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.
36686 Commits
32 Branches
844MB
Tree: fc386f2eea
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'fc386f2eea'
${ noResults }
FFmpeg/libavcodec/svq3.c

1295 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 "dsputil.h"

  44. #include "avcodec.h"

  45. #include "mpegvideo.h"

  46. #include "h264.h"

  47. 

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

  49. 

  50. #include "h264_mvpred.h"

  51. #include "golomb.h"

  52. #include "rectangle.h"

  53. #include "vdpau_internal.h"

  54. 

  55. #if CONFIG_ZLIB

  56. #include <zlib.h>

  57. #endif

  58. 

  59. #include "svq1.h"

  60. #include "svq3.h"

  61. 

  62. /**

  63.  * @file

  64.  * svq3 decoder.

  65.  */

  66. 

  67. typedef struct {

  68.     H264Context h;

  69.     Picture *cur_pic;

  70.     Picture *next_pic;

  71.     Picture *last_pic;

  72.     int halfpel_flag;

  73.     int thirdpel_flag;

  74.     int unknown_flag;

  75.     int next_slice_index;

  76.     uint32_t watermark_key;

  77.     int adaptive_quant;

  78.     int next_p_frame_damaged;

  79.     int h_edge_pos;

  80.     int v_edge_pos;

  81.     int last_frame_output;

  82. } SVQ3Context;

  83. 

  84. #define FULLPEL_MODE  1

  85. #define HALFPEL_MODE  2

  86. #define THIRDPEL_MODE 3

  87. #define PREDICT_MODE  4

  88. 

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

  90.  * o-->o-->o   o

  91.  *         |  /|

  92.  * o   o   o / o

  93.  * | / |   |/  |

  94.  * o   o   o   o

  95.  *   /

  96.  * o-->o-->o-->o

  97.  */

  98. static const uint8_t svq3_scan[16] = {

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

  100.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

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

  102.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  103. };

  104. 

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

  106.     { 0, 0 },

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

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

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

  110.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  111.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  112.     { 2, 4 }, { 3, 3 }, { 4, 2 },

  113.     { 4, 3 }, { 3, 4 },

  114.     { 4, 4 }

  115. };

  116. 

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

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

  119.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

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

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

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

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

  129.       { 4,  2,  0,  1,  3 }, { 2, 0,  1,  4,  3 }, { 4, 2,  1,  0,  3 } },

  130. };

  131. 

  132. static const struct {

  133.     uint8_t run;

  134.     uint8_t level;

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

  136.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  137.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  138.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  139.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  140. };

  141. 

  142. static const uint32_t svq3_dequant_coeff[32] = {

  143.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  144.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  145.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  146.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  147. };

  148. 

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

  150. {

  151.     const int qmul = svq3_dequant_coeff[qp];

  152. #define stride 16

  153.     int i;

  154.     int temp[16];

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

  156. 

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

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

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

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

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

  162. 

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

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

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

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

  167.     }

  168. 

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

  170.         const int offset = x_offset[i];

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

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

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

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

  175. 

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

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

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

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

  180.     }

  181. }

  182. #undef stride

  183. 

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

  185.                         int stride, int qp, int dc)

  186. {

  187.     const int qmul = svq3_dequant_coeff[qp];

  188.     int i;

  189. 

  190.     if (dc) {

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

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

  193.         block[0] = 0;

  194.     }

  195. 

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

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

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

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

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

  201. 

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

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

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

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

  206.     }

  207. 

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

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

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

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

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

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

  214. 

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

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

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

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

  219.     }

  220. }

  221. 

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

  223.                                     int index, const int type)

  224. {

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

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

  227. 

  228.     int run, level, limit;

  229.     unsigned vlc;

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

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

  232. 

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

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

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

  236.             vlc      = vlc + 1 >> 1;

  237. 

