63504, 63001, 62500, 62001, 61504, 61009, 60516, 60025, 59536, 59049, 58564, 58081,

  36.     57600, 57121, 56644, 56169, 55696, 55225, 54756, 54289, 53824, 53361, 52900, 52441, 51984, 51529, 51076, 50625,

  37.     50176, 49729, 49284, 48841, 48400, 47961, 47524, 47089, 46656, 46225, 45796, 45369, 44944, 44521, 44100, 43681,

  38.     43264, 42849, 42436, 42025, 41616, 41209, 40804, 40401, 40000, 39601, 39204, 38809, 38416, 38025, 37636, 37249,

  39.     36864, 36481, 36100, 35721, 35344, 34969, 34596, 34225, 33856, 33489, 33124, 32761, 32400, 32041, 31684, 31329,

  40.     30976, 30625, 30276, 29929, 29584, 29241, 28900, 28561, 28224, 27889, 27556, 27225, 26896, 26569, 26244, 25921,

  41.     25600, 25281, 24964, 24649, 24336, 24025, 23716, 23409, 23104, 22801, 22500, 22201, 21904, 21609, 21316, 21025,

  42.     20736, 20449, 20164, 19881, 19600, 19321, 19044, 18769, 18496, 18225, 17956, 17689, 17424, 17161, 16900, 16641,

  43.     16384, 16129, 15876, 15625, 15376, 15129, 14884, 14641, 14400, 14161, 13924, 13689, 13456, 13225, 12996, 12769,

  44.     12544, 12321, 12100, 11881, 11664, 11449, 11236, 11025, 10816, 10609, 10404, 10201, 10000,  9801,  9604,  9409,

  45.      9216,  9025,  8836,  8649,  8464,  8281,  8100,  7921,  7744,  7569,  7396,  7225,  7056,  6889,  6724,  6561,

  46.      6400,  6241,  6084,  5929,  5776,  5625,  5476,  5329,  5184,  5041,  4900,  4761,  4624,  4489,  4356,  4225,

  47.      4096,  3969,  3844,  3721,  3600,  3481,  3364,  3249,  3136,  3025,  2916,  2809,  2704,  2601,  2500,  2401,

  48.      2304,  2209,  2116,  2025,  1936,  1849,  1764,  1681,  1600,  1521,  1444,  1369,  1296,  1225,  1156,  1089,

  49.      1024,   961,   900,   841,   784,   729,   676,   625,   576,   529,   484,   441,   400,   361,   324,   289,

  50.       256,   225,   196,   169,   144,   121,   100,    81,    64,    49,    36,    25,    16,     9,     4,     1,

  51.         0,     1,     4,     9,    16,    25,    36,    49,    64,    81,   100,   121,   144,   169,   196,   225,

  52.       256,   289,   324,   361,   400,   441,   484,   529,   576,   625,   676,   729,   784,   841,   900,   961,

  53.      1024,  1089,  1156,  1225,  1296,  1369,  1444,  1521,  1600,  1681,  1764,  1849,  1936,  2025,  2116,  2209,

  54.      2304,  2401,  2500,  2601,  2704,  2809,  2916,  3025,  3136,  3249,  3364,  3481,  3600,  3721,  3844,  3969,

  55.      4096,  4225,  4356,  4489,  4624,  4761,  4900,  5041,  5184,  5329,  5476,  5625,  5776,  5929,  6084,  6241,

  56.      6400,  6561,  6724,  6889,  7056,  7225,  7396,  7569,  7744,  7921,  8100,  8281,  8464,  8649,  8836,  9025,

  57.      9216,  9409,  9604,  9801, 10000, 10201, 10404, 10609, 10816, 11025, 11236, 11449, 11664, 11881, 12100, 12321,

  58.     12544, 12769, 12996, 13225, 13456, 13689, 13924, 14161, 14400, 14641, 14884, 15129, 15376, 15625, 15876, 16129,

  59.     16384, 16641, 16900, 17161, 17424, 17689, 17956, 18225, 18496, 18769, 19044, 19321, 19600, 19881, 20164, 20449,

