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.
36155 Commits
32 Branches
1.0GB
Tree: 89b51b570d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '89b51b570d'
${ noResults }
FFmpeg/libavcodec/vc1dsp.c

855 lines
28KB
Raw Blame History 

  1. /*

  2.  * VC-1 and WMV3 decoder - DSP functions

  3.  * Copyright (c) 2006 Konstantin Shishkov

  4.  *

  5.  * This file is part of Libav.

  6.  *

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

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

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

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

  11.  *

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

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

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

  15.  * Lesser General Public License for more details.

  16.  *

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

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

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

  20.  */

  21. 

  22. /**

  23. * @file

  24.  * VC-1 and WMV3 decoder

  25.  *

  26.  */

  27. 

  28. #include "vc1dsp.h"

  29. #include "libavutil/common.h"

  30. 

  31. 

  32. /** Apply overlap transform to horizontal edge

  33. */

  34. static void vc1_v_overlap_c(uint8_t* src, int stride)

  35. {

  36.     int i;

  37.     int a, b, c, d;

  38.     int d1, d2;

  39.     int rnd = 1;

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

  41.         a = src[-2*stride];

  42.         b = src[-stride];

  43.         c = src[0];

  44.         d = src[stride];

  45.         d1 = (a - d + 3 + rnd) >> 3;

  46.         d2 = (a - d + b - c + 4 - rnd) >> 3;

  47. 

  48.         src[-2*stride] = a - d1;

  49.         src[-stride] = av_clip_uint8(b - d2);

  50.         src[0] = av_clip_uint8(c + d2);

  51.         src[stride] = d + d1;

  52.         src++;

  53.         rnd = !rnd;

  54.     }

  55. }

  56. 

  57. /** Apply overlap transform to vertical edge

  58. */

  59. static void vc1_h_overlap_c(uint8_t* src, int stride)

  60. {

  61.     int i;

  62.     int a, b, c, d;

  63.     int d1, d2;

  64.     int rnd = 1;

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

  66.         a = src[-2];

  67.         b = src[-1];

  68.         c = src[0];

  69.         d = src[1];

  70.         d1 = (a - d + 3 + rnd) >> 3;

  71.         d2 = (a - d + b - c + 4 - rnd) >> 3;

  72. 

  73.         src[-2] = a - d1;

  74.         src[-1] = av_clip_uint8(b - d2);

  75.         src[0] = av_clip_uint8(c + d2);

  76.         src[1] = d + d1;

  77.         src += stride;

  78.         rnd = !rnd;

  79.     }

  80. }

  81. 

  82. static void vc1_v_s_overlap_c(DCTELEM *top,  DCTELEM *bottom)

  83. {

  84.     int i;

  85.     int a, b, c, d;

  86.     int d1, d2;

  87.     int rnd1 = 4, rnd2 = 3;

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

  89.         a = top[48];

  90.         b = top[56];

  91.         c = bottom[0];

  92.         d = bottom[8];

  93.         d1 = a - d;

  94.         d2 = a - d + b - c;

  95. 

  96.         top[48]   = ((a << 3) - d1 + rnd1) >> 3;

  97.         top[56]   = ((b << 3) - d2 + rnd2) >> 3;

  98.         bottom[0] = ((c << 3) + d2 + rnd1) >> 3;

  99.         bottom[8] = ((d << 3) + d1 + rnd2) >> 3;

  100. 

  101.         bottom++;

  102.         top++;

  103.         rnd2 = 7 - rnd2;

  104.         rnd1 = 7 - rnd1;

  105.     }

  106. }

  107. 

  108. static void vc1_h_s_overlap_c(DCTELEM *left, DCTELEM *right)

  109. {

  110.     int i;

  111.     int a, b, c, d;

  112.     int d1, d2;

  113.     int rnd1 = 4, rnd2 = 3;

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

  115.         a = left[6];

  116.         b = left[7];

  117.         c = right[0];

  118.         d = right[1];

  119.         d1 = a - d;

  120.         d2 = a - d + b - c;

  121. 

  122.         left[6]  = ((a << 3) - d1 + rnd1) >> 3;

  123.         left[7]  = ((b << 3) - d2 + rnd2) >> 3;

  124.         right[0] = ((c << 3) + d2 + rnd1) >> 3;

  125.         right[1] = ((d << 3) + d1 + rnd2) >> 3;

  126. 

  127.         right += 8;

  128.         left += 8;

  129.         rnd2 = 7 - rnd2;

  130.         rnd1 = 7 - rnd1;

  131.     }

  132. }

  133. 

