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.
39047 Commits
32 Branches
449MB
Tree: dc4d726bab
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'dc4d726bab'
${ noResults }
FFmpeg/libavcodec/svq3.c

1329 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 luma_dc_zigzag_scan[16] = {

  108.     0 * 16 + 0 * 64, 1 * 16 + 0 * 64, 2 * 16 + 0 * 64, 0 * 16 + 2 * 64,

  109.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

  110.     1 * 16 + 2 * 64, 2 * 16 + 2 * 64, 3 * 16 + 2 * 64, 0 * 16 + 3 * 64,

  111.     3 * 16 + 1 * 64, 1 * 16 + 3 * 64, 2 * 16 + 3 * 64, 3 * 16 + 3 * 64,

  112. };

  113. 

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

  115.     { 0, 0 },

  116.     { 1, 0 }, { 0, 1 },

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

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

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

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

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

  122.     { 4, 3 }, { 3, 4 },

  123.     { 4, 4 }

  124. };

  125. 

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

  127.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  128.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

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

  135.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  3,  4 }, { 1, 2,  3,  0,  4 },

  136.       { 2,  0,  1,  3,  4 }, { 2, 1,  3,  0,  4 }, { 2, 0,  4,  3,  1 } },

  137.     { { 0,  2, -1, -1, -1 }, { 0, 2,  4,  1,  3 }, { 1, 4,  2,  0,  3 },

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

  139. };

  140. 

  141. static const struct {

  142.     uint8_t run;

  143.     uint8_t level;

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

  145.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  146.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  147.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  148.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  149. };

  150. 

  151. static const uint32_t svq3_dequant_coeff[32] = {

  152.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  153.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  154.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  155.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  156. };

  157. 

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

  159. {

  160.     const int qmul = svq3_dequant_coeff[qp];

  161. #define stride 16

  162.     int i;

  163.     int temp[16];

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

  165. 

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

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

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

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

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

  171. 

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

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

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

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

  176.     }

  177. 

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

  179.         const int offset = x_offset[i];

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

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

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

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

  184. 

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

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

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

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

  189.     }

  190. }

  191. #undef stride

  192. 

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

  194.                         int stride, int qp, int dc)

  195. {

  196.     const int qmul = svq3_dequant_coeff[qp];

  197.     int i;

  198. 

  199.     if (dc) {

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

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

  202.         block[0] = 0;

  203.     }

  204. 

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

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

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

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

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

  210. 

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

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

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

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

  215.     }

  216. 

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

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

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

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

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

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

  223. 

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

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

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

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

  228.     }

  229. 

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

  231. }

  232. 

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

  234.                                     int index, const int type)

  235. {

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

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

  238. 

  239.     int run, level, limit;

  240.     unsigned vlc;

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

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

  243. 

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

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

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

  247.             vlc      = vlc + 1 >> 1;

  248. 

  249.             if (type == 3) {

  250.                 if (vlc < 3) {

  251.                     run   = 0;

  252.                     level = vlc;

  253.                 } else if (vlc < 4) {

  254.                     run   = 1;

  255.                     level = 1;

  256.                 } else {

  257.                     run   = vlc & 0x3;

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

  259.                 }

  260.             } else {

  261.                 if (vlc < 16) {

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

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

  264.                 } else if (intra) {

  265.                     run   = vlc & 0x7;

  266.                     level = (vlc >> 3) +

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

  268.                 } else {

  269.                     run   = vlc & 0xF;

  270.                     level = (vlc >> 4) +

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

  272.                 }

  273.             }

  274. 

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

  276.                 return -1;

  277. 

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

  279.         }

  280. 

  281.         if (type != 2) {

  282.             break;

  283.         }

  284.     }

  285. 

  286.     return 0;

  287. }

  288. 

