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.
677 Commits
32 Branches
1.0GB
Tree: 6000abfa89
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '6000abfa89'
${ noResults }
FFmpeg/libavcodec/dsputil.c

1358 lines
51KB
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 & 32/64 bit based MC by Michael Niedermayer <michaelni@gmx.at>

  20.  */

  21. #include "avcodec.h"

  22. #include "dsputil.h"

  23. #include "simple_idct.h"

  24. 

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

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

  27. void (*diff_pixels)(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride);

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

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

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

  31. void (*clear_blocks)(DCTELEM *blocks);

  32. 

  33. op_pixels_abs_func pix_abs16x16;

  34. op_pixels_abs_func pix_abs16x16_x2;

  35. op_pixels_abs_func pix_abs16x16_y2;

  36. op_pixels_abs_func pix_abs16x16_xy2;

  37. 

  38. op_pixels_abs_func pix_abs8x8;

  39. op_pixels_abs_func pix_abs8x8_x2;

  40. op_pixels_abs_func pix_abs8x8_y2;

  41. op_pixels_abs_func pix_abs8x8_xy2;

  42. 

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

  44. UINT32 squareTbl[512];

  45. 

  46. extern UINT16 default_intra_matrix[64];

  47. extern UINT16 default_non_intra_matrix[64];

  48. extern UINT16 ff_mpeg4_default_intra_matrix[64];

  49. extern UINT16 ff_mpeg4_default_non_intra_matrix[64];

  50. 

  51. UINT8 zigzag_direct[64] = {

  52.     0, 1, 8, 16, 9, 2, 3, 10,

  53.     17, 24, 32, 25, 18, 11, 4, 5,

  54.     12, 19, 26, 33, 40, 48, 41, 34,

  55.     27, 20, 13, 6, 7, 14, 21, 28,

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

  57.     29, 22, 15, 23, 30, 37, 44, 51,

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

  59.     53, 60, 61, 54, 47, 55, 62, 63

  60. };

  61. 

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

  63. UINT16 __align8 inv_zigzag_direct16[64];

  64. 

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

  66. UINT8 zigzag_direct_noperm[64];

  67. 

  68. UINT8 ff_alternate_horizontal_scan[64] = {

  69.     0,  1,  2,  3,  8,  9, 16, 17, 

  70.     10, 11,  4,  5,  6,  7, 15, 14,

  71.     13, 12, 19, 18, 24, 25, 32, 33, 

  72.     26, 27, 20, 21, 22, 23, 28, 29,

  73.     30, 31, 34, 35, 40, 41, 48, 49, 

  74.     42, 43, 36, 37, 38, 39, 44, 45,

  75.     46, 47, 50, 51, 56, 57, 58, 59, 

  76.     52, 53, 54, 55, 60, 61, 62, 63,

  77. };

  78. 

  79. UINT8 ff_alternate_vertical_scan[64] = {

  80.     0,  8, 16, 24,  1,  9,  2, 10, 

  81.     17, 25, 32, 40, 48, 56, 57, 49,

  82.     41, 33, 26, 18,  3, 11,  4, 12, 

  83.     19, 27, 34, 42, 50, 58, 35, 43,

  84.     51, 59, 20, 28,  5, 13,  6, 14, 

  85.     21, 29, 36, 44, 52, 60, 37, 45,

  86.     53, 61, 22, 30,  7, 15, 23, 31, 

  87.     38, 46, 54, 62, 39, 47, 55, 63,

  88. };

  89. 

  90. #ifdef SIMPLE_IDCT

  91. 

  92. /* Input permutation for the simple_idct_mmx */

  93. static UINT8 simple_mmx_permutation[64]={

  94. 	0x00, 0x08, 0x04, 0x09, 0x01, 0x0C, 0x05, 0x0D, 

  95. 	0x10, 0x18, 0x14, 0x19, 0x11, 0x1C, 0x15, 0x1D, 

  96. 	0x20, 0x28, 0x24, 0x29, 0x21, 0x2C, 0x25, 0x2D, 

  97. 	0x12, 0x1A, 0x16, 0x1B, 0x13, 0x1E, 0x17, 0x1F, 

  98. 	0x02, 0x0A, 0x06, 0x0B, 0x03, 0x0E, 0x07, 0x0F, 

  99. 	0x30, 0x38, 0x34, 0x39, 0x31, 0x3C, 0x35, 0x3D, 

  100. 	0x22, 0x2A, 0x26, 0x2B, 0x23, 0x2E, 0x27, 0x2F, 

  101. 	0x32, 0x3A, 0x36, 0x3B, 0x33, 0x3E, 0x37, 0x3F,

  102. };

  103. #endif

  104. 

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

  106. UINT32 inverse[256]={

  107.          0, 4294967295U,2147483648U,1431655766, 1073741824,  858993460,  715827883,  613566757, 

  108.  536870912,  477218589,  429496730,  390451573,  357913942,  330382100,  306783379,  286331154, 

  109.  268435456,  252645136,  238609295,  226050911,  214748365,  204522253,  195225787,  186737709, 

  110.  178956971,  171798692,  165191050,  159072863,  153391690,  148102321,  143165577,  138547333, 

  111.  134217728,  130150525,  126322568,  122713352,  119304648,  116080198,  113025456,  110127367, 

  112.  107374183,  104755300,  102261127,   99882961,   97612894,   95443718,   93368855,   91382283, 

  113.   89478486,   87652394,   85899346,   84215046,   82595525,   81037119,   79536432,   78090315, 

  114.   76695845,   75350304,   74051161,   72796056,   71582789,   70409300,   69273667,   68174085, 

  115.   67108864,   66076420,   65075263,   64103990,   63161284,   62245903,   61356676,   60492498, 

  116.   59652324,   58835169,   58040099,   57266231,   56512728,   55778797,   55063684,   54366675, 

  117.   53687092,   53024288,   52377650,   51746594,   51130564,   50529028,   49941481,   49367441, 

  118.   48806447,   48258060,   47721859,   47197443,   46684428,   46182445,   45691142,   45210183, 

  119.   44739243,   44278014,   43826197,   43383509,   42949673,   42524429,   42107523,   41698712, 

  120.   41297763,   40904451,   40518560,   40139882,   39768216,   39403370,   39045158,   38693400, 