  134. /**

  135.  * VC-1 in-loop deblocking filter for one line

  136.  * @param src source block type

  137.  * @param stride block stride

  138.  * @param pq block quantizer

  139.  * @return whether other 3 pairs should be filtered or not

  140.  * @see 8.6

  141.  */

  142. static av_always_inline int vc1_filter_line(uint8_t* src, int stride, int pq){

  143.     int a0 = (2*(src[-2*stride] - src[ 1*stride]) - 5*(src[-1*stride] - src[ 0*stride]) + 4) >> 3;

  144.     int a0_sign = a0 >> 31;        /* Store sign */

  145.     a0 = (a0 ^ a0_sign) - a0_sign; /* a0 = FFABS(a0); */

  146.     if(a0 < pq){

  147.         int a1 = FFABS((2*(src[-4*stride] - src[-1*stride]) - 5*(src[-3*stride] - src[-2*stride]) + 4) >> 3);

  148.         int a2 = FFABS((2*(src[ 0*stride] - src[ 3*stride]) - 5*(src[ 1*stride] - src[ 2*stride]) + 4) >> 3);

  149.         if(a1 < a0 || a2 < a0){

  150.             int clip = src[-1*stride] - src[ 0*stride];

  151.             int clip_sign = clip >> 31;

  152.             clip = ((clip ^ clip_sign) - clip_sign)>>1;

  153.             if(clip){

  154.                 int a3 = FFMIN(a1, a2);

  155.                 int d = 5 * (a3 - a0);

  156.                 int d_sign = (d >> 31);

  157.                 d = ((d ^ d_sign) - d_sign) >> 3;

  158.                 d_sign ^= a0_sign;

  159. 

  160.                 if( d_sign ^ clip_sign )

  161.                     d = 0;

  162.                 else{

  163.                     d = FFMIN(d, clip);

  164.                     d = (d ^ d_sign) - d_sign;          /* Restore sign */

  165.                     src[-1*stride] = av_clip_uint8(src[-1*stride] - d);

  166.                     src[ 0*stride] = av_clip_uint8(src[ 0*stride] + d);

  167.                 }

  168.                 return 1;

  169.             }

  170.         }

  171.     }

  172.     return 0;

  173. }

  174. 

  175. /**

  176.  * VC-1 in-loop deblocking filter

  177.  * @param src source block type

  178.  * @param step distance between horizontally adjacent elements

  179.  * @param stride distance between vertically adjacent elements

  180.  * @param len edge length to filter (4 or 8 pixels)

  181.  * @param pq block quantizer

  182.  * @see 8.6

  183.  */

  184. static inline void vc1_loop_filter(uint8_t* src, int step, int stride, int len, int pq)

  185. {

  186.     int i;

  187.     int filt3;

  188. 

  189.     for(i = 0; i < len; i += 4){

  190.         filt3 = vc1_filter_line(src + 2*step, stride, pq);

  191.         if(filt3){

  192.             vc1_filter_line(src + 0*step, stride, pq);

  193.             vc1_filter_line(src + 1*step, stride, pq);

  194.             vc1_filter_line(src + 3*step, stride, pq);

  195.         }

  196.         src += step * 4;

  197.     }

  198. }

  199. 

  200. static void vc1_v_loop_filter4_c(uint8_t *src, int stride, int pq)

  201. {

  202.     vc1_loop_filter(src, 1, stride, 4, pq);

  203. }

  204. 

  205. static void vc1_h_loop_filter4_c(uint8_t *src, int stride, int pq)

  206. {

  207.     vc1_loop_filter(src, stride, 1, 4, pq);

  208. }

  209. 

  210. static void vc1_v_loop_filter8_c(uint8_t *src, int stride, int pq)

  211. {

  212.     vc1_loop_filter(src, 1, stride, 8, pq);

  213. }

  214. 

  215. static void vc1_h_loop_filter8_c(uint8_t *src, int stride, int pq)

  216. {

  217.     vc1_loop_filter(src, stride, 1, 8, pq);

  218. }

  219. 

  220. static void vc1_v_loop_filter16_c(uint8_t *src, int stride, int pq)

  221. {

  222.     vc1_loop_filter(src, 1, stride, 16, pq);

  223. }

  224. 

  225. static void vc1_h_loop_filter16_c(uint8_t *src, int stride, int pq)

  226. {

  227.     vc1_loop_filter(src, stride, 1, 16, pq);

  228. }

  229. 

  230. /** Do inverse transform on 8x8 block

  231. */

  232. static void vc1_inv_trans_8x8_dc_c(uint8_t *dest, int linesize, DCTELEM *block)

  233. {

  234.     int i;

  235.     int dc = block[0];

  236.     dc = (3 * dc +  1) >> 1;

  237.     dc = (3 * dc + 16) >> 5;

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

  239.         dest[0] = av_clip_uint8(dest[0] + dc);

  240.         dest[1] = av_clip_uint8(dest[1] + dc);

  241.         dest[2] = av_clip_uint8(dest[2] + dc);

  242.         dest[3] = av_clip_uint8(dest[3] + dc);

  243.         dest[4] = av_clip_uint8(dest[4] + dc);

  244.         dest[5] = av_clip_uint8(dest[5] + dc);

  245.         dest[6] = av_clip_uint8(dest[6] + dc);

  246.         dest[7] = av_clip_uint8(dest[7] + dc);

  247.         dest += linesize;

  248.     }

  249. }

  250. 

  251. static void vc1_inv_trans_8x8_c(DCTELEM block[64])

  252. {

  253.     int i;

  254.     register int t1,t2,t3,t4,t5,t6,t7,t8;

  255.     DCTELEM *src, *dst, temp[64];

  256. 

