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.
39555 Commits
32 Branches
325MB
Tree: cc8163e1a3
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'cc8163e1a3'
${ noResults }
FFmpeg/libavcodec/svq3.c

1334 lines
47KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The Libav Project

  3.  *

  4.  * This file is part of Libav.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

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

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

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

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

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

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

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

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

  31.  *

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

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

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

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

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

  37.  *

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

  39.  * correctly decodes this file:

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

  41.  */

  42. 

  43. #include <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. 

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

  52. 

  53. #include "h264_mvpred.h"

  54. #include "golomb.h"

  55. #include "hpeldsp.h"

  56. #include "rectangle.h"

  57. 

  58. #if CONFIG_ZLIB

  59. #include <zlib.h>

  60. #endif

  61. 

  62. #include "svq1.h"

  63. #include "svq3.h"

  64. 

  65. /**

  66.  * @file

  67.  * svq3 decoder.

  68.  */

  69. 

  70. typedef struct {

  71.     H264Context h;

  72.     HpelDSPContext hdsp;

  73.     H264Picture *cur_pic;

  74.     H264Picture *next_pic;

  75.     H264Picture *last_pic;

  76.     int halfpel_flag;

  77.     int thirdpel_flag;

  78.     int unknown_flag;

  79.     int next_slice_index;

  80.     uint32_t watermark_key;

  81.     int adaptive_quant;

  82.     int next_p_frame_damaged;

  83.     int h_edge_pos;

  84.     int v_edge_pos;

  85.     int last_frame_output;

  86. } SVQ3Context;

  87. 

  88. #define FULLPEL_MODE  1

  89. #define HALFPEL_MODE  2

  90. #define THIRDPEL_MODE 3

  91. #define PREDICT_MODE  4

  92. 

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

  94.  * o-->o-->o   o

  95.  *         |  /|

  96.  * o   o   o / o

  97.  * | / |   |/  |

  98.  * o   o   o   o

  99.  *   /

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

  101.  */

  102. static const uint8_t svq3_scan[16] = {

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

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

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

  106.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  107. };

  108. 

  109. static const uint8_t luma_dc_zigzag_scan[16] = {

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

  111.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

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

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

  114. };

  115. 

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

  117.     { 0, 0 },

  118.     { 1, 0 }, { 0, 1 },

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

  120.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  121.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

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

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

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

  125.     { 4, 4 }

  126. };

  127. 

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

  129.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

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

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

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

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

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

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

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

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

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

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

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

  141. };

  142. 

  143. static const struct {

  144.     uint8_t run;

  145.     uint8_t level;

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

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

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

  149.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  150.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  151. };

  152. 

  153. static const uint32_t svq3_dequant_coeff[32] = {

  154.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  155.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  156.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  157.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  158. };

  159. 

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

  161. {

  162.     const int qmul = svq3_dequant_coeff[qp];

  163. #define stride 16

  164.     int i;

  165.     int temp[16];

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

  167. 

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

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

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

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

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

  173. 

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

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

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

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

  178.     }

  179. 

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

  181.         const int offset = x_offset[i];

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

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

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

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

  186. 

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

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

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

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

  191.     }

  192. }

  193. #undef stride

  194. 

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

  196.                         int stride, int qp, int dc)

  197. {

  198.     const int qmul = svq3_dequant_coeff[qp];

  199.     int i;

  200. 

  201.     if (dc) {

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

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

  204.         block[0] = 0;

  205.     }

  206. 

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

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

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

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

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

  212. 

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

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

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

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

  217.     }

  218. 

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

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

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

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

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

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

  225. 

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

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

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

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

  230.     }

  231. 

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

  233. }

  234. 

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

  236.                                     int index, const int type)

  237. {

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

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

  240. 

  241.     int run, level, limit;

  242.     unsigned vlc;

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

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

  245. 

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

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

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

  249.             vlc      = vlc + 1 >> 1;

  250. 

  251.             if (type == 3) {

  252.                 if (vlc < 3) {

  253.                     run   = 0;

  254.                     level = vlc;

  255.                 } else if (vlc < 4) {

  256.                     run   = 1;

  257.                     level = 1;

  258.                 } else {

  259.                     run   = vlc & 0x3;

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

  261.                 }

  262.             } else {

  263.                 if (vlc < 16) {

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

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

  266.                 } else if (intra) {

  267.                     run   = vlc & 0x7;

  268.                     level = (vlc >> 3) +

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

  270.                 } else {

  271.                     run   = vlc & 0xF;

  272.                     level = (vlc >> 4) +

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

  274.                 }

  275.             }

  276. 