  121.   38347923,   38008561,   37675152,   37347542,   37025581,   36709123,   36398028,   36092163, 

  122.   35791395,   35495598,   35204650,   34918434,   34636834,   34359739,   34087043,   33818641, 

  123.   33554432,   33294321,   33038210,   32786010,   32537632,   32292988,   32051995,   31814573, 

  124.   31580642,   31350127,   31122952,   30899046,   30678338,   30460761,   30246249,   30034737, 

  125.   29826162,   29620465,   29417585,   29217465,   29020050,   28825284,   28633116,   28443493, 

  126.   28256364,   28071682,   27889399,   27709467,   27531842,   27356480,   27183338,   27012373, 

  127.   26843546,   26676816,   26512144,   26349493,   26188825,   26030105,   25873297,   25718368, 

  128.   25565282,   25414008,   25264514,   25116768,   24970741,   24826401,   24683721,   24542671, 

  129.   24403224,   24265352,   24129030,   23994231,   23860930,   23729102,   23598722,   23469767, 

  130.   23342214,   23216040,   23091223,   22967740,   22845571,   22724695,   22605092,   22486740, 

  131.   22369622,   22253717,   22139007,   22025474,   21913099,   21801865,   21691755,   21582751, 

  132.   21474837,   21367997,   21262215,   21157475,   21053762,   20951060,   20849356,   20748635, 

  133.   20648882,   20550083,   20452226,   20355296,   20259280,   20164166,   20069941,   19976593, 

  134.   19884108,   19792477,   19701685,   19611723,   19522579,   19434242,   19346700,   19259944, 

  135.   19173962,   19088744,   19004281,   18920561,   18837576,   18755316,   18673771,   18592933, 

  136.   18512791,   18433337,   18354562,   18276457,   18199014,   18122225,   18046082,   17970575, 

  137.   17895698,   17821442,   17747799,   17674763,   17602325,   17530479,   17459217,   17388532, 

  138.   17318417,   17248865,   17179870,   17111424,   17043522,   16976156,   16909321,   16843010,

  139. };

  140. 

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

  142. UINT8 zigzag_end[64];

  143. 

  144. UINT8 permutation[64];

  145. //UINT8 invPermutation[64];

  146. 

  147. static void build_zigzag_end()

  148. {

  149.     int lastIndex;

  150.     int lastIndexAfterPerm=0;

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

  152.     {

  153.         if(zigzag_direct[lastIndex] > lastIndexAfterPerm) 

  154.             lastIndexAfterPerm= zigzag_direct[lastIndex];

  155.         zigzag_end[lastIndex]= lastIndexAfterPerm + 1;

  156.     }

  157. }

  158. 

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

  160. {

  161.     DCTELEM *p;

  162.     const UINT8 *pix;

  163.     int i;

  164. 

  165.     /* read the pixels */

  166.     p = block;

  167.     pix = pixels;

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

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

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

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

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

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

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

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

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

  177.         pix += line_size;

  178.         p += 8;

  179.     }

  180. }

  181. 

  182. void diff_pixels_c(DCTELEM *block, const UINT8 *s1, const UINT8 *s2, int stride){

  183.     DCTELEM *p;

  184.     int i;

  185. 

  186.     /* read the pixels */

  187.     p = block;

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

  189.         p[0] = s1[0] - s2[0];

  190.         p[1] = s1[1] - s2[1];

  191.         p[2] = s1[2] - s2[2];

  192.         p[3] = s1[3] - s2[3];

  193.         p[4] = s1[4] - s2[4];

  194.         p[5] = s1[5] - s2[5];

  195.         p[6] = s1[6] - s2[6];

  196.         p[7] = s1[7] - s2[7];

  197.         s1 += stride;

  198.         s2 += stride;

  199.         p += 8;

  200.     }

  201. }

  202. 

  203. 

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

  205. {

  206.     const DCTELEM *p;

  207.     UINT8 *pix;

  208.     int i;

  209.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  210.     

  211.     /* read the pixels */

  212.     p = block;

  213.     pix = pixels;

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

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

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

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

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

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

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

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

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

  223.         pix += line_size;

  224.         p += 8;

  225.     }

  226. }

  227. 

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

  229. {

  230.     const DCTELEM *p;

  231.     UINT8 *pix;

  232.     int i;

  233.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  234.     

  235.     /* read the pixels */

  236.     p = block;

  237.     pix = pixels;

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

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

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

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

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

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

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

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

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

  247.         pix += line_size;

  248.         p += 8;

  249.     }

  250. }

  251. 

  252. //FIXME someone with a alignemtent picky cpu should change these

  253. 

  254. #define LD32(a) (*((uint32_t*)(a)))

  255. #define LD64(a) (*((uint64_t*)(a)))

  256. 

  257. #if 0

  258. 

  259. #define PIXOP2(OPNAME, OP) \

  260. void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  261. {\

  262.     int i;\

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

  264.         OP(*((uint64_t*)block), LD64(pixels));\

  265.         pixels+=line_size;\

  266.         block +=line_size;\

  267.     }\

  268. }\

  269. \

  270. void OPNAME ## _no_rnd_pixels_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  271. {\

  272.     int i;\

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

  274.         const uint64_t a= LD64(pixels  );\

  275.         const uint64_t b= LD64(pixels+1);\

  276.         OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\

  277.         pixels+=line_size;\

  278.         block +=line_size;\

  279.     }\

  280. }\

  281. \

  282. void OPNAME ## _pixels_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  283. {\

  284.     int i;\

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

  286.         const uint64_t a= LD64(pixels  );\

  287.         const uint64_t b= LD64(pixels+1);\

  288.         OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\

  289.         pixels+=line_size;\

  290.         block +=line_size;\

  291.     }\

  292. }\

  293. \

  294. void OPNAME ## _no_rnd_pixels_y2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  295. {\

  296.     int i;\

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

  298.         const uint64_t a= LD64(pixels          );\

  299.         const uint64_t b= LD64(pixels+line_size);\

  300.         OP(*((uint64_t*)block), (a&b) + (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\

  301.         pixels+=line_size;\

  302.         block +=line_size;\

  303.     }\

  304. }\

  305. \

  306. void OPNAME ## _pixels_y2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  307. {\

