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.
79648 Commits
32 Branches
732MB
Tree: f8d1bb2f2c
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'f8d1bb2f2c'
${ noResults }
FFmpeg/libavcodec/svq3.c

1544 lines
55KB
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 <inttypes.h>

  44. 

  45. #include "libavutil/attributes.h"

  46. #include "internal.h"

  47. #include "avcodec.h"

  48. #include "mpegutils.h"

  49. #include "h264.h"

  50. #include "h264_mvpred.h"

  51. #include "h264data.h"

  52. #include "golomb.h"

  53. #include "hpeldsp.h"

  54. #include "mathops.h"

  55. #include "rectangle.h"

  56. #include "tpeldsp.h"

  57. #include "vdpau_internal.h"

  58. 

  59. #if CONFIG_ZLIB

  60. #include <zlib.h>

  61. #endif

  62. 

  63. #include "svq1.h"

  64. #include "svq3.h"

  65. 

  66. /**

  67.  * @file

  68.  * svq3 decoder.

  69.  */

  70. 

  71. typedef struct SVQ3Context {

  72.     H264Context h;

  73.     HpelDSPContext hdsp;

  74.     TpelDSPContext tdsp;

  75.     H264Picture *cur_pic;

  76.     H264Picture *next_pic;

  77.     H264Picture *last_pic;

  78.     GetBitContext gb;

  79.     uint8_t *slice_buf;

  80.     int slice_size;

  81.     int halfpel_flag;

  82.     int thirdpel_flag;

  83.     int has_watermark;

  84.     uint32_t watermark_key;

  85.     uint8_t *buf;

  86.     int buf_size;

  87.     int adaptive_quant;

  88.     int next_p_frame_damaged;

  89.     int h_edge_pos;

  90.     int v_edge_pos;

  91.     int last_frame_output;

  92. } SVQ3Context;

  93. 

  94. #define FULLPEL_MODE  1

  95. #define HALFPEL_MODE  2

  96. #define THIRDPEL_MODE 3

  97. #define PREDICT_MODE  4

  98. 

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

  100.  * o-->o-->o   o

  101.  *         |  /|

  102.  * o   o   o / o

  103.  * | / |   |/  |

  104.  * o   o   o   o

  105.  *   /

  106.  * o-->o-->o-->o

  107.  */

  108. static const uint8_t svq3_scan[16] = {

  109.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  110.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

  111.     0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4,

  112.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  113. };

  114. 

  115. static const uint8_t luma_dc_zigzag_scan[16] = {

  116.     0 * 16 + 0 * 64, 1 * 16 + 0 * 64, 2 * 16 + 0 * 64, 0 * 16 + 2 * 64,

  117.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

  118.     1 * 16 + 2 * 64, 2 * 16 + 2 * 64, 3 * 16 + 2 * 64, 0 * 16 + 3 * 64,

  119.     3 * 16 + 1 * 64, 1 * 16 + 3 * 64, 2 * 16 + 3 * 64, 3 * 16 + 3 * 64,

  120. };

  121. 

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

  123.     { 0, 0 },

  124.     { 1, 0 }, { 0, 1 },

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

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

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

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

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

  130.     { 4, 3 }, { 3, 4 },

  131.     { 4, 4 }

  132. };

  133. 

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

  135.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  136.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

  142.       { 2,  1,  0,  4,  3 }, { 2, 1,  3,  4,  0 }, { 2, 4,  1,  0,  3 } },

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

  144.       { 2,  0,  1,  3,  4 }, { 2, 1,  3,  0,  4 }, { 2, 0,  4,  3,  1 } },

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

  146.       { 4,  2,  0,  1,  3 }, { 2, 0,  1,  4,  3 }, { 4, 2,  1,  0,  3 } },

  147. };

  148. 

  149. static const struct {

  150.     uint8_t run;

  151.     uint8_t level;

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

  153.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  154.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  155.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  156.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  157. };

  158. 

  159. static const uint32_t svq3_dequant_coeff[32] = {

  160.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  161.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  162.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  163.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  164. };

  165. 

  166. static int svq3_decode_end(AVCodecContext *avctx);

  167. 

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

  169. {

  170.     const int qmul = svq3_dequant_coeff[qp];

  171. #define stride 16

  172.     int i;

  173.     int temp[16];

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

  175. 

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

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

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

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

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

  181. 

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

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

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

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

  186.     }

  187. 

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

  189.         const int offset = x_offset[i];

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

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

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

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

  194. 

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

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

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

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

  199.     }

  200. }

  201. #undef stride

  202. 

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

  204.                         int stride, int qp, int dc)

  205. {

  206.     const int qmul = svq3_dequant_coeff[qp];

  207.     int i;

  208. 

  209.     if (dc) {

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

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

  212.         block[0] = 0;

  213.     }

  214. 

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

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

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

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

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

  220. 

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

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

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

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

  225.     }

  226. 

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

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

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

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

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

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

  233. 

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

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

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

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

  238.     }

  239. 

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

  241. }

  242. 

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

  244.                                     int index, const int type)

  245. {

  246.     static const uint8_t *const scan_patterns[4] = {

  247.         luma_dc_zigzag_scan, ff_zigzag_scan, svq3_scan, ff_h264_chroma_dc_scan

  248.     };

  249. 

  250.     int run, level, sign, limit;

  251.     unsigned vlc;

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

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

  254. 

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

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

  257.             if ((int32_t)vlc < 0)

  258.                 return -1;

  259. 

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

  261.             vlc      = vlc + 1 >> 1;

  262. 

  263.             if (type == 3) {

  264.                 if (vlc < 3) {

  265.                     run   = 0;

  266.                     level = vlc;

  267.                 } else if (vlc < 4) {

  268.                     run   = 1;

  269.                     level = 1;

  270.                 } else {

  271.                     run   = vlc & 0x3;

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

  273.                 }

  274.             } else {

  275.                 if (vlc < 16U) {

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

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

  278.                 } else if (intra) {

  279.                     run   = vlc & 0x7;

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

  281.                 } else {

  282.                     run   = vlc & 0xF;

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

  284.                 }

  285.             }

  286. 

  287. 

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

  289.                 return -1;

  290. 

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

  292.         }

  293. 

