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.
50020 Commits
32 Branches
411MB
Tree: ddbbfd95be
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ddbbfd95be'
${ noResults }
FFmpeg/libavcodec/svq3.c

1207 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. #include "svq3.h"

  61. 

  62. /**

  63.  * @file

  64.  * svq3 decoder.

  65.  */

  66. 

  67. typedef struct {

  68.     H264Context h;

  69.     int halfpel_flag;

  70.     int thirdpel_flag;

  71.     int unknown_flag;

  72.     int next_slice_index;

  73.     uint32_t watermark_key;

  74.     uint8_t *buf;

  75.     int buf_size;

  76. } SVQ3Context;

  77. 

  78. #define FULLPEL_MODE  1

  79. #define HALFPEL_MODE  2

  80. #define THIRDPEL_MODE 3

  81. #define PREDICT_MODE  4

  82. 

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

  84.  * o-->o-->o   o

  85.  *         |  /|

  86.  * o   o   o / o

  87.  * | / |   |/  |

  88.  * o   o   o   o

  89.  *   /

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

  91.  */

  92. static const uint8_t svq3_scan[16] = {

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

  94.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

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

  96.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  97. };

  98. 

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

  100.     { 0, 0 },

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

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

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

  104.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  105.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  106.     { 2, 4 }, { 3, 3 }, { 4, 2 },

  107.     { 4, 3 }, { 3, 4 },

  108.     { 4, 4 }

  109. };

  110. 

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

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

  113.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

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

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

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

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

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

  124. };

  125. 

  126. static const struct {

  127.     uint8_t run;

  128.     uint8_t level;

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

  130.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  131.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  132.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  133.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  134. };

  135. 

  136. static const uint32_t svq3_dequant_coeff[32] = {

  137.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  138.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  139.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  140.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  141. };

  142. 

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

  144. {

  145.     const int qmul = svq3_dequant_coeff[qp];

  146. #define stride 16

  147.     int i;

  148.     int temp[16];

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

  150. 

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

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

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

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

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

  156. 

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

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

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

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

  161.     }

  162. 

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

  164.         const int offset = x_offset[i];

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

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

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

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

  169. 

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

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

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

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

  174.     }

  175. }

  176. #undef stride

  177. 

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

  179.                         int stride, int qp, int dc)

  180. {

  181.     const int qmul = svq3_dequant_coeff[qp];

  182.     int i;

  183. 

  184.     if (dc) {

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

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

  187.         block[0] = 0;

  188.     }

  189. 

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

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

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

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

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

  195. 

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

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

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

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

  200.     }

  201. 

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

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

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

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

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

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

  208. 

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

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

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

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

  213.     }

  214. }

  215. 

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

  217.                                     int index, const int type)

  218. {

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

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

  221. 

  222.     int run, level, sign, limit;

  223.     unsigned vlc;

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

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

  226. 

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

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

  229.             if ((int32_t)vlc < 0)

  230.                 return -1;

  231. 

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

  233.             vlc      = vlc + 1 >> 1;

  234. 

  235.             if (type == 3) {

  236.                 if (vlc < 3) {

  237.                     run   = 0;

  238.                     level = vlc;

  239.                 } else if (vlc < 4) {

  240.                     run   = 1;

  241.                     level = 1;

  242.                 } else {

  243.                     run   = vlc & 0x3;

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

  245.                 }

  246.             } else {

  247.                 if (vlc < 16U) {

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

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

  250.                 } else if (intra) {

  251.                     run   = vlc & 0x7;

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

  253.                 } else {

  254.                     run   = vlc & 0xF;

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

  256.                 }

  257.             }

  258. 

  259. 

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

  261.                 return -1;

  262. 

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

  264.         }

  265. 

  266.         if (type != 2) {

  267.             break;

  268.         }

  269.     }

  270. 

  271.     return 0;

  272. }

  273. 

  274. static inline void svq3_mc_dir_part(MpegEncContext *s,

  275.                                     int x, int y, int width, int height,

  276.                                     int mx, int my, int dxy,

  277.                                     int thirdpel, int dir, int avg)

  278. {

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

  280.     uint8_t *src, *dest;

  281.     int i, emu = 0;

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

  283. 