  60.     20736, 21025, 21316, 21609, 21904, 22201, 22500, 22801, 23104, 23409, 23716, 24025, 24336, 24649, 24964, 25281,

  61.     25600, 25921, 26244, 26569, 26896, 27225, 27556, 27889, 28224, 28561, 28900, 29241, 29584, 29929, 30276, 30625,

  62.     30976, 31329, 31684, 32041, 32400, 32761, 33124, 33489, 33856, 34225, 34596, 34969, 35344, 35721, 36100, 36481,

  63.     36864, 37249, 37636, 38025, 38416, 38809, 39204, 39601, 40000, 40401, 40804, 41209, 41616, 42025, 42436, 42849,

  64.     43264, 43681, 44100, 44521, 44944, 45369, 45796, 46225, 46656, 47089, 47524, 47961, 48400, 48841, 49284, 49729,

  65.     50176, 50625, 51076, 51529, 51984, 52441, 52900, 53361, 53824, 54289, 54756, 55225, 55696, 56169, 56644, 57121,

  66.     57600, 58081, 58564, 59049, 59536, 60025, 60516, 61009, 61504, 62001, 62500, 63001, 63504, 64009, 64516, 65025,

  67. };

  68. 

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

  70.                   ptrdiff_t stride, int h)

  71. {

  72.     int s = 0, i;

  73.     const uint32_t *sq = ff_square_tab + 256;

  74. 

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

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

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

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

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

  80.         pix1 += stride;

  81.         pix2 += stride;

  82.     }

  83.     return s;

  84. }

  85. 

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

  87.                   ptrdiff_t stride, int h)

  88. {

  89.     int s = 0, i;

  90.     const uint32_t *sq = ff_square_tab + 256;

  91. 

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

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

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

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

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

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

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

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

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

  101.         pix1 += stride;

  102.         pix2 += stride;

  103.     }

  104.     return s;

  105. }

  106. 

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

  108.                    ptrdiff_t stride, int h)

  109. {

  110.     int s = 0, i;

  111.     const uint32_t *sq = ff_square_tab + 256;

  112. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  130. 

  131.         pix1 += stride;

  132.         pix2 += stride;

  133.     }

  134.     return s;

  135. }

  136. 

  137. static int sum_abs_dctelem_c(int16_t *block)

  138. {

  139.     int sum = 0, i;

  140. 

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

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

  143.     return sum;

  144. }

  145. 

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

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

  148. 

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

  150.                               ptrdiff_t stride, int h)

  151. {

  152.     int s = 0, i;

  153. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  171.         pix1 += stride;

  172.         pix2 += stride;

  173.     }

  174.     return s;

  175. }

  176. 

  177. static inline int pix_median_abs16_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  178.                              ptrdiff_t stride, int h)

  179. {

  180.     int s = 0, i, j;

  181. 

  182. #define V(x) (pix1[x] - pix2[x])

  183. 

  184.     s    += abs(V(0));

  185.     s    += abs(V(1) - V(0));

  186.     s    += abs(V(2) - V(1));

  187.     s    += abs(V(3) - V(2));

  188.     s    += abs(V(4) - V(3));

  189.     s    += abs(V(5) - V(4));

  190.     s    += abs(V(6) - V(5));

  191.     s    += abs(V(7) - V(6));

  192.     s    += abs(V(8) - V(7));

  193.     s    += abs(V(9) - V(8));

  194.     s    += abs(V(10) - V(9));

  195.     s    += abs(V(11) - V(10));

  196.     s    += abs(V(12) - V(11));

  197.     s    += abs(V(13) - V(12));

  198.     s    += abs(V(14) - V(13));

  199.     s    += abs(V(15) - V(14));

  200. 

  201.     pix1 += stride;

  202.     pix2 += stride;

  203. 