  294.         if (type != 2) {

  295.             break;

  296.         }

  297.     }

  298. 

  299.     return 0;

  300. }

  301. 

  302. static inline void svq3_mc_dir_part(SVQ3Context *s,

  303.                                     int x, int y, int width, int height,

  304.                                     int mx, int my, int dxy,

  305.                                     int thirdpel, int dir, int avg)

  306. {

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

  308.     H264SliceContext *sl = &h->slice_ctx[0];

  309.     const H264Picture *pic = (dir == 0) ? s->last_pic : s->next_pic;

  310.     uint8_t *src, *dest;

  311.     int i, emu = 0;

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

  313. 

  314.     mx += x;

  315.     my += y;

  316. 

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

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

  319.         emu = 1;

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

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

  322.     }

  323. 

  324.     /* form component predictions */

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

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

  327. 

  328.     if (emu) {

  329.         h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src,

  330.                                  sl->linesize, sl->linesize,

  331.                                  width + 1, height + 1,

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

  333.         src = sl->edge_emu_buffer;

  334.     }

  335.     if (thirdpel)

  336.         (avg ? s->tdsp.avg_tpel_pixels_tab

  337.              : s->tdsp.put_tpel_pixels_tab)[dxy](dest, src, sl->linesize,

  338.                                                  width, height);

  339.     else

  340.         (avg ? s->hdsp.avg_pixels_tab

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

  342.                                                        height);

  343. 

  344.     if (!(h->flags & AV_CODEC_FLAG_GRAY)) {

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

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

  347.         width  = width  >> 1;

  348.         height = height >> 1;

  349.         blocksize++;

  350. 

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

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

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

  354. 

  355.             if (emu) {

  356.                 h->vdsp.emulated_edge_mc(sl->edge_emu_buffer, src,

  357.                                          sl->uvlinesize, sl->uvlinesize,

  358.                                          width + 1, height + 1,

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

  360.                                          s->v_edge_pos >> 1);

  361.                 src = sl->edge_emu_buffer;

  362.             }

  363.             if (thirdpel)

  364.                 (avg ? s->tdsp.avg_tpel_pixels_tab

  365.                      : s->tdsp.put_tpel_pixels_tab)[dxy](dest, src,

  366.                                                          sl->uvlinesize,

  367.                                                          width, height);

  368.             else

  369.                 (avg ? s->hdsp.avg_pixels_tab

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

  371.                                                                sl->uvlinesize,

  372.                                                                height);

  373.         }

  374.     }

  375. }

  376. 

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

  378.                               int dir, int avg)

  379. {

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

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

  382.     H264SliceContext *sl    = &h->slice_ctx[0];

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

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

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

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

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

  388. 

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

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

  391.             const int b_xy = (4 * sl->mb_x + (j >> 2)) +

  392.                              (4 * sl->mb_y + (i >> 2)) * h->b_stride;

  393.             int dxy;

  394.             x = 16 * sl->mb_x + j;

  395.             y = 16 * sl->mb_y + i;

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

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

  398. 

  399.             if (mode != PREDICT_MODE) {

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

  401.             } else {

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

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

  404. 

  405.                 if (dir == 0) {

  406.                     mx = mx * h->frame_num_offset /

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

  408.                     my = my * h->frame_num_offset /

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

  410.                 } else {

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

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

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

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

  415.                 }

  416.             }

  417. 

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

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

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

  421. 

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

  423.             if (mode == PREDICT_MODE) {

  424.                 dx = dy = 0;

  425.             } else {

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

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

  428. 

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

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

  431.                     return -1;

  432.                 }

  433.             }

  434. 

  435.             /* compute motion vector */

  436.             if (mode == THIRDPEL_MODE) {

  437.                 int fx, fy;

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

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

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

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

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

  443. 

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

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

  446.                 mx += mx;

  447.                 my += my;

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

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

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

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

  452. 

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

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

  455.                 mx *= 3;

  456.                 my *= 3;

  457.             } else {

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

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

  460. 

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

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

  463.                 mx *= 6;

  464.                 my *= 6;

  465.             }

  466. 

  467.             /* update mv_cache */

  468.             if (mode != PREDICT_MODE) {

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

  470. 

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

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

  473. 

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

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

  476.                 }

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

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

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

  480.                     AV_WN32A(sl->mv_cache[dir][scan8[k]], mv);

  481.             }

  482. 

  483.             /* write back motion vectors */

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

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

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

  487.         }

  488. 

  489.     return 0;

  490. }

  491. 

  492. static av_always_inline void hl_decode_mb_idct_luma(const H264Context *h, H264SliceContext *sl,

  493.                                                     int mb_type, const int *block_offset,

  494.                                                     int linesize, uint8_t *dest_y)

  495. {

  496.     int i;

  497.     if (!IS_INTRA4x4(mb_type)) {

  498.         for (i = 0; i < 16; i++)

  499.             if (sl->non_zero_count_cache[scan8[i]] || sl->mb[i * 16]) {

  500.                 uint8_t *const ptr = dest_y + block_offset[i];

  501.                 ff_svq3_add_idct_c(ptr, sl->mb + i * 16, linesize,

  502.                                    sl->qscale, IS_INTRA(mb_type) ? 1 : 0);

  503.             }

  504.     }

  505. }

  506. 

  507. static av_always_inline int dctcoef_get(int16_t *mb, int index)

  508. {

  509.     return AV_RN16A(mb + index);

  510. }

  511. 

  512. static av_always_inline void hl_decode_mb_predict_luma(const H264Context *h,

  513.                                                        H264SliceContext *sl,

  514.                                                        int mb_type,

  515.                                                        const int *block_offset,

  516.                                                        int linesize,

  517.                                                        uint8_t *dest_y)

  518. {

  519.     int i;

  520.     int qscale = sl->qscale;

  521. 

  522.     if (IS_INTRA4x4(mb_type)) {

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

  524.             uint8_t *const ptr = dest_y + block_offset[i];

  525.             const int dir      = sl->intra4x4_pred_mode_cache[scan8[i]];

  526. 