  284.     mx += x;

  285.     my += y;

  286. 

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

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

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

  290.             emu = 1;

  291. 

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

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

  294.     }

  295. 

  296.     /* form component predictions */

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

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

  299. 

  300.     if (emu) {

  301.         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, s->linesize,

  302.                                  width + 1, height + 1,

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

  304.         src = s->edge_emu_buffer;

  305.     }

  306.     if (thirdpel)

  307.         (avg ? s->dsp.avg_tpel_pixels_tab

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

  309.                                                 width, height);

  310.     else

  311.         (avg ? s->dsp.avg_pixels_tab

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

  313.                                                       height);

  314. 

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

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

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

  318.         width  = width  >> 1;

  319.         height = height >> 1;

  320.         blocksize++;

  321. 

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

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

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

  325. 

  326.             if (emu) {

  327.                 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src, s->uvlinesize,

  328.                                          width + 1, height + 1,

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

  330.                                          s->v_edge_pos >> 1);

  331.                 src = s->edge_emu_buffer;

  332.             }

  333.             if (thirdpel)

  334.                 (avg ? s->dsp.avg_tpel_pixels_tab

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

  336.                                                         s->uvlinesize,

  337.                                                         width, height);

  338.             else

  339.                 (avg ? s->dsp.avg_pixels_tab

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

  341.                                                               s->uvlinesize,

  342.                                                               height);

  343.         }

  344.     }

  345. }

  346. 

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

  348.                               int dir, int avg)

  349. {

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

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

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

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

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

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

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

  357. 

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

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

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

  361.                              (4 * s->mb_y + (i >> 2)) * h->b_stride;

  362.             int dxy;

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

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

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

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

  367. 

  368.             if (mode != PREDICT_MODE) {

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

  370.             } else {

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

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

  373. 

  374.                 if (dir == 0) {

  375.                     mx = mx * h->frame_num_offset /

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

  377.                     my = my * h->frame_num_offset /

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

  379.                 } else {

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

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

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

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

  384.                 }

  385.             }

  386. 

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

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

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

  390. 

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

  392.             if (mode == PREDICT_MODE) {

  393.                 dx = dy = 0;

  394.             } else {

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

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

  397. 

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

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

  400.                     return -1;

  401.                 }

  402.             }

  403. 

  404.             /* compute motion vector */

  405.             if (mode == THIRDPEL_MODE) {

  406.                 int fx, fy;

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

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

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

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

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

  412. 

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

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

  415.                 mx += mx;

  416.                 my += my;

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

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

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

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

  421. 

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

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

  424.                 mx *= 3;

  425.                 my *= 3;

  426.             } else {

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

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

  429. 

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

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

  432.                 mx *= 6;

  433.                 my *= 6;

  434.             }

  435. 

  436.             /* update mv_cache */

  437.             if (mode != PREDICT_MODE) {

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

  439. 

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

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

  442. 

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

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

  445.                 }

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

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

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

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

  450.             }

  451. 

  452.             /* write back motion vectors */

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

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

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

  456.         }

  457. 

  458.     return 0;

  459. }

  460. 

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

  462. {

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

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

  465.     int cbp = 0;

  466.     uint32_t vlc;

  467.     int8_t *top, *left;

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

  469.     const int mb_xy         = h->mb_xy;

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

  471. 

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

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

  474.     h->topright_samples_available = 0xFFFF;

  475. 

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

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

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

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

  480.                              0, 0, 0, 0, 0, 0);

  481. 

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

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

  484.                                  0, 0, 0, 0, 1, 1);

  485. 

  486.             mb_type = MB_TYPE_SKIP;

  487.         } else {

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

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

  490.                 return -1;

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

  492.                 return -1;

  493. 

  494.             mb_type = MB_TYPE_16x16;

  495.         }

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

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

  498.             mode = THIRDPEL_MODE;

  499.         else if (svq3->halfpel_flag &&

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

  501.             mode = HALFPEL_MODE;

  502.         else

  503.             mode = FULLPEL_MODE;

  504. 

  505.         /* fill caches */

  506.         /* note ref_cache should contain here:

  507.          *  ????????

  508.          *  ???11111

  509.          *  N??11111

  510.          *  N??11111

  511.          *  N??11111

  512.          */

  513. 