  257.     src = block;

  258.     dst = temp;

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

  260.         t1 = 12 * (src[ 0] + src[32]) + 4;

  261.         t2 = 12 * (src[ 0] - src[32]) + 4;

  262.         t3 = 16 * src[16] +  6 * src[48];

  263.         t4 =  6 * src[16] - 16 * src[48];

  264. 

  265.         t5 = t1 + t3;

  266.         t6 = t2 + t4;

  267.         t7 = t2 - t4;

  268.         t8 = t1 - t3;

  269. 

  270.         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];

  271.         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];

  272.         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];

  273.         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];

  274. 

  275.         dst[0] = (t5 + t1) >> 3;

  276.         dst[1] = (t6 + t2) >> 3;

  277.         dst[2] = (t7 + t3) >> 3;

  278.         dst[3] = (t8 + t4) >> 3;

  279.         dst[4] = (t8 - t4) >> 3;

  280.         dst[5] = (t7 - t3) >> 3;

  281.         dst[6] = (t6 - t2) >> 3;

  282.         dst[7] = (t5 - t1) >> 3;

  283. 

  284.         src += 1;

  285.         dst += 8;

  286.     }

  287. 

  288.     src = temp;

  289.     dst = block;

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

  291.         t1 = 12 * (src[ 0] + src[32]) + 64;

  292.         t2 = 12 * (src[ 0] - src[32]) + 64;

  293.         t3 = 16 * src[16] +  6 * src[48];

  294.         t4 =  6 * src[16] - 16 * src[48];

  295. 

  296.         t5 = t1 + t3;

  297.         t6 = t2 + t4;

  298.         t7 = t2 - t4;

  299.         t8 = t1 - t3;

  300. 

  301.         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];

  302.         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];

  303.         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];

  304.         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];

  305. 

  306.         dst[ 0] = (t5 + t1) >> 7;

  307.         dst[ 8] = (t6 + t2) >> 7;

  308.         dst[16] = (t7 + t3) >> 7;

  309.         dst[24] = (t8 + t4) >> 7;

  310.         dst[32] = (t8 - t4 + 1) >> 7;

  311.         dst[40] = (t7 - t3 + 1) >> 7;

  312.         dst[48] = (t6 - t2 + 1) >> 7;

  313.         dst[56] = (t5 - t1 + 1) >> 7;

  314. 

  315.         src++;

  316.         dst++;

  317.     }

  318. }

  319. 

  320. /** Do inverse transform on 8x4 part of block

  321. */

  322. static void vc1_inv_trans_8x4_dc_c(uint8_t *dest, int linesize, DCTELEM *block)

  323. {

  324.     int i;

  325.     int dc = block[0];

  326.     dc = ( 3 * dc +  1) >> 1;

  327.     dc = (17 * dc + 64) >> 7;

  328.     for(i = 0; i < 4; i++){

  329.         dest[0] = av_clip_uint8(dest[0] + dc);

  330.         dest[1] = av_clip_uint8(dest[1] + dc);

  331.         dest[2] = av_clip_uint8(dest[2] + dc);

  332.         dest[3] = av_clip_uint8(dest[3] + dc);

  333.         dest[4] = av_clip_uint8(dest[4] + dc);

  334.         dest[5] = av_clip_uint8(dest[5] + dc);

  335.         dest[6] = av_clip_uint8(dest[6] + dc);

  336.         dest[7] = av_clip_uint8(dest[7] + dc);

  337.         dest += linesize;

  338.     }

  339. }

  340. 

  341. static void vc1_inv_trans_8x4_c(uint8_t *dest, int linesize, DCTELEM *block)

  342. {

  343.     int i;

  344.     register int t1,t2,t3,t4,t5,t6,t7,t8;

  345.     DCTELEM *src, *dst;

  346. 

  347.     src = block;

  348.     dst = block;

  349.     for(i = 0; i < 4; i++){

  350.         t1 = 12 * (src[0] + src[4]) + 4;

  351.         t2 = 12 * (src[0] - src[4]) + 4;

  352.         t3 = 16 * src[2] +  6 * src[6];

  353.         t4 =  6 * src[2] - 16 * src[6];

  354. 

  355.         t5 = t1 + t3;

  356.         t6 = t2 + t4;

  357.         t7 = t2 - t4;

  358.         t8 = t1 - t3;

  359. 

  360.         t1 = 16 * src[1] + 15 * src[3] +  9 * src[5] +  4 * src[7];

  361.         t2 = 15 * src[1] -  4 * src[3] - 16 * src[5] -  9 * src[7];

  362.         t3 =  9 * src[1] - 16 * src[3] +  4 * src[5] + 15 * src[7];

  363.         t4 =  4 * src[1] -  9 * src[3] + 15 * src[5] - 16 * src[7];

  364. 

  365.         dst[0] = (t5 + t1) >> 3;

  366.         dst[1] = (t6 + t2) >> 3;

  367.         dst[2] = (t7 + t3) >> 3;

  368.         dst[3] = (t8 + t4) >> 3;

  369.         dst[4] = (t8 - t4) >> 3;

  370.         dst[5] = (t7 - t3) >> 3;

  371.         dst[6] = (t6 - t2) >> 3;

  372.         dst[7] = (t5 - t1) >> 3;

  373. 

