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.
36760 Commits
32 Branches
325MB
Tree: 7e350379f8
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '7e350379f8'
${ noResults }
FFmpeg/libavcodec/svq3.c

1294 lines
46KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The Libav Project

  3.  *

  4.  * This file is part of Libav.

  5.  *

  6.  * Libav is free software; you can redistribute it and/or

  7.  * modify it under the terms of the GNU Lesser General Public

  8.  * License as published by the Free Software Foundation; either

  9.  * version 2.1 of the License, or (at your option) any later version.

  10.  *

  11.  * Libav is distributed in the hope that it will be useful,

  12.  * but WITHOUT ANY WARRANTY; without even the implied warranty of

  13.  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU

  14.  * Lesser General Public License for more details.

  15.  *

  16.  * You should have received a copy of the GNU Lesser General Public

  17.  * License along with Libav; if not, write to the Free Software

  18.  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

  23.  * SVQ3 data is transported within Apple Quicktime files. Quicktime files

  24.  * have stsd atoms to describe media trak properties. A stsd atom for a

  25.  * video trak contains 1 or more ImageDescription atoms. These atoms begin

  26.  * with the 4-byte length of the atom followed by the codec fourcc. Some

  27.  * decoders need information in this atom to operate correctly. Such

  28.  * is the case with SVQ3. In order to get the best use out of this decoder,

  29.  * the calling app must make the SVQ3 ImageDescription atom available

  30.  * via the AVCodecContext's extradata[_size] field:

  31.  *

  32.  * AVCodecContext.extradata = pointer to ImageDescription, first characters

  33.  * are expected to be 'S', 'V', 'Q', and '3', NOT the 4-byte atom length

  34.  * AVCodecContext.extradata_size = size of ImageDescription atom memory

  35.  * buffer (which will be the same as the ImageDescription atom size field

  36.  * from the QT file, minus 4 bytes since the length is missing)

  37.  *

  38.  * You will know you have these parameters passed correctly when the decoder

  39.  * correctly decodes this file:

  40.  *  http://samples.libav.org/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov

  41.  */

  42. #include "internal.h"

  43. #include "avcodec.h"

  44. #include "mpegvideo.h"

  45. #include "h264.h"

  46. 

  47. #include "h264data.h" // FIXME FIXME FIXME

  48. 

  49. #include "h264_mvpred.h"

  50. #include "golomb.h"

  51. #include "rectangle.h"

  52. #include "vdpau_internal.h"

  53. 

  54. #if CONFIG_ZLIB

  55. #include <zlib.h>

  56. #endif

  57. 

  58. #include "svq1.h"

  59. #include "svq3.h"

  60. 

  61. /**

  62.  * @file

  63.  * svq3 decoder.

  64.  */

  65. 

  66. typedef struct {

  67.     H264Context h;

  68.     Picture *cur_pic;

  69.     Picture *next_pic;

  70.     Picture *last_pic;

  71.     int halfpel_flag;

  72.     int thirdpel_flag;

  73.     int unknown_flag;

  74.     int next_slice_index;

  75.     uint32_t watermark_key;

  76.     int adaptive_quant;

  77.     int next_p_frame_damaged;

  78.     int h_edge_pos;

  79.     int v_edge_pos;

  80.     int last_frame_output;

  81. } SVQ3Context;

  82. 

  83. #define FULLPEL_MODE  1

  84. #define HALFPEL_MODE  2

  85. #define THIRDPEL_MODE 3

  86. #define PREDICT_MODE  4

  87. 

  88. /* dual scan (from some older h264 draft)

  89.  * o-->o-->o   o

  90.  *         |  /|

  91.  * o   o   o / o

  92.  * | / |   |/  |

  93.  * o   o   o   o

  94.  *   /

  95.  * o-->o-->o-->o

  96.  */

  97. static const uint8_t svq3_scan[16] = {

  98.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  99.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

  100.     0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4,

  101.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  102. };

  103. 

  104. static const uint8_t svq3_pred_0[25][2] = {

  105.     { 0, 0 },

  106.     { 1, 0 }, { 0, 1 },

  107.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  108.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  109.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  110.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  111.     { 2, 4 }, { 3, 3 }, { 4, 2 },

  112.     { 4, 3 }, { 3, 4 },

  113.     { 4, 4 }

  114. };

  115. 

  116. static const int8_t svq3_pred_1[6][6][5] = {

  117.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  118.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

  119.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  4,  3 }, { 0, 1,  2,  4,  3 },

  120.       { 0,  2,  1,  4,  3 }, { 2, 0,  1,  3,  4 }, { 0, 4,  2,  1,  3 } },

  121.     { { 2,  0, -1, -1, -1 }, { 2, 1,  0,  4,  3 }, { 1, 2,  4,  0,  3 },

  122.       { 2,  1,  0,  4,  3 }, { 2, 1,  4,  3,  0 }, { 1, 2,  4,  0,  3 } },

  123.     { { 2,  0, -1, -1, -1 }, { 2, 0,  1,  4,  3 }, { 1, 2,  0,  4,  3 },

  124.       { 2,  1,  0,  4,  3 }, { 2, 1,  3,  4,  0 }, { 2, 4,  1,  0,  3 } },

  125.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  3,  4 }, { 1, 2,  3,  0,  4 },

  126.       { 2,  0,  1,  3,  4 }, { 2, 1,  3,  0,  4 }, { 2, 0,  4,  3,  1 } },

  127.     { { 0,  2, -1, -1, -1 }, { 0, 2,  4,  1,  3 }, { 1, 4,  2,  0,  3 },

  128.       { 4,  2,  0,  1,  3 }, { 2, 0,  1,  4,  3 }, { 4, 2,  1,  0,  3 } },

  129. };

  130. 

  131. static const struct {

  132.     uint8_t run;

  133.     uint8_t level;

  134. } svq3_dct_tables[2][16] = {

  135.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  136.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  137.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  138.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  139. };

  140. 

  141. static const uint32_t svq3_dequant_coeff[32] = {

  142.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  143.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  144.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  145.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  146. };

  147. 

  148. void ff_svq3_luma_dc_dequant_idct_c(int16_t *output, int16_t *input, int qp)

  149. {

  150.     const int qmul = svq3_dequant_coeff[qp];

  151. #define stride 16

  152.     int i;

  153.     int temp[16];

  154.     static const uint8_t x_offset[4] = { 0, 1 * stride, 4 * stride, 5 * stride };

  155. 

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

  157.         const int z0 = 13 * (input[4 * i + 0] +      input[4 * i + 2]);

  158.         const int z1 = 13 * (input[4 * i + 0] -      input[4 * i + 2]);

  159.         const int z2 =  7 *  input[4 * i + 1] - 17 * input[4 * i + 3];

  160.         const int z3 = 17 *  input[4 * i + 1] +  7 * input[4 * i + 3];

  161. 

  162.         temp[4 * i + 0] = z0 + z3;

  163.         temp[4 * i + 1] = z1 + z2;

  164.         temp[4 * i + 2] = z1 - z2;

  165.         temp[4 * i + 3] = z0 - z3;

  166.     }

  167. 