  289. static inline void svq3_mc_dir_part(SVQ3Context *s,

  290.                                     int x, int y, int width, int height,

  291.                                     int mx, int my, int dxy,

  292.                                     int thirdpel, int dir, int avg)

  293. {

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

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

  296.     uint8_t *src, *dest;

  297.     int i, emu = 0;

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

  299. 

  300.     mx += x;

  301.     my += y;

  302. 

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

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

  305.         emu = 1;

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

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

  308.     }

  309. 

  310.     /* form component predictions */

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

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

  313. 

  314.     if (emu) {

  315.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src,

  316.                                  h->linesize, h->linesize,

  317.                                  width + 1, height + 1,

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

  319.         src = h->edge_emu_buffer;

  320.     }

  321.     if (thirdpel)

  322.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  324.                                                 width, height);

  325.     else

  326.         (avg ? s->hdsp.avg_pixels_tab

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

  328.                                                        height);

  329. 

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

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

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

  333.         width  = width  >> 1;

  334.         height = height >> 1;

  335.         blocksize++;

  336. 

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

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

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

  340. 

  341.             if (emu) {

  342.                 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src,

  343.                                          h->uvlinesize, h->uvlinesize,

  344.                                          width + 1, height + 1,

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

  346.                                          s->v_edge_pos >> 1);

  347.                 src = h->edge_emu_buffer;

  348.             }

  349.             if (thirdpel)

  350.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  352.                                                         h->uvlinesize,

  353.                                                         width, height);

  354.             else

  355.                 (avg ? s->hdsp.avg_pixels_tab

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

  357.                                                                h->uvlinesize,

  358.                                                                height);

  359.         }

  360.     }

  361. }

  362. 

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

  364.                               int dir, int avg)

  365. {

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

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

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

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

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

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

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

  373. 

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

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

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

  377.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  378.             int dxy;

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

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

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

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

  383. 

  384.             if (mode != PREDICT_MODE) {

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

  386.             } else {

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

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

  389. 

  390.                 if (dir == 0) {

  391.                     mx = mx * h->frame_num_offset /

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

  393.                     my = my * h->frame_num_offset /

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

  395.                 } else {

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

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

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

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

  400.                 }

  401.             }

  402. 

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

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

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

  406. 

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

  408.             if (mode == PREDICT_MODE) {

  409.                 dx = dy = 0;

  410.             } else {

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

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

  413. 

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

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

  416.                     return -1;

  417.                 }

  418.             }

  419. 

  420.             /* compute motion vector */

  421.             if (mode == THIRDPEL_MODE) {

  422.                 int fx, fy;

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

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

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

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

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

  428. 

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

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

  431.                 mx += mx;

  432.                 my += my;

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

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

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

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

  437. 

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

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

  440.                 mx *= 3;

  441.                 my *= 3;

  442.             } else {

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

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

  445. 

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

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

  448.                 mx *= 6;

  449.                 my *= 6;

  450.             }

  451. 

  452.             /* update mv_cache */

  453.             if (mode != PREDICT_MODE) {

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

  455. 

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

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

  458. 

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

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

  461.                 }

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

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

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

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

  466.             }

  467. 

  468.             /* write back motion vectors */

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

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

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

  472.         }

  473. 

  474.     return 0;

  475. }

  476. 

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

  478. {

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

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

  481.     int cbp = 0;

  482.     uint32_t vlc;

  483.     int8_t *top, *left;

  484.     const int mb_xy         = h->mb_xy;

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

  486. 

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

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

  489.     h->topright_samples_available = 0xFFFF;

  490. 

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

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

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

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

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

  496. 

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

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

  499.                                  0, 0, 0, 0, 1, 1);

  500. 

  501.             mb_type = MB_TYPE_SKIP;

  502.         } else {

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

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

  505.                 return -1;

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

  507.                 return -1;

  508. 

  509.             mb_type = MB_TYPE_16x16;

  510.         }

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

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

  513.             mode = THIRDPEL_MODE;

  514.         else if (s->halfpel_flag &&

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

  516.             mode = HALFPEL_MODE;

  517.         else

  518.             mode = FULLPEL_MODE;

  519. 

