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.
33469 Commits
32 Branches
283MB
Tree: 4b4a6ebd68
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '4b4a6ebd68'
${ noResults }
FFmpeg/libavcodec/vc1dsp.c

846 lines
27KB
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 FFmpeg.

  6.  *

  7.  * FFmpeg 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.  * FFmpeg 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 FFmpeg; 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. 

  30. 

  31. /** Apply overlap transform to horizontal edge

  32. */

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

  34. {

  35.     int i;

  36.     int a, b, c, d;

  37.     int d1, d2;

  38.     int rnd = 1;

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

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

  41.         b = src[-stride];

  42.         c = src[0];

  43.         d = src[stride];

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

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

  46. 

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

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

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

  50.         src[stride] = d + d1;

  51.         src++;

  52.         rnd = !rnd;

  53.     }

  54. }

  55. 

  56. /** Apply overlap transform to vertical edge

  57. */

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

  59. {

  60.     int i;

  61.     int a, b, c, d;

  62.     int d1, d2;

  63.     int rnd = 1;

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

  65.         a = src[-2];

  66.         b = src[-1];

  67.         c = src[0];

  68.         d = src[1];

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

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

  71. 

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

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

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

  75.         src[1] = d + d1;

  76.         src += stride;

  77.         rnd = !rnd;

  78.     }

  79. }

  80. 

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

  82. {

  83.     int i;

  84.     int a, b, c, d;

  85.     int d1, d2;

  86.     int rnd1 = 4, rnd2 = 3;

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

  88.         a = top[48];

  89.         b = top[56];

  90.         c = bottom[0];

  91.         d = bottom[8];

  92.         d1 = a - d;

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

  94. 

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

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

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

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

  99. 

  100.         bottom++;

  101.         top++;

  102.         rnd2 = 7 - rnd2;

  103.         rnd1 = 7 - rnd1;

  104.     }

  105. }

  106. 

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

  108. {

  109.     int i;

  110.     int a, b, c, d;

  111.     int d1, d2;

  112.     int rnd1 = 4, rnd2 = 3;

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

  114.         a = left[6];

  115.         b = left[7];

  116.         c = right[0];

  117.         d = right[1];

  118.         d1 = a - d;

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

  120. 

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

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

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

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

  125. 

  126.         right += 8;

  127.         left += 8;

  128.         rnd2 = 7 - rnd2;

  129.         rnd1 = 7 - rnd1;

  130.     }

  131. }

  132. 

  133. /**

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

  135.  * @param src source block type

  136.  * @param stride block stride

  137.  * @param pq block quantizer

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

  139.  * @see 8.6

  140.  */

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

  142.     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  143. 

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

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

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

  147.     if(a0 < pq){

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

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

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

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

  152.             int clip_sign = clip >> 31;

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

  154.             if(clip){

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

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

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

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

  159.                 d_sign ^= a0_sign;

  160. 

  161.                 if( d_sign ^ clip_sign )

  162.                     d = 0;

  163.                 else{

  164.                     d = FFMIN(d, clip);

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

  166.                     src[-1*stride] = cm[src[-1*stride] - d];

  167.                     src[ 0*stride] = cm[src[ 0*stride] + d];

  168.                 }

  169.                 return 1;

  170.             }

  171.         }

  172.     }

  173.     return 0;

  174. }

  175. 

  176. /**

  177.  * VC-1 in-loop deblocking filter

  178.  * @param src source block type

  179.  * @param step distance between horizontally adjacent elements

  180.  * @param stride distance between vertically adjacent elements

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

  182.  * @param pq block quantizer

  183.  * @see 8.6

  184.  */

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

  186. {

  187.     int i;

  188.     int filt3;

  189. 

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

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

  192.         if(filt3){

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

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

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

  196.         }

  197.         src += step * 4;

  198.     }

  199. }

  200. 

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

  202. {

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

  204. }

  205. 

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

  207. {

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

  209. }

  210. 

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

  212. {

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

  214. }

  215. 

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

  217. {

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

  219. }

  220. 

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

  222. {

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

  224. }

  225. 

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

  227. {

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

  229. }

  230. 

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

  232. */

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

  234. {

  235.     int i;

  236.     int dc = block[0];

  237.     const uint8_t *cm;

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

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

  240.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

  242.         dest[0] = cm[dest[0]];

  243.         dest[1] = cm[dest[1]];

  244.         dest[2] = cm[dest[2]];

  245.         dest[3] = cm[dest[3]];

  246.         dest[4] = cm[dest[4]];

  247.         dest[5] = cm[dest[5]];

  248.         dest[6] = cm[dest[6]];

  249.         dest[7] = cm[dest[7]];

  250.         dest += linesize;

  251.     }

  252. }

  253. 

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

  255. {

  256.     int i;

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

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

  259. 

  260.     src = block;

  261.     dst = temp;

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

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

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

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

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

  267. 

  268.         t5 = t1 + t3;

  269.         t6 = t2 + t4;

  270.         t7 = t2 - t4;

  271.         t8 = t1 - t3;

  272. 

