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.
36514 Commits
32 Branches
325MB
Tree: 75d5156ac1
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '75d5156ac1'
${ noResults }
FFmpeg/libavcodec/svq3.c

1176 lines
42KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The Libav Project

  3.  *

  4.  * This file is part of Libav.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

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

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

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

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

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

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

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

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

  31.  *

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

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

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

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

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

  37.  *

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

  39.  * correctly decodes this file:

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

  41.  */

  42. #include "internal.h"

  43. #include "dsputil.h"

  44. #include "avcodec.h"

  45. #include "mpegvideo.h"

  46. #include "h264.h"

  47. 

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

  49. 

  50. #include "h264_mvpred.h"

  51. #include "golomb.h"

  52. #include "rectangle.h"

  53. #include "vdpau_internal.h"

  54. 

  55. #if CONFIG_ZLIB

  56. #include <zlib.h>

  57. #endif

  58. 

  59. #include "svq1.h"

  60. #include "svq3.h"

  61. 

  62. /**

  63.  * @file

  64.  * svq3 decoder.

  65.  */

  66. 

  67. typedef struct {

  68.     H264Context h;

  69.     int halfpel_flag;

  70.     int thirdpel_flag;

  71.     int unknown_flag;

  72.     int next_slice_index;

  73.     uint32_t watermark_key;

  74. } SVQ3Context;

  75. 

  76. #define FULLPEL_MODE  1

  77. #define HALFPEL_MODE  2

  78. #define THIRDPEL_MODE 3

  79. #define PREDICT_MODE  4

  80. 

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

  82.  * o-->o-->o   o

  83.  *         |  /|

  84.  * o   o   o / o

  85.  * | / |   |/  |

  86.  * o   o   o   o

  87.  *   /

  88.  * o-->o-->o-->o

  89.  */

  90. static const uint8_t svq3_scan[16] = {

  91.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  92.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

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

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

  95. };

  96. 

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

  98.     { 0, 0 },

  99.     { 1, 0 }, { 0, 1 },

  100.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  101.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  102.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

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

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

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

  106.     { 4, 4 }

  107. };

  108. 

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

  110.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

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

  112.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  4,  3 }, { 0, 1,  2,  4,  3 },

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

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

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

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

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

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

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

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

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

  122. };

  123. 

  124. static const struct {

  125.     uint8_t run;

  126.     uint8_t level;

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

  128.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  129.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

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

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

  132. };

  133. 

  134. static const uint32_t svq3_dequant_coeff[32] = {

  135.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  136.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  137.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  138.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  139. };

  140. 

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

  142. {

  143.     const int qmul = svq3_dequant_coeff[qp];

  144. #define stride 16

  145.     int i;

  146.     int temp[16];

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

  148. 

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

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

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

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

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

  154. 

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

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

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

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

  159.     }

  160. 

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

  162.         const int offset = x_offset[i];

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

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

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

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

  167. 

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

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

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

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

  172.     }

  173. }

  174. #undef stride

  175. 

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

  177.                         int stride, int qp, int dc)

  178. {

  179.     const int qmul = svq3_dequant_coeff[qp];

  180.     int i;

  181. 

  182.     if (dc) {

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

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

  185.         block[0] = 0;

  186.     }

  187. 

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

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

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

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

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

  193. 

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

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

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

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

  198.     }

  199. 

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

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

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

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

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

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

  206. 

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

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

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

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

  211.     }

  212. }

  213. 

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

  215.                                     int index, const int type)

  216. {

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

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

  219. 

  220.     int run, level, limit;

  221.     unsigned vlc;

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

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

  224. 

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

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

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

  228.             vlc      = vlc + 1 >> 1;

  229. 

  230.             if (type == 3) {

  231.                 if (vlc < 3) {

  232.                     run   = 0;

  233.                     level = vlc;

  234.                 } else if (vlc < 4) {

  235.                     run   = 1;

  236.                     level = 1;

  237.                 } else {

  238.                     run   = vlc & 0x3;

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

  240.                 }

  241.             } else {

  242.                 if (vlc < 16) {

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

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

  245.                 } else if (intra) {

  246.                     run   = vlc & 0x7;

  247.                     level = (vlc >> 3) +

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

  249.                 } else {

  250.                     run   = vlc & 0xF;

  251.                     level = (vlc >> 4) +

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

  253.                 }

  254.             }

  255. 

