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.
28324 Commits
32 Branches
766MB
Tree: 13dfce3d44
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '13dfce3d44'
${ noResults }
FFmpeg/libavcodec/vc1dsp.c

758 lines
25KB
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. /**

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

  83.  * @param src source block type

  84.  * @param stride block stride

  85.  * @param pq block quantizer

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

  87.  * @see 8.6

  88.  */

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

  90.     uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  91. 

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

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

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

  95.     if(a0 < pq){

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

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

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

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

  100.             int clip_sign = clip >> 31;

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

  102.             if(clip){

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

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

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

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

  107.                 d_sign ^= a0_sign;

  108. 

  109.                 if( d_sign ^ clip_sign )

  110.                     d = 0;

  111.                 else{

  112.                     d = FFMIN(d, clip);

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

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

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

  116.                 }

  117.                 return 1;

  118.             }

  119.         }

  120.     }

  121.     return 0;

  122. }

  123. 

  124. /**

  125.  * VC-1 in-loop deblocking filter

  126.  * @param src source block type

  127.  * @param step distance between horizontally adjacent elements

  128.  * @param stride distance between vertically adjacent elements

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

  130.  * @param pq block quantizer

  131.  * @see 8.6

  132.  */

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

  134. {

  135.     int i;

  136.     int filt3;

  137. 

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

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

  140.         if(filt3){

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

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

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

  144.         }

  145.         src += step * 4;

  146.     }

  147. }

  148. 

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

  150. {

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

  152. }

  153. 

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

  155. {

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

  157. }

  158. 

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

  160. {

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

  162. }

  163. 

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

  165. {

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

  167. }

  168. 

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

  170. {

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

  172. }

  173. 

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

  175. {

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

  177. }

  178. 

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

  180. */

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

  182. {

  183.     int i;

  184.     int dc = block[0];

  185.     const uint8_t *cm;

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

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

  188.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

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

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

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

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

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

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

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

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

  198.         dest += linesize;

  199.     }

  200. }

  201. 

  202. static av_always_inline void vc1_inv_trans_8x8_c(DCTELEM block[64], int shl, int sub)

  203. {

  204.     int i;

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

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

  207. 

  208.     src = block;

  209.     dst = temp;

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

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

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

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

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

  215. 

  216.         t5 = t1 + t3;

  217.         t6 = t2 + t4;

  218.         t7 = t2 - t4;

  219.         t8 = t1 - t3;

  220. 

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

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

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

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

  225. 

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

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

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

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

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

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

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

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

  234. 

  235.         src += 1;

  236.         dst += 8;

  237.     }

  238. 

  239.     src = temp;

  240.     dst = block;

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

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

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

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

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

  246. 

  247.         t5 = t1 + t3;

  248.         t6 = t2 + t4;

  249.         t7 = t2 - t4;

  250.         t8 = t1 - t3;

  251. 

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

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

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

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

  256. 

  257.         dst[ 0] = (((t5 + t1    ) >> 7) - sub) << shl;

  258.         dst[ 8] = (((t6 + t2    ) >> 7) - sub) << shl;

  259.         dst[16] = (((t7 + t3    ) >> 7) - sub) << shl;

  260.         dst[24] = (((t8 + t4    ) >> 7) - sub) << shl;

  261.         dst[32] = (((t8 - t4 + 1) >> 7) - sub) << shl;

  262.         dst[40] = (((t7 - t3 + 1) >> 7) - sub) << shl;

  263.         dst[48] = (((t6 - t2 + 1) >> 7) - sub) << shl;

  264.         dst[56] = (((t5 - t1 + 1) >> 7) - sub) << shl;

  265. 

  266.         src++;

  267.         dst++;

  268.     }

  269. }

  270. 

  271. static void vc1_inv_trans_8x8_add_c(uint8_t *dest, int linesize, DCTELEM *block)

  272. {

  273.     vc1_inv_trans_8x8_c(block, 0, 0);

  274.     ff_add_pixels_clamped_c(block, dest, linesize);

  275. }

  276. 

  277. static void vc1_inv_trans_8x8_put_signed_c(uint8_t *dest, int linesize, DCTELEM *block)

  278. {

  279.     vc1_inv_trans_8x8_c(block, 0, 0);

  280.     ff_put_signed_pixels_clamped_c(block, dest, linesize);

  281. }

  282. 

  283. static void vc1_inv_trans_8x8_put_signed_rangered_c(uint8_t *dest, int linesize, DCTELEM *block)

  284. {

  285.     vc1_inv_trans_8x8_c(block, 1, 0);

  286.     ff_put_signed_pixels_clamped_c(block, dest, linesize);

  287. }

  288. 

