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.
47832 Commits
32 Branches
421MB
Tree: b6e7041f90
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'b6e7041f90'
${ noResults }
FFmpeg/libavcodec/svq3.c

1209 lines
43KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The FFmpeg Project

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

  6.  * FFmpeg 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.  * FFmpeg 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 FFmpeg; 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.mplayerhq.hu/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. 

  61. /**

  62.  * @file

  63.  * svq3 decoder.

  64.  */

  65. 

  66. typedef struct {

  67.     H264Context h;

  68.     int halfpel_flag;

  69.     int thirdpel_flag;

  70.     int unknown_flag;

  71.     int next_slice_index;

  72.     uint32_t watermark_key;

  73.     uint8_t *buf;

  74.     int buf_size;

  75. } SVQ3Context;

  76. 

  77. #define FULLPEL_MODE  1

  78. #define HALFPEL_MODE  2

  79. #define THIRDPEL_MODE 3

  80. #define PREDICT_MODE  4

  81. 

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

  83.  * o-->o-->o   o

  84.  *         |  /|

  85.  * o   o   o / o

  86.  * | / |   |/  |

  87.  * o   o   o   o

  88.  *   /

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

  90.  */

  91. static const uint8_t svq3_scan[16] = {

  92.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  93.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

  94.     0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4,

  95.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  96. };

  97. 

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

  99.     { 0, 0 },

  100.     { 1, 0 }, { 0, 1 },

  101.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  102.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  103.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  104.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  105.     { 2, 4 }, { 3, 3 }, { 4, 2 },

  106.     { 4, 3 }, { 3, 4 },

  107.     { 4, 4 }

  108. };

  109. 

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

  111.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  112.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

  113.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  4,  3 }, { 0, 1,  2,  4,  3 },

  114.       { 0,  2,  1,  4,  3 }, { 2, 0,  1,  3,  4 }, { 0, 4,  2,  1,  3 } },

  115.     { { 2,  0, -1, -1, -1 }, { 2, 1,  0,  4,  3 }, { 1, 2,  4,  0,  3 },

  116.       { 2,  1,  0,  4,  3 }, { 2, 1,  4,  3,  0 }, { 1, 2,  4,  0,  3 } },

  117.     { { 2,  0, -1, -1, -1 }, { 2, 0,  1,  4,  3 }, { 1, 2,  0,  4,  3 },

  118.       { 2,  1,  0,  4,  3 }, { 2, 1,  3,  4,  0 }, { 2, 4,  1,  0,  3 } },

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

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

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

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

  123. };

  124. 

  125. static const struct {

  126.     uint8_t run;

  127.     uint8_t level;

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

  129.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  130.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  131.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  132.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  133. };

  134. 

  135. static const uint32_t svq3_dequant_coeff[32] = {

  136.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  137.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  138.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  139.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  140. };

  141. 

  142. void ff_svq3_luma_dc_dequant_idct_c(DCTELEM *output, DCTELEM *input, int qp)

  143. {

  144.     const int qmul = svq3_dequant_coeff[qp];

  145. #define stride 16

  146.     int i;

  147.     int temp[16];

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

  149. 

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

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

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

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

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

  155. 

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

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

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

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

  160.     }

  161. 

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

  163.         const int offset = x_offset[i];

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

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

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

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

  168. 

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

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

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

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

  173.     }

  174. }

  175. #undef stride

  176. 

  177. void ff_svq3_add_idct_c(uint8_t *dst, DCTELEM *block,

  178.                         int stride, int qp, int dc)

  179. {

  180.     const int qmul = svq3_dequant_coeff[qp];

  181.     int i;

  182. 

  183.     if (dc) {

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

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

  186.         block[0] = 0;

  187.     }

  188. 

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

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

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

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

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

  194. 

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

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

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

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

  199.     }

  200. 

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

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

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

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

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

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

  207. 

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

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

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

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

  212.     }

  213. }

  214. 

  215. static inline int svq3_decode_block(GetBitContext *gb, DCTELEM *block,

  216.                                     int index, const int type)

  217. {

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

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

  220. 

  221.     int run, level, sign, limit;

  222.     unsigned vlc;

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

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

  225. 

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

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

  228.             if ((int32_t)vlc < 0)

  229.                 return -1;

  230. 

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

  232.             vlc      = vlc + 1 >> 1;

  233. 