  374.         src += 8;

  375.         dst += 8;

  376.     }

  377. 

  378.     src = block;

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

  380.         t1 = 17 * (src[ 0] + src[16]) + 64;

  381.         t2 = 17 * (src[ 0] - src[16]) + 64;

  382.         t3 = 22 * src[ 8] + 10 * src[24];

  383.         t4 = 22 * src[24] - 10 * src[ 8];

  384. 

  385.         dest[0*linesize] = av_clip_uint8(dest[0*linesize] + ((t1 + t3) >> 7));

  386.         dest[1*linesize] = av_clip_uint8(dest[1*linesize] + ((t2 - t4) >> 7));

  387.         dest[2*linesize] = av_clip_uint8(dest[2*linesize] + ((t2 + t4) >> 7));

  388.         dest[3*linesize] = av_clip_uint8(dest[3*linesize] + ((t1 - t3) >> 7));

  389. 

  390.         src ++;

  391.         dest++;

  392.     }

  393. }

  394. 

  395. /** Do inverse transform on 4x8 parts of block

  396. */

  397. static void vc1_inv_trans_4x8_dc_c(uint8_t *dest, int linesize, DCTELEM *block)

  398. {

  399.     int i;

  400.     int dc = block[0];

  401.     dc = (17 * dc +  4) >> 3;

  402.     dc = (12 * dc + 64) >> 7;

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

  404.         dest[0] = av_clip_uint8(dest[0] + dc);

  405.         dest[1] = av_clip_uint8(dest[1] + dc);

  406.         dest[2] = av_clip_uint8(dest[2] + dc);

  407.         dest[3] = av_clip_uint8(dest[3] + dc);

  408.         dest += linesize;

  409.     }

  410. }

  411. 

  412. static void vc1_inv_trans_4x8_c(uint8_t *dest, int linesize, DCTELEM *block)

  413. {

  414.     int i;

  415.     register int t1,t2,t3,t4,t5,t6,t7,t8;

  416.     DCTELEM *src, *dst;

  417. 

  418.     src = block;

  419.     dst = block;

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

  421.         t1 = 17 * (src[0] + src[2]) + 4;

  422.         t2 = 17 * (src[0] - src[2]) + 4;

  423.         t3 = 22 * src[1] + 10 * src[3];

  424.         t4 = 22 * src[3] - 10 * src[1];

  425. 

  426.         dst[0] = (t1 + t3) >> 3;

  427.         dst[1] = (t2 - t4) >> 3;

  428.         dst[2] = (t2 + t4) >> 3;

  429.         dst[3] = (t1 - t3) >> 3;

  430. 

  431.         src += 8;

  432.         dst += 8;

  433.     }

  434. 

  435.     src = block;

  436.     for(i = 0; i < 4; i++){

  437.         t1 = 12 * (src[ 0] + src[32]) + 64;

  438.         t2 = 12 * (src[ 0] - src[32]) + 64;

  439.         t3 = 16 * src[16] +  6 * src[48];

  440.         t4 =  6 * src[16] - 16 * src[48];

  441. 

  442.         t5 = t1 + t3;

  443.         t6 = t2 + t4;

  444.         t7 = t2 - t4;

  445.         t8 = t1 - t3;

  446. 

  447.         t1 = 16 * src[ 8] + 15 * src[24] +  9 * src[40] +  4 * src[56];

  448.         t2 = 15 * src[ 8] -  4 * src[24] - 16 * src[40] -  9 * src[56];

  449.         t3 =  9 * src[ 8] - 16 * src[24] +  4 * src[40] + 15 * src[56];

  450.         t4 =  4 * src[ 8] -  9 * src[24] + 15 * src[40] - 16 * src[56];

  451. 

  452.         dest[0*linesize] = av_clip_uint8(dest[0*linesize] + ((t5 + t1) >> 7));

  453.         dest[1*linesize] = av_clip_uint8(dest[1*linesize] + ((t6 + t2) >> 7));

  454.         dest[2*linesize] = av_clip_uint8(dest[2*linesize] + ((t7 + t3) >> 7));

  455.         dest[3*linesize] = av_clip_uint8(dest[3*linesize] + ((t8 + t4) >> 7));

  456.         dest[4*linesize] = av_clip_uint8(dest[4*linesize] + ((t8 - t4 + 1) >> 7));

  457.         dest[5*linesize] = av_clip_uint8(dest[5*linesize] + ((t7 - t3 + 1) >> 7));

  458.         dest[6*linesize] = av_clip_uint8(dest[6*linesize] + ((t6 - t2 + 1) >> 7));

  459.         dest[7*linesize] = av_clip_uint8(dest[7*linesize] + ((t5 - t1 + 1) >> 7));

  460. 

  461.         src ++;

  462.         dest++;

  463.     }

  464. }

  465. 

