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.
50910 Commits
32 Branches
836MB
Tree: ed69c69a01
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ed69c69a01'
${ noResults }
FFmpeg/libavcodec/svq3.c

1351 lines
48KB
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 "avcodec.h"

  44. #include "mpegvideo.h"

  45. #include "h264.h"

  46. 

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

  48. 

  49. #include "h264_mvpred.h"

  50. #include "golomb.h"

  51. #include "hpeldsp.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.     HpelDSPContext hdsp;

  70.     Picture *cur_pic;

  71.     Picture *next_pic;

  72.     Picture *last_pic;

  73.     int halfpel_flag;

  74.     int thirdpel_flag;

  75.     int unknown_flag;

  76.     int next_slice_index;

  77.     uint32_t watermark_key;

  78.     uint8_t *buf;

  79.     int buf_size;

  80.     int adaptive_quant;

  81.     int next_p_frame_damaged;

  82.     int h_edge_pos;

  83.     int v_edge_pos;

  84.     int last_frame_output;

  85. } SVQ3Context;

  86. 

  87. #define FULLPEL_MODE  1

  88. #define HALFPEL_MODE  2

  89. #define THIRDPEL_MODE 3

  90. #define PREDICT_MODE  4

  91. 

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

  93.  * o-->o-->o   o

  94.  *         |  /|

  95.  * o   o   o / o

  96.  * | / |   |/  |

  97.  * o   o   o   o

  98.  *   /

  99.  * o-->o-->o-->o

  100.  */

  101. static const uint8_t svq3_scan[16] = {

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

  103.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

  104.     0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4,

  105.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  106. };

  107. 

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

  109.     { 0, 0 },

  110.     { 1, 0 }, { 0, 1 },

  111.     { 0, 2 }, { 1, 1 }, { 2, 0 },

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

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

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

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

  116.     { 4, 3 }, { 3, 4 },

  117.     { 4, 4 }

  118. };

  119. 

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

  121.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  122.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

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

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

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

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

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

  133. };

  134. 

  135. static const struct {

  136.     uint8_t run;

  137.     uint8_t level;

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

  139.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  140.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  141.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  142.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  143. };

  144. 

  145. static const uint32_t svq3_dequant_coeff[32] = {

  146.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  147.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  148.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  149.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  150. };

  151. 

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

  153. {

  154.     const int qmul = svq3_dequant_coeff[qp];

  155. #define stride 16

  156.     int i;

  157.     int temp[16];

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

  159. 

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

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

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

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

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

  165. 

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

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

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

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

  170.     }

  171. 

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

  173.         const int offset = x_offset[i];

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

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

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

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

  178. 

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

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

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

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

  183.     }

  184. }

  185. #undef stride

  186. 

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

  188.                         int stride, int qp, int dc)

  189. {

  190.     const int qmul = svq3_dequant_coeff[qp];

  191.     int i;

  192. 

  193.     if (dc) {

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

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

  196.         block[0] = 0;

  197.     }

  198. 

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

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

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

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

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

  204. 

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

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

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

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

  209.     }

  210. 

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

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

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

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

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

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

  217. 

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

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

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

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

  222.     }

  223. 

  224.     memset(block, 0, 16 * sizeof(int16_t));

  225. }

  226. 

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

  228.                                     int index, const int type)

  229. {

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

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

  232. 

  233.     int run, level, sign, limit;

  234.     unsigned vlc;

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

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

  237. 

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

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

  240.             if ((int32_t)vlc < 0)

  241.                 return -1;

  242. 

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

  244.             vlc      = vlc + 1 >> 1;

  245. 

  246.             if (type == 3) {

  247.                 if (vlc < 3) {

  248.                     run   = 0;

  249.                     level = vlc;

  250.                 } else if (vlc < 4) {

  251.                     run   = 1;

  252.                     level = 1;

  253.                 } else {

  254.                     run   = vlc & 0x3;

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

  256.                 }

  257.             } else {

  258.                 if (vlc < 16U) {

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

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

  261.                 } else if (intra) {

  262.                     run   = vlc & 0x7;

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

  264.                 } else {

  265.                     run   = vlc & 0xF;

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

  267.                 }

  268.             }

  269. 