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

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

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

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

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

  519.             } else {

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

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

  522.             }

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

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

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

  526.                        4 * 2 * sizeof(int16_t));

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

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

  529. 

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

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

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

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

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

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

  536.                 } else

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

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

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

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

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

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

  543.                 } else

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

  545.             } else

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

  547.                        PART_NOT_AVAILABLE, 8);

  548. 

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

  550.                 break;

  551.         }

  552. 

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

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

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

  556.                 return -1;

  557.         } else {        /* AV_PICTURE_TYPE_B */

  558.             if (mb_type != 2) {

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

  560.                     return -1;

  561.             } else {

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

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

  564.                            0, 4 * 2 * sizeof(int16_t));

  565.             }

  566.             if (mb_type != 1) {

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

  568.                     return -1;

  569.             } else {

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

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

  572.                            0, 4 * 2 * sizeof(int16_t));

  573.             }

  574.         }

  575. 

  576.         mb_type = MB_TYPE_16x16;

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

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

  579. 

  580.         if (mb_type == 8) {

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

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

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

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

  585.                     h->left_samples_available = 0x5F5F;

  586.             }

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

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

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

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

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

  592. 

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

  594.                     h->top_samples_available = 0x33FF;

  595.             }

  596. 

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

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

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

  600. 

  601.                 if (vlc >= 25U) {

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

  603.                     return -1;

  604.                 }

  605. 

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

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

  608. 

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

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

  611. 

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

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

  614.                     return -1;

  615.                 }

  616.             }

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

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

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

  620.         }

  621. 

  622.         write_back_intra_pred_mode(h);

  623. 

  624.         if (mb_type == 8) {

  625.             ff_h264_check_intra4x4_pred_mode(h);

  626. 

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

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

  629.         } else {

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

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

  632. 

  633.             h->top_samples_available  = 0x33FF;

  634.             h->left_samples_available = 0x5F5F;

  635.         }

  636. 

  637.         mb_type = MB_TYPE_INTRA4x4;

  638.     } else {                      /* INTRA16x16 */

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

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

  641. 

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

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

  644.             return -1;

  645.         }

  646. 

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

  648.         mb_type = MB_TYPE_INTRA16x16;

  649.     }

  650. 

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

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

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

  654.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  658.                        0, 4 * 2 * sizeof(int16_t));

  659.         }

  660.     }

  661.     if (!IS_INTRA4x4(mb_type)) {

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

  663.     }

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

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

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

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

  668.     }

  669. 

  670.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  674.             return -1;

  675.         }

  676. 

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

  678.                                 : golomb_to_inter_cbp[vlc];

  679.     }

  680.     if (IS_INTRA16x16(mb_type) ||

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

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

  683. 

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

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

  686.             return -1;

  687.         }

  688.     }

  689.     if (IS_INTRA16x16(mb_type)) {

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

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

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

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

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

  695.             return -1;

  696.         }

  697.     }

  698. 

  699.     if (cbp) {

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

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

  702. 

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

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

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

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

  707.                                  2 * (j & 2) + 4 * (i & 2))

  708.                               : (4 * i + j);

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

  710. 

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

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

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

  714.                         return -1;

  715.                     }

  716.                 }

  717.             }

  718. 

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

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

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

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

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

  724.                     return -1;

  725.                 }

  726. 

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

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

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

  730.                         k                                 = 16 * i + j;

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

  732. 

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

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

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

  736.                             return -1;

  737.                         }

  738.                     }

  739.                 }

  740.             }

  741.         }

  742.     }

  743. 

  744.     h->cbp                              = cbp;

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

  746. 

  747.     if (IS_INTRA(mb_type))

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

  749. 

  750.     return 0;

  751. }

  752. 

  753. static int svq3_decode_slice_header(AVCodecContext *avctx)

  754. {

  755.     SVQ3Context *svq3 = avctx->priv_data;

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

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

  758.     const int mb_xy   = h->mb_xy;

  759.     int i, header;

  760.     unsigned slice_id;

  761. 

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

  763. 

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

  765.         /* TODO: what? */

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

  767.         return -1;

  768.     } else {

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

  770. 

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

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

  773.                                  8 * length;

  774. 