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

  257.                 return -1;

  258. 

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

  260.         }

  261. 

  262.         if (type != 2) {

  263.             break;

  264.         }

  265.     }

  266. 

  267.     return 0;

  268. }

  269. 

  270. static inline void svq3_mc_dir_part(MpegEncContext *s,

  271.                                     int x, int y, int width, int height,

  272.                                     int mx, int my, int dxy,

  273.                                     int thirdpel, int dir, int avg)

  274. {

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

  276.     uint8_t *src, *dest;

  277.     int i, emu = 0;

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

  279. 

  280.     mx += x;

  281.     my += y;

  282. 

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

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

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

  286.             emu = 1;

  287. 

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

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

  290.     }

  291. 

  292.     /* form component predictions */

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

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

  295. 

  296.     if (emu) {

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

  298.                                  width + 1, height + 1,

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

  300.         src = s->edge_emu_buffer;

  301.     }

  302.     if (thirdpel)

  303.         (avg ? s->dsp.avg_tpel_pixels_tab

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

  305.                                                 width, height);

  306.     else

  307.         (avg ? s->dsp.avg_pixels_tab

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

  309.                                                       height);

  310. 

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

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

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

  314.         width  = width  >> 1;

  315.         height = height >> 1;

  316.         blocksize++;

  317. 

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

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

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

  321. 

  322.             if (emu) {

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

  324.                                          width + 1, height + 1,

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

  326.                                          s->v_edge_pos >> 1);

  327.                 src = s->edge_emu_buffer;

  328.             }

  329.             if (thirdpel)

  330.                 (avg ? s->dsp.avg_tpel_pixels_tab

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

  332.                                                         s->uvlinesize,

  333.                                                         width, height);

  334.             else

  335.                 (avg ? s->dsp.avg_pixels_tab

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

  337.                                                               s->uvlinesize,

  338.                                                               height);

  339.         }

  340.     }

  341. }

  342. 

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

  344.                               int dir, int avg)

  345. {

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

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

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

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

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

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

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

  353. 

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

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

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

  357.                              (4 * s->mb_y + (i >> 2)) * h->b_stride;

  358.             int dxy;

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

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

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

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

  363. 

  364.             if (mode != PREDICT_MODE) {

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

  366.             } else {

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

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

  369. 

  370.                 if (dir == 0) {

  371.                     mx = mx * h->frame_num_offset /

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

  373.                     my = my * h->frame_num_offset /

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

  375.                 } else {

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

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

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

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

  380.                 }

  381.             }

  382. 

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

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

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

  386. 

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

  388.             if (mode == PREDICT_MODE) {

  389.                 dx = dy = 0;

  390.             } else {

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

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

  393. 

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

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

  396.                     return -1;

  397.                 }

  398.             }

  399. 

  400.             /* compute motion vector */

  401.             if (mode == THIRDPEL_MODE) {

  402.                 int fx, fy;

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

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

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

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

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

  408. 

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

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

  411.                 mx += mx;

  412.                 my += my;

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

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

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

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

  417. 

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

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

  420.                 mx *= 3;

  421.                 my *= 3;

  422.             } else {

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

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

  425. 

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

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

  428.                 mx *= 6;

  429.                 my *= 6;

  430.             }

  431. 

  432.             /* update mv_cache */

  433.             if (mode != PREDICT_MODE) {

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

  435. 

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

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

  438. 

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

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

  441.                 }

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

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

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

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

  446.             }

  447. 

  448.             /* write back motion vectors */

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

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

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

  452.         }

  453. 

  454.     return 0;

  455. }

  456. 

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

  458. {

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

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

  461.     int cbp = 0;

  462.     uint32_t vlc;

  463.     int8_t *top, *left;

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

  465.     const int mb_xy         = h->mb_xy;

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

  467. 

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

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

  470.     h->topright_samples_available = 0xFFFF;

  471. 

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

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

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

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

  476.                              0, 0, 0, 0, 0, 0);

  477. 

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

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

  480.                                  0, 0, 0, 0, 1, 1);

  481. 

  482.             mb_type = MB_TYPE_SKIP;

  483.         } else {

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

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

  486.                 return -1;

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

  488.                 return -1;

  489. 

