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.
44689 Commits
32 Branches
824MB
Tree: c5714b51aa
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c5714b51aa'
${ noResults }
FFmpeg/libavcodec/vc1dsp.c

962 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. #include "libavutil/common.h"

  28. #include "h264chroma.h"

  29. #include "qpeldsp.h"

  30. #include "vc1dsp.h"

  31. #include "startcode.h"

  32. 

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

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

  35. {

  36.     int i;

  37.     int a, b, c, d;

  38.     int d1, d2;

  39.     int rnd = 1;

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

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

  42.         b  = src[-stride];

  43.         c  = src[0];

  44.         d  = src[stride];

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

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

  47. 

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

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

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

  51.         src[stride]      = d + d1;

  52.         src++;

  53.         rnd = !rnd;

  54.     }

  55. }

  56. 

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

  58. 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(int16_t *top, int16_t *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(int16_t *left, int16_t *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. {

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

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

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

  146. 

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

  148.     if (a0 < pq) {

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

  150.                         5 * (src[-3 * stride] - src[-2 * stride]) + 4) >> 3);

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

  152.                         5 * (src[ 1 * stride] - src[ 2 * stride]) + 4) >> 3);

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

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

  155.             int clip_sign = clip >> 31;

  156. 

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

  158.             if (clip) {

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

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

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

  162. 

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

  164.                 d_sign ^= a0_sign;

  165. 

  166.                 if (d_sign ^ clip_sign)

  167.                     d = 0;

  168.                 else {

  169.                     d = FFMIN(d, clip);

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

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

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

  173.                 }

  174.                 return 1;

  175.             }

  176.         }

  177.     }

  178.     return 0;

  179. }

  180. 

  181. /**

  182.  * VC-1 in-loop deblocking filter

  183.  * @param src source block type

  184.  * @param step distance between horizontally adjacent elements

  185.  * @param stride distance between vertically adjacent elements

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

  187.  * @param pq block quantizer

  188.  * @see 8.6

  189.  */

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

  191.                                    int len, int pq)

  192. {

  193.     int i;

  194.     int filt3;

  195. 

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

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

  198.         if (filt3) {

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

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

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

  202.         }

  203.         src += step * 4;

  204.     }

  205. }

  206. 

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

  208. {

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

  210. }

  211. 

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

  213. {

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

  215. }

  216. 

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

  218. {

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

  220. }

  221. 

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

  223. {

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

  225. }

  226. 

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

  228. {

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

  230. }

  231. 

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

  233. {

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

  235. }

  236. 

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

  238. static void vc1_inv_trans_8x8_dc_c(uint8_t *dest, ptrdiff_t stride, int16_t *block)

  239. {

  240.     int i;

  241.     int dc = block[0];

  242. 

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

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

  245. 

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

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

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

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

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

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

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

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

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

  255.         dest += stride;

  256.     }

  257. }

  258. 

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

  260. {

  261.     int i;

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

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

  264. 

  265.     src = block;

  266.     dst = temp;

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

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

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

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

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

  272. 

  273.         t5 = t1 + t3;

  274.         t6 = t2 + t4;

  275.         t7 = t2 - t4;

  276.         t8 = t1 - t3;

  277. 

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

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

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

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

  282. 

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

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

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

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

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

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

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

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

  291. 

  292.         src += 1;

  293.         dst += 8;

  294.     }

  295. 

  296.     src = temp;

  297.     dst = block;

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

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

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

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

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

  303. 

  304.         t5 = t1 + t3;

  305.         t6 = t2 + t4;

  306.         t7 = t2 - t4;

  307.         t8 = t1 - t3;

  308. 

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

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

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

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

  313. 

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

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

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

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

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

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

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

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

  322. 

  323.         src++;

  324.         dst++;

  325.     }

  326. }

  327. 

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

  329. static void vc1_inv_trans_8x4_dc_c(uint8_t *dest, ptrdiff_t stride, int16_t *block)

  330. {

  331.     int i;

  332.     int dc = block[0];

  333. 

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

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

  336. 

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

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

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

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

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

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

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

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

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

  346.         dest += stride;

  347.     }

  348. }

  349. 

  350. static void vc1_inv_trans_8x4_c(uint8_t *dest, ptrdiff_t stride, int16_t *block)

  351. {

  352.     int i;

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

  354.     int16_t *src, *dst;

  355. 

  356.     src = block;

  357.     dst = block;

  358. 

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

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

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

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

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

  364. 

  365.         t5 = t1 + t3;

  366.         t6 = t2 + t4;

  367.         t7 = t2 - t4;

  368.         t8 = t1 - t3;

  369. 

