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.
37957 Commits
32 Branches
498MB
Tree: a90cff137b
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'a90cff137b'
${ noResults }
FFmpeg/libavcodec/svq3.c

1319 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. 

  56. #if CONFIG_ZLIB

  57. #include <zlib.h>

  58. #endif

  59. 

  60. #include "svq1.h"

  61. #include "svq3.h"

  62. 

  63. /**

  64.  * @file

  65.  * svq3 decoder.

  66.  */

  67. 

  68. typedef struct {

  69.     H264Context h;

  70.     HpelDSPContext hdsp;

  71.     Picture *cur_pic;

  72.     Picture *next_pic;

  73.     Picture *last_pic;

  74.     int halfpel_flag;

  75.     int thirdpel_flag;

  76.     int unknown_flag;

  77.     int next_slice_index;

  78.     uint32_t watermark_key;

  79.     int adaptive_quant;

  80.     int next_p_frame_damaged;

  81.     int h_edge_pos;

  82.     int v_edge_pos;

  83.     int last_frame_output;

  84. } SVQ3Context;

  85. 

  86. #define FULLPEL_MODE  1

  87. #define HALFPEL_MODE  2

  88. #define THIRDPEL_MODE 3

  89. #define PREDICT_MODE  4

  90. 

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

  92.  * o-->o-->o   o

  93.  *         |  /|

  94.  * o   o   o / o

  95.  * | / |   |/  |

  96.  * o   o   o   o

  97.  *   /

  98.  * o-->o-->o-->o

  99.  */

  100. static const uint8_t svq3_scan[16] = {

  101.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  102.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

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

  104.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  105. };

  106. 

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

  108.     { 0, 0 },

  109.     { 1, 0 }, { 0, 1 },

  110.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  111.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

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

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

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

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

  116.     { 4, 4 }

  117. };

  118. 

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

  120.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

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

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

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

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

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

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

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

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

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

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

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

  132. };

  133. 

  134. static const struct {

  135.     uint8_t run;

  136.     uint8_t level;

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

  138.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  139.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  140.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  141.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  142. };

  143. 

  144. static const uint32_t svq3_dequant_coeff[32] = {

  145.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  146.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  147.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  148.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  149. };

  150. 

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

  152. {

  153.     const int qmul = svq3_dequant_coeff[qp];

  154. #define stride 16

  155.     int i;

  156.     int temp[16];

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

  158. 

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

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

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

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

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

  164. 

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

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

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

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

  169.     }

  170. 

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

  172.         const int offset = x_offset[i];

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

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

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

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

  177. 

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

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

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

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

  182.     }

  183. }

  184. #undef stride

  185. 

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

  187.                         int stride, int qp, int dc)

  188. {

  189.     const int qmul = svq3_dequant_coeff[qp];

  190.     int i;

  191. 

  192.     if (dc) {

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

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

  195.         block[0] = 0;

  196.     }

  197. 

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

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

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

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

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

  203. 

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

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

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

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

  208.     }

  209. 

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

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

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

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

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

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

  216. 

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

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

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

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

  221.     }

  222. 

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

  224. }

  225. 

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

  227.                                     int index, const int type)

  228. {

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

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

  231. 

  232.     int run, level, limit;

  233.     unsigned vlc;

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

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

  236. 

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

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

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

  240.             vlc      = vlc + 1 >> 1;

  241. 

  242.             if (type == 3) {

  243.                 if (vlc < 3) {

  244.                     run   = 0;

  245.                     level = vlc;

  246.                 } else if (vlc < 4) {

  247.                     run   = 1;

  248.                     level = 1;

  249.                 } else {

  250.                     run   = vlc & 0x3;

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

  252.                 }

  253.             } else {

  254.                 if (vlc < 16) {

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

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

  257.                 } else if (intra) {

  258.                     run   = vlc & 0x7;

  259.                     level = (vlc >> 3) +

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

  261.                 } else {

  262.                     run   = vlc & 0xF;

  263.                     level = (vlc >> 4) +

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

  265.                 }

  266.             }

  267. 