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

  278.                 return -1;

  279. 

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

  281.         }

  282. 

  283.         if (type != 2) {

  284.             break;

  285.         }

  286.     }

  287. 

  288.     return 0;

  289. }

  290. 

  291. static inline void svq3_mc_dir_part(SVQ3Context *s,

  292.                                     int x, int y, int width, int height,

  293.                                     int mx, int my, int dxy,

  294.                                     int thirdpel, int dir, int avg)

  295. {

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

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

  298.     uint8_t *src, *dest;

  299.     int i, emu = 0;

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

  301. 

  302.     mx += x;

  303.     my += y;

  304. 

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

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

  307.         emu = 1;

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

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

  310.     }

  311. 

  312.     /* form component predictions */

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

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

  315. 

  316.     if (emu) {

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

  318.                                  h->linesize, h->linesize,

  319.                                  width + 1, height + 1,

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

  321.         src = h->edge_emu_buffer;

  322.     }

  323.     if (thirdpel)

  324.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  326.                                                 width, height);

  327.     else

  328.         (avg ? s->hdsp.avg_pixels_tab

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

  330.                                                        height);

  331. 

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

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

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

  335.         width  = width  >> 1;

  336.         height = height >> 1;

  337.         blocksize++;

  338. 

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

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

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

  342. 

  343.             if (emu) {

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

  345.                                          h->uvlinesize, h->uvlinesize,

  346.                                          width + 1, height + 1,

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

  348.                                          s->v_edge_pos >> 1);

  349.                 src = h->edge_emu_buffer;

  350.             }

  351.             if (thirdpel)

  352.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  354.                                                         h->uvlinesize,

  355.                                                         width, height);

  356.             else

  357.                 (avg ? s->hdsp.avg_pixels_tab

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

  359.                                                                h->uvlinesize,

  360.                                                                height);

  361.         }

  362.     }

  363. }

  364. 

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

  366.                               int dir, int avg)

  367. {

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

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

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

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

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

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

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

  375. 

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

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

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

  379.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  380.             int dxy;

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

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

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

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

  385. 

  386.             if (mode != PREDICT_MODE) {

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

  388.             } else {

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

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

  391. 

  392.                 if (dir == 0) {

  393.                     mx = mx * h->frame_num_offset /

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

  395.                     my = my * h->frame_num_offset /

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

  397.                 } else {

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

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

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

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

  402.                 }

  403.             }

  404. 

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

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

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

  408. 

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

  410.             if (mode == PREDICT_MODE) {

  411.                 dx = dy = 0;

  412.             } else {

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

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

  415. 

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

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

  418.                     return -1;

  419.                 }

  420.             }

  421. 

  422.             /* compute motion vector */

  423.             if (mode == THIRDPEL_MODE) {

  424.                 int fx, fy;

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

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

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

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

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

  430. 

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

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

  433.                 mx += mx;

  434.                 my += my;

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

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

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

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

  439. 

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

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

  442.                 mx *= 3;

  443.                 my *= 3;

  444.             } else {

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

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

  447. 

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

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

  450.                 mx *= 6;

  451.                 my *= 6;

  452.             }

  453. 

  454.             /* update mv_cache */

  455.             if (mode != PREDICT_MODE) {

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

  457. 

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

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

  460. 

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

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

  463.                 }

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

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

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

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

  468.             }

  469. 

  470.             /* write back motion vectors */

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

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

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

  474.         }

  475. 

  476.     return 0;

  477. }

  478. 

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

  480. {

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

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

  483.     int cbp = 0;

  484.     uint32_t vlc;

  485.     int8_t *top, *left;

  486.     const int mb_xy         = h->mb_xy;

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

  488. 

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

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

  491.     h->topright_samples_available = 0xFFFF;

  492. 

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

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

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

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

  497.                              0, 0, 0, 0, 0, 0);

  498. 

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

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

  501.                                  0, 0, 0, 0, 1, 1);

  502. 

  503.             mb_type = MB_TYPE_SKIP;

  504.         } else {

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

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

  507.                 return -1;

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

  509.                 return -1;

  510. 