  204.     for (i = 1; i < h; i++) {

  205.         s    += abs(V(0) - V(-stride));

  206.         for (j = 1; j < 16; j++)

  207.             s    += abs(V(j) - mid_pred(V(j-stride), V(j-1), V(j-stride) + V(j-1) - V(j-stride-1)));

  208.         pix1 += stride;

  209.         pix2 += stride;

  210. 

  211.     }

  212. #undef V

  213.     return s;

  214. }

  215. 

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

  217.                           ptrdiff_t stride, int h)

  218. {

  219.     int s = 0, i;

  220. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  238.         pix1 += stride;

  239.         pix2 += stride;

  240.     }

  241.     return s;

  242. }

  243. 

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

  245.                           ptrdiff_t stride, int h)

  246. {

  247.     int s = 0, i;

  248.     uint8_t *pix3 = pix2 + stride;

  249. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  267.         pix1 += stride;

  268.         pix2 += stride;

  269.         pix3 += stride;

  270.     }

  271.     return s;

  272. }

  273. 

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

  275.                            ptrdiff_t stride, int h)

  276. {

  277.     int s = 0, i;

  278.     uint8_t *pix3 = pix2 + stride;

  279. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  297.         pix1 += stride;

  298.         pix2 += stride;

  299.         pix3 += stride;

  300.     }

  301.     return s;

  302. }

  303. 

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

  305.                              ptrdiff_t stride, int h)

  306. {

  307.     int s = 0, i;

  308. 

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

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

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

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

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

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

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

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

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

  318.         pix1 += stride;

  319.         pix2 += stride;

  320.     }

  321.     return s;

  322. }

  323. 

  324. static inline int pix_median_abs8_c(MpegEncContext *v, uint8_t *pix1, uint8_t *pix2,

  325.                              ptrdiff_t stride, int h)

  326. {

  327.     int s = 0, i, j;

  328. 

  329. #define V(x) (pix1[x] - pix2[x])

  330. 

  331.     s    += abs(V(0));

  332.     s    += abs(V(1) - V(0));

  333.     s    += abs(V(2) - V(1));

  334.     s    += abs(V(3) - V(2));

  335.     s    += abs(V(4) - V(3));

  336.     s    += abs(V(5) - V(4));

  337.     s    += abs(V(6) - V(5));

  338.     s    += abs(V(7) - V(6));

  339. 

  340.     pix1 += stride;

  341.     pix2 += stride;

  342. 

  343.     for (i = 1; i < h; i++) {

  344.         s    += abs(V(0) - V(-stride));

  345.         for (j = 1; j < 8; j++)

  346.             s    += abs(V(j) - mid_pred(V(j-stride), V(j-1), V(j-stride) + V(j-1) - V(j-stride-1)));

  347.         pix1 += stride;

  348.         pix2 += stride;

  349. 

  350.     }

  351. #undef V

  352.     return s;

  353. }

  354. 

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

  356.                          ptrdiff_t stride, int h)

  357. {

  358.     int s = 0, i;

  359. 

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

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

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

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

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

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

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

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

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

  369.         pix1 += stride;

  370.         pix2 += stride;

  371.     }

  372.     return s;

  373. }

  374. 

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

  376.                          ptrdiff_t stride, int h)

  377. {

  378.     int s = 0, i;

  379.     uint8_t *pix3 = pix2 + stride;

  380. 

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

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

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

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

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

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

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

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

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

  390.         pix1 += stride;

  391.         pix2 += stride;

  392.         pix3 += stride;

  393.     }

  394.     return s;

  395. }

  396. 

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

  398.                           ptrdiff_t stride, int h)

  399. {

  400.     int s = 0, i;

  401.     uint8_t *pix3 = pix2 + stride;

  402. 

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

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

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

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

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

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

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

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

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

  412.         pix1 += stride;

  413.         pix2 += stride;

  414.         pix3 += stride;

  415.     }

  416.     return s;

  417. }

  418. 

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

  420.                     ptrdiff_t stride, int h)

  421. {

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

  423. 

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

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

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

  427.         if (y + 1 < h) {

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

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

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

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

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

  433.         }

  434.         s1 += stride;

  435.         s2 += stride;

  436.     }

  437. 

