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.
509 Commits
32 Branches
325MB
Tree: ba6802debf
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ba6802debf'
${ noResults }
FFmpeg/libavcodec/dsputil.c

987 lines
40KB
Raw Blame History 

  1. /*

  2.  * DSP utils

  3.  * Copyright (c) 2000, 2001 Gerard Lantau.

  4.  *

  5.  * This program is free software; you can redistribute it and/or modify

  6.  * it under the terms of the GNU General Public License as published by

  7.  * the Free Software Foundation; either version 2 of the License, or

  8.  * (at your option) any later version.

  9.  *

  10.  * This program is distributed in the hope that it will be useful,

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

  12.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the

  13.  * GNU General Public License for more details.

  14.  *

  15.  * You should have received a copy of the GNU General Public License

  16.  * along with this program; if not, write to the Free Software

  17.  * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.

  18.  *

  19.  * gmc & q-pel support by Michael Niedermayer <michaelni@gmx.at>

  20.  */

  21. #include <stdlib.h>

  22. #include <stdio.h>

  23. #include <math.h>

  24. #include "avcodec.h"

  25. #include "dsputil.h"

  26. #include "simple_idct.h"

  27. 

  28. void (*ff_idct)(DCTELEM *block);

  29. void (*get_pixels)(DCTELEM *block, const UINT8 *pixels, int line_size);

  30. void (*put_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);

  31. void (*add_pixels_clamped)(const DCTELEM *block, UINT8 *pixels, int line_size);

  32. void (*gmc1)(UINT8 *dst, UINT8 *src, int srcStride, int h, int x16, int y16, int rounder);

  33. 

  34. op_pixels_abs_func pix_abs16x16;

  35. op_pixels_abs_func pix_abs16x16_x2;

  36. op_pixels_abs_func pix_abs16x16_y2;

  37. op_pixels_abs_func pix_abs16x16_xy2;

  38. 

  39. op_pixels_abs_func pix_abs8x8;

  40. op_pixels_abs_func pix_abs8x8_x2;

  41. op_pixels_abs_func pix_abs8x8_y2;

  42. op_pixels_abs_func pix_abs8x8_xy2;

  43. 

  44. UINT8 cropTbl[256 + 2 * MAX_NEG_CROP];

  45. UINT32 squareTbl[512];

  46. 

  47. extern UINT16 default_intra_matrix[64];

  48. extern UINT16 default_non_intra_matrix[64];

  49. 

  50. UINT8 zigzag_direct[64] = {

  51.     0, 1, 8, 16, 9, 2, 3, 10,

  52.     17, 24, 32, 25, 18, 11, 4, 5,

  53.     12, 19, 26, 33, 40, 48, 41, 34,

  54.     27, 20, 13, 6, 7, 14, 21, 28,

  55.     35, 42, 49, 56, 57, 50, 43, 36,

  56.     29, 22, 15, 23, 30, 37, 44, 51,

  57.     58, 59, 52, 45, 38, 31, 39, 46,

  58.     53, 60, 61, 54, 47, 55, 62, 63

  59. };

  60. 

  61. /* not permutated inverse zigzag_direct + 1 for MMX quantizer */

  62. UINT16 __align8 inv_zigzag_direct16[64];

  63. 

  64. /* not permutated zigzag_direct for MMX quantizer */

  65. UINT8 zigzag_direct_noperm[64];

  66. 

  67. UINT8 ff_alternate_horizontal_scan[64] = {

  68.     0,  1,  2,  3,  8,  9, 16, 17, 

  69.     10, 11,  4,  5,  6,  7, 15, 14,

  70.     13, 12, 19, 18, 24, 25, 32, 33, 

  71.     26, 27, 20, 21, 22, 23, 28, 29,

  72.     30, 31, 34, 35, 40, 41, 48, 49, 

  73.     42, 43, 36, 37, 38, 39, 44, 45,

  74.     46, 47, 50, 51, 56, 57, 58, 59, 

  75.     52, 53, 54, 55, 60, 61, 62, 63,

  76. };

  77. 

  78. UINT8 ff_alternate_vertical_scan[64] = {

  79.     0,  8, 16, 24,  1,  9,  2, 10, 

  80.     17, 25, 32, 40, 48, 56, 57, 49,

  81.     41, 33, 26, 18,  3, 11,  4, 12, 

  82.     19, 27, 34, 42, 50, 58, 35, 43,

  83.     51, 59, 20, 28,  5, 13,  6, 14, 

  84.     21, 29, 36, 44, 52, 60, 37, 45,

  85.     53, 61, 22, 30,  7, 15, 23, 31, 

  86.     38, 46, 54, 62, 39, 47, 55, 63,

  87. };

  88. 

  89. /* Input permutation for the simple_idct_mmx */

  90. static UINT8 simple_mmx_permutation[64]={

  91. 	0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, 

  92. 	0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, 

  93. 	0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, 

  94. 	0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, 

  95. 	0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, 

  96. 	0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, 

  97. 	0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, 

  98. 	0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  99. };

  100. 