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

  269.                 return -1;

  270. 

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

  272.         }

  273. 

  274.         if (type != 2) {

  275.             break;

  276.         }

  277.     }

  278. 

  279.     return 0;

  280. }

  281. 

  282. static inline void svq3_mc_dir_part(SVQ3Context *s,

  283.                                     int x, int y, int width, int height,

  284.                                     int mx, int my, int dxy,

  285.                                     int thirdpel, int dir, int avg)

  286. {

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

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

  289.     uint8_t *src, *dest;

  290.     int i, emu = 0;

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

  292. 

  293.     mx += x;

  294.     my += y;

  295. 

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

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

  298.         emu = 1;

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

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

  301.     }

  302. 

  303.     /* form component predictions */

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

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

  306. 

  307.     if (emu) {

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

  309.                                  width + 1, height + 1,

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

  311.         src = h->edge_emu_buffer;

  312.     }

  313.     if (thirdpel)

  314.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  316.                                                 width, height);

  317.     else

  318.         (avg ? s->hdsp.avg_pixels_tab

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

  320.                                                        height);

  321. 

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

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

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

  325.         width  = width  >> 1;

  326.         height = height >> 1;

  327.         blocksize++;

  328. 

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

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

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

  332. 

  333.             if (emu) {

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

  335.                                          width + 1, height + 1,

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

  337.                                          s->v_edge_pos >> 1);

  338.                 src = h->edge_emu_buffer;

  339.             }

  340.             if (thirdpel)

  341.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  343.                                                         h->uvlinesize,

  344.                                                         width, height);

  345.             else

  346.                 (avg ? s->hdsp.avg_pixels_tab

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

  348.                                                                h->uvlinesize,

  349.                                                                height);

  350.         }

  351.     }

  352. }

  353. 

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

  355.                               int dir, int avg)

  356. {

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

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

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

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

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

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

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

  364. 

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

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

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

  368.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  369.             int dxy;

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

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

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

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

  374. 

  375.             if (mode != PREDICT_MODE) {

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

  377.             } else {

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

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

  380. 

  381.                 if (dir == 0) {

  382.                     mx = mx * h->frame_num_offset /

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

  384.                     my = my * h->frame_num_offset /

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

  386.                 } else {

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

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

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

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

  391.                 }

  392.             }

  393. 

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

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

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

  397. 

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

  399.             if (mode == PREDICT_MODE) {

  400.                 dx = dy = 0;

  401.             } else {

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

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

  404. 

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

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

  407.                     return -1;

  408.                 }

  409.             }

  410. 

  411.             /* compute motion vector */

  412.             if (mode == THIRDPEL_MODE) {

  413.                 int fx, fy;

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

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

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

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

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

  419. 

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

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

  422.                 mx += mx;

  423.                 my += my;

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

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

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

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

  428. 

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

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

  431.                 mx *= 3;

  432.                 my *= 3;

  433.             } else {

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

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

  436. 

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

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

  439.                 mx *= 6;

  440.                 my *= 6;

  441.             }

  442. 

  443.             /* update mv_cache */

  444.             if (mode != PREDICT_MODE) {

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

  446. 

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

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

  449. 

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

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

  452.                 }

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

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

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

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

  457.             }

  458. 

  459.             /* write back motion vectors */

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

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

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

  463.         }

  464. 

  465.     return 0;

  466. }

  467. 

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

  469. {

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

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

  472.     int cbp = 0;

  473.     uint32_t vlc;

  474.     int8_t *top, *left;

  475.     const int mb_xy         = h->mb_xy;

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

  477. 

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

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

  480.     h->topright_samples_available = 0xFFFF;

  481. 

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

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

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

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

  486.                              0, 0, 0, 0, 0, 0);

  487. 

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

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

  490.                                  0, 0, 0, 0, 1, 1);

  491. 