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

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

  777.             return -1;

  778.         }

  779. 

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

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

  782. 

  783.         if (svq3->watermark_key) {

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

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

  786.                     header ^ svq3->watermark_key);

  787.         }

  788.         if (length > 0) {

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

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

  791.         }

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

  793.     }

  794. 

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

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

  797.         return -1;

  798.     }

  799. 

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

  801. 

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

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

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

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

  806.     } else {

  807.         skip_bits1(&s->gb);

  808.         s->mb_skip_run = 0;

  809.     }

  810. 

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

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

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

  814. 

  815.     /* unknown fields */

  816.     skip_bits1(&s->gb);

  817. 

  818.     if (svq3->unknown_flag)

  819.         skip_bits1(&s->gb);

  820. 

  821.     skip_bits1(&s->gb);

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

  823. 

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

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

  826. 

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

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

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

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

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

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

  833.     }

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

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

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

  837. 

  838.         if (s->mb_x > 0)

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

  840.     }

  841. 

  842.     return 0;

  843. }

  844. 

  845. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  846. {

  847.     SVQ3Context *svq3 = avctx->priv_data;

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

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

  850.     int m;

  851.     unsigned char *extradata;

  852.     unsigned char *extradata_end;

  853.     unsigned int size;

  854.     int marker_found = 0;

  855. 

  856.     if (ff_h264_decode_init(avctx) < 0)

  857.         return -1;

  858. 

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

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

  861.     s->unrestricted_mv = 1;

  862.     h->is_complex      = 1;

  863.     h->sps.chroma_format_idc = 1;

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

  865. 

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

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

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

  875.     if (extradata) {

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

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

  878.                 marker_found = 1;

  879.                 break;

  880.             }

  881.             extradata++;

  882.         }

  883.     }

  884. 

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

  886.     if (marker_found) {

  887.         GetBitContext gb;

  888.         int frame_size_code;

  889. 

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

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

  892.             return AVERROR_INVALIDDATA;

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

  894. 

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

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

  897.         switch (frame_size_code) {

  898.         case 0:

  899.             avctx->width  = 160;

  900.             avctx->height = 120;

  901.             break;

  902.         case 1:

  903.             avctx->width  = 128;

  904.             avctx->height =  96;

  905.             break;

  906.         case 2:

  907.             avctx->width  = 176;

  908.             avctx->height = 144;

  909.             break;

  910.         case 3:

  911.             avctx->width  = 352;

  912.             avctx->height = 288;

  913.             break;

  914.         case 4:

  915.             avctx->width  = 704;

  916.             avctx->height = 576;

  917.             break;

  918.         case 5:

  919.             avctx->width  = 240;

  920.             avctx->height = 180;

  921.             break;

  922.         case 6:

  923.             avctx->width  = 320;

  924.             avctx->height = 240;

  925.             break;

  926.         case 7:

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

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

  929.             break;

  930.         }

  931. 

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

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

  934. 

  935.         /* unknown fields */

  936.         skip_bits1(&gb);

  937.         skip_bits1(&gb);

  938.         skip_bits1(&gb);

  939.         skip_bits1(&gb);

  940. 

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

  942. 

  943.         /* unknown field */

  944.         skip_bits1(&gb);

  945. 

  946.         while (get_bits1(&gb))

  947.             skip_bits(&gb, 8);

  948. 

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

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

  951.         if (svq3->unknown_flag) {

  952. #if CONFIG_ZLIB

  953.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  954.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  955.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  958.             int u4                    = svq3_get_ue_golomb(&gb);

  959.             unsigned long buf_len     = watermark_width *

  960.                                         watermark_height * 4;

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

  962.             uint8_t *buf;

  963. 

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

  965.                 return -1;

  966. 

  967.             buf = av_malloc(buf_len);

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

  969.                    watermark_width, watermark_height);

  970.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  974.                            size - offset) != Z_OK) {

  975.                 av_log(avctx, AV_LOG_ERROR,

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

  977.                 av_free(buf);

  978.                 return -1;

  979.             }

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

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

  982.                                   svq3->watermark_key;

  983.             av_log(avctx, AV_LOG_DEBUG,

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

  985.             av_free(buf);

  986. #else

  987.             av_log(avctx, AV_LOG_ERROR,

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

  989.             return -1;

  990. #endif

  991.         }

  992.     }

  993. 

