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.
59070 Commits
32 Branches
618MB
Tree: c90cdf4b64
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c90cdf4b64'
${ noResults }
FFmpeg/libavcodec/svq3.c

1362 lines
48KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The FFmpeg Project

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

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

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

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

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

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

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

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

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

  31.  *

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

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

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

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

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

  37.  *

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

  39.  * correctly decodes this file:

  40.  *  http://samples.mplayerhq.hu/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov

  41.  */

  42. 

  43. #include "libavutil/attributes.h"

  44. #include "internal.h"

  45. #include "avcodec.h"

  46. #include "mpegvideo.h"

  47. #include "h264.h"

  48. 

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

  50. 

  51. #include "h264_mvpred.h"

  52. #include "golomb.h"

  53. #include "hpeldsp.h"

  54. #include "rectangle.h"

  55. #include "vdpau_internal.h"

  56. 

  57. #if CONFIG_ZLIB

  58. #include <zlib.h>

  59. #endif

  60. 

  61. #include "svq1.h"

  62. #include "svq3.h"

  63. 

  64. /**

  65.  * @file

  66.  * svq3 decoder.

  67.  */

  68. 

  69. typedef struct {

  70.     H264Context h;

  71.     HpelDSPContext hdsp;

  72.     Picture *cur_pic;

  73.     Picture *next_pic;

  74.     Picture *last_pic;

  75.     int halfpel_flag;

  76.     int thirdpel_flag;

  77.     int unknown_flag;

  78.     int next_slice_index;

  79.     uint32_t watermark_key;

  80.     uint8_t *buf;

  81.     int buf_size;

  82.     int adaptive_quant;

  83.     int next_p_frame_damaged;

  84.     int h_edge_pos;

  85.     int v_edge_pos;

  86.     int last_frame_output;

  87. } SVQ3Context;

  88. 

  89. #define FULLPEL_MODE  1

  90. #define HALFPEL_MODE  2

  91. #define THIRDPEL_MODE 3

  92. #define PREDICT_MODE  4

  93. 

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

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

  96.  *         |  /|

  97.  * o   o   o / o

  98.  * | / |   |/  |

  99.  * o   o   o   o

  100.  *   /

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

  102.  */

  103. static const uint8_t svq3_scan[16] = {

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

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

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

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

  108. };

  109. 

  110. static const uint8_t luma_dc_zigzag_scan[16] = {

  111.     0 * 16 + 0 * 64, 1 * 16 + 0 * 64, 2 * 16 + 0 * 64, 0 * 16 + 2 * 64,

  112.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

  113.     1 * 16 + 2 * 64, 2 * 16 + 2 * 64, 3 * 16 + 2 * 64, 0 * 16 + 3 * 64,

  114.     3 * 16 + 1 * 64, 1 * 16 + 3 * 64, 2 * 16 + 3 * 64, 3 * 16 + 3 * 64,

  115. };

  116. 

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

  118.     { 0, 0 },

  119.     { 1, 0 }, { 0, 1 },

  120.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  121.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

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

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

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

  125.     { 4, 3 }, { 3, 4 },

  126.     { 4, 4 }

  127. };

  128. 

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

  130.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  131.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

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

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

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

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

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

  142. };

  143. 

  144. static const struct {

  145.     uint8_t run;

  146.     uint8_t level;

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

  148.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  149.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  150.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  151.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  152. };

  153. 

  154. static const uint32_t svq3_dequant_coeff[32] = {

  155.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  156.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  157.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  158.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  159. };

  160. 

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

  162. {

  163.     const int qmul = svq3_dequant_coeff[qp];

  164. #define stride 16

  165.     int i;

  166.     int temp[16];

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

  168. 

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

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

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

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

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

  174. 

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

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

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

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

  179.     }

  180. 

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

  182.         const int offset = x_offset[i];

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

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

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

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

  187. 

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

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

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

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

  192.     }

  193. }

  194. #undef stride

  195. 

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

  197.                         int stride, int qp, int dc)

  198. {

  199.     const int qmul = svq3_dequant_coeff[qp];

  200.     int i;

  201. 

  202.     if (dc) {

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

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

  205.         block[0] = 0;

  206.     }

  207. 

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

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

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

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

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

  213. 

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

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

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

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

  218.     }

  219. 

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

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

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

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

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

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

  226. 

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

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

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

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

  231.     }

  232. 

