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.
42730 Commits
32 Branches
283MB
Tree: b06cb15b9d
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b06cb15b9d'
${ noResults }
FFmpeg/libavcodec/vc1dsp.c

963 lines
34KB
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 "libavutil/common.h"

  29. #include "h264chroma.h"

  30. #include "qpeldsp.h"

  31. #include "vc1dsp.h"

  32. #include "startcode.h"

  33. 

  34. /* Apply overlap transform to horizontal edge */

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

  36. {

  37.     int i;

  38.     int a, b, c, d;

  39.     int d1, d2;

  40.     int rnd = 1;

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

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

  43.         b  = src[-stride];

  44.         c  = src[0];

  45.         d  = src[stride];

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

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

  48. 

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

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

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

  52.         src[stride]      = d + d1;

  53.         src++;

  54.         rnd = !rnd;

  55.     }

  56. }

  57. 

  58. /* Apply overlap transform to vertical edge */

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

  60. {

  61.     int i;

  62.     int a, b, c, d;

  63.     int d1, d2;

  64.     int rnd = 1;

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

  66.         a  = src[-2];

  67.         b  = src[-1];

  68.         c  = src[0];

  69.         d  = src[1];

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

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

  72. 

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

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

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

  76.         src[1]  = d + d1;

  77.         src    += stride;

  78.         rnd     = !rnd;

  79.     }

  80. }

  81. 

  82. static void vc1_v_s_overlap_c(int16_t *top, int16_t *bottom)

  83. {

  84.     int i;

  85.     int a, b, c, d;

  86.     int d1, d2;

  87.     int rnd1 = 4, rnd2 = 3;

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

  89.         a  = top[48];

  90.         b  = top[56];

  91.         c  = bottom[0];

  92.         d  = bottom[8];

  93.         d1 = a - d;

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

  95. 

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

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

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

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

  100. 

  101.         bottom++;

  102.         top++;

  103.         rnd2 = 7 - rnd2;

  104.         rnd1 = 7 - rnd1;

  105.     }

  106. }

  107. 

  108. static void vc1_h_s_overlap_c(int16_t *left, int16_t *right)

  109. {

  110.     int i;

  111.     int a, b, c, d;

  112.     int d1, d2;

  113.     int rnd1 = 4, rnd2 = 3;

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

  115.         a  = left[6];

  116.         b  = left[7];

  117.         c  = right[0];

  118.         d  = right[1];

  119.         d1 = a - d;

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

  121. 

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

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

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

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

  126. 

  127.         right += 8;

  128.         left  += 8;

  129.         rnd2   = 7 - rnd2;

  130.         rnd1   = 7 - rnd1;

  131.     }

  132. }

  133. 

  134. /**

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

  136.  * @param src source block type

  137.  * @param stride block stride

  138.  * @param pq block quantizer

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

  140.  * @see 8.6

  141.  */

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

  143. {

  144.     int a0 = (2 * (src[-2 * stride] - src[1 * stride]) -

  145.               5 * (src[-1 * stride] - src[0 * stride]) + 4) >> 3;

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

  147. 

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

  149.     if (a0 < pq) {

  150.         int a1 = FFABS((2 * (src[-4 * stride] - src[-1 * stride]) -

  151.                         5 * (src[-3 * stride] - src[-2 * stride]) + 4) >> 3);

  152.         int a2 = FFABS((2 * (src[ 0 * stride] - src[ 3 * stride]) -

  153.                         5 * (src[ 1 * stride] - src[ 2 * stride]) + 4) >> 3);

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

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

  156.             int clip_sign = clip >> 31;

  157. 

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

  159.             if (clip) {

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

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

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

  163. 

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

  165.                 d_sign ^= a0_sign;

  166. 

  167.                 if (d_sign ^ clip_sign)

  168.                     d = 0;

  169.                 else {

  170.                     d = FFMIN(d, clip);

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

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

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

  174.                 }

  175.                 return 1;

  176.             }

  177.         }

  178.     }

  179.     return 0;

  180. }

  181. 