  101. /* a*inverse[b]>>32 == a/b for all 0<=a<=65536 && 2<=b<=255 */

  102. UINT32 inverse[256]={

  103.          0, 4294967295U,2147483648U,1431655766, 1073741824,  858993460,  715827883,  613566757, 

  104.  536870912,  477218589,  429496730,  390451573,  357913942,  330382100,  306783379,  286331154, 

  105.  268435456,  252645136,  238609295,  226050911,  214748365,  204522253,  195225787,  186737709, 

  106.  178956971,  171798692,  165191050,  159072863,  153391690,  148102321,  143165577,  138547333, 

  107.  134217728,  130150525,  126322568,  122713352,  119304648,  116080198,  113025456,  110127367, 

  108.  107374183,  104755300,  102261127,   99882961,   97612894,   95443718,   93368855,   91382283, 

  109.   89478486,   87652394,   85899346,   84215046,   82595525,   81037119,   79536432,   78090315, 

  110.   76695845,   75350304,   74051161,   72796056,   71582789,   70409300,   69273667,   68174085, 

  111.   67108864,   66076420,   65075263,   64103990,   63161284,   62245903,   61356676,   60492498, 

  112.   59652324,   58835169,   58040099,   57266231,   56512728,   55778797,   55063684,   54366675, 

  113.   53687092,   53024288,   52377650,   51746594,   51130564,   50529028,   49941481,   49367441, 

  114.   48806447,   48258060,   47721859,   47197443,   46684428,   46182445,   45691142,   45210183, 

  115.   44739243,   44278014,   43826197,   43383509,   42949673,   42524429,   42107523,   41698712, 

  116.   41297763,   40904451,   40518560,   40139882,   39768216,   39403370,   39045158,   38693400, 

  117.   38347923,   38008561,   37675152,   37347542,   37025581,   36709123,   36398028,   36092163, 

  118.   35791395,   35495598,   35204650,   34918434,   34636834,   34359739,   34087043,   33818641, 

  119.   33554432,   33294321,   33038210,   32786010,   32537632,   32292988,   32051995,   31814573, 

  120.   31580642,   31350127,   31122952,   30899046,   30678338,   30460761,   30246249,   30034737, 

  121.   29826162,   29620465,   29417585,   29217465,   29020050,   28825284,   28633116,   28443493, 

  122.   28256364,   28071682,   27889399,   27709467,   27531842,   27356480,   27183338,   27012373, 

  123.   26843546,   26676816,   26512144,   26349493,   26188825,   26030105,   25873297,   25718368, 

  124.   25565282,   25414008,   25264514,   25116768,   24970741,   24826401,   24683721,   24542671, 

  125.   24403224,   24265352,   24129030,   23994231,   23860930,   23729102,   23598722,   23469767, 

  126.   23342214,   23216040,   23091223,   22967740,   22845571,   22724695,   22605092,   22486740, 

  127.   22369622,   22253717,   22139007,   22025474,   21913099,   21801865,   21691755,   21582751, 

  128.   21474837,   21367997,   21262215,   21157475,   21053762,   20951060,   20849356,   20748635, 

  129.   20648882,   20550083,   20452226,   20355296,   20259280,   20164166,   20069941,   19976593, 

  130.   19884108,   19792477,   19701685,   19611723,   19522579,   19434242,   19346700,   19259944, 

  131.   19173962,   19088744,   19004281,   18920561,   18837576,   18755316,   18673771,   18592933, 

  132.   18512791,   18433337,   18354562,   18276457,   18199014,   18122225,   18046082,   17970575, 

  133.   17895698,   17821442,   17747799,   17674763,   17602325,   17530479,   17459217,   17388532, 

  134.   17318417,   17248865,   17179870,   17111424,   17043522,   16976156,   16909321,   16843010,

  135. };

  136. 

  137. /* used to skip zeros at the end */

  138. UINT8 zigzag_end[64];

  139. 

  140. UINT8 permutation[64];

  141. //UINT8 invPermutation[64];

  142. 

  143. static void build_zigzag_end()

  144. {

  145.     int lastIndex;

  146.     int lastIndexAfterPerm=0;

  147.     for(lastIndex=0; lastIndex<64; lastIndex++)

  148.     {

  149.         if(zigzag_direct[lastIndex] > lastIndexAfterPerm) 

  150.             lastIndexAfterPerm= zigzag_direct[lastIndex];

  151.         zigzag_end[lastIndex]= lastIndexAfterPerm + 1;

  152.     }

  153. }

  154. 

  155. void get_pixels_c(DCTELEM *block, const UINT8 *pixels, int line_size)

  156. {

  157.     DCTELEM *p;

  158.     const UINT8 *pix;

  159.     int i;

  160. 

  161.     /* read the pixels */

  162.     p = block;

  163.     pix = pixels;

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

  165.         p[0] = pix[0];

  166.         p[1] = pix[1];

  167.         p[2] = pix[2];

  168.         p[3] = pix[3];

  169.         p[4] = pix[4];

  170.         p[5] = pix[5];

  171.         p[6] = pix[6];

  172.         p[7] = pix[7];

  173.         pix += line_size;

  174.         p += 8;

  175.     }

  176. }

  177. 

  178. void put_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size)

  179. {

  180.     const DCTELEM *p;

  181.     UINT8 *pix;

  182.     int i;

  183.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  184.     