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

  234. }

  235. 

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

  237.                                     int index, const int type)

  238. {

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

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

  241. 

  242.     int run, level, sign, limit;

  243.     unsigned vlc;

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

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

  246. 

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

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

  249.             if ((int32_t)vlc < 0)

  250.                 return -1;

  251. 

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

  253.             vlc      = vlc + 1 >> 1;

  254. 

  255.             if (type == 3) {

  256.                 if (vlc < 3) {

  257.                     run   = 0;

  258.                     level = vlc;

  259.                 } else if (vlc < 4) {

  260.                     run   = 1;

  261.                     level = 1;

  262.                 } else {

  263.                     run   = vlc & 0x3;

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

  265.                 }

  266.             } else {

  267.                 if (vlc < 16U) {

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

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

  270.                 } else if (intra) {

  271.                     run   = vlc & 0x7;

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

  273.                 } else {

  274.                     run   = vlc & 0xF;

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

  276.                 }

  277.             }

  278. 

  279. 

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

  281.                 return -1;

  282. 

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

  284.         }

  285. 

  286.         if (type != 2) {

  287.             break;

  288.         }

  289.     }

  290. 

  291.     return 0;

  292. }

  293. 

  294. static inline void svq3_mc_dir_part(SVQ3Context *s,

  295.                                     int x, int y, int width, int height,

  296.                                     int mx, int my, int dxy,

  297.                                     int thirdpel, int dir, int avg)

  298. {

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

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

  301.     uint8_t *src, *dest;

  302.     int i, emu = 0;

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

  304. 

  305.     mx += x;

  306.     my += y;

  307. 

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

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

  310.         emu = 1;

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

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

  313.     }

  314. 

  315.     /* form component predictions */

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

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

  318. 

  319.     if (emu) {

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

  321.                                  h->linesize, h->linesize,

  322.                                  width + 1, height + 1,

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

  324.         src = h->edge_emu_buffer;

  325.     }

  326.     if (thirdpel)

  327.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  329.                                                 width, height);

  330.     else

  331.         (avg ? s->hdsp.avg_pixels_tab

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

  333.                                                        height);

  334. 

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

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

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

  338.         width  = width  >> 1;

  339.         height = height >> 1;

  340.         blocksize++;

  341. 

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

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

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

  345. 

  346.             if (emu) {

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

  348.                                          h->uvlinesize, h->uvlinesize,

  349.                                          width + 1, height + 1,

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

  351.                                          s->v_edge_pos >> 1);

  352.                 src = h->edge_emu_buffer;

  353.             }

  354.             if (thirdpel)

  355.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  357.                                                         h->uvlinesize,

  358.                                                         width, height);

  359.             else

  360.                 (avg ? s->hdsp.avg_pixels_tab

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

  362.                                                                h->uvlinesize,

  363.                                                                height);

  364.         }

  365.     }

  366. }

  367. 

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

  369.                               int dir, int avg)

  370. {

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

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

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

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

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

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

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

  378. 

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

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

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

  382.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  383.             int dxy;

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

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

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

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

  388. 

  389.             if (mode != PREDICT_MODE) {

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

  391.             } else {

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

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

  394. 

  395.                 if (dir == 0) {

  396.                     mx = mx * h->frame_num_offset /

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

  398.                     my = my * h->frame_num_offset /

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

  400.                 } else {

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

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

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

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

  405.                 }

  406.             }

  407. 

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

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

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

  411. 

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

  413.             if (mode == PREDICT_MODE) {

  414.                 dx = dy = 0;

  415.             } else {

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

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

  418. 

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

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

  421.                     return -1;

  422.                 }

  423.             }

  424. 

  425.             /* compute motion vector */

  426.             if (mode == THIRDPEL_MODE) {

  427.                 int fx, fy;

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

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

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

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

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

  433. 

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

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

  436.                 mx += mx;

  437.                 my += my;

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

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

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

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

  442. 

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

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

  445.                 mx *= 3;

  446.                 my *= 3;

  447.             } else {

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

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

  450. 

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

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

  453.                 mx *= 6;

  454.                 my *= 6;

  455.             }

  456. 

  457.             /* update mv_cache */

  458.             if (mode != PREDICT_MODE) {

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

  460. 

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

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

  463. 

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

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

  466.                 }

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

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

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

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

  471.             }

  472. 

  473.             /* write back motion vectors */

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

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

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

  477.         }

  478. 