  527.             uint8_t *topright;

  528.             int nnz, tr;

  529.             if (dir == DIAG_DOWN_LEFT_PRED || dir == VERT_LEFT_PRED) {

  530.                 const int topright_avail = (sl->topright_samples_available << i) & 0x8000;

  531.                 assert(sl->mb_y || linesize <= block_offset[i]);

  532.                 if (!topright_avail) {

  533.                     tr       = ptr[3 - linesize] * 0x01010101u;

  534.                     topright = (uint8_t *)&tr;

  535.                 } else

  536.                     topright = ptr + 4 - linesize;

  537.             } else

  538.                 topright = NULL;

  539. 

  540.             h->hpc.pred4x4[dir](ptr, topright, linesize);

  541.             nnz = sl->non_zero_count_cache[scan8[i]];

  542.             if (nnz) {

  543.                 ff_svq3_add_idct_c(ptr, sl->mb + i * 16, linesize, qscale, 0);

  544.             }

  545.         }

  546.     } else {

  547.         h->hpc.pred16x16[sl->intra16x16_pred_mode](dest_y, linesize);

  548.         ff_svq3_luma_dc_dequant_idct_c(sl->mb,

  549.                                        sl->mb_luma_dc[0], qscale);

  550.     }

  551. }

  552. 

  553. static void hl_decode_mb(const H264Context *h, H264SliceContext *sl)

  554. {

  555.     const int mb_x    = sl->mb_x;

  556.     const int mb_y    = sl->mb_y;

  557.     const int mb_xy   = sl->mb_xy;

  558.     const int mb_type = h->cur_pic.mb_type[mb_xy];

  559.     uint8_t *dest_y, *dest_cb, *dest_cr;

  560.     int linesize, uvlinesize;

  561.     int i, j;

  562.     const int *block_offset = &h->block_offset[0];

  563.     const int block_h   = 16 >> h->chroma_y_shift;

  564. 

  565.     dest_y  = h->cur_pic.f->data[0] + (mb_x     + mb_y * sl->linesize)  * 16;

  566.     dest_cb = h->cur_pic.f->data[1] +  mb_x * 8 + mb_y * sl->uvlinesize * block_h;

  567.     dest_cr = h->cur_pic.f->data[2] +  mb_x * 8 + mb_y * sl->uvlinesize * block_h;

  568. 

  569.     h->vdsp.prefetch(dest_y  + (sl->mb_x & 3) * 4 * sl->linesize   + 64, sl->linesize,      4);

  570.     h->vdsp.prefetch(dest_cb + (sl->mb_x & 7)     * sl->uvlinesize + 64, dest_cr - dest_cb, 2);

  571. 

  572.     h->list_counts[mb_xy] = sl->list_count;

  573. 

  574.     linesize   = sl->mb_linesize   = sl->linesize;

  575.     uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;

  576. 

  577.     if (IS_INTRA(mb_type)) {

  578.         h->hpc.pred8x8[sl->chroma_pred_mode](dest_cb, uvlinesize);

  579.         h->hpc.pred8x8[sl->chroma_pred_mode](dest_cr, uvlinesize);

  580. 

  581.         hl_decode_mb_predict_luma(h, sl, mb_type, block_offset, linesize, dest_y);

  582.     }

  583. 

  584.     hl_decode_mb_idct_luma(h, sl, mb_type, block_offset, linesize, dest_y);

  585. 

  586.     if (sl->cbp & 0x30) {

  587.         uint8_t *dest[2] = { dest_cb, dest_cr };

  588.         h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + 16 * 16 * 1,

  589.                                                h->dequant4_coeff[IS_INTRA(mb_type) ? 1 : 4][sl->chroma_qp[0]][0]);

  590.         h->h264dsp.h264_chroma_dc_dequant_idct(sl->mb + 16 * 16 * 2,

  591.                                                h->dequant4_coeff[IS_INTRA(mb_type) ? 2 : 5][sl->chroma_qp[1]][0]);

  592.         for (j = 1; j < 3; j++) {

  593.             for (i = j * 16; i < j * 16 + 4; i++)

  594.                 if (sl->non_zero_count_cache[scan8[i]] || sl->mb[i * 16]) {

  595.                     uint8_t *const ptr = dest[j - 1] + block_offset[i];

  596.                     ff_svq3_add_idct_c(ptr, sl->mb + i * 16,

  597.                                        uvlinesize,

  598.                                        ff_h264_chroma_qp[0][sl->qscale + 12] - 12, 2);

  599.                 }

  600.         }

  601.     }

  602. }

  603. 

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

  605. {

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

  607.     H264SliceContext *sl = &h->slice_ctx[0];

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

  609.     int cbp = 0;

  610.     uint32_t vlc;

  611.     int8_t *top, *left;

  612.     const int mb_xy         = sl->mb_xy;

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

  614. 

  615.     sl->top_samples_available      = (sl->mb_y == 0) ? 0x33FF : 0xFFFF;

  616.     sl->left_samples_available     = (sl->mb_x == 0) ? 0x5F5F : 0xFFFF;

  617.     sl->topright_samples_available = 0xFFFF;

  618. 

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

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

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

  622.             svq3_mc_dir_part(s, 16 * sl->mb_x, 16 * sl->mb_y, 16, 16,

  623.                              0, 0, 0, 0, 0, 0);

  624. 

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

  626.                 svq3_mc_dir_part(s, 16 * sl->mb_x, 16 * sl->mb_y, 16, 16,

  627.                                  0, 0, 0, 0, 1, 1);

  628. 

  629.             mb_type = MB_TYPE_SKIP;

  630.         } else {

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

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

  633.                 return -1;

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

  635.                 return -1;

  636. 

  637.             mb_type = MB_TYPE_16x16;

  638.         }

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

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

  641.             mode = THIRDPEL_MODE;

  642.         else if (s->halfpel_flag &&

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

  644.             mode = HALFPEL_MODE;

  645.         else

  646.             mode = FULLPEL_MODE;

  647. 

  648.         /* fill caches */

  649.         /* note ref_cache should contain here:

  650.          *  ????????

  651.          *  ???11111

  652.          *  N??11111

  653.          *  N??11111

  654.          *  N??11111

  655.          */

  656. 

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

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

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

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

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

  662.             } else {

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

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

  665.             }

  666.             if (sl->mb_y > 0) {

  667.                 memcpy(sl->mv_cache[m][scan8[0] - 1 * 8],

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

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

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

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

  672. 

