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.
34063 Commits
32 Branches
283MB
Tree: 25b51b2c44
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '25b51b2c44'
${ noResults }
FFmpeg/libavcodec/vc1dsp.c

854 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. 

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

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

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

  145.     if(a0 < pq){

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

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

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

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

  150.             int clip_sign = clip >> 31;

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

  152.             if(clip){

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

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

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

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

  157.                 d_sign ^= a0_sign;

  158. 

  159.                 if( d_sign ^ clip_sign )

  160.                     d = 0;

  161.                 else{

  162.                     d = FFMIN(d, clip);

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

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

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

  166.                 }

  167.                 return 1;

  168.             }

  169.         }

  170.     }

  171.     return 0;

  172. }

  173. 

  174. /**

  175.  * VC-1 in-loop deblocking filter

  176.  * @param src source block type

  177.  * @param step distance between horizontally adjacent elements

  178.  * @param stride distance between vertically adjacent elements

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

  180.  * @param pq block quantizer

  181.  * @see 8.6

  182.  */

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

  184. {

  185.     int i;

  186.     int filt3;

  187. 

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

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

  190.         if(filt3){

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

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

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

  194.         }

  195.         src += step * 4;

  196.     }

  197. }

  198. 

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

  200. {

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

  202. }

  203. 

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

  205. {

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

  207. }

  208. 

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

  210. {

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

  212. }

  213. 

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

  215. {

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

  217. }

  218. 

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

  220. {

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

  222. }

  223. 

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

  225. {

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

  227. }

  228. 

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

  230. */

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

  232. {

  233.     int i;

  234.     int dc = block[0];

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

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

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

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

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

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

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

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

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

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

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

  246.         dest += linesize;

  247.     }

  248. }

  249. 

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

  251. {

  252.     int i;

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

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

  255. 

  256.     src = block;

  257.     dst = temp;

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

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

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

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

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

  263. 

  264.         t5 = t1 + t3;

  265.         t6 = t2 + t4;

  266.         t7 = t2 - t4;

  267.         t8 = t1 - t3;

  268. 

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

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

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

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

  273. 

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

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

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

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

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

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

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

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

  282. 

  283.         src += 1;

  284.         dst += 8;

  285.     }

  286. 

  287.     src = temp;

  288.     dst = block;

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

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

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

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

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

  294. 

  295.         t5 = t1 + t3;

  296.         t6 = t2 + t4;

  297.         t7 = t2 - t4;

  298.         t8 = t1 - t3;

  299. 

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

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

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

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

  304. 

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

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

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

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

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

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

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

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

  313. 

  314.         src++;

  315.         dst++;

  316.     }

  317. }

  318. 

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

  320. */

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

  322. {

  323.     int i;

  324.     int dc = block[0];

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

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

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

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

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

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

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

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

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

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

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

  336.         dest += linesize;

  337.     }

  338. }

  339. 

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

  341. {

  342.     int i;

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

  344.     DCTELEM *src, *dst;

  345. 

  346.     src = block;

  347.     dst = block;

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

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

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

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

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

  353. 

  354.         t5 = t1 + t3;

  355.         t6 = t2 + t4;

  356.         t7 = t2 - t4;

  357.         t8 = t1 - t3;

  358. 

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

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

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

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

  363. 

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

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

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

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

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

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

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

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

  372. 

  373.         src += 8;

  374.         dst += 8;

  375.     }

  376. 

  377.     src = block;

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

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

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

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

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

  383. 

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

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

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

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

  388. 

  389.         src ++;

  390.         dest++;

  391.     }

  392. }

  393. 

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

  395. */

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

  397. {

  398.     int i;

  399.     int dc = block[0];

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

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

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

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

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

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

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

  407.         dest += linesize;

  408.     }

  409. }

  410. 

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

  412. {

  413.     int i;

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

  415.     DCTELEM *src, *dst;

  416. 

  417.     src = block;

  418.     dst = block;

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

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

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

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

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

  424. 

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

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

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

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

  429. 

  430.         src += 8;

  431.         dst += 8;

  432.     }

  433. 

  434.     src = block;

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

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

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

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

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

  440. 

  441.         t5 = t1 + t3;

  442.         t6 = t2 + t4;

  443.         t7 = t2 - t4;

  444.         t8 = t1 - t3;

  445. 

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

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

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

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

  450. 

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

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

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

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

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

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

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

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

  459. 

  460.         src ++;

  461.         dest++;

  462.     }

  463. }

  464. 

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

  466. */

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

  468. {

  469.     int i;

  470.     int dc = block[0];

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

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

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

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

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

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

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

  478.         dest += linesize;

  479.     }

  480. }

  481. 

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

  483. {

  484.     int i;

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

  486.     DCTELEM *src, *dst;

  487. 

  488.     src = block;

  489.     dst = block;

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

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

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

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

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

  495. 

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

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

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

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

  500. 

  501.         src += 8;

  502.         dst += 8;

  503.     }

  504. 

  505.     src = block;

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

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

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

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

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

  511. 

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

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

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

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

  516. 

  517.         src ++;

  518.         dest++;

  519.     }

  520. }

  521. 

  522. /* motion compensation functions */

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

  524. #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                 \

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

  526. {                                                                       \

  527.     switch(mode){                                                       \

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

  529.         return 0;                                                       \

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

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

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

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

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

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

  536.     }                                                                   \

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

  538. }

  539. 

  540. VC1_MSPEL_FILTER_16B(ver, uint8_t)

  541. VC1_MSPEL_FILTER_16B(hor, int16_t)

  542. 

  543. 