  511.             mb_type = MB_TYPE_16x16;

  512.         }

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

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

  515.             mode = THIRDPEL_MODE;

  516.         else if (s->halfpel_flag &&

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

  518.             mode = HALFPEL_MODE;

  519.         else

  520.             mode = FULLPEL_MODE;

  521. 

  522.         /* fill caches */

  523.         /* note ref_cache should contain here:

  524.          *  ????????

  525.          *  ???11111

  526.          *  N??11111

  527.          *  N??11111

  528.          *  N??11111

  529.          */

  530. 

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

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

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

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

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

  536.             } else {

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

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

  539.             }

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

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

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

  543.                        4 * 2 * sizeof(int16_t));

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

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

  546. 

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

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

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

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

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

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

  553.                 } else

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

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

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

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

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

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

  560.                 } else

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

  562.             } else

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

  564.                        PART_NOT_AVAILABLE, 8);

  565. 

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

  567.                 break;

  568.         }

  569. 

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

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

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

  573.                 return -1;

  574.         } else {        /* AV_PICTURE_TYPE_B */

  575.             if (mb_type != 2) {

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

  577.                     return -1;

  578.             } else {

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

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

  581.                            0, 4 * 2 * sizeof(int16_t));

  582.             }

  583.             if (mb_type != 1) {

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

  585.                     return -1;

  586.             } else {

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

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

  589.                            0, 4 * 2 * sizeof(int16_t));

  590.             }

  591.         }

  592. 

  593.         mb_type = MB_TYPE_16x16;

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

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

  596. 

  597.         if (mb_type == 8) {

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

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

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

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

  602.                     h->left_samples_available = 0x5F5F;

  603.             }

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

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

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

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

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

  609. 

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

  611.                     h->top_samples_available = 0x33FF;

  612.             }

  613. 

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

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

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

  617. 

  618.                 if (vlc >= 25) {

  619.                     av_log(h->avctx, AV_LOG_ERROR,

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

  621.                     return -1;

  622.                 }

  623. 

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

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

  626. 

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

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

  629. 

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

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

  632.                     return -1;

  633.                 }

  634.             }

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

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

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

  638.         }

  639. 

  640.         write_back_intra_pred_mode(h);

  641. 

  642.         if (mb_type == 8) {

  643.             ff_h264_check_intra4x4_pred_mode(h);

  644. 

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

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

  647.         } else {

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

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

  650. 

  651.             h->top_samples_available  = 0x33FF;

  652.             h->left_samples_available = 0x5F5F;

  653.         }

  654. 

  655.         mb_type = MB_TYPE_INTRA4x4;

  656.     } else {                      /* INTRA16x16 */

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

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

  659. 

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

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

  662.             return h->intra16x16_pred_mode;

  663.         }

  664. 

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

  666.         mb_type = MB_TYPE_INTRA16x16;

  667.     }

  668. 

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

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

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

  672.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  676.                        0, 4 * 2 * sizeof(int16_t));

  677.         }

  678.     }

  679.     if (!IS_INTRA4x4(mb_type)) {

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

  681.     }

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

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

  684.     }

  685. 

  686.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  690.             return -1;

  691.         }

  692. 

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

  694.                                 : golomb_to_inter_cbp[vlc];

  695.     }

  696.     if (IS_INTRA16x16(mb_type) ||

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

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

  699. 

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

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

  702.             return -1;

  703.         }

  704.     }

  705.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  709.             av_log(h->avctx, AV_LOG_ERROR,

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

  711.             return -1;

  712.         }

  713.     }

  714. 

  715.     if (cbp) {

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

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

  718. 

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

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

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

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

  723.                                  2 * (j & 2) + 4 * (i & 2))

  724.                               : (4 * i + j);

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

  726. 

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

  728.                         av_log(h->avctx, AV_LOG_ERROR,

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

  730.                         return -1;

  731.                     }

  732.                 }

  733.             }

  734. 

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

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

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

  738.                     av_log(h->avctx, AV_LOG_ERROR,

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

  740.                     return -1;

  741.                 }

  742. 

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

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

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

  746.                         k                                 = 16 * i + j;

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

  748. 

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

  750.                             av_log(h->avctx, AV_LOG_ERROR,

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

  752.                             return -1;

  753.                         }

  754.                     }

  755.                 }

  756.             }

  757.         }

  758.     }

  759. 

  760.     h->cbp                              = cbp;

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

  762. 