  492.             mb_type = MB_TYPE_SKIP;

  493.         } else {

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

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

  496.                 return -1;

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

  498.                 return -1;

  499. 

  500.             mb_type = MB_TYPE_16x16;

  501.         }

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

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

  504.             mode = THIRDPEL_MODE;

  505.         else if (s->halfpel_flag &&

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

  507.             mode = HALFPEL_MODE;

  508.         else

  509.             mode = FULLPEL_MODE;

  510. 

  511.         /* fill caches */

  512.         /* note ref_cache should contain here:

  513.          *  ????????

  514.          *  ???11111

  515.          *  N??11111

  516.          *  N??11111

  517.          *  N??11111

  518.          */

  519. 

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

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

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

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

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

  525.             } else {

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

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

  528.             }

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

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

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

  532.                        4 * 2 * sizeof(int16_t));

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

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

  535. 

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

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

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

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

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

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

  542.                 } else

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

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

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

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

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

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

  549.                 } else

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

  551.             } else

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

  553.                        PART_NOT_AVAILABLE, 8);

  554. 

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

  556.                 break;

  557.         }

  558. 

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

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

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

  562.                 return -1;

  563.         } else {        /* AV_PICTURE_TYPE_B */

  564.             if (mb_type != 2) {

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

  566.                     return -1;

  567.             } else {

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

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

  570.                            0, 4 * 2 * sizeof(int16_t));

  571.             }

  572.             if (mb_type != 1) {

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

  574.                     return -1;

  575.             } else {

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

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

  578.                            0, 4 * 2 * sizeof(int16_t));

  579.             }

  580.         }

  581. 

  582.         mb_type = MB_TYPE_16x16;

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

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

  585. 

  586.         if (mb_type == 8) {

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

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

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

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

  591.                     h->left_samples_available = 0x5F5F;

  592.             }

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

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

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

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

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

  598. 

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

  600.                     h->top_samples_available = 0x33FF;

  601.             }

  602. 

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

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

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

  606. 

  607.                 if (vlc >= 25) {

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

  609.                     return -1;

  610.                 }

  611. 

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

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

  614. 

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

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

  617. 

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

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

  620.                     return -1;

  621.                 }

  622.             }

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

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

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

  626.         }

  627. 

  628.         write_back_intra_pred_mode(h);

  629. 

  630.         if (mb_type == 8) {

  631.             ff_h264_check_intra4x4_pred_mode(h);

  632. 

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

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

  635.         } else {

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

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

  638. 

  639.             h->top_samples_available  = 0x33FF;

  640.             h->left_samples_available = 0x5F5F;

  641.         }

  642. 

  643.         mb_type = MB_TYPE_INTRA4x4;

  644.     } else {                      /* INTRA16x16 */

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

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

  647. 

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

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

  650.             return -1;

  651.         }

  652. 

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

  654.         mb_type = MB_TYPE_INTRA16x16;

  655.     }

  656. 

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

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

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

  660.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  664.                        0, 4 * 2 * sizeof(int16_t));

  665.         }

  666.     }

  667.     if (!IS_INTRA4x4(mb_type)) {

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

  669.     }

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

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

  672.     }

  673. 

  674.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  678.             return -1;

  679.         }

  680. 

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

  682.                                 : golomb_to_inter_cbp[vlc];

  683.     }

  684.     if (IS_INTRA16x16(mb_type) ||

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

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

  687. 

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

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

  690.             return -1;

  691.         }

  692.     }

  693.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  697.             av_log(h->avctx, AV_LOG_ERROR,

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

  699.             return -1;

  700.         }

  701.     }

  702. 

  703.     if (cbp) {

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

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

  706. 

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

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

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

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

  711.                                  2 * (j & 2) + 4 * (i & 2))

  712.                               : (4 * i + j);

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

  714. 