  185.     /* read the pixels */

  186.     p = block;

  187.     pix = pixels;

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

  189.         pix[0] = cm[p[0]];

  190.         pix[1] = cm[p[1]];

  191.         pix[2] = cm[p[2]];

  192.         pix[3] = cm[p[3]];

  193.         pix[4] = cm[p[4]];

  194.         pix[5] = cm[p[5]];

  195.         pix[6] = cm[p[6]];

  196.         pix[7] = cm[p[7]];

  197.         pix += line_size;

  198.         p += 8;

  199.     }

  200. }

  201. 

  202. void add_pixels_clamped_c(const DCTELEM *block, UINT8 *pixels, int line_size)

  203. {

  204.     const DCTELEM *p;

  205.     UINT8 *pix;

  206.     int i;

  207.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  208.     

  209.     /* read the pixels */

  210.     p = block;

  211.     pix = pixels;

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

  213.         pix[0] = cm[pix[0] + p[0]];

  214.         pix[1] = cm[pix[1] + p[1]];

  215.         pix[2] = cm[pix[2] + p[2]];

  216.         pix[3] = cm[pix[3] + p[3]];

  217.         pix[4] = cm[pix[4] + p[4]];

  218.         pix[5] = cm[pix[5] + p[5]];

  219.         pix[6] = cm[pix[6] + p[6]];

  220.         pix[7] = cm[pix[7] + p[7]];

  221.         pix += line_size;

  222.         p += 8;

  223.     }

  224. }

  225. 

  226. #define PIXOP(BTYPE, OPNAME, OP, INCR)                                                   \

  227.                                                                                          \

  228. static void OPNAME ## _pixels(BTYPE *block, const UINT8 *pixels, int line_size, int h)    \

  229. {                                                                                        \

  230.     BTYPE *p;                                                                            \

  231.     const UINT8 *pix;                                                                    \

  232.                                                                                          \

  233.     p = block;                                                                           \

  234.     pix = pixels;                                                                        \

  235.     do {                                                                                 \

  236.         OP(p[0], pix[0]);                                                                  \

  237.         OP(p[1], pix[1]);                                                                  \

  238.         OP(p[2], pix[2]);                                                                  \

  239.         OP(p[3], pix[3]);                                                                  \

  240.         OP(p[4], pix[4]);                                                                  \

  241.         OP(p[5], pix[5]);                                                                  \

  242.         OP(p[6], pix[6]);                                                                  \

  243.         OP(p[7], pix[7]);                                                                  \

  244.         pix += line_size;                                                                \

  245.         p += INCR;                                                                       \

  246.     } while (--h);;                                                                       \

  247. }                                                                                        \

  248.                                                                                          \

  249. static void OPNAME ## _pixels_x2(BTYPE *block, const UINT8 *pixels, int line_size, int h)     \

  250. {                                                                                        \

  251.     BTYPE *p;                                                                          \

  252.     const UINT8 *pix;                                                                    \

  253.                                                                                          \

  254.     p = block;                                                                           \

  255.     pix = pixels;                                                                        \

  256.     do {                                                                   \

  257.         OP(p[0], avg2(pix[0], pix[1]));                                                    \

  258.         OP(p[1], avg2(pix[1], pix[2]));                                                    \

  259.         OP(p[2], avg2(pix[2], pix[3]));                                                    \

  260.         OP(p[3], avg2(pix[3], pix[4]));                                                    \

  261.         OP(p[4], avg2(pix[4], pix[5]));                                                    \

  262.         OP(p[5], avg2(pix[5], pix[6]));                                                    \

  263.         OP(p[6], avg2(pix[6], pix[7]));                                                    \

  264.         OP(p[7], avg2(pix[7], pix[8]));                                                    \

  265.         pix += line_size;                                                                \

  266.         p += INCR;                                                                       \

  267.     } while (--h);                                                                        \

  268. }                                                                                        \

  269.                                                                                          \

  270. static void OPNAME ## _pixels_y2(BTYPE *block, const UINT8 *pixels, int line_size, int h)     \

  271. {                                                                                        \

  272.     BTYPE *p;                                                                          \

  273.     const UINT8 *pix;                                                                    \

  274.     const UINT8 *pix1;                                                                   \

  275.                                                                                          \

  276.     p = block;                                                                           \

  277.     pix = pixels;                                                                        \

  278.     pix1 = pixels + line_size;                                                           \

  279.     do {                                                                                 \

  280.         OP(p[0], avg2(pix[0], pix1[0]));                                                   \

  281.         OP(p[1], avg2(pix[1], pix1[1]));                                                   \

  282.         OP(p[2], avg2(pix[2], pix1[2]));                                                   \

  283.         OP(p[3], avg2(pix[3], pix1[3]));                                                   \

  284.         OP(p[4], avg2(pix[4], pix1[4]));                                                   \

  285.         OP(p[5], avg2(pix[5], pix1[5]));                                                   \

  286.         OP(p[6], avg2(pix[6], pix1[6]));                                                   \

  287.         OP(p[7], avg2(pix[7], pix1[7]));                                                   \

  288.         pix += line_size;                                                                \

  289.         pix1 += line_size;                                                               \

  290.         p += INCR;                                                                       \

  291.     } while(--h);                                                                         \

  292. }                                                                                        \

  293.                                                                                          \

  294. static void OPNAME ## _pixels_xy2(BTYPE *block, const UINT8 *pixels, int line_size, int h)    \

  295. {                                                                                        \