  238.             if (type == 3) {

  239.                 if (vlc < 3) {

  240.                     run   = 0;

  241.                     level = vlc;

  242.                 } else if (vlc < 4) {

  243.                     run   = 1;

  244.                     level = 1;

  245.                 } else {

  246.                     run   = vlc & 0x3;

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

  248.                 }

  249.             } else {

  250.                 if (vlc < 16) {

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

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

  253.                 } else if (intra) {

  254.                     run   = vlc & 0x7;

  255.                     level = (vlc >> 3) +

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

  257.                 } else {

  258.                     run   = vlc & 0xF;

  259.                     level = (vlc >> 4) +

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

  261.                 }

  262.             }

  263. 

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

  265.                 return -1;

  266. 

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

  268.         }

  269. 

  270.         if (type != 2) {

  271.             break;

  272.         }

  273.     }

  274. 

  275.     return 0;

  276. }

  277. 

  278. static inline void svq3_mc_dir_part(SVQ3Context *s,

  279.                                     int x, int y, int width, int height,

  280.                                     int mx, int my, int dxy,

  281.                                     int thirdpel, int dir, int avg)

  282. {

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

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

  285.     uint8_t *src, *dest;

  286.     int i, emu = 0;

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

  288. 

  289.     mx += x;

  290.     my += y;

  291. 

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

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

  294.         emu = 1;

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

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

  297.     }

  298. 

  299.     /* form component predictions */

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

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

  302. 

  303.     if (emu) {

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

  305.                                  width + 1, height + 1,

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

  307.         src = h->edge_emu_buffer;

  308.     }

  309.     if (thirdpel)

  310.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  312.                                                 width, height);

  313.     else

  314.         (avg ? h->dsp.avg_pixels_tab

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

  316.                                                       height);

  317. 

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

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

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

  321.         width  = width  >> 1;

  322.         height = height >> 1;

  323.         blocksize++;

  324. 

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

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

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

  328. 

  329.             if (emu) {

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

  331.                                          width + 1, height + 1,

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

  333.                                          s->v_edge_pos >> 1);

  334.                 src = h->edge_emu_buffer;

  335.             }

  336.             if (thirdpel)

  337.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  339.                                                         h->uvlinesize,

  340.                                                         width, height);

  341.             else

  342.                 (avg ? h->dsp.avg_pixels_tab

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

  344.                                                               h->uvlinesize,

  345.                                                               height);

  346.         }

  347.     }

  348. }

  349. 

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

  351.                               int dir, int avg)

  352. {

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

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

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

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

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

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

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

  360. 

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

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

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

  364.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  365.             int dxy;

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

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

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

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

  370. 

  371.             if (mode != PREDICT_MODE) {

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

  373.             } else {

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

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

  376. 

  377.                 if (dir == 0) {

  378.                     mx = mx * h->frame_num_offset /

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

  380.                     my = my * h->frame_num_offset /

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

  382.                 } else {

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

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

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

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

  387.                 }

  388.             }

  389. 

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

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

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

  393. 

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

  395.             if (mode == PREDICT_MODE) {

  396.                 dx = dy = 0;

  397.             } else {

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

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

  400. 

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

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

  403.                     return -1;

  404.                 }

  405.             }

  406. 

  407.             /* compute motion vector */

  408.             if (mode == THIRDPEL_MODE) {

  409.                 int fx, fy;

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

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

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

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

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

  415. 

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

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

  418.                 mx += mx;

  419.                 my += my;

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

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

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

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

  424. 

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

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

  427.                 mx *= 3;

  428.                 my *= 3;

  429.             } else {

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

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

  432. 

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

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

  435.                 mx *= 6;

  436.                 my *= 6;

  437.             }

  438. 

  439.             /* update mv_cache */

  440.             if (mode != PREDICT_MODE) {

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

  442. 

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

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

  445. 

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

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

  448.                 }

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

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

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

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

  453.             }

  454. 

  455.             /* write back motion vectors */

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

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

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

  459.         }

  460. 

  461.     return 0;

  462. }

  463. 