  438.     if (c)

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

  440.     else

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

  442. }

  443. 

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

  445.                    ptrdiff_t stride, int h)

  446. {

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

  448. 

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

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

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

  452.         if (y + 1 < h) {

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

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

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

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

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

  458.         }

  459.         s1 += stride;

  460.         s2 += stride;

  461.     }

  462. 

  463.     if (c)

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

  465.     else

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

  467. }

  468. 

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

  470.                     ptrdiff_t stride, int h)

  471. {

  472.     return 0;

  473. }

  474. 

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

  476. {

  477.     int i;

  478. 

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

  480. 

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

  482.         switch (type & 0xFF) {

  483.         case FF_CMP_SAD:

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

  485.             break;

  486.         case FF_CMP_MEDIAN_SAD:

  487.             cmp[i] = c->median_sad[i];

  488.             break;

  489.         case FF_CMP_SATD:

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

  491.             break;

  492.         case FF_CMP_SSE:

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

  494.             break;

  495.         case FF_CMP_DCT:

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

  497.             break;

  498.         case FF_CMP_DCT264:

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

  500.             break;

  501.         case FF_CMP_DCTMAX:

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

  503.             break;

  504.         case FF_CMP_PSNR:

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

  506.             break;

  507.         case FF_CMP_BIT:

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

  509.             break;

  510.         case FF_CMP_RD:

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

  512.             break;

  513.         case FF_CMP_VSAD:

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

  515.             break;

  516.         case FF_CMP_VSSE:

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

  518.             break;

  519.         case FF_CMP_ZERO:

  520.             cmp[i] = zero_cmp;

  521.             break;

  522.         case FF_CMP_NSSE:

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

  524.             break;

  525. #if CONFIG_DWT

  526.         case FF_CMP_W53:

  527.             cmp[i]= c->w53[i];

  528.             break;

  529.         case FF_CMP_W97:

  530.             cmp[i]= c->w97[i];

  531.             break;

  532. #endif

  533.         default:

  534.             av_log(NULL, AV_LOG_ERROR,

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

  536.         }

  537.     }

  538. }

  539. 

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

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

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

  543. 

  544. #define BUTTERFLY1(x, y)                        \

  545.     {                                           \

  546.         int a, b;                               \

  547.         a = x;                                  \

  548.         b = y;                                  \

  549.         x = a + b;                              \

  550.         y = a - b;                              \

  551.     }

  552. 

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

  554. 

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

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

  557. {

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

  559. 

  560.     av_assert2(h == 8);

  561. 

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

  563.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

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

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

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

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

  576. 

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

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

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

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

  581. 

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

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

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

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

  586.     }

  587. 

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

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

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

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

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

  593. 

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

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

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

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

  598. 

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

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

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

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

  603.     }

  604.     return sum;

  605. }

  606. 

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

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

  609. {

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

  611. 

  612.     av_assert2(h == 8);

  613. 

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

  615.         // FIXME: try pointer walks

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

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

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

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

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

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

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

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

  624. 

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

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

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

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

  629. 

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

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

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

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

  634.     }

  635. 

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

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

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

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

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

  641. 

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

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

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

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

  646. 

  647.         sum +=

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

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

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

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

  652.     }

  653. 

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

  655. 

  656.     return sum;

  657. }

  658. 

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

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

  661. {

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

  663. 

  664.     av_assert2(h == 8);

  665. 

  666.     s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);

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

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

  669. }

  670. 

  671. #if CONFIG_GPL

  672. #define DCT8_1D                                         \

  673.     {                                                   \

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

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

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

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

  678.         const int a0  = s07 + s34;                      \

  679.         const int a1  = s16 + s25;                      \

  680.         const int a2  = s07 - s34;                      \

  681.         const int a3  = s16 - s25;                      \

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  698.     }

  699. 

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

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

  702. {

  703.     int16_t dct[8][8];

  704.     int i, sum = 0;

  705. 

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

  707. 

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

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

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

  711.         DCT8_1D

  712. #undef SRC

  713. #undef DST

  714. 