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

  545.  */

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

  547. {

  548.     switch(mode){

  549.     case 0: //no shift

  550.         return src[0];

  551.     case 1: // 1/4 shift

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

  553.     case 2: // 1/2 shift

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

  555.     case 3: // 3/4 shift

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

  557.     }

  558.     return 0; //should not occur

  559. }

  560. 

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

  562.  */

  563. #define VC1_MSPEL_MC(OP, OPNAME)\

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

  565. {\

  566.     int     i, j;\

  567. \

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

  569.         int r;\

  570. \

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

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

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

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

  575. \

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

  577. \

  578.             src -= 1;\

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

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

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

  582.                 src += stride;\

  583.                 tptr += 11;\

  584.             }\

  585. \

  586.             r = 64-rnd;\

  587.             tptr = tmp+1;\

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

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

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

  591.                 dst += stride;\

  592.                 tptr += 11;\

  593.             }\

  594. \

  595.             return;\

  596.         }\

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

  598.             r = 1-rnd;\

  599. \

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

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

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

  603.                 src += stride;\

  604.                 dst += stride;\

  605.             }\

  606.             return;\

  607.         }\

  608.     }\

  609. \

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

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

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

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

  614.         dst += stride;\

  615.         src += stride;\

  616.     }\

  617. }

  618. 

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

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

  621. 

  622. VC1_MSPEL_MC(op_put, put_)

  623. VC1_MSPEL_MC(op_avg, avg_)

  624. 

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

  626. 

  627. #define PUT_VC1_MSPEL(a, b)\

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

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

  630. }\

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

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

  633. }

  634. 

  635. PUT_VC1_MSPEL(1, 0)

  636. PUT_VC1_MSPEL(2, 0)

  637. PUT_VC1_MSPEL(3, 0)

  638. 

  639. PUT_VC1_MSPEL(0, 1)

  640. PUT_VC1_MSPEL(1, 1)

  641. PUT_VC1_MSPEL(2, 1)

  642. PUT_VC1_MSPEL(3, 1)

  643. 

  644. PUT_VC1_MSPEL(0, 2)

  645. PUT_VC1_MSPEL(1, 2)

  646. PUT_VC1_MSPEL(2, 2)

  647. PUT_VC1_MSPEL(3, 2)

  648. 

  649. PUT_VC1_MSPEL(0, 3)

  650. PUT_VC1_MSPEL(1, 3)

  651. PUT_VC1_MSPEL(2, 3)

  652. PUT_VC1_MSPEL(3, 3)

  653. 

  654. 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){

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

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

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

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

  659.     int i;

  660. 

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

  662. 

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

  664.     {

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

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

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

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

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

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

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

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

  673.         dst+= stride;

  674.         src+= stride;

  675.     }

  676. }

  677. 

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

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

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

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

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

  683.     int i;

  684. 

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

  686. 

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

  688.     {

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

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

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

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

  693.         dst+= stride;

  694.         src+= stride;

  695.     }

  696. }

  697. 

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

  699. 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){

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

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

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

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

  704.     int i;

  705. 

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

  707. 

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

  709.     {

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

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

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

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

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

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

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

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

  718.         dst+= stride;

  719.         src+= stride;

  720.     }

  721. }

  722. 

  723. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  724. 

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

  726. {

  727.     while (count--) {

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

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

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

  731.         offset += advance;

  732.     }

  733. }

  734. 

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

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

  737.                                             int alpha, int scaled, int width)

  738. {

  739.     int a1, b1, a2, b2;

  740.     while (width--) {

  741.         a1 = *src1a++;

  742.         if (scaled) {

  743.             b1 = *src1b++;

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

  745.         }

  746.         if (two_sprites) {

  747.             a2 = *src2a++;

  748.             if (scaled > 1) {

  749.                 b2 = *src2b++;

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

  751.             }

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

  753.         }

  754.         *dst++ = a1;

  755.     }

  756. }

  757. 

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

  759. {

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

  761. }

  762. 

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

  764. {

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

  766. }

  767. 

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

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

  770. {

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

  772. }

  773. 

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

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

  776.                                        int alpha, int width)

  777. {

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

  779. }

  780. 

  781. #endif

  782. 

  783. av_cold void ff_vc1dsp_init(VC1DSPContext* dsp) {

  784.     dsp->vc1_inv_trans_8x8 = vc1_inv_trans_8x8_c;

  785.     dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;

  786.     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;

  787.     dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;

  788.     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;

  789.     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;

  790.     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;

  791.     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;

  792.     dsp->vc1_h_overlap = vc1_h_overlap_c;

  793.     dsp->vc1_v_overlap = vc1_v_overlap_c;

  794.     dsp->vc1_h_s_overlap = vc1_h_s_overlap_c;

  795.     dsp->vc1_v_s_overlap = vc1_v_s_overlap_c;

  796.     dsp->vc1_v_loop_filter4 = vc1_v_loop_filter4_c;

  797.     dsp->vc1_h_loop_filter4 = vc1_h_loop_filter4_c;

  798.     dsp->vc1_v_loop_filter8 = vc1_v_loop_filter8_c;

  799.     dsp->vc1_h_loop_filter8 = vc1_h_loop_filter8_c;

  800.     dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_c;

  801.     dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_c;

  802. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  819. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  836. 

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

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

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

  840. 

  841. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  842.     dsp->sprite_h = sprite_h_c;

  843.     dsp->sprite_v_single = sprite_v_single_c;

  844.     dsp->sprite_v_double_noscale = sprite_v_double_noscale_c;

  845.     dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;

  846.     dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;

  847. #endif

  848. 

  849.     if (HAVE_ALTIVEC)

  850.         ff_vc1dsp_init_altivec(dsp);

  851.     if (HAVE_MMX)

  852.         ff_vc1dsp_init_mmx(dsp);

  853. }