  234.             if (type == 3) {

  235.                 if (vlc < 3) {

  236.                     run   = 0;

  237.                     level = vlc;

  238.                 } else if (vlc < 4) {

  239.                     run   = 1;

  240.                     level = 1;

  241.                 } else {

  242.                     run   = vlc & 0x3;

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

  244.                 }

  245.             } else {

  246.                 if (vlc < 16U) {

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

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

  249.                 } else if (intra) {

  250.                     run   = vlc & 0x7;

  251.                     level = (vlc >> 3) + ((run == 0) ? 8 : ((run < 2) ? 2 : ((run < 5) ? 0 : -1)));

  252.                 } else {

  253.                     run   = vlc & 0xF;

  254.                     level = (vlc >> 4) + ((run == 0) ? 4 : ((run < 3) ? 2 : ((run < 10) ? 1 : 0)));

  255.                 }

  256.             }

  257. 

  258. 

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

  260.                 return -1;

  261. 

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

  263.         }

  264. 

  265.         if (type != 2) {

  266.             break;

  267.         }

  268.     }

  269. 

  270.     return 0;

  271. }

  272. 

  273. static inline void svq3_mc_dir_part(MpegEncContext *s,

  274.                                     int x, int y, int width, int height,

  275.                                     int mx, int my, int dxy,

  276.                                     int thirdpel, int dir, int avg)

  277. {

  278.     const Picture *pic = (dir == 0) ? &s->last_picture : &s->next_picture;

  279.     uint8_t *src, *dest;

  280.     int i, emu = 0;

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

  282. 

  283.     mx += x;

  284.     my += y;

  285. 

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

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

  288.         if ((s->flags & CODEC_FLAG_EMU_EDGE))

  289.             emu = 1;

  290. 

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

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

  293.     }

  294. 

  295.     /* form component predictions */

  296.     dest = s->current_picture.f.data[0] + x + y * s->linesize;

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

  298. 

  299.     if (emu) {

  300.         s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize,

  301.                                 width + 1, height + 1,

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

  303.         src = s->edge_emu_buffer;

  304.     }

  305.     if (thirdpel)

  306.         (avg ? s->dsp.avg_tpel_pixels_tab

  307.              : s->dsp.put_tpel_pixels_tab)[dxy](dest, src, s->linesize,

  308.                                                 width, height);

  309.     else

  310.         (avg ? s->dsp.avg_pixels_tab

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

  312.                                                       height);

  313. 

  314.     if (!(s->flags & CODEC_FLAG_GRAY)) {

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

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

  317.         width  = width  >> 1;

  318.         height = height >> 1;

  319.         blocksize++;

  320. 

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

  322.             dest = s->current_picture.f.data[i] + (x >> 1) + (y >> 1) * s->uvlinesize;

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

  324. 

  325.             if (emu) {

  326.                 s->dsp.emulated_edge_mc(s->edge_emu_buffer, src, s->uvlinesize,

  327.                                         width + 1, height + 1,

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

  329.                                         s->v_edge_pos >> 1);

  330.                 src = s->edge_emu_buffer;

  331.             }

  332.             if (thirdpel)

  333.                 (avg ? s->dsp.avg_tpel_pixels_tab

  334.                      : s->dsp.put_tpel_pixels_tab)[dxy](dest, src,

  335.                                                         s->uvlinesize,

  336.                                                         width, height);

  337.             else

  338.                 (avg ? s->dsp.avg_pixels_tab

  339.                      : s->dsp.put_pixels_tab)[blocksize][dxy](dest, src,

  340.                                                               s->uvlinesize,

  341.                                                               height);

  342.         }

  343.     }

  344. }

  345. 

  346. static inline int svq3_mc_dir(H264Context *h, int size, int mode,

  347.                               int dir, int avg)

  348. {

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

  350.     MpegEncContext *const s = (MpegEncContext *)h;

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

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

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

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

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

  356. 

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

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

  359.             const int b_xy = (4 * s->mb_x + (j >> 2)) +

  360.                              (4 * s->mb_y + (i >> 2)) * h->b_stride;

  361.             int dxy;

  362.             x = 16 * s->mb_x + j;

  363.             y = 16 * s->mb_y + i;

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

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

  366. 