  308.     int i;\

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

  310.         const uint64_t a= LD64(pixels          );\

  311.         const uint64_t b= LD64(pixels+line_size);\

  312.         OP(*((uint64_t*)block), (a|b) - (((a^b)&0xFEFEFEFEFEFEFEFEULL)>>1));\

  313.         pixels+=line_size;\

  314.         block +=line_size;\

  315.     }\

  316. }\

  317. \

  318. void OPNAME ## _pixels_xy2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  319. {\

  320.         int i;\

  321.         const uint64_t a= LD64(pixels  );\

  322.         const uint64_t b= LD64(pixels+1);\

  323.         uint64_t l0=  (a&0x0303030303030303ULL)\

  324.                     + (b&0x0303030303030303ULL)\

  325.                     + 0x0202020202020202ULL;\

  326.         uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\

  327.                    + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\

  328.         uint64_t l1,h1;\

  329. \

  330.         pixels+=line_size;\

  331.         for(i=0; i<h; i+=2){\

  332.             uint64_t a= LD64(pixels  );\

  333.             uint64_t b= LD64(pixels+1);\

  334.             l1=  (a&0x0303030303030303ULL)\

  335.                + (b&0x0303030303030303ULL);\

  336.             h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\

  337.               + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\

  338.             OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\

  339.             pixels+=line_size;\

  340.             block +=line_size;\

  341.             a= LD64(pixels  );\

  342.             b= LD64(pixels+1);\

  343.             l0=  (a&0x0303030303030303ULL)\

  344.                + (b&0x0303030303030303ULL)\

  345.                + 0x0202020202020202ULL;\

  346.             h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\

  347.               + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\

  348.             OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\

  349.             pixels+=line_size;\

  350.             block +=line_size;\

  351.         }\

  352. }\

  353. \

  354. void OPNAME ## _no_rnd_pixels_xy2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  355. {\

  356.         int i;\

  357.         const uint64_t a= LD64(pixels  );\

  358.         const uint64_t b= LD64(pixels+1);\

  359.         uint64_t l0=  (a&0x0303030303030303ULL)\

  360.                     + (b&0x0303030303030303ULL)\

  361.                     + 0x0101010101010101ULL;\

  362.         uint64_t h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\

  363.                    + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\

  364.         uint64_t l1,h1;\

  365. \

  366.         pixels+=line_size;\

  367.         for(i=0; i<h; i+=2){\

  368.             uint64_t a= LD64(pixels  );\

  369.             uint64_t b= LD64(pixels+1);\

  370.             l1=  (a&0x0303030303030303ULL)\

  371.                + (b&0x0303030303030303ULL);\

  372.             h1= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\

  373.               + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\

  374.             OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\

  375.             pixels+=line_size;\

  376.             block +=line_size;\

  377.             a= LD64(pixels  );\

  378.             b= LD64(pixels+1);\

  379.             l0=  (a&0x0303030303030303ULL)\

  380.                + (b&0x0303030303030303ULL)\

  381.                + 0x0101010101010101ULL;\

  382.             h0= ((a&0xFCFCFCFCFCFCFCFCULL)>>2)\

  383.               + ((b&0xFCFCFCFCFCFCFCFCULL)>>2);\

  384.             OP(*((uint64_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0F0F0F0F0FULL));\

  385.             pixels+=line_size;\

  386.             block +=line_size;\

  387.         }\

  388. }\

  389. \

  390. void (*OPNAME ## _pixels_tab[4])(uint8_t *block, const uint8_t *pixels, int line_size, int h) = {\

  391.     OPNAME ## _pixels,\

  392.     OPNAME ## _pixels_x2,\

  393.     OPNAME ## _pixels_y2,\

  394.     OPNAME ## _pixels_xy2,\

  395. };\

  396. \

  397. void (*OPNAME ## _no_rnd_pixels_tab[4])(uint8_t *block, const uint8_t *pixels, int line_size, int h) = {\

  398.     OPNAME ## _pixels,\

  399.     OPNAME ## _no_rnd_pixels_x2,\

  400.     OPNAME ## _no_rnd_pixels_y2,\

  401.     OPNAME ## _no_rnd_pixels_xy2,\

  402. };

  403. 

  404. #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEFEFEFEFEULL)>>1) )

  405. #else // 64 bit variant

  406. 

  407. #define PIXOP2(OPNAME, OP) \

  408. void OPNAME ## _pixels(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  409. {\

  410.     int i;\

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

  412.         OP(*((uint32_t*)(block  )), LD32(pixels  ));\

  413.         OP(*((uint32_t*)(block+4)), LD32(pixels+4));\

  414.         pixels+=line_size;\

  415.         block +=line_size;\

  416.     }\

  417. }\

  418. \

  419. void OPNAME ## _no_rnd_pixels_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  420. {\

  421.     int i;\

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

  423.         int j;\

  424.         for(j=0; j<2; j++){\

  425.             const uint32_t a= LD32(pixels  );\

  426.             const uint32_t b= LD32(pixels+1);\

  427.             OP(*((uint32_t*)block), (a&b) + (((a^b)&0xFEFEFEFEUL)>>1));\

  428.             pixels+=4;\

  429.             block +=4;\

  430.         }\

  431.         pixels+=line_size-8;\

  432.         block +=line_size-8;\

  433.     }\

  434. }\

  435. \

  436. void OPNAME ## _pixels_x2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  437. {\

  438.     int i;\

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

  440.         int j;\

  441.         for(j=0; j<2; j++){\

  442.             const uint32_t a= LD32(pixels  );\

  443.             const uint32_t b= LD32(pixels+1);\

  444.             OP(*((uint32_t*)block), (a|b) - (((a^b)&0xFEFEFEFEUL)>>1));\

  445.             pixels+=4;\

  446.             block +=4;\

  447.         }\

  448.         pixels+=line_size-8;\

  449.         block +=line_size-8;\

  450.     }\

  451. }\

  452. \

  453. void OPNAME ## _no_rnd_pixels_y2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  454. {\