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

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

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

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

  277. 

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

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

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

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

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

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

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

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

  286. 

  287.         src += 1;

  288.         dst += 8;

  289.     }

  290. 

  291.     src = temp;

  292.     dst = block;

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

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

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

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

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

  298. 

  299.         t5 = t1 + t3;

  300.         t6 = t2 + t4;

  301.         t7 = t2 - t4;

  302.         t8 = t1 - t3;

  303. 

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

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

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

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

  308. 

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

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

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

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

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

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

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

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

  317. 

  318.         src++;

  319.         dst++;

  320.     }

  321. }

  322. 

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

  324. */

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

  326. {

  327.     int i;

  328.     int dc = block[0];

  329.     const uint8_t *cm;

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

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

  332.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

  334.         dest[0] = cm[dest[0]];

  335.         dest[1] = cm[dest[1]];

  336.         dest[2] = cm[dest[2]];

  337.         dest[3] = cm[dest[3]];

  338.         dest[4] = cm[dest[4]];

  339.         dest[5] = cm[dest[5]];

  340.         dest[6] = cm[dest[6]];

  341.         dest[7] = cm[dest[7]];

  342.         dest += linesize;

  343.     }

  344. }

  345. 

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

  347. {

  348.     int i;

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

  350.     DCTELEM *src, *dst;

  351.     const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  352. 

  353.     src = block;

  354.     dst = block;

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

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

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

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

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

  360. 

  361.         t5 = t1 + t3;

  362.         t6 = t2 + t4;

  363.         t7 = t2 - t4;

  364.         t8 = t1 - t3;

  365. 

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

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

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

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

  370. 

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

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

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

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

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

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

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

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

  379. 

  380.         src += 8;

  381.         dst += 8;

  382.     }

  383. 

  384.     src = block;

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

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

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

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

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

  390. 

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

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

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

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

  395. 

  396.         src ++;

  397.         dest++;

  398.     }

  399. }

  400. 

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

  402. */

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

  404. {

  405.     int i;

  406.     int dc = block[0];

  407.     const uint8_t *cm;

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

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

  410.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

  412.         dest[0] = cm[dest[0]];

  413.         dest[1] = cm[dest[1]];

  414.         dest[2] = cm[dest[2]];

  415.         dest[3] = cm[dest[3]];

  416.         dest += linesize;

  417.     }

  418. }

  419. 

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

  421. {

  422.     int i;

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

  424.     DCTELEM *src, *dst;

  425.     const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  426. 

  427.     src = block;

  428.     dst = block;

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

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

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

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

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

  434. 

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

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

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

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

  439. 

  440.         src += 8;

  441.         dst += 8;

  442.     }

  443. 

  444.     src = block;

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

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

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

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

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

  450. 

  451.         t5 = t1 + t3;

  452.         t6 = t2 + t4;

  453.         t7 = t2 - t4;

  454.         t8 = t1 - t3;

  455. 

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

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

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

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

  460. 

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

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

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

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

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

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

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

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

  469. 

  470.         src ++;

  471.         dest++;

  472.     }

  473. }

  474. 

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

  476. */

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

  478. {

  479.     int i;

  480.     int dc = block[0];

  481.     const uint8_t *cm;

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

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

  484.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

  486.         dest[0] = cm[dest[0]];

  487.         dest[1] = cm[dest[1]];

  488.         dest[2] = cm[dest[2]];

  489.         dest[3] = cm[dest[3]];

  490.         dest += linesize;

  491.     }

  492. }

  493. 

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

  495. {

  496.     int i;

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

  498.     DCTELEM *src, *dst;

  499.     const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  500. 

  501.     src = block;

  502.     dst = block;

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

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

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

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

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

  508. 

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

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

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

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

  513. 

  514.         src += 8;

  515.         dst += 8;

  516.     }

  517. 

  518.     src = block;

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

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

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

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

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

  524. 

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

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

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

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

  529. 

  530.         src ++;

  531.         dest++;

  532.     }

  533. }

  534. 

  535. /* motion compensation functions */

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

  537. #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                 \

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

  539. {                                                                       \

  540.     switch(mode){                                                       \

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

  542.         return 0;                                                       \

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

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

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

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

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

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

  549.     }                                                                   \

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

  551. }

  552. 

  553. VC1_MSPEL_FILTER_16B(ver, uint8_t);

  554. VC1_MSPEL_FILTER_16B(hor, int16_t);

  555. 

  556. 

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

  558.  */

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

  560. {

  561.     switch(mode){

  562.     case 0: //no shift

  563.         return src[0];

  564.     case 1: // 1/4 shift

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

  566.     case 2: // 1/2 shift

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

  568.     case 3: // 3/4 shift

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

  570.     }

  571.     return 0; //should not occur

  572. }

  573. 