  270. 

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

  272.                 return -1;

  273. 

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

  275.         }

  276. 

  277.         if (type != 2) {

  278.             break;

  279.         }

  280.     }

  281. 

  282.     return 0;

  283. }

  284. 

  285. static inline void svq3_mc_dir_part(SVQ3Context *s,

  286.                                     int x, int y, int width, int height,

  287.                                     int mx, int my, int dxy,

  288.                                     int thirdpel, int dir, int avg)

  289. {

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

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

  292.     uint8_t *src, *dest;

  293.     int i, emu = 0;

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

  295. 

  296.     mx += x;

  297.     my += y;

  298. 

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

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

  301.         emu = 1;

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

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

  304.     }

  305. 

  306.     /* form component predictions */

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

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

  309. 

  310.     if (emu) {

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

  312.                                  width + 1, height + 1,

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

  314.         src = h->edge_emu_buffer;

  315.     }

  316.     if (thirdpel)

  317.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  319.                                                 width, height);

  320.     else

  321.         (avg ? s->hdsp.avg_pixels_tab

  322.              : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src, h->linesize,

  323.                                                        height);

  324. 

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

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

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

  328.         width  = width  >> 1;

  329.         height = height >> 1;

  330.         blocksize++;

  331. 

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

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

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

  335. 

  336.             if (emu) {

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

  338.                                          width + 1, height + 1,

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

  340.                                          s->v_edge_pos >> 1);

  341.                 src = h->edge_emu_buffer;

  342.             }

  343.             if (thirdpel)

  344.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  346.                                                         h->uvlinesize,

  347.                                                         width, height);

  348.             else

  349.                 (avg ? s->hdsp.avg_pixels_tab

  350.                      : s->hdsp.put_pixels_tab)[blocksize][dxy](dest, src,

  351.                                                                h->uvlinesize,

  352.                                                                height);

  353.         }

  354.     }

  355. }

  356. 

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

  358.                               int dir, int avg)

  359. {

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

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

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

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

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

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

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

  367. 

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

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

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

  371.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  372.             int dxy;

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

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

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

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

  377. 

  378.             if (mode != PREDICT_MODE) {

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

  380.             } else {

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

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

  383. 

  384.                 if (dir == 0) {

  385.                     mx = mx * h->frame_num_offset /

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

  387.                     my = my * h->frame_num_offset /

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

  389.                 } else {

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

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

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

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

  394.                 }

  395.             }

  396. 

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

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

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

  400. 

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

  402.             if (mode == PREDICT_MODE) {

  403.                 dx = dy = 0;

  404.             } else {

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

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

  407. 

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

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

  410.                     return -1;

  411.                 }

  412.             }

  413. 

  414.             /* compute motion vector */

  415.             if (mode == THIRDPEL_MODE) {

  416.                 int fx, fy;

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

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

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

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

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

  422. 

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

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

  425.                 mx += mx;

  426.                 my += my;

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

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

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

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

  431. 

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

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

  434.                 mx *= 3;

  435.                 my *= 3;

  436.             } else {

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

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

  439. 

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

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

  442.                 mx *= 6;

  443.                 my *= 6;

  444.             }

  445. 

  446.             /* update mv_cache */

  447.             if (mode != PREDICT_MODE) {

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

  449. 

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

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

  452. 

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

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

  455.                 }

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

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

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

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

  460.             }

  461. 

  462.             /* write back motion vectors */

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

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

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

  466.         }

  467. 

  468.     return 0;

  469. }

  470. 

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

  472. {

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

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

  475.     int cbp = 0;

  476.     uint32_t vlc;

  477.     int8_t *top, *left;

  478.     const int mb_xy         = h->mb_xy;

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

  480. 

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

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

  483.     h->topright_samples_available = 0xFFFF;

  484. 

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

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

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

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

  489.                              0, 0, 0, 0, 0, 0);

  490. 

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

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

  493.                                  0, 0, 0, 0, 1, 1);

  494. 

  495.             mb_type = MB_TYPE_SKIP;

  496.         } else {

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

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

  499.                 return -1;

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

  501.                 return -1;

  502. 