  367.             if (mode != PREDICT_MODE) {

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

  369.             } else {

  370.                 mx = s->next_picture.f.motion_val[0][b_xy][0] << 1;

  371.                 my = s->next_picture.f.motion_val[0][b_xy][1] << 1;

  372. 

  373.                 if (dir == 0) {

  374.                     mx = mx * h->frame_num_offset /

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

  376.                     my = my * h->frame_num_offset /

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

  378.                 } else {

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

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

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

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

  383.                 }

  384.             }

  385. 

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

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

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

  389. 

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

  391.             if (mode == PREDICT_MODE) {

  392.                 dx = dy = 0;

  393.             } else {

  394.                 dy = svq3_get_se_golomb(&s->gb);

  395.                 dx = svq3_get_se_golomb(&s->gb);

  396. 

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

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

  399.                     return -1;

  400.                 }

  401.             }

  402. 

  403.             /* compute motion vector */

  404.             if (mode == THIRDPEL_MODE) {

  405.                 int fx, fy;

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

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

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

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

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

  411. 

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

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

  414.                 mx += mx;

  415.                 my += my;

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

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

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

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

  420. 

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

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

  423.                 mx *= 3;

  424.                 my *= 3;

  425.             } else {

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

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

  428. 

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

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

  431.                 mx *= 6;

  432.                 my *= 6;

  433.             }

  434. 

  435.             /* update mv_cache */

  436.             if (mode != PREDICT_MODE) {

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

  438. 

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

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

  441. 

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

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

  444.                 }

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

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

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

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

  449.             }

  450. 

  451.             /* write back motion vectors */

  452.             fill_rectangle(s->current_picture.f.motion_val[dir][b_xy],

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

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

  455.         }

  456. 

  457.     return 0;

  458. }

  459. 

  460. static int svq3_decode_mb(SVQ3Context *svq3, unsigned int mb_type)

  461. {

  462.     H264Context *h = &svq3->h;

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

  464.     int cbp = 0;

  465.     uint32_t vlc;

  466.     int8_t *top, *left;

  467.     MpegEncContext *const s = (MpegEncContext *)h;

  468.     const int mb_xy         = h->mb_xy;

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

  470. 

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

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

  473.     h->topright_samples_available = 0xFFFF;

  474. 

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

  476.         if (s->pict_type == AV_PICTURE_TYPE_P ||

  477.             s->next_picture.f.mb_type[mb_xy] == -1) {

  478.             svq3_mc_dir_part(s, 16 * s->mb_x, 16 * s->mb_y, 16, 16,

  479.                              0, 0, 0, 0, 0, 0);

  480. 

  481.             if (s->pict_type == AV_PICTURE_TYPE_B)

  482.                 svq3_mc_dir_part(s, 16 * s->mb_x, 16 * s->mb_y, 16, 16,

  483.                                  0, 0, 0, 0, 1, 1);

  484. 

  485.             mb_type = MB_TYPE_SKIP;

  486.         } else {

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

  488.             if (svq3_mc_dir(h, mb_type, PREDICT_MODE, 0, 0) < 0)

  489.                 return -1;

  490.             if (svq3_mc_dir(h, mb_type, PREDICT_MODE, 1, 1) < 0)

  491.                 return -1;

  492. 

  493.             mb_type = MB_TYPE_16x16;

  494.         }

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

  496.         if (svq3->thirdpel_flag && svq3->halfpel_flag == !get_bits1(&s->gb))

  497.             mode = THIRDPEL_MODE;

  498.         else if (svq3->halfpel_flag &&

  499.                  svq3->thirdpel_flag == !get_bits1(&s->gb))

  500.             mode = HALFPEL_MODE;

  501.         else

  502.             mode = FULLPEL_MODE;

  503. 

  504.         /* fill caches */

  505.         /* note ref_cache should contain here:

  506.          *  ????????

  507.          *  ???11111

  508.          *  N??11111

  509.          *  N??11111

  510.          *  N??11111

  511.          */

  512. 

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

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

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

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

  517.                               s->current_picture.f.motion_val[m][b_xy - 1 + i * h->b_stride]);

  518.             } else {

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

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

  521.             }

  522.             if (s->mb_y > 0) {

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

  524.                        s->current_picture.f.motion_val[m][b_xy - h->b_stride],

  525.                        4 * 2 * sizeof(int16_t));

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

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

  528. 