  182. /**

  183.  * VC-1 in-loop deblocking filter

  184.  * @param src source block type

  185.  * @param step distance between horizontally adjacent elements

  186.  * @param stride distance between vertically adjacent elements

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

  188.  * @param pq block quantizer

  189.  * @see 8.6

  190.  */

  191. static inline void vc1_loop_filter(uint8_t *src, int step, int stride,

  192.                                    int len, int pq)

  193. {

  194.     int i;

  195.     int filt3;

  196. 

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

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

  199.         if (filt3) {

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

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

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

  203.         }

  204.         src += step * 4;

  205.     }

  206. }

  207. 

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

  209. {

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

  211. }

  212. 

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

  214. {

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

  216. }

  217. 

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

  219. {

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

  221. }

  222. 

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

  224. {

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

  226. }

  227. 

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

  229. {

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

  231. }

  232. 

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

  234. {

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

  236. }

  237. 

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

  239. static void vc1_inv_trans_8x8_dc_c(uint8_t *dest, int linesize, int16_t *block)

  240. {

  241.     int i;

  242.     int dc = block[0];

  243. 

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

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

  246. 

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

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

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

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

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

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

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

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

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

  256.         dest += linesize;

  257.     }

  258. }

  259. 

  260. static void vc1_inv_trans_8x8_c(int16_t block[64])

  261. {

  262.     int i;

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

  264.     int16_t *src, *dst, temp[64];

  265. 

  266.     src = block;

  267.     dst = temp;

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

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

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

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

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

  273. 

  274.         t5 = t1 + t3;

  275.         t6 = t2 + t4;

  276.         t7 = t2 - t4;

  277.         t8 = t1 - t3;

  278. 

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

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

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

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

  283. 

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

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

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

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

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

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

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

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

  292. 

  293.         src += 1;

  294.         dst += 8;

  295.     }

  296. 

  297.     src = temp;

  298.     dst = block;

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

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

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

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

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

  304. 

  305.         t5 = t1 + t3;

  306.         t6 = t2 + t4;

  307.         t7 = t2 - t4;

  308.         t8 = t1 - t3;

  309. 

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

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

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

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

  314. 

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

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

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

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

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

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

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

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

  323. 

  324.         src++;

  325.         dst++;

  326.     }

  327. }

  328. 

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

  330. static void vc1_inv_trans_8x4_dc_c(uint8_t *dest, int linesize, int16_t *block)

  331. {

  332.     int i;

  333.     int dc = block[0];

  334. 

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

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

  337. 

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

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

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

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

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

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

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

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

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

  347.         dest += linesize;

  348.     }

  349. }

  350. 

  351. static void vc1_inv_trans_8x4_c(uint8_t *dest, int linesize, int16_t *block)

  352. {

  353.     int i;

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

  355.     int16_t *src, *dst;

  356. 

  357.     src = block;

  358.     dst = block;

  359. 

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

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

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

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

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

  365. 

  366.         t5 = t1 + t3;

  367.         t6 = t2 + t4;

  368.         t7 = t2 - t4;

  369.         t8 = t1 - t3;

  370. 

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

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

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

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

  375. 

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

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

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

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

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

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

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

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

  384. 

  385.         src += 8;

  386.         dst += 8;

  387.     }

  388. 

  389.     src = block;

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

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

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

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

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

  395. 

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

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

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

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

  400. 

  401.         src++;

  402.         dest++;

  403.     }

  404. }

  405. 

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

  407. static void vc1_inv_trans_4x8_dc_c(uint8_t *dest, int linesize, int16_t *block)

  408. {

  409.     int i;

  410.     int dc = block[0];

  411. 

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

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

  414. 

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

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

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

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

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

  420.         dest += linesize;

  421.     }

  422. }

  423. 

  424. static void vc1_inv_trans_4x8_c(uint8_t *dest, int linesize, int16_t *block)

  425. {

  426.     int i;

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

  428.     int16_t *src, *dst;

  429. 

