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.
55617 Commits
32 Branches
366MB
Tree: a0c6c8e53e
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a0c6c8e53e'
${ noResults }
FFmpeg/libavcodec/svq3.c

1353 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. 

  43. #include "libavutil/attributes.h"

  44. #include "internal.h"

  45. #include "avcodec.h"

  46. #include "mpegvideo.h"

  47. #include "h264.h"

  48. 

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

  50. 

  51. #include "h264_mvpred.h"

  52. #include "golomb.h"

  53. #include "hpeldsp.h"

  54. #include "rectangle.h"

  55. #include "vdpau_internal.h"

  56. 

  57. #if CONFIG_ZLIB

  58. #include <zlib.h>

  59. #endif

  60. 

  61. #include "svq1.h"

  62. #include "svq3.h"

  63. 

  64. /**

  65.  * @file

  66.  * svq3 decoder.

  67.  */

  68. 

  69. typedef struct {

  70.     H264Context h;

  71.     HpelDSPContext hdsp;

  72.     Picture *cur_pic;

  73.     Picture *next_pic;

  74.     Picture *last_pic;

  75.     int halfpel_flag;

  76.     int thirdpel_flag;

  77.     int unknown_flag;

  78.     int next_slice_index;

  79.     uint32_t watermark_key;

  80.     uint8_t *buf;

  81.     int buf_size;

  82.     int adaptive_quant;

  83.     int next_p_frame_damaged;

  84.     int h_edge_pos;

  85.     int v_edge_pos;

  86.     int last_frame_output;

  87. } SVQ3Context;

  88. 

  89. #define FULLPEL_MODE  1

  90. #define HALFPEL_MODE  2

  91. #define THIRDPEL_MODE 3

  92. #define PREDICT_MODE  4

  93. 

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

  95.  * o-->o-->o   o

  96.  *         |  /|

  97.  * o   o   o / o

  98.  * | / |   |/  |

  99.  * o   o   o   o

  100.  *   /

  101.  * o-->o-->o-->o

  102.  */

  103. static const uint8_t svq3_scan[16] = {

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

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

  106.     0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4,

  107.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  108. };

  109. 

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

  111.     { 0, 0 },

  112.     { 1, 0 }, { 0, 1 },

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

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

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

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

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

  118.     { 4, 3 }, { 3, 4 },

  119.     { 4, 4 }

  120. };

  121. 

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

  123.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  124.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

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

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

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

  133.     { { 0,  2, -1, -1, -1 }, { 0, 2,  4,  1,  3 }, { 1, 4,  2,  0,  3 },

  134.       { 4,  2,  0,  1,  3 }, { 2, 0,  1,  4,  3 }, { 4, 2,  1,  0,  3 } },

  135. };

  136. 

  137. static const struct {

  138.     uint8_t run;

  139.     uint8_t level;

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

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

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

  143.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  144.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  145. };

  146. 

  147. static const uint32_t svq3_dequant_coeff[32] = {

  148.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  149.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  150.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  151.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  152. };

  153. 

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

  155. {

  156.     const int qmul = svq3_dequant_coeff[qp];

  157. #define stride 16

  158.     int i;

  159.     int temp[16];

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

  161. 

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

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

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

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

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

  167. 

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

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

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

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

  172.     }

  173. 

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

  175.         const int offset = x_offset[i];

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

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

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

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

  180. 

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

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

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

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

  185.     }

  186. }

  187. #undef stride

  188. 

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

  190.                         int stride, int qp, int dc)

  191. {

  192.     const int qmul = svq3_dequant_coeff[qp];

  193.     int i;

  194. 

  195.     if (dc) {

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

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

  198.         block[0] = 0;

  199.     }

  200. 

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

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

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

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

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

  206. 

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

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

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

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

  211.     }

  212. 

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

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

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

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

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

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

  219. 

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

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

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

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

  224.     }

  225. 

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

  227. }

  228. 

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

  230.                                     int index, const int type)

  231. {

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

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

  234. 

  235.     int run, level, sign, limit;

  236.     unsigned vlc;

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

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

  239. 

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

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

  242.             if ((int32_t)vlc < 0)

  243.                 return -1;

  244. 

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

  246.             vlc      = vlc + 1 >> 1;

  247. 