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

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

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

  718.             DCT8_1D

  719. #undef SRC

  720. #undef DST

  721.             return sum;

  722. }

  723. #endif

  724. 

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

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

  727. {

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

  729.     int sum = 0, i;

  730. 

  731.     av_assert2(h == 8);

  732. 

  733.     s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);

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

  735. 

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

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

  738. 

  739.     return sum;

  740. }

  741. 

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

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

  744. {

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

  746.     int16_t *const bak = temp + 64;

  747.     int sum = 0, i;

  748. 

  749.     av_assert2(h == 8);

  750.     s->mb_intra = 0;

  751. 

  752.     s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);

  753. 

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

  755. 

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

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

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

  759.     ff_simple_idct_int16_8bit(temp); // FIXME

  760. 

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

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

  763. 

  764.     return sum;

  765. }

  766. 

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

  768.                    ptrdiff_t stride, int h)

  769. {

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

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

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

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

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

  775.     const int esc_length = s->ac_esc_length;

  776.     uint8_t *length, *last_length;

  777. 

  778.     av_assert2(h == 8);

  779. 

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

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

  782. 

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

  784. 

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

  786.     last                               =

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

  788. 

  789.     bits = 0;

  790. 

  791.     if (s->mb_intra) {

  792.         start_i     = 1;

  793.         length      = s->intra_ac_vlc_length;

  794.         last_length = s->intra_ac_vlc_last_length;

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

  796.     } else {

  797.         start_i     = 0;

  798.         length      = s->inter_ac_vlc_length;

  799.         last_length = s->inter_ac_vlc_last_length;

  800.     }

  801. 

  802.     if (last >= start_i) {

  803.         run = 0;

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

  805.             int j = scantable[i];

  806.             level = temp[j];

  807. 

  808.             if (level) {

  809.                 level += 64;

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

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

  812.                 else

  813.                     bits += esc_length;

  814.                 run = 0;

  815.             } else

  816.                 run++;

  817.         }

  818.         i = scantable[last];

  819. 

  820.         level = temp[i] + 64;

  821. 

  822.         av_assert2(level - 64);

  823. 

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

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

  826.         } else

  827.             bits += esc_length;

  828.     }

  829. 

  830.     if (last >= 0) {

  831.         if (s->mb_intra)

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

  833.         else

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

  835.     }

  836. 

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

  838. 

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

  840. 

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

  842. }

  843. 

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

  845.                     ptrdiff_t stride, int h)

  846. {

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

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

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

  850.     const int esc_length = s->ac_esc_length;

  851.     uint8_t *length, *last_length;

  852. 

  853.     av_assert2(h == 8);

  854. 

  855.     s->pdsp.diff_pixels_unaligned(temp, src1, src2, stride);

  856. 

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

  858.     last                               =

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

  860. 

  861.     bits = 0;

  862. 

  863.     if (s->mb_intra) {

  864.         start_i     = 1;

  865.         length      = s->intra_ac_vlc_length;

  866.         last_length = s->intra_ac_vlc_last_length;

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

  868.     } else {

  869.         start_i     = 0;

  870.         length      = s->inter_ac_vlc_length;

  871.         last_length = s->inter_ac_vlc_last_length;

  872.     }

  873. 

  874.     if (last >= start_i) {

  875.         run = 0;

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

  877.             int j = scantable[i];

  878.             level = temp[j];

  879. 

  880.             if (level) {

  881.                 level += 64;

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

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

  884.                 else

  885.                     bits += esc_length;

  886.                 run = 0;

  887.             } else

  888.                 run++;

  889.         }

  890.         i = scantable[last];

  891. 

  892.         level = temp[i] + 64;

  893. 

  894.         av_assert2(level - 64);

  895. 

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

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

  898.         else

  899.             bits += esc_length;

  900.     }

  901. 

  902.     return bits;

  903. }

  904. 