  455.     int i;\

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

  457.         int j;\

  458.         for(j=0; j<2; j++){\

  459.             const uint32_t a= LD32(pixels          );\

  460.             const uint32_t b= LD32(pixels+line_size);\

  461.             OP(*((uint32_t*)block), (a&b) + (((a^b)&0xFEFEFEFEUL)>>1));\

  462.             pixels+=4;\

  463.             block +=4;\

  464.         }\

  465.         pixels+=line_size-8;\

  466.         block +=line_size-8;\

  467.     }\

  468. }\

  469. \

  470. void OPNAME ## _pixels_y2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  471. {\

  472.     int i;\

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

  474.         int j;\

  475.         for(j=0; j<2; j++){\

  476.             const uint32_t a= LD32(pixels          );\

  477.             const uint32_t b= LD32(pixels+line_size);\

  478.             OP(*((uint32_t*)block), (a|b) - (((a^b)&0xFEFEFEFEUL)>>1));\

  479.             pixels+=4;\

  480.             block +=4;\

  481.         }\

  482.         pixels+=line_size-8;\

  483.         block +=line_size-8;\

  484.     }\

  485. }\

  486. \

  487. void OPNAME ## _pixels_xy2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  488. {\

  489.     int j;\

  490.     for(j=0; j<2; j++){\

  491.         int i;\

  492.         const uint32_t a= LD32(pixels  );\

  493.         const uint32_t b= LD32(pixels+1);\

  494.         uint32_t l0=  (a&0x03030303UL)\

  495.                     + (b&0x03030303UL)\

  496.                     + 0x02020202UL;\

  497.         uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\

  498.                    + ((b&0xFCFCFCFCUL)>>2);\

  499.         uint32_t l1,h1;\

  500. \

  501.         pixels+=line_size;\

  502.         for(i=0; i<h; i+=2){\

  503.             uint32_t a= LD32(pixels  );\

  504.             uint32_t b= LD32(pixels+1);\

  505.             l1=  (a&0x03030303UL)\

  506.                + (b&0x03030303UL);\

  507.             h1= ((a&0xFCFCFCFCUL)>>2)\

  508.               + ((b&0xFCFCFCFCUL)>>2);\

  509.             OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\

  510.             pixels+=line_size;\

  511.             block +=line_size;\

  512.             a= LD32(pixels  );\

  513.             b= LD32(pixels+1);\

  514.             l0=  (a&0x03030303UL)\

  515.                + (b&0x03030303UL)\

  516.                + 0x02020202UL;\

  517.             h0= ((a&0xFCFCFCFCUL)>>2)\

  518.               + ((b&0xFCFCFCFCUL)>>2);\

  519.             OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\

  520.             pixels+=line_size;\

  521.             block +=line_size;\

  522.         }\

  523.         pixels+=4-line_size*(h+1);\

  524.         block +=4-line_size*h;\

  525.     }\

  526. }\

  527. \

  528. void OPNAME ## _no_rnd_pixels_xy2(uint8_t *block, const uint8_t *pixels, int line_size, int h)\

  529. {\

  530.     int j;\

  531.     for(j=0; j<2; j++){\

  532.         int i;\

  533.         const uint32_t a= LD32(pixels  );\

  534.         const uint32_t b= LD32(pixels+1);\

  535.         uint32_t l0=  (a&0x03030303UL)\

  536.                     + (b&0x03030303UL)\

  537.                     + 0x01010101UL;\

  538.         uint32_t h0= ((a&0xFCFCFCFCUL)>>2)\

  539.                    + ((b&0xFCFCFCFCUL)>>2);\

  540.         uint32_t l1,h1;\

  541. \

  542.         pixels+=line_size;\

  543.         for(i=0; i<h; i+=2){\

  544.             uint32_t a= LD32(pixels  );\

  545.             uint32_t b= LD32(pixels+1);\

  546.             l1=  (a&0x03030303UL)\

  547.                + (b&0x03030303UL);\

  548.             h1= ((a&0xFCFCFCFCUL)>>2)\

  549.               + ((b&0xFCFCFCFCUL)>>2);\

  550.             OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\

  551.             pixels+=line_size;\

  552.             block +=line_size;\

  553.             a= LD32(pixels  );\

  554.             b= LD32(pixels+1);\

  555.             l0=  (a&0x03030303UL)\

  556.                + (b&0x03030303UL)\

  557.                + 0x01010101UL;\

  558.             h0= ((a&0xFCFCFCFCUL)>>2)\

  559.               + ((b&0xFCFCFCFCUL)>>2);\

  560.             OP(*((uint32_t*)block), h0+h1+(((l0+l1)>>2)&0x0F0F0F0FUL));\

  561.             pixels+=line_size;\

  562.             block +=line_size;\

  563.         }\

  564.         pixels+=4-line_size*(h+1);\

  565.         block +=4-line_size*h;\

  566.     }\

  567. }\

  568. \

  569. void (*OPNAME ## _pixels_tab[4])(uint8_t *block, const uint8_t *pixels, int line_size, int h) = {\

  570.     OPNAME ## _pixels,\

  571.     OPNAME ## _pixels_x2,\

  572.     OPNAME ## _pixels_y2,\

  573.     OPNAME ## _pixels_xy2,\

  574. };\

  575. \

  576. void (*OPNAME ## _no_rnd_pixels_tab[4])(uint8_t *block, const uint8_t *pixels, int line_size, int h) = {\

  577.     OPNAME ## _pixels,\

  578.     OPNAME ## _no_rnd_pixels_x2,\

  579.     OPNAME ## _no_rnd_pixels_y2,\

  580.     OPNAME ## _no_rnd_pixels_xy2,\

  581. };

  582. #define op_avg(a, b) a = ( ((a)|(b)) - ((((a)^(b))&0xFEFEFEFEUL)>>1) )

  583. #endif

  584. 

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

  586. 

  587. PIXOP2(avg, op_avg)

  588. PIXOP2(put, op_put)

  589. #undef op_avg

  590. #undef op_put

  591. 