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

  169.         const int offset = x_offset[i];

  170.         const int z0     = 13 * (temp[4 * 0 + i] +      temp[4 * 2 + i]);

  171.         const int z1     = 13 * (temp[4 * 0 + i] -      temp[4 * 2 + i]);

  172.         const int z2     =  7 *  temp[4 * 1 + i] - 17 * temp[4 * 3 + i];

  173.         const int z3     = 17 *  temp[4 * 1 + i] +  7 * temp[4 * 3 + i];

  174. 

  175.         output[stride *  0 + offset] = (z0 + z3) * qmul + 0x80000 >> 20;

  176.         output[stride *  2 + offset] = (z1 + z2) * qmul + 0x80000 >> 20;

  177.         output[stride *  8 + offset] = (z1 - z2) * qmul + 0x80000 >> 20;

  178.         output[stride * 10 + offset] = (z0 - z3) * qmul + 0x80000 >> 20;

  179.     }

  180. }

  181. #undef stride

  182. 

  183. void ff_svq3_add_idct_c(uint8_t *dst, int16_t *block,

  184.                         int stride, int qp, int dc)

  185. {

  186.     const int qmul = svq3_dequant_coeff[qp];

  187.     int i;

  188. 

  189.     if (dc) {

  190.         dc       = 13 * 13 * (dc == 1 ? 1538 * block[0]

  191.                                       : qmul * (block[0] >> 3) / 2);

  192.         block[0] = 0;

  193.     }

  194. 

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

  196.         const int z0 = 13 * (block[0 + 4 * i] +      block[2 + 4 * i]);

  197.         const int z1 = 13 * (block[0 + 4 * i] -      block[2 + 4 * i]);

  198.         const int z2 =  7 *  block[1 + 4 * i] - 17 * block[3 + 4 * i];

  199.         const int z3 = 17 *  block[1 + 4 * i] +  7 * block[3 + 4 * i];

  200. 

  201.         block[0 + 4 * i] = z0 + z3;

  202.         block[1 + 4 * i] = z1 + z2;

  203.         block[2 + 4 * i] = z1 - z2;

  204.         block[3 + 4 * i] = z0 - z3;

  205.     }

  206. 

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

  208.         const int z0 = 13 * (block[i + 4 * 0] +      block[i + 4 * 2]);

  209.         const int z1 = 13 * (block[i + 4 * 0] -      block[i + 4 * 2]);

  210.         const int z2 =  7 *  block[i + 4 * 1] - 17 * block[i + 4 * 3];

  211.         const int z3 = 17 *  block[i + 4 * 1] +  7 * block[i + 4 * 3];

  212.         const int rr = (dc + 0x80000);

  213. 

  214.         dst[i + stride * 0] = av_clip_uint8(dst[i + stride * 0] + ((z0 + z3) * qmul + rr >> 20));

  215.         dst[i + stride * 1] = av_clip_uint8(dst[i + stride * 1] + ((z1 + z2) * qmul + rr >> 20));

  216.         dst[i + stride * 2] = av_clip_uint8(dst[i + stride * 2] + ((z1 - z2) * qmul + rr >> 20));

  217.         dst[i + stride * 3] = av_clip_uint8(dst[i + stride * 3] + ((z0 - z3) * qmul + rr >> 20));

  218.     }

  219. }

  220. 

  221. static inline int svq3_decode_block(GetBitContext *gb, int16_t *block,

  222.                                     int index, const int type)

  223. {

  224.     static const uint8_t *const scan_patterns[4] =

  225.     { luma_dc_zigzag_scan, zigzag_scan, svq3_scan, chroma_dc_scan };

  226. 

  227.     int run, level, limit;

  228.     unsigned vlc;

  229.     const int intra           = 3 * type >> 2;

  230.     const uint8_t *const scan = scan_patterns[type];

  231. 

  232.     for (limit = (16 >> intra); index < 16; index = limit, limit += 8) {

  233.         for (; (vlc = svq3_get_ue_golomb(gb)) != 0; index++) {

  234.             int sign = (vlc & 1) ? 0 : -1;

  235.             vlc      = vlc + 1 >> 1;

  236. 

  237.             if (type == 3) {

  238.                 if (vlc < 3) {

  239.                     run   = 0;

  240.                     level = vlc;

  241.                 } else if (vlc < 4) {

  242.                     run   = 1;

  243.                     level = 1;

  244.                 } else {

  245.                     run   = vlc & 0x3;

  246.                     level = (vlc + 9 >> 2) - run;

  247.                 }

  248.             } else {

  249.                 if (vlc < 16) {

  250.                     run   = svq3_dct_tables[intra][vlc].run;

  251.                     level = svq3_dct_tables[intra][vlc].level;

  252.                 } else if (intra) {

  253.                     run   = vlc & 0x7;

  254.                     level = (vlc >> 3) +

  255.                             ((run == 0) ? 8 : ((run < 2) ? 2 : ((run < 5) ? 0 : -1)));

  256.                 } else {

  257.                     run   = vlc & 0xF;

  258.                     level = (vlc >> 4) +

  259.                             ((run == 0) ? 4 : ((run < 3) ? 2 : ((run < 10) ? 1 : 0)));

  260.                 }

  261.             }

  262. 

  263.             if ((index += run) >= limit)

  264.                 return -1;

  265. 

  266.             block[scan[index]] = (level ^ sign) - sign;

  267.         }

  268. 

  269.         if (type != 2) {

  270.             break;

  271.         }

  272.     }

  273. 

  274.     return 0;

  275. }

  276. 

  277. static inline void svq3_mc_dir_part(SVQ3Context *s,

  278.                                     int x, int y, int width, int height,

  279.                                     int mx, int my, int dxy,

  280.                                     int thirdpel, int dir, int avg)

  281. {

  282.     H264Context *h     = &s->h;

  283.     const Picture *pic = (dir == 0) ? s->last_pic : s->next_pic;

  284.     uint8_t *src, *dest;

  285.     int i, emu = 0;

  286.     int blocksize = 2 - (width >> 3); // 16->0, 8->1, 4->2

  287. 

  288.     mx += x;

  289.     my += y;

  290. 

  291.     if (mx < 0 || mx >= s->h_edge_pos - width  - 1 ||

  292.         my < 0 || my >= s->v_edge_pos - height - 1) {

  293.         emu = 1;

  294.         mx = av_clip(mx, -16, s->h_edge_pos - width  + 15);

  295.         my = av_clip(my, -16, s->v_edge_pos - height + 15);

  296.     }

  297. 

  298.     /* form component predictions */

  299.     dest = h->cur_pic.f.data[0] + x + y * h->linesize;

  300.     src  = pic->f.data[0] + mx + my * h->linesize;

  301. 

  302.     if (emu) {

  303.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->linesize,

  304.                                  width + 1, height + 1,

  305.                                  mx, my, s->h_edge_pos, s->v_edge_pos);

  306.         src = h->edge_emu_buffer;

  307.     }

  308.     if (thirdpel)

  309.         (avg ? h->dsp.avg_tpel_pixels_tab

  310.              : h->dsp.put_tpel_pixels_tab)[dxy](dest, src, h->linesize,

  311.                                                 width, height);

  312.     else

  313.         (avg ? h->dsp.avg_pixels_tab

  314.              : h->dsp.put_pixels_tab)[blocksize][dxy](dest, src, h->linesize,

  315.                                                       height);

  316. 