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

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

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

  676.                     sl->ref_cache[m][scan8[0] + 4 - 1 * 8] =

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

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

  679.                 } else

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

  681.                 if (sl->mb_x > 0) {

  682.                     AV_COPY32(sl->mv_cache[m][scan8[0] - 1 - 1 * 8],

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

  684.                     sl->ref_cache[m][scan8[0] - 1 - 1 * 8] =

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

  686.                 } else

  687.                     sl->ref_cache[m][scan8[0] - 1 - 1 * 8] = PART_NOT_AVAILABLE;

  688.             } else

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

  690.                        PART_NOT_AVAILABLE, 8);

  691. 

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

  693.                 break;

  694.         }

  695. 

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

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

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

  699.                 return -1;

  700.         } else {        /* AV_PICTURE_TYPE_B */

  701.             if (mb_type != 2) {

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

  703.                     return -1;

  704.             } else {

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

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

  707.                            0, 4 * 2 * sizeof(int16_t));

  708.             }

  709.             if (mb_type != 1) {

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

  711.                     return -1;

  712.             } else {

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

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

  715.                            0, 4 * 2 * sizeof(int16_t));

  716.             }

  717.         }

  718. 

  719.         mb_type = MB_TYPE_16x16;

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

  721.         memset(sl->intra4x4_pred_mode_cache, -1, 8 * 5 * sizeof(int8_t));

  722. 

  723.         if (mb_type == 8) {

  724.             if (sl->mb_x > 0) {

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

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

  727.                 if (sl->intra4x4_pred_mode_cache[scan8[0] - 1] == -1)

  728.                     sl->left_samples_available = 0x5F5F;

  729.             }

  730.             if (sl->mb_y > 0) {

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

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

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

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

  735. 

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

  737.                     sl->top_samples_available = 0x33FF;

  738.             }

  739. 

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

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

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

  743. 

  744.                 if (vlc >= 25U) {

  745.                     av_log(h->avctx, AV_LOG_ERROR,

  746.                            "luma prediction:%"PRIu32"\n", vlc);

  747.                     return -1;

  748.                 }

  749. 

  750.                 left = &sl->intra4x4_pred_mode_cache[scan8[i] - 1];

  751.                 top  = &sl->intra4x4_pred_mode_cache[scan8[i] - 8];

  752. 

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

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

  755. 

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

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

  758.                     return -1;

  759.                 }

  760.             }

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

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

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

  764.         }

  765. 

  766.         write_back_intra_pred_mode(h, sl);

  767. 

  768.         if (mb_type == 8) {

  769.             ff_h264_check_intra4x4_pred_mode(h, sl);

  770. 

  771.             sl->top_samples_available  = (sl->mb_y == 0) ? 0x33FF : 0xFFFF;

  772.             sl->left_samples_available = (sl->mb_x == 0) ? 0x5F5F : 0xFFFF;

  773.         } else {

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

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

  776. 

  777.             sl->top_samples_available  = 0x33FF;

  778.             sl->left_samples_available = 0x5F5F;

  779.         }

  780. 

  781.         mb_type = MB_TYPE_INTRA4x4;

  782.     } else {                      /* INTRA16x16 */

  783.         dir = ff_h264_i_mb_type_info[mb_type - 8].pred_mode;

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

  785. 

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

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

  788.             return sl->intra16x16_pred_mode;

  789.         }

  790. 

  791.         cbp     = ff_h264_i_mb_type_info[mb_type - 8].cbp;

  792.         mb_type = MB_TYPE_INTRA16x16;

  793.     }

  794. 

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

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

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

  798.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  802.                        0, 4 * 2 * sizeof(int16_t));

  803.         }

  804.     }

  805.     if (!IS_INTRA4x4(mb_type)) {

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

  807.     }

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

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

  810.     }

  811. 

  812.     if (!IS_INTRA16x16(mb_type) &&

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

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

  815.             av_log(h->avctx, AV_LOG_ERROR, "cbp_vlc=%"PRIu32"\n", vlc);

  816.             return -1;

  817.         }

  818. 

  819.         cbp = IS_INTRA(mb_type) ? ff_h264_golomb_to_intra4x4_cbp[vlc]

  820.                                 : ff_h264_golomb_to_inter_cbp[vlc];

  821.     }

  822.     if (IS_INTRA16x16(mb_type) ||

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

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

  825. 

  826.         if (sl->qscale > 31u) {

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

  828.             return -1;

  829.         }

  830.     }

  831.     if (IS_INTRA16x16(mb_type)) {

  832.         AV_ZERO128(sl->mb_luma_dc[0] + 0);

  833.         AV_ZERO128(sl->mb_luma_dc[0] + 8);

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

  835.             av_log(h->avctx, AV_LOG_ERROR,

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

  837.             return -1;

  838.         }

  839.     }

  840. 

  841.     if (cbp) {

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

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

  844. 

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

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

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

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

  849.                                  2 * (j & 2) + 4 * (i & 2))

  850.                               : (4 * i + j);

  851.                     sl->non_zero_count_cache[scan8[k]] = 1;

  852. 

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

  854.                         av_log(h->avctx, AV_LOG_ERROR,

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

  856.                         return -1;

  857.                     }

  858.                 }

  859.             }

  860. 

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

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

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

  864.                     av_log(h->avctx, AV_LOG_ERROR,

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

  866.                     return -1;

  867.                 }

  868. 

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

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

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

  872.                         k                                 = 16 * i + j;

  873.                         sl->non_zero_count_cache[scan8[k]] = 1;

  874. 

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

  876.                             av_log(h->avctx, AV_LOG_ERROR,

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

  878.                             return -1;

  879.                         }

  880.                     }

  881.                 }

  882.             }

  883.         }

  884.     }

  885. 

  886.     sl->cbp                   = cbp;

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

  888. 

  889.     if (IS_INTRA(mb_type))

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

  891. 

  892.     return 0;

  893. }

  894. 

  895. static int svq3_decode_slice_header(AVCodecContext *avctx)

  896. {

  897.     SVQ3Context *s = avctx->priv_data;

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

  899.     H264SliceContext *sl = &h->slice_ctx[0];

  900.     const int mb_xy   = sl->mb_xy;

  901.     int i, header;

  902.     unsigned slice_id;

  903. 