  479.     return 0;

  480. }

  481. 

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

  483. {

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

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

  486.     int cbp = 0;

  487.     uint32_t vlc;

  488.     int8_t *top, *left;

  489.     const int mb_xy         = h->mb_xy;

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

  491. 

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

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

  494.     h->topright_samples_available = 0xFFFF;

  495. 

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

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

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

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

  500.                              0, 0, 0, 0, 0, 0);

  501. 

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

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

  504.                                  0, 0, 0, 0, 1, 1);

  505. 

  506.             mb_type = MB_TYPE_SKIP;

  507.         } else {

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

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

  510.                 return -1;

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

  512.                 return -1;

  513. 

  514.             mb_type = MB_TYPE_16x16;

  515.         }

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

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

  518.             mode = THIRDPEL_MODE;

  519.         else if (s->halfpel_flag &&

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

  521.             mode = HALFPEL_MODE;

  522.         else

  523.             mode = FULLPEL_MODE;

  524. 

  525.         /* fill caches */

  526.         /* note ref_cache should contain here:

  527.          *  ????????

  528.          *  ???11111

  529.          *  N??11111

  530.          *  N??11111

  531.          *  N??11111

  532.          */

  533. 

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

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

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

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

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

  539.             } else {

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

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

  542.             }

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

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

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

  546.                        4 * 2 * sizeof(int16_t));

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

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

  549. 

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

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

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

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

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

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

  556.                 } else

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

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

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

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

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

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

  563.                 } else

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

  565.             } else

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

  567.                        PART_NOT_AVAILABLE, 8);

  568. 

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

  570.                 break;

  571.         }

  572. 

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

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

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

  576.                 return -1;

  577.         } else {        /* AV_PICTURE_TYPE_B */

  578.             if (mb_type != 2) {

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

  580.                     return -1;

  581.             } else {

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

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

  584.                            0, 4 * 2 * sizeof(int16_t));

  585.             }

  586.             if (mb_type != 1) {

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

  588.                     return -1;

  589.             } else {

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

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

  592.                            0, 4 * 2 * sizeof(int16_t));

  593.             }

  594.         }

  595. 

  596.         mb_type = MB_TYPE_16x16;

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

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

  599. 

  600.         if (mb_type == 8) {

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

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

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

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

  605.                     h->left_samples_available = 0x5F5F;

  606.             }

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

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

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

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

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

  612. 

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

  614.                     h->top_samples_available = 0x33FF;

  615.             }

  616. 

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

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

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

  620. 

  621.                 if (vlc >= 25U) {

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

  623.                     return -1;

  624.                 }

  625. 

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

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

  628. 

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

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

  631. 

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

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

  634.                     return -1;

  635.                 }

  636.             }

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

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

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

  640.         }

  641. 

  642.         write_back_intra_pred_mode(h);

  643. 

  644.         if (mb_type == 8) {

  645.             ff_h264_check_intra4x4_pred_mode(h);

  646. 

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

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

  649.         } else {

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

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

  652. 

  653.             h->top_samples_available  = 0x33FF;

  654.             h->left_samples_available = 0x5F5F;

  655.         }

  656. 

  657.         mb_type = MB_TYPE_INTRA4x4;

  658.     } else {                      /* INTRA16x16 */

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

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

  661. 

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

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

  664.             return h->intra16x16_pred_mode;

  665.         }

  666. 

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

  668.         mb_type = MB_TYPE_INTRA16x16;

  669.     }

  670. 

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

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

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

  674.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  678.                        0, 4 * 2 * sizeof(int16_t));

  679.         }

  680.     }

  681.     if (!IS_INTRA4x4(mb_type)) {

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

  683.     }

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

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

  686.     }

  687. 

  688.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  692.             return -1;

  693.         }

  694. 

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

  696.                                 : golomb_to_inter_cbp[vlc];

  697.     }

  698.     if (IS_INTRA16x16(mb_type) ||

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

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

  701. 

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

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

  704.             return -1;

  705.         }

  706.     }

  707.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  711.             av_log(h->avctx, AV_LOG_ERROR,

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

  713.             return -1;

  714.         }

  715.     }

  716. 

  717.     if (cbp) {

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

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

  720. 

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

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

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

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

  725.                                  2 * (j & 2) + 4 * (i & 2))

  726.                               : (4 * i + j);

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

  728. 