  317.     if (!(h->flags & CODEC_FLAG_GRAY)) {

  318.         mx     = mx + (mx < (int) x) >> 1;

  319.         my     = my + (my < (int) y) >> 1;

  320.         width  = width  >> 1;

  321.         height = height >> 1;

  322.         blocksize++;

  323. 

  324.         for (i = 1; i < 3; i++) {

  325.             dest = h->cur_pic.f.data[i] + (x >> 1) + (y >> 1) * h->uvlinesize;

  326.             src  = pic->f.data[i] + mx + my * h->uvlinesize;

  327. 

  328.             if (emu) {

  329.                 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->uvlinesize,

  330.                                          width + 1, height + 1,

  331.                                          mx, my, (s->h_edge_pos >> 1),

  332.                                          s->v_edge_pos >> 1);

  333.                 src = h->edge_emu_buffer;

  334.             }

  335.             if (thirdpel)

  336.                 (avg ? h->dsp.avg_tpel_pixels_tab

  337.                      : h->dsp.put_tpel_pixels_tab)[dxy](dest, src,

  338.                                                         h->uvlinesize,

  339.                                                         width, height);

  340.             else

  341.                 (avg ? h->dsp.avg_pixels_tab

  342.                      : h->dsp.put_pixels_tab)[blocksize][dxy](dest, src,

  343.                                                               h->uvlinesize,

  344.                                                               height);

  345.         }

  346.     }

  347. }

  348. 

  349. static inline int svq3_mc_dir(SVQ3Context *s, int size, int mode,

  350.                               int dir, int avg)

  351. {

  352.     int i, j, k, mx, my, dx, dy, x, y;

  353.     H264Context *h          = &s->h;

  354.     const int part_width    = ((size & 5) == 4) ? 4 : 16 >> (size & 1);

  355.     const int part_height   = 16 >> ((unsigned)(size + 1) / 3);

  356.     const int extra_width   = (mode == PREDICT_MODE) ? -16 * 6 : 0;

  357.     const int h_edge_pos    = 6 * (s->h_edge_pos - part_width)  - extra_width;

  358.     const int v_edge_pos    = 6 * (s->v_edge_pos - part_height) - extra_width;

  359. 

  360.     for (i = 0; i < 16; i += part_height)

  361.         for (j = 0; j < 16; j += part_width) {

  362.             const int b_xy = (4 * h->mb_x + (j >> 2)) +

  363.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  364.             int dxy;

  365.             x = 16 * h->mb_x + j;

  366.             y = 16 * h->mb_y + i;

  367.             k = (j >> 2 & 1) + (i >> 1 & 2) +

  368.                 (j >> 1 & 4) + (i      & 8);

  369. 

  370.             if (mode != PREDICT_MODE) {

  371.                 pred_motion(h, k, part_width >> 2, dir, 1, &mx, &my);

  372.             } else {

  373.                 mx = s->next_pic->f.motion_val[0][b_xy][0] << 1;

  374.                 my = s->next_pic->f.motion_val[0][b_xy][1] << 1;

  375. 

  376.                 if (dir == 0) {

  377.                     mx = mx * h->frame_num_offset /

  378.                          h->prev_frame_num_offset + 1 >> 1;

  379.                     my = my * h->frame_num_offset /

  380.                          h->prev_frame_num_offset + 1 >> 1;

  381.                 } else {

  382.                     mx = mx * (h->frame_num_offset - h->prev_frame_num_offset) /

  383.                          h->prev_frame_num_offset + 1 >> 1;

  384.                     my = my * (h->frame_num_offset - h->prev_frame_num_offset) /

  385.                          h->prev_frame_num_offset + 1 >> 1;

  386.                 }

  387.             }

  388. 

  389.             /* clip motion vector prediction to frame border */

  390.             mx = av_clip(mx, extra_width - 6 * x, h_edge_pos - 6 * x);

  391.             my = av_clip(my, extra_width - 6 * y, v_edge_pos - 6 * y);

  392. 

  393.             /* get (optional) motion vector differential */

  394.             if (mode == PREDICT_MODE) {

  395.                 dx = dy = 0;

  396.             } else {

  397.                 dy = svq3_get_se_golomb(&h->gb);

  398.                 dx = svq3_get_se_golomb(&h->gb);

  399. 

  400.                 if (dx == INVALID_VLC || dy == INVALID_VLC) {

  401.                     av_log(h->avctx, AV_LOG_ERROR, "invalid MV vlc\n");

  402.                     return -1;

  403.                 }

  404.             }

  405. 

  406.             /* compute motion vector */

  407.             if (mode == THIRDPEL_MODE) {

  408.                 int fx, fy;

  409.                 mx  = (mx + 1 >> 1) + dx;

  410.                 my  = (my + 1 >> 1) + dy;

  411.                 fx  = (unsigned)(mx + 0x3000) / 3 - 0x1000;

  412.                 fy  = (unsigned)(my + 0x3000) / 3 - 0x1000;

  413.                 dxy = (mx - 3 * fx) + 4 * (my - 3 * fy);

  414. 

  415.                 svq3_mc_dir_part(s, x, y, part_width, part_height,

  416.                                  fx, fy, dxy, 1, dir, avg);

  417.                 mx += mx;

  418.                 my += my;

  419.             } else if (mode == HALFPEL_MODE || mode == PREDICT_MODE) {

  420.                 mx  = (unsigned)(mx + 1 + 0x3000) / 3 + dx - 0x1000;

  421.                 my  = (unsigned)(my + 1 + 0x3000) / 3 + dy - 0x1000;

  422.                 dxy = (mx & 1) + 2 * (my & 1);

  423. 

  424.                 svq3_mc_dir_part(s, x, y, part_width, part_height,

  425.                                  mx >> 1, my >> 1, dxy, 0, dir, avg);

  426.                 mx *= 3;

  427.                 my *= 3;

  428.             } else {

  429.                 mx = (unsigned)(mx + 3 + 0x6000) / 6 + dx - 0x1000;

  430.                 my = (unsigned)(my + 3 + 0x6000) / 6 + dy - 0x1000;

  431. 

  432.                 svq3_mc_dir_part(s, x, y, part_width, part_height,

  433.                                  mx, my, 0, 0, dir, avg);

  434.                 mx *= 6;

  435.                 my *= 6;

  436.             }

  437. 

  438.             /* update mv_cache */

  439.             if (mode != PREDICT_MODE) {

  440.                 int32_t mv = pack16to32(mx, my);

  441. 

  442.                 if (part_height == 8 && i < 8) {

  443.                     AV_WN32A(h->mv_cache[dir][scan8[k] + 1 * 8], mv);

  444. 