  503.             mb_type = MB_TYPE_16x16;

  504.         }

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

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

  507.             mode = THIRDPEL_MODE;

  508.         else if (s->halfpel_flag &&

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

  510.             mode = HALFPEL_MODE;

  511.         else

  512.             mode = FULLPEL_MODE;

  513. 

  514.         /* fill caches */

  515.         /* note ref_cache should contain here:

  516.          *  ????????

  517.          *  ???11111

  518.          *  N??11111

  519.          *  N??11111

  520.          *  N??11111

  521.          */

  522. 

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

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

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

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

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

  528.             } else {

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

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

  531.             }

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

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

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

  535.                        4 * 2 * sizeof(int16_t));

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

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

  538. 

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

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

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

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

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

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

  545.                 } else

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

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

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

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

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

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

  552.                 } else

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

  554.             } else

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

  556.                        PART_NOT_AVAILABLE, 8);

  557. 

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

  559.                 break;

  560.         }

  561. 

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

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

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

  565.                 return -1;

  566.         } else {        /* AV_PICTURE_TYPE_B */

  567.             if (mb_type != 2) {

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

  569.                     return -1;

  570.             } else {

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

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

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

  574.             }

  575.             if (mb_type != 1) {

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

  577.                     return -1;

  578.             } else {

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

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

  581.                            0, 4 * 2 * sizeof(int16_t));

  582.             }

  583.         }

  584. 

  585.         mb_type = MB_TYPE_16x16;

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

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

  588. 

  589.         if (mb_type == 8) {

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

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

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

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

  594.                     h->left_samples_available = 0x5F5F;

  595.             }

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

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

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

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

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

  601. 

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

  603.                     h->top_samples_available = 0x33FF;

  604.             }

  605. 

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

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

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

  609. 

  610.                 if (vlc >= 25U) {

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

  612.                     return -1;

  613.                 }

  614. 

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

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

  617. 

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

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

  620. 

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

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

  623.                     return -1;

  624.                 }

  625.             }

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

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

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

  629.         }

  630. 

  631.         write_back_intra_pred_mode(h);

  632. 

  633.         if (mb_type == 8) {

  634.             ff_h264_check_intra4x4_pred_mode(h);

  635. 

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

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

  638.         } else {

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

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

  641. 

  642.             h->top_samples_available  = 0x33FF;

  643.             h->left_samples_available = 0x5F5F;

  644.         }

  645. 

  646.         mb_type = MB_TYPE_INTRA4x4;

  647.     } else {                      /* INTRA16x16 */

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

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

  650. 

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

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

  653.             return -1;

  654.         }

  655. 

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

  657.         mb_type = MB_TYPE_INTRA16x16;

  658.     }

  659. 

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

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

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

  663.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  667.                        0, 4 * 2 * sizeof(int16_t));

  668.         }

  669.     }

  670.     if (!IS_INTRA4x4(mb_type)) {

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

  672.     }

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

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

  675.     }

  676. 

  677.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  681.             return -1;

  682.         }

  683. 

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

  685.                                 : golomb_to_inter_cbp[vlc];

  686.     }

  687.     if (IS_INTRA16x16(mb_type) ||

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

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

  690. 

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

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

  693.             return -1;

  694.         }

  695.     }

  696.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  700.             av_log(h->avctx, AV_LOG_ERROR,

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

  702.             return -1;

  703.         }

  704.     }

  705. 

  706.     if (cbp) {

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

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

  709. 

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

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

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

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

  714.                                  2 * (j & 2) + 4 * (i & 2))

  715.                               : (4 * i + j);

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

  717. 

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

  719.                         av_log(h->avctx, AV_LOG_ERROR,

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

  721.                         return -1;

  722.                     }

  723.                 }

  724.             }

  725. 

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

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

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

  729.                     av_log(h->avctx, AV_LOG_ERROR,

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

  731.                     return -1;

  732.                 }

  733. 

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

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

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

  737.                         k                                 = 16 * i + j;

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

  739. 

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

  741.                             av_log(h->avctx, AV_LOG_ERROR,

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

  743.                             return -1;

  744.                         }

  745.                     }

  746.                 }

  747.             }

  748.         }

  749.     }

  750. 