  905. #define VSAD_INTRA(size)                                                \

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

  907.                                     uint8_t *s, uint8_t *dummy,         \

  908.                                     ptrdiff_t stride, int h)            \

  909. {                                                                       \

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

  911.                                                                         \

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

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

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

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

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

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

  918.         }                                                               \

  919.         s += stride;                                                    \

  920.     }                                                                   \

  921.                                                                         \

  922.     return score;                                                       \

  923. }

  924. VSAD_INTRA(8)

  925. VSAD_INTRA(16)

  926. 

  927. #define VSAD(size)                                                             \

  928. static int vsad ## size ## _c(MpegEncContext *c,                               \

  929.                               uint8_t *s1, uint8_t *s2,                        \

  930.                               ptrdiff_t stride, int h)                               \

  931. {                                                                              \

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

  933.                                                                                \

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

  935.         for (x = 0; x < size; x++)                                             \

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

  937.         s1 += stride;                                                          \

  938.         s2 += stride;                                                          \

  939.     }                                                                          \

  940.                                                                                \

  941.     return score;                                                              \

  942. }

  943. VSAD(8)

  944. VSAD(16)

  945. 

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

  947. #define VSSE_INTRA(size)                                                \

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

  949.                                     uint8_t *s, uint8_t *dummy,         \

  950.                                     ptrdiff_t stride, int h)            \

  951. {                                                                       \

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

  953.                                                                         \

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

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

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

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

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

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

  960.         }                                                               \

  961.         s += stride;                                                    \

  962.     }                                                                   \

  963.                                                                         \

  964.     return score;                                                       \

  965. }

  966. VSSE_INTRA(8)

  967. VSSE_INTRA(16)

  968. 

  969. #define VSSE(size)                                                             \

  970. static int vsse ## size ## _c(MpegEncContext *c, uint8_t *s1, uint8_t *s2,     \

  971.                               ptrdiff_t stride, int h)                         \

  972. {                                                                              \

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

  974.                                                                                \

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

  976.         for (x = 0; x < size; x++)                                             \

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

  978.         s1 += stride;                                                          \

  979.         s2 += stride;                                                          \

  980.     }                                                                          \

  981.                                                                                \

  982.     return score;                                                              \

  983. }

  984. VSSE(8)

  985. VSSE(16)

  986. 

  987. #define WRAPPER8_16_SQ(name8, name16)                                   \

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

  989.                   ptrdiff_t stride, int h)                              \

  990. {                                                                       \

  991.     int score = 0;                                                      \

  992.                                                                         \

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

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

  995.     if (h == 16) {                                                      \

  996.         dst   += 8 * stride;                                            \

  997.         src   += 8 * stride;                                            \

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

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

  1000.     }                                                                   \

  1001.     return score;                                                       \

  1002. }

  1003. 

  1004. WRAPPER8_16_SQ(hadamard8_diff8x8_c, hadamard8_diff16_c)

  1005. WRAPPER8_16_SQ(hadamard8_intra8x8_c, hadamard8_intra16_c)

  1006. WRAPPER8_16_SQ(dct_sad8x8_c, dct_sad16_c)

  1007. #if CONFIG_GPL

  1008. WRAPPER8_16_SQ(dct264_sad8x8_c, dct264_sad16_c)

  1009. #endif

  1010. WRAPPER8_16_SQ(dct_max8x8_c, dct_max16_c)

  1011. WRAPPER8_16_SQ(quant_psnr8x8_c, quant_psnr16_c)

  1012. WRAPPER8_16_SQ(rd8x8_c, rd16_c)

  1013. WRAPPER8_16_SQ(bit8x8_c, bit16_c)

  1014. 

  1015. int ff_check_alignment(void)

  1016. {

  1017.     static int did_fail = 0;

  1018.     LOCAL_ALIGNED_16(int, aligned, [4]);

  1019. 