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

  730.                         av_log(h->avctx, AV_LOG_ERROR,

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

  732.                         return -1;

  733.                     }

  734.                 }

  735.             }

  736. 

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

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

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

  740.                     av_log(h->avctx, AV_LOG_ERROR,

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

  742.                     return -1;

  743.                 }

  744. 

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

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

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

  748.                         k                                 = 16 * i + j;

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

  750. 

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

  752.                             av_log(h->avctx, AV_LOG_ERROR,

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

  754.                             return -1;

  755.                         }

  756.                     }

  757.                 }

  758.             }

  759.         }

  760.     }

  761. 

  762.     h->cbp                              = cbp;

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

  764. 

  765.     if (IS_INTRA(mb_type))

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

  767. 

  768.     return 0;

  769. }

  770. 

  771. static int svq3_decode_slice_header(AVCodecContext *avctx)

  772. {

  773.     SVQ3Context *s = avctx->priv_data;

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

  775.     const int mb_xy   = h->mb_xy;

  776.     int i, header;

  777.     unsigned slice_id;

  778. 

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

  780. 

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

  782.         /* TODO: what? */

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

  784.         return -1;

  785.     } else {

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

  787. 

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

  789.                               8 * show_bits(&h->gb, 8 * length) +

  790.                               8 * length;

  791. 

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

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

  794.             return -1;

  795.         }

  796. 

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

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

  799. 

  800.         if (s->watermark_key) {

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

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

  803.                     header ^ s->watermark_key);

  804.         }

  805.         if (length > 0) {

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

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

  808.         }

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

  810.     }

  811. 

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

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

  814.         return -1;

  815.     }

  816. 

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

  818. 

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

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

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

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

  823.     } else {

  824.         skip_bits1(&h->gb);

  825.         h->mb_skip_run = 0;

  826.     }

  827. 

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

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

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

  831. 

  832.     /* unknown fields */

  833.     skip_bits1(&h->gb);

  834. 

  835.     if (s->unknown_flag)

  836.         skip_bits1(&h->gb);

  837. 

  838.     skip_bits1(&h->gb);

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

  840. 

  841.     if (skip_1stop_8data_bits(&h->gb) < 0)

  842.         return AVERROR_INVALIDDATA;

  843. 

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

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

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

  847.                -1, 4 * sizeof(int8_t));

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

  849.                -1, 8 * sizeof(int8_t) * h->mb_x);

  850.     }

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

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

  853.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  854. 

  855.         if (h->mb_x > 0)

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

  857.     }

  858. 

  859.     return 0;

  860. }

  861. 

  862. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  863. {

  864.     SVQ3Context *s = avctx->priv_data;

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

  866.     int m;

  867.     unsigned char *extradata;

  868.     unsigned char *extradata_end;

  869.     unsigned int size;

  870.     int marker_found = 0;

  871. 

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

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

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

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

  876.         av_freep(&s->cur_pic);

  877.         av_freep(&s->last_pic);

  878.         av_freep(&s->next_pic);

  879.         return AVERROR(ENOMEM);

  880.     }

  881. 

  882.     if (ff_h264_decode_init(avctx) < 0)

  883.         return -1;

  884. 

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

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

  887.     h->is_complex      = 1;

  888.     h->sps.chroma_format_idc = 1;

  889.     h->picture_structure = PICT_FRAME;

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

  891. 

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

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

  894. 

  895.     s->halfpel_flag  = 1;

  896.     s->thirdpel_flag = 1;

  897.     s->unknown_flag  = 0;

  898. 

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

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

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

  902.     if (extradata) {

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

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

  905.                 marker_found = 1;

  906.                 break;

  907.             }

  908.             extradata++;

  909.         }

  910.     }

  911. 

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

  913.     if (marker_found) {

  914.         GetBitContext gb;

  915.         int frame_size_code;

  916. 

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

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

  919.             return AVERROR_INVALIDDATA;

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

  921. 

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

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

  924.         switch (frame_size_code) {

  925.         case 0:

  926.             avctx->width  = 160;

  927.             avctx->height = 120;

  928.             break;

  929.         case 1:

  930.             avctx->width  = 128;

  931.             avctx->height =  96;

  932.             break;

  933.         case 2:

  934.             avctx->width  = 176;

  935.             avctx->height = 144;

  936.             break;

  937.         case 3:

  938.             avctx->width  = 352;

  939.             avctx->height = 288;

  940.             break;

  941.         case 4:

  942.             avctx->width  = 704;

  943.             avctx->height = 576;

  944.             break;

  945.         case 5:

  946.             avctx->width  = 240;

  947.             avctx->height = 180;

  948.             break;

  949.         case 6:

  950.             avctx->width  = 320;

  951.             avctx->height = 240;

  952.             break;

  953.         case 7:

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

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

  956.             break;

  957.         }

  958. 