  751.     h->cbp                              = cbp;

  752.     h->cur_pic.mb_type[mb_xy] = mb_type;

  753. 

  754.     if (IS_INTRA(mb_type))

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

  756. 

  757.     return 0;

  758. }

  759. 

  760. static int svq3_decode_slice_header(AVCodecContext *avctx)

  761. {

  762.     SVQ3Context *s = avctx->priv_data;

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

  764.     const int mb_xy   = h->mb_xy;

  765.     int i, header;

  766.     unsigned slice_id;

  767. 

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

  769. 

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

  771.         /* TODO: what? */

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

  773.         return -1;

  774.     } else {

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

  776. 

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

  778.                               8 * show_bits(&h->gb, 8 * length) +

  779.                               8 * length;

  780. 

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

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

  783.             return -1;

  784.         }

  785. 

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

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

  788. 

  789.         if (s->watermark_key) {

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

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

  792.                     header ^ s->watermark_key);

  793.         }

  794.         if (length > 0) {

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

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

  797.         }

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

  799.     }

  800. 

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

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

  803.         return -1;

  804.     }

  805. 

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

  807. 

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

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

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

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

  812.     } else {

  813.         skip_bits1(&h->gb);

  814.         h->mb_skip_run = 0;

  815.     }

  816. 

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

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

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

  820. 

  821.     /* unknown fields */

  822.     skip_bits1(&h->gb);

  823. 

  824.     if (s->unknown_flag)

  825.         skip_bits1(&h->gb);

  826. 

  827.     skip_bits1(&h->gb);

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

  829. 

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

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

  832. 

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

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

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

  836.                -1, 4 * sizeof(int8_t));

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

  838.                -1, 8 * sizeof(int8_t) * h->mb_x);

  839.     }

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

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

  842.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  843. 

  844.         if (h->mb_x > 0)

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

  846.     }

  847. 

  848.     return 0;

  849. }

  850. 

  851. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  852. {

  853.     SVQ3Context *s = avctx->priv_data;

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

  855.     int m;

  856.     unsigned char *extradata;

  857.     unsigned char *extradata_end;

  858.     unsigned int size;

  859.     int marker_found = 0;

  860. 

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

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

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

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

  865.         av_freep(&s->cur_pic);

  866.         av_freep(&s->last_pic);

  867.         av_freep(&s->next_pic);

  868.         return AVERROR(ENOMEM);

  869.     }

  870. 

  871.     if (ff_h264_decode_init(avctx) < 0)

  872.         return -1;

  873. 

  874.     ff_hpeldsp_init(&s->hdsp, avctx->flags);

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

  876.     h->is_complex      = 1;

  877.     h->sps.chroma_format_idc = 1;

  878.     h->picture_structure = PICT_FRAME;

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

  880. 

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

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

  883. 

  884.     s->halfpel_flag  = 1;

  885.     s->thirdpel_flag = 1;

  886.     s->unknown_flag  = 0;

  887. 

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

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

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

  891.     if (extradata) {

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

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

  894.                 marker_found = 1;

  895.                 break;

  896.             }

  897.             extradata++;

  898.         }

  899.     }

  900. 

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

  902.     if (marker_found) {

  903.         GetBitContext gb;

  904.         int frame_size_code;

  905. 

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

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

  908.             return AVERROR_INVALIDDATA;

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

  910. 

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

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

  913.         switch (frame_size_code) {

  914.         case 0:

  915.             avctx->width  = 160;

  916.             avctx->height = 120;

  917.             break;

  918.         case 1:

  919.             avctx->width  = 128;

  920.             avctx->height =  96;

  921.             break;

  922.         case 2:

  923.             avctx->width  = 176;

  924.             avctx->height = 144;

  925.             break;

  926.         case 3:

  927.             avctx->width  = 352;

  928.             avctx->height = 288;

  929.             break;

  930.         case 4:

  931.             avctx->width  = 704;

  932.             avctx->height = 576;

  933.             break;

  934.         case 5:

  935.             avctx->width  = 240;

  936.             avctx->height = 180;

  937.             break;

  938.         case 6:

  939.             avctx->width  = 320;

  940.             avctx->height = 240;

  941.             break;

  942.         case 7:

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

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

  945.             break;

  946.         }

  947. 