  445.                     if (part_width == 8 && j < 8)

  446.                         AV_WN32A(h->mv_cache[dir][scan8[k] + 1 + 1 * 8], mv);

  447.                 }

  448.                 if (part_width == 8 && j < 8)

  449.                     AV_WN32A(h->mv_cache[dir][scan8[k] + 1], mv);

  450.                 if (part_width == 4 || part_height == 4)

  451.                     AV_WN32A(h->mv_cache[dir][scan8[k]], mv);

  452.             }

  453. 

  454.             /* write back motion vectors */

  455.             fill_rectangle(h->cur_pic.f.motion_val[dir][b_xy],

  456.                            part_width >> 2, part_height >> 2, h->b_stride,

  457.                            pack16to32(mx, my), 4);

  458.         }

  459. 

  460.     return 0;

  461. }

  462. 

  463. static int svq3_decode_mb(SVQ3Context *s, unsigned int mb_type)

  464. {

  465.     H264Context *h = &s->h;

  466.     int i, j, k, m, dir, mode;

  467.     int cbp = 0;

  468.     uint32_t vlc;

  469.     int8_t *top, *left;

  470.     const int mb_xy         = h->mb_xy;

  471.     const int b_xy          = 4 * h->mb_x + 4 * h->mb_y * h->b_stride;

  472. 

  473.     h->top_samples_available      = (h->mb_y == 0) ? 0x33FF : 0xFFFF;

  474.     h->left_samples_available     = (h->mb_x == 0) ? 0x5F5F : 0xFFFF;

  475.     h->topright_samples_available = 0xFFFF;

  476. 

  477.     if (mb_type == 0) {           /* SKIP */

  478.         if (h->pict_type == AV_PICTURE_TYPE_P ||

  479.             s->next_pic->f.mb_type[mb_xy] == -1) {

  480.             svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16,

  481.                              0, 0, 0, 0, 0, 0);

  482. 

  483.             if (h->pict_type == AV_PICTURE_TYPE_B)

  484.                 svq3_mc_dir_part(s, 16 * h->mb_x, 16 * h->mb_y, 16, 16,

  485.                                  0, 0, 0, 0, 1, 1);

  486. 

  487.             mb_type = MB_TYPE_SKIP;

  488.         } else {

  489.             mb_type = FFMIN(s->next_pic->f.mb_type[mb_xy], 6);

  490.             if (svq3_mc_dir(s, mb_type, PREDICT_MODE, 0, 0) < 0)

  491.                 return -1;

  492.             if (svq3_mc_dir(s, mb_type, PREDICT_MODE, 1, 1) < 0)

  493.                 return -1;

  494. 

  495.             mb_type = MB_TYPE_16x16;

  496.         }

  497.     } else if (mb_type < 8) {     /* INTER */

  498.         if (s->thirdpel_flag && s->halfpel_flag == !get_bits1(&h->gb))

  499.             mode = THIRDPEL_MODE;

  500.         else if (s->halfpel_flag &&

  501.                  s->thirdpel_flag == !get_bits1(&h->gb))

  502.             mode = HALFPEL_MODE;

  503.         else

  504.             mode = FULLPEL_MODE;

  505. 

  506.         /* fill caches */

  507.         /* note ref_cache should contain here:

  508.          *  ????????

  509.          *  ???11111

  510.          *  N??11111

  511.          *  N??11111

  512.          *  N??11111

  513.          */

  514. 

  515.         for (m = 0; m < 2; m++) {

  516.             if (h->mb_x > 0 && h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6] != -1) {

  517.                 for (i = 0; i < 4; i++)

  518.                     AV_COPY32(h->mv_cache[m][scan8[0] - 1 + i * 8],

  519.                               h->cur_pic.f.motion_val[m][b_xy - 1 + i * h->b_stride]);

  520.             } else {

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

  522.                     AV_ZERO32(h->mv_cache[m][scan8[0] - 1 + i * 8]);

  523.             }

  524.             if (h->mb_y > 0) {

  525.                 memcpy(h->mv_cache[m][scan8[0] - 1 * 8],

  526.                        h->cur_pic.f.motion_val[m][b_xy - h->b_stride],

  527.                        4 * 2 * sizeof(int16_t));

  528.                 memset(&h->ref_cache[m][scan8[0] - 1 * 8],

  529.                        (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1, 4);

  530. 

  531.                 if (h->mb_x < h->mb_width - 1) {

  532.                     AV_COPY32(h->mv_cache[m][scan8[0] + 4 - 1 * 8],

  533.                               h->cur_pic.f.motion_val[m][b_xy - h->b_stride + 4]);

  534.                     h->ref_cache[m][scan8[0] + 4 - 1 * 8] =

  535.                         (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride + 1] + 6] == -1 ||

  536.                          h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride]] == -1) ? PART_NOT_AVAILABLE : 1;

  537.                 } else

  538.                     h->ref_cache[m][scan8[0] + 4 - 1 * 8] = PART_NOT_AVAILABLE;

  539.                 if (h->mb_x > 0) {

  540.                     AV_COPY32(h->mv_cache[m][scan8[0] - 1 - 1 * 8],

  541.                               h->cur_pic.f.motion_val[m][b_xy - h->b_stride - 1]);

  542.                     h->ref_cache[m][scan8[0] - 1 - 1 * 8] =

  543.                         (h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride - 1] + 3] == -1) ? PART_NOT_AVAILABLE : 1;

  544.                 } else

  545.                     h->ref_cache[m][scan8[0] - 1 - 1 * 8] = PART_NOT_AVAILABLE;

  546.             } else

  547.                 memset(&h->ref_cache[m][scan8[0] - 1 * 8 - 1],

  548.                        PART_NOT_AVAILABLE, 8);

  549. 

  550.             if (h->pict_type != AV_PICTURE_TYPE_B)

  551.                 break;

  552.         }

  553. 

  554.         /* decode motion vector(s) and form prediction(s) */

  555.         if (h->pict_type == AV_PICTURE_TYPE_P) {

  556.             if (svq3_mc_dir(s, mb_type - 1, mode, 0, 0) < 0)

  557.                 return -1;

  558.         } else {        /* AV_PICTURE_TYPE_B */

  559.             if (mb_type != 2) {

  560.                 if (svq3_mc_dir(s, 0, mode, 0, 0) < 0)

  561.                     return -1;

  562.             } else {

  563.                 for (i = 0; i < 4; i++)

  564.                     memset(h->cur_pic.f.motion_val[0][b_xy + i * h->b_stride],

  565.                            0, 4 * 2 * sizeof(int16_t));

  566.             }

  567.             if (mb_type != 1) {

  568.                 if (svq3_mc_dir(s, 0, mode, 1, mb_type == 3) < 0)

  569.                     return -1;

  570.             } else {

  571.                 for (i = 0; i < 4; i++)

  572.                     memset(h->cur_pic.f.motion_val[1][b_xy + i * h->b_stride],

  573.                            0, 4 * 2 * sizeof(int16_t));

  574.             }

  575.         }

  576. 

  577.         mb_type = MB_TYPE_16x16;

  578.     } else if (mb_type == 8 || mb_type == 33) {   /* INTRA4x4 */

  579.         memset(h->intra4x4_pred_mode_cache, -1, 8 * 5 * sizeof(int8_t));

  580. 