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

  575.  */

  576. #define VC1_MSPEL_MC(OP, OPNAME)\

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

  578. {\

  579.     int     i, j;\

  580. \

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

  582.         int r;\

  583. \

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

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

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

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

  588. \

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

  590. \

  591.             src -= 1;\

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

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

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

  595.                 src += stride;\

  596.                 tptr += 11;\

  597.             }\

  598. \

  599.             r = 64-rnd;\

  600.             tptr = tmp+1;\

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

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

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

  604.                 dst += stride;\

  605.                 tptr += 11;\

  606.             }\

  607. \

  608.             return;\

  609.         }\

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

  611.             r = 1-rnd;\

  612. \

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

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

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

  616.                 src += stride;\

  617.                 dst += stride;\

  618.             }\

  619.             return;\

  620.         }\

  621.     }\

  622. \

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

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

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

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

  627.         dst += stride;\

  628.         src += stride;\

  629.     }\

  630. }

  631. 

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

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

  634. 

  635. VC1_MSPEL_MC(op_put, put_)

  636. VC1_MSPEL_MC(op_avg, avg_)

  637. 

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

  639. 

  640. #define PUT_VC1_MSPEL(a, b)\

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

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

  643. }\

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

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

  646. }

  647. 

  648. PUT_VC1_MSPEL(1, 0)

  649. PUT_VC1_MSPEL(2, 0)

  650. PUT_VC1_MSPEL(3, 0)

  651. 

  652. PUT_VC1_MSPEL(0, 1)

  653. PUT_VC1_MSPEL(1, 1)

  654. PUT_VC1_MSPEL(2, 1)

  655. PUT_VC1_MSPEL(3, 1)

  656. 

  657. PUT_VC1_MSPEL(0, 2)

  658. PUT_VC1_MSPEL(1, 2)

  659. PUT_VC1_MSPEL(2, 2)

  660. PUT_VC1_MSPEL(3, 2)

  661. 

  662. PUT_VC1_MSPEL(0, 3)

  663. PUT_VC1_MSPEL(1, 3)

  664. PUT_VC1_MSPEL(2, 3)

  665. PUT_VC1_MSPEL(3, 3)

  666. 

  667. 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){

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

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

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

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

  672.     int i;

  673. 

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

  675. 

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

  677.     {

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

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

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

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

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

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

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

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

  686.         dst+= stride;

  687.         src+= stride;

  688.     }

  689. }

  690. 

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

  692. 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){

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

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

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

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

  697.     int i;

  698. 

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

  700. 

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

  702.     {

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

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

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

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

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

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

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

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

  711.         dst+= stride;

  712.         src+= stride;

  713.     }

  714. }

  715. 

  716. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  717. 

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

  719. {

  720.     while (count--) {

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

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

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

  724.         offset += advance;

  725.     }

  726. }

  727. 

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

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

  730.                                             int alpha, int scaled, int width)

  731. {

  732.     int a1, b1, a2, b2;

  733.     while (width--) {

  734.         a1 = *src1a++;

  735.         if (scaled) {

  736.             b1 = *src1b++;

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

  738.         }

  739.         if (two_sprites) {

  740.             a2 = *src2a++;

  741.             if (scaled > 1) {

  742.                 b2 = *src2b++;

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

  744.             }

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

  746.         }

  747.         *dst++ = a1;

  748.     }

  749. }

  750. 

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

  752. {

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

  754. }

  755. 

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

  757. {

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

  759. }

  760. 

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

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

  763. {

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

  765. }

  766. 

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

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

  769.                                        int alpha, int width)

  770. {

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

  772. }

  773. 

  774. #endif

  775. 

  776. av_cold void ff_vc1dsp_init(VC1DSPContext* dsp) {

  777.     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;

  778.     dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;

  779.     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;

  780.     dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;

  781.     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;

  782.     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;

  783.     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;

  784.     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;

  785.     dsp->vc1_h_overlap = vc1_h_overlap_c;

  786.     dsp->vc1_v_overlap = vc1_v_overlap_c;

  787.     dsp->vc1_h_s_overlap = vc1_h_s_overlap_c;

  788.     dsp->vc1_v_s_overlap = vc1_v_s_overlap_c;

  789.     dsp->vc1_v_loop_filter4 = vc1_v_loop_filter4_c;

  790.     dsp->vc1_h_loop_filter4 = vc1_h_loop_filter4_c;

  791.     dsp->vc1_v_loop_filter8 = vc1_v_loop_filter8_c;

  792.     dsp->vc1_h_loop_filter8 = vc1_h_loop_filter8_c;

  793.     dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_c;

  794.     dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_c;

  795. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  812. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  829. 

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

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

  832. 

  833. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  834.     dsp->sprite_h = sprite_h_c;

  835.     dsp->sprite_v_single = sprite_v_single_c;

  836.     dsp->sprite_v_double_noscale = sprite_v_double_noscale_c;

  837.     dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;

  838.     dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;

  839. #endif

  840. 

  841.     if (HAVE_ALTIVEC)

  842.         ff_vc1dsp_init_altivec(dsp);

  843.     if (HAVE_MMX)

  844.         ff_vc1dsp_init_mmx(dsp);

  845. }