  289. static void vc1_inv_trans_8x8_put_c(uint8_t *dest, int linesize, DCTELEM *block)

  290. {

  291.     vc1_inv_trans_8x8_c(block, 0, 0);

  292.     ff_put_pixels_clamped_c(block, dest, linesize);

  293. }

  294. 

  295. static void vc1_inv_trans_8x8_put_rangered_c(uint8_t *dest, int linesize, DCTELEM *block)

  296. {

  297.     vc1_inv_trans_8x8_c(block, 1, 64);

  298.     ff_put_pixels_clamped_c(block, dest, linesize);

  299. }

  300. 

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

  302. */

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

  304. {

  305.     int i;

  306.     int dc = block[0];

  307.     const uint8_t *cm;

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

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

  310.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

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

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

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

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

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

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

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

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

  320.         dest += linesize;

  321.     }

  322. }

  323. 

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

  325. {

  326.     int i;

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

  328.     DCTELEM *src, *dst;

  329.     const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  330. 

  331.     src = block;

  332.     dst = block;

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

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

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

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

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

  338. 

  339.         t5 = t1 + t3;

  340.         t6 = t2 + t4;

  341.         t7 = t2 - t4;

  342.         t8 = t1 - t3;

  343. 

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

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

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

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

  348. 

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

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

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

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

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

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

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

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

  357. 

  358.         src += 8;

  359.         dst += 8;

  360.     }

  361. 

  362.     src = block;

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

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

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

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

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

  368. 

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

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

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

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

  373. 

  374.         src ++;

  375.         dest++;

  376.     }

  377. }

  378. 

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

  380. */

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

  382. {

  383.     int i;

  384.     int dc = block[0];

  385.     const uint8_t *cm;

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

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

  388.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

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

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

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

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

  394.         dest += linesize;

  395.     }

  396. }

  397. 

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

  399. {

  400.     int i;

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

  402.     DCTELEM *src, *dst;

  403.     const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  404. 

  405.     src = block;

  406.     dst = block;

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

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

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

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

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

  412. 

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

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

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

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

  417. 

  418.         src += 8;

  419.         dst += 8;

  420.     }

  421. 

  422.     src = block;

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

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

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

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

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

  428. 

  429.         t5 = t1 + t3;

  430.         t6 = t2 + t4;

  431.         t7 = t2 - t4;

  432.         t8 = t1 - t3;

  433. 

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

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

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

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

  438. 

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

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

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

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

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

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

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

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

  447. 

  448.         src ++;

  449.         dest++;

  450.     }

  451. }

  452. 

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

  454. */

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

  456. {

  457.     int i;

  458.     int dc = block[0];

  459.     const uint8_t *cm;

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

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

  462.     cm = ff_cropTbl + MAX_NEG_CROP + dc;

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

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

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

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

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

  468.         dest += linesize;

  469.     }

  470. }

  471. 

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

  473. {

  474.     int i;

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

  476.     DCTELEM *src, *dst;

  477.     const uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;

  478. 

  479.     src = block;

  480.     dst = block;

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

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

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

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

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

  486. 

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

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

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

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

  491. 

  492.         src += 8;

  493.         dst += 8;

  494.     }

  495. 

  496.     src = block;

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

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

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

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

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

  502. 

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

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

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

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

  507. 

  508.         src ++;

  509.         dest++;

  510.     }

  511. }

  512. 

  513. /* motion compensation functions */

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

  515. #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                 \

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

  517. {                                                                       \

  518.     switch(mode){                                                       \

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

  520.         return 0;                                                       \

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

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

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

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

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

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

  527.     }                                                                   \

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

  529. }

  530. 

  531. VC1_MSPEL_FILTER_16B(ver, uint8_t);

  532. VC1_MSPEL_FILTER_16B(hor, int16_t);

  533. 

  534. 

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

  536.  */

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

  538. {

  539.     switch(mode){

  540.     case 0: //no shift

  541.         return src[0];

  542.     case 1: // 1/4 shift

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

  544.     case 2: // 1/2 shift

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

  546.     case 3: // 3/4 shift

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

  548.     }

  549.     return 0; //should not occur

  550. }

  551. 