  296.     BTYPE *p;                                                                          \

  297.     const UINT8 *pix;                                                                    \

  298.     const UINT8 *pix1;                                                                   \

  299.                                                                                          \

  300.     p = block;                                                                           \

  301.     pix = pixels;                                                                        \

  302.     pix1 = pixels + line_size;                                                           \

  303.     do {                                                                   \

  304.         OP(p[0], avg4(pix[0], pix[1], pix1[0], pix1[1]));                                  \

  305.         OP(p[1], avg4(pix[1], pix[2], pix1[1], pix1[2]));                                  \

  306.         OP(p[2], avg4(pix[2], pix[3], pix1[2], pix1[3]));                                  \

  307.         OP(p[3], avg4(pix[3], pix[4], pix1[3], pix1[4]));                                  \

  308.         OP(p[4], avg4(pix[4], pix[5], pix1[4], pix1[5]));                                  \

  309.         OP(p[5], avg4(pix[5], pix[6], pix1[5], pix1[6]));                                  \

  310.         OP(p[6], avg4(pix[6], pix[7], pix1[6], pix1[7]));                                  \

  311.         OP(p[7], avg4(pix[7], pix[8], pix1[7], pix1[8]));                                  \

  312.         pix += line_size;                                                                \

  313.         pix1 += line_size;                                                               \

  314.         p += INCR;                                                                       \

  315.     } while(--h);                                                                         \

  316. }                                                                                        \

  317.                                                                                          \

  318. void (*OPNAME ## _pixels_tab[4])(BTYPE *block, const UINT8 *pixels, int line_size, int h) = { \

  319.     OPNAME ## _pixels,                                                                   \

  320.     OPNAME ## _pixels_x2,                                                                \

  321.     OPNAME ## _pixels_y2,                                                                \

  322.     OPNAME ## _pixels_xy2,                                                               \

  323. };

  324. 

  325. 

  326. /* rounding primitives */

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

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

  329. 

  330. #define op_put(a, b) a = b

  331. #define op_avg(a, b) a = avg2(a, b)

  332. #define op_sub(a, b) a -= b

  333. 

  334. PIXOP(UINT8, put, op_put, line_size)

  335. PIXOP(UINT8, avg, op_avg, line_size)

  336. 

  337. PIXOP(DCTELEM, sub, op_sub, 8)

  338. 

  339. /* not rounding primitives */

  340. #undef avg2

  341. #undef avg4

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

  343. #define avg4(a,b,c,d) ((a+b+c+d+1)>>2)

  344. 

  345. PIXOP(UINT8, put_no_rnd, op_put, line_size)

  346. PIXOP(UINT8, avg_no_rnd, op_avg, line_size)

  347. 

  348. /* motion estimation */

  349. 

  350. #undef avg2

  351. #undef avg4

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

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

  354. 

  355. static void gmc1_c(UINT8 *dst, UINT8 *src, int srcStride, int h, int x16, int y16, int rounder)

  356. {

  357.     const int A=(16-x16)*(16-y16);

  358.     const int B=(   x16)*(16-y16);

  359.     const int C=(16-x16)*(   y16);

  360.     const int D=(   x16)*(   y16);

  361.     int i;

  362.     rounder= 128 - rounder;

  363. 

  364.     for(i=0; i<h; i++)

  365.     {

  366.         dst[0]= (A*src[0] + B*src[1] + C*src[srcStride+0] + D*src[srcStride+1] + rounder)>>8;

  367.         dst[1]= (A*src[1] + B*src[2] + C*src[srcStride+1] + D*src[srcStride+2] + rounder)>>8;

  368.         dst[2]= (A*src[2] + B*src[3] + C*src[srcStride+2] + D*src[srcStride+3] + rounder)>>8;

  369.         dst[3]= (A*src[3] + B*src[4] + C*src[srcStride+3] + D*src[srcStride+4] + rounder)>>8;

  370.         dst[4]= (A*src[4] + B*src[5] + C*src[srcStride+4] + D*src[srcStride+5] + rounder)>>8;

  371.         dst[5]= (A*src[5] + B*src[6] + C*src[srcStride+5] + D*src[srcStride+6] + rounder)>>8;

  372.         dst[6]= (A*src[6] + B*src[7] + C*src[srcStride+6] + D*src[srcStride+7] + rounder)>>8;

  373.         dst[7]= (A*src[7] + B*src[8] + C*src[srcStride+7] + D*src[srcStride+8] + rounder)>>8;

  374.         dst+= srcStride;

  375.         src+= srcStride;

  376.     }

  377. }

  378. 