  904.     header = get_bits(&s->gb, 8);

  905. 

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

  907.         /* TODO: what? */

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

  909.         return -1;

  910.     } else {

  911.         int slice_bits, slice_bytes, slice_length;

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

  913. 

  914.         slice_length = show_bits(&s->gb, 8 * length);

  915.         slice_bits   = slice_length * 8;

  916.         slice_bytes  = slice_length + length - 1;

  917. 

  918.         if (slice_bytes > get_bits_left(&s->gb)) {

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

  920.             return -1;

  921.         }

  922. 

  923.         skip_bits(&s->gb, 8);

  924. 

  925.         av_fast_malloc(&s->slice_buf, &s->slice_size, slice_bytes + AV_INPUT_BUFFER_PADDING_SIZE);

  926.         if (!s->slice_buf)

  927.             return AVERROR(ENOMEM);

  928. 

  929.         memcpy(s->slice_buf, s->gb.buffer + s->gb.index / 8, slice_bytes);

  930. 

  931.         init_get_bits(&h->gb, s->slice_buf, slice_bits);

  932. 

  933.         if (s->watermark_key) {

  934.             uint32_t header = AV_RL32(&h->gb.buffer[1]);

  935.             AV_WL32(&h->gb.buffer[1], header ^ s->watermark_key);

  936.         }

  937.         if (length > 0) {

  938.             memmove(s->slice_buf, &s->slice_buf[slice_length], length - 1);

  939.         }

  940.         skip_bits_long(&s->gb, slice_bytes * 8);

  941.     }

  942. 

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

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

  945.         return -1;

  946.     }

  947. 

  948.     sl->slice_type = ff_h264_golomb_to_pict_type[slice_id];

  949. 

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

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

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

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

  954.     } else {

  955.         skip_bits1(&h->gb);

  956.         sl->mb_skip_run = 0;

  957.     }

  958. 

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

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

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

  962. 

  963.     /* unknown fields */

  964.     skip_bits1(&h->gb);

  965. 

  966.     if (s->has_watermark)

  967.         skip_bits1(&h->gb);

  968. 

  969.     skip_bits1(&h->gb);

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

  971. 

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

  973.         return AVERROR_INVALIDDATA;

  974. 

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

  976.     if (sl->mb_x > 0) {

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

  978.                -1, 4 * sizeof(int8_t));

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

  980.                -1, 8 * sizeof(int8_t) * sl->mb_x);

  981.     }

  982.     if (sl->mb_y > 0) {

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

  984.                -1, 8 * sizeof(int8_t) * (h->mb_width - sl->mb_x));

  985. 

  986.         if (sl->mb_x > 0)

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

  988.     }

  989. 

  990.     return 0;

  991. }

  992. 

  993. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  994. {

  995.     SVQ3Context *s = avctx->priv_data;

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

  997.     H264SliceContext *sl;

  998.     int m;

  999.     unsigned char *extradata;

  1000.     unsigned char *extradata_end;

  1001.     unsigned int size;

  1002.     int marker_found = 0;

  1003.     int ret;

  1004. 

  1005.     s->cur_pic  = av_mallocz(sizeof(*s->cur_pic));

  1006.     s->last_pic = av_mallocz(sizeof(*s->last_pic));

  1007.     s->next_pic = av_mallocz(sizeof(*s->next_pic));

  1008.     if (!s->next_pic || !s->last_pic || !s->cur_pic) {

  1009.         ret = AVERROR(ENOMEM);

  1010.         goto fail;

  1011.     }

  1012. 

  1013.     s->cur_pic->f  = av_frame_alloc();

  1014.     s->last_pic->f = av_frame_alloc();

  1015.     s->next_pic->f = av_frame_alloc();

  1016.     if (!s->cur_pic->f || !s->last_pic->f || !s->next_pic->f)

  1017.         return AVERROR(ENOMEM);

  1018. 

  1019.     if ((ret = ff_h264_decode_init(avctx)) < 0)

  1020.         goto fail;

  1021. 

  1022.     // we will overwrite it later during decoding

  1023.     av_frame_free(&h->cur_pic.f);

  1024. 

  1025.     av_frame_free(&h->last_pic_for_ec.f);

  1026. 

  1027.     ff_h264dsp_init(&h->h264dsp, 8, 1);

  1028.     av_assert0(h->sps.bit_depth_chroma == 0);

  1029.     ff_h264_pred_init(&h->hpc, AV_CODEC_ID_SVQ3, 8, 1);

  1030.     ff_videodsp_init(&h->vdsp, 8);

  1031. 

  1032.     memset(h->pps.scaling_matrix4, 16, 6 * 16 * sizeof(uint8_t));

  1033.     memset(h->pps.scaling_matrix8, 16, 2 * 64 * sizeof(uint8_t));

  1034. 

  1035.     avctx->bits_per_raw_sample = 8;

  1036.     h->sps.bit_depth_luma = 8;

  1037.     h->chroma_format_idc = 1;

  1038. 

  1039.     ff_hpeldsp_init(&s->hdsp, avctx->flags);

  1040.     ff_tpeldsp_init(&s->tdsp);

  1041. 

  1042.     sl = h->slice_ctx;

  1043. 

  1044.     h->flags           = avctx->flags;

  1045.     sl->is_complex     = 1;

  1046.     h->sps.chroma_format_idc = 1;

  1047.     h->picture_structure = PICT_FRAME;

  1048.     avctx->pix_fmt     = AV_PIX_FMT_YUVJ420P;

  1049.     avctx->color_range = AVCOL_RANGE_JPEG;

  1050. 

  1051.     h->slice_ctx[0].chroma_qp[0] = h->slice_ctx[0].chroma_qp[1] = 4;

  1052.     h->chroma_x_shift = h->chroma_y_shift = 1;

  1053. 

  1054.     s->halfpel_flag  = 1;

  1055.     s->thirdpel_flag = 1;

  1056.     s->has_watermark = 0;

  1057. 

  1058.     /* prowl for the "SEQH" marker in the extradata */

  1059.     extradata     = (unsigned char *)avctx->extradata;

  1060.     extradata_end = avctx->extradata + avctx->extradata_size;

  1061.     if (extradata) {

  1062.         for (m = 0; m + 8 < avctx->extradata_size; m++) {

  1063.             if (!memcmp(extradata, "SEQH", 4)) {

  1064.                 marker_found = 1;

  1065.                 break;

  1066.             }

  1067.             extradata++;

  1068.         }

  1069.     }

  1070. 