  520.         /* fill caches */

  521.         /* note ref_cache should contain here:

  522.          *  ????????

  523.          *  ???11111

  524.          *  N??11111

  525.          *  N??11111

  526.          *  N??11111

  527.          */

  528. 

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

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

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

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

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

  534.             } else {

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

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

  537.             }

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

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

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

  541.                        4 * 2 * sizeof(int16_t));

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

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

  544. 

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

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

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

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

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

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

  551.                 } else

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

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

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

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

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

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

  558.                 } else

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

  560.             } else

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

  562.                        PART_NOT_AVAILABLE, 8);

  563. 

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

  565.                 break;

  566.         }

  567. 

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

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

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

  571.                 return -1;

  572.         } else {        /* AV_PICTURE_TYPE_B */

  573.             if (mb_type != 2) {

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

  575.                     return -1;

  576.             } else {

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

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

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

  580.             }

  581.             if (mb_type != 1) {

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

  583.                     return -1;

  584.             } else {

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

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

  587.                            0, 4 * 2 * sizeof(int16_t));

  588.             }

  589.         }

  590. 

  591.         mb_type = MB_TYPE_16x16;

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

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

  594. 

  595.         if (mb_type == 8) {

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

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

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

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

  600.                     h->left_samples_available = 0x5F5F;

  601.             }

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

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

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

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

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

  607. 

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

  609.                     h->top_samples_available = 0x33FF;

  610.             }

  611. 

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

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

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

  615. 

  616.                 if (vlc >= 25) {

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

  618.                     return -1;

  619.                 }

  620. 

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

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

  623. 

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

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

  626. 

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

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

  629.                     return -1;

  630.                 }

  631.             }

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

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

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

  635.         }

  636. 

  637.         write_back_intra_pred_mode(h);

  638. 

  639.         if (mb_type == 8) {

  640.             ff_h264_check_intra4x4_pred_mode(h);

  641. 

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

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

  644.         } else {

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

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

  647. 

  648.             h->top_samples_available  = 0x33FF;

  649.             h->left_samples_available = 0x5F5F;

  650.         }

  651. 

  652.         mb_type = MB_TYPE_INTRA4x4;

  653.     } else {                      /* INTRA16x16 */

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

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

  656. 

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

  658.             av_log(h->avctx, AV_LOG_ERROR, "ff_h264_check_intra_pred_mode < 0\n");

  659.             return h->intra16x16_pred_mode;

  660.         }

  661. 

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

  663.         mb_type = MB_TYPE_INTRA16x16;

  664.     }

  665. 

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

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

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

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

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

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

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

  673.                        0, 4 * 2 * sizeof(int16_t));

  674.         }

  675.     }

  676.     if (!IS_INTRA4x4(mb_type)) {

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

  678.     }

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

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

  681.     }

  682. 

  683.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  687.             return -1;

  688.         }

  689. 

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

  691.                                 : golomb_to_inter_cbp[vlc];

  692.     }

  693.     if (IS_INTRA16x16(mb_type) ||

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

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

  696. 

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

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

  699.             return -1;

  700.         }

  701.     }

  702.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  706.             av_log(h->avctx, AV_LOG_ERROR,

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

  708.             return -1;

  709.         }

  710.     }

  711. 

  712.     if (cbp) {

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

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

  715. 

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

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

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

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

  720.                                  2 * (j & 2) + 4 * (i & 2))

  721.                               : (4 * i + j);

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

  723. 

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

  725.                         av_log(h->avctx, AV_LOG_ERROR,

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

  727.                         return -1;

  728.                     }

  729.                 }

  730.             }

  731. 

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

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

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

  735.                     av_log(h->avctx, AV_LOG_ERROR,

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

  737.                     return -1;

  738.                 }

  739. 

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

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

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

  743.                         k                                 = 16 * i + j;

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

  745. 