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

  716.                         av_log(h->avctx, AV_LOG_ERROR,

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

  718.                         return -1;

  719.                     }

  720.                 }

  721.             }

  722. 

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

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

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

  726.                     av_log(h->avctx, AV_LOG_ERROR,

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

  728.                     return -1;

  729.                 }

  730. 

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

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

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

  734.                         k                                 = 16 * i + j;

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

  736. 

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

  738.                             av_log(h->avctx, AV_LOG_ERROR,

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

  740.                             return -1;

  741.                         }

  742.                     }

  743.                 }

  744.             }

  745.         }

  746.     }

  747. 

  748.     h->cbp                              = cbp;

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

  750. 

  751.     if (IS_INTRA(mb_type))

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

  753. 

  754.     return 0;

  755. }

  756. 

  757. static int svq3_decode_slice_header(AVCodecContext *avctx)

  758. {

  759.     SVQ3Context *s = avctx->priv_data;

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

  761.     const int mb_xy   = h->mb_xy;

  762.     int i, header;

  763.     unsigned slice_id;

  764. 

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

  766. 

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

  768.         /* TODO: what? */

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

  770.         return -1;

  771.     } else {

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

  773. 

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

  775.                               8 * show_bits(&h->gb, 8 * length) +

  776.                               8 * length;

  777. 

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

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

  780.             return -1;

  781.         }

  782. 

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

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

  785. 

  786.         if (s->watermark_key) {

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

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

  789.                     header ^ s->watermark_key);

  790.         }

  791.         if (length > 0) {

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

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

  794.         }

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

  796.     }

  797. 

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

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

  800.         return -1;

  801.     }

  802. 

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

  804. 

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

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

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

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

  809.     } else {

  810.         skip_bits1(&h->gb);

  811.         h->mb_skip_run = 0;

  812.     }

  813. 

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

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

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

  817. 

  818.     /* unknown fields */

  819.     skip_bits1(&h->gb);

  820. 

  821.     if (s->unknown_flag)

  822.         skip_bits1(&h->gb);

  823. 

  824.     skip_bits1(&h->gb);

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

  826. 

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

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

  829. 

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

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

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

  833.                -1, 4 * sizeof(int8_t));

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

  835.                -1, 8 * sizeof(int8_t) * h->mb_x);

  836.     }

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

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

  839.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  840. 

  841.         if (h->mb_x > 0)

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

  843.     }

  844. 

  845.     return 0;

  846. }

  847. 

  848. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  849. {

  850.     SVQ3Context *s = avctx->priv_data;

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

  852.     int m;

  853.     unsigned char *extradata;

  854.     unsigned char *extradata_end;

  855.     unsigned int size;

  856.     int marker_found = 0;

  857. 

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

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

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

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

  862.         av_freep(&s->cur_pic);

  863.         av_freep(&s->last_pic);

  864.         av_freep(&s->next_pic);

  865.         return AVERROR(ENOMEM);

  866.     }

  867. 

  868.     if (ff_h264_decode_init(avctx) < 0)

  869.         return -1;

  870. 

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

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

  873.     h->is_complex      = 1;

  874.     h->picture_structure = PICT_FRAME;

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

  876. 

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

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

  879. 

  880.     s->halfpel_flag  = 1;

  881.     s->thirdpel_flag = 1;

  882.     s->unknown_flag  = 0;

  883. 

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

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

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

  887.     if (extradata) {

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

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

  890.                 marker_found = 1;

  891.                 break;

  892.             }

  893.             extradata++;

  894.         }

  895.     }

  896. 

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

  898.     if (marker_found) {

  899.         GetBitContext gb;

  900.         int frame_size_code;

  901. 

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

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

  904.             return AVERROR_INVALIDDATA;

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

  906. 