  1071.     /* if a match was found, parse the extra data */

  1072.     if (marker_found) {

  1073.         GetBitContext gb;

  1074.         int frame_size_code;

  1075.         int unk0, unk1, unk2, unk3, unk4;

  1076. 

  1077.         size = AV_RB32(&extradata[4]);

  1078.         if (size > extradata_end - extradata - 8) {

  1079.             ret = AVERROR_INVALIDDATA;

  1080.             goto fail;

  1081.         }

  1082.         init_get_bits(&gb, extradata + 8, size * 8);

  1083. 

  1084.         /* 'frame size code' and optional 'width, height' */

  1085.         frame_size_code = get_bits(&gb, 3);

  1086.         switch (frame_size_code) {

  1087.         case 0:

  1088.             avctx->width  = 160;

  1089.             avctx->height = 120;

  1090.             break;

  1091.         case 1:

  1092.             avctx->width  = 128;

  1093.             avctx->height =  96;

  1094.             break;

  1095.         case 2:

  1096.             avctx->width  = 176;

  1097.             avctx->height = 144;

  1098.             break;

  1099.         case 3:

  1100.             avctx->width  = 352;

  1101.             avctx->height = 288;

  1102.             break;

  1103.         case 4:

  1104.             avctx->width  = 704;

  1105.             avctx->height = 576;

  1106.             break;

  1107.         case 5:

  1108.             avctx->width  = 240;

  1109.             avctx->height = 180;

  1110.             break;

  1111.         case 6:

  1112.             avctx->width  = 320;

  1113.             avctx->height = 240;

  1114.             break;

  1115.         case 7:

  1116.             avctx->width  = get_bits(&gb, 12);

  1117.             avctx->height = get_bits(&gb, 12);

  1118.             break;

  1119.         }

  1120. 

  1121.         s->halfpel_flag  = get_bits1(&gb);

  1122.         s->thirdpel_flag = get_bits1(&gb);

  1123. 

  1124.         /* unknown fields */

  1125.         unk0 = get_bits1(&gb);

  1126.         unk1 = get_bits1(&gb);

  1127.         unk2 = get_bits1(&gb);

  1128.         unk3 = get_bits1(&gb);

  1129. 

  1130.         h->low_delay = get_bits1(&gb);

  1131. 

  1132.         /* unknown field */

  1133.         unk4 = get_bits1(&gb);

  1134. 

  1135.         av_log(avctx, AV_LOG_DEBUG, "Unknown fields %d %d %d %d %d\n",

  1136.                unk0, unk1, unk2, unk3, unk4);

  1137. 

  1138.         if (skip_1stop_8data_bits(&gb) < 0) {

  1139.             ret = AVERROR_INVALIDDATA;

  1140.             goto fail;

  1141.         }

  1142. 

  1143.         s->has_watermark  = get_bits1(&gb);

  1144.         avctx->has_b_frames = !h->low_delay;

  1145.         if (s->has_watermark) {

  1146. #if CONFIG_ZLIB

  1147.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  1148.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  1149.             int u1                    = svq3_get_ue_golomb(&gb);

  1150.             int u2                    = get_bits(&gb, 8);

  1151.             int u3                    = get_bits(&gb, 2);

  1152.             int u4                    = svq3_get_ue_golomb(&gb);

  1153.             unsigned long buf_len     = watermark_width *

  1154.                                         watermark_height * 4;

  1155.             int offset                = get_bits_count(&gb) + 7 >> 3;

  1156.             uint8_t *buf;

  1157. 

  1158.             if (watermark_height <= 0 ||

  1159.                 (uint64_t)watermark_width * 4 > UINT_MAX / watermark_height) {

  1160.                 ret = -1;

  1161.                 goto fail;

  1162.             }

  1163. 

  1164.             buf = av_malloc(buf_len);

  1165.             if (!buf) {

  1166.                 ret = AVERROR(ENOMEM);

  1167.                 goto fail;

  1168.             }

  1169.             av_log(avctx, AV_LOG_DEBUG, "watermark size: %ux%u\n",

  1170.                    watermark_width, watermark_height);

  1171.             av_log(avctx, AV_LOG_DEBUG,

  1172.                    "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n",

  1173.                    u1, u2, u3, u4, offset);

  1174.             if (uncompress(buf, &buf_len, extradata + 8 + offset,

  1175.                            size - offset) != Z_OK) {

  1176.                 av_log(avctx, AV_LOG_ERROR,

  1177.                        "could not uncompress watermark logo\n");

  1178.                 av_free(buf);

  1179.                 ret = -1;

  1180.                 goto fail;

  1181.             }

  1182.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1183.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1184.             av_log(avctx, AV_LOG_DEBUG,

  1185.                    "watermark key %#"PRIx32"\n", s->watermark_key);

  1186.             av_free(buf);

  1187. #else

  1188.             av_log(avctx, AV_LOG_ERROR,

  1189.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1190.             ret = -1;

  1191.             goto fail;

  1192. #endif

  1193.         }

  1194.     }

  1195. 

  1196.     h->width  = avctx->width;

  1197.     h->height = avctx->height;

  1198.     h->mb_width  = (h->width + 15) / 16;

  1199.     h->mb_height = (h->height + 15) / 16;

  1200.     h->mb_stride = h->mb_width + 1;

  1201.     h->mb_num    = h->mb_width * h->mb_height;

  1202.     h->b_stride = 4 * h->mb_width;

  1203.     s->h_edge_pos = h->mb_width * 16;

  1204.     s->v_edge_pos = h->mb_height * 16;

  1205. 

  1206.     if ((ret = ff_h264_alloc_tables(h)) < 0) {

  1207.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1208.         goto fail;

  1209.     }

  1210. 

  1211.     return 0;

  1212. fail:

  1213.     svq3_decode_end(avctx);

  1214.     return ret;

  1215. }

  1216. 