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

  465. {

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

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

  468.     int cbp = 0;

  469.     uint32_t vlc;

  470.     int8_t *top, *left;

  471.     const int mb_xy         = h->mb_xy;

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

  473. 

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

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

  476.     h->topright_samples_available = 0xFFFF;

  477. 

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

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

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

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

  482.                              0, 0, 0, 0, 0, 0);

  483. 

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

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

  486.                                  0, 0, 0, 0, 1, 1);

  487. 

  488.             mb_type = MB_TYPE_SKIP;

  489.         } else {

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

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

  492.                 return -1;

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

  494.                 return -1;

  495. 

  496.             mb_type = MB_TYPE_16x16;

  497.         }

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

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

  500.             mode = THIRDPEL_MODE;

  501.         else if (s->halfpel_flag &&

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

  503.             mode = HALFPEL_MODE;

  504.         else

  505.             mode = FULLPEL_MODE;

  506. 

  507.         /* fill caches */

  508.         /* note ref_cache should contain here:

  509.          *  ????????

  510.          *  ???11111

  511.          *  N??11111

  512.          *  N??11111

  513.          *  N??11111

  514.          */

  515. 

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

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

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

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

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

  521.             } else {

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

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

  524.             }

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

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

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

  528.                        4 * 2 * sizeof(int16_t));

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

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

  531. 

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

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

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

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

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

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

  538.                 } else

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

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

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

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

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

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

  545.                 } else

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

  547.             } else

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

  549.                        PART_NOT_AVAILABLE, 8);

  550. 

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

  552.                 break;

  553.         }

  554. 

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

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

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

  558.                 return -1;

  559.         } else {        /* AV_PICTURE_TYPE_B */

  560.             if (mb_type != 2) {

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

  562.                     return -1;

  563.             } else {

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

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

  566.                            0, 4 * 2 * sizeof(int16_t));

  567.             }

  568.             if (mb_type != 1) {

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

  570.                     return -1;

  571.             } else {

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

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

  574.                            0, 4 * 2 * sizeof(int16_t));

  575.             }

  576.         }

  577. 

  578.         mb_type = MB_TYPE_16x16;

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

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

  581. 

  582.         if (mb_type == 8) {

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

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

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

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

  587.                     h->left_samples_available = 0x5F5F;

  588.             }

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

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

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

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

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

  594. 

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

  596.                     h->top_samples_available = 0x33FF;

  597.             }

  598. 

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

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

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

  602. 

  603.                 if (vlc >= 25) {

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

  605.                     return -1;

  606.                 }

  607. 

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

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

  610. 

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

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

  613. 

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

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

  616.                     return -1;

  617.                 }

  618.             }

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

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

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

  622.         }

  623. 

  624.         write_back_intra_pred_mode(h);

  625. 

  626.         if (mb_type == 8) {

  627.             ff_h264_check_intra4x4_pred_mode(h);

  628. 

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

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

  631.         } else {

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

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

  634. 

  635.             h->top_samples_available  = 0x33FF;

  636.             h->left_samples_available = 0x5F5F;

  637.         }

  638. 

  639.         mb_type = MB_TYPE_INTRA4x4;

  640.     } else {                      /* INTRA16x16 */

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

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

  643. 

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

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

  646.             return -1;

  647.         }

  648. 

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

  650.         mb_type = MB_TYPE_INTRA16x16;

  651.     }

  652. 

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

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

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

  656.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  660.                        0, 4 * 2 * sizeof(int16_t));

  661.         }

  662.     }

  663.     if (!IS_INTRA4x4(mb_type)) {

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

  665.     }

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

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

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

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

  670.     }

  671. 

  672.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  676.             return -1;

  677.         }

  678. 

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

  680.                                 : golomb_to_inter_cbp[vlc];

  681.     }

  682.     if (IS_INTRA16x16(mb_type) ||

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

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

  685. 

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

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

  688.             return -1;

  689.         }

  690.     }

  691.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  695.             av_log(h->avctx, AV_LOG_ERROR,

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

  697.             return -1;

  698.         }

  699.     }

  700. 