  490.             mb_type = MB_TYPE_16x16;

  491.         }

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

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

  494.             mode = THIRDPEL_MODE;

  495.         else if (svq3->halfpel_flag &&

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

  497.             mode = HALFPEL_MODE;

  498.         else

  499.             mode = FULLPEL_MODE;

  500. 

  501.         /* fill caches */

  502.         /* note ref_cache should contain here:

  503.          *  ????????

  504.          *  ???11111

  505.          *  N??11111

  506.          *  N??11111

  507.          *  N??11111

  508.          */

  509. 

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

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

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

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

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

  515.             } else {

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

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

  518.             }

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

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

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

  522.                        4 * 2 * sizeof(int16_t));

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

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

  525. 

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

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

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

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

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

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

  532.                 } else

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

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

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

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

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

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

  539.                 } else

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

  541.             } else

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

  543.                        PART_NOT_AVAILABLE, 8);

  544. 

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

  546.                 break;

  547.         }

  548. 

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

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

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

  552.                 return -1;

  553.         } else {        /* AV_PICTURE_TYPE_B */

  554.             if (mb_type != 2)

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

  556.                     return -1;

  557.             else

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

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

  560.                            0, 4 * 2 * sizeof(int16_t));

  561.             if (mb_type != 1)

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

  563.                     return -1;

  564.             else

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

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

  567.                            0, 4 * 2 * sizeof(int16_t));

  568.         }

  569. 

  570.         mb_type = MB_TYPE_16x16;

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

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

  573. 

  574.         if (mb_type == 8) {

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

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

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

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

  579.                     h->left_samples_available = 0x5F5F;

  580.             }

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

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

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

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

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

  586. 

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

  588.                     h->top_samples_available = 0x33FF;

  589.             }

  590. 

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

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

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

  594. 

  595.                 if (vlc >= 25) {

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

  597.                     return -1;

  598.                 }

  599. 

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

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

  602. 

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

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

  605. 

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

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

  608.                     return -1;

  609.                 }

  610.             }

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

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

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

  614.         }

  615. 

  616.         write_back_intra_pred_mode(h);

  617. 

  618.         if (mb_type == 8) {

  619.             ff_h264_check_intra4x4_pred_mode(h);

  620. 

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

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

  623.         } else {

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

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

  626. 

  627.             h->top_samples_available  = 0x33FF;

  628.             h->left_samples_available = 0x5F5F;

  629.         }

  630. 

  631.         mb_type = MB_TYPE_INTRA4x4;

  632.     } else {                      /* INTRA16x16 */

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

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

  635. 

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

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

  638.             return -1;

  639.         }

  640. 

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

  642.         mb_type = MB_TYPE_INTRA16x16;

  643.     }

  644. 

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

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

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

  648.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  652.                        0, 4 * 2 * sizeof(int16_t));

  653.         }

  654.     }

  655.     if (!IS_INTRA4x4(mb_type)) {

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

  657.     }

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

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

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

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

  662.     }

  663. 

  664.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  668.             return -1;

  669.         }

  670. 

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

  672.                                 : golomb_to_inter_cbp[vlc];

  673.     }

  674.     if (IS_INTRA16x16(mb_type) ||

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

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

  677. 

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

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

  680.             return -1;

  681.         }

  682.     }

  683.     if (IS_INTRA16x16(mb_type)) {

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

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

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

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

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

  689.             return -1;

  690.         }

  691.     }

  692. 

  693.     if (cbp) {

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

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

  696. 

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

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

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

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

  701.                                  2 * (j & 2) + 4 * (i & 2))

  702.                               : (4 * i + j);

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

  704. 

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

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

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

  708.                         return -1;

  709.                     }

  710.                 }

  711.             }

  712. 

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

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

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

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

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

  718.                     return -1;

  719.                 }

  720. 

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

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

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

  724.                         k                                 = 16 * i + j;

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

  726. 

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

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

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

  730.                             return -1;

  731.                         }

  732.                     }

  733.                 }

  734.             }

  735.         }

  736.     }

  737. 

  738.     h->cbp                              = cbp;

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

  740. 

  741.     if (IS_INTRA(mb_type))

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

  743. 

  744.     return 0;

  745. }

  746. 