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

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

  961. 

  962.         /* unknown fields */

  963.         skip_bits1(&gb);

  964.         skip_bits1(&gb);

  965.         skip_bits1(&gb);

  966.         skip_bits1(&gb);

  967. 

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

  969. 

  970.         /* unknown field */

  971.         skip_bits1(&gb);

  972. 

  973.         if (skip_1stop_8data_bits(&gb) < 0)

  974.             return AVERROR_INVALIDDATA;

  975. 

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

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

  978.         if (s->unknown_flag) {

  979. #if CONFIG_ZLIB

  980.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  981.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  982.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  985.             int u4                    = svq3_get_ue_golomb(&gb);

  986.             unsigned long buf_len     = watermark_width *

  987.                                         watermark_height * 4;

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

  989.             uint8_t *buf;

  990. 

  991.             if (watermark_height <= 0 ||

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

  993.                 return -1;

  994. 

  995.             buf = av_malloc(buf_len);

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

  997.                    watermark_width, watermark_height);

  998.             av_log(avctx, AV_LOG_DEBUG,

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

  1000.                    u1, u2, u3, u4, offset);

  1001.             if (uncompress(buf, &buf_len, extradata + 8 + offset,

  1002.                            size - offset) != Z_OK) {

  1003.                 av_log(avctx, AV_LOG_ERROR,

  1004.                        "could not uncompress watermark logo\n");

  1005.                 av_free(buf);

  1006.                 return -1;

  1007.             }

  1008.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1009.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1010.             av_log(avctx, AV_LOG_DEBUG,

  1011.                    "watermark key %#x\n", s->watermark_key);

  1012.             av_free(buf);

  1013. #else

  1014.             av_log(avctx, AV_LOG_ERROR,

  1015.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1016.             return -1;

  1017. #endif

  1018.         }

  1019.     }

  1020. 

  1021.     h->width  = avctx->width;

  1022.     h->height = avctx->height;

  1023.     h->mb_width  = (h->width + 15) / 16;

  1024.     h->mb_height = (h->height + 15) / 16;

  1025.     h->mb_stride = h->mb_width + 1;

  1026.     h->mb_num    = h->mb_width * h->mb_height;

  1027.     h->b_stride = 4 * h->mb_width;

  1028.     s->h_edge_pos = h->mb_width * 16;

  1029.     s->v_edge_pos = h->mb_height * 16;

  1030. 

  1031.     if (ff_h264_alloc_tables(h) < 0) {

  1032.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1033.         return AVERROR(ENOMEM);

  1034.     }

  1035. 

  1036.     return 0;

  1037. }

  1038. 

  1039. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1040. {

  1041.     int i;

  1042.     for (i = 0; i < 2; i++) {

  1043.         av_buffer_unref(&pic->motion_val_buf[i]);

  1044.         av_buffer_unref(&pic->ref_index_buf[i]);

  1045.     }

  1046.     av_buffer_unref(&pic->mb_type_buf);

  1047. 

  1048.     av_frame_unref(&pic->f);

  1049. }

  1050. 

  1051. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1052. {

  1053.     SVQ3Context *s = avctx->priv_data;

  1054.     H264Context *h = &s->h;

  1055.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1056.     const int mb_array_size = h->mb_stride * h->mb_height;

  1057.     const int b4_stride     = h->mb_width * 4 + 1;

  1058.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1059.     int ret;

  1060. 

  1061.     if (!pic->motion_val_buf[0]) {

  1062.         int i;

  1063. 

  1064.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1065.         if (!pic->mb_type_buf)

  1066.             return AVERROR(ENOMEM);

  1067.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1068. 

  1069.         for (i = 0; i < 2; i++) {

  1070.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1071.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1072.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1073.                 ret = AVERROR(ENOMEM);

  1074.                 goto fail;

  1075.             }

  1076. 

  1077.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1078.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1079.         }

  1080.     }

  1081.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1082. 