  592. /* FIXME this stuff could be removed as its ot really used anymore */

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

  594.                                                                                          \

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

  596. {                                                                                        \

  597.     BTYPE *p;                                                                            \

  598.     const UINT8 *pix;                                                                    \

  599.                                                                                          \

  600.     p = block;                                                                           \

  601.     pix = pixels;                                                                        \

  602.     do {                                                                                 \

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

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

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

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

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

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

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

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

  611.         pix += line_size;                                                                \

  612.         p += INCR;                                                                       \

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

  614. }                                                                                        \

  615.                                                                                          \

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

  617. {                                                                                        \

  618.     BTYPE *p;                                                                          \

  619.     const UINT8 *pix;                                                                    \

  620.                                                                                          \

  621.     p = block;                                                                           \

  622.     pix = pixels;                                                                        \

  623.     do {                                                                   \

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

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

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

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

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

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

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

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

  632.         pix += line_size;                                                                \

  633.         p += INCR;                                                                       \

  634.     } while (--h);                                                                        \

  635. }                                                                                        \

  636.                                                                                          \

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

  638. {                                                                                        \

  639.     BTYPE *p;                                                                          \

  640.     const UINT8 *pix;                                                                    \

  641.     const UINT8 *pix1;                                                                   \

  642.                                                                                          \

  643.     p = block;                                                                           \

  644.     pix = pixels;                                                                        \

  645.     pix1 = pixels + line_size;                                                           \

  646.     do {                                                                                 \

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

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

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

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

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

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

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

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

  655.         pix += line_size;                                                                \

  656.         pix1 += line_size;                                                               \

  657.         p += INCR;                                                                       \

  658.     } while(--h);                                                                         \

  659. }                                                                                        \

  660.                                                                                          \

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

  662. {                                                                                        \

  663.     BTYPE *p;                                                                          \

  664.     const UINT8 *pix;                                                                    \

  665.     const UINT8 *pix1;                                                                   \

  666.                                                                                          \

  667.     p = block;                                                                           \

  668.     pix = pixels;                                                                        \

  669.     pix1 = pixels + line_size;                                                           \

  670.     do {                                                                   \

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

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

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

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

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

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

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

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

  679.         pix += line_size;                                                                \

  680.         pix1 += line_size;                                                               \

  681.         p += INCR;                                                                       \

  682.     } while(--h);                                                                         \

  683. }                                                                                        \

  684.                                                                                          \

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

  686.     OPNAME ## _pixels,                                                                   \

  687.     OPNAME ## _pixels_x2,                                                                \

  688.     OPNAME ## _pixels_y2,                                                                \

  689.     OPNAME ## _pixels_xy2,                                                               \

  690. };

  691. 

  692. 

  693. /* rounding primitives */

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

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

  696. 

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

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

  699. 

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

  701. 

  702. /* not rounding primitives */

  703. #undef avg2

  704. #undef avg4

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

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

  707. 

  708. /* motion estimation */

  709. 

  710. #undef avg2

  711. #undef avg4

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

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

  714. 

  715. /* end of removeale stuff */

  716. 

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

  718. {

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

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

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

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

  723.     int i;

  724.     rounder= 128 - rounder;

  725. 

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

  727.     {

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

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

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

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

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

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

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

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

  736.         dst+= srcStride;

  737.         src+= srcStride;

  738.     }

  739. }

  740. 

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

  742. {

  743.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  744.     int i;

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

  746.     {

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

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

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

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

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

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

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

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

  755.         dst+=dstStride;

  756.         src+=srcStride;

  757.     }

  758. }

  759. 

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

  761. {

  762.     UINT8 *cm = cropTbl + MAX_NEG_CROP;

  763.     int i;

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

  765.     {

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  783.         dst++;

  784.         src++;

  785.     }

  786. }

  787. 

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

  789. {

  790.     int i;

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

  792.     {

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

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

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

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

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

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

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

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

  801.         dst+=dstStride;

  802.         src+=srcStride;

  803.     }

  804. }

  805. 

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

  807. {

  808.     int i;

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

  810.     {

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

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

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

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

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

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

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

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

  819.         dst+=dstStride;

  820.         src1+=srcStride;

  821.         src2+=8;

  822.     }

  823. }

  824. 

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

  826. {

  827.     int i;

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

  829.     {

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

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

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

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

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

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

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

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

  838.         dst+=dstStride;

  839.         src1+=srcStride;

  840.         src2+=8;

  841.         src3+=8;

  842.         src4+=8;

  843.     }

  844. }

  845. 

  846. #define QPEL_MC(r, name) \

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

  848. {\

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

  850. }\

  851. \

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

  853. {\

  854.     UINT8 half[64];\

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

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

  857. }\

  858. \

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

  860. {\

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

  862. }\

  863. \

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

  865. {\

  866.     UINT8 half[64];\

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

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

  869. }\

  870. \

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

  872. {\

  873.     UINT8 half[64];\

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

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

  876. }\

  877. \

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

  879. {\

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

  881. }\

  882. \

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

  884. {\

  885.     UINT8 half[64];\

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

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

  888. }\

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

  890. {\

  891.     UINT8 halfH[72];\

  892.     UINT8 halfV[64];\

  893.     UINT8 halfHV[64];\

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

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

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

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

  898. }\

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

  900. {\

  901.     UINT8 halfH[72];\

  902.     UINT8 halfV[64];\

  903.     UINT8 halfHV[64];\

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

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

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

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

  908. }\

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

  910. {\

  911.     UINT8 halfH[72];\

  912.     UINT8 halfV[64];\

  913.     UINT8 halfHV[64];\

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

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

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

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

  918. }\

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

  920. {\

  921.     UINT8 halfH[72];\

  922.     UINT8 halfV[64];\

  923.     UINT8 halfHV[64];\

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

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

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

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

  928. }\

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

  930. {\

  931.     UINT8 halfH[72];\

  932.     UINT8 halfHV[64];\

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

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

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

  936. }\

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

  938. {\

  939.     UINT8 halfH[72];\

  940.     UINT8 halfHV[64];\

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

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

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

  944. }\

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

  946. {\

  947.     UINT8 halfH[72];\

  948.     UINT8 halfV[64];\

  949.     UINT8 halfHV[64];\

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

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

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

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

  954. }\

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

  956. {\

  957.     UINT8 halfH[72];\

  958.     UINT8 halfV[64];\

  959.     UINT8 halfHV[64];\

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

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

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

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

  964. }\

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

  966. {\

  967.     UINT8 halfH[72];\

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

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

  970. }\

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

  972.     qpel_mc00_c ## name,                                                                   \

  973.     qpel_mc10_c ## name,                                                                   \

  974.     qpel_mc20_c ## name,                                                                   \

  975.     qpel_mc30_c ## name,                                                                   \

  976.     qpel_mc01_c ## name,                                                                   \

  977.     qpel_mc11_c ## name,                                                                   \

  978.     qpel_mc21_c ## name,                                                                   \

  979.     qpel_mc31_c ## name,                                                                   \

  980.     qpel_mc02_c ## name,                                                                   \

  981.     qpel_mc12_c ## name,                                                                   \

  982.     qpel_mc22_c ## name,                                                                   \

  983.     qpel_mc32_c ## name,                                                                   \

  984.     qpel_mc03_c ## name,                                                                   \

  985.     qpel_mc13_c ## name,                                                                   \

  986.     qpel_mc23_c ## name,                                                                   \

  987.     qpel_mc33_c ## name,                                                                   \

  988. };

  989. 

