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.
37585 Commits
32 Branches
292MB
Tree: e0d6e2fc67
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'e0d6e2fc67'
${ noResults }
FFmpeg/libavcodec/svq3.c

1320 lines
47KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The Libav Project

  3.  *

  4.  * This file is part of Libav.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

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

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

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

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

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

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

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

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

  31.  *

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

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

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

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

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

  37.  *

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

  39.  * correctly decodes this file:

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

  41.  */

  42. 

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

  241.             vlc      = vlc + 1 >> 1;

  242. 

  243.             if (type == 3) {

  244.                 if (vlc < 3) {

  245.                     run   = 0;

  246.                     level = vlc;

  247.                 } else if (vlc < 4) {

  248.                     run   = 1;

  249.                     level = 1;

  250.                 } else {

  251.                     run   = vlc & 0x3;

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

  253.                 }

  254.             } else {

  255.                 if (vlc < 16) {

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

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

  258.                 } else if (intra) {

  259.                     run   = vlc & 0x7;

  260.                     level = (vlc >> 3) +

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

  262.                 } else {

  263.                     run   = vlc & 0xF;

  264.                     level = (vlc >> 4) +

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

  266.                 }

  267.             }

  268. 

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

  270.                 return -1;

  271. 

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

  273.         }

  274. 

  275.         if (type != 2) {

  276.             break;

  277.         }

  278.     }

  279. 

  280.     return 0;

  281. }

  282. 

  283. static inline void svq3_mc_dir_part(SVQ3Context *s,

  284.                                     int x, int y, int width, int height,

  285.                                     int mx, int my, int dxy,

  286.                                     int thirdpel, int dir, int avg)

  287. {

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

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

  290.     uint8_t *src, *dest;

  291.     int i, emu = 0;

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

  293. 

  294.     mx += x;

  295.     my += y;

  296. 

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

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

  299.         emu = 1;

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

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

  302.     }

  303. 

  304.     /* form component predictions */

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

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

  307. 

  308.     if (emu) {

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

  310.                                  width + 1, height + 1,

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

  312.         src = h->edge_emu_buffer;

  313.     }

  314.     if (thirdpel)

  315.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  317.                                                 width, height);

  318.     else

  319.         (avg ? s->hdsp.avg_pixels_tab

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

  321.                                                        height);

  322. 

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

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

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

  326.         width  = width  >> 1;

  327.         height = height >> 1;

  328.         blocksize++;

  329. 

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

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

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

  333. 

  334.             if (emu) {

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

  336.                                          width + 1, height + 1,

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

  338.                                          s->v_edge_pos >> 1);

  339.                 src = h->edge_emu_buffer;

  340.             }

  341.             if (thirdpel)

  342.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  344.                                                         h->uvlinesize,

  345.                                                         width, height);

  346.             else

  347.                 (avg ? s->hdsp.avg_pixels_tab

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

  349.                                                                h->uvlinesize,

  350.                                                                height);

  351.         }

  352.     }

  353. }

  354. 

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

  356.                               int dir, int avg)

  357. {

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

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

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

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

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

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

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

  365. 

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

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

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

  369.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  370.             int dxy;

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

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

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

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

  375. 

  376.             if (mode != PREDICT_MODE) {

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

  378.             } else {

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

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

  381. 

  382.                 if (dir == 0) {

  383.                     mx = mx * h->frame_num_offset /

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

  385.                     my = my * h->frame_num_offset /

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

  387.                 } else {

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

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

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

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

  392.                 }

  393.             }

  394. 

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

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

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

  398. 

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

  400.             if (mode == PREDICT_MODE) {

  401.                 dx = dy = 0;

  402.             } else {

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

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

  405. 

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

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

  408.                     return -1;

  409.                 }

  410.             }

  411. 

  412.             /* compute motion vector */

  413.             if (mode == THIRDPEL_MODE) {

  414.                 int fx, fy;

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

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

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

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

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

  420. 

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

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

  423.                 mx += mx;

  424.                 my += my;

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

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

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

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

  429. 

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

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

  432.                 mx *= 3;

  433.                 my *= 3;

  434.             } else {

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

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

  437. 