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

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

  909.         switch (frame_size_code) {

  910.         case 0:

  911.             avctx->width  = 160;

  912.             avctx->height = 120;

  913.             break;

  914.         case 1:

  915.             avctx->width  = 128;

  916.             avctx->height =  96;

  917.             break;

  918.         case 2:

  919.             avctx->width  = 176;

  920.             avctx->height = 144;

  921.             break;

  922.         case 3:

  923.             avctx->width  = 352;

  924.             avctx->height = 288;

  925.             break;

  926.         case 4:

  927.             avctx->width  = 704;

  928.             avctx->height = 576;

  929.             break;

  930.         case 5:

  931.             avctx->width  = 240;

  932.             avctx->height = 180;

  933.             break;

  934.         case 6:

  935.             avctx->width  = 320;

  936.             avctx->height = 240;

  937.             break;

  938.         case 7:

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

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

  941.             break;

  942.         }

  943. 

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

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

  946. 

  947.         /* unknown fields */

  948.         skip_bits1(&gb);

  949.         skip_bits1(&gb);

  950.         skip_bits1(&gb);

  951.         skip_bits1(&gb);

  952. 

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

  954. 

  955.         /* unknown field */

  956.         skip_bits1(&gb);

  957. 

  958.         while (get_bits1(&gb))

  959.             skip_bits(&gb, 8);

  960. 

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

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

  963.         if (s->unknown_flag) {

  964. #if CONFIG_ZLIB

  965.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  966.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  967.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  970.             int u4                    = svq3_get_ue_golomb(&gb);

  971.             unsigned long buf_len     = watermark_width *

  972.                                         watermark_height * 4;

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

  974.             uint8_t *buf;

  975. 

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

  977.                 return -1;

  978. 

  979.             buf = av_malloc(buf_len);

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

  981.                    watermark_width, watermark_height);

  982.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  986.                            size - offset) != Z_OK) {

  987.                 av_log(avctx, AV_LOG_ERROR,

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

  989.                 av_free(buf);

  990.                 return -1;

  991.             }

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

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

  994.             av_log(avctx, AV_LOG_DEBUG,

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

  996.             av_free(buf);

  997. #else

  998.             av_log(avctx, AV_LOG_ERROR,

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

  1000.             return -1;

  1001. #endif

  1002.         }

  1003.     }

  1004. 

  1005.     h->width  = avctx->width;

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

  1007.     h->mb_width  = (h->width + 15) / 16;

  1008.     h->mb_height = (h->height + 15) / 16;

  1009.     h->mb_stride = h->mb_width + 1;

  1010.     h->mb_num    = h->mb_width * h->mb_height;

  1011.     h->b_stride = 4 * h->mb_width;

  1012.     s->h_edge_pos = h->mb_width * 16;

  1013.     s->v_edge_pos = h->mb_height * 16;

  1014. 

  1015.     if (ff_h264_alloc_tables(h) < 0) {

  1016.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1017.         return AVERROR(ENOMEM);

  1018.     }

  1019. 

  1020.     return 0;

  1021. }

  1022. 

  1023. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1024. {

  1025.     int i;

  1026.     for (i = 0; i < 2; i++) {

  1027.         av_buffer_unref(&pic->motion_val_buf[i]);

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

  1029.     }

  1030.     av_buffer_unref(&pic->mb_type_buf);

  1031. 

  1032.     av_frame_unref(&pic->f);

  1033. }

  1034. 

  1035. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1036. {

  1037.     SVQ3Context *s = avctx->priv_data;

  1038.     H264Context *h = &s->h;

  1039.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1040.     const int mb_array_size = h->mb_stride * h->mb_height;

  1041.     const int b4_stride     = h->mb_width * 4 + 1;

  1042.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1043.     int ret;

  1044. 

  1045.     if (!pic->motion_val_buf[0]) {

  1046.         int i;

  1047. 

  1048.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1049.         if (!pic->mb_type_buf)

  1050.             return AVERROR(ENOMEM);

  1051.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1052. 

  1053.         for (i = 0; i < 2; i++) {

  1054.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1055.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1056.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1057.                 ret = AVERROR(ENOMEM);

  1058.                 goto fail;

  1059.             }

  1060. 