  430.     src = block;

  431.     dst = block;

  432. 

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

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

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

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

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

  438. 

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

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

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

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

  443. 

  444.         src += 8;

  445.         dst += 8;

  446.     }

  447. 

  448.     src = block;

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

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

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

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

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

  454. 

  455.         t5 = t1 + t3;

  456.         t6 = t2 + t4;

  457.         t7 = t2 - t4;

  458.         t8 = t1 - t3;

  459. 

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

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

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

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

  464. 

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

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

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

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

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

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

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

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

  473. 

  474.         src++;

  475.         dest++;

  476.     }

  477. }

  478. 

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

  480. static void vc1_inv_trans_4x4_dc_c(uint8_t *dest, int linesize, int16_t *block)

  481. {

  482.     int i;

  483.     int dc = block[0];

  484. 

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

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

  487. 

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

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

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

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

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

  493.         dest += linesize;

  494.     }

  495. }

  496. 

  497. static void vc1_inv_trans_4x4_c(uint8_t *dest, int linesize, int16_t *block)

  498. {

  499.     int i;

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

  501.     int16_t *src, *dst;

  502. 

  503.     src = block;

  504.     dst = block;

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

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

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

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

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

  510. 

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

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

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

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

  515. 

  516.         src += 8;

  517.         dst += 8;

  518.     }

  519. 

  520.     src = block;

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

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

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

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

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

  526. 

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

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

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

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

  531. 

  532.         src++;

  533.         dest++;

  534.     }

  535. }

  536. 

  537. /* motion compensation functions */

  538. 

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

  540. #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                       \

  541. static av_always_inline int vc1_mspel_ ## DIR ## _filter_16bits(const TYPE *src, \

  542.                                                                 int stride,   \

  543.                                                                 int mode)     \

  544. {                                                                             \

  545.     switch(mode) {                                                            \

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

  547.         return 0;                                                             \

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

  549.         return -4 * src[-stride] + 53 * src[0] +                              \

  550.                18 * src[stride]  -  3 * src[stride * 2];                      \

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

  552.         return -1 * src[-stride] +  9 * src[0] +                              \

  553.                 9 * src[stride]  -  1 * src[stride * 2];                      \

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

  555.         return -3 * src[-stride] + 18 * src[0] +                              \

  556.                53 * src[stride]  -  4 * src[stride * 2];                      \

  557.     }                                                                         \

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

  559. }

  560. 

  561. VC1_MSPEL_FILTER_16B(ver, uint8_t)

  562. VC1_MSPEL_FILTER_16B(hor, int16_t)

  563. 

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

  565. static av_always_inline int vc1_mspel_filter(const uint8_t *src, int stride,

  566.                                              int mode, int r)

  567. {

  568.     switch (mode) {

  569.     case 0: // no shift

  570.         return src[0];

  571.     case 1: // 1/4 shift

  572.         return (-4 * src[-stride] + 53 * src[0] +

  573.                 18 * src[stride]  -  3 * src[stride * 2] + 32 - r) >> 6;

  574.     case 2: // 1/2 shift

  575.         return (-1 * src[-stride] +  9 * src[0] +

  576.                  9 * src[stride]  -  1 * src[stride * 2] + 8 - r) >> 4;

  577.     case 3: // 3/4 shift

  578.         return (-3 * src[-stride] + 18 * src[0] +

  579.                 53 * src[stride]  -  4 * src[stride * 2] + 32 - r) >> 6;

  580.     }

  581.     return 0; // should not occur

  582. }

  583. 