  763.     if (IS_INTRA(mb_type))

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

  765. 

  766.     return 0;

  767. }

  768. 

  769. static int svq3_decode_slice_header(AVCodecContext *avctx)

  770. {

  771.     SVQ3Context *s = avctx->priv_data;

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

  773.     const int mb_xy   = h->mb_xy;

  774.     int i, header;

  775.     unsigned slice_id;

  776. 

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

  778. 

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

  780.         /* TODO: what? */

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

  782.         return -1;

  783.     } else {

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

  785. 

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

  787.                               8 * show_bits(&h->gb, 8 * length) +

  788.                               8 * length;

  789. 

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

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

  792.             return -1;

  793.         }

  794. 

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

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

  797. 

  798.         if (s->watermark_key) {

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

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

  801.                     header ^ s->watermark_key);

  802.         }

  803.         if (length > 0) {

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

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

  806.         }

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

  808.     }

  809. 

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

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

  812.         return -1;

  813.     }

  814. 

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

  816. 

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

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

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

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

  821.     } else {

  822.         skip_bits1(&h->gb);

  823.         h->mb_skip_run = 0;

  824.     }

  825. 

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

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

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

  829. 

  830.     /* unknown fields */

  831.     skip_bits1(&h->gb);

  832. 

  833.     if (s->unknown_flag)

  834.         skip_bits1(&h->gb);

  835. 

  836.     skip_bits1(&h->gb);

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

  838. 

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

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

  841. 

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

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

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

  845.                -1, 4 * sizeof(int8_t));

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

  847.                -1, 8 * sizeof(int8_t) * h->mb_x);

  848.     }

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

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

  851.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  852. 

  853.         if (h->mb_x > 0)

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

  855.     }

  856. 

  857.     return 0;

  858. }

  859. 

  860. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  861. {

  862.     SVQ3Context *s = avctx->priv_data;

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

  864.     int m;

  865.     unsigned char *extradata;

  866.     unsigned char *extradata_end;

  867.     unsigned int size;

  868.     int marker_found = 0;

  869. 

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

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

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

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

  874.         av_freep(&s->cur_pic);

  875.         av_freep(&s->last_pic);

  876.         av_freep(&s->next_pic);

  877.         return AVERROR(ENOMEM);

  878.     }

  879. 

  880.     if (ff_h264_decode_init(avctx) < 0)

  881.         return -1;

  882. 

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

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

  885.     h->is_complex      = 1;

  886.     h->picture_structure = PICT_FRAME;

  887.     avctx->pix_fmt     = AV_PIX_FMT_YUVJ420P;

  888.     avctx->color_range = AVCOL_RANGE_JPEG;

  889. 

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

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

  892. 

  893.     s->halfpel_flag  = 1;

  894.     s->thirdpel_flag = 1;

  895.     s->unknown_flag  = 0;

  896. 

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

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

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

  900.     if (extradata) {

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

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

  903.                 marker_found = 1;

  904.                 break;

  905.             }

  906.             extradata++;

  907.         }

  908.     }

  909. 

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

  911.     if (marker_found) {

  912.         GetBitContext gb;

  913.         int frame_size_code;

  914. 

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

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

  917.             return AVERROR_INVALIDDATA;

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

  919. 

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

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

  922.         switch (frame_size_code) {

  923.         case 0:

  924.             avctx->width  = 160;

  925.             avctx->height = 120;

  926.             break;

  927.         case 1:

  928.             avctx->width  = 128;

  929.             avctx->height =  96;

  930.             break;

  931.         case 2:

  932.             avctx->width  = 176;

  933.             avctx->height = 144;

  934.             break;

  935.         case 3:

  936.             avctx->width  = 352;

  937.             avctx->height = 288;

  938.             break;

  939.         case 4:

  940.             avctx->width  = 704;

  941.             avctx->height = 576;

  942.             break;

  943.         case 5:

  944.             avctx->width  = 240;

  945.             avctx->height = 180;

  946.             break;

  947.         case 6:

  948.             avctx->width  = 320;

  949.             avctx->height = 240;

  950.             break;

  951.         case 7:

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

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

  954.             break;

  955.         }

  956. 

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

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

  959. 

  960.         /* unknown fields */

  961.         skip_bits1(&gb);

  962.         skip_bits1(&gb);

  963.         skip_bits1(&gb);

  964.         skip_bits1(&gb);

  965. 