  747. static int svq3_decode_slice_header(AVCodecContext *avctx)

  748. {

  749.     SVQ3Context *svq3 = avctx->priv_data;

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

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

  752.     const int mb_xy   = h->mb_xy;

  753.     int i, header;

  754.     unsigned slice_id;

  755. 

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

  757. 

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

  759.         /* TODO: what? */

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

  761.         return -1;

  762.     } else {

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

  764. 

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

  766.                                  8 * show_bits(&s->gb, 8 * length) +

  767.                                  8 * length;

  768. 

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

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

  771.             return -1;

  772.         }

  773. 

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

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

  776. 

  777.         if (svq3->watermark_key) {

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

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

  780.                     header ^ svq3->watermark_key);

  781.         }

  782.         if (length > 0) {

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

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

  785.         }

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

  787.     }

  788. 

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

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

  791.         return -1;

  792.     }

  793. 

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

  795. 

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

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

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

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

  800.     } else {

  801.         skip_bits1(&s->gb);

  802.         s->mb_skip_run = 0;

  803.     }

  804. 

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

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

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

  808. 

  809.     /* unknown fields */

  810.     skip_bits1(&s->gb);

  811. 

  812.     if (svq3->unknown_flag)

  813.         skip_bits1(&s->gb);

  814. 

  815.     skip_bits1(&s->gb);

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

  817. 

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

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

  820. 

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

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

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

  824.                -1, 4 * sizeof(int8_t));

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

  826.                -1, 8 * sizeof(int8_t) * s->mb_x);

  827.     }

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

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

  830.                -1, 8 * sizeof(int8_t) * (s->mb_width - s->mb_x));

  831. 

  832.         if (s->mb_x > 0)

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

  834.     }

  835. 

  836.     return 0;

  837. }

  838. 

  839. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  840. {

  841.     SVQ3Context *svq3 = avctx->priv_data;

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

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

  844.     int m;

  845.     unsigned char *extradata;

  846.     unsigned char *extradata_end;

  847.     unsigned int size;

  848.     int marker_found = 0;

  849. 

  850.     if (ff_h264_decode_init(avctx) < 0)

  851.         return -1;

  852. 

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

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

  855.     s->unrestricted_mv = 1;

  856.     h->is_complex      = 1;

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

  858. 

  859.     if (!s->context_initialized) {

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

  861. 

  862.         svq3->halfpel_flag  = 1;

  863.         svq3->thirdpel_flag = 1;

  864.         svq3->unknown_flag  = 0;

  865. 

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

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

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

  869.         if (extradata) {

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

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

  872.                     marker_found = 1;

  873.                     break;

  874.                 }

  875.                 extradata++;

  876.             }

  877.         }

  878. 

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

  880.         if (marker_found) {

  881.             GetBitContext gb;

  882.             int frame_size_code;

  883. 

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

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

  886.                 return AVERROR_INVALIDDATA;

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

  888. 

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

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

  891.             switch (frame_size_code) {

  892.             case 0:

  893.                 avctx->width  = 160;

  894.                 avctx->height = 120;

  895.                 break;

  896.             case 1:

  897.                 avctx->width  = 128;

  898.                 avctx->height =  96;

  899.                 break;

  900.             case 2:

  901.                 avctx->width  = 176;

  902.                 avctx->height = 144;

  903.                 break;

  904.             case 3:

  905.                 avctx->width  = 352;

  906.                 avctx->height = 288;

  907.                 break;

  908.             case 4:

  909.                 avctx->width  = 704;

  910.                 avctx->height = 576;

  911.                 break;

  912.             case 5:

  913.                 avctx->width  = 240;

  914.                 avctx->height = 180;

  915.                 break;

  916.             case 6:

  917.                 avctx->width  = 320;

  918.                 avctx->height = 240;

  919.                 break;

  920.             case 7:

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

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

  923.                 break;

  924.             }

  925. 

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

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

  928. 

  929.             /* unknown fields */

  930.             skip_bits1(&gb);

  931.             skip_bits1(&gb);

  932.             skip_bits1(&gb);

  933.             skip_bits1(&gb);

  934. 

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

  936. 

  937.             /* unknown field */

  938.             skip_bits1(&gb);

  939. 

  940.             while (get_bits1(&gb))

  941.                 skip_bits(&gb, 8);

  942. 