  990. QPEL_MC(0, _rnd)

  991. QPEL_MC(1, _no_rnd)

  992. 

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

  994. {

  995.     int s, i;

  996. 

  997.     s = 0;

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

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

  1000.         s += abs(pix1[1] - pix2[1]);

  1001.         s += abs(pix1[2] - pix2[2]);

  1002.         s += abs(pix1[3] - pix2[3]);

  1003.         s += abs(pix1[4] - pix2[4]);

  1004.         s += abs(pix1[5] - pix2[5]);

  1005.         s += abs(pix1[6] - pix2[6]);

  1006.         s += abs(pix1[7] - pix2[7]);

  1007.         s += abs(pix1[8] - pix2[8]);

  1008.         s += abs(pix1[9] - pix2[9]);

  1009.         s += abs(pix1[10] - pix2[10]);

  1010.         s += abs(pix1[11] - pix2[11]);

  1011.         s += abs(pix1[12] - pix2[12]);

  1012.         s += abs(pix1[13] - pix2[13]);

  1013.         s += abs(pix1[14] - pix2[14]);

  1014.         s += abs(pix1[15] - pix2[15]);

  1015.         pix1 += line_size;

  1016.         pix2 += line_size;

  1017.     }

  1018.     return s;

  1019. }

  1020. 

  1021. int pix_abs16x16_x2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  1022. {

  1023.     int s, i;

  1024. 

  1025.     s = 0;

  1026.     for(i=0;i<16;i++) {

  1027.         s += abs(pix1[0] - avg2(pix2[0], pix2[1]));

  1028.         s += abs(pix1[1] - avg2(pix2[1], pix2[2]));

  1029.         s += abs(pix1[2] - avg2(pix2[2], pix2[3]));

  1030.         s += abs(pix1[3] - avg2(pix2[3], pix2[4]));

  1031.         s += abs(pix1[4] - avg2(pix2[4], pix2[5]));

  1032.         s += abs(pix1[5] - avg2(pix2[5], pix2[6]));

  1033.         s += abs(pix1[6] - avg2(pix2[6], pix2[7]));

  1034.         s += abs(pix1[7] - avg2(pix2[7], pix2[8]));

  1035.         s += abs(pix1[8] - avg2(pix2[8], pix2[9]));

  1036.         s += abs(pix1[9] - avg2(pix2[9], pix2[10]));

  1037.         s += abs(pix1[10] - avg2(pix2[10], pix2[11]));

  1038.         s += abs(pix1[11] - avg2(pix2[11], pix2[12]));

  1039.         s += abs(pix1[12] - avg2(pix2[12], pix2[13]));

  1040.         s += abs(pix1[13] - avg2(pix2[13], pix2[14]));

  1041.         s += abs(pix1[14] - avg2(pix2[14], pix2[15]));

  1042.         s += abs(pix1[15] - avg2(pix2[15], pix2[16]));

  1043.         pix1 += line_size;

  1044.         pix2 += line_size;

  1045.     }

  1046.     return s;

  1047. }

  1048. 

  1049. int pix_abs16x16_y2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  1050. {

  1051.     int s, i;

  1052.     UINT8 *pix3 = pix2 + line_size;

  1053. 

  1054.     s = 0;

  1055.     for(i=0;i<16;i++) {

  1056.         s += abs(pix1[0] - avg2(pix2[0], pix3[0]));

  1057.         s += abs(pix1[1] - avg2(pix2[1], pix3[1]));

  1058.         s += abs(pix1[2] - avg2(pix2[2], pix3[2]));

  1059.         s += abs(pix1[3] - avg2(pix2[3], pix3[3]));

  1060.         s += abs(pix1[4] - avg2(pix2[4], pix3[4]));

  1061.         s += abs(pix1[5] - avg2(pix2[5], pix3[5]));

  1062.         s += abs(pix1[6] - avg2(pix2[6], pix3[6]));

  1063.         s += abs(pix1[7] - avg2(pix2[7], pix3[7]));

  1064.         s += abs(pix1[8] - avg2(pix2[8], pix3[8]));

  1065.         s += abs(pix1[9] - avg2(pix2[9], pix3[9]));

  1066.         s += abs(pix1[10] - avg2(pix2[10], pix3[10]));

  1067.         s += abs(pix1[11] - avg2(pix2[11], pix3[11]));

  1068.         s += abs(pix1[12] - avg2(pix2[12], pix3[12]));

  1069.         s += abs(pix1[13] - avg2(pix2[13], pix3[13]));

  1070.         s += abs(pix1[14] - avg2(pix2[14], pix3[14]));

  1071.         s += abs(pix1[15] - avg2(pix2[15], pix3[15]));

  1072.         pix1 += line_size;

  1073.         pix2 += line_size;

  1074.         pix3 += line_size;

  1075.     }

  1076.     return s;

  1077. }

  1078. 

  1079. int pix_abs16x16_xy2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  1080. {

  1081.     int s, i;

  1082.     UINT8 *pix3 = pix2 + line_size;

  1083. 