  581.         if (mb_type == 8) {

  582.             if (h->mb_x > 0) {

  583.                 for (i = 0; i < 4; i++)

  584.                     h->intra4x4_pred_mode_cache[scan8[0] - 1 + i * 8] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - 1] + 6 - i];

  585.                 if (h->intra4x4_pred_mode_cache[scan8[0] - 1] == -1)

  586.                     h->left_samples_available = 0x5F5F;

  587.             }

  588.             if (h->mb_y > 0) {

  589.                 h->intra4x4_pred_mode_cache[4 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 0];

  590.                 h->intra4x4_pred_mode_cache[5 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 1];

  591.                 h->intra4x4_pred_mode_cache[6 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 2];

  592.                 h->intra4x4_pred_mode_cache[7 + 8 * 0] = h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride] + 3];

  593. 

  594.                 if (h->intra4x4_pred_mode_cache[4 + 8 * 0] == -1)

  595.                     h->top_samples_available = 0x33FF;

  596.             }

  597. 

  598.             /* decode prediction codes for luma blocks */

  599.             for (i = 0; i < 16; i += 2) {

  600.                 vlc = svq3_get_ue_golomb(&h->gb);

  601. 

  602.                 if (vlc >= 25) {

  603.                     av_log(h->avctx, AV_LOG_ERROR, "luma prediction:%d\n", vlc);

  604.                     return -1;

  605.                 }

  606. 

  607.                 left = &h->intra4x4_pred_mode_cache[scan8[i] - 1];

  608.                 top  = &h->intra4x4_pred_mode_cache[scan8[i] - 8];

  609. 

  610.                 left[1] = svq3_pred_1[top[0] + 1][left[0] + 1][svq3_pred_0[vlc][0]];

  611.                 left[2] = svq3_pred_1[top[1] + 1][left[1] + 1][svq3_pred_0[vlc][1]];

  612. 

  613.                 if (left[1] == -1 || left[2] == -1) {

  614.                     av_log(h->avctx, AV_LOG_ERROR, "weird prediction\n");

  615.                     return -1;

  616.                 }

  617.             }

  618.         } else {    /* mb_type == 33, DC_128_PRED block type */

  619.             for (i = 0; i < 4; i++)

  620.                 memset(&h->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_PRED, 4);

  621.         }

  622. 

  623.         write_back_intra_pred_mode(h);

  624. 

  625.         if (mb_type == 8) {

  626.             ff_h264_check_intra4x4_pred_mode(h);

  627. 

  628.             h->top_samples_available  = (h->mb_y == 0) ? 0x33FF : 0xFFFF;

  629.             h->left_samples_available = (h->mb_x == 0) ? 0x5F5F : 0xFFFF;

  630.         } else {

  631.             for (i = 0; i < 4; i++)

  632.                 memset(&h->intra4x4_pred_mode_cache[scan8[0] + 8 * i], DC_128_PRED, 4);

  633. 

  634.             h->top_samples_available  = 0x33FF;

  635.             h->left_samples_available = 0x5F5F;

  636.         }

  637. 

  638.         mb_type = MB_TYPE_INTRA4x4;

  639.     } else {                      /* INTRA16x16 */

  640.         dir = i_mb_type_info[mb_type - 8].pred_mode;

  641.         dir = (dir >> 1) ^ 3 * (dir & 1) ^ 1;

  642. 

  643.         if ((h->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, dir, 0)) == -1) {

  644.             av_log(h->avctx, AV_LOG_ERROR, "check_intra_pred_mode = -1\n");

  645.             return -1;

  646.         }

  647. 

  648.         cbp     = i_mb_type_info[mb_type - 8].cbp;

  649.         mb_type = MB_TYPE_INTRA16x16;

  650.     }

  651. 

  652.     if (!IS_INTER(mb_type) && h->pict_type != AV_PICTURE_TYPE_I) {

  653.         for (i = 0; i < 4; i++)

  654.             memset(h->cur_pic.f.motion_val[0][b_xy + i * h->b_stride],

  655.                    0, 4 * 2 * sizeof(int16_t));

  656.         if (h->pict_type == AV_PICTURE_TYPE_B) {

  657.             for (i = 0; i < 4; i++)

  658.                 memset(h->cur_pic.f.motion_val[1][b_xy + i * h->b_stride],

  659.                        0, 4 * 2 * sizeof(int16_t));

  660.         }

  661.     }

  662.     if (!IS_INTRA4x4(mb_type)) {

  663.         memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy], DC_PRED, 8);

  664.     }

  665.     if (!IS_SKIP(mb_type) || h->pict_type == AV_PICTURE_TYPE_B) {

  666.         memset(h->non_zero_count_cache + 8, 0, 14 * 8 * sizeof(uint8_t));

  667.         h->dsp.clear_blocks(h->mb +   0);

  668.         h->dsp.clear_blocks(h->mb + 384);

  669.     }

  670. 

  671.     if (!IS_INTRA16x16(mb_type) &&

  672.         (!IS_SKIP(mb_type) || h->pict_type == AV_PICTURE_TYPE_B)) {

  673.         if ((vlc = svq3_get_ue_golomb(&h->gb)) >= 48) {

  674.             av_log(h->avctx, AV_LOG_ERROR, "cbp_vlc=%d\n", vlc);

  675.             return -1;

  676.         }

  677. 

  678.         cbp = IS_INTRA(mb_type) ? golomb_to_intra4x4_cbp[vlc]

  679.                                 : golomb_to_inter_cbp[vlc];

  680.     }

  681.     if (IS_INTRA16x16(mb_type) ||

  682.         (h->pict_type != AV_PICTURE_TYPE_I && s->adaptive_quant && cbp)) {

  683.         h->qscale += svq3_get_se_golomb(&h->gb);

  684. 

  685.         if (h->qscale > 31u) {

  686.             av_log(h->avctx, AV_LOG_ERROR, "qscale:%d\n", h->qscale);

  687.             return -1;

  688.         }

  689.     }

  690.     if (IS_INTRA16x16(mb_type)) {

  691.         AV_ZERO128(h->mb_luma_dc[0] + 0);

  692.         AV_ZERO128(h->mb_luma_dc[0] + 8);

  693.         if (svq3_decode_block(&h->gb, h->mb_luma_dc[0], 0, 1)) {

  694.             av_log(h->avctx, AV_LOG_ERROR,

  695.                    "error while decoding intra luma dc\n");

  696.             return -1;

  697.         }

  698.     }

  699. 

  700.     if (cbp) {

  701.         const int index = IS_INTRA16x16(mb_type) ? 1 : 0;

  702.         const int type  = ((h->qscale < 24 && IS_INTRA4x4(mb_type)) ? 2 : 1);

  703. 

  704.         for (i = 0; i < 4; i++)

  705.             if ((cbp & (1 << i))) {

  706.                 for (j = 0; j < 4; j++) {

  707.                     k = index ? (1 * (j & 1) + 2 * (i & 1) +

  708.                                  2 * (j & 2) + 4 * (i & 2))

  709.                               : (4 * i + j);

  710.                     h->non_zero_count_cache[scan8[k]] = 1;

  711. 