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

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

  945.             if (svq3->unknown_flag) {

  946. #if CONFIG_ZLIB

  947.                 unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  948.                 unsigned watermark_height = svq3_get_ue_golomb(&gb);

  949.                 int u1                    = svq3_get_ue_golomb(&gb);

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

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

  952.                 int u4                    = svq3_get_ue_golomb(&gb);

  953.                 unsigned long buf_len     = watermark_width *

  954.                                             watermark_height * 4;

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

  956.                 uint8_t *buf;

  957. 

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

  959.                     return -1;

  960. 

  961.                 buf = av_malloc(buf_len);

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

  963.                        watermark_width, watermark_height);

  964.                 av_log(avctx, AV_LOG_DEBUG,

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

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

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

  968.                                size - offset) != Z_OK) {

  969.                     av_log(avctx, AV_LOG_ERROR,

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

  971.                     av_free(buf);

  972.                     return -1;

  973.                 }

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

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

  976.                                       svq3->watermark_key;

  977.                 av_log(avctx, AV_LOG_DEBUG,

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

  979.                 av_free(buf);

  980. #else

  981.                 av_log(avctx, AV_LOG_ERROR,

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

  983.                 return -1;

  984. #endif

  985.             }

  986.         }

  987. 

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

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

  990. 

  991.         if (ff_MPV_common_init(s) < 0)

  992.             return -1;

  993. 

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

  995. 

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

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

  998.             return AVERROR(ENOMEM);

  999.         }

  1000.     }

  1001. 

  1002.     return 0;

  1003. }

  1004. 

  1005. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1006.                              int *got_frame, AVPacket *avpkt)

  1007. {

  1008.     const uint8_t *buf = avpkt->data;

  1009.     SVQ3Context *svq3  = avctx->priv_data;

  1010.     H264Context *h     = &svq3->h;

  1011.     MpegEncContext *s  = &h->s;

  1012.     int buf_size       = avpkt->size;

  1013.     int m;

  1014. 

  1015.     /* special case for last picture */

  1016.     if (buf_size == 0) {

  1017.         if (s->next_picture_ptr && !s->low_delay) {

  1018.             *(AVFrame *) data   = s->next_picture.f;

  1019.             s->next_picture_ptr = NULL;

  1020.             *got_frame          = 1;

  1021.         }

  1022.         return 0;

  1023.     }

  1024. 

  1025.     init_get_bits(&s->gb, buf, 8 * buf_size);

  1026. 

  1027.     s->mb_x = s->mb_y = h->mb_xy = 0;

  1028. 

  1029.     if (svq3_decode_slice_header(avctx))

  1030.         return -1;

  1031. 

  1032.     s->pict_type      = h->slice_type;

  1033.     s->picture_number = h->slice_num;

  1034. 

  1035.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1036.         av_log(h->s.avctx, AV_LOG_DEBUG,

  1037.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1038.                av_get_picture_type_char(s->pict_type),

  1039.                svq3->halfpel_flag, svq3->thirdpel_flag,

  1040.                s->adaptive_quant, s->qscale, h->slice_num);

  1041. 

  1042.     /* for skipping the frame */

  1043.     s->current_picture.f.pict_type = s->pict_type;

  1044.     s->current_picture.f.key_frame = (s->pict_type == AV_PICTURE_TYPE_I);

  1045. 

  1046.     /* Skip B-frames if we do not have reference frames. */

  1047.     if (s->last_picture_ptr == NULL && s->pict_type == AV_PICTURE_TYPE_B)

  1048.         return 0;

  1049.     if (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B ||

  1050.         avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I ||

  1051.         avctx->skip_frame >= AVDISCARD_ALL)

  1052.         return 0;

  1053. 

  1054.     if (s->next_p_frame_damaged) {

  1055.         if (s->pict_type == AV_PICTURE_TYPE_B)

  1056.             return 0;

  1057.         else

  1058.             s->next_p_frame_damaged = 0;

  1059.     }

  1060. 

  1061.     if (ff_h264_frame_start(h) < 0)

  1062.         return -1;

  1063. 

  1064.     if (s->pict_type == AV_PICTURE_TYPE_B) {

  1065.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1066. 