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

  747.                             av_log(h->avctx, AV_LOG_ERROR,

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

  749.                             return -1;

  750.                         }

  751.                     }

  752.                 }

  753.             }

  754.         }

  755.     }

  756. 

  757.     h->cbp                              = cbp;

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

  759. 

  760.     if (IS_INTRA(mb_type))

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

  762. 

  763.     return 0;

  764. }

  765. 

  766. static int svq3_decode_slice_header(AVCodecContext *avctx)

  767. {

  768.     SVQ3Context *s = avctx->priv_data;

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

  770.     const int mb_xy   = h->mb_xy;

  771.     int i, header;

  772.     unsigned slice_id;

  773. 

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

  775. 

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

  777.         /* TODO: what? */

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

  779.         return -1;

  780.     } else {

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

  782. 

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

  784.                               8 * show_bits(&h->gb, 8 * length) +

  785.                               8 * length;

  786. 

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

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

  789.             return -1;

  790.         }

  791. 

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

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

  794. 

  795.         if (s->watermark_key) {

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

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

  798.                     header ^ s->watermark_key);

  799.         }

  800.         if (length > 0) {

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

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

  803.         }

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

  805.     }

  806. 

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

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

  809.         return -1;

  810.     }

  811. 

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

  813. 

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

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

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

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

  818.     } else {

  819.         skip_bits1(&h->gb);

  820.         h->mb_skip_run = 0;

  821.     }

  822. 

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

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

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

  826. 

  827.     /* unknown fields */

  828.     skip_bits1(&h->gb);

  829. 

  830.     if (s->unknown_flag)

  831.         skip_bits1(&h->gb);

  832. 

  833.     skip_bits1(&h->gb);

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

  835. 

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

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

  838. 

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

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

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

  842.                -1, 4 * sizeof(int8_t));

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

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

  845.     }

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

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

  848.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  849. 

  850.         if (h->mb_x > 0)

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

  852.     }

  853. 

  854.     return 0;

  855. }

  856. 

  857. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  858. {

  859.     SVQ3Context *s = avctx->priv_data;

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

  861.     int m;

  862.     unsigned char *extradata;

  863.     unsigned char *extradata_end;

  864.     unsigned int size;

  865.     int marker_found = 0;

  866. 

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

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

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

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

  871.         av_freep(&s->cur_pic);

  872.         av_freep(&s->last_pic);

  873.         av_freep(&s->next_pic);

  874.         return AVERROR(ENOMEM);

  875.     }

  876. 

  877.     if (ff_h264_decode_init(avctx) < 0)

  878.         return -1;

  879. 

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

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

  882.     h->is_complex      = 1;

  883.     h->picture_structure = PICT_FRAME;

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

  885. 

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

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

  888. 

  889.     s->halfpel_flag  = 1;

  890.     s->thirdpel_flag = 1;

  891.     s->unknown_flag  = 0;

  892. 

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

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

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

  896.     if (extradata) {

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

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

  899.                 marker_found = 1;

  900.                 break;

  901.             }

  902.             extradata++;

  903.         }

  904.     }

  905. 

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

  907.     if (marker_found) {

  908.         GetBitContext gb;

  909.         int frame_size_code;

  910. 

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

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

  913.             return AVERROR_INVALIDDATA;

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

  915. 

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

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

  918.         switch (frame_size_code) {

  919.         case 0:

  920.             avctx->width  = 160;

  921.             avctx->height = 120;

  922.             break;

  923.         case 1:

  924.             avctx->width  = 128;

  925.             avctx->height =  96;

  926.             break;

  927.         case 2:

  928.             avctx->width  = 176;

  929.             avctx->height = 144;

  930.             break;

  931.         case 3:

  932.             avctx->width  = 352;

  933.             avctx->height = 288;

  934.             break;

  935.         case 4:

  936.             avctx->width  = 704;

  937.             avctx->height = 576;

  938.             break;

  939.         case 5:

  940.             avctx->width  = 240;

  941.             avctx->height = 180;

  942.             break;

  943.         case 6:

  944.             avctx->width  = 320;

  945.             avctx->height = 240;

  946.             break;

  947.         case 7:

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

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

  950.             break;

  951.         }

  952. 