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

  967. 

  968.         /* unknown field */

  969.         skip_bits1(&gb);

  970. 

  971.         while (get_bits1(&gb))

  972.             skip_bits(&gb, 8);

  973. 

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

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

  976.         if (s->unknown_flag) {

  977. #if CONFIG_ZLIB

  978.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  979.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  980.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  983.             int u4                    = svq3_get_ue_golomb(&gb);

  984.             unsigned long buf_len     = watermark_width *

  985.                                         watermark_height * 4;

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

  987.             uint8_t *buf;

  988. 

  989.             if (watermark_height > 0 &&

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

  991.                 return -1;

  992. 

  993.             buf = av_malloc(buf_len);

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

  995.                    watermark_width, watermark_height);

  996.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  1000.                            size - offset) != Z_OK) {

  1001.                 av_log(avctx, AV_LOG_ERROR,

  1002.                        "could not uncompress watermark logo\n");

  1003.                 av_free(buf);

  1004.                 return -1;

  1005.             }

  1006.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1007.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1008.             av_log(avctx, AV_LOG_DEBUG,

  1009.                    "watermark key %#"PRIx32"\n", s->watermark_key);

  1010.             av_free(buf);

  1011. #else

  1012.             av_log(avctx, AV_LOG_ERROR,

  1013.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1014.             return -1;

  1015. #endif

  1016.         }

  1017.     }

  1018. 

  1019.     h->width  = avctx->width;

  1020.     h->height = avctx->height;

  1021.     h->mb_width  = (h->width + 15) / 16;

  1022.     h->mb_height = (h->height + 15) / 16;

  1023.     h->mb_stride = h->mb_width + 1;

  1024.     h->mb_num    = h->mb_width * h->mb_height;

  1025.     h->b_stride = 4 * h->mb_width;

  1026.     s->h_edge_pos = h->mb_width * 16;

  1027.     s->v_edge_pos = h->mb_height * 16;

  1028. 

  1029.     if (ff_h264_alloc_tables(h) < 0) {

  1030.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1031.         return AVERROR(ENOMEM);

  1032.     }

  1033. 

  1034.     return 0;

  1035. }

  1036. 

  1037. static void free_picture(AVCodecContext *avctx, H264Picture *pic)

  1038. {

  1039.     int i;

  1040.     for (i = 0; i < 2; i++) {

  1041.         av_buffer_unref(&pic->motion_val_buf[i]);

  1042.         av_buffer_unref(&pic->ref_index_buf[i]);

  1043.     }

  1044.     av_buffer_unref(&pic->mb_type_buf);

  1045. 

  1046.     av_frame_unref(&pic->f);

  1047. }

  1048. 

  1049. static int get_buffer(AVCodecContext *avctx, H264Picture *pic)

  1050. {

  1051.     SVQ3Context *s = avctx->priv_data;

  1052.     H264Context *h = &s->h;

  1053.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1054.     const int mb_array_size = h->mb_stride * h->mb_height;

  1055.     const int b4_stride     = h->mb_width * 4 + 1;

  1056.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1057.     int ret;

  1058. 

  1059.     if (!pic->motion_val_buf[0]) {

  1060.         int i;

  1061. 

  1062.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1063.         if (!pic->mb_type_buf)

  1064.             return AVERROR(ENOMEM);

  1065.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1066. 

  1067.         for (i = 0; i < 2; i++) {

  1068.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1069.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1070.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1071.                 ret = AVERROR(ENOMEM);

  1072.                 goto fail;

  1073.             }

  1074. 

  1075.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1076.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1077.         }

  1078.     }

  1079.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1080. 

  1081.     ret = ff_get_buffer(avctx, &pic->f,

  1082.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1083.     if (ret < 0)

  1084.         goto fail;

  1085. 

  1086.     if (!h->edge_emu_buffer) {

  1087.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1088.         if (!h->edge_emu_buffer)

  1089.             return AVERROR(ENOMEM);

  1090.     }

  1091. 

  1092.     h->linesize   = pic->f.linesize[0];

  1093.     h->uvlinesize = pic->f.linesize[1];

  1094. 

  1095.     return 0;

  1096. fail:

  1097.     free_picture(avctx, pic);

  1098.     return ret;

  1099. }

  1100. 