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

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

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

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

  374. 

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

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

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

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

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

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

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

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

  383. 

  384.         src += 8;

  385.         dst += 8;

  386.     }

  387. 

  388.     src = block;

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

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

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

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

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

  394. 

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

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

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

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

  399. 

  400.         src++;

  401.         dest++;

  402.     }

  403. }

  404. 

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

  406. static void vc1_inv_trans_4x8_dc_c(uint8_t *dest, ptrdiff_t stride, int16_t *block)

  407. {

  408.     int i;

  409.     int dc = block[0];

  410. 

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

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

  413. 

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

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

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

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

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

  419.         dest += stride;

  420.     }

  421. }

  422. 

  423. static void vc1_inv_trans_4x8_c(uint8_t *dest, ptrdiff_t stride, int16_t *block)

  424. {

  425.     int i;

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

  427.     int16_t *src, *dst;

  428. 

  429.     src = block;

  430.     dst = block;

  431. 

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

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

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

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

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

  437. 

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

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

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

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

  442. 

  443.         src += 8;

  444.         dst += 8;

  445.     }

  446. 

  447.     src = block;

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

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

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

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

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

  453. 

  454.         t5 = t1 + t3;

  455.         t6 = t2 + t4;

  456.         t7 = t2 - t4;

  457.         t8 = t1 - t3;

  458. 

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

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

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

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

  463. 

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

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

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

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

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

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

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

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

  472. 

  473.         src++;

  474.         dest++;

  475.     }

  476. }

  477. 

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

  479. static void vc1_inv_trans_4x4_dc_c(uint8_t *dest, ptrdiff_t stride, int16_t *block)

  480. {

  481.     int i;

  482.     int dc = block[0];

  483. 

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

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

  486. 

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

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

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

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

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

  492.         dest += stride;

  493.     }

  494. }

  495. 

  496. static void vc1_inv_trans_4x4_c(uint8_t *dest, ptrdiff_t stride, int16_t *block)

  497. {

  498.     int i;

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

  500.     int16_t *src, *dst;

  501. 

  502.     src = block;

  503.     dst = block;

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

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

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

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

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

  509. 

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

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

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

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

  514. 

  515.         src += 8;

  516.         dst += 8;

  517.     }

  518. 

  519.     src = block;

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

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

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

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

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

  525. 

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

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

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

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

  530. 

  531.         src++;

  532.         dest++;

  533.     }

  534. }

  535. 

  536. /* motion compensation functions */

  537. 

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

  539. #define VC1_MSPEL_FILTER_16B(DIR, TYPE)                                       \

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

  541.                                                                 int stride,   \

  542.                                                                 int mode)     \

  543. {                                                                             \

  544.     switch(mode) {                                                            \

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

  546.         return 0;                                                             \

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

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

  549.                18 * src[stride]  -  3 * src[stride * 2];                      \

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

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

  552.                 9 * src[stride]  -  1 * src[stride * 2];                      \

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

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

  555.                53 * src[stride]  -  4 * src[stride * 2];                      \

  556.     }                                                                         \

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

  558. }

  559. 

  560. VC1_MSPEL_FILTER_16B(ver, uint8_t)

  561. VC1_MSPEL_FILTER_16B(hor, int16_t)

  562. 

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

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

  565.                                              int mode, int r)

  566. {

  567.     switch (mode) {

  568.     case 0: // no shift

  569.         return src[0];

  570.     case 1: // 1/4 shift

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

  572.                 18 * src[stride]  -  3 * src[stride * 2] + 32 - r) >> 6;

  573.     case 2: // 1/2 shift

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

  575.                  9 * src[stride]  -  1 * src[stride * 2] + 8 - r) >> 4;

  576.     case 3: // 3/4 shift

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

  578.                 53 * src[stride]  -  4 * src[stride * 2] + 32 - r) >> 6;

  579.     }

  580.     return 0; // should not occur

  581. }

  582. 