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

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

  955. 

  956.         /* unknown fields */

  957.         skip_bits1(&gb);

  958.         skip_bits1(&gb);

  959.         skip_bits1(&gb);

  960.         skip_bits1(&gb);

  961. 

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

  963. 

  964.         /* unknown field */

  965.         skip_bits1(&gb);

  966. 

  967.         while (get_bits1(&gb))

  968.             skip_bits(&gb, 8);

  969. 

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

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

  972.         if (s->unknown_flag) {

  973. #if CONFIG_ZLIB

  974.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  975.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  976.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  979.             int u4                    = svq3_get_ue_golomb(&gb);

  980.             unsigned long buf_len     = watermark_width *

  981.                                         watermark_height * 4;

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

  983.             uint8_t *buf;

  984. 

  985.             if (watermark_height > 0 &&

  986.                 (uint64_t)watermark_width * 4 > UINT_MAX / watermark_height)

  987.                 return -1;

  988. 

  989.             buf = av_malloc(buf_len);

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

  991.                    watermark_width, watermark_height);

  992.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  996.                            size - offset) != Z_OK) {

  997.                 av_log(avctx, AV_LOG_ERROR,

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

  999.                 av_free(buf);

  1000.                 return -1;

  1001.             }

  1002.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1003.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1004.             av_log(avctx, AV_LOG_DEBUG,

  1005.                    "watermark key %#x\n", s->watermark_key);

  1006.             av_free(buf);

  1007. #else

  1008.             av_log(avctx, AV_LOG_ERROR,

  1009.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1010.             return -1;

  1011. #endif

  1012.         }

  1013.     }

  1014. 

  1015.     h->width  = avctx->width;

  1016.     h->height = avctx->height;

  1017.     h->mb_width  = (h->width + 15) / 16;

  1018.     h->mb_height = (h->height + 15) / 16;

  1019.     h->mb_stride = h->mb_width + 1;

  1020.     h->mb_num    = h->mb_width * h->mb_height;

  1021.     h->b_stride = 4 * h->mb_width;

  1022.     s->h_edge_pos = h->mb_width * 16;

  1023.     s->v_edge_pos = h->mb_height * 16;

  1024. 

  1025.     if (ff_h264_alloc_tables(h) < 0) {

  1026.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1027.         return AVERROR(ENOMEM);

  1028.     }

  1029. 

  1030.     return 0;

  1031. }

  1032. 

  1033. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1034. {

  1035.     int i;

  1036.     for (i = 0; i < 2; i++) {

  1037.         av_buffer_unref(&pic->motion_val_buf[i]);

  1038.         av_buffer_unref(&pic->ref_index_buf[i]);

  1039.     }

  1040.     av_buffer_unref(&pic->mb_type_buf);

  1041. 

  1042.     av_frame_unref(&pic->f);

  1043. }

  1044. 

  1045. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1046. {

  1047.     SVQ3Context *s = avctx->priv_data;

  1048.     H264Context *h = &s->h;

  1049.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1050.     const int mb_array_size = h->mb_stride * h->mb_height;

  1051.     const int b4_stride     = h->mb_width * 4 + 1;

  1052.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1053.     int ret;

  1054. 

  1055.     if (!pic->motion_val_buf[0]) {

  1056.         int i;

  1057. 

  1058.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1059.         if (!pic->mb_type_buf)

  1060.             return AVERROR(ENOMEM);

  1061.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1062. 

  1063.         for (i = 0; i < 2; i++) {

  1064.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1065.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1066.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1067.                 ret = AVERROR(ENOMEM);

  1068.                 goto fail;

  1069.             }

  1070. 

  1071.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1072.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1073.         }

  1074.     }

  1075.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1076. 