  248.             if (type == 3) {

  249.                 if (vlc < 3) {

  250.                     run   = 0;

  251.                     level = vlc;

  252.                 } else if (vlc < 4) {

  253.                     run   = 1;

  254.                     level = 1;

  255.                 } else {

  256.                     run   = vlc & 0x3;

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

  258.                 }

  259.             } else {

  260.                 if (vlc < 16U) {

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

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

  263.                 } else if (intra) {

  264.                     run   = vlc & 0x7;

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

  266.                 } else {

  267.                     run   = vlc & 0xF;

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

  269.                 }

  270.             }

  271. 

  272. 

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

  274.                 return -1;

  275. 

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

  277.         }

  278. 

  279.         if (type != 2) {

  280.             break;

  281.         }

  282.     }

  283. 

  284.     return 0;

  285. }

  286. 

  287. static inline void svq3_mc_dir_part(SVQ3Context *s,

  288.                                     int x, int y, int width, int height,

  289.                                     int mx, int my, int dxy,

  290.                                     int thirdpel, int dir, int avg)

  291. {

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

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

  294.     uint8_t *src, *dest;

  295.     int i, emu = 0;

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

  297. 

  298.     mx += x;

  299.     my += y;

  300. 

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

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

  303.         emu = 1;

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

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

  306.     }

  307. 

  308.     /* form component predictions */

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

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

  311. 

  312.     if (emu) {

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

  314.                                  width + 1, height + 1,

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

  316.         src = h->edge_emu_buffer;

  317.     }

  318.     if (thirdpel)

  319.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  321.                                                 width, height);

  322.     else

  323.         (avg ? s->hdsp.avg_pixels_tab

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

  325.                                                        height);

  326. 

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

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

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

  330.         width  = width  >> 1;

  331.         height = height >> 1;

  332.         blocksize++;

  333. 

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

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

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

  337. 

  338.             if (emu) {

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

  340.                                          width + 1, height + 1,

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

  342.                                          s->v_edge_pos >> 1);

  343.                 src = h->edge_emu_buffer;

  344.             }

  345.             if (thirdpel)

  346.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  348.                                                         h->uvlinesize,

  349.                                                         width, height);

  350.             else

  351.                 (avg ? s->hdsp.avg_pixels_tab

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

  353.                                                                h->uvlinesize,

  354.                                                                height);

  355.         }

  356.     }

  357. }

  358. 

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

  360.                               int dir, int avg)

  361. {

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

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

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

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

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

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

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

  369. 

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

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

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

  373.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  374.             int dxy;

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

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

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

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

  379. 

  380.             if (mode != PREDICT_MODE) {

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

  382.             } else {

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

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

  385. 

  386.                 if (dir == 0) {

  387.                     mx = mx * h->frame_num_offset /

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

  389.                     my = my * h->frame_num_offset /

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

  391.                 } else {

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

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

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

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

  396.                 }

  397.             }

  398. 

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

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

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

  402. 

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

  404.             if (mode == PREDICT_MODE) {

  405.                 dx = dy = 0;

  406.             } else {

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

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

  409. 

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

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

  412.                     return -1;

  413.                 }

  414.             }

  415. 

  416.             /* compute motion vector */

  417.             if (mode == THIRDPEL_MODE) {

  418.                 int fx, fy;

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

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

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

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

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

  424. 

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

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

  427.                 mx += mx;

  428.                 my += my;

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

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

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

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

  433. 

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

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

  436.                 mx *= 3;

  437.                 my *= 3;

  438.             } else {

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

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

  441. 

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

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

  444.                 mx *= 6;

  445.                 my *= 6;

  446.             }

  447. 

  448.             /* update mv_cache */

  449.             if (mode != PREDICT_MODE) {

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

  451. 

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

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

  454. 

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

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

  457.                 }

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

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

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

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

  462.             }

  463. 

  464.             /* write back motion vectors */

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

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

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

  468.         }

  469. 

  470.     return 0;

  471. }

  472. 

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

  474. {

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

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

  477.     int cbp = 0;

  478.     uint32_t vlc;

  479.     int8_t *top, *left;

  480.     const int mb_xy         = h->mb_xy;

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

  482. 