  379. static void qpel_h_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int h, int r)

  380. {

  381.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  382.     int i;

  383.     for(i=0; i<h; i++)

  384.     {

  385.         dst[0]= cm[(((src[0]+src[1])*20 - (src[0]+src[2])*6 + (src[1]+src[3])*3 - (src[2]+src[4]) + r)>>5)];

  386.         dst[1]= cm[(((src[1]+src[2])*20 - (src[0]+src[3])*6 + (src[0]+src[4])*3 - (src[1]+src[5]) + r)>>5)];

  387.         dst[2]= cm[(((src[2]+src[3])*20 - (src[1]+src[4])*6 + (src[0]+src[5])*3 - (src[0]+src[6]) + r)>>5)];

  388.         dst[3]= cm[(((src[3]+src[4])*20 - (src[2]+src[5])*6 + (src[1]+src[6])*3 - (src[0]+src[7]) + r)>>5)];

  389.         dst[4]= cm[(((src[4]+src[5])*20 - (src[3]+src[6])*6 + (src[2]+src[7])*3 - (src[1]+src[8]) + r)>>5)];

  390.         dst[5]= cm[(((src[5]+src[6])*20 - (src[4]+src[7])*6 + (src[3]+src[8])*3 - (src[2]+src[8]) + r)>>5)];

  391.         dst[6]= cm[(((src[6]+src[7])*20 - (src[5]+src[8])*6 + (src[4]+src[8])*3 - (src[3]+src[7]) + r)>>5)];

  392.         dst[7]= cm[(((src[7]+src[8])*20 - (src[6]+src[8])*6 + (src[5]+src[7])*3 - (src[4]+src[6]) + r)>>5)];

  393.         dst+=dstStride;

  394.         src+=srcStride;

  395.     }

  396. }

  397. 

  398. static void qpel_v_lowpass(UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int w, int r)

  399. {

  400.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  401.     int i;

  402.     for(i=0; i<w; i++)

  403.     {

  404.         const int src0= src[0*srcStride];

  405.         const int src1= src[1*srcStride];

  406.         const int src2= src[2*srcStride];

  407.         const int src3= src[3*srcStride];

  408.         const int src4= src[4*srcStride];

  409.         const int src5= src[5*srcStride];

  410.         const int src6= src[6*srcStride];

  411.         const int src7= src[7*srcStride];

  412.         const int src8= src[8*srcStride];

  413.         dst[0*dstStride]= cm[(((src0+src1)*20 - (src0+src2)*6 + (src1+src3)*3 - (src2+src4) + r)>>5)];

  414.         dst[1*dstStride]= cm[(((src1+src2)*20 - (src0+src3)*6 + (src0+src4)*3 - (src1+src5) + r)>>5)];

  415.         dst[2*dstStride]= cm[(((src2+src3)*20 - (src1+src4)*6 + (src0+src5)*3 - (src0+src6) + r)>>5)];

  416.         dst[3*dstStride]= cm[(((src3+src4)*20 - (src2+src5)*6 + (src1+src6)*3 - (src0+src7) + r)>>5)];

  417.         dst[4*dstStride]= cm[(((src4+src5)*20 - (src3+src6)*6 + (src2+src7)*3 - (src1+src8) + r)>>5)];

  418.         dst[5*dstStride]= cm[(((src5+src6)*20 - (src4+src7)*6 + (src3+src8)*3 - (src2+src8) + r)>>5)];

  419.         dst[6*dstStride]= cm[(((src6+src7)*20 - (src5+src8)*6 + (src4+src8)*3 - (src3+src7) + r)>>5)];

  420.         dst[7*dstStride]= cm[(((src7+src8)*20 - (src6+src8)*6 + (src5+src7)*3 - (src4+src6) + r)>>5)];

  421.         dst++;

  422.         src++;

  423.     }

  424. }

  425. 

  426. static inline void put_block(UINT8 *dst, UINT8 *src, int dstStride, int srcStride)

  427. {

  428.     int i;

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

  430.     {

  431.         dst[0]= src[0];

  432.         dst[1]= src[1];

  433.         dst[2]= src[2];

  434.         dst[3]= src[3];

  435.         dst[4]= src[4];

  436.         dst[5]= src[5];

  437.         dst[6]= src[6];

  438.         dst[7]= src[7];

  439.         dst+=dstStride;

  440.         src+=srcStride;

  441.     }

  442. }

  443. 

  444. static inline void avg2_block(UINT8 *dst, UINT8 *src1, UINT8 *src2, int dstStride, int srcStride, int r)

  445. {

  446.     int i;

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

  448.     {

  449.         dst[0]= (src1[0] + src2[0] + r)>>1;

  450.         dst[1]= (src1[1] + src2[1] + r)>>1;

  451.         dst[2]= (src1[2] + src2[2] + r)>>1;

  452.         dst[3]= (src1[3] + src2[3] + r)>>1;

  453.         dst[4]= (src1[4] + src2[4] + r)>>1;

  454.         dst[5]= (src1[5] + src2[5] + r)>>1;

  455.         dst[6]= (src1[6] + src2[6] + r)>>1;

  456.         dst[7]= (src1[7] + src2[7] + r)>>1;

  457.         dst+=dstStride;

  458.         src1+=srcStride;

  459.         src2+=8;

  460.     }

  461. }

  462. 

  463. static inline void avg4_block(UINT8 *dst, UINT8 *src1, UINT8 *src2, UINT8 *src3, UINT8 *src4, int dstStride, int srcStride, int r)

  464. {

  465.     int i;

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

  467.     {

  468.         dst[0]= (src1[0] + src2[0] + src3[0] + src4[0] + r)>>2;

  469.         dst[1]= (src1[1] + src2[1] + src3[1] + src4[1] + r)>>2;

  470.         dst[2]= (src1[2] + src2[2] + src3[2] + src4[2] + r)>>2;

  471.         dst[3]= (src1[3] + src2[3] + src3[3] + src4[3] + r)>>2;

  472.         dst[4]= (src1[4] + src2[4] + src3[4] + src4[4] + r)>>2;

  473.         dst[5]= (src1[5] + src2[5] + src3[5] + src4[5] + r)>>2;

  474.         dst[6]= (src1[6] + src2[6] + src3[6] + src4[6] + r)>>2;

  475.         dst[7]= (src1[7] + src2[7] + src3[7] + src4[7] + r)>>2;

  476.         dst+=dstStride;

  477.         src1+=srcStride;

  478.         src2+=8;

  479.         src3+=8;

  480.         src4+=8;

  481.     }

  482. }

  483. 