  712.                     if (svq3_decode_block(&h->gb, &h->mb[16 * k], index, type)) {

  713.                         av_log(h->avctx, AV_LOG_ERROR,

  714.                                "error while decoding block\n");

  715.                         return -1;

  716.                     }

  717.                 }

  718.             }

  719. 

  720.         if ((cbp & 0x30)) {

  721.             for (i = 1; i < 3; ++i)

  722.                 if (svq3_decode_block(&h->gb, &h->mb[16 * 16 * i], 0, 3)) {

  723.                     av_log(h->avctx, AV_LOG_ERROR,

  724.                            "error while decoding chroma dc block\n");

  725.                     return -1;

  726.                 }

  727. 

  728.             if ((cbp & 0x20)) {

  729.                 for (i = 1; i < 3; i++) {

  730.                     for (j = 0; j < 4; j++) {

  731.                         k                                 = 16 * i + j;

  732.                         h->non_zero_count_cache[scan8[k]] = 1;

  733. 

  734.                         if (svq3_decode_block(&h->gb, &h->mb[16 * k], 1, 1)) {

  735.                             av_log(h->avctx, AV_LOG_ERROR,

  736.                                    "error while decoding chroma ac block\n");

  737.                             return -1;

  738.                         }

  739.                     }

  740.                 }

  741.             }

  742.         }

  743.     }

  744. 

  745.     h->cbp                              = cbp;

  746.     h->cur_pic.f.mb_type[mb_xy] = mb_type;

  747. 

  748.     if (IS_INTRA(mb_type))

  749.         h->chroma_pred_mode = ff_h264_check_intra_pred_mode(h, DC_PRED8x8, 1);

  750. 

  751.     return 0;

  752. }

  753. 

  754. static int svq3_decode_slice_header(AVCodecContext *avctx)

  755. {

  756.     SVQ3Context *s = avctx->priv_data;

  757.     H264Context *h    = &s->h;

  758.     const int mb_xy   = h->mb_xy;

  759.     int i, header;

  760.     unsigned slice_id;

  761. 

  762.     header = get_bits(&h->gb, 8);

  763. 

  764.     if (((header & 0x9F) != 1 && (header & 0x9F) != 2) || (header & 0x60) == 0) {

  765.         /* TODO: what? */

  766.         av_log(avctx, AV_LOG_ERROR, "unsupported slice header (%02X)\n", header);

  767.         return -1;

  768.     } else {

  769.         int length = header >> 5 & 3;

  770. 

  771.         s->next_slice_index = get_bits_count(&h->gb) +

  772.                               8 * show_bits(&h->gb, 8 * length) +

  773.                               8 * length;

  774. 

  775.         if (s->next_slice_index > h->gb.size_in_bits) {

  776.             av_log(avctx, AV_LOG_ERROR, "slice after bitstream end\n");

  777.             return -1;

  778.         }

  779. 

  780.         h->gb.size_in_bits = s->next_slice_index - 8 * (length - 1);

  781.         skip_bits(&h->gb, 8);

  782. 

  783.         if (s->watermark_key) {

  784.             uint32_t header = AV_RL32(&h->gb.buffer[(get_bits_count(&h->gb) >> 3) + 1]);

  785.             AV_WL32(&h->gb.buffer[(get_bits_count(&h->gb) >> 3) + 1],

  786.                     header ^ s->watermark_key);

  787.         }

  788.         if (length > 0) {

  789.             memcpy((uint8_t *) &h->gb.buffer[get_bits_count(&h->gb) >> 3],

  790.                    &h->gb.buffer[h->gb.size_in_bits >> 3], length - 1);

  791.         }

  792.         skip_bits_long(&h->gb, 0);

  793.     }

  794. 

  795.     if ((slice_id = svq3_get_ue_golomb(&h->gb)) >= 3) {

  796.         av_log(h->avctx, AV_LOG_ERROR, "illegal slice type %d \n", slice_id);

  797.         return -1;

  798.     }

  799. 

  800.     h->slice_type = golomb_to_pict_type[slice_id];

  801. 

  802.     if ((header & 0x9F) == 2) {

  803.         i              = (h->mb_num < 64) ? 6 : (1 + av_log2(h->mb_num - 1));

  804.         h->mb_skip_run = get_bits(&h->gb, i) -

  805.                          (h->mb_y * h->mb_width + h->mb_x);

  806.     } else {

  807.         skip_bits1(&h->gb);

  808.         h->mb_skip_run = 0;

  809.     }

  810. 

  811.     h->slice_num      = get_bits(&h->gb, 8);

  812.     h->qscale         = get_bits(&h->gb, 5);

  813.     s->adaptive_quant = get_bits1(&h->gb);

  814. 

  815.     /* unknown fields */

  816.     skip_bits1(&h->gb);

  817. 

  818.     if (s->unknown_flag)

  819.         skip_bits1(&h->gb);

  820. 

  821.     skip_bits1(&h->gb);

  822.     skip_bits(&h->gb, 2);

  823. 

  824.     while (get_bits1(&h->gb))

  825.         skip_bits(&h->gb, 8);

  826. 

  827.     /* reset intra predictors and invalidate motion vector references */

  828.     if (h->mb_x > 0) {

  829.         memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy - 1] + 3,

  830.                -1, 4 * sizeof(int8_t));

  831.         memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_x],

  832.                -1, 8 * sizeof(int8_t) * h->mb_x);

  833.     }

  834.     if (h->mb_y > 0) {

  835.         memset(h->intra4x4_pred_mode + h->mb2br_xy[mb_xy - h->mb_stride],

  836.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  837. 

  838.         if (h->mb_x > 0)

  839.             h->intra4x4_pred_mode[h->mb2br_xy[mb_xy - h->mb_stride - 1] + 3] = -1;

  840.     }

  841. 

  842.     return 0;

  843. }

  844. 

  845. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  846. {

  847.     SVQ3Context *s = avctx->priv_data;

  848.     H264Context *h = &s->h;

  849.     int m;

  850.     unsigned char *extradata;

  851.     unsigned char *extradata_end;

  852.     unsigned int size;

  853.     int marker_found = 0;

  854. 

  855.     s->cur_pic  = av_mallocz(sizeof(*s->cur_pic));

  856.     s->last_pic = av_mallocz(sizeof(*s->last_pic));

  857.     s->next_pic = av_mallocz(sizeof(*s->next_pic));

  858.     if (!s->next_pic || !s->last_pic || !s->cur_pic) {

  859.         av_freep(&s->cur_pic);

  860.         av_freep(&s->last_pic);

  861.         av_freep(&s->next_pic);

  862.         return AVERROR(ENOMEM);

  863.     }

  864. 

  865.     if (ff_h264_decode_init(avctx) < 0)

  866.         return -1;

  867. 

  868.     h->flags           = avctx->flags;

  869.     h->is_complex      = 1;

  870.     h->picture_structure = PICT_FRAME;

  871.     avctx->pix_fmt     = avctx->codec->pix_fmts[0];

  872. 

  873.     h->chroma_qp[0] = h->chroma_qp[1] = 4;

  874.     h->chroma_x_shift = h->chroma_y_shift = 1;

  875. 