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

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

  950. 

  951.         /* unknown fields */

  952.         skip_bits1(&gb);

  953.         skip_bits1(&gb);

  954.         skip_bits1(&gb);

  955.         skip_bits1(&gb);

  956. 

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

  958. 

  959.         /* unknown field */

  960.         skip_bits1(&gb);

  961. 

  962.         while (get_bits1(&gb))

  963.             skip_bits(&gb, 8);

  964. 

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

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

  967.         if (s->unknown_flag) {

  968. #if CONFIG_ZLIB

  969.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  970.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  971.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  974.             int u4                    = svq3_get_ue_golomb(&gb);

  975.             unsigned long buf_len     = watermark_width *

  976.                                         watermark_height * 4;

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

  978.             uint8_t *buf;

  979. 

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

  981.                 return -1;

  982. 

  983.             buf = av_malloc(buf_len);

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

  985.                    watermark_width, watermark_height);

  986.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  990.                            size - offset) != Z_OK) {

  991.                 av_log(avctx, AV_LOG_ERROR,

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

  993.                 av_free(buf);

  994.                 return -1;

  995.             }

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

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

  998.             av_log(avctx, AV_LOG_DEBUG,

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

  1000.             av_free(buf);

  1001. #else

  1002.             av_log(avctx, AV_LOG_ERROR,

  1003.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1004.             return -1;

  1005. #endif

  1006.         }

  1007.     }

  1008. 

  1009.     h->width  = avctx->width;

  1010.     h->height = avctx->height;

  1011.     h->mb_width  = (h->width + 15) / 16;

  1012.     h->mb_height = (h->height + 15) / 16;

  1013.     h->mb_stride = h->mb_width + 1;

  1014.     h->mb_num    = h->mb_width * h->mb_height;

  1015.     h->b_stride = 4 * h->mb_width;

  1016.     s->h_edge_pos = h->mb_width * 16;

  1017.     s->v_edge_pos = h->mb_height * 16;

  1018. 

  1019.     if (ff_h264_alloc_tables(h) < 0) {

  1020.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1021.         return AVERROR(ENOMEM);

  1022.     }

  1023. 

  1024.     return 0;

  1025. }

  1026. 

  1027. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1028. {

  1029.     int i;

  1030.     for (i = 0; i < 2; i++) {

  1031.         av_buffer_unref(&pic->motion_val_buf[i]);

  1032.         av_buffer_unref(&pic->ref_index_buf[i]);

  1033.     }

  1034.     av_buffer_unref(&pic->mb_type_buf);

  1035. 

  1036.     av_frame_unref(&pic->f);

  1037. }

  1038. 

  1039. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1040. {

  1041.     SVQ3Context *s = avctx->priv_data;

  1042.     H264Context *h = &s->h;

  1043.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1044.     const int mb_array_size = h->mb_stride * h->mb_height;

  1045.     const int b4_stride     = h->mb_width * 4 + 1;

  1046.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1047.     int ret;

  1048. 

  1049.     if (!pic->motion_val_buf[0]) {

  1050.         int i;

  1051. 

  1052.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1053.         if (!pic->mb_type_buf)

  1054.             return AVERROR(ENOMEM);

  1055.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1056. 

  1057.         for (i = 0; i < 2; i++) {

  1058.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1059.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1060.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1061.                 ret = AVERROR(ENOMEM);

  1062.                 goto fail;

  1063.             }

  1064. 

  1065.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1066.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1067.         }

  1068.     }

  1069.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1070. 

  1071.     ret = ff_get_buffer(avctx, &pic->f,

  1072.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1073.     if (ret < 0)

  1074.         goto fail;

  1075. 

  1076.     if (!h->edge_emu_buffer) {

  1077.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1078.         if (!h->edge_emu_buffer)

  1079.             return AVERROR(ENOMEM);

  1080.     }

  1081. 

  1082.     h->linesize   = pic->f.linesize[0];

  1083.     h->uvlinesize = pic->f.linesize[1];

  1084. 

  1085.     return 0;

  1086. fail:

  1087.     free_picture(avctx, pic);

  1088.     return ret;

  1089. }

  1090. 