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

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

  485.     h->topright_samples_available = 0xFFFF;

  486. 

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

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

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

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

  491.                              0, 0, 0, 0, 0, 0);

  492. 

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

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

  495.                                  0, 0, 0, 0, 1, 1);

  496. 

  497.             mb_type = MB_TYPE_SKIP;

  498.         } else {

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

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

  501.                 return -1;

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

  503.                 return -1;

  504. 

  505.             mb_type = MB_TYPE_16x16;

  506.         }

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

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

  509.             mode = THIRDPEL_MODE;

  510.         else if (s->halfpel_flag &&

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

  512.             mode = HALFPEL_MODE;

  513.         else

  514.             mode = FULLPEL_MODE;

  515. 

  516.         /* fill caches */

  517.         /* note ref_cache should contain here:

  518.          *  ????????

  519.          *  ???11111

  520.          *  N??11111

  521.          *  N??11111

  522.          *  N??11111

  523.          */

  524. 

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

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

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

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

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

  530.             } else {

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

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

  533.             }

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

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

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

  537.                        4 * 2 * sizeof(int16_t));

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

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

  540. 

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

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

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

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

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

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

  547.                 } else

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

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

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

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

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

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

  554.                 } else

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

  556.             } else

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

  558.                        PART_NOT_AVAILABLE, 8);

  559. 

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

  561.                 break;

  562.         }

  563. 

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

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

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

  567.                 return -1;

  568.         } else {        /* AV_PICTURE_TYPE_B */

  569.             if (mb_type != 2) {

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

  571.                     return -1;

  572.             } else {

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

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

  575.                            0, 4 * 2 * sizeof(int16_t));

  576.             }

  577.             if (mb_type != 1) {

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

  579.                     return -1;

  580.             } else {

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

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

  583.                            0, 4 * 2 * sizeof(int16_t));

  584.             }

  585.         }

  586. 

  587.         mb_type = MB_TYPE_16x16;

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

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

  590. 

  591.         if (mb_type == 8) {

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

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

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

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

  596.                     h->left_samples_available = 0x5F5F;

  597.             }

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

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

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

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

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

  603. 

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

  605.                     h->top_samples_available = 0x33FF;

  606.             }

  607. 

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

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

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

  611. 

  612.                 if (vlc >= 25U) {

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

  614.                     return -1;

  615.                 }

  616. 

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

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

  619. 

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

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

  622. 

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

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

  625.                     return -1;

  626.                 }

  627.             }

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

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

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

  631.         }

  632. 

  633.         write_back_intra_pred_mode(h);

  634. 

  635.         if (mb_type == 8) {

  636.             ff_h264_check_intra4x4_pred_mode(h);

  637. 

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

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

  640.         } else {

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

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

  643. 

  644.             h->top_samples_available  = 0x33FF;

  645.             h->left_samples_available = 0x5F5F;

  646.         }

  647. 

  648.         mb_type = MB_TYPE_INTRA4x4;

  649.     } else {                      /* INTRA16x16 */

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

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

  652. 

  653.         if ((h->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(h, dir, 0)) < 0) {

  654.             av_log(h->avctx, AV_LOG_ERROR, "check_intra_pred_mode < 0\n");

  655.             return -1;

  656.         }

  657. 

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

  659.         mb_type = MB_TYPE_INTRA16x16;

  660.     }

  661. 

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

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

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

  665.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  669.                        0, 4 * 2 * sizeof(int16_t));

  670.         }

  671.     }

  672.     if (!IS_INTRA4x4(mb_type)) {

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

  674.     }

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

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

  677.     }

  678. 

  679.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  683.             return -1;

  684.         }

  685. 

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

  687.                                 : golomb_to_inter_cbp[vlc];

  688.     }

  689.     if (IS_INTRA16x16(mb_type) ||

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

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

  692. 

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

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

  695.             return -1;

  696.         }

  697.     }

  698.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  702.             av_log(h->avctx, AV_LOG_ERROR,

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

  704.             return -1;

  705.         }

  706.     }

  707. 

  708.     if (cbp) {

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

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

  711. 

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

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

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

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

  716.                                  2 * (j & 2) + 4 * (i & 2))

  717.                               : (4 * i + j);

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

  719. 