  529.                 if (s->mb_x < s->mb_width - 1) {

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

  531.                               s->current_picture.f.motion_val[m][b_xy - h->b_stride + 4]);

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

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

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

  535.                 } else

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

  537.                 if (s->mb_x > 0) {

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

  539.                               s->current_picture.f.motion_val[m][b_xy - h->b_stride - 1]);

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

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

  542.                 } else

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

  544.             } else

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

  546.                        PART_NOT_AVAILABLE, 8);

  547. 

  548.             if (s->pict_type != AV_PICTURE_TYPE_B)

  549.                 break;

  550.         }

  551. 

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

  553.         if (s->pict_type == AV_PICTURE_TYPE_P) {

  554.             if (svq3_mc_dir(h, mb_type - 1, mode, 0, 0) < 0)

  555.                 return -1;

  556.         } else {        /* AV_PICTURE_TYPE_B */

  557.             if (mb_type != 2) {

  558.                 if (svq3_mc_dir(h, 0, mode, 0, 0) < 0)

  559.                     return -1;

  560.             } else {

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

  562.                     memset(s->current_picture.f.motion_val[0][b_xy + i * h->b_stride],

  563.                            0, 4 * 2 * sizeof(int16_t));

  564.             }

  565.             if (mb_type != 1) {

  566.                 if (svq3_mc_dir(h, 0, mode, 1, mb_type == 3) < 0)

  567.                     return -1;

  568.             } else {

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

  570.                     memset(s->current_picture.f.motion_val[1][b_xy + i * h->b_stride],

  571.                            0, 4 * 2 * sizeof(int16_t));

  572.             }

  573.         }

  574. 

  575.         mb_type = MB_TYPE_16x16;

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

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

  578. 

  579.         if (mb_type == 8) {

  580.             if (s->mb_x > 0) {

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

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

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

  584.                     h->left_samples_available = 0x5F5F;

  585.             }

  586.             if (s->mb_y > 0) {

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

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

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

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

  591. 

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

  593.                     h->top_samples_available = 0x33FF;

  594.             }

  595. 

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

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

  598.                 vlc = svq3_get_ue_golomb(&s->gb);

  599. 

  600.                 if (vlc >= 25U) {

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

  602.                     return -1;

  603.                 }

  604. 

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

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

  607. 

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

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

  610. 

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

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

  613.                     return -1;

  614.                 }

  615.             }

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

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

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

  619.         }

  620. 

  621.         write_back_intra_pred_mode(h);

  622. 

  623.         if (mb_type == 8) {

  624.             ff_h264_check_intra4x4_pred_mode(h);

  625. 

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

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

  628.         } else {

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

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

  631. 

  632.             h->top_samples_available  = 0x33FF;

  633.             h->left_samples_available = 0x5F5F;

  634.         }

  635. 

  636.         mb_type = MB_TYPE_INTRA4x4;

  637.     } else {                      /* INTRA16x16 */

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

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

  640. 

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

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

  643.             return -1;

  644.         }

  645. 

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

  647.         mb_type = MB_TYPE_INTRA16x16;

  648.     }

  649. 

  650.     if (!IS_INTER(mb_type) && s->pict_type != AV_PICTURE_TYPE_I) {

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

  652.             memset(s->current_picture.f.motion_val[0][b_xy + i * h->b_stride],

  653.                    0, 4 * 2 * sizeof(int16_t));

  654.         if (s->pict_type == AV_PICTURE_TYPE_B) {

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

  656.                 memset(s->current_picture.f.motion_val[1][b_xy + i * h->b_stride],

  657.                        0, 4 * 2 * sizeof(int16_t));

  658.         }

  659.     }

  660.     if (!IS_INTRA4x4(mb_type)) {

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

  662.     }

  663.     if (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B) {

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

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

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

  667.     }

  668. 

  669.     if (!IS_INTRA16x16(mb_type) &&

  670.         (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B)) {

  671.         if ((vlc = svq3_get_ue_golomb(&s->gb)) >= 48U){

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

  673.             return -1;

  674.         }

  675. 

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

  677.                                 : golomb_to_inter_cbp[vlc];

  678.     }

  679.     if (IS_INTRA16x16(mb_type) ||

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

  681.         s->qscale += svq3_get_se_golomb(&s->gb);

  682. 