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

  585. #define VC1_MSPEL_MC(OP, OPNAME)                                              \

  586. static av_always_inline void OPNAME ## vc1_mspel_mc(uint8_t *dst,             \

  587.                                                     const uint8_t *src,       \

  588.                                                     int stride,               \

  589.                                                     int hmode,                \

  590.                                                     int vmode,                \

  591.                                                     int rnd)                  \

  592. {                                                                             \

  593.     int i, j;                                                                 \

  594.                                                                               \

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

  596.         int r;                                                                \

  597.                                                                               \

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

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

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

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

  602.                                                                               \

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

  604.                                                                               \

  605.             src -= 1;                                                         \

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

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

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

  609.                 src  += stride;                                               \

  610.                 tptr += 11;                                                   \

  611.             }                                                                 \

  612.                                                                               \

  613.             r    = 64 - rnd;                                                  \

  614.             tptr = tmp + 1;                                                   \

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

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

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

  618.                 dst  += stride;                                               \

  619.                 tptr += 11;                                                   \

  620.             }                                                                 \

  621.                                                                               \

  622.             return;                                                           \

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

  624.             r = 1 - rnd;                                                      \

  625.                                                                               \

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

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

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

  629.                 src += stride;                                                \

  630.                 dst += stride;                                                \

  631.             }                                                                 \

  632.             return;                                                           \

  633.         }                                                                     \

  634.     }                                                                         \

  635.                                                                               \

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

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

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

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

  640.         dst += stride;                                                        \

  641.         src += stride;                                                        \

  642.     }                                                                         \

  643. }

  644. 

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

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

  647. 

  648. VC1_MSPEL_MC(op_put, put_)

  649. VC1_MSPEL_MC(op_avg, avg_)

  650. 

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

  652. 

  653. #define PUT_VC1_MSPEL(a, b)                                                   \

  654. static void put_vc1_mspel_mc ## a ## b ## _c(uint8_t *dst,                    \

  655.                                              const uint8_t *src,              \

  656.                                              ptrdiff_t stride, int rnd)       \

  657. {                                                                             \

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

  659. }                                                                             \

  660. static void avg_vc1_mspel_mc ## a ## b ## _c(uint8_t *dst,                    \

  661.                                              const uint8_t *src,              \

  662.                                              ptrdiff_t stride, int rnd)       \

  663. {                                                                             \

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

  665. }

  666. 

  667. PUT_VC1_MSPEL(1, 0)

  668. PUT_VC1_MSPEL(2, 0)

  669. PUT_VC1_MSPEL(3, 0)

  670. 

  671. PUT_VC1_MSPEL(0, 1)

  672. PUT_VC1_MSPEL(1, 1)

  673. PUT_VC1_MSPEL(2, 1)

  674. PUT_VC1_MSPEL(3, 1)

  675. 

  676. PUT_VC1_MSPEL(0, 2)

  677. PUT_VC1_MSPEL(1, 2)

  678. PUT_VC1_MSPEL(2, 2)

  679. PUT_VC1_MSPEL(3, 2)

  680. 

  681. PUT_VC1_MSPEL(0, 3)

  682. PUT_VC1_MSPEL(1, 3)

  683. PUT_VC1_MSPEL(2, 3)

  684. PUT_VC1_MSPEL(3, 3)

  685. 

  686. 

  687. static void put_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src,

  688.                                  ptrdiff_t stride, int rnd)

  689. {

  690.     ff_put_pixels8x8_c(dst, src, stride);

  691. }

  692. 

  693. static void avg_vc1_mspel_mc00_c(uint8_t *dst, const uint8_t *src,

  694.                                  ptrdiff_t stride, int rnd)

  695. {

  696.     ff_avg_pixels8x8_c(dst, src, stride);

  697. }

  698. 

  699. #define chroma_mc(a) \

  700.     ((A * src[a] + B * src[a + 1] + \

  701.       C * src[stride + a] + D * src[stride + a + 1] + 32 - 4) >> 6)

  702. static void put_no_rnd_vc1_chroma_mc8_c(uint8_t *dst /* align 8 */,

  703.                                         uint8_t *src /* align 1 */,

  704.                                         int stride, int h, int x, int y)

  705. {

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

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

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

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

  710.     int i;

  711. 

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

  713. 