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

  721.                         av_log(h->avctx, AV_LOG_ERROR,

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

  723.                         return -1;

  724.                     }

  725.                 }

  726.             }

  727. 

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

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

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

  731.                     av_log(h->avctx, AV_LOG_ERROR,

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

  733.                     return -1;

  734.                 }

  735. 

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

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

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

  739.                         k                                 = 16 * i + j;

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

  741. 

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

  743.                             av_log(h->avctx, AV_LOG_ERROR,

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

  745.                             return -1;

  746.                         }

  747.                     }

  748.                 }

  749.             }

  750.         }

  751.     }

  752. 

  753.     h->cbp                              = cbp;

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

  755. 

  756.     if (IS_INTRA(mb_type))

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

  758. 

  759.     return 0;

  760. }

  761. 

  762. static int svq3_decode_slice_header(AVCodecContext *avctx)

  763. {

  764.     SVQ3Context *s = avctx->priv_data;

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

  766.     const int mb_xy   = h->mb_xy;

  767.     int i, header;

  768.     unsigned slice_id;

  769. 

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

  771. 

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

  773.         /* TODO: what? */

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

  775.         return -1;

  776.     } else {

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

  778. 

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

  780.                               8 * show_bits(&h->gb, 8 * length) +

  781.                               8 * length;

  782. 

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

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

  785.             return -1;

  786.         }

  787. 

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

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

  790. 

  791.         if (s->watermark_key) {

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

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

  794.                     header ^ s->watermark_key);

  795.         }

  796.         if (length > 0) {

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

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

  799.         }

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

  801.     }

  802. 

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

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

  805.         return -1;

  806.     }

  807. 

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

  809. 

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

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

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

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

  814.     } else {

  815.         skip_bits1(&h->gb);

  816.         h->mb_skip_run = 0;

  817.     }

  818. 

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

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

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

  822. 

  823.     /* unknown fields */

  824.     skip_bits1(&h->gb);

  825. 

  826.     if (s->unknown_flag)

  827.         skip_bits1(&h->gb);

  828. 

  829.     skip_bits1(&h->gb);

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

  831. 

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

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

  834. 

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

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

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

  838.                -1, 4 * sizeof(int8_t));

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

  840.                -1, 8 * sizeof(int8_t) * h->mb_x);

  841.     }

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

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

  844.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  845. 

  846.         if (h->mb_x > 0)

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

  848.     }

  849. 

  850.     return 0;

  851. }

  852. 

  853. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  854. {

  855.     SVQ3Context *s = avctx->priv_data;

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

  857.     int m;

  858.     unsigned char *extradata;

  859.     unsigned char *extradata_end;

  860.     unsigned int size;

  861.     int marker_found = 0;

  862. 

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

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

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

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

  867.         av_freep(&s->cur_pic);

  868.         av_freep(&s->last_pic);

  869.         av_freep(&s->next_pic);

  870.         return AVERROR(ENOMEM);

  871.     }

  872. 

  873.     if (ff_h264_decode_init(avctx) < 0)

  874.         return -1;

  875. 

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

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

  878.     h->is_complex      = 1;

  879.     h->sps.chroma_format_idc = 1;

  880.     h->picture_structure = PICT_FRAME;

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

  882. 

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

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

  885. 

  886.     s->halfpel_flag  = 1;

  887.     s->thirdpel_flag = 1;

  888.     s->unknown_flag  = 0;

  889. 

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

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

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

  893.     if (extradata) {

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

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

  896.                 marker_found = 1;

  897.                 break;

  898.             }

  899.             extradata++;

  900.         }

  901.     }

  902. 

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

  904.     if (marker_found) {

  905.         GetBitContext gb;

  906.         int frame_size_code;

  907. 

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

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

  910.             return AVERROR_INVALIDDATA;

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

  912. 