  1083.     ret = ff_get_buffer(avctx, &pic->f,

  1084.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1085.     if (ret < 0)

  1086.         goto fail;

  1087. 

  1088.     if (!h->edge_emu_buffer) {

  1089.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1090.         if (!h->edge_emu_buffer)

  1091.             return AVERROR(ENOMEM);

  1092.     }

  1093. 

  1094.     h->linesize   = pic->f.linesize[0];

  1095.     h->uvlinesize = pic->f.linesize[1];

  1096. 

  1097.     return 0;

  1098. fail:

  1099.     free_picture(avctx, pic);

  1100.     return ret;

  1101. }

  1102. 

  1103. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1104.                              int *got_frame, AVPacket *avpkt)

  1105. {

  1106.     SVQ3Context *s     = avctx->priv_data;

  1107.     H264Context *h     = &s->h;

  1108.     int buf_size       = avpkt->size;

  1109.     int left;

  1110.     uint8_t *buf;

  1111.     int ret, m, i;

  1112. 

  1113.     /* special case for last picture */

  1114.     if (buf_size == 0) {

  1115.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1116.             ret = av_frame_ref(data, &s->next_pic->f);

  1117.             if (ret < 0)

  1118.                 return ret;

  1119.             s->last_frame_output = 1;

  1120.             *got_frame          = 1;

  1121.         }

  1122.         return 0;

  1123.     }

  1124. 

  1125.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1126. 

  1127.     if (s->watermark_key) {

  1128.         av_fast_padded_malloc(&s->buf, &s->buf_size, buf_size);

  1129.         if (!s->buf)

  1130.             return AVERROR(ENOMEM);

  1131.         memcpy(s->buf, avpkt->data, buf_size);

  1132.         buf = s->buf;

  1133.     } else {

  1134.         buf = avpkt->data;

  1135.     }

  1136. 

  1137.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1138. 

  1139.     if (svq3_decode_slice_header(avctx))

  1140.         return -1;

  1141. 

  1142.     h->pict_type = h->slice_type;

  1143. 

  1144.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1145.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1146. 

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

  1148. 

  1149.     /* for skipping the frame */

  1150.     s->cur_pic->f.pict_type = h->pict_type;

  1151.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1152. 

  1153.     ret = get_buffer(avctx, s->cur_pic);

  1154.     if (ret < 0)

  1155.         return ret;

  1156. 

  1157.     h->cur_pic_ptr = s->cur_pic;

  1158.     av_frame_unref(&h->cur_pic.f);

  1159.     h->cur_pic     = *s->cur_pic;

  1160.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1161.     if (ret < 0)

  1162.         return ret;

  1163. 

  1164.     for (i = 0; i < 16; i++) {

  1165.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1166.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1167.     }

  1168.     for (i = 0; i < 16; i++) {

  1169.         h->block_offset[16 + i]      =

  1170.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1171.         h->block_offset[48 + 16 + i] =

  1172.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1173.     }

  1174. 

  1175.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1176.         if (!s->last_pic->f.data[0]) {

  1177.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1178.             ret = get_buffer(avctx, s->last_pic);

  1179.             if (ret < 0)

  1180.                 return ret;

  1181.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1182.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1183.                    s->last_pic->f.linesize[1]);

  1184.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1185.                    s->last_pic->f.linesize[2]);

  1186.         }

  1187. 

  1188.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1189.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1190.             ret = get_buffer(avctx, s->next_pic);

  1191.             if (ret < 0)

  1192.                 return ret;

  1193.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1194.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1195.                    s->next_pic->f.linesize[1]);

  1196.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1197.                    s->next_pic->f.linesize[2]);

  1198.         }

  1199.     }

  1200. 

  1201.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1202.         av_log(h->avctx, AV_LOG_DEBUG,

  1203.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1204.                av_get_picture_type_char(h->pict_type),

  1205.                s->halfpel_flag, s->thirdpel_flag,

  1206.                s->adaptive_quant, h->qscale, h->slice_num);

  1207. 

  1208.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1209.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1210.         avctx->skip_frame >= AVDISCARD_ALL)

  1211.         return 0;

  1212. 

  1213.     if (s->next_p_frame_damaged) {

  1214.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1215.             return 0;

  1216.         else

  1217.             s->next_p_frame_damaged = 0;

  1218.     }

  1219. 

  1220.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1221.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1222. 