  1077.     ret = ff_get_buffer(avctx, &pic->f,

  1078.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1079.     if (ret < 0)

  1080.         goto fail;

  1081. 

  1082.     if (!h->edge_emu_buffer) {

  1083.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1084.         if (!h->edge_emu_buffer)

  1085.             return AVERROR(ENOMEM);

  1086.     }

  1087. 

  1088.     h->linesize   = pic->f.linesize[0];

  1089.     h->uvlinesize = pic->f.linesize[1];

  1090. 

  1091.     return 0;

  1092. fail:

  1093.     free_picture(avctx, pic);

  1094.     return ret;

  1095. }

  1096. 

  1097. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1098.                              int *got_frame, AVPacket *avpkt)

  1099. {

  1100.     const uint8_t *buf = avpkt->data;

  1101.     SVQ3Context *s     = avctx->priv_data;

  1102.     H264Context *h     = &s->h;

  1103.     int buf_size       = avpkt->size;

  1104.     int ret, m, i;

  1105. 

  1106.     /* special case for last picture */

  1107.     if (buf_size == 0) {

  1108.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1109.             ret = av_frame_ref(data, &s->next_pic->f);

  1110.             if (ret < 0)

  1111.                 return ret;

  1112.             s->last_frame_output = 1;

  1113.             *got_frame          = 1;

  1114.         }

  1115.         return 0;

  1116.     }

  1117. 

  1118.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1119. 

  1120.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1121. 

  1122.     if (svq3_decode_slice_header(avctx))

  1123.         return -1;

  1124. 

  1125.     h->pict_type = h->slice_type;

  1126. 

  1127.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1128.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1129. 

  1130.     av_frame_unref(&s->cur_pic->f);

  1131. 

  1132.     /* for skipping the frame */

  1133.     s->cur_pic->f.pict_type = h->pict_type;

  1134.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1135. 

  1136.     ret = get_buffer(avctx, s->cur_pic);

  1137.     if (ret < 0)

  1138.         return ret;

  1139. 

  1140.     h->cur_pic_ptr = s->cur_pic;

  1141.     av_frame_unref(&h->cur_pic.f);

  1142.     h->cur_pic     = *s->cur_pic;

  1143.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1144.     if (ret < 0)

  1145.         return ret;

  1146. 

  1147.     for (i = 0; i < 16; i++) {

  1148.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1149.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1150.     }

  1151.     for (i = 0; i < 16; i++) {

  1152.         h->block_offset[16 + i]      =

  1153.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1154.         h->block_offset[48 + 16 + i] =

  1155.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1156.     }

  1157. 

  1158.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1159.         if (!s->last_pic->f.data[0]) {

  1160.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1161.             ret = get_buffer(avctx, s->last_pic);

  1162.             if (ret < 0)

  1163.                 return ret;

  1164.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1165.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1166.                    s->last_pic->f.linesize[1]);

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

  1168.                    s->last_pic->f.linesize[2]);

  1169.         }

  1170. 

  1171.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1172.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1173.             ret = get_buffer(avctx, s->next_pic);

  1174.             if (ret < 0)

  1175.                 return ret;

  1176.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1177.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1178.                    s->next_pic->f.linesize[1]);

  1179.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1180.                    s->next_pic->f.linesize[2]);

  1181.         }

  1182.     }

  1183. 

  1184.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1185.         av_log(h->avctx, AV_LOG_DEBUG,

  1186.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1187.                av_get_picture_type_char(h->pict_type),

  1188.                s->halfpel_flag, s->thirdpel_flag,

  1189.                s->adaptive_quant, h->qscale, h->slice_num);

  1190. 

  1191.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1192.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1193.         avctx->skip_frame >= AVDISCARD_ALL)

  1194.         return 0;

  1195. 

  1196.     if (s->next_p_frame_damaged) {

  1197.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1198.             return 0;

  1199.         else

  1200.             s->next_p_frame_damaged = 0;

  1201.     }

  1202. 