  683.         if (s->qscale > 31u) {

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

  685.             return -1;

  686.         }

  687.     }

  688.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  692.             av_log(h->s.avctx, AV_LOG_ERROR,

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

  694.             return -1;

  695.         }

  696.     }

  697. 

  698.     if (cbp) {

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

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

  701. 

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

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

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

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

  706.                                  2 * (j & 2) + 4 * (i & 2))

  707.                               : (4 * i + j);

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

  709. 

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

  711.                         av_log(h->s.avctx, AV_LOG_ERROR,

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

  713.                         return -1;

  714.                     }

  715.                 }

  716.             }

  717. 

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

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

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

  721.                     av_log(h->s.avctx, AV_LOG_ERROR,

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

  723.                     return -1;

  724.                 }

  725. 

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

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

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

  729.                         k                                 = 16 * i + j;

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

  731. 

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

  733.                             av_log(h->s.avctx, AV_LOG_ERROR,

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

  735.                             return -1;

  736.                         }

  737.                     }

  738.                 }

  739.             }

  740.         }

  741.     }

  742. 

  743.     h->cbp                              = cbp;

  744.     s->current_picture.f.mb_type[mb_xy] = mb_type;

  745. 

  746.     if (IS_INTRA(mb_type))

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

  748. 

  749.     return 0;

  750. }

  751. 

  752. static int svq3_decode_slice_header(AVCodecContext *avctx)

  753. {

  754.     SVQ3Context *svq3 = avctx->priv_data;

  755.     H264Context *h    = &svq3->h;

  756.     MpegEncContext *s = &h->s;

  757.     const int mb_xy   = h->mb_xy;

  758.     int i, header;

  759.     unsigned slice_id;

  760. 

  761.     header = get_bits(&s->gb, 8);

  762. 

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

  764.         /* TODO: what? */

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

  766.         return -1;

  767.     } else {

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

  769. 

  770.         svq3->next_slice_index = get_bits_count(&s->gb) +

  771.                                  8 * show_bits(&s->gb, 8 * length) +

  772.                                  8 * length;

  773. 

  774.         if (svq3->next_slice_index > s->gb.size_in_bits) {

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

  776.             return -1;

  777.         }

  778. 

  779.         s->gb.size_in_bits = svq3->next_slice_index - 8 * (length - 1);

  780.         skip_bits(&s->gb, 8);

  781. 

  782.         if (svq3->watermark_key) {

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

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

  785.                     header ^ svq3->watermark_key);

  786.         }

  787.         if (length > 0) {

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

  789.                    &s->gb.buffer[s->gb.size_in_bits >> 3], length - 1);

  790.         }

  791.         skip_bits_long(&s->gb, 0);

  792.     }

  793. 

  794.     if ((slice_id = svq3_get_ue_golomb(&s->gb)) >= 3) {

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

  796.         return -1;

  797.     }

  798. 

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

  800. 

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

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

  803.         s->mb_skip_run = get_bits(&s->gb, i) -

  804.                          (s->mb_y * s->mb_width + s->mb_x);

  805.     } else {

  806.         skip_bits1(&s->gb);

  807.         s->mb_skip_run = 0;

  808.     }

  809. 

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

  811.     s->qscale         = get_bits(&s->gb, 5);

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

  813. 

  814.     /* unknown fields */

  815.     skip_bits1(&s->gb);

  816. 

  817.     if (svq3->unknown_flag)

  818.         skip_bits1(&s->gb);

  819. 

  820.     skip_bits1(&s->gb);

  821.     skip_bits(&s->gb, 2);

  822. 

  823.     while (get_bits1(&s->gb))

  824.         skip_bits(&s->gb, 8);

  825. 

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

  827.     if (s->mb_x > 0) {

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

  829.                -1, 4 * sizeof(int8_t));

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

  831.                -1, 8 * sizeof(int8_t) * s->mb_x);

  832.     }

  833.     if (s->mb_y > 0) {

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

  835.                -1, 8 * sizeof(int8_t) * (s->mb_width - s->mb_x));

  836. 

  837.         if (s->mb_x > 0)

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

  839.     }

  840. 

  841.     return 0;

  842. }

  843. 