  876.     s->halfpel_flag  = 1;

  877.     s->thirdpel_flag = 1;

  878.     s->unknown_flag  = 0;

  879. 

  880.     /* prowl for the "SEQH" marker in the extradata */

  881.     extradata     = (unsigned char *)avctx->extradata;

  882.     extradata_end = avctx->extradata + avctx->extradata_size;

  883.     if (extradata) {

  884.         for (m = 0; m + 8 < avctx->extradata_size; m++) {

  885.             if (!memcmp(extradata, "SEQH", 4)) {

  886.                 marker_found = 1;

  887.                 break;

  888.             }

  889.             extradata++;

  890.         }

  891.     }

  892. 

  893.     /* if a match was found, parse the extra data */

  894.     if (marker_found) {

  895.         GetBitContext gb;

  896.         int frame_size_code;

  897. 

  898.         size = AV_RB32(&extradata[4]);

  899.         if (size > extradata_end - extradata - 8)

  900.             return AVERROR_INVALIDDATA;

  901.         init_get_bits(&gb, extradata + 8, size * 8);

  902. 

  903.         /* 'frame size code' and optional 'width, height' */

  904.         frame_size_code = get_bits(&gb, 3);

  905.         switch (frame_size_code) {

  906.         case 0:

  907.             avctx->width  = 160;

  908.             avctx->height = 120;

  909.             break;

  910.         case 1:

  911.             avctx->width  = 128;

  912.             avctx->height =  96;

  913.             break;

  914.         case 2:

  915.             avctx->width  = 176;

  916.             avctx->height = 144;

  917.             break;

  918.         case 3:

  919.             avctx->width  = 352;

  920.             avctx->height = 288;

  921.             break;

  922.         case 4:

  923.             avctx->width  = 704;

  924.             avctx->height = 576;

  925.             break;

  926.         case 5:

  927.             avctx->width  = 240;

  928.             avctx->height = 180;

  929.             break;

  930.         case 6:

  931.             avctx->width  = 320;

  932.             avctx->height = 240;

  933.             break;

  934.         case 7:

  935.             avctx->width  = get_bits(&gb, 12);

  936.             avctx->height = get_bits(&gb, 12);

  937.             break;

  938.         }

  939. 

  940.         s->halfpel_flag  = get_bits1(&gb);

  941.         s->thirdpel_flag = get_bits1(&gb);

  942. 

  943.         /* unknown fields */

  944.         skip_bits1(&gb);

  945.         skip_bits1(&gb);

  946.         skip_bits1(&gb);

  947.         skip_bits1(&gb);

  948. 

  949.         h->low_delay = get_bits1(&gb);

  950. 

  951.         /* unknown field */

  952.         skip_bits1(&gb);

  953. 

  954.         while (get_bits1(&gb))

  955.             skip_bits(&gb, 8);

  956. 

  957.         s->unknown_flag  = get_bits1(&gb);

  958.         avctx->has_b_frames = !h->low_delay;

  959.         if (s->unknown_flag) {

  960. #if CONFIG_ZLIB

  961.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  962.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  963.             int u1                    = svq3_get_ue_golomb(&gb);

  964.             int u2                    = get_bits(&gb, 8);

  965.             int u3                    = get_bits(&gb, 2);

  966.             int u4                    = svq3_get_ue_golomb(&gb);

  967.             unsigned long buf_len     = watermark_width *

  968.                                         watermark_height * 4;

  969.             int offset                = get_bits_count(&gb) + 7 >> 3;

  970.             uint8_t *buf;

  971. 

  972.             if ((uint64_t)watermark_width * 4 > UINT_MAX / watermark_height)

  973.                 return -1;

  974. 

  975.             buf = av_malloc(buf_len);

  976.             av_log(avctx, AV_LOG_DEBUG, "watermark size: %dx%d\n",

  977.                    watermark_width, watermark_height);

  978.             av_log(avctx, AV_LOG_DEBUG,

  979.                    "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n",

  980.                    u1, u2, u3, u4, offset);

  981.             if (uncompress(buf, &buf_len, extradata + 8 + offset,

  982.                            size - offset) != Z_OK) {

  983.                 av_log(avctx, AV_LOG_ERROR,

  984.                        "could not uncompress watermark logo\n");

  985.                 av_free(buf);

  986.                 return -1;

  987.             }

  988.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  989.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  990.             av_log(avctx, AV_LOG_DEBUG,

  991.                    "watermark key %#x\n", s->watermark_key);

  992.             av_free(buf);

  993. #else

  994.             av_log(avctx, AV_LOG_ERROR,

  995.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  996.             return -1;

  997. #endif

  998.         }

  999.     }

  1000. 

  1001.     h->width  = avctx->width;

  1002.     h->height = avctx->height;

  1003.     h->mb_width  = (h->width + 15) / 16;

  1004.     h->mb_height = (h->height + 15) / 16;

  1005.     h->mb_stride = h->mb_width + 1;

  1006.     h->mb_num    = h->mb_width * h->mb_height;

  1007.     h->b_stride = 4 * h->mb_width;

  1008.     s->h_edge_pos = h->mb_width * 16;

  1009.     s->v_edge_pos = h->mb_height * 16;

  1010. 

  1011.     if (ff_h264_alloc_tables(h) < 0) {

  1012.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1013.         return AVERROR(ENOMEM);

  1014.     }

  1015. 

  1016.     return 0;

  1017. }

  1018. 

  1019. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1020. {

  1021.     SVQ3Context *s = avctx->priv_data;

  1022.     H264Context *h = &s->h;

  1023.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1024.     const int mb_array_size = h->mb_stride * h->mb_height;

  1025.     const int b4_stride     = h->mb_width * 4 + 1;

  1026.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1027.     int ret;

  1028. 

  1029.     if (!pic->motion_val_base[0]) {

  1030.         int i;

  1031. 

  1032.         pic->mb_type_base = av_mallocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1033.         if (!pic->mb_type_base)

  1034.             return AVERROR(ENOMEM);

  1035.         pic->f.mb_type = pic->mb_type_base + 2 * h->mb_stride + 1;

  1036. 

  1037.         for (i = 0; i < 2; i++) {

  1038.             pic->motion_val_base[i] = av_mallocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1039.             pic->f.ref_index[i]     = av_mallocz(4 * mb_array_size);

  1040.             if (!pic->motion_val_base[i] || !pic->f.ref_index[i])

  1041.                 return AVERROR(ENOMEM);

  1042. 

  1043.             pic->f.motion_val[i] = pic->motion_val_base[i] + 4;

  1044.         }

  1045.     }

  1046.     pic->f.motion_subsample_log2 = 2;

  1047.     pic->f.reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1048. 

  1049.     ret = ff_get_buffer(avctx, &pic->f);

  1050.     if (!h->edge_emu_buffer) {

  1051.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1052.         if (!h->edge_emu_buffer)

  1053.             return AVERROR(ENOMEM);

  1054.     }

  1055. 

  1056.     h->linesize   = pic->f.linesize[0];

  1057.     h->uvlinesize = pic->f.linesize[1];

  1058. 

  1059.     return ret;

  1060. }

  1061. 