  1020.     if ((intptr_t)aligned & 15) {

  1021.         if (!did_fail) {

  1022. #if HAVE_MMX || HAVE_ALTIVEC

  1023.             av_log(NULL, AV_LOG_ERROR,

  1024.                 "Compiler did not align stack variables. Libavcodec has been miscompiled\n"

  1025.                 "and may be very slow or crash. This is not a bug in libavcodec,\n"

  1026.                 "but in the compiler. You may try recompiling using gcc >= 4.2.\n"

  1027.                 "Do not report crashes to FFmpeg developers.\n");

  1028. #endif

  1029.             did_fail=1;

  1030.         }

  1031.         return -1;

  1032.     }

  1033.     return 0;

  1034. }

  1035. 

  1036. av_cold void ff_me_cmp_init(MECmpContext *c, AVCodecContext *avctx)

  1037. {

  1038.     ff_check_alignment();

  1039. 

  1040.     c->sum_abs_dctelem = sum_abs_dctelem_c;

  1041. 

  1042.     /* TODO [0] 16  [1] 8 */

  1043.     c->pix_abs[0][0] = pix_abs16_c;

  1044.     c->pix_abs[0][1] = pix_abs16_x2_c;

  1045.     c->pix_abs[0][2] = pix_abs16_y2_c;

  1046.     c->pix_abs[0][3] = pix_abs16_xy2_c;

  1047.     c->pix_abs[1][0] = pix_abs8_c;

  1048.     c->pix_abs[1][1] = pix_abs8_x2_c;

  1049.     c->pix_abs[1][2] = pix_abs8_y2_c;

  1050.     c->pix_abs[1][3] = pix_abs8_xy2_c;

  1051. 

  1052. #define SET_CMP_FUNC(name)                      \

  1053.     c->name[0] = name ## 16_c;                  \

  1054.     c->name[1] = name ## 8x8_c;

  1055. 

  1056.     SET_CMP_FUNC(hadamard8_diff)

  1057.     c->hadamard8_diff[4] = hadamard8_intra16_c;

  1058.     c->hadamard8_diff[5] = hadamard8_intra8x8_c;

  1059.     SET_CMP_FUNC(dct_sad)

  1060.     SET_CMP_FUNC(dct_max)

  1061. #if CONFIG_GPL

  1062.     SET_CMP_FUNC(dct264_sad)

  1063. #endif

  1064.     c->sad[0] = pix_abs16_c;

  1065.     c->sad[1] = pix_abs8_c;

  1066.     c->sse[0] = sse16_c;

  1067.     c->sse[1] = sse8_c;

  1068.     c->sse[2] = sse4_c;

  1069.     SET_CMP_FUNC(quant_psnr)

  1070.     SET_CMP_FUNC(rd)

  1071.     SET_CMP_FUNC(bit)

  1072.     c->vsad[0] = vsad16_c;

  1073.     c->vsad[1] = vsad8_c;

  1074.     c->vsad[4] = vsad_intra16_c;

  1075.     c->vsad[5] = vsad_intra8_c;

  1076.     c->vsse[0] = vsse16_c;

  1077.     c->vsse[1] = vsse8_c;

  1078.     c->vsse[4] = vsse_intra16_c;

  1079.     c->vsse[5] = vsse_intra8_c;

  1080.     c->nsse[0] = nsse16_c;

  1081.     c->nsse[1] = nsse8_c;

  1082. #if CONFIG_SNOW_DECODER || CONFIG_SNOW_ENCODER

  1083.     ff_dsputil_init_dwt(c);

  1084. #endif

  1085. 

  1086.     if (ARCH_ALPHA)

  1087.         ff_me_cmp_init_alpha(c, avctx);

  1088.     if (ARCH_ARM)

  1089.         ff_me_cmp_init_arm(c, avctx);

  1090.     if (ARCH_PPC)

  1091.         ff_me_cmp_init_ppc(c, avctx);

  1092.     if (ARCH_X86)

  1093.         ff_me_cmp_init_x86(c, avctx);

  1094.     if (ARCH_MIPS)

  1095.         ff_me_cmp_init_mips(c, avctx);

  1096. 

  1097.     c->median_sad[0] = pix_median_abs16_c;

  1098.     c->median_sad[1] = pix_median_abs8_c;

  1099. }