  701.     if (cbp) {

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

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

  704. 

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

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

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

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

  709.                                  2 * (j & 2) + 4 * (i & 2))

  710.                               : (4 * i + j);

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

  712. 

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

  714.                         av_log(h->avctx, AV_LOG_ERROR,

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

  716.                         return -1;

  717.                     }

  718.                 }

  719.             }

  720. 

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

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

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

  724.                     av_log(h->avctx, AV_LOG_ERROR,

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

  726.                     return -1;

  727.                 }

  728. 

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

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

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

  732.                         k                                 = 16 * i + j;

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

  734. 

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

  736.                             av_log(h->avctx, AV_LOG_ERROR,

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

  738.                             return -1;

  739.                         }

  740.                     }

  741.                 }

  742.             }

  743.         }

  744.     }

  745. 

  746.     h->cbp                              = cbp;

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

  748. 

  749.     if (IS_INTRA(mb_type))

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

  751. 

  752.     return 0;

  753. }

  754. 

  755. static int svq3_decode_slice_header(AVCodecContext *avctx)

  756. {

  757.     SVQ3Context *s = avctx->priv_data;

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

  759.     const int mb_xy   = h->mb_xy;

  760.     int i, header;

  761.     unsigned slice_id;

  762. 

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

  764. 

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

  766.         /* TODO: what? */

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

  768.         return -1;

  769.     } else {

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

  771. 

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

  773.                               8 * show_bits(&h->gb, 8 * length) +

  774.                               8 * length;

  775. 

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

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

  778.             return -1;

  779.         }

  780. 

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

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

  783. 

  784.         if (s->watermark_key) {

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

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

  787.                     header ^ s->watermark_key);

  788.         }

  789.         if (length > 0) {

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

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

  792.         }

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

  794.     }

  795. 

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

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

  798.         return -1;

  799.     }

  800. 

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

  802. 

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

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

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

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

  807.     } else {

  808.         skip_bits1(&h->gb);

  809.         h->mb_skip_run = 0;

  810.     }

  811. 

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

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

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

  815. 

  816.     /* unknown fields */

  817.     skip_bits1(&h->gb);

  818. 

  819.     if (s->unknown_flag)

  820.         skip_bits1(&h->gb);

  821. 

  822.     skip_bits1(&h->gb);

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

  824. 

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

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

  827. 

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

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

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

  831.                -1, 4 * sizeof(int8_t));

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

  833.                -1, 8 * sizeof(int8_t) * h->mb_x);

  834.     }

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

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

  837.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  838. 

  839.         if (h->mb_x > 0)

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

  841.     }

  842. 

  843.     return 0;

  844. }

  845. 

  846. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  847. {

  848.     SVQ3Context *s = avctx->priv_data;

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

  850.     int m;

  851.     unsigned char *extradata;

  852.     unsigned char *extradata_end;

  853.     unsigned int size;

  854.     int marker_found = 0;

  855. 

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

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

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

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

  860.         av_freep(&s->cur_pic);

  861.         av_freep(&s->last_pic);

  862.         av_freep(&s->next_pic);

  863.         return AVERROR(ENOMEM);

  864.     }

  865. 

  866.     if (ff_h264_decode_init(avctx) < 0)

  867.         return -1;

  868. 

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

  870.     h->is_complex      = 1;

  871.     h->picture_structure = PICT_FRAME;

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

  873. 

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

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

  876. 

  877.     s->halfpel_flag  = 1;

  878.     s->thirdpel_flag = 1;

  879.     s->unknown_flag  = 0;

  880. 

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

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

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

  884.     if (extradata) {

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

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

  887.                 marker_found = 1;

  888.                 break;

  889.             }

  890.             extradata++;

  891.         }

  892.     }

  893. 

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

  895.     if (marker_found) {

  896.         GetBitContext gb;

  897.         int frame_size_code;

  898. 