  1061.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1062.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1063.         }

  1064.     }

  1065.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1066. 

  1067.     ret = ff_get_buffer(avctx, &pic->f,

  1068.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1069.     if (ret < 0)

  1070.         goto fail;

  1071. 

  1072.     if (!h->edge_emu_buffer) {

  1073.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1074.         if (!h->edge_emu_buffer)

  1075.             return AVERROR(ENOMEM);

  1076.     }

  1077. 

  1078.     h->linesize   = pic->f.linesize[0];

  1079.     h->uvlinesize = pic->f.linesize[1];

  1080. 

  1081.     return 0;

  1082. fail:

  1083.     free_picture(avctx, pic);

  1084.     return ret;

  1085. }

  1086. 

  1087. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1088.                              int *got_frame, AVPacket *avpkt)

  1089. {

  1090.     const uint8_t *buf = avpkt->data;

  1091.     SVQ3Context *s     = avctx->priv_data;

  1092.     H264Context *h     = &s->h;

  1093.     int buf_size       = avpkt->size;

  1094.     int ret, m, i;

  1095. 

  1096.     /* special case for last picture */

  1097.     if (buf_size == 0) {

  1098.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1099.             ret = av_frame_ref(data, &s->next_pic->f);

  1100.             if (ret < 0)

  1101.                 return ret;

  1102.             s->last_frame_output = 1;

  1103.             *got_frame          = 1;

  1104.         }

  1105.         return 0;

  1106.     }

  1107. 

  1108.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1109. 

  1110.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1111. 

  1112.     if (svq3_decode_slice_header(avctx))

  1113.         return -1;

  1114. 

  1115.     h->pict_type = h->slice_type;

  1116. 

  1117.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1118.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1119. 

  1120.     av_frame_unref(&s->cur_pic->f);

  1121. 

  1122.     /* for skipping the frame */

  1123.     s->cur_pic->f.pict_type = h->pict_type;

  1124.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1125. 

  1126.     ret = get_buffer(avctx, s->cur_pic);

  1127.     if (ret < 0)

  1128.         return ret;

  1129. 

  1130.     h->cur_pic_ptr = s->cur_pic;

  1131.     av_frame_unref(&h->cur_pic.f);

  1132.     h->cur_pic     = *s->cur_pic;

  1133.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1134.     if (ret < 0)

  1135.         return ret;

  1136. 

  1137.     for (i = 0; i < 16; i++) {

  1138.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1139.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1140.     }

  1141.     for (i = 0; i < 16; i++) {

  1142.         h->block_offset[16 + i]      =

  1143.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1144.         h->block_offset[48 + 16 + i] =

  1145.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1146.     }

  1147. 

  1148.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1149.         if (!s->last_pic->f.data[0]) {

  1150.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1151.             ret = get_buffer(avctx, s->last_pic);

  1152.             if (ret < 0)

  1153.                 return ret;

  1154.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1155.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1156.                    s->last_pic->f.linesize[1]);

  1157.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1158.                    s->last_pic->f.linesize[2]);

  1159.         }

  1160. 

  1161.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1162.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1163.             ret = get_buffer(avctx, s->next_pic);

  1164.             if (ret < 0)

  1165.                 return ret;

  1166.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1167.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1168.                    s->next_pic->f.linesize[1]);

  1169.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1170.                    s->next_pic->f.linesize[2]);

  1171.         }

  1172.     }

  1173. 

  1174.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1175.         av_log(h->avctx, AV_LOG_DEBUG,

  1176.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1177.                av_get_picture_type_char(h->pict_type),

  1178.                s->halfpel_flag, s->thirdpel_flag,

  1179.                s->adaptive_quant, h->qscale, h->slice_num);

  1180. 

  1181.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1182.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1183.         avctx->skip_frame >= AVDISCARD_ALL)

  1184.         return 0;

  1185. 