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

  1204.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1205. 

  1206.         if (h->frame_num_offset < 0)

  1207.             h->frame_num_offset += 256;

  1208.         if (h->frame_num_offset == 0 ||

  1209.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1210.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1211.             return -1;

  1212.         }

  1213.     } else {

  1214.         h->prev_frame_num        = h->frame_num;

  1215.         h->frame_num             = h->slice_num;

  1216.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1217. 

  1218.         if (h->prev_frame_num_offset < 0)

  1219.             h->prev_frame_num_offset += 256;

  1220.     }

  1221. 

  1222.     for (m = 0; m < 2; m++) {

  1223.         int i;

  1224.         for (i = 0; i < 4; i++) {

  1225.             int j;

  1226.             for (j = -1; j < 4; j++)

  1227.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1228.             if (i < 3)

  1229.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1230.         }

  1231.     }

  1232. 

  1233.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1234.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1235.             unsigned mb_type;

  1236.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1237. 

  1238.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1239.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1240.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1241.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1242.                 h->gb.size_in_bits = 8 * buf_size;

  1243. 

  1244.                 if (svq3_decode_slice_header(avctx))

  1245.                     return -1;

  1246. 

  1247.                 /* TODO: support s->mb_skip_run */

  1248.             }

  1249. 

  1250.             mb_type = svq3_get_ue_golomb(&h->gb);

  1251. 

  1252.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1253.                 mb_type += 8;

  1254.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1255.                 mb_type += 4;

  1256.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1257.                 av_log(h->avctx, AV_LOG_ERROR,

  1258.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1259.                 return -1;

  1260.             }

  1261. 

  1262.             if (mb_type != 0)

  1263.                 ff_h264_hl_decode_mb(h);

  1264. 

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

  1266.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1267.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1268.         }

  1269. 

  1270.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1271.                            16 * h->mb_y, 16, h->picture_structure, 0,

  1272.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1273.     }

  1274. 

  1275.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1276.         ret = av_frame_ref(data, &s->cur_pic->f);

  1277.     else if (s->last_pic->f.data[0])

  1278.         ret = av_frame_ref(data, &s->last_pic->f);

  1279.     if (ret < 0)

  1280.         return ret;

  1281. 

  1282.     /* Do not output the last pic after seeking. */

  1283.     if (s->last_pic->f.data[0] || h->low_delay)

  1284.         *got_frame = 1;

  1285. 

  1286.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1287.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1288.     } else {

  1289.         av_frame_unref(&s->cur_pic->f);

  1290.     }

  1291. 

  1292.     return buf_size;

  1293. }

  1294. 

  1295. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1296. {

  1297.     SVQ3Context *s = avctx->priv_data;

  1298.     H264Context *h = &s->h;

  1299. 

  1300.     free_picture(avctx, s->cur_pic);

  1301.     free_picture(avctx, s->next_pic);

  1302.     free_picture(avctx, s->last_pic);

  1303.     av_freep(&s->cur_pic);

  1304.     av_freep(&s->next_pic);

  1305.     av_freep(&s->last_pic);

  1306. 

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

  1308. 

  1309.     ff_h264_free_context(h);

  1310. 

  1311.     return 0;

  1312. }

  1313. 

  1314. AVCodec ff_svq3_decoder = {

  1315.     .name           = "svq3",

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

  1317.     .type           = AVMEDIA_TYPE_VIDEO,

  1318.     .id             = AV_CODEC_ID_SVQ3,

  1319.     .priv_data_size = sizeof(SVQ3Context),

  1320.     .init           = svq3_decode_init,

  1321.     .close          = svq3_decode_end,

  1322.     .decode         = svq3_decode_frame,

  1323.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1324.                       CODEC_CAP_DR1             |

  1325.                       CODEC_CAP_DELAY,

  1326.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1327.                                                      AV_PIX_FMT_NONE},

  1328. };