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

  715.         dst[0] = chroma_mc(0);

  716.         dst[1] = chroma_mc(1);

  717.         dst[2] = chroma_mc(2);

  718.         dst[3] = chroma_mc(3);

  719.         dst[4] = chroma_mc(4);

  720.         dst[5] = chroma_mc(5);

  721.         dst[6] = chroma_mc(6);

  722.         dst[7] = chroma_mc(7);

  723.         dst += stride;

  724.         src += stride;

  725.     }

  726. }

  727. 

  728. static void put_no_rnd_vc1_chroma_mc4_c(uint8_t *dst, uint8_t *src,

  729.                                         int stride, int h, int x, int y)

  730. {

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

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

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

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

  735.     int i;

  736. 

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

  738. 

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

  740.         dst[0] = chroma_mc(0);

  741.         dst[1] = chroma_mc(1);

  742.         dst[2] = chroma_mc(2);

  743.         dst[3] = chroma_mc(3);

  744.         dst += stride;

  745.         src += stride;

  746.     }

  747. }

  748. 

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

  750. static void avg_no_rnd_vc1_chroma_mc8_c(uint8_t *dst /* align 8 */,

  751.                                         uint8_t *src /* align 1 */,

  752.                                         int stride, int h, int x, int y)

  753. {

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

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

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

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

  758.     int i;

  759. 

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

  761. 

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

  763.         dst[0] = avg2(dst[0], chroma_mc(0));

  764.         dst[1] = avg2(dst[1], chroma_mc(1));

  765.         dst[2] = avg2(dst[2], chroma_mc(2));

  766.         dst[3] = avg2(dst[3], chroma_mc(3));

  767.         dst[4] = avg2(dst[4], chroma_mc(4));

  768.         dst[5] = avg2(dst[5], chroma_mc(5));

  769.         dst[6] = avg2(dst[6], chroma_mc(6));

  770.         dst[7] = avg2(dst[7], chroma_mc(7));

  771.         dst += stride;

  772.         src += stride;

  773.     }

  774. }

  775. 

  776. static void avg_no_rnd_vc1_chroma_mc4_c(uint8_t *dst /* align 8 */,

  777.                                         uint8_t *src /* align 1 */,

  778.                                         int stride, int h, int x, int y)

  779. {

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

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

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

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

  784.     int i;

  785. 

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

  787. 

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

  789.         dst[0] = avg2(dst[0], chroma_mc(0));

  790.         dst[1] = avg2(dst[1], chroma_mc(1));

  791.         dst[2] = avg2(dst[2], chroma_mc(2));

  792.         dst[3] = avg2(dst[3], chroma_mc(3));

  793.         dst += stride;

  794.         src += stride;

  795.     }

  796. }

  797. 

  798. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  799. 

  800. static void sprite_h_c(uint8_t *dst, const uint8_t *src, int offset,

  801.                        int advance, int count)

  802. {

  803.     while (count--) {

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

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

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

  807.         offset += advance;

  808.     }

  809. }

  810. 

  811. static av_always_inline void sprite_v_template(uint8_t *dst,

  812.                                                const uint8_t *src1a,

  813.                                                const uint8_t *src1b,

  814.                                                int offset1,

  815.                                                int two_sprites,

  816.                                                const uint8_t *src2a,

  817.                                                const uint8_t *src2b,

  818.                                                int offset2,

  819.                                                int alpha, int scaled,

  820.                                                int width)

  821. {

  822.     int a1, b1, a2, b2;

  823.     while (width--) {

  824.         a1 = *src1a++;

  825.         if (scaled) {

  826.             b1 = *src1b++;

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

  828.         }

  829.         if (two_sprites) {

  830.             a2 = *src2a++;

  831.             if (scaled > 1) {

  832.                 b2 = *src2b++;

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

  834.             }

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

  836.         }

  837.         *dst++ = a1;

  838.     }

  839. }

  840. 

  841. static void sprite_v_single_c(uint8_t *dst, const uint8_t *src1a,

  842.                               const uint8_t *src1b,

  843.                               int offset, int width)

  844. {

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

  846. }

  847. 