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

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

  440.                 mx *= 6;

  441.                 my *= 6;

  442.             }

  443. 

  444.             /* update mv_cache */

  445.             if (mode != PREDICT_MODE) {

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

  447. 

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

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

  450. 

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

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

  453.                 }

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

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

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

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

  458.             }

  459. 

  460.             /* write back motion vectors */

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

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

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

  464.         }

  465. 

  466.     return 0;

  467. }

  468. 

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

  470. {

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

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

  473.     int cbp = 0;

  474.     uint32_t vlc;

  475.     int8_t *top, *left;

  476.     const int mb_xy         = h->mb_xy;

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

  478. 

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

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

  481.     h->topright_samples_available = 0xFFFF;

  482. 

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

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

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

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

  487.                              0, 0, 0, 0, 0, 0);

  488. 

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

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

  491.                                  0, 0, 0, 0, 1, 1);

  492. 

  493.             mb_type = MB_TYPE_SKIP;

  494.         } else {

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

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

  497.                 return -1;

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

  499.                 return -1;

  500. 

  501.             mb_type = MB_TYPE_16x16;

  502.         }

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

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

  505.             mode = THIRDPEL_MODE;

  506.         else if (s->halfpel_flag &&

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

  508.             mode = HALFPEL_MODE;

  509.         else

  510.             mode = FULLPEL_MODE;

  511. 

  512.         /* fill caches */

  513.         /* note ref_cache should contain here:

  514.          *  ????????

  515.          *  ???11111

  516.          *  N??11111

  517.          *  N??11111

  518.          *  N??11111

  519.          */

  520. 

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

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

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

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

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

  526.             } else {

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

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

  529.             }

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

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

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

  533.                        4 * 2 * sizeof(int16_t));

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

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

  536. 

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

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

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

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

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

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

  543.                 } else

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

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

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

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

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

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

  550.                 } else

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

  552.             } else

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

  554.                        PART_NOT_AVAILABLE, 8);

  555. 

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

  557.                 break;

  558.         }

  559. 

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

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

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

  563.                 return -1;

  564.         } else {        /* AV_PICTURE_TYPE_B */

  565.             if (mb_type != 2) {

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

  567.                     return -1;

  568.             } else {

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

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

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

  572.             }

  573.             if (mb_type != 1) {

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

  575.                     return -1;

  576.             } else {

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

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

  579.                            0, 4 * 2 * sizeof(int16_t));

  580.             }

  581.         }

  582. 

  583.         mb_type = MB_TYPE_16x16;

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

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

  586. 

  587.         if (mb_type == 8) {

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

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

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

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

  592.                     h->left_samples_available = 0x5F5F;

  593.             }

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

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

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

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

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

  599. 

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

  601.                     h->top_samples_available = 0x33FF;

  602.             }

  603. 

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

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

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

  607. 

  608.                 if (vlc >= 25) {

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

  610.                     return -1;

  611.                 }

  612. 

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

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

  615. 

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

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

  618. 

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

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

  621.                     return -1;

  622.                 }

  623.             }

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

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

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

  627.         }

  628. 

  629.         write_back_intra_pred_mode(h);

  630. 

  631.         if (mb_type == 8) {

  632.             ff_h264_check_intra4x4_pred_mode(h);

  633. 

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

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

  636.         } else {

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

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

  639. 

  640.             h->top_samples_available  = 0x33FF;

  641.             h->left_samples_available = 0x5F5F;

  642.         }

  643. 

  644.         mb_type = MB_TYPE_INTRA4x4;

  645.     } else {                      /* INTRA16x16 */

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

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

  648. 

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

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

  651.             return -1;

  652.         }

  653. 

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

  655.         mb_type = MB_TYPE_INTRA16x16;

  656.     }

  657. 

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

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

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

  661.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

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

  666.         }

  667.     }

  668.     if (!IS_INTRA4x4(mb_type)) {

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

  670.     }

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

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

  673.     }

  674. 

  675.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  679.             return -1;

  680.         }

  681. 