  1062. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1063.                              int *got_frame, AVPacket *avpkt)

  1064. {

  1065.     const uint8_t *buf = avpkt->data;

  1066.     SVQ3Context *s     = avctx->priv_data;

  1067.     H264Context *h     = &s->h;

  1068.     int buf_size       = avpkt->size;

  1069.     int ret, m, i;

  1070. 

  1071.     /* special case for last picture */

  1072.     if (buf_size == 0) {

  1073.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1074.             *(AVFrame *) data   = s->next_pic->f;

  1075.             s->last_frame_output = 1;

  1076.             *got_frame          = 1;

  1077.         }

  1078.         return 0;

  1079.     }

  1080. 

  1081.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1082. 

  1083.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1084. 

  1085.     if (svq3_decode_slice_header(avctx))

  1086.         return -1;

  1087. 

  1088.     h->pict_type = h->slice_type;

  1089. 

  1090.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1091.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1092. 

  1093.     if (s->cur_pic->f.data[0])

  1094.         avctx->release_buffer(avctx, &s->cur_pic->f);

  1095. 

  1096.     /* for skipping the frame */

  1097.     s->cur_pic->f.pict_type = h->pict_type;

  1098.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1099. 

  1100.     ret = get_buffer(avctx, s->cur_pic);

  1101.     if (ret < 0)

  1102.         return ret;

  1103. 

  1104.     h->cur_pic_ptr = s->cur_pic;

  1105.     h->cur_pic     = *s->cur_pic;

  1106. 

  1107.     for (i = 0; i < 16; i++) {

  1108.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1109.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1110.     }

  1111.     for (i = 0; i < 16; i++) {

  1112.         h->block_offset[16 + i]      =

  1113.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1114.         h->block_offset[48 + 16 + i] =

  1115.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1116.     }

  1117. 

  1118.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1119.         if (!s->last_pic->f.data[0]) {

  1120.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1121.             ret = get_buffer(avctx, s->last_pic);

  1122.             if (ret < 0)

  1123.                 return ret;

  1124.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1125.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1126.                    s->last_pic->f.linesize[1]);

  1127.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1128.                    s->last_pic->f.linesize[2]);

  1129.         }

  1130. 

  1131.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1132.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1133.             ret = get_buffer(avctx, s->next_pic);

  1134.             if (ret < 0)

  1135.                 return ret;

  1136.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1137.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1138.                    s->next_pic->f.linesize[1]);

  1139.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1140.                    s->next_pic->f.linesize[2]);

  1141.         }

  1142.     }

  1143. 

  1144.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1145.         av_log(h->avctx, AV_LOG_DEBUG,

  1146.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1147.                av_get_picture_type_char(h->pict_type),

  1148.                s->halfpel_flag, s->thirdpel_flag,

  1149.                s->adaptive_quant, h->qscale, h->slice_num);

  1150. 

  1151.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1152.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1153.         avctx->skip_frame >= AVDISCARD_ALL)

  1154.         return 0;

  1155. 

  1156.     if (s->next_p_frame_damaged) {

  1157.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1158.             return 0;

  1159.         else

  1160.             s->next_p_frame_damaged = 0;

  1161.     }

  1162. 

  1163.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1164.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1165. 

  1166.         if (h->frame_num_offset < 0)

  1167.             h->frame_num_offset += 256;

  1168.         if (h->frame_num_offset == 0 ||

  1169.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1170.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1171.             return -1;

  1172.         }

  1173.     } else {

  1174.         h->prev_frame_num        = h->frame_num;

  1175.         h->frame_num             = h->slice_num;

  1176.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1177. 

  1178.         if (h->prev_frame_num_offset < 0)

  1179.             h->prev_frame_num_offset += 256;

  1180.     }

  1181. 

  1182.     for (m = 0; m < 2; m++) {

  1183.         int i;

  1184.         for (i = 0; i < 4; i++) {

  1185.             int j;

  1186.             for (j = -1; j < 4; j++)

  1187.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1188.             if (i < 3)

  1189.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1190.         }

  1191.     }

  1192. 

  1193.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1194.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1195.             unsigned mb_type;

  1196.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1197. 

  1198.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1199.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1200.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1201.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1202.                 h->gb.size_in_bits = 8 * buf_size;

  1203. 

  1204.                 if (svq3_decode_slice_header(avctx))

  1205.                     return -1;

  1206. 

  1207.                 /* TODO: support s->mb_skip_run */

  1208.             }

  1209. 

  1210.             mb_type = svq3_get_ue_golomb(&h->gb);

  1211. 

  1212.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1213.                 mb_type += 8;

  1214.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1215.                 mb_type += 4;

  1216.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1217.                 av_log(h->avctx, AV_LOG_ERROR,

  1218.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1219.                 return -1;

  1220.             }

  1221. 

  1222.             if (mb_type != 0)

  1223.                 ff_h264_hl_decode_mb(h);

  1224. 

  1225.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1226.                 h->cur_pic.f.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1227.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1228.         }

  1229. 

  1230.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1231.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1232.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1233.     }

  1234. 

  1235.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1236.         *(AVFrame *)data = s->cur_pic->f;

  1237.     else

  1238.         *(AVFrame *)data = s->last_pic->f;

  1239. 

  1240.     /* Do not output the last pic after seeking. */

  1241.     if (s->last_pic->f.data[0] || h->low_delay)

  1242.         *got_frame = 1;

  1243. 

  1244.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1245.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1246.     }

  1247. 

  1248.     return buf_size;

  1249. }

  1250. 

  1251. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1252. {

  1253.     int i;

  1254.     for (i = 0; i < 2; i++) {

  1255.         av_freep(&pic->motion_val_base[i]);

  1256.         av_freep(&pic->f.ref_index[i]);

  1257.     }

  1258.     av_freep(&pic->mb_type_base);

  1259. 

  1260.     if (pic->f.data[0])

  1261.         avctx->release_buffer(avctx, &pic->f);

  1262.     av_freep(&pic);

  1263. }

  1264. 

  1265. static int svq3_decode_end(AVCodecContext *avctx)

  1266. {

  1267.     SVQ3Context *s = avctx->priv_data;

  1268.     H264Context *h = &s->h;

  1269. 

  1270.     free_picture(avctx, s->cur_pic);

  1271.     free_picture(avctx, s->next_pic);

  1272.     free_picture(avctx, s->last_pic);

  1273. 

  1274.     ff_h264_free_context(h);

  1275. 

  1276.     return 0;

  1277. }

  1278. 

  1279. AVCodec ff_svq3_decoder = {

  1280.     .name           = "svq3",

  1281.     .type           = AVMEDIA_TYPE_VIDEO,

  1282.     .id             = AV_CODEC_ID_SVQ3,

  1283.     .priv_data_size = sizeof(SVQ3Context),

  1284.     .init           = svq3_decode_init,

  1285.     .close          = svq3_decode_end,

  1286.     .decode         = svq3_decode_frame,

  1287.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1288.                       CODEC_CAP_DR1             |

  1289.                       CODEC_CAP_DELAY,

  1290.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1291.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1292.                                                      AV_PIX_FMT_NONE},

  1293. };