  848. static void sprite_v_double_noscale_c(uint8_t *dst, const uint8_t *src1a,

  849.                                       const uint8_t *src2a,

  850.                                       int alpha, int width)

  851. {

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

  853. }

  854. 

  855. static void sprite_v_double_onescale_c(uint8_t *dst,

  856.                                        const uint8_t *src1a,

  857.                                        const uint8_t *src1b,

  858.                                        int offset1,

  859.                                        const uint8_t *src2a,

  860.                                        int alpha, int width)

  861. {

  862.     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, NULL, 0, alpha, 1,

  863.                       width);

  864. }

  865. 

  866. static void sprite_v_double_twoscale_c(uint8_t *dst,

  867.                                        const uint8_t *src1a,

  868.                                        const uint8_t *src1b,

  869.                                        int offset1,

  870.                                        const uint8_t *src2a,

  871.                                        const uint8_t *src2b,

  872.                                        int offset2,

  873.                                        int alpha,

  874.                                        int width)

  875. {

  876.     sprite_v_template(dst, src1a, src1b, offset1, 1, src2a, src2b, offset2,

  877.                       alpha, 2, width);

  878. }

  879. 

  880. #endif /* CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER */

  881. 

  882. av_cold void ff_vc1dsp_init(VC1DSPContext *dsp)

  883. {

  884.     dsp->vc1_inv_trans_8x8    = vc1_inv_trans_8x8_c;

  885.     dsp->vc1_inv_trans_4x8    = vc1_inv_trans_4x8_c;

  886.     dsp->vc1_inv_trans_8x4    = vc1_inv_trans_8x4_c;

  887.     dsp->vc1_inv_trans_4x4    = vc1_inv_trans_4x4_c;

  888.     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;

  889.     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;

  890.     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;

  891.     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;

  892. 

  893.     dsp->vc1_h_overlap        = vc1_h_overlap_c;

  894.     dsp->vc1_v_overlap        = vc1_v_overlap_c;

  895.     dsp->vc1_h_s_overlap      = vc1_h_s_overlap_c;

  896.     dsp->vc1_v_s_overlap      = vc1_v_s_overlap_c;

  897. 

  898.     dsp->vc1_v_loop_filter4   = vc1_v_loop_filter4_c;

  899.     dsp->vc1_h_loop_filter4   = vc1_h_loop_filter4_c;

  900.     dsp->vc1_v_loop_filter8   = vc1_v_loop_filter8_c;

  901.     dsp->vc1_h_loop_filter8   = vc1_h_loop_filter8_c;

  902.     dsp->vc1_v_loop_filter16  = vc1_v_loop_filter16_c;

  903.     dsp->vc1_h_loop_filter16  = vc1_h_loop_filter16_c;

  904. 

  905.     dsp->put_vc1_mspel_pixels_tab[0]  = put_vc1_mspel_mc00_c;

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  921. 

  922.     dsp->avg_vc1_mspel_pixels_tab[0]  = avg_vc1_mspel_mc00_c;

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  938. 

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

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

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

  942.     dsp->avg_no_rnd_vc1_chroma_pixels_tab[1] = avg_no_rnd_vc1_chroma_mc4_c;

  943. 

  944. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  945.     dsp->sprite_h                 = sprite_h_c;

  946.     dsp->sprite_v_single          = sprite_v_single_c;

  947.     dsp->sprite_v_double_noscale  = sprite_v_double_noscale_c;

  948.     dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;

  949.     dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;

  950. #endif /* CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER */

  951. 

  952.     dsp->startcode_find_candidate = ff_startcode_find_candidate_c;

  953. 

  954.     if (ARCH_AARCH64)

  955.         ff_vc1dsp_init_aarch64(dsp);

  956.     if (ARCH_ARM)

  957.         ff_vc1dsp_init_arm(dsp);

  958.     if (ARCH_PPC)

  959.         ff_vc1dsp_init_ppc(dsp);

  960.     if (ARCH_X86)

  961.         ff_vc1dsp_init_x86(dsp);

  962. }