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

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

  901.             return AVERROR_INVALIDDATA;

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

  903. 

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

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

  906.         switch (frame_size_code) {

  907.         case 0:

  908.             avctx->width  = 160;

  909.             avctx->height = 120;

  910.             break;

  911.         case 1:

  912.             avctx->width  = 128;

  913.             avctx->height =  96;

  914.             break;

  915.         case 2:

  916.             avctx->width  = 176;

  917.             avctx->height = 144;

  918.             break;

  919.         case 3:

  920.             avctx->width  = 352;

  921.             avctx->height = 288;

  922.             break;

  923.         case 4:

  924.             avctx->width  = 704;

  925.             avctx->height = 576;

  926.             break;

  927.         case 5:

  928.             avctx->width  = 240;

  929.             avctx->height = 180;

  930.             break;

  931.         case 6:

  932.             avctx->width  = 320;

  933.             avctx->height = 240;

  934.             break;

  935.         case 7:

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

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

  938.             break;

  939.         }

  940. 

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

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

  943. 

  944.         /* unknown fields */

  945.         skip_bits1(&gb);

  946.         skip_bits1(&gb);

  947.         skip_bits1(&gb);

  948.         skip_bits1(&gb);

  949. 

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

  951. 

  952.         /* unknown field */

  953.         skip_bits1(&gb);

  954. 

  955.         while (get_bits1(&gb))

  956.             skip_bits(&gb, 8);

  957. 

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

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

  960.         if (s->unknown_flag) {

  961. #if CONFIG_ZLIB

  962.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  963.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  964.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  967.             int u4                    = svq3_get_ue_golomb(&gb);

  968.             unsigned long buf_len     = watermark_width *

  969.                                         watermark_height * 4;

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

  971.             uint8_t *buf;

  972. 

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

  974.                 return -1;

  975. 

  976.             buf = av_malloc(buf_len);

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

  978.                    watermark_width, watermark_height);

  979.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  983.                            size - offset) != Z_OK) {

  984.                 av_log(avctx, AV_LOG_ERROR,

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

  986.                 av_free(buf);

  987.                 return -1;

  988.             }

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

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

  991.             av_log(avctx, AV_LOG_DEBUG,

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

  993.             av_free(buf);

  994. #else

  995.             av_log(avctx, AV_LOG_ERROR,

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

  997.             return -1;

  998. #endif

  999.         }

  1000.     }

  1001. 

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

  1003.     h->height = avctx->height;

  1004.     h->mb_width  = (h->width + 15) / 16;

  1005.     h->mb_height = (h->height + 15) / 16;

  1006.     h->mb_stride = h->mb_width + 1;

  1007.     h->mb_num    = h->mb_width * h->mb_height;

  1008.     h->b_stride = 4 * h->mb_width;

  1009.     s->h_edge_pos = h->mb_width * 16;

  1010.     s->v_edge_pos = h->mb_height * 16;

  1011. 

  1012.     if (ff_h264_alloc_tables(h) < 0) {

  1013.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1014.         return AVERROR(ENOMEM);

  1015.     }

  1016. 

  1017.     return 0;

  1018. }

  1019. 

  1020. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1021. {

  1022.     SVQ3Context *s = avctx->priv_data;

  1023.     H264Context *h = &s->h;

  1024.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1025.     const int mb_array_size = h->mb_stride * h->mb_height;

  1026.     const int b4_stride     = h->mb_width * 4 + 1;

  1027.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1028.     int ret;

  1029. 

  1030.     if (!pic->motion_val_base[0]) {

  1031.         int i;

  1032. 

  1033.         pic->mb_type_base = av_mallocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1034.         if (!pic->mb_type_base)

  1035.             return AVERROR(ENOMEM);

  1036.         pic->f.mb_type = pic->mb_type_base + 2 * h->mb_stride + 1;

  1037. 