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

  584. #define VC1_MSPEL_MC(OP, OPNAME)                                              \

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

  586.                                                     const uint8_t *src,       \

  587.                                                     int stride,               \

  588.                                                     int hmode,                \

  589.                                                     int vmode,                \

  590.                                                     int rnd)                  \

  591. {                                                                             \

  592.     int i, j;                                                                 \

  593.                                                                               \

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

  595.         int r;                                                                \

  596.                                                                               \

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

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

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

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

  601.                                                                               \

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

  603.                                                                               \

  604.             src -= 1;                                                         \

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

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

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

  608.                 src  += stride;                                               \

  609.                 tptr += 11;                                                   \

  610.             }                                                                 \

  611.                                                                               \

  612.             r    = 64 - rnd;                                                  \

  613.             tptr = tmp + 1;                                                   \

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

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

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

  617.                 dst  += stride;                                               \

  618.                 tptr += 11;                                                   \

  619.             }                                                                 \

  620.                                                                               \

  621.             return;                                                           \

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

  623.             r = 1 - rnd;                                                      \

  624.                                                                               \

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

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

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

  628.                 src += stride;                                                \

  629.                 dst += stride;                                                \

  630.             }                                                                 \

  631.             return;                                                           \

  632.         }                                                                     \

  633.     }                                                                         \

  634.                                                                               \

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

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

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

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

  639.         dst += stride;                                                        \

  640.         src += stride;                                                        \

  641.     }                                                                         \

  642. }

  643. 

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

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

  646. 

  647. VC1_MSPEL_MC(op_put, put_)

  648. VC1_MSPEL_MC(op_avg, avg_)

  649. 

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

  651. 

  652. #define PUT_VC1_MSPEL(a, b)                                                   \

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

  654.                                              const uint8_t *src,              \

  655.                                              ptrdiff_t stride, int rnd)       \

  656. {                                                                             \

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

  658. }                                                                             \

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

  660.                                              const uint8_t *src,              \

  661.                                              ptrdiff_t stride, int rnd)       \

  662. {                                                                             \

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

  664. }

  665. 

  666. PUT_VC1_MSPEL(1, 0)

  667. PUT_VC1_MSPEL(2, 0)

  668. PUT_VC1_MSPEL(3, 0)

  669. 

  670. PUT_VC1_MSPEL(0, 1)

  671. PUT_VC1_MSPEL(1, 1)

  672. PUT_VC1_MSPEL(2, 1)

  673. PUT_VC1_MSPEL(3, 1)

  674. 

  675. PUT_VC1_MSPEL(0, 2)

  676. PUT_VC1_MSPEL(1, 2)

  677. PUT_VC1_MSPEL(2, 2)

  678. PUT_VC1_MSPEL(3, 2)

  679. 

  680. PUT_VC1_MSPEL(0, 3)

  681. PUT_VC1_MSPEL(1, 3)

  682. PUT_VC1_MSPEL(2, 3)

  683. PUT_VC1_MSPEL(3, 3)

  684. 

  685. 

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

  687.                                  ptrdiff_t stride, int rnd)

  688. {

  689.     ff_put_pixels8x8_c(dst, src, stride);

  690. }

  691. 

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

  693.                                  ptrdiff_t stride, int rnd)

  694. {

  695.     ff_avg_pixels8x8_c(dst, src, stride);

  696. }

  697. 

  698. #define chroma_mc(a) \

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

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

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

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

  703.                                         ptrdiff_t stride, int h, int x, int y)

  704. {

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

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

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

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

  709.     int i;

  710. 

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

  712. 

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

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

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

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

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

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

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

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

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

  722.         dst += stride;

  723.         src += stride;

  724.     }

  725. }

  726. 

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

  728.                                         ptrdiff_t stride, int h, int x, int y)

  729. {

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

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

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

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

  734.     int i;

  735. 

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

  737. 

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

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

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

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

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

  743.         dst += stride;

  744.         src += stride;

  745.     }

  746. }

  747. 

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

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

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

  751.                                         ptrdiff_t stride, int h, int x, int y)

  752. {

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

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

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

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

  757.     int i;

  758. 

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

  760. 

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

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

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

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

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

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

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

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

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

  770.         dst += stride;

  771.         src += stride;

  772.     }

  773. }

  774. 

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

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

  777.                                         ptrdiff_t stride, int h, int x, int y)

  778. {

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

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

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

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

  783.     int i;

  784. 

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

  786. 

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

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

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

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

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

  792.         dst += stride;

  793.         src += stride;

  794.     }

  795. }

  796. 

  797. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  798. 

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

  800.                        int advance, int count)

  801. {

  802.     while (count--) {

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

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

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

  806.         offset += advance;

  807.     }

  808. }

  809. 