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

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

  915.         switch (frame_size_code) {

  916.         case 0:

  917.             avctx->width  = 160;

  918.             avctx->height = 120;

  919.             break;

  920.         case 1:

  921.             avctx->width  = 128;

  922.             avctx->height =  96;

  923.             break;

  924.         case 2:

  925.             avctx->width  = 176;

  926.             avctx->height = 144;

  927.             break;

  928.         case 3:

  929.             avctx->width  = 352;

  930.             avctx->height = 288;

  931.             break;

  932.         case 4:

  933.             avctx->width  = 704;

  934.             avctx->height = 576;

  935.             break;

  936.         case 5:

  937.             avctx->width  = 240;

  938.             avctx->height = 180;

  939.             break;

  940.         case 6:

  941.             avctx->width  = 320;

  942.             avctx->height = 240;

  943.             break;

  944.         case 7:

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

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

  947.             break;

  948.         }

  949. 

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

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

  952. 

  953.         /* unknown fields */

  954.         skip_bits1(&gb);

  955.         skip_bits1(&gb);

  956.         skip_bits1(&gb);

  957.         skip_bits1(&gb);

  958. 

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

  960. 

  961.         /* unknown field */

  962.         skip_bits1(&gb);

  963. 

  964.         while (get_bits1(&gb))

  965.             skip_bits(&gb, 8);

  966. 

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

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

  969.         if (s->unknown_flag) {

  970. #if CONFIG_ZLIB

  971.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  972.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  973.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  976.             int u4                    = svq3_get_ue_golomb(&gb);

  977.             unsigned long buf_len     = watermark_width *

  978.                                         watermark_height * 4;

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

  980.             uint8_t *buf;

  981. 

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

  983.                 return -1;

  984. 

  985.             buf = av_malloc(buf_len);

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

  987.                    watermark_width, watermark_height);

  988.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  992.                            size - offset) != Z_OK) {

  993.                 av_log(avctx, AV_LOG_ERROR,

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

  995.                 av_free(buf);

  996.                 return -1;

  997.             }

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

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

  1000.             av_log(avctx, AV_LOG_DEBUG,

  1001.                    "watermark key %#x\n", s->watermark_key);

  1002.             av_free(buf);

  1003. #else

  1004.             av_log(avctx, AV_LOG_ERROR,

  1005.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1006.             return -1;

  1007. #endif

  1008.         }

  1009.     }

  1010. 

  1011.     h->width  = avctx->width;

  1012.     h->height = avctx->height;

  1013.     h->mb_width  = (h->width + 15) / 16;

  1014.     h->mb_height = (h->height + 15) / 16;

  1015.     h->mb_stride = h->mb_width + 1;

  1016.     h->mb_num    = h->mb_width * h->mb_height;

  1017.     h->b_stride = 4 * h->mb_width;

  1018.     s->h_edge_pos = h->mb_width * 16;

  1019.     s->v_edge_pos = h->mb_height * 16;

  1020. 

  1021.     if (ff_h264_alloc_tables(h) < 0) {

  1022.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1023.         return AVERROR(ENOMEM);

  1024.     }

  1025. 

  1026.     return 0;

  1027. }

  1028. 

  1029. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1030. {

  1031.     int i;

  1032.     for (i = 0; i < 2; i++) {

  1033.         av_buffer_unref(&pic->motion_val_buf[i]);

  1034.         av_buffer_unref(&pic->ref_index_buf[i]);

  1035.     }

  1036.     av_buffer_unref(&pic->mb_type_buf);

  1037. 

  1038.     av_frame_unref(&pic->f);

  1039. }

  1040. 

  1041. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1042. {

  1043.     SVQ3Context *s = avctx->priv_data;

  1044.     H264Context *h = &s->h;

  1045.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1046.     const int mb_array_size = h->mb_stride * h->mb_height;

  1047.     const int b4_stride     = h->mb_width * 4 + 1;

  1048.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1049.     int ret;

  1050. 

  1051.     if (!pic->motion_val_buf[0]) {

  1052.         int i;

  1053. 

  1054.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1055.         if (!pic->mb_type_buf)

  1056.             return AVERROR(ENOMEM);

  1057.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1058. 

  1059.         for (i = 0; i < 2; i++) {

  1060.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1061.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1062.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1063.                 ret = AVERROR(ENOMEM);

  1064.                 goto fail;

  1065.             }

  1066. 

  1067.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1068.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1069.         }

  1070.     }

  1071.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1072. 

  1073.     ret = ff_get_buffer(avctx, &pic->f,

  1074.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1075.     if (ret < 0)

  1076.         goto fail;

  1077. 