  484. #define QPEL_MC(r, name) \

  485. static void qpel_mc00_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  486. {\

  487.     put_block(dst, src, dstStride, srcStride);\

  488. }\

  489. \

  490. static void qpel_mc10_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  491. {\

  492.     UINT8 half[64];\

  493.     qpel_h_lowpass(half, src, 8, srcStride, 8, 16-r);\

  494.     avg2_block(dst, src, half, dstStride, srcStride, 1-r);\

  495. }\

  496. \

  497. static void qpel_mc20_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  498. {\

  499.     qpel_h_lowpass(dst, src, dstStride, srcStride, 8, 16-r);\

  500. }\

  501. \

  502. static void qpel_mc30_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  503. {\

  504.     UINT8 half[64];\

  505.     qpel_h_lowpass(half, src, 8, srcStride, 8, 16-r);\

  506.     avg2_block(dst, src+1, half, dstStride, srcStride, 1-r);\

  507. }\

  508. \

  509. static void qpel_mc01_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  510. {\

  511.     UINT8 half[64];\

  512.     qpel_v_lowpass(half, src, 8, srcStride, 8, 16-r);\

  513.     avg2_block(dst, src, half, dstStride, srcStride, 1-r);\

  514. }\

  515. \

  516. static void qpel_mc02_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  517. {\

  518.     qpel_v_lowpass(dst, src, dstStride, srcStride, 8, 16-r);\

  519. }\

  520. \

  521. static void qpel_mc03_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  522. {\

  523.     UINT8 half[64];\

  524.     qpel_v_lowpass(half, src, 8, srcStride, 8, 16-r);\

  525.     avg2_block(dst, src+srcStride, half, dstStride, srcStride, 1-r);\

  526. }\

  527. static void qpel_mc11_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  528. {\

  529.     UINT8 halfH[72];\

  530.     UINT8 halfV[64];\

  531.     UINT8 halfHV[64];\

  532.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  533.     qpel_v_lowpass(halfV, src, 8, srcStride, 8, 16-r);\

  534.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  535.     avg4_block(dst, src, halfH, halfV, halfHV, dstStride, srcStride, 2-r);\

  536. }\

  537. static void qpel_mc31_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  538. {\

  539.     UINT8 halfH[72];\

  540.     UINT8 halfV[64];\

  541.     UINT8 halfHV[64];\

  542.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  543.     qpel_v_lowpass(halfV, src+1, 8, srcStride, 8, 16-r);\

  544.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  545.     avg4_block(dst, src+1, halfH, halfV, halfHV, dstStride, srcStride, 2-r);\

  546. }\

  547. static void qpel_mc13_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  548. {\

  549.     UINT8 halfH[72];\

  550.     UINT8 halfV[64];\

  551.     UINT8 halfHV[64];\

  552.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  553.     qpel_v_lowpass(halfV, src, 8, srcStride, 8, 16-r);\

  554.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  555.     avg4_block(dst, src+srcStride, halfH+8, halfV, halfHV, dstStride, srcStride, 2-r);\

  556. }\

  557. static void qpel_mc33_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  558. {\

  559.     UINT8 halfH[72];\

  560.     UINT8 halfV[64];\

  561.     UINT8 halfHV[64];\

  562.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  563.     qpel_v_lowpass(halfV, src+1, 8, srcStride, 8, 16-r);\

  564.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  565.     avg4_block(dst, src+srcStride+1, halfH+8, halfV, halfHV, dstStride, srcStride, 2-r);\

  566. }\

  567. static void qpel_mc21_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  568. {\

  569.     UINT8 halfH[72];\

  570.     UINT8 halfHV[64];\

  571.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  572.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  573.     avg2_block(dst, halfH, halfHV, dstStride, 8, 1-r);\

  574. }\

  575. static void qpel_mc23_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  576. {\

  577.     UINT8 halfH[72];\

  578.     UINT8 halfHV[64];\

  579.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  580.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  581.     avg2_block(dst, halfH+8, halfHV, dstStride, 8, 1-r);\

  582. }\

  583. static void qpel_mc12_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  584. {\

  585.     UINT8 halfH[72];\

  586.     UINT8 halfV[64];\

  587.     UINT8 halfHV[64];\

  588.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  589.     qpel_v_lowpass(halfV, src, 8, srcStride, 8, 16-r);\

  590.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  591.     avg2_block(dst, halfV, halfHV, dstStride, 8, 1-r);\

  592. }\

  593. static void qpel_mc32_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  594. {\

  595.     UINT8 halfH[72];\

  596.     UINT8 halfV[64];\

  597.     UINT8 halfHV[64];\

  598.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  599.     qpel_v_lowpass(halfV, src+1, 8, srcStride, 8, 16-r);\