  1186.     if (s->next_p_frame_damaged) {

  1187.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1188.             return 0;

  1189.         else

  1190.             s->next_p_frame_damaged = 0;

  1191.     }

  1192. 

  1193.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1194.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1195. 

  1196.         if (h->frame_num_offset < 0)

  1197.             h->frame_num_offset += 256;

  1198.         if (h->frame_num_offset == 0 ||

  1199.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1200.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1201.             return -1;

  1202.         }

  1203.     } else {

  1204.         h->prev_frame_num        = h->frame_num;

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

  1206.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1207. 

  1208.         if (h->prev_frame_num_offset < 0)

  1209.             h->prev_frame_num_offset += 256;

  1210.     }

  1211. 

  1212.     for (m = 0; m < 2; m++) {

  1213.         int i;

  1214.         for (i = 0; i < 4; i++) {

  1215.             int j;

  1216.             for (j = -1; j < 4; j++)

  1217.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1218.             if (i < 3)

  1219.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1220.         }

  1221.     }

  1222. 

  1223.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1224.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1225.             unsigned mb_type;

  1226.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1227. 

  1228.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1229.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1230.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1231.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1232.                 h->gb.size_in_bits = 8 * buf_size;

  1233. 

  1234.                 if (svq3_decode_slice_header(avctx))

  1235.                     return -1;

  1236. 

  1237.                 /* TODO: support s->mb_skip_run */

  1238.             }

  1239. 

  1240.             mb_type = svq3_get_ue_golomb(&h->gb);

  1241. 

  1242.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1243.                 mb_type += 8;

  1244.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1245.                 mb_type += 4;

  1246.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1247.                 av_log(h->avctx, AV_LOG_ERROR,

  1248.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1249.                 return -1;

  1250.             }

  1251. 

  1252.             if (mb_type != 0)

  1253.                 ff_h264_hl_decode_mb(h);

  1254. 

  1255.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1256.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1257.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1258.         }

  1259. 

  1260.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1261.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1262.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1263.     }

  1264. 

  1265.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1266.         ret = av_frame_ref(data, &s->cur_pic->f);

  1267.     else if (s->last_pic->f.data[0])

  1268.         ret = av_frame_ref(data, &s->last_pic->f);

  1269.     if (ret < 0)

  1270.         return ret;

  1271. 

  1272.     /* Do not output the last pic after seeking. */

  1273.     if (s->last_pic->f.data[0] || h->low_delay)

  1274.         *got_frame = 1;

  1275. 

  1276.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1277.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1278.     } else {

  1279.         av_frame_unref(&s->cur_pic->f);

  1280.     }

  1281. 

  1282.     return buf_size;

  1283. }

  1284. 

  1285. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1286. {

  1287.     SVQ3Context *s = avctx->priv_data;

  1288.     H264Context *h = &s->h;

  1289. 

  1290.     free_picture(avctx, s->cur_pic);

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

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

  1293.     av_freep(&s->cur_pic);

  1294.     av_freep(&s->next_pic);

  1295.     av_freep(&s->last_pic);

  1296. 

  1297.     av_frame_unref(&h->cur_pic.f);

  1298. 

  1299.     ff_h264_free_context(h);

  1300. 

  1301.     return 0;

  1302. }

  1303. 

  1304. AVCodec ff_svq3_decoder = {

  1305.     .name           = "svq3",

  1306.     .type           = AVMEDIA_TYPE_VIDEO,

  1307.     .id             = AV_CODEC_ID_SVQ3,

  1308.     .priv_data_size = sizeof(SVQ3Context),

  1309.     .init           = svq3_decode_init,

  1310.     .close          = svq3_decode_end,

  1311.     .decode         = svq3_decode_frame,

  1312.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1313.                       CODEC_CAP_DR1             |

  1314.                       CODEC_CAP_DELAY,

  1315.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1316.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1317.                                                      AV_PIX_FMT_NONE},

  1318. };