  1078.     if (!h->edge_emu_buffer) {

  1079.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1080.         if (!h->edge_emu_buffer)

  1081.             return AVERROR(ENOMEM);

  1082.     }

  1083. 

  1084.     h->linesize   = pic->f.linesize[0];

  1085.     h->uvlinesize = pic->f.linesize[1];

  1086. 

  1087.     return 0;

  1088. fail:

  1089.     free_picture(avctx, pic);

  1090.     return ret;

  1091. }

  1092. 

  1093. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1094.                              int *got_frame, AVPacket *avpkt)

  1095. {

  1096.     SVQ3Context *s     = avctx->priv_data;

  1097.     H264Context *h     = &s->h;

  1098.     int buf_size       = avpkt->size;

  1099.     int left;

  1100.     uint8_t *buf;

  1101.     int ret, m, i;

  1102. 

  1103.     /* special case for last picture */

  1104.     if (buf_size == 0) {

  1105.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1106.             ret = av_frame_ref(data, &s->next_pic->f);

  1107.             if (ret < 0)

  1108.                 return ret;

  1109.             s->last_frame_output = 1;

  1110.             *got_frame          = 1;

  1111.         }

  1112.         return 0;

  1113.     }

  1114. 

  1115.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1116. 

  1117.     if (s->watermark_key) {

  1118.         av_fast_malloc(&s->buf, &s->buf_size,

  1119.                        buf_size+FF_INPUT_BUFFER_PADDING_SIZE);

  1120.         if (!s->buf)

  1121.             return AVERROR(ENOMEM);

  1122.         memcpy(s->buf, avpkt->data, buf_size);

  1123.         buf = s->buf;

  1124.     } else {

  1125.         buf = avpkt->data;

  1126.     }

  1127. 

  1128.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1129. 

  1130.     if (svq3_decode_slice_header(avctx))

  1131.         return -1;

  1132. 

  1133.     h->pict_type = h->slice_type;

  1134. 

  1135.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1136.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1137. 

  1138.     av_frame_unref(&s->cur_pic->f);

  1139. 

  1140.     /* for skipping the frame */

  1141.     s->cur_pic->f.pict_type = h->pict_type;

  1142.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1143. 

  1144.     ret = get_buffer(avctx, s->cur_pic);

  1145.     if (ret < 0)

  1146.         return ret;

  1147. 

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

  1149.     av_frame_unref(&h->cur_pic.f);

  1150.     h->cur_pic     = *s->cur_pic;

  1151.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1152.     if (ret < 0)

  1153.         return ret;

  1154. 

  1155.     for (i = 0; i < 16; i++) {

  1156.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1157.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1158.     }

  1159.     for (i = 0; i < 16; i++) {

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

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

  1162.         h->block_offset[48 + 16 + i] =

  1163.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1164.     }

  1165. 

  1166.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1167.         if (!s->last_pic->f.data[0]) {

  1168.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1169.             ret = get_buffer(avctx, s->last_pic);

  1170.             if (ret < 0)

  1171.                 return ret;

  1172.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1173.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1174.                    s->last_pic->f.linesize[1]);

  1175.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1176.                    s->last_pic->f.linesize[2]);

  1177.         }

  1178. 

  1179.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1180.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1181.             ret = get_buffer(avctx, s->next_pic);

  1182.             if (ret < 0)

  1183.                 return ret;

  1184.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1185.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1186.                    s->next_pic->f.linesize[1]);

  1187.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1188.                    s->next_pic->f.linesize[2]);

  1189.         }

  1190.     }

  1191. 

  1192.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1193.         av_log(h->avctx, AV_LOG_DEBUG,

  1194.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1195.                av_get_picture_type_char(h->pict_type),

  1196.                s->halfpel_flag, s->thirdpel_flag,

  1197.                s->adaptive_quant, h->qscale, h->slice_num);

  1198. 

  1199.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1200.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1201.         avctx->skip_frame >= AVDISCARD_ALL)

  1202.         return 0;

  1203. 

  1204.     if (s->next_p_frame_damaged) {

  1205.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1206.             return 0;

  1207.         else

  1208.             s->next_p_frame_damaged = 0;

  1209.     }

  1210. 