  466. /** Do inverse transform on 4x4 part of block

  467. */

  468. static void vc1_inv_trans_4x4_dc_c(uint8_t *dest, int linesize, DCTELEM *block)

  469. {

  470.     int i;

  471.     int dc = block[0];

  472.     dc = (17 * dc +  4) >> 3;

  473.     dc = (17 * dc + 64) >> 7;

  474.     for(i = 0; i < 4; i++){

  475.         dest[0] = av_clip_uint8(dest[0] + dc);

  476.         dest[1] = av_clip_uint8(dest[1] + dc);

  477.         dest[2] = av_clip_uint8(dest[2] + dc);

  478.         dest[3] = av_clip_uint8(dest[3] + dc);

  479.         dest += linesize;

  480.     }

  481. }

  482. 

  483. static void vc1_inv_trans_4x4_c(uint8_t *dest, int linesize, DCTELEM *block)

  484. {

  485.     int i;

  486.     register int t1,t2,t3,t4;

  487.     DCTELEM *src, *dst;

  488. 

  489.     src = block;

  490.     dst = block;

  491.     for(i = 0; i < 4; i++){

  492.         t1 = 17 * (src[0] + src[2]) + 4;

  493.         t2 = 17 * (src[0] - src[2]) + 4;

  494.         t3 = 22 * src[1] + 10 * src[3];

  495.         t4 = 22 * src[3] - 10 * src[1];

  496. 

  497.         dst[0] = (t1 + t3) >> 3;

  498.         dst[1] = (t2 - t4) >> 3;

  499.         dst[2] = (t2 + t4) >> 3;

  500.         dst[3] = (t1 - t3) >> 3;

  501. 

  502.         src += 8;

  503.         dst += 8;

  504.     }

  505. 

  506.     src = block;

  507.     for(i = 0; i < 4; i++){

  508.         t1 = 17 * (src[ 0] + src[16]) + 64;

  509.         t2 = 17 * (src[ 0] - src[16]) + 64;

  510.         t3 = 22 * src[ 8] + 10 * src[24];

  511.         t4 = 22 * src[24] - 10 * src[ 8];

  512. 

  513.         dest[0*linesize] = av_clip_uint8(dest[0*linesize] + ((t1 + t3) >> 7));

  514.         dest[1*linesize] = av_clip_uint8(dest[1*linesize] + ((t2 - t4) >> 7));

  515.         dest[2*linesize] = av_clip_uint8(dest[2*linesize] + ((t2 + t4) >> 7));

  516.         dest[3*linesize] = av_clip_uint8(dest[3*linesize] + ((t1 - t3) >> 7));

  517. 

  518.         src ++;

  519.         dest++;

  520.     }

  521. }

  522. 

  523. /* motion compensation functions */

  524. /** Filter in case of 2 filters */

  525. #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                 \

  526. static av_always_inline int vc1_mspel_ ## DIR ## _filter_16bits(const TYPE *src, int stride, int mode) \

  527. {                                                                       \

  528.     switch(mode){                                                       \

  529.     case 0: /* no shift - should not occur */                           \

  530.         return 0;                                                       \

  531.     case 1: /* 1/4 shift */                                             \

  532.         return -4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2]; \

  533.     case 2: /* 1/2 shift */                                             \

  534.         return -src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2]; \

  535.     case 3: /* 3/4 shift */                                             \

  536.         return -3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2]; \

  537.     }                                                                   \

  538.     return 0; /* should not occur */                                    \

  539. }

  540. 

  541. VC1_MSPEL_FILTER_16B(ver, uint8_t)

  542. VC1_MSPEL_FILTER_16B(hor, int16_t)

  543. 

  544. 

  545. /** Filter used to interpolate fractional pel values

  546.  */

  547. static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride, int mode, int r)

  548. {

  549.     switch(mode){

  550.     case 0: //no shift

  551.         return src[0];

  552.     case 1: // 1/4 shift

  553.         return (-4*src[-stride] + 53*src[0] + 18*src[stride] - 3*src[stride*2] + 32 - r) >> 6;

  554.     case 2: // 1/2 shift

  555.         return (-src[-stride] + 9*src[0] + 9*src[stride] - src[stride*2] + 8 - r) >> 4;

  556.     case 3: // 3/4 shift

  557.         return (-3*src[-stride] + 18*src[0] + 53*src[stride] - 4*src[stride*2] + 32 - r) >> 6;

  558.     }

  559.     return 0; //should not occur

  560. }

  561. 