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

  683.                                 : golomb_to_inter_cbp[vlc];

  684.     }

  685.     if (IS_INTRA16x16(mb_type) ||

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

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

  688. 

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

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

  691.             return -1;

  692.         }

  693.     }

  694.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  698.             av_log(h->avctx, AV_LOG_ERROR,

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

  700.             return -1;

  701.         }

  702.     }

  703. 

  704.     if (cbp) {

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

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

  707. 

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

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

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

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

  712.                                  2 * (j & 2) + 4 * (i & 2))

  713.                               : (4 * i + j);

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

  715. 

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

  717.                         av_log(h->avctx, AV_LOG_ERROR,

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

  719.                         return -1;

  720.                     }

  721.                 }

  722.             }

  723. 

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

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

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

  727.                     av_log(h->avctx, AV_LOG_ERROR,

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

  729.                     return -1;

  730.                 }

  731. 

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

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

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

  735.                         k                                 = 16 * i + j;

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

  737. 

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

  739.                             av_log(h->avctx, AV_LOG_ERROR,

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

  741.                             return -1;

  742.                         }

  743.                     }

  744.                 }

  745.             }

  746.         }

  747.     }

  748. 

  749.     h->cbp                              = cbp;

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

  751. 

  752.     if (IS_INTRA(mb_type))

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

  754. 

  755.     return 0;

  756. }

  757. 

  758. static int svq3_decode_slice_header(AVCodecContext *avctx)

  759. {

  760.     SVQ3Context *s = avctx->priv_data;

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

  762.     const int mb_xy   = h->mb_xy;

  763.     int i, header;

  764.     unsigned slice_id;

  765. 

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

  767. 

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

  769.         /* TODO: what? */

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

  771.         return -1;

  772.     } else {

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

  774. 

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

  776.                               8 * show_bits(&h->gb, 8 * length) +

  777.                               8 * length;

  778. 

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

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

  781.             return -1;

  782.         }

  783. 

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

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

  786. 

  787.         if (s->watermark_key) {

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

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

  790.                     header ^ s->watermark_key);

  791.         }

  792.         if (length > 0) {

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

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

  795.         }

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

  797.     }

  798. 

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

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

  801.         return -1;

  802.     }

  803. 

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

  805. 

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

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

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

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

  810.     } else {

  811.         skip_bits1(&h->gb);

  812.         h->mb_skip_run = 0;

  813.     }

  814. 

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

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

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

  818. 

  819.     /* unknown fields */

  820.     skip_bits1(&h->gb);

  821. 

  822.     if (s->unknown_flag)

  823.         skip_bits1(&h->gb);

  824. 

  825.     skip_bits1(&h->gb);

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

  827. 

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

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

  830. 

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

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

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

  834.                -1, 4 * sizeof(int8_t));

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

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

  837.     }

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

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

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

  841. 

  842.         if (h->mb_x > 0)

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

  844.     }

  845. 

  846.     return 0;

  847. }

  848. 

  849. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  850. {

  851.     SVQ3Context *s = avctx->priv_data;

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

  853.     int m;

  854.     unsigned char *extradata;

  855.     unsigned char *extradata_end;

  856.     unsigned int size;

  857.     int marker_found = 0;

  858. 

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

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

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

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

  863.         av_freep(&s->cur_pic);

  864.         av_freep(&s->last_pic);

  865.         av_freep(&s->next_pic);

  866.         return AVERROR(ENOMEM);

  867.     }

  868. 

  869.     if (ff_h264_decode_init(avctx) < 0)

  870.         return -1;

  871. 

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

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

  874.     h->is_complex      = 1;

  875.     h->picture_structure = PICT_FRAME;

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

  877. 

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

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

  880. 

  881.     s->halfpel_flag  = 1;

  882.     s->thirdpel_flag = 1;

  883.     s->unknown_flag  = 0;

  884. 

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

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

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

  888.     if (extradata) {

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

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

  891.                 marker_found = 1;

  892.                 break;

  893.             }

  894.             extradata++;

  895.         }

  896.     }

  897. 

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

  899.     if (marker_found) {

  900.         GetBitContext gb;

  901.         int frame_size_code;

  902. 