  600.     qpel_v_lowpass(halfHV, halfH, 8, 8, 8, 16-r);\

  601.     avg2_block(dst, halfV, halfHV, dstStride, 8, 1-r);\

  602. }\

  603. static void qpel_mc22_c ## name (UINT8 *dst, UINT8 *src, int dstStride, int srcStride, int mx, int my)\

  604. {\

  605.     UINT8 halfH[72];\

  606.     qpel_h_lowpass(halfH, src, 8, srcStride, 9, 16-r);\

  607.     qpel_v_lowpass(dst, halfH, dstStride, 8, 8, 16-r);\

  608. }\

  609. qpel_mc_func qpel_mc ## name ## _tab[16]={ \

  610.     qpel_mc00_c ## name,                                                                   \

  611.     qpel_mc10_c ## name,                                                                   \

  612.     qpel_mc20_c ## name,                                                                   \

  613.     qpel_mc30_c ## name,                                                                   \

  614.     qpel_mc01_c ## name,                                                                   \

  615.     qpel_mc11_c ## name,                                                                   \

  616.     qpel_mc21_c ## name,                                                                   \

  617.     qpel_mc31_c ## name,                                                                   \

  618.     qpel_mc02_c ## name,                                                                   \

  619.     qpel_mc12_c ## name,                                                                   \

  620.     qpel_mc22_c ## name,                                                                   \

  621.     qpel_mc32_c ## name,                                                                   \

  622.     qpel_mc03_c ## name,                                                                   \

  623.     qpel_mc13_c ## name,                                                                   \

  624.     qpel_mc23_c ## name,                                                                   \

  625.     qpel_mc33_c ## name,                                                                   \

  626. };

  627. 

  628. QPEL_MC(0, _rnd)

  629. QPEL_MC(1, _no_rnd)

  630. 

  631. int pix_abs16x16_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  632. {

  633.     int s, i;

  634. 

  635.     s = 0;

  636.     for(i=0;i<16;i++) {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  653.         pix1 += line_size;

  654.         pix2 += line_size;

  655.     }

  656.     return s;

  657. }

  658. 

  659. int pix_abs16x16_x2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  660. {

  661.     int s, i;

  662. 

  663.     s = 0;

  664.     for(i=0;i<16;i++) {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  681.         pix1 += line_size;

  682.         pix2 += line_size;

  683.     }

  684.     return s;

  685. }

  686. 

  687. int pix_abs16x16_y2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  688. {

  689.     int s, i;

  690.     UINT8 *pix3 = pix2 + line_size;

  691. 

  692.     s = 0;

  693.     for(i=0;i<16;i++) {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  710.         pix1 += line_size;

  711.         pix2 += line_size;

  712.         pix3 += line_size;

  713.     }

  714.     return s;

  715. }

  716. 

  717. int pix_abs16x16_xy2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  718. {

  719.     int s, i;

  720.     UINT8 *pix3 = pix2 + line_size;

  721. 

  722.     s = 0;

  723.     for(i=0;i<16;i++) {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  740.         pix1 += line_size;

  741.         pix2 += line_size;

  742.         pix3 += line_size;

  743.     }

  744.     return s;

  745. }

  746. 

  747. int pix_abs8x8_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  748. {

  749.     int s, i;

  750. 

  751.     s = 0;

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

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

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

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

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

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

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

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

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

  761.         pix1 += line_size;

  762.         pix2 += line_size;

  763.     }

  764.     return s;

  765. }

  766. 

  767. int pix_abs8x8_x2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  768. {

  769.     int s, i;

  770. 

  771.     s = 0;

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

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

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

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

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

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

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

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

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

  781.         pix1 += line_size;

  782.         pix2 += line_size;

  783.     }

  784.     return s;

  785. }

  786. 

  787. int pix_abs8x8_y2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  788. {

  789.     int s, i;

  790.     UINT8 *pix3 = pix2 + line_size;

  791. 

  792.     s = 0;

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

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

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

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

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

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

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

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

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

  802.         pix1 += line_size;

  803.         pix2 += line_size;

  804.         pix3 += line_size;

  805.     }

  806.     return s;

  807. }

  808. 

  809. int pix_abs8x8_xy2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  810. {

  811.     int s, i;

  812.     UINT8 *pix3 = pix2 + line_size;

  813. 

  814.     s = 0;

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

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

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

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

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

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

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

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

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

  824.         pix1 += line_size;

  825.         pix2 += line_size;

  826.         pix3 += line_size;

  827.     }

  828.     return s;

  829. }

  830. 

  831. /* permute block according so that it corresponds to the MMX idct

  832.    order */

  833. #ifdef SIMPLE_IDCT

  834.  /* general permutation, but perhaps slightly slower */

  835. void block_permute(INT16 *block)

  836. {

  837. 	int i;

  838. 	INT16 temp[64];

  839. 

  840. 	for(i=0; i<64; i++) temp[ block_permute_op(i) ] = block[i];

  841. 

  842. 	for(i=0; i<64; i++) block[i] = temp[i];

  843. }

  844. #else

  845. 

  846. void block_permute(INT16 *block)

  847. {

  848.     int tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;

  849.     int i;

  850. 