  562. /** Function used to do motion compensation with bicubic interpolation

  563.  */

  564. #define VC1_MSPEL_MC(OP, OPNAME)\

  565. static av_always_inline void OPNAME ## vc1_mspel_mc(uint8_t *dst, const uint8_t *src, int stride, int hmode, int vmode, int rnd)\

  566. {\

  567.     int     i, j;\

  568. \

  569.     if (vmode) { /* Horizontal filter to apply */\

  570.         int r;\

  571. \

  572.         if (hmode) { /* Vertical filter to apply, output to tmp */\

  573.             static const int shift_value[] = { 0, 5, 1, 5 };\

  574.             int              shift = (shift_value[hmode]+shift_value[vmode])>>1;\

  575.             int16_t          tmp[11*8], *tptr = tmp;\

  576. \

  577.             r = (1<<(shift-1)) + rnd-1;\

  578. \

  579.             src -= 1;\

  580.             for(j = 0; j < 8; j++) {\

  581.                 for(i = 0; i < 11; i++)\

  582.                     tptr[i] = (vc1_mspel_ver_filter_16bits(src + i, stride, vmode)+r)>>shift;\

  583.                 src += stride;\

  584.                 tptr += 11;\

  585.             }\

  586. \

  587.             r = 64-rnd;\

  588.             tptr = tmp+1;\

  589.             for(j = 0; j < 8; j++) {\

  590.                 for(i = 0; i < 8; i++)\

  591.                     OP(dst[i], (vc1_mspel_hor_filter_16bits(tptr + i, 1, hmode)+r)>>7);\

  592.                 dst += stride;\

  593.                 tptr += 11;\

  594.             }\

  595. \

  596.             return;\

  597.         }\

  598.         else { /* No horizontal filter, output 8 lines to dst */\

  599.             r = 1-rnd;\

  600. \

  601.             for(j = 0; j < 8; j++) {\

  602.                 for(i = 0; i < 8; i++)\

  603.                     OP(dst[i], vc1_mspel_filter(src + i, stride, vmode, r));\

  604.                 src += stride;\

  605.                 dst += stride;\

  606.             }\

  607.             return;\

  608.         }\

  609.     }\

  610. \

  611.     /* Horizontal mode with no vertical mode */\

  612.     for(j = 0; j < 8; j++) {\

  613.         for(i = 0; i < 8; i++)\

  614.             OP(dst[i], vc1_mspel_filter(src + i, 1, hmode, rnd));\

  615.         dst += stride;\

  616.         src += stride;\

  617.     }\

  618. }

  619. 

  620. #define op_put(a, b) a = av_clip_uint8(b)

  621. #define op_avg(a, b) a = (a + av_clip_uint8(b) + 1) >> 1

  622. 

  623. VC1_MSPEL_MC(op_put, put_)

  624. VC1_MSPEL_MC(op_avg, avg_)

  625. 

  626. /* pixel functions - really are entry points to vc1_mspel_mc */

  627. 

  628. #define PUT_VC1_MSPEL(a, b)\

  629. static void put_vc1_mspel_mc ## a ## b ##_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \

  630.      put_vc1_mspel_mc(dst, src, stride, a, b, rnd);                         \

  631. }\

  632. static void avg_vc1_mspel_mc ## a ## b ##_c(uint8_t *dst, const uint8_t *src, int stride, int rnd) { \

  633.      avg_vc1_mspel_mc(dst, src, stride, a, b, rnd);                         \

  634. }

  635. 

  636. PUT_VC1_MSPEL(1, 0)

  637. PUT_VC1_MSPEL(2, 0)

  638. PUT_VC1_MSPEL(3, 0)

  639. 

  640. PUT_VC1_MSPEL(0, 1)

  641. PUT_VC1_MSPEL(1, 1)

  642. PUT_VC1_MSPEL(2, 1)

  643. PUT_VC1_MSPEL(3, 1)

  644. 

  645. PUT_VC1_MSPEL(0, 2)

  646. PUT_VC1_MSPEL(1, 2)

  647. PUT_VC1_MSPEL(2, 2)

  648. PUT_VC1_MSPEL(3, 2)

  649. 

  650. PUT_VC1_MSPEL(0, 3)

  651. PUT_VC1_MSPEL(1, 3)

  652. PUT_VC1_MSPEL(2, 3)

  653. PUT_VC1_MSPEL(3, 3)

  654. 

  655. static void put_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){

  656.     const int A=(8-x)*(8-y);

  657.     const int B=(  x)*(8-y);

  658.     const int C=(8-x)*(  y);

  659.     const int D=(  x)*(  y);

  660.     int i;

  661. 

  662.     assert(x<8 && y<8 && x>=0 && y>=0);

  663. 

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

  665.     {

  666.         dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6;

  667.         dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6;

  668.         dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6;

  669.         dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6;

  670.         dst[4] = (A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6;

  671.         dst[5] = (A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6;

  672.         dst[6] = (A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6;

  673.         dst[7] = (A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6;

  674.         dst+= stride;

  675.         src+= stride;

  676.     }

  677. }

  678. 

  679. static void put_no_rnd_vc1_chroma_mc4_c(uint8_t *dst, uint8_t *src, int stride, int h, int x, int y){

  680.     const int A=(8-x)*(8-y);

  681.     const int B=(  x)*(8-y);

  682.     const int C=(8-x)*(  y);

  683.     const int D=(  x)*(  y);

  684.     int i;

  685. 

  686.     assert(x<8 && y<8 && x>=0 && y>=0);

  687. 

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

  689.     {

  690.         dst[0] = (A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6;

  691.         dst[1] = (A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6;

  692.         dst[2] = (A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6;

  693.         dst[3] = (A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6;

  694.         dst+= stride;

  695.         src+= stride;

  696.     }

  697. }

  698. 