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

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

  996. 

  997.     if (ff_MPV_common_init(s) < 0)

  998.         return -1;

  999. 

  1000.     h->b_stride = 4 * s->mb_width;

  1001. 

  1002.     if (ff_h264_alloc_tables(h) < 0) {

  1003.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1004.         return AVERROR(ENOMEM);

  1005.     }

  1006. 

  1007.     return 0;

  1008. }

  1009. 

  1010. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1011.                              int *got_frame, AVPacket *avpkt)

  1012. {

  1013.     SVQ3Context *svq3  = avctx->priv_data;

  1014.     H264Context *h     = &svq3->h;

  1015.     MpegEncContext *s  = &h->s;

  1016.     int buf_size       = avpkt->size;

  1017.     int m, left;

  1018.     uint8_t *buf;

  1019. 

  1020.     /* special case for last picture */

  1021.     if (buf_size == 0) {

  1022.         if (s->next_picture_ptr && !s->low_delay) {

  1023.             *(AVFrame *) data   = s->next_picture.f;

  1024.             s->next_picture_ptr = NULL;

  1025.             *got_frame          = 1;

  1026.         }

  1027.         return 0;

  1028.     }

  1029. 

  1030.     s->mb_x = s->mb_y = h->mb_xy = 0;

  1031. 

  1032.     if (svq3->watermark_key) {

  1033.         av_fast_malloc(&svq3->buf, &svq3->buf_size,

  1034.                        buf_size+FF_INPUT_BUFFER_PADDING_SIZE);

  1035.         if (!svq3->buf)

  1036.             return AVERROR(ENOMEM);

  1037.         memcpy(svq3->buf, avpkt->data, buf_size);

  1038.         buf = svq3->buf;

  1039.     } else {

  1040.         buf = avpkt->data;

  1041.     }

  1042. 

  1043.     init_get_bits(&s->gb, buf, 8 * buf_size);

  1044. 

  1045.     if (svq3_decode_slice_header(avctx))

  1046.         return -1;

  1047. 

  1048.     s->pict_type      = h->slice_type;

  1049.     s->picture_number = h->slice_num;

  1050. 

  1051.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1052.         av_log(h->s.avctx, AV_LOG_DEBUG,

  1053.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1054.                av_get_picture_type_char(s->pict_type),

  1055.                svq3->halfpel_flag, svq3->thirdpel_flag,

  1056.                s->adaptive_quant, s->qscale, h->slice_num);

  1057. 

  1058.     /* for skipping the frame */

  1059.     s->current_picture.f.pict_type = s->pict_type;

  1060.     s->current_picture.f.key_frame = (s->pict_type == AV_PICTURE_TYPE_I);

  1061. 

  1062.     /* Skip B-frames if we do not have reference frames. */

  1063.     if (s->last_picture_ptr == NULL && s->pict_type == AV_PICTURE_TYPE_B)

  1064.         return 0;

  1065.     if (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B ||

  1066.         avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I ||

  1067.         avctx->skip_frame >= AVDISCARD_ALL)

  1068.         return 0;

  1069. 

  1070.     if (s->next_p_frame_damaged) {

  1071.         if (s->pict_type == AV_PICTURE_TYPE_B)

  1072.             return 0;

  1073.         else

  1074.             s->next_p_frame_damaged = 0;

  1075.     }

  1076. 

  1077.     if (ff_h264_frame_start(h) < 0)

  1078.         return -1;

  1079. 

  1080.     if (s->pict_type == AV_PICTURE_TYPE_B) {

  1081.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1082. 

  1083.         if (h->frame_num_offset < 0)

  1084.             h->frame_num_offset += 256;

  1085.         if (h->frame_num_offset == 0 ||

  1086.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1087.             av_log(h->s.avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1088.             return -1;

  1089.         }

  1090.     } else {

  1091.         h->prev_frame_num        = h->frame_num;

  1092.         h->frame_num             = h->slice_num;

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

  1094. 

  1095.         if (h->prev_frame_num_offset < 0)

  1096.             h->prev_frame_num_offset += 256;

  1097.     }

  1098. 