  1091. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1092.                              int *got_frame, AVPacket *avpkt)

  1093. {

  1094.     SVQ3Context *s     = avctx->priv_data;

  1095.     H264Context *h     = &s->h;

  1096.     int buf_size       = avpkt->size;

  1097.     int left;

  1098.     uint8_t *buf;

  1099.     int ret, m, i;

  1100. 

  1101.     /* special case for last picture */

  1102.     if (buf_size == 0) {

  1103.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1104.             ret = av_frame_ref(data, &s->next_pic->f);

  1105.             if (ret < 0)

  1106.                 return ret;

  1107.             s->last_frame_output = 1;

  1108.             *got_frame          = 1;

  1109.         }

  1110.         return 0;

  1111.     }

  1112. 

  1113.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1114. 

  1115.     if (s->watermark_key) {

  1116.         av_fast_malloc(&s->buf, &s->buf_size,

  1117.                        buf_size+FF_INPUT_BUFFER_PADDING_SIZE);

  1118.         if (!s->buf)

  1119.             return AVERROR(ENOMEM);

  1120.         memcpy(s->buf, avpkt->data, buf_size);

  1121.         buf = s->buf;

  1122.     } else {

  1123.         buf = avpkt->data;

  1124.     }

  1125. 

  1126.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1127. 

  1128.     if (svq3_decode_slice_header(avctx))

  1129.         return -1;

  1130. 

  1131.     h->pict_type = h->slice_type;

  1132. 

  1133.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1134.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1135. 

  1136.     av_frame_unref(&s->cur_pic->f);

  1137. 

  1138.     /* for skipping the frame */

  1139.     s->cur_pic->f.pict_type = h->pict_type;

  1140.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1141. 

  1142.     ret = get_buffer(avctx, s->cur_pic);

  1143.     if (ret < 0)

  1144.         return ret;

  1145. 

  1146.     h->cur_pic_ptr = s->cur_pic;

  1147.     av_frame_unref(&h->cur_pic.f);

  1148.     h->cur_pic     = *s->cur_pic;

  1149.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1150.     if (ret < 0)

  1151.         return ret;

  1152. 

  1153.     for (i = 0; i < 16; i++) {

  1154.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1155.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1156.     }

  1157.     for (i = 0; i < 16; i++) {

  1158.         h->block_offset[16 + i]      =

  1159.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1160.         h->block_offset[48 + 16 + i] =

  1161.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1162.     }

  1163. 

  1164.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1165.         if (!s->last_pic->f.data[0]) {

  1166.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1167.             ret = get_buffer(avctx, s->last_pic);

  1168.             if (ret < 0)

  1169.                 return ret;

  1170.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1171.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1172.                    s->last_pic->f.linesize[1]);

  1173.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1174.                    s->last_pic->f.linesize[2]);

  1175.         }

  1176. 

  1177.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1178.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1179.             ret = get_buffer(avctx, s->next_pic);

  1180.             if (ret < 0)

  1181.                 return ret;

  1182.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1183.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1184.                    s->next_pic->f.linesize[1]);

  1185.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1186.                    s->next_pic->f.linesize[2]);

  1187.         }

  1188.     }

  1189. 

  1190.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1191.         av_log(h->avctx, AV_LOG_DEBUG,

  1192.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1193.                av_get_picture_type_char(h->pict_type),

  1194.                s->halfpel_flag, s->thirdpel_flag,

  1195.                s->adaptive_quant, h->qscale, h->slice_num);

  1196. 

  1197.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1198.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1199.         avctx->skip_frame >= AVDISCARD_ALL)

  1200.         return 0;

  1201. 

  1202.     if (s->next_p_frame_damaged) {

  1203.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1204.             return 0;

  1205.         else

  1206.             s->next_p_frame_damaged = 0;

  1207.     }

  1208. 

  1209.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1210.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1211. 

  1212.         if (h->frame_num_offset < 0)

  1213.             h->frame_num_offset += 256;

  1214.         if (h->frame_num_offset == 0 ||

  1215.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1216.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1217.             return -1;

  1218.         }

  1219.     } else {

  1220.         h->prev_frame_num        = h->frame_num;

  1221.         h->frame_num             = h->slice_num;

  1222.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1223. 