  810. static av_always_inline void sprite_v_template(uint8_t *dst,

  811.                                                const uint8_t *src1a,

  812.                                                const uint8_t *src1b,

  813.                                                int offset1,

  814.                                                int two_sprites,

  815.                                                const uint8_t *src2a,

  816.                                                const uint8_t *src2b,

  817.                                                int offset2,

  818.                                                int alpha, int scaled,

  819.                                                int width)

  820. {

  821.     int a1, b1, a2, b2;

  822.     while (width--) {

  823.         a1 = *src1a++;

  824.         if (scaled) {

  825.             b1 = *src1b++;

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

  827.         }

  828.         if (two_sprites) {

  829.             a2 = *src2a++;

  830.             if (scaled > 1) {

  831.                 b2 = *src2b++;

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

  833.             }

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

  835.         }

  836.         *dst++ = a1;

  837.     }

  838. }

  839. 

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

  841.                               const uint8_t *src1b,

  842.                               int offset, int width)

  843. {

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

  845. }

  846. 

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

  848.                                       const uint8_t *src2a,

  849.                                       int alpha, int width)

  850. {

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

  852. }

  853. 

  854. static void sprite_v_double_onescale_c(uint8_t *dst,

  855.                                        const uint8_t *src1a,

  856.                                        const uint8_t *src1b,

  857.                                        int offset1,

  858.                                        const uint8_t *src2a,

  859.                                        int alpha, int width)

  860. {

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

  862.                       width);

  863. }

  864. 

  865. static void sprite_v_double_twoscale_c(uint8_t *dst,

  866.                                        const uint8_t *src1a,

  867.                                        const uint8_t *src1b,

  868.                                        int offset1,

  869.                                        const uint8_t *src2a,

  870.                                        const uint8_t *src2b,

  871.                                        int offset2,

  872.                                        int alpha,

  873.                                        int width)

  874. {

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

  876.                       alpha, 2, width);

  877. }

  878. 

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

  880. 

  881. av_cold void ff_vc1dsp_init(VC1DSPContext *dsp)

  882. {

  883.     dsp->vc1_inv_trans_8x8    = vc1_inv_trans_8x8_c;

  884.     dsp->vc1_inv_trans_4x8    = vc1_inv_trans_4x8_c;

  885.     dsp->vc1_inv_trans_8x4    = vc1_inv_trans_8x4_c;

  886.     dsp->vc1_inv_trans_4x4    = vc1_inv_trans_4x4_c;

  887.     dsp->vc1_inv_trans_8x8_dc = vc1_inv_trans_8x8_dc_c;

  888.     dsp->vc1_inv_trans_4x8_dc = vc1_inv_trans_4x8_dc_c;

  889.     dsp->vc1_inv_trans_8x4_dc = vc1_inv_trans_8x4_dc_c;

  890.     dsp->vc1_inv_trans_4x4_dc = vc1_inv_trans_4x4_dc_c;

  891. 

  892.     dsp->vc1_h_overlap        = vc1_h_overlap_c;

  893.     dsp->vc1_v_overlap        = vc1_v_overlap_c;

  894.     dsp->vc1_h_s_overlap      = vc1_h_s_overlap_c;

  895.     dsp->vc1_v_s_overlap      = vc1_v_s_overlap_c;

  896. 

  897.     dsp->vc1_v_loop_filter4   = vc1_v_loop_filter4_c;

  898.     dsp->vc1_h_loop_filter4   = vc1_h_loop_filter4_c;

  899.     dsp->vc1_v_loop_filter8   = vc1_v_loop_filter8_c;

  900.     dsp->vc1_h_loop_filter8   = vc1_h_loop_filter8_c;

  901.     dsp->vc1_v_loop_filter16  = vc1_v_loop_filter16_c;

  902.     dsp->vc1_h_loop_filter16  = vc1_h_loop_filter16_c;

  903. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  920. 

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

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

  937. 

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

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

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

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

  942. 

  943. #if CONFIG_WMV3IMAGE_DECODER || CONFIG_VC1IMAGE_DECODER

  944.     dsp->sprite_h                 = sprite_h_c;

  945.     dsp->sprite_v_single          = sprite_v_single_c;

  946.     dsp->sprite_v_double_noscale  = sprite_v_double_noscale_c;

  947.     dsp->sprite_v_double_onescale = sprite_v_double_onescale_c;

  948.     dsp->sprite_v_double_twoscale = sprite_v_double_twoscale_c;

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

  950. 

  951.     dsp->startcode_find_candidate = ff_startcode_find_candidate_c;

  952. 

  953.     if (ARCH_AARCH64)

  954.         ff_vc1dsp_init_aarch64(dsp);

  955.     if (ARCH_ARM)

  956.         ff_vc1dsp_init_arm(dsp);

  957.     if (ARCH_PPC)

  958.         ff_vc1dsp_init_ppc(dsp);

  959.     if (ARCH_X86)

  960.         ff_vc1dsp_init_x86(dsp);

  961. }