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

  852.         tmp1 = block[1];

  853.         tmp2 = block[2];

  854.         tmp3 = block[3];

  855.         tmp4 = block[4];

  856.         tmp5 = block[5];

  857.         tmp6 = block[6];

  858.         block[1] = tmp2;

  859.         block[2] = tmp4;

  860.         block[3] = tmp6;

  861.         block[4] = tmp1;

  862.         block[5] = tmp3;

  863.         block[6] = tmp5;

  864.         block += 8;

  865.     }

  866. }

  867. #endif

  868. 

  869. void dsputil_init(void)

  870. {

  871.     int i, j;

  872.     int use_permuted_idct;

  873. 

  874.     for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;

  875.     for(i=0;i<MAX_NEG_CROP;i++) {

  876.         cropTbl[i] = 0;

  877.         cropTbl[i + MAX_NEG_CROP + 256] = 255;

  878.     }

  879. 

  880.     for(i=0;i<512;i++) {

  881.         squareTbl[i] = (i - 256) * (i - 256);

  882.     }

  883. 

  884. #ifdef SIMPLE_IDCT

  885.     ff_idct = simple_idct;

  886. #else

  887.     ff_idct = j_rev_dct;

  888. #endif

  889.     get_pixels = get_pixels_c;

  890.     put_pixels_clamped = put_pixels_clamped_c;

  891.     add_pixels_clamped = add_pixels_clamped_c;

  892.     gmc1= gmc1_c;

  893. 

  894.     pix_abs16x16     = pix_abs16x16_c;

  895.     pix_abs16x16_x2  = pix_abs16x16_x2_c;

  896.     pix_abs16x16_y2  = pix_abs16x16_y2_c;

  897.     pix_abs16x16_xy2 = pix_abs16x16_xy2_c;

  898.     pix_abs8x8     = pix_abs8x8_c;

  899.     pix_abs8x8_x2  = pix_abs8x8_x2_c;

  900.     pix_abs8x8_y2  = pix_abs8x8_y2_c;

  901.     pix_abs8x8_xy2 = pix_abs8x8_xy2_c;

  902.     av_fdct = jpeg_fdct_ifast;

  903. 

  904.     use_permuted_idct = 1;

  905. 

  906. #ifdef HAVE_MMX

  907.     dsputil_init_mmx();

  908. #endif

  909. #ifdef ARCH_ARMV4L

  910.     dsputil_init_armv4l();

  911. #endif

  912. #ifdef HAVE_MLIB

  913.     dsputil_init_mlib();

  914.     use_permuted_idct = 0;

  915. #endif

  916. #ifdef ARCH_ALPHA

  917.     dsputil_init_alpha();

  918.     use_permuted_idct = 0;

  919. #endif

  920. 

  921. #ifdef SIMPLE_IDCT

  922.     if(ff_idct == simple_idct) use_permuted_idct=0;

  923. #endif

  924. 

  925.     if(use_permuted_idct)

  926. #ifdef SIMPLE_IDCT

  927.         for(i=0; i<64; i++) permutation[i]= simple_mmx_permutation[i];

  928. #else

  929.         for(i=0; i<64; i++) permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);

  930. #endif

  931.     else

  932.         for(i=0; i<64; i++) permutation[i]=i;

  933. 

  934.     for(i=0; i<64; i++) inv_zigzag_direct16[zigzag_direct[i]]= i+1;

  935.     for(i=0; i<64; i++) zigzag_direct_noperm[i]= zigzag_direct[i];

  936.     

  937.     if (use_permuted_idct) {

  938.         /* permute for IDCT */

  939.         for(i=0;i<64;i++) {

  940.             j = zigzag_direct[i];

  941.             zigzag_direct[i] = block_permute_op(j);

  942.             j = ff_alternate_horizontal_scan[i];

  943.             ff_alternate_horizontal_scan[i] = block_permute_op(j);

  944.             j = ff_alternate_vertical_scan[i];

  945.             ff_alternate_vertical_scan[i] = block_permute_op(j);

  946.         }

  947.         block_permute(default_intra_matrix);

  948.         block_permute(default_non_intra_matrix);

  949.     }

  950.     

  951.     build_zigzag_end();

  952. }

  953. 

  954. void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],

  955.               int orig_linesize[3], int coded_linesize,

  956.               AVCodecContext *avctx)

  957. {

  958.     int quad, diff, x, y;

  959.     UINT8 *orig, *coded;

  960.     UINT32 *sq = squareTbl + 256;

  961.     

  962.     quad = 0;

  963.     diff = 0;

  964.     

  965.     /* Luminance */

  966.     orig = orig_image[0];

  967.     coded = coded_image[0];

  968.     

  969.     for (y=0;y<avctx->height;y++) {

  970.         for (x=0;x<avctx->width;x++) {

  971.             diff = *(orig + x) - *(coded + x);

  972.             quad += sq[diff];

  973.         }

  974.         orig += orig_linesize[0];

  975.         coded += coded_linesize;

  976.     }

  977.    

  978.     avctx->psnr_y = (float) quad / (float) (avctx->width * avctx->height);

  979.     

  980.     if (avctx->psnr_y) {

  981.         avctx->psnr_y = (float) (255 * 255) / avctx->psnr_y;

  982.         avctx->psnr_y = 10 * (float) log10 (avctx->psnr_y); 

  983.     } else

  984.         avctx->psnr_y = 99.99;

  985. }