  1038.         for (i = 0; i < 2; i++) {

  1039.             pic->motion_val_base[i] = av_mallocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1040.             pic->f.ref_index[i]     = av_mallocz(4 * mb_array_size);

  1041.             if (!pic->motion_val_base[i] || !pic->f.ref_index[i])

  1042.                 return AVERROR(ENOMEM);

  1043. 

  1044.             pic->f.motion_val[i] = pic->motion_val_base[i] + 4;

  1045.         }

  1046.     }

  1047.     pic->f.motion_subsample_log2 = 2;

  1048.     pic->f.reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1049. 

  1050.     ret = ff_get_buffer(avctx, &pic->f);

  1051.     if (!h->edge_emu_buffer) {

  1052.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1053.         if (!h->edge_emu_buffer)

  1054.             return AVERROR(ENOMEM);

  1055.     }

  1056. 

  1057.     h->linesize   = pic->f.linesize[0];

  1058.     h->uvlinesize = pic->f.linesize[1];

  1059. 

  1060.     return ret;

  1061. }

  1062. 

  1063. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1064.                              int *got_frame, AVPacket *avpkt)

  1065. {

  1066.     const uint8_t *buf = avpkt->data;

  1067.     SVQ3Context *s     = avctx->priv_data;

  1068.     H264Context *h     = &s->h;

  1069.     int buf_size       = avpkt->size;

  1070.     int ret, m, i;

  1071. 

  1072.     /* special case for last picture */

  1073.     if (buf_size == 0) {

  1074.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1075.             *(AVFrame *) data   = s->next_pic->f;

  1076.             s->last_frame_output = 1;

  1077.             *got_frame          = 1;

  1078.         }

  1079.         return 0;

  1080.     }

  1081. 

  1082.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1083. 

  1084.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1085. 

  1086.     if (svq3_decode_slice_header(avctx))

  1087.         return -1;

  1088. 

  1089.     h->pict_type = h->slice_type;

  1090. 

  1091.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1092.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1093. 

  1094.     if (s->cur_pic->f.data[0])

  1095.         avctx->release_buffer(avctx, &s->cur_pic->f);

  1096. 

  1097.     /* for skipping the frame */

  1098.     s->cur_pic->f.pict_type = h->pict_type;

  1099.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1100. 

  1101.     ret = get_buffer(avctx, s->cur_pic);

  1102.     if (ret < 0)

  1103.         return ret;

  1104. 

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

  1106.     h->cur_pic     = *s->cur_pic;

  1107. 

  1108.     for (i = 0; i < 16; i++) {

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

  1110.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1111.     }

  1112.     for (i = 0; i < 16; i++) {

  1113.         h->block_offset[16 + i]      =

  1114.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1115.         h->block_offset[48 + 16 + i] =

  1116.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1117.     }

  1118. 

  1119.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1120.         if (!s->last_pic->f.data[0]) {

  1121.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1122.             ret = get_buffer(avctx, s->last_pic);

  1123.             if (ret < 0)

  1124.                 return ret;

  1125.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1126.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1127.                    s->last_pic->f.linesize[1]);

  1128.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1129.                    s->last_pic->f.linesize[2]);

  1130.         }

  1131. 

  1132.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1133.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1134.             ret = get_buffer(avctx, s->next_pic);

  1135.             if (ret < 0)

  1136.                 return ret;

  1137.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1138.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1139.                    s->next_pic->f.linesize[1]);

  1140.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1141.                    s->next_pic->f.linesize[2]);

  1142.         }

  1143.     }

  1144. 

  1145.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1146.         av_log(h->avctx, AV_LOG_DEBUG,

  1147.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1148.                av_get_picture_type_char(h->pict_type),

  1149.                s->halfpel_flag, s->thirdpel_flag,

  1150.                s->adaptive_quant, h->qscale, h->slice_num);

  1151. 

  1152.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1153.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1154.         avctx->skip_frame >= AVDISCARD_ALL)

  1155.         return 0;

  1156. 