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

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

  905.             return AVERROR_INVALIDDATA;

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

  907. 

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

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

  910.         switch (frame_size_code) {

  911.         case 0:

  912.             avctx->width  = 160;

  913.             avctx->height = 120;

  914.             break;

  915.         case 1:

  916.             avctx->width  = 128;

  917.             avctx->height =  96;

  918.             break;

  919.         case 2:

  920.             avctx->width  = 176;

  921.             avctx->height = 144;

  922.             break;

  923.         case 3:

  924.             avctx->width  = 352;

  925.             avctx->height = 288;

  926.             break;

  927.         case 4:

  928.             avctx->width  = 704;

  929.             avctx->height = 576;

  930.             break;

  931.         case 5:

  932.             avctx->width  = 240;

  933.             avctx->height = 180;

  934.             break;

  935.         case 6:

  936.             avctx->width  = 320;

  937.             avctx->height = 240;

  938.             break;

  939.         case 7:

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

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

  942.             break;

  943.         }

  944. 

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

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

  947. 

  948.         /* unknown fields */

  949.         skip_bits1(&gb);

  950.         skip_bits1(&gb);

  951.         skip_bits1(&gb);

  952.         skip_bits1(&gb);

  953. 

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

  955. 

  956.         /* unknown field */

  957.         skip_bits1(&gb);

  958. 

  959.         while (get_bits1(&gb))

  960.             skip_bits(&gb, 8);

  961. 

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

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

  964.         if (s->unknown_flag) {

  965. #if CONFIG_ZLIB

  966.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  967.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  968.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  971.             int u4                    = svq3_get_ue_golomb(&gb);

  972.             unsigned long buf_len     = watermark_width *

  973.                                         watermark_height * 4;

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

  975.             uint8_t *buf;

  976. 

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

  978.                 return -1;

  979. 

  980.             buf = av_malloc(buf_len);

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

  982.                    watermark_width, watermark_height);

  983.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  987.                            size - offset) != Z_OK) {

  988.                 av_log(avctx, AV_LOG_ERROR,

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

  990.                 av_free(buf);

  991.                 return -1;

  992.             }

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

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

  995.             av_log(avctx, AV_LOG_DEBUG,

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

  997.             av_free(buf);

  998. #else

  999.             av_log(avctx, AV_LOG_ERROR,

  1000.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1001.             return -1;

  1002. #endif

  1003.         }

  1004.     }

  1005. 

  1006.     h->width  = avctx->width;

  1007.     h->height = avctx->height;

  1008.     h->mb_width  = (h->width + 15) / 16;

  1009.     h->mb_height = (h->height + 15) / 16;

  1010.     h->mb_stride = h->mb_width + 1;

  1011.     h->mb_num    = h->mb_width * h->mb_height;

  1012.     h->b_stride = 4 * h->mb_width;

  1013.     s->h_edge_pos = h->mb_width * 16;

  1014.     s->v_edge_pos = h->mb_height * 16;

  1015. 

  1016.     if (ff_h264_alloc_tables(h) < 0) {

  1017.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1018.         return AVERROR(ENOMEM);

  1019.     }

  1020. 

  1021.     return 0;

  1022. }

  1023. 

  1024. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1025. {

  1026.     int i;

  1027.     for (i = 0; i < 2; i++) {

  1028.         av_buffer_unref(&pic->motion_val_buf[i]);

  1029.         av_buffer_unref(&pic->ref_index_buf[i]);

  1030.     }

  1031.     av_buffer_unref(&pic->mb_type_buf);

  1032. 

  1033.     av_frame_unref(&pic->f);

  1034. }

  1035. 

  1036. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1037. {

  1038.     SVQ3Context *s = avctx->priv_data;

  1039.     H264Context *h = &s->h;

  1040.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1041.     const int mb_array_size = h->mb_stride * h->mb_height;

  1042.     const int b4_stride     = h->mb_width * 4 + 1;

  1043.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1044.     int ret;

  1045. 

  1046.     if (!pic->motion_val_buf[0]) {

  1047.         int i;

  1048. 