  1223.         if (h->frame_num_offset < 0)

  1224.             h->frame_num_offset += 256;

  1225.         if (h->frame_num_offset == 0 ||

  1226.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1227.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1228.             return -1;

  1229.         }

  1230.     } else {

  1231.         h->prev_frame_num        = h->frame_num;

  1232.         h->frame_num             = h->slice_num;

  1233.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1234. 

  1235.         if (h->prev_frame_num_offset < 0)

  1236.             h->prev_frame_num_offset += 256;

  1237.     }

  1238. 

  1239.     for (m = 0; m < 2; m++) {

  1240.         int i;

  1241.         for (i = 0; i < 4; i++) {

  1242.             int j;

  1243.             for (j = -1; j < 4; j++)

  1244.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1245.             if (i < 3)

  1246.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1247.         }

  1248.     }

  1249. 

  1250.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1251.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1252.             unsigned mb_type;

  1253.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1254. 

  1255.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1256.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1257.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1258.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1259.                 h->gb.size_in_bits = 8 * buf_size;

  1260. 

  1261.                 if (svq3_decode_slice_header(avctx))

  1262.                     return -1;

  1263. 

  1264.                 /* TODO: support s->mb_skip_run */

  1265.             }

  1266. 

  1267.             mb_type = svq3_get_ue_golomb(&h->gb);

  1268. 

  1269.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1270.                 mb_type += 8;

  1271.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1272.                 mb_type += 4;

  1273.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1274.                 av_log(h->avctx, AV_LOG_ERROR,

  1275.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1276.                 return -1;

  1277.             }

  1278. 

  1279.             if (mb_type != 0 || h->cbp)

  1280.                 ff_h264_hl_decode_mb(h);

  1281. 

  1282.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1283.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1284.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1285.         }

  1286. 

  1287.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1288.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1289.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1290.     }

  1291. 

  1292.     left = buf_size*8 - get_bits_count(&h->gb);

  1293. 

  1294.     if (h->mb_y != h->mb_height || h->mb_x != h->mb_width) {

  1295.         av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, h->mb_y, h->mb_x, left);

  1296.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1297.     }

  1298. 

  1299.     if (left < 0) {

  1300.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1301.         return -1;

  1302.     }

  1303. 

  1304.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1305.         ret = av_frame_ref(data, &s->cur_pic->f);

  1306.     else if (s->last_pic->f.data[0])

  1307.         ret = av_frame_ref(data, &s->last_pic->f);

  1308.     if (ret < 0)

  1309.         return ret;

  1310. 

  1311.     /* Do not output the last pic after seeking. */

  1312.     if (s->last_pic->f.data[0] || h->low_delay)

  1313.         *got_frame = 1;

  1314. 

  1315.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1316.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1317.     } else {

  1318.         av_frame_unref(&s->cur_pic->f);

  1319.     }

  1320. 

  1321.     return buf_size;

  1322. }

  1323. 

  1324. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1325. {

  1326.     SVQ3Context *s = avctx->priv_data;

  1327.     H264Context *h = &s->h;

  1328. 

  1329.     free_picture(avctx, s->cur_pic);

  1330.     free_picture(avctx, s->next_pic);

  1331.     free_picture(avctx, s->last_pic);

  1332.     av_freep(&s->cur_pic);

  1333.     av_freep(&s->next_pic);

  1334.     av_freep(&s->last_pic);

  1335. 

  1336.     av_frame_unref(&h->cur_pic.f);

  1337. 

  1338.     ff_h264_free_context(h);

  1339. 

  1340.     av_freep(&s->buf);

  1341.     s->buf_size = 0;

  1342.     av_freep(&h->edge_emu_buffer);

  1343. 

  1344.     return 0;

  1345. }

  1346. 

  1347. AVCodec ff_svq3_decoder = {

  1348.     .name           = "svq3",

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

  1350.     .type           = AVMEDIA_TYPE_VIDEO,

  1351.     .id             = AV_CODEC_ID_SVQ3,

  1352.     .priv_data_size = sizeof(SVQ3Context),

  1353.     .init           = svq3_decode_init,

  1354.     .close          = svq3_decode_end,

  1355.     .decode         = svq3_decode_frame,

  1356.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1357.                       CODEC_CAP_DR1             |

  1358.                       CODEC_CAP_DELAY,

  1359.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1360.                                                      AV_PIX_FMT_NONE},

  1361. };