  1224.         if (h->prev_frame_num_offset < 0)

  1225.             h->prev_frame_num_offset += 256;

  1226.     }

  1227. 

  1228.     for (m = 0; m < 2; m++) {

  1229.         int i;

  1230.         for (i = 0; i < 4; i++) {

  1231.             int j;

  1232.             for (j = -1; j < 4; j++)

  1233.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1234.             if (i < 3)

  1235.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1236.         }

  1237.     }

  1238. 

  1239.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1240.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1241.             unsigned mb_type;

  1242.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1243. 

  1244.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1245.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1246.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1247.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1248.                 h->gb.size_in_bits = 8 * buf_size;

  1249. 

  1250.                 if (svq3_decode_slice_header(avctx))

  1251.                     return -1;

  1252. 

  1253.                 /* TODO: support s->mb_skip_run */

  1254.             }

  1255. 

  1256.             mb_type = svq3_get_ue_golomb(&h->gb);

  1257. 

  1258.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1259.                 mb_type += 8;

  1260.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1261.                 mb_type += 4;

  1262.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1263.                 av_log(h->avctx, AV_LOG_ERROR,

  1264.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1265.                 return -1;

  1266.             }

  1267. 

  1268.             if (mb_type != 0 || h->cbp)

  1269.                 ff_h264_hl_decode_mb(h);

  1270. 

  1271.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1272.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1273.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1274.         }

  1275. 

  1276.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1277.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1278.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1279.     }

  1280. 

  1281.     left = buf_size*8 - get_bits_count(&h->gb);

  1282. 

  1283.     if (h->mb_y != h->mb_height || h->mb_x != h->mb_width) {

  1284.         av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, h->mb_y, h->mb_x, left);

  1285.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1286.     }

  1287. 

  1288.     if (left < 0) {

  1289.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1290.         return -1;

  1291.     }

  1292. 

  1293.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1294.         ret = av_frame_ref(data, &s->cur_pic->f);

  1295.     else if (s->last_pic->f.data[0])

  1296.         ret = av_frame_ref(data, &s->last_pic->f);

  1297.     if (ret < 0)

  1298.         return ret;

  1299. 

  1300.     /* Do not output the last pic after seeking. */

  1301.     if (s->last_pic->f.data[0] || h->low_delay)

  1302.         *got_frame = 1;

  1303. 

  1304.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1305.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1306.     } else {

  1307.         av_frame_unref(&s->cur_pic->f);

  1308.     }

  1309. 

  1310.     return buf_size;

  1311. }

  1312. 

  1313. static int svq3_decode_end(AVCodecContext *avctx)

  1314. {

  1315.     SVQ3Context *s = avctx->priv_data;

  1316.     H264Context *h = &s->h;

  1317. 

  1318.     free_picture(avctx, s->cur_pic);

  1319.     free_picture(avctx, s->next_pic);

  1320.     free_picture(avctx, s->last_pic);

  1321.     av_freep(&s->cur_pic);

  1322.     av_freep(&s->next_pic);

  1323.     av_freep(&s->last_pic);

  1324. 

  1325.     av_frame_unref(&h->cur_pic.f);

  1326. 

  1327.     ff_h264_free_context(h);

  1328. 

  1329.     av_freep(&s->buf);

  1330.     s->buf_size = 0;

  1331.     av_freep(&h->edge_emu_buffer);

  1332. 

  1333.     return 0;

  1334. }

  1335. 

  1336. AVCodec ff_svq3_decoder = {

  1337.     .name           = "svq3",

  1338.     .type           = AVMEDIA_TYPE_VIDEO,

  1339.     .id             = AV_CODEC_ID_SVQ3,

  1340.     .priv_data_size = sizeof(SVQ3Context),

  1341.     .init           = svq3_decode_init,

  1342.     .close          = svq3_decode_end,

  1343.     .decode         = svq3_decode_frame,

  1344.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1345.                       CODEC_CAP_DR1             |

  1346.                       CODEC_CAP_DELAY,

  1347.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1348.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1349.                                                      AV_PIX_FMT_NONE},

  1350. };