  1211.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1212.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1213. 

  1214.         if (h->frame_num_offset < 0)

  1215.             h->frame_num_offset += 256;

  1216.         if (h->frame_num_offset == 0 ||

  1217.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1218.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1219.             return -1;

  1220.         }

  1221.     } else {

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

  1223.         h->frame_num             = h->slice_num;

  1224.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1225. 

  1226.         if (h->prev_frame_num_offset < 0)

  1227.             h->prev_frame_num_offset += 256;

  1228.     }

  1229. 

  1230.     for (m = 0; m < 2; m++) {

  1231.         int i;

  1232.         for (i = 0; i < 4; i++) {

  1233.             int j;

  1234.             for (j = -1; j < 4; j++)

  1235.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1236.             if (i < 3)

  1237.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1238.         }

  1239.     }

  1240. 

  1241.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1242.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1243.             unsigned mb_type;

  1244.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1245. 

  1246.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1247.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1248.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1249.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1250.                 h->gb.size_in_bits = 8 * buf_size;

  1251. 

  1252.                 if (svq3_decode_slice_header(avctx))

  1253.                     return -1;

  1254. 

  1255.                 /* TODO: support s->mb_skip_run */

  1256.             }

  1257. 

  1258.             mb_type = svq3_get_ue_golomb(&h->gb);

  1259. 

  1260.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1261.                 mb_type += 8;

  1262.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1263.                 mb_type += 4;

  1264.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1265.                 av_log(h->avctx, AV_LOG_ERROR,

  1266.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1267.                 return -1;

  1268.             }

  1269. 

  1270.             if (mb_type != 0 || h->cbp)

  1271.                 ff_h264_hl_decode_mb(h);

  1272. 

  1273.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1274.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1275.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1276.         }

  1277. 

  1278.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1279.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1280.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1281.     }

  1282. 

  1283.     left = buf_size*8 - get_bits_count(&h->gb);

  1284. 

  1285.     if (h->mb_y != h->mb_height || h->mb_x != h->mb_width) {

  1286.         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);

  1287.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1288.     }

  1289. 

  1290.     if (left < 0) {

  1291.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1292.         return -1;

  1293.     }

  1294. 

  1295.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

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

  1297.     else if (s->last_pic->f.data[0])

  1298.         ret = av_frame_ref(data, &s->last_pic->f);

  1299.     if (ret < 0)

  1300.         return ret;

  1301. 

  1302.     /* Do not output the last pic after seeking. */

  1303.     if (s->last_pic->f.data[0] || h->low_delay)

  1304.         *got_frame = 1;

  1305. 

  1306.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1307.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1308.     } else {

  1309.         av_frame_unref(&s->cur_pic->f);

  1310.     }

  1311. 

  1312.     return buf_size;

  1313. }

  1314. 

  1315. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1316. {

  1317.     SVQ3Context *s = avctx->priv_data;

  1318.     H264Context *h = &s->h;

  1319. 

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

  1321.     free_picture(avctx, s->next_pic);

  1322.     free_picture(avctx, s->last_pic);

  1323.     av_freep(&s->cur_pic);

  1324.     av_freep(&s->next_pic);

  1325.     av_freep(&s->last_pic);

  1326. 

  1327.     av_frame_unref(&h->cur_pic.f);

  1328. 

  1329.     ff_h264_free_context(h);

  1330. 

  1331.     av_freep(&s->buf);

  1332.     s->buf_size = 0;

  1333.     av_freep(&h->edge_emu_buffer);

  1334. 

  1335.     return 0;

  1336. }

  1337. 

  1338. AVCodec ff_svq3_decoder = {

  1339.     .name           = "svq3",

  1340.     .type           = AVMEDIA_TYPE_VIDEO,

  1341.     .id             = AV_CODEC_ID_SVQ3,

  1342.     .priv_data_size = sizeof(SVQ3Context),

  1343.     .init           = svq3_decode_init,

  1344.     .close          = svq3_decode_end,

  1345.     .decode         = svq3_decode_frame,

  1346.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1347.                       CODEC_CAP_DR1             |

  1348.                       CODEC_CAP_DELAY,

  1349.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1350.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1351.                                                      AV_PIX_FMT_NONE},

  1352. };