  1157.     if (s->next_p_frame_damaged) {

  1158.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1159.             return 0;

  1160.         else

  1161.             s->next_p_frame_damaged = 0;

  1162.     }

  1163. 

  1164.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1165.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1166. 

  1167.         if (h->frame_num_offset < 0)

  1168.             h->frame_num_offset += 256;

  1169.         if (h->frame_num_offset == 0 ||

  1170.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1171.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1172.             return -1;

  1173.         }

  1174.     } else {

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

  1176.         h->frame_num             = h->slice_num;

  1177.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1178. 

  1179.         if (h->prev_frame_num_offset < 0)

  1180.             h->prev_frame_num_offset += 256;

  1181.     }

  1182. 

  1183.     for (m = 0; m < 2; m++) {

  1184.         int i;

  1185.         for (i = 0; i < 4; i++) {

  1186.             int j;

  1187.             for (j = -1; j < 4; j++)

  1188.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1189.             if (i < 3)

  1190.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1191.         }

  1192.     }

  1193. 

  1194.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1195.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1196.             unsigned mb_type;

  1197.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1198. 

  1199.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1200.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1201.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1202.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1203.                 h->gb.size_in_bits = 8 * buf_size;

  1204. 

  1205.                 if (svq3_decode_slice_header(avctx))

  1206.                     return -1;

  1207. 

  1208.                 /* TODO: support s->mb_skip_run */

  1209.             }

  1210. 

  1211.             mb_type = svq3_get_ue_golomb(&h->gb);

  1212. 

  1213.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1214.                 mb_type += 8;

  1215.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1216.                 mb_type += 4;

  1217.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1218.                 av_log(h->avctx, AV_LOG_ERROR,

  1219.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1220.                 return -1;

  1221.             }

  1222. 

  1223.             if (mb_type != 0)

  1224.                 ff_h264_hl_decode_mb(h);

  1225. 

  1226.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1227.                 h->cur_pic.f.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1228.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1229.         }

  1230. 

  1231.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1232.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1233.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1234.     }

  1235. 

  1236.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1237.         *(AVFrame *)data = s->cur_pic->f;

  1238.     else

  1239.         *(AVFrame *)data = s->last_pic->f;

  1240. 

  1241.     /* Do not output the last pic after seeking. */

  1242.     if (s->last_pic->f.data[0] || h->low_delay)

  1243.         *got_frame = 1;

  1244. 

  1245.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1246.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1247.     }

  1248. 

  1249.     return buf_size;

  1250. }

  1251. 

  1252. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1253. {

  1254.     int i;

  1255.     for (i = 0; i < 2; i++) {

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

  1257.         av_freep(&pic->f.ref_index[i]);

  1258.     }

  1259.     av_freep(&pic->mb_type_base);

  1260. 

  1261.     if (pic->f.data[0])

  1262.         avctx->release_buffer(avctx, &pic->f);

  1263.     av_freep(&pic);

  1264. }

  1265. 

  1266. static int svq3_decode_end(AVCodecContext *avctx)

  1267. {

  1268.     SVQ3Context *s = avctx->priv_data;

  1269.     H264Context *h = &s->h;

  1270. 

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

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

  1273.     free_picture(avctx, s->last_pic);

  1274. 

  1275.     ff_h264_free_context(h);

  1276. 

  1277.     return 0;

  1278. }

  1279. 

  1280. AVCodec ff_svq3_decoder = {

  1281.     .name           = "svq3",

  1282.     .type           = AVMEDIA_TYPE_VIDEO,

  1283.     .id             = AV_CODEC_ID_SVQ3,

  1284.     .priv_data_size = sizeof(SVQ3Context),

  1285.     .init           = svq3_decode_init,

  1286.     .close          = svq3_decode_end,

  1287.     .decode         = svq3_decode_frame,

  1288.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1289.                       CODEC_CAP_DR1             |

  1290.                       CODEC_CAP_DELAY,

  1291.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1292.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1293.                                                      AV_PIX_FMT_NONE},

  1294. };