  1067.         if (h->frame_num_offset < 0)

  1068.             h->frame_num_offset += 256;

  1069.         if (h->frame_num_offset == 0 ||

  1070.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1071.             av_log(h->s.avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1072.             return -1;

  1073.         }

  1074.     } else {

  1075.         h->prev_frame_num        = h->frame_num;

  1076.         h->frame_num             = h->slice_num;

  1077.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1078. 

  1079.         if (h->prev_frame_num_offset < 0)

  1080.             h->prev_frame_num_offset += 256;

  1081.     }

  1082. 

  1083.     for (m = 0; m < 2; m++) {

  1084.         int i;

  1085.         for (i = 0; i < 4; i++) {

  1086.             int j;

  1087.             for (j = -1; j < 4; j++)

  1088.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1089.             if (i < 3)

  1090.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1091.         }

  1092.     }

  1093. 

  1094.     for (s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {

  1095.         for (s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {

  1096.             unsigned mb_type;

  1097.             h->mb_xy = s->mb_x + s->mb_y * s->mb_stride;

  1098. 

  1099.             if ((get_bits_count(&s->gb) + 7) >= s->gb.size_in_bits &&

  1100.                 ((get_bits_count(&s->gb) & 7) == 0 ||

  1101.                  show_bits(&s->gb, -get_bits_count(&s->gb) & 7) == 0)) {

  1102.                 skip_bits(&s->gb, svq3->next_slice_index - get_bits_count(&s->gb));

  1103.                 s->gb.size_in_bits = 8 * buf_size;

  1104. 

  1105.                 if (svq3_decode_slice_header(avctx))

  1106.                     return -1;

  1107. 

  1108.                 /* TODO: support s->mb_skip_run */

  1109.             }

  1110. 

  1111.             mb_type = svq3_get_ue_golomb(&s->gb);

  1112. 

  1113.             if (s->pict_type == AV_PICTURE_TYPE_I)

  1114.                 mb_type += 8;

  1115.             else if (s->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1116.                 mb_type += 4;

  1117.             if (mb_type > 33 || svq3_decode_mb(svq3, mb_type)) {

  1118.                 av_log(h->s.avctx, AV_LOG_ERROR,

  1119.                        "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  1120.                 return -1;

  1121.             }

  1122. 

  1123.             if (mb_type != 0)

  1124.                 ff_h264_hl_decode_mb(h);

  1125. 

  1126.             if (s->pict_type != AV_PICTURE_TYPE_B && !s->low_delay)

  1127.                 s->current_picture.f.mb_type[s->mb_x + s->mb_y * s->mb_stride] =

  1128.                     (s->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1129.         }

  1130. 

  1131.         ff_draw_horiz_band(s, 16 * s->mb_y, 16);

  1132.     }

  1133. 

  1134.     ff_MPV_frame_end(s);

  1135. 

  1136.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay)

  1137.         *(AVFrame *)data = s->current_picture.f;

  1138.     else

  1139.         *(AVFrame *)data = s->last_picture.f;

  1140. 

  1141.     /* Do not output the last pic after seeking. */

  1142.     if (s->last_picture_ptr || s->low_delay)

  1143.         *got_frame = 1;

  1144. 

  1145.     return buf_size;

  1146. }

  1147. 

  1148. static int svq3_decode_end(AVCodecContext *avctx)

  1149. {

  1150.     SVQ3Context *svq3 = avctx->priv_data;

  1151.     H264Context *h    = &svq3->h;

  1152.     MpegEncContext *s = &h->s;

  1153. 

  1154.     ff_h264_free_context(h);

  1155. 

  1156.     ff_MPV_common_end(s);

  1157. 

  1158.     return 0;

  1159. }

  1160. 

  1161. AVCodec ff_svq3_decoder = {

  1162.     .name           = "svq3",

  1163.     .type           = AVMEDIA_TYPE_VIDEO,

  1164.     .id             = AV_CODEC_ID_SVQ3,

  1165.     .priv_data_size = sizeof(SVQ3Context),

  1166.     .init           = svq3_decode_init,

  1167.     .close          = svq3_decode_end,

  1168.     .decode         = svq3_decode_frame,

  1169.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1170.                       CODEC_CAP_DR1             |

  1171.                       CODEC_CAP_DELAY,

  1172.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1173.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1174.                                                      AV_PIX_FMT_NONE},

  1175. };