  844. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  845. {

  846.     SVQ3Context *svq3 = avctx->priv_data;

  847.     H264Context *h    = &svq3->h;

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

  849.     int m;

  850.     unsigned char *extradata;

  851.     unsigned char *extradata_end;

  852.     unsigned int size;

  853.     int marker_found = 0;

  854. 

  855.     if (ff_h264_decode_init(avctx) < 0)

  856.         return -1;

  857. 

  858.     s->flags           = avctx->flags;

  859.     s->flags2          = avctx->flags2;

  860.     s->unrestricted_mv = 1;

  861.     h->is_complex      = 1;

  862.     h->sps.chroma_format_idc = 1;

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

  864. 

  865.     if (!s->context_initialized) {

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

  867. 

  868.         svq3->halfpel_flag  = 1;

  869.         svq3->thirdpel_flag = 1;

  870.         svq3->unknown_flag  = 0;

  871. 

  872. 

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

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

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

  876.         if (extradata) {

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

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

  879.                     marker_found = 1;

  880.                     break;

  881.                 }

  882.                 extradata++;

  883.             }

  884.         }

  885. 

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

  887.         if (marker_found) {

  888.             GetBitContext gb;

  889.             int frame_size_code;

  890. 

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

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

  893.                 return AVERROR_INVALIDDATA;

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

  895. 

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

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

  898.             switch (frame_size_code) {

  899.             case 0:

  900.                 avctx->width  = 160;

  901.                 avctx->height = 120;

  902.                 break;

  903.             case 1:

  904.                 avctx->width  = 128;

  905.                 avctx->height =  96;

  906.                 break;

  907.             case 2:

  908.                 avctx->width  = 176;

  909.                 avctx->height = 144;

  910.                 break;

  911.             case 3:

  912.                 avctx->width  = 352;

  913.                 avctx->height = 288;

  914.                 break;

  915.             case 4:

  916.                 avctx->width  = 704;

  917.                 avctx->height = 576;

  918.                 break;

  919.             case 5:

  920.                 avctx->width  = 240;

  921.                 avctx->height = 180;

  922.                 break;

  923.             case 6:

  924.                 avctx->width  = 320;

  925.                 avctx->height = 240;

  926.                 break;

  927.             case 7:

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

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

  930.                 break;

  931.             }

  932. 

  933.             svq3->halfpel_flag  = get_bits1(&gb);

  934.             svq3->thirdpel_flag = get_bits1(&gb);

  935. 

  936.             /* unknown fields */

  937.             skip_bits1(&gb);

  938.             skip_bits1(&gb);

  939.             skip_bits1(&gb);

  940.             skip_bits1(&gb);

  941. 

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

  943. 

  944.             /* unknown field */

  945.             skip_bits1(&gb);

  946. 

  947.             while (get_bits1(&gb))

  948.                 skip_bits(&gb, 8);

  949. 

  950.             svq3->unknown_flag  = get_bits1(&gb);

  951.             avctx->has_b_frames = !s->low_delay;

  952.             if (svq3->unknown_flag) {

  953. #if CONFIG_ZLIB

  954.                 unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  955.                 unsigned watermark_height = svq3_get_ue_golomb(&gb);

  956.                 int u1                    = svq3_get_ue_golomb(&gb);

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

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

  959.                 int u4                    = svq3_get_ue_golomb(&gb);

  960.                 unsigned long buf_len     = watermark_width *

  961.                                             watermark_height * 4;

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

  963.                 uint8_t *buf;

  964. 

  965.                 if (watermark_height <= 0 || (uint64_t)watermark_width*4 > UINT_MAX/watermark_height)

  966.                     return -1;

  967. 

  968.                 buf = av_malloc(buf_len);

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

  970.                        watermark_width, watermark_height);

  971.                 av_log(avctx, AV_LOG_DEBUG,

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

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

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

  975.                                size - offset) != Z_OK) {

  976.                     av_log(avctx, AV_LOG_ERROR,

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

  978.                     av_free(buf);

  979.                     return -1;

  980.                 }

  981.                 svq3->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  982.                 svq3->watermark_key = svq3->watermark_key << 16 |

  983.                                       svq3->watermark_key;

  984.                 av_log(avctx, AV_LOG_DEBUG,

  985.                        "watermark key %#x\n", svq3->watermark_key);

  986.                 av_free(buf);

  987. #else

  988.                 av_log(avctx, AV_LOG_ERROR,

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

  990.                 return -1;

  991. #endif

  992.             }

  993.         }

  994. 