  1049.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1050.         if (!pic->mb_type_buf)

  1051.             return AVERROR(ENOMEM);

  1052.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1053. 

  1054.         for (i = 0; i < 2; i++) {

  1055.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1056.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1057.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1058.                 ret = AVERROR(ENOMEM);

  1059.                 goto fail;

  1060.             }

  1061. 

  1062.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1063.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1064.         }

  1065.     }

  1066.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1067. 

  1068.     ret = ff_get_buffer(avctx, &pic->f,

  1069.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1070.     if (ret < 0)

  1071.         goto fail;

  1072. 

  1073.     if (!h->edge_emu_buffer) {

  1074.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1075.         if (!h->edge_emu_buffer)

  1076.             return AVERROR(ENOMEM);

  1077.     }

  1078. 

  1079.     h->linesize   = pic->f.linesize[0];

  1080.     h->uvlinesize = pic->f.linesize[1];

  1081. 

  1082.     return 0;

  1083. fail:

  1084.     free_picture(avctx, pic);

  1085.     return ret;

  1086. }

  1087. 

  1088. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1089.                              int *got_frame, AVPacket *avpkt)

  1090. {

  1091.     const uint8_t *buf = avpkt->data;

  1092.     SVQ3Context *s     = avctx->priv_data;

  1093.     H264Context *h     = &s->h;

  1094.     int buf_size       = avpkt->size;

  1095.     int ret, m, i;

  1096. 

  1097.     /* special case for last picture */

  1098.     if (buf_size == 0) {

  1099.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1100.             ret = av_frame_ref(data, &s->next_pic->f);

  1101.             if (ret < 0)

  1102.                 return ret;

  1103.             s->last_frame_output = 1;

  1104.             *got_frame          = 1;

  1105.         }

  1106.         return 0;

  1107.     }

  1108. 

  1109.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1110. 

  1111.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1112. 

  1113.     if (svq3_decode_slice_header(avctx))

  1114.         return -1;

  1115. 

  1116.     h->pict_type = h->slice_type;

  1117. 

  1118.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1119.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1120. 

  1121.     av_frame_unref(&s->cur_pic->f);

  1122. 

  1123.     /* for skipping the frame */

  1124.     s->cur_pic->f.pict_type = h->pict_type;

  1125.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1126. 

  1127.     ret = get_buffer(avctx, s->cur_pic);

  1128.     if (ret < 0)

  1129.         return ret;

  1130. 

  1131.     h->cur_pic_ptr = s->cur_pic;

  1132.     av_frame_unref(&h->cur_pic.f);

  1133.     h->cur_pic     = *s->cur_pic;

  1134.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1135.     if (ret < 0)

  1136.         return ret;

  1137. 

  1138.     for (i = 0; i < 16; i++) {

  1139.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1140.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1141.     }

  1142.     for (i = 0; i < 16; i++) {

  1143.         h->block_offset[16 + i]      =

  1144.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1145.         h->block_offset[48 + 16 + i] =

  1146.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1147.     }

  1148. 

  1149.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1150.         if (!s->last_pic->f.data[0]) {

  1151.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1152.             ret = get_buffer(avctx, s->last_pic);

  1153.             if (ret < 0)

  1154.                 return ret;

  1155.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1156.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1157.                    s->last_pic->f.linesize[1]);

  1158.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1159.                    s->last_pic->f.linesize[2]);

  1160.         }

  1161. 

  1162.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1163.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1164.             ret = get_buffer(avctx, s->next_pic);

  1165.             if (ret < 0)

  1166.                 return ret;

  1167.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1168.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1169.                    s->next_pic->f.linesize[1]);

  1170.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1171.                    s->next_pic->f.linesize[2]);

  1172.         }

  1173.     }

  1174. 

  1175.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1176.         av_log(h->avctx, AV_LOG_DEBUG,

  1177.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1178.                av_get_picture_type_char(h->pict_type),

  1179.                s->halfpel_flag, s->thirdpel_flag,

  1180.                s->adaptive_quant, h->qscale, h->slice_num);

  1181. 

  1182.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1183.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1184.         avctx->skip_frame >= AVDISCARD_ALL)

  1185.         return 0;

  1186. 