  1217. static void free_picture(AVCodecContext *avctx, H264Picture *pic)

  1218. {

  1219.     int i;

  1220.     for (i = 0; i < 2; i++) {

  1221.         av_buffer_unref(&pic->motion_val_buf[i]);

  1222.         av_buffer_unref(&pic->ref_index_buf[i]);

  1223.     }

  1224.     av_buffer_unref(&pic->mb_type_buf);

  1225. 

  1226.     av_frame_unref(pic->f);

  1227. }

  1228. 

  1229. static int get_buffer(AVCodecContext *avctx, H264Picture *pic)

  1230. {

  1231.     SVQ3Context *s = avctx->priv_data;

  1232.     H264Context *h = &s->h;

  1233.     H264SliceContext *sl = &h->slice_ctx[0];

  1234.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1235.     const int mb_array_size = h->mb_stride * h->mb_height;

  1236.     const int b4_stride     = h->mb_width * 4 + 1;

  1237.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1238.     int ret;

  1239. 

  1240.     if (!pic->motion_val_buf[0]) {

  1241.         int i;

  1242. 

  1243.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1244.         if (!pic->mb_type_buf)

  1245.             return AVERROR(ENOMEM);

  1246.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1247. 

  1248.         for (i = 0; i < 2; i++) {

  1249.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1250.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1251.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1252.                 ret = AVERROR(ENOMEM);

  1253.                 goto fail;

  1254.             }

  1255. 

  1256.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1257.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1258.         }

  1259.     }

  1260.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1261. 

  1262.     ret = ff_get_buffer(avctx, pic->f,

  1263.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1264.     if (ret < 0)

  1265.         goto fail;

  1266. 

  1267.     if (!sl->edge_emu_buffer) {

  1268.         sl->edge_emu_buffer = av_mallocz_array(pic->f->linesize[0], 17);

  1269.         if (!sl->edge_emu_buffer)

  1270.             return AVERROR(ENOMEM);

  1271.     }

  1272. 

  1273.     sl->linesize   = pic->f->linesize[0];

  1274.     sl->uvlinesize = pic->f->linesize[1];

  1275. 

  1276.     return 0;

  1277. fail:

  1278.     free_picture(avctx, pic);

  1279.     return ret;

  1280. }

  1281. 

  1282. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1283.                              int *got_frame, AVPacket *avpkt)

  1284. {

  1285.     SVQ3Context *s     = avctx->priv_data;

  1286.     H264Context *h     = &s->h;

  1287.     H264SliceContext *sl = &h->slice_ctx[0];

  1288.     int buf_size       = avpkt->size;

  1289.     int left;

  1290.     uint8_t *buf;

  1291.     int ret, m, i;

  1292. 

  1293.     /* special case for last picture */

  1294.     if (buf_size == 0) {

  1295.         if (s->next_pic->f->data[0] && !h->low_delay && !s->last_frame_output) {

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

  1297.             if (ret < 0)

  1298.                 return ret;

  1299.             s->last_frame_output = 1;

  1300.             *got_frame          = 1;

  1301.         }

  1302.         return 0;

  1303.     }

  1304. 

  1305.     sl->mb_x = sl->mb_y = sl->mb_xy = 0;

  1306. 

  1307.     if (s->watermark_key) {

  1308.         av_fast_padded_malloc(&s->buf, &s->buf_size, buf_size);

  1309.         if (!s->buf)

  1310.             return AVERROR(ENOMEM);

  1311.         memcpy(s->buf, avpkt->data, buf_size);

  1312.         buf = s->buf;

  1313.     } else {

  1314.         buf = avpkt->data;

  1315.     }

  1316. 

  1317.     ret = init_get_bits(&s->gb, buf, 8 * buf_size);

  1318.     if (ret < 0)

  1319.         return ret;

  1320. 

  1321.     if (svq3_decode_slice_header(avctx))

  1322.         return -1;

  1323. 

  1324.     h->pict_type = sl->slice_type;

  1325. 

  1326.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1327.         FFSWAP(H264Picture*, s->next_pic, s->last_pic);

  1328. 

  1329.     av_frame_unref(s->cur_pic->f);

  1330. 

  1331.     /* for skipping the frame */

  1332.     s->cur_pic->f->pict_type = h->pict_type;

  1333.     s->cur_pic->f->key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1334. 

  1335.     ret = get_buffer(avctx, s->cur_pic);

  1336.     if (ret < 0)

  1337.         return ret;

  1338. 

  1339.     h->cur_pic_ptr = s->cur_pic;

  1340.     h->cur_pic     = *s->cur_pic;

  1341. 

  1342.     for (i = 0; i < 16; i++) {

  1343.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);

  1344.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);

  1345.     }

  1346.     for (i = 0; i < 16; i++) {

  1347.         h->block_offset[16 + i]      =

  1348.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1349.         h->block_offset[48 + 16 + i] =

  1350.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1351.     }

  1352. 

  1353.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1354.         if (!s->last_pic->f->data[0]) {

  1355.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1356.             av_frame_unref(s->last_pic->f);

  1357.             ret = get_buffer(avctx, s->last_pic);

  1358.             if (ret < 0)

  1359.                 return ret;

  1360.             memset(s->last_pic->f->data[0], 0, avctx->height * s->last_pic->f->linesize[0]);

  1361.             memset(s->last_pic->f->data[1], 0x80, (avctx->height / 2) *

  1362.                    s->last_pic->f->linesize[1]);

  1363.             memset(s->last_pic->f->data[2], 0x80, (avctx->height / 2) *

  1364.                    s->last_pic->f->linesize[2]);

  1365.         }

  1366. 

  1367.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f->data[0]) {

  1368.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1369.             av_frame_unref(s->next_pic->f);

  1370.             ret = get_buffer(avctx, s->next_pic);

  1371.             if (ret < 0)

  1372.                 return ret;

  1373.             memset(s->next_pic->f->data[0], 0, avctx->height * s->next_pic->f->linesize[0]);

  1374.             memset(s->next_pic->f->data[1], 0x80, (avctx->height / 2) *

  1375.                    s->next_pic->f->linesize[1]);

  1376.             memset(s->next_pic->f->data[2], 0x80, (avctx->height / 2) *

  1377.                    s->next_pic->f->linesize[2]);

  1378.         }

  1379.     }

  1380. 