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

  700. static void avg_no_rnd_vc1_chroma_mc8_c(uint8_t *dst/*align 8*/, uint8_t *src/*align 1*/, int stride, int h, int x, int y){

  701.     const int A=(8-x)*(8-y);

  702.     const int B=(  x)*(8-y);

  703.     const int C=(8-x)*(  y);

  704.     const int D=(  x)*(  y);

  705.     int i;

  706. 

  707.     assert(x<8 && y<8 && x>=0 && y>=0);

  708. 

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

  710.     {

  711.         dst[0] = avg2(dst[0], ((A*src[0] + B*src[1] + C*src[stride+0] + D*src[stride+1] + 32 - 4) >> 6));

  712.         dst[1] = avg2(dst[1], ((A*src[1] + B*src[2] + C*src[stride+1] + D*src[stride+2] + 32 - 4) >> 6));

  713.         dst[2] = avg2(dst[2], ((A*src[2] + B*src[3] + C*src[stride+2] + D*src[stride+3] + 32 - 4) >> 6));

  714.         dst[3] = avg2(dst[3], ((A*src[3] + B*src[4] + C*src[stride+3] + D*src[stride+4] + 32 - 4) >> 6));

  715.         dst[4] = avg2(dst[4], ((A*src[4] + B*src[5] + C*src[stride+4] + D*src[stride+5] + 32 - 4) >> 6));

  716.         dst[5] = avg2(dst[5], ((A*src[5] + B*src[6] + C*src[stride+5] + D*src[stride+6] + 32 - 4) >> 6));

  717.         dst[6] = avg2(dst[6], ((A*src[6] + B*src[7] + C*src[stride+6] + D*src[stride+7] + 32 - 4) >> 6));

  718.         dst[7] = avg2(dst[7], ((A*src[7] + B*src[8] + C*src[stride+7] + D*src[stride+8] + 32 - 4) >> 6));

  719.         dst+= stride;

  720.         src+= stride;

  721.     }

  722. }

  723. 

  724. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  725. 

  726. static void sprite_h_c(uint8_t *dst, const uint8_t *src, int offset, int advance, int count)

  727. {

  728.     while (count--) {

  729.         int a = src[(offset >> 16)    ];

  730.         int b = src[(offset >> 16) + 1];

  731.         *dst++ = a + ((b - a) * (offset&0xFFFF) >> 16);

  732.         offset += advance;

  733.     }

  734. }

  735. 

  736. static av_always_inline void sprite_v_template(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset1,

  737.                                             int two_sprites, const uint8_t *src2a, const uint8_t *src2b, int offset2,

  738.                                             int alpha, int scaled, int width)

  739. {

  740.     int a1, b1, a2, b2;

  741.     while (width--) {

  742.         a1 = *src1a++;

  743.         if (scaled) {

  744.             b1 = *src1b++;

  745.             a1 = a1 + ((b1 - a1) * offset1 >> 16);

  746.         }

  747.         if (two_sprites) {

  748.             a2 = *src2a++;

  749.             if (scaled > 1) {

  750.                 b2 = *src2b++;

  751.                 a2 = a2 + ((b2 - a2) * offset2 >> 16);

  752.             }

  753.             a1 = a1 + ((a2 - a1) * alpha >> 16);

  754.         }

  755.         *dst++ = a1;

  756.     }

  757. }

  758. 

  759. static void sprite_v_single_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset, int width)

  760. {

  761.     sprite_v_template(dst, src1a, src1b, offset, 0, NULL, NULL, 0, 0, 1, width);

  762. }

  763. 

  764. static void sprite_v_double_noscale_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src2a, int alpha, int width)

  765. {

  766.     sprite_v_template(dst, src1a, NULL, 0, 1, src2a, NULL, 0, alpha, 0, width);

  767. }

  768. 

  769. static void sprite_v_double_onescale_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset1,

  770.                                                      const uint8_t *src2a, int alpha, int width)

  771. {

  772.     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, NULL, 0, alpha, 1, width);

  773. }

  774. 

  775. static void sprite_v_double_twoscale_c(uint8_t *dst, const uint8_t *src1a, const uint8_t *src1b, int offset1,

  776.                                                      const uint8_t *src2a, const uint8_t *src2b, int offset2,

  777.                                        int alpha, int width)

  778. {

  779.     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, src2b, offset2, alpha, 2, width);

  780. }

  781. 

  782. #endif

  783. 

  784. av_cold void ff_vc1dsp_init(VC1DSPContext* dsp) {

  785.     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;

  786.     dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;

  787.     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;

  788.     dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;

  789.     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;

  790.     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;

  791.     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;

  792.     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;

  793.     dsp->vc1_h_overlap = vc1_h_overlap_c;

  794.     dsp->vc1_v_overlap = vc1_v_overlap_c;

  795.     dsp->vc1_h_s_overlap = vc1_h_s_overlap_c;

  796.     dsp->vc1_v_s_overlap = vc1_v_s_overlap_c;

  797.     dsp->vc1_v_loop_filter4 = vc1_v_loop_filter4_c;

  798.     dsp->vc1_h_loop_filter4 = vc1_h_loop_filter4_c;

  799.     dsp->vc1_v_loop_filter8 = vc1_v_loop_filter8_c;

  800.     dsp->vc1_h_loop_filter8 = vc1_h_loop_filter8_c;

  801.     dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_c;

  802.     dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_c;

  803. 