  1084.     s = 0;

  1085.     for(i=0;i<16;i++) {

  1086.         s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));

  1087.         s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));

  1088.         s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));

  1089.         s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));

  1090.         s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));

  1091.         s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));

  1092.         s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));

  1093.         s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));

  1094.         s += abs(pix1[8] - avg4(pix2[8], pix2[9], pix3[8], pix3[9]));

  1095.         s += abs(pix1[9] - avg4(pix2[9], pix2[10], pix3[9], pix3[10]));

  1096.         s += abs(pix1[10] - avg4(pix2[10], pix2[11], pix3[10], pix3[11]));

  1097.         s += abs(pix1[11] - avg4(pix2[11], pix2[12], pix3[11], pix3[12]));

  1098.         s += abs(pix1[12] - avg4(pix2[12], pix2[13], pix3[12], pix3[13]));

  1099.         s += abs(pix1[13] - avg4(pix2[13], pix2[14], pix3[13], pix3[14]));

  1100.         s += abs(pix1[14] - avg4(pix2[14], pix2[15], pix3[14], pix3[15]));

  1101.         s += abs(pix1[15] - avg4(pix2[15], pix2[16], pix3[15], pix3[16]));

  1102.         pix1 += line_size;

  1103.         pix2 += line_size;

  1104.         pix3 += line_size;

  1105.     }

  1106.     return s;

  1107. }

  1108. 

  1109. int pix_abs8x8_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  1110. {

  1111.     int s, i;

  1112. 

  1113.     s = 0;

  1114.     for(i=0;i<8;i++) {

  1115.         s += abs(pix1[0] - pix2[0]);

  1116.         s += abs(pix1[1] - pix2[1]);

  1117.         s += abs(pix1[2] - pix2[2]);

  1118.         s += abs(pix1[3] - pix2[3]);

  1119.         s += abs(pix1[4] - pix2[4]);

  1120.         s += abs(pix1[5] - pix2[5]);

  1121.         s += abs(pix1[6] - pix2[6]);

  1122.         s += abs(pix1[7] - pix2[7]);

  1123.         pix1 += line_size;

  1124.         pix2 += line_size;

  1125.     }

  1126.     return s;

  1127. }

  1128. 

  1129. int pix_abs8x8_x2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  1130. {

  1131.     int s, i;

  1132. 

  1133.     s = 0;

  1134.     for(i=0;i<8;i++) {

  1135.         s += abs(pix1[0] - avg2(pix2[0], pix2[1]));

  1136.         s += abs(pix1[1] - avg2(pix2[1], pix2[2]));

  1137.         s += abs(pix1[2] - avg2(pix2[2], pix2[3]));

  1138.         s += abs(pix1[3] - avg2(pix2[3], pix2[4]));

  1139.         s += abs(pix1[4] - avg2(pix2[4], pix2[5]));

  1140.         s += abs(pix1[5] - avg2(pix2[5], pix2[6]));

  1141.         s += abs(pix1[6] - avg2(pix2[6], pix2[7]));

  1142.         s += abs(pix1[7] - avg2(pix2[7], pix2[8]));

  1143.         pix1 += line_size;

  1144.         pix2 += line_size;

  1145.     }

  1146.     return s;

  1147. }

  1148. 

  1149. int pix_abs8x8_y2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  1150. {

  1151.     int s, i;

  1152.     UINT8 *pix3 = pix2 + line_size;

  1153. 

  1154.     s = 0;

  1155.     for(i=0;i<8;i++) {

  1156.         s += abs(pix1[0] - avg2(pix2[0], pix3[0]));

  1157.         s += abs(pix1[1] - avg2(pix2[1], pix3[1]));

  1158.         s += abs(pix1[2] - avg2(pix2[2], pix3[2]));

  1159.         s += abs(pix1[3] - avg2(pix2[3], pix3[3]));

  1160.         s += abs(pix1[4] - avg2(pix2[4], pix3[4]));

  1161.         s += abs(pix1[5] - avg2(pix2[5], pix3[5]));

  1162.         s += abs(pix1[6] - avg2(pix2[6], pix3[6]));

  1163.         s += abs(pix1[7] - avg2(pix2[7], pix3[7]));

  1164.         pix1 += line_size;

  1165.         pix2 += line_size;

  1166.         pix3 += line_size;

  1167.     }

  1168.     return s;

  1169. }

  1170. 

  1171. int pix_abs8x8_xy2_c(UINT8 *pix1, UINT8 *pix2, int line_size)

  1172. {

  1173.     int s, i;

  1174.     UINT8 *pix3 = pix2 + line_size;

  1175. 

  1176.     s = 0;

  1177.     for(i=0;i<8;i++) {

  1178.         s += abs(pix1[0] - avg4(pix2[0], pix2[1], pix3[0], pix3[1]));

  1179.         s += abs(pix1[1] - avg4(pix2[1], pix2[2], pix3[1], pix3[2]));

  1180.         s += abs(pix1[2] - avg4(pix2[2], pix2[3], pix3[2], pix3[3]));

  1181.         s += abs(pix1[3] - avg4(pix2[3], pix2[4], pix3[3], pix3[4]));

  1182.         s += abs(pix1[4] - avg4(pix2[4], pix2[5], pix3[4], pix3[5]));

  1183.         s += abs(pix1[5] - avg4(pix2[5], pix2[6], pix3[5], pix3[6]));

  1184.         s += abs(pix1[6] - avg4(pix2[6], pix2[7], pix3[6], pix3[7]));

  1185.         s += abs(pix1[7] - avg4(pix2[7], pix2[8], pix3[7], pix3[8]));

  1186.         pix1 += line_size;

  1187.         pix2 += line_size;

  1188.         pix3 += line_size;

  1189.     }

  1190.     return s;

  1191. }

  1192. 

  1193. /* permute block according so that it corresponds to the MMX idct

  1194.    order */

  1195. #ifdef SIMPLE_IDCT

  1196.  /* general permutation, but perhaps slightly slower */

  1197. void block_permute(INT16 *block)

  1198. {

  1199. 	int i;

  1200. 	INT16 temp[64];

  1201. 

  1202. 	for(i=0; i<64; i++) temp[ block_permute_op(i) ] = block[i];

  1203. 

  1204. 	for(i=0; i<64; i++) block[i] = temp[i];

  1205. }

  1206. #else

  1207. 

  1208. void block_permute(INT16 *block)

  1209. {

  1210.     int tmp1, tmp2, tmp3, tmp4, tmp5, tmp6;

  1211.     int i;

  1212. 