  995.         s->width  = avctx->width;

  996.         s->height = avctx->height;

  997. 

  998.         if (ff_MPV_common_init(s) < 0)

  999.             return -1;

  1000. 

  1001.         h->b_stride = 4 * s->mb_width;

  1002. 

  1003.         if (ff_h264_alloc_tables(h) < 0) {

  1004.             av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1005.             return AVERROR(ENOMEM);

  1006.         }

  1007.     }

  1008. 

  1009.     return 0;

  1010. }

  1011. 

  1012. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1013.                              int *got_frame, AVPacket *avpkt)

  1014. {

  1015.     SVQ3Context *svq3  = avctx->priv_data;

  1016.     H264Context *h     = &svq3->h;

  1017.     MpegEncContext *s  = &h->s;

  1018.     int buf_size       = avpkt->size;

  1019.     int m, left;

  1020.     uint8_t *buf;

  1021. 

  1022.     /* special case for last picture */

  1023.     if (buf_size == 0) {

  1024.         if (s->next_picture_ptr && !s->low_delay) {

  1025.             *(AVFrame *) data   = s->next_picture.f;

  1026.             s->next_picture_ptr = NULL;

  1027.             *got_frame          = 1;

  1028.         }

  1029.         return 0;

  1030.     }

  1031. 

  1032.     s->mb_x = s->mb_y = h->mb_xy = 0;

  1033. 

  1034.     if (svq3->watermark_key) {

  1035.         av_fast_malloc(&svq3->buf, &svq3->buf_size,

  1036.                        buf_size+FF_INPUT_BUFFER_PADDING_SIZE);

  1037.         if (!svq3->buf)

  1038.             return AVERROR(ENOMEM);

  1039.         memcpy(svq3->buf, avpkt->data, buf_size);

  1040.         buf = svq3->buf;

  1041.     } else {

  1042.         buf = avpkt->data;

  1043.     }

  1044. 

  1045.     init_get_bits(&s->gb, buf, 8 * buf_size);

  1046. 

  1047.     if (svq3_decode_slice_header(avctx))

  1048.         return -1;

  1049. 

  1050.     s->pict_type      = h->slice_type;

  1051.     s->picture_number = h->slice_num;

  1052. 

  1053.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1054.         av_log(h->s.avctx, AV_LOG_DEBUG,

  1055.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1056.                av_get_picture_type_char(s->pict_type),

  1057.                svq3->halfpel_flag, svq3->thirdpel_flag,

  1058.                s->adaptive_quant, s->qscale, h->slice_num);

  1059. 

  1060.     /* for skipping the frame */

  1061.     s->current_picture.f.pict_type = s->pict_type;

  1062.     s->current_picture.f.key_frame = (s->pict_type == AV_PICTURE_TYPE_I);

  1063. 

  1064.     /* Skip B-frames if we do not have reference frames. */

  1065.     if (s->last_picture_ptr == NULL && s->pict_type == AV_PICTURE_TYPE_B)

  1066.         return 0;

  1067.     if (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B ||

  1068.         avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I ||

  1069.         avctx->skip_frame >= AVDISCARD_ALL)

  1070.         return 0;

  1071. 

  1072.     if (s->next_p_frame_damaged) {

  1073.         if (s->pict_type == AV_PICTURE_TYPE_B)

  1074.             return 0;

  1075.         else

  1076.             s->next_p_frame_damaged = 0;

  1077.     }

  1078. 

  1079.     if (ff_h264_frame_start(h) < 0)

  1080.         return -1;

  1081. 

  1082.     if (s->pict_type == AV_PICTURE_TYPE_B) {

  1083.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1084. 

  1085.         if (h->frame_num_offset < 0)

  1086.             h->frame_num_offset += 256;

  1087.         if (h->frame_num_offset == 0 ||

  1088.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1089.             av_log(h->s.avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1090.             return -1;

  1091.         }

  1092.     } else {

  1093.         h->prev_frame_num        = h->frame_num;

  1094.         h->frame_num             = h->slice_num;

  1095.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1096. 

  1097.         if (h->prev_frame_num_offset < 0)

  1098.             h->prev_frame_num_offset += 256;

  1099.     }

  1100. 