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

  553.  */

  554. #define VC1_MSPEL_MC(OP, OPNAME)\

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

  556. {\

  557.     int     i, j;\

  558. \

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

  560.         int r;\

  561. \

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

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

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

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

  566. \

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

  568. \

  569.             src -= 1;\

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

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

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

  573.                 src += stride;\

  574.                 tptr += 11;\

  575.             }\

  576. \

  577.             r = 64-rnd;\

  578.             tptr = tmp+1;\

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

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

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

  582.                 dst += stride;\

  583.                 tptr += 11;\

  584.             }\

  585. \

  586.             return;\

  587.         }\

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

  589.             r = 1-rnd;\

  590. \

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

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

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

  594.                 src += stride;\

  595.                 dst += stride;\

  596.             }\

  597.             return;\

  598.         }\

  599.     }\

  600. \

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

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

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

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

  605.         dst += stride;\

  606.         src += stride;\

  607.     }\

  608. }

  609. 

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

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

  612. 

  613. VC1_MSPEL_MC(op_put, put_)

  614. VC1_MSPEL_MC(op_avg, avg_)

  615. 

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

  617. 

  618. #define PUT_VC1_MSPEL(a, b)\

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

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

  621. }\

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

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

  624. }

  625. 

  626. PUT_VC1_MSPEL(1, 0)

  627. PUT_VC1_MSPEL(2, 0)

  628. PUT_VC1_MSPEL(3, 0)

  629. 

  630. PUT_VC1_MSPEL(0, 1)

  631. PUT_VC1_MSPEL(1, 1)

  632. PUT_VC1_MSPEL(2, 1)

  633. PUT_VC1_MSPEL(3, 1)

  634. 

  635. PUT_VC1_MSPEL(0, 2)

  636. PUT_VC1_MSPEL(1, 2)

  637. PUT_VC1_MSPEL(2, 2)

  638. PUT_VC1_MSPEL(3, 2)

  639. 

  640. PUT_VC1_MSPEL(0, 3)

  641. PUT_VC1_MSPEL(1, 3)

  642. PUT_VC1_MSPEL(2, 3)

  643. PUT_VC1_MSPEL(3, 3)

  644. 

  645. 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){

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

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

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

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

  650.     int i;

  651. 

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

  653. 

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

  655.     {

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

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

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

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

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

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

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

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

  664.         dst+= stride;

  665.         src+= stride;

  666.     }

  667. }

  668. 

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

  670. 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){

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

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

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

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

  675.     int i;

  676. 

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

  678. 

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

  680.     {

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

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

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

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

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

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

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

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

  689.         dst+= stride;

  690.         src+= stride;

  691.     }

  692. }

  693. 

  694. av_cold void ff_vc1dsp_init(VC1DSPContext* dsp) {

  695.     dsp->vc1_inv_trans_8x8_add = vc1_inv_trans_8x8_add_c;

  696.     dsp->vc1_inv_trans_8x8_put_signed[0] = vc1_inv_trans_8x8_put_signed_c;

  697.     dsp->vc1_inv_trans_8x8_put_signed[1] = vc1_inv_trans_8x8_put_signed_rangered_c;

  698.     dsp->vc1_inv_trans_8x8_put[0] = vc1_inv_trans_8x8_put_c;

  699.     dsp->vc1_inv_trans_8x8_put[1] = vc1_inv_trans_8x8_put_rangered_c;

  700.     dsp->vc1_inv_trans_4x8 = vc1_inv_trans_4x8_c;

  701.     dsp->vc1_inv_trans_8x4 = vc1_inv_trans_8x4_c;

  702.     dsp->vc1_inv_trans_4x4 = vc1_inv_trans_4x4_c;

  703.     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;

  704.     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;

  705.     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;

  706.     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;

  707.     dsp->vc1_h_overlap = vc1_h_overlap_c;

  708.     dsp->vc1_v_overlap = vc1_v_overlap_c;

  709.     dsp->vc1_v_loop_filter4 = vc1_v_loop_filter4_c;

  710.     dsp->vc1_h_loop_filter4 = vc1_h_loop_filter4_c;

  711.     dsp->vc1_v_loop_filter8 = vc1_v_loop_filter8_c;

  712.     dsp->vc1_h_loop_filter8 = vc1_h_loop_filter8_c;

  713.     dsp->vc1_v_loop_filter16 = vc1_v_loop_filter16_c;

  714.     dsp->vc1_h_loop_filter16 = vc1_h_loop_filter16_c;

  715. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  732. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  749. 

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

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

  752. 

  753.     if (HAVE_ALTIVEC)

  754.         ff_vc1dsp_init_altivec(dsp);

  755.     if (HAVE_MMX)

  756.         ff_vc1dsp_init_mmx(dsp);

  757. }