  1213.     for(i=0;i<8;i++) {

  1214.         tmp1 = block[1];

  1215.         tmp2 = block[2];

  1216.         tmp3 = block[3];

  1217.         tmp4 = block[4];

  1218.         tmp5 = block[5];

  1219.         tmp6 = block[6];

  1220.         block[1] = tmp2;

  1221.         block[2] = tmp4;

  1222.         block[3] = tmp6;

  1223.         block[4] = tmp1;

  1224.         block[5] = tmp3;

  1225.         block[6] = tmp5;

  1226.         block += 8;

  1227.     }

  1228. }

  1229. #endif

  1230. 

  1231. void clear_blocks_c(DCTELEM *blocks)

  1232. {

  1233.     memset(blocks, 0, sizeof(DCTELEM)*6*64);

  1234. }

  1235. 

  1236. void dsputil_init(void)

  1237. {

  1238.     int i, j;

  1239.     int use_permuted_idct;

  1240. 

  1241.     for(i=0;i<256;i++) cropTbl[i + MAX_NEG_CROP] = i;

  1242.     for(i=0;i<MAX_NEG_CROP;i++) {

  1243.         cropTbl[i] = 0;

  1244.         cropTbl[i + MAX_NEG_CROP + 256] = 255;

  1245.     }

  1246. 

  1247.     for(i=0;i<512;i++) {

  1248.         squareTbl[i] = (i - 256) * (i - 256);

  1249.     }

  1250. 

  1251. #ifdef SIMPLE_IDCT

  1252.     ff_idct = simple_idct;

  1253. #else

  1254.     ff_idct = j_rev_dct;

  1255. #endif

  1256.     get_pixels = get_pixels_c;

  1257.     diff_pixels = diff_pixels_c;

  1258.     put_pixels_clamped = put_pixels_clamped_c;

  1259.     add_pixels_clamped = add_pixels_clamped_c;

  1260.     gmc1= gmc1_c;

  1261.     clear_blocks= clear_blocks_c;

  1262. 

  1263.     pix_abs16x16     = pix_abs16x16_c;

  1264.     pix_abs16x16_x2  = pix_abs16x16_x2_c;

  1265.     pix_abs16x16_y2  = pix_abs16x16_y2_c;

  1266.     pix_abs16x16_xy2 = pix_abs16x16_xy2_c;

  1267.     pix_abs8x8     = pix_abs8x8_c;

  1268.     pix_abs8x8_x2  = pix_abs8x8_x2_c;

  1269.     pix_abs8x8_y2  = pix_abs8x8_y2_c;

  1270.     pix_abs8x8_xy2 = pix_abs8x8_xy2_c;

  1271.     av_fdct = jpeg_fdct_ifast;

  1272. 

  1273.     use_permuted_idct = 1;

  1274. 

  1275. #ifdef HAVE_MMX

  1276.     dsputil_init_mmx();

  1277. #endif

  1278. #ifdef ARCH_ARMV4L

  1279.     dsputil_init_armv4l();

  1280. #endif

  1281. #ifdef HAVE_MLIB

  1282.     dsputil_init_mlib();

  1283.     use_permuted_idct = 0;

  1284. #endif

  1285. #ifdef ARCH_ALPHA

  1286.     dsputil_init_alpha();

  1287.     use_permuted_idct = 0;

  1288. #endif

  1289. 

  1290. #ifdef SIMPLE_IDCT

  1291.     if(ff_idct == simple_idct) use_permuted_idct=0;

  1292. #endif

  1293. 

  1294.     if(use_permuted_idct)

  1295. #ifdef SIMPLE_IDCT

  1296.         for(i=0; i<64; i++) permutation[i]= simple_mmx_permutation[i];

  1297. #else

  1298.         for(i=0; i<64; i++) permutation[i]= (i & 0x38) | ((i & 6) >> 1) | ((i & 1) << 2);

  1299. #endif

  1300.     else

  1301.         for(i=0; i<64; i++) permutation[i]=i;

  1302. 

  1303.     for(i=0; i<64; i++) inv_zigzag_direct16[zigzag_direct[i]]= i+1;

  1304.     for(i=0; i<64; i++) zigzag_direct_noperm[i]= zigzag_direct[i];

  1305.     

  1306.     if (use_permuted_idct) {

  1307.         /* permute for IDCT */

  1308.         for(i=0;i<64;i++) {

  1309.             j = zigzag_direct[i];

  1310.             zigzag_direct[i] = block_permute_op(j);

  1311.             j = ff_alternate_horizontal_scan[i];

  1312.             ff_alternate_horizontal_scan[i] = block_permute_op(j);

  1313.             j = ff_alternate_vertical_scan[i];

  1314.             ff_alternate_vertical_scan[i] = block_permute_op(j);

  1315.         }

  1316.         block_permute(default_intra_matrix);

  1317.         block_permute(default_non_intra_matrix);

  1318.         block_permute(ff_mpeg4_default_intra_matrix);

  1319.         block_permute(ff_mpeg4_default_non_intra_matrix);

  1320.     }

  1321.     

  1322.     build_zigzag_end();

  1323. }

  1324. 

  1325. void get_psnr(UINT8 *orig_image[3], UINT8 *coded_image[3],

  1326.               int orig_linesize[3], int coded_linesize,

  1327.               AVCodecContext *avctx)

  1328. {

  1329.     int quad, diff, x, y;

  1330.     UINT8 *orig, *coded;

  1331.     UINT32 *sq = squareTbl + 256;

  1332.     

  1333.     quad = 0;

  1334.     diff = 0;

  1335.     

  1336.     /* Luminance */

  1337.     orig = orig_image[0];

  1338.     coded = coded_image[0];

  1339.     

  1340.     for (y=0;y<avctx->height;y++) {

  1341.         for (x=0;x<avctx->width;x++) {

  1342.             diff = *(orig + x) - *(coded + x);

  1343.             quad += sq[diff];

  1344.         }

  1345.         orig += orig_linesize[0];

  1346.         coded += coded_linesize;

  1347.     }

  1348.    

  1349.     avctx->psnr_y = (float) quad / (float) (avctx->width * avctx->height);

  1350.     

  1351.     if (avctx->psnr_y) {

  1352.         avctx->psnr_y = (float) (255 * 255) / avctx->psnr_y;

  1353.         avctx->psnr_y = 10 * (float) log10 (avctx->psnr_y); 

  1354.     } else

  1355.         avctx->psnr_y = 99.99;

  1356. }