  1099.     for (m = 0; m < 2; m++) {

  1100.         int i;

  1101.         for (i = 0; i < 4; i++) {

  1102.             int j;

  1103.             for (j = -1; j < 4; j++)

  1104.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1105.             if (i < 3)

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

  1107.         }

  1108.     }

  1109. 

  1110.     for (s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {

  1111.         for (s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {

  1112.             unsigned mb_type;

  1113.             h->mb_xy = s->mb_x + s->mb_y * s->mb_stride;

  1114. 

  1115.             if ((get_bits_count(&s->gb) + 7) >= s->gb.size_in_bits &&

  1116.                 ((get_bits_count(&s->gb) & 7) == 0 ||

  1117.                  show_bits(&s->gb, -get_bits_count(&s->gb) & 7) == 0)) {

  1118.                 skip_bits(&s->gb, svq3->next_slice_index - get_bits_count(&s->gb));

  1119.                 s->gb.size_in_bits = 8 * buf_size;

  1120. 

  1121.                 if (svq3_decode_slice_header(avctx))

  1122.                     return -1;

  1123. 

  1124.                 /* TODO: support s->mb_skip_run */

  1125.             }

  1126. 

  1127.             mb_type = svq3_get_ue_golomb(&s->gb);

  1128. 

  1129.             if (s->pict_type == AV_PICTURE_TYPE_I)

  1130.                 mb_type += 8;

  1131.             else if (s->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1132.                 mb_type += 4;

  1133.             if (mb_type > 33 || svq3_decode_mb(svq3, mb_type)) {

  1134.                 av_log(h->s.avctx, AV_LOG_ERROR,

  1135.                        "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  1136.                 return -1;

  1137.             }

  1138. 

  1139.             if (mb_type != 0)

  1140.                 ff_h264_hl_decode_mb(h);

  1141. 

  1142.             if (s->pict_type != AV_PICTURE_TYPE_B && !s->low_delay)

  1143.                 s->current_picture.f.mb_type[s->mb_x + s->mb_y * s->mb_stride] =

  1144.                     (s->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1145.         }

  1146. 

  1147.         ff_mpeg_draw_horiz_band(s, 16 * s->mb_y, 16);

  1148.     }

  1149. 

  1150.     left = buf_size*8 - get_bits_count(&s->gb);

  1151. 

  1152.     if (s->mb_y != s->mb_height || s->mb_x != s->mb_width) {

  1153.         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);

  1154.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1155.     }

  1156. 

  1157.     if (left < 0) {

  1158.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1159.         return -1;

  1160.     }

  1161. 

  1162.     ff_MPV_frame_end(s);

  1163. 

  1164.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay)

  1165.         *(AVFrame *)data = s->current_picture.f;

  1166.     else

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

  1168. 

  1169.     /* Do not output the last pic after seeking. */

  1170.     if (s->last_picture_ptr || s->low_delay)

  1171.         *got_frame = 1;

  1172. 

  1173.     return buf_size;

  1174. }

  1175. 

  1176. static int svq3_decode_end(AVCodecContext *avctx)

  1177. {

  1178.     SVQ3Context *svq3 = avctx->priv_data;

  1179.     H264Context *h    = &svq3->h;

  1180.     MpegEncContext *s = &h->s;

  1181. 

  1182.     ff_h264_free_context(h);

  1183. 

  1184.     ff_MPV_common_end(s);

  1185. 

  1186.     av_freep(&svq3->buf);

  1187.     svq3->buf_size = 0;

  1188. 

  1189.     return 0;

  1190. }

  1191. 

  1192. AVCodec ff_svq3_decoder = {

  1193.     .name           = "svq3",

  1194.     .type           = AVMEDIA_TYPE_VIDEO,

  1195.     .id             = AV_CODEC_ID_SVQ3,

  1196.     .priv_data_size = sizeof(SVQ3Context),

  1197.     .init           = svq3_decode_init,

  1198.     .close          = svq3_decode_end,

  1199.     .decode         = svq3_decode_frame,

  1200.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1201.                       CODEC_CAP_DR1             |

  1202.                       CODEC_CAP_DELAY,

  1203.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1204.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1205.                                                      AV_PIX_FMT_NONE},

  1206. };