  804.     dsp->put_vc1_mspel_pixels_tab[ 0] = ff_put_pixels8x8_c;

  805.     dsp->put_vc1_mspel_pixels_tab[ 1] = put_vc1_mspel_mc10_c;

  806.     dsp->put_vc1_mspel_pixels_tab[ 2] = put_vc1_mspel_mc20_c;

  807.     dsp->put_vc1_mspel_pixels_tab[ 3] = put_vc1_mspel_mc30_c;

  808.     dsp->put_vc1_mspel_pixels_tab[ 4] = put_vc1_mspel_mc01_c;

  809.     dsp->put_vc1_mspel_pixels_tab[ 5] = put_vc1_mspel_mc11_c;

  810.     dsp->put_vc1_mspel_pixels_tab[ 6] = put_vc1_mspel_mc21_c;

  811.     dsp->put_vc1_mspel_pixels_tab[ 7] = put_vc1_mspel_mc31_c;

  812.     dsp->put_vc1_mspel_pixels_tab[ 8] = put_vc1_mspel_mc02_c;

  813.     dsp->put_vc1_mspel_pixels_tab[ 9] = put_vc1_mspel_mc12_c;

  814.     dsp->put_vc1_mspel_pixels_tab[10] = put_vc1_mspel_mc22_c;

  815.     dsp->put_vc1_mspel_pixels_tab[11] = put_vc1_mspel_mc32_c;

  816.     dsp->put_vc1_mspel_pixels_tab[12] = put_vc1_mspel_mc03_c;

  817.     dsp->put_vc1_mspel_pixels_tab[13] = put_vc1_mspel_mc13_c;

  818.     dsp->put_vc1_mspel_pixels_tab[14] = put_vc1_mspel_mc23_c;

  819.     dsp->put_vc1_mspel_pixels_tab[15] = put_vc1_mspel_mc33_c;

  820. 

  821.     dsp->avg_vc1_mspel_pixels_tab[ 0] = ff_avg_pixels8x8_c;

  822.     dsp->avg_vc1_mspel_pixels_tab[ 1] = avg_vc1_mspel_mc10_c;

  823.     dsp->avg_vc1_mspel_pixels_tab[ 2] = avg_vc1_mspel_mc20_c;

  824.     dsp->avg_vc1_mspel_pixels_tab[ 3] = avg_vc1_mspel_mc30_c;

  825.     dsp->avg_vc1_mspel_pixels_tab[ 4] = avg_vc1_mspel_mc01_c;

  826.     dsp->avg_vc1_mspel_pixels_tab[ 5] = avg_vc1_mspel_mc11_c;

  827.     dsp->avg_vc1_mspel_pixels_tab[ 6] = avg_vc1_mspel_mc21_c;

  828.     dsp->avg_vc1_mspel_pixels_tab[ 7] = avg_vc1_mspel_mc31_c;

  829.     dsp->avg_vc1_mspel_pixels_tab[ 8] = avg_vc1_mspel_mc02_c;

  830.     dsp->avg_vc1_mspel_pixels_tab[ 9] = avg_vc1_mspel_mc12_c;

  831.     dsp->avg_vc1_mspel_pixels_tab[10] = avg_vc1_mspel_mc22_c;

  832.     dsp->avg_vc1_mspel_pixels_tab[11] = avg_vc1_mspel_mc32_c;

  833.     dsp->avg_vc1_mspel_pixels_tab[12] = avg_vc1_mspel_mc03_c;

  834.     dsp->avg_vc1_mspel_pixels_tab[13] = avg_vc1_mspel_mc13_c;

  835.     dsp->avg_vc1_mspel_pixels_tab[14] = avg_vc1_mspel_mc23_c;

  836.     dsp->avg_vc1_mspel_pixels_tab[15] = avg_vc1_mspel_mc33_c;

  837. 

  838.     dsp->put_no_rnd_vc1_chroma_pixels_tab[0]= put_no_rnd_vc1_chroma_mc8_c;

  839.     dsp->avg_no_rnd_vc1_chroma_pixels_tab[0]= avg_no_rnd_vc1_chroma_mc8_c;

  840.     dsp->put_no_rnd_vc1_chroma_pixels_tab[1] = put_no_rnd_vc1_chroma_mc4_c;

  841. 

  842. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  843.     dsp->sprite_h = sprite_h_c;

  844.     dsp->sprite_v_single = sprite_v_single_c;

  845.     dsp->sprite_v_double_noscale = sprite_v_double_noscale_c;

  846.     dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;

  847.     dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;

  848. #endif

  849. 

  850.     if (HAVE_ALTIVEC)

  851.         ff_vc1dsp_init_altivec(dsp);

  852.     if (ARCH_X86)

  853.         ff_vc1dsp_init_x86(dsp);

  854. }