  1381.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1382.         av_log(h->avctx, AV_LOG_DEBUG,

  1383.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1384.                av_get_picture_type_char(h->pict_type),

  1385.                s->halfpel_flag, s->thirdpel_flag,

  1386.                s->adaptive_quant, h->slice_ctx[0].qscale, sl->slice_num);

  1387. 

  1388.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1389.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1390.         avctx->skip_frame >= AVDISCARD_ALL)

  1391.         return 0;

  1392. 

  1393.     if (s->next_p_frame_damaged) {

  1394.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1395.             return 0;

  1396.         else

  1397.             s->next_p_frame_damaged = 0;

  1398.     }

  1399. 

  1400.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1401.         h->frame_num_offset = sl->slice_num - h->prev_frame_num;

  1402. 

  1403.         if (h->frame_num_offset < 0)

  1404.             h->frame_num_offset += 256;

  1405.         if (h->frame_num_offset == 0 ||

  1406.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1407.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1408.             return -1;

  1409.         }

  1410.     } else {

  1411.         h->prev_frame_num        = h->frame_num;

  1412.         h->frame_num             = sl->slice_num;

  1413.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1414. 

  1415.         if (h->prev_frame_num_offset < 0)

  1416.             h->prev_frame_num_offset += 256;

  1417.     }

  1418. 

  1419.     for (m = 0; m < 2; m++) {

  1420.         int i;

  1421.         for (i = 0; i < 4; i++) {

  1422.             int j;

  1423.             for (j = -1; j < 4; j++)

  1424.                 sl->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1425.             if (i < 3)

  1426.                 sl->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1427.         }

  1428.     }

  1429. 

  1430.     for (sl->mb_y = 0; sl->mb_y < h->mb_height; sl->mb_y++) {

  1431.         for (sl->mb_x = 0; sl->mb_x < h->mb_width; sl->mb_x++) {

  1432.             unsigned mb_type;

  1433.             sl->mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;

  1434. 

  1435.             if ((get_bits_left(&h->gb)) <= 7) {

  1436.                 if (((get_bits_count(&h->gb) & 7) == 0 ||

  1437.                     show_bits(&h->gb, get_bits_left(&h->gb) & 7) == 0)) {

  1438. 

  1439.                     if (svq3_decode_slice_header(avctx))

  1440.                         return -1;

  1441.                 }

  1442.                 /* TODO: support s->mb_skip_run */

  1443.             }

  1444. 

  1445.             mb_type = svq3_get_ue_golomb(&h->gb);

  1446. 

  1447.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1448.                 mb_type += 8;

  1449.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1450.                 mb_type += 4;

  1451.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1452.                 av_log(h->avctx, AV_LOG_ERROR,

  1453.                        "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);

  1454.                 return -1;

  1455.             }

  1456. 

  1457.             if (mb_type != 0 || sl->cbp)

  1458.                 hl_decode_mb(h, &h->slice_ctx[0]);

  1459. 

  1460.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1461.                 h->cur_pic.mb_type[sl->mb_x + sl->mb_y * h->mb_stride] =

  1462.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1463.         }

  1464. 

  1465.         ff_draw_horiz_band(avctx, s->cur_pic->f,

  1466.                            s->last_pic->f->data[0] ? s->last_pic->f : NULL,

  1467.                            16 * sl->mb_y, 16, h->picture_structure, 0,

  1468.                            h->low_delay);

  1469.     }

  1470. 

  1471.     left = buf_size*8 - get_bits_count(&h->gb);

  1472. 

  1473.     if (sl->mb_y != h->mb_height || sl->mb_x != h->mb_width) {

  1474.         av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, sl->mb_y, sl->mb_x, left);

  1475.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1476.     }

  1477. 

  1478.     if (left < 0) {

  1479.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1480.         return -1;

  1481.     }

  1482. 

  1483.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1484.         ret = av_frame_ref(data, s->cur_pic->f);

  1485.     else if (s->last_pic->f->data[0])

  1486.         ret = av_frame_ref(data, s->last_pic->f);

  1487.     if (ret < 0)

  1488.         return ret;

  1489. 

  1490.     /* Do not output the last pic after seeking. */

  1491.     if (s->last_pic->f->data[0] || h->low_delay)

  1492.         *got_frame = 1;

  1493. 

  1494.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1495.         FFSWAP(H264Picture*, s->cur_pic, s->next_pic);

  1496.     } else {

  1497.         av_frame_unref(s->cur_pic->f);

  1498.     }

  1499. 

  1500.     return buf_size;

  1501. }

  1502. 

  1503. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1504. {

  1505.     SVQ3Context *s = avctx->priv_data;

  1506.     H264Context *h = &s->h;

  1507. 

  1508.     free_picture(avctx, s->cur_pic);

  1509.     free_picture(avctx, s->next_pic);

  1510.     free_picture(avctx, s->last_pic);

  1511.     av_frame_free(&s->cur_pic->f);

  1512.     av_frame_free(&s->next_pic->f);

  1513.     av_frame_free(&s->last_pic->f);

  1514.     av_freep(&s->cur_pic);

  1515.     av_freep(&s->next_pic);

  1516.     av_freep(&s->last_pic);

  1517.     av_freep(&s->slice_buf);

  1518. 

  1519.     memset(&h->cur_pic, 0, sizeof(h->cur_pic));

  1520. 

  1521.     ff_h264_free_context(h);

  1522. 

  1523.     av_freep(&s->buf);

  1524.     s->buf_size = 0;

  1525. 

  1526.     return 0;

  1527. }

  1528. 

  1529. AVCodec ff_svq3_decoder = {

  1530.     .name           = "svq3",

  1531.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1532.     .type           = AVMEDIA_TYPE_VIDEO,

  1533.     .id             = AV_CODEC_ID_SVQ3,

  1534.     .priv_data_size = sizeof(SVQ3Context),

  1535.     .init           = svq3_decode_init,

  1536.     .close          = svq3_decode_end,

  1537.     .decode         = svq3_decode_frame,

  1538.     .capabilities   = AV_CODEC_CAP_DRAW_HORIZ_BAND |

  1539.                       AV_CODEC_CAP_DR1             |

  1540.                       AV_CODEC_CAP_DELAY,

  1541.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1542.                                                      AV_PIX_FMT_NONE},

  1543. };