  1101.     for (m = 0; m < 2; m++) {

  1102.         int i;

  1103.         for (i = 0; i < 4; i++) {

  1104.             int j;

  1105.             for (j = -1; j < 4; j++)

  1106.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1107.             if (i < 3)

  1108.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1109.         }

  1110.     }

  1111. 

  1112.     for (s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {

  1113.         for (s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {

  1114.             unsigned mb_type;

  1115.             h->mb_xy = s->mb_x + s->mb_y * s->mb_stride;

  1116. 

  1117.             if ((get_bits_count(&s->gb) + 7) >= s->gb.size_in_bits &&

  1118.                 ((get_bits_count(&s->gb) & 7) == 0 ||

  1119.                  show_bits(&s->gb, -get_bits_count(&s->gb) & 7) == 0)) {

  1120.                 skip_bits(&s->gb, svq3->next_slice_index - get_bits_count(&s->gb));

  1121.                 s->gb.size_in_bits = 8 * buf_size;

  1122. 

  1123.                 if (svq3_decode_slice_header(avctx))

  1124.                     return -1;

  1125. 

  1126.                 /* TODO: support s->mb_skip_run */

  1127.             }

  1128. 

  1129.             mb_type = svq3_get_ue_golomb(&s->gb);

  1130. 

  1131.             if (s->pict_type == AV_PICTURE_TYPE_I)

  1132.                 mb_type += 8;

  1133.             else if (s->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1134.                 mb_type += 4;

  1135.             if (mb_type > 33 || svq3_decode_mb(svq3, mb_type)) {

  1136.                 av_log(h->s.avctx, AV_LOG_ERROR,

  1137.                        "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  1138.                 return -1;

  1139.             }

  1140. 

  1141.             if (mb_type != 0)

  1142.                 ff_h264_hl_decode_mb(h);

  1143. 

  1144.             if (s->pict_type != AV_PICTURE_TYPE_B && !s->low_delay)

  1145.                 s->current_picture.f.mb_type[s->mb_x + s->mb_y * s->mb_stride] =

  1146.                     (s->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1147.         }

  1148. 

  1149.         ff_draw_horiz_band(s, 16 * s->mb_y, 16);

  1150.     }

  1151. 

  1152.     left = buf_size*8 - get_bits_count(&s->gb);

  1153. 

  1154.     if (s->mb_y != s->mb_height || s->mb_x != s->mb_width) {

  1155.         av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, s->mb_y, s->mb_x, left);

  1156.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1157.     }

  1158. 

  1159.     if (left < 0) {

  1160.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1161.         return -1;

  1162.     }

  1163. 

  1164.     ff_MPV_frame_end(s);

  1165. 

  1166.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay)

  1167.         *(AVFrame *)data = s->current_picture.f;

  1168.     else

  1169.         *(AVFrame *)data = s->last_picture.f;

  1170. 

  1171.     /* Do not output the last pic after seeking. */

  1172.     if (s->last_picture_ptr || s->low_delay)

  1173.         *got_frame = 1;

  1174. 

  1175.     return buf_size;

  1176. }

  1177. 

  1178. static int svq3_decode_end(AVCodecContext *avctx)

  1179. {

  1180.     SVQ3Context *svq3 = avctx->priv_data;

  1181.     H264Context *h    = &svq3->h;

  1182.     MpegEncContext *s = &h->s;

  1183. 

  1184.     ff_h264_free_context(h);

  1185. 

  1186.     ff_MPV_common_end(s);

  1187. 

  1188.     av_freep(&svq3->buf);

  1189.     svq3->buf_size = 0;

  1190. 

  1191.     return 0;

  1192. }

  1193. 

  1194. AVCodec ff_svq3_decoder = {

  1195.     .name           = "svq3",

  1196.     .type           = AVMEDIA_TYPE_VIDEO,

  1197.     .id             = AV_CODEC_ID_SVQ3,

  1198.     .priv_data_size = sizeof(SVQ3Context),

  1199.     .init           = svq3_decode_init,

  1200.     .close          = svq3_decode_end,

  1201.     .decode         = svq3_decode_frame,

  1202.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1203.                       CODEC_CAP_DR1             |

  1204.                       CODEC_CAP_DELAY,

  1205.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1206.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1207.                                                      AV_PIX_FMT_NONE},

  1208. };