  1187.     if (s->next_p_frame_damaged) {

  1188.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1189.             return 0;

  1190.         else

  1191.             s->next_p_frame_damaged = 0;

  1192.     }

  1193. 

  1194.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1195.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1196. 

  1197.         if (h->frame_num_offset < 0)

  1198.             h->frame_num_offset += 256;

  1199.         if (h->frame_num_offset == 0 ||

  1200.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1201.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1202.             return -1;

  1203.         }

  1204.     } else {

  1205.         h->prev_frame_num        = h->frame_num;

  1206.         h->frame_num             = h->slice_num;

  1207.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1208. 

  1209.         if (h->prev_frame_num_offset < 0)

  1210.             h->prev_frame_num_offset += 256;

  1211.     }

  1212. 

  1213.     for (m = 0; m < 2; m++) {

  1214.         int i;

  1215.         for (i = 0; i < 4; i++) {

  1216.             int j;

  1217.             for (j = -1; j < 4; j++)

  1218.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1219.             if (i < 3)

  1220.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1221.         }

  1222.     }

  1223. 

  1224.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1225.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1226.             unsigned mb_type;

  1227.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1228. 

  1229.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1230.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1231.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1232.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1233.                 h->gb.size_in_bits = 8 * buf_size;

  1234. 

  1235.                 if (svq3_decode_slice_header(avctx))

  1236.                     return -1;

  1237. 

  1238.                 /* TODO: support s->mb_skip_run */

  1239.             }

  1240. 

  1241.             mb_type = svq3_get_ue_golomb(&h->gb);

  1242. 

  1243.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1244.                 mb_type += 8;

  1245.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1246.                 mb_type += 4;

  1247.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1248.                 av_log(h->avctx, AV_LOG_ERROR,

  1249.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1250.                 return -1;

  1251.             }

  1252. 

  1253.             if (mb_type != 0)

  1254.                 ff_h264_hl_decode_mb(h);

  1255. 

  1256.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1257.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1258.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1259.         }

  1260. 

  1261.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1262.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1263.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1264.     }

  1265. 

  1266.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1267.         ret = av_frame_ref(data, &s->cur_pic->f);

  1268.     else if (s->last_pic->f.data[0])

  1269.         ret = av_frame_ref(data, &s->last_pic->f);

  1270.     if (ret < 0)

  1271.         return ret;

  1272. 

  1273.     /* Do not output the last pic after seeking. */

  1274.     if (s->last_pic->f.data[0] || h->low_delay)

  1275.         *got_frame = 1;

  1276. 

  1277.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1278.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1279.     } else {

  1280.         av_frame_unref(&s->cur_pic->f);

  1281.     }

  1282. 

  1283.     return buf_size;

  1284. }

  1285. 

  1286. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1287. {

  1288.     SVQ3Context *s = avctx->priv_data;

  1289.     H264Context *h = &s->h;

  1290. 

  1291.     free_picture(avctx, s->cur_pic);

  1292.     free_picture(avctx, s->next_pic);

  1293.     free_picture(avctx, s->last_pic);

  1294.     av_freep(&s->cur_pic);

  1295.     av_freep(&s->next_pic);

  1296.     av_freep(&s->last_pic);

  1297. 

  1298.     av_frame_unref(&h->cur_pic.f);

  1299. 

  1300.     ff_h264_free_context(h);

  1301. 

  1302.     return 0;

  1303. }

  1304. 

  1305. AVCodec ff_svq3_decoder = {

  1306.     .name           = "svq3",

  1307.     .type           = AVMEDIA_TYPE_VIDEO,

  1308.     .id             = AV_CODEC_ID_SVQ3,

  1309.     .priv_data_size = sizeof(SVQ3Context),

  1310.     .init           = svq3_decode_init,

  1311.     .close          = svq3_decode_end,

  1312.     .decode         = svq3_decode_frame,

  1313.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1314.                       CODEC_CAP_DR1             |

  1315.                       CODEC_CAP_DELAY,

  1316.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1317.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1318.                                                      AV_PIX_FMT_NONE},

  1319. };