  1101. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1102.                              int *got_frame, AVPacket *avpkt)

  1103. {

  1104.     const uint8_t *buf = avpkt->data;

  1105.     SVQ3Context *s     = avctx->priv_data;

  1106.     H264Context *h     = &s->h;

  1107.     int buf_size       = avpkt->size;

  1108.     int ret, m, i;

  1109. 

  1110.     /* special case for last picture */

  1111.     if (buf_size == 0) {

  1112.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1113.             ret = av_frame_ref(data, &s->next_pic->f);

  1114.             if (ret < 0)

  1115.                 return ret;

  1116.             s->last_frame_output = 1;

  1117.             *got_frame          = 1;

  1118.         }

  1119.         return 0;

  1120.     }

  1121. 

  1122.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1123. 

  1124.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1125. 

  1126.     if (svq3_decode_slice_header(avctx))

  1127.         return -1;

  1128. 

  1129.     h->pict_type = h->slice_type;

  1130. 

  1131.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1132.         FFSWAP(H264Picture*, s->next_pic, s->last_pic);

  1133. 

  1134.     av_frame_unref(&s->cur_pic->f);

  1135. 

  1136.     /* for skipping the frame */

  1137.     s->cur_pic->f.pict_type = h->pict_type;

  1138.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1139. 

  1140.     ret = get_buffer(avctx, s->cur_pic);

  1141.     if (ret < 0)

  1142.         return ret;

  1143. 

  1144.     h->cur_pic_ptr = s->cur_pic;

  1145.     av_frame_unref(&h->cur_pic.f);

  1146.     h->cur_pic     = *s->cur_pic;

  1147.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1148.     if (ret < 0)

  1149.         return ret;

  1150. 

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

  1152.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1153.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1154.     }

  1155.     for (i = 0; i < 16; i++) {

  1156.         h->block_offset[16 + i]      =

  1157.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1158.         h->block_offset[48 + 16 + i] =

  1159.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1160.     }

  1161. 

  1162.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1163.         if (!s->last_pic->f.data[0]) {

  1164.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1165.             ret = get_buffer(avctx, s->last_pic);

  1166.             if (ret < 0)

  1167.                 return ret;

  1168.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1169.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1170.                    s->last_pic->f.linesize[1]);

  1171.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1172.                    s->last_pic->f.linesize[2]);

  1173.         }

  1174. 

  1175.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1176.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1177.             ret = get_buffer(avctx, s->next_pic);

  1178.             if (ret < 0)

  1179.                 return ret;

  1180.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1181.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1182.                    s->next_pic->f.linesize[1]);

  1183.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1184.                    s->next_pic->f.linesize[2]);

  1185.         }

  1186.     }

  1187. 

  1188.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1189.         av_log(h->avctx, AV_LOG_DEBUG,

  1190.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1191.                av_get_picture_type_char(h->pict_type),

  1192.                s->halfpel_flag, s->thirdpel_flag,

  1193.                s->adaptive_quant, h->qscale, h->slice_num);

  1194. 

  1195.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1196.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1197.         avctx->skip_frame >= AVDISCARD_ALL)

  1198.         return 0;

  1199. 

  1200.     if (s->next_p_frame_damaged) {

  1201.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1202.             return 0;

  1203.         else

  1204.             s->next_p_frame_damaged = 0;

  1205.     }

  1206. 

  1207.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1208.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1209. 

  1210.         if (h->frame_num_offset < 0)

  1211.             h->frame_num_offset += 256;

  1212.         if (h->frame_num_offset == 0 ||

  1213.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1214.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1215.             return -1;

  1216.         }

  1217.     } else {

  1218.         h->prev_frame_num        = h->frame_num;

  1219.         h->frame_num             = h->slice_num;

  1220.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1221. 

  1222.         if (h->prev_frame_num_offset < 0)

  1223.             h->prev_frame_num_offset += 256;

  1224.     }

  1225. 

  1226.     for (m = 0; m < 2; m++) {

  1227.         int i;

  1228.         for (i = 0; i < 4; i++) {

  1229.             int j;

  1230.             for (j = -1; j < 4; j++)

  1231.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1232.             if (i < 3)

  1233.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1234.         }

  1235.     }

  1236. 

  1237.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1238.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1239.             unsigned mb_type;

  1240.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1241. 

  1242.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1243.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1244.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1245.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1246.                 h->gb.size_in_bits = 8 * buf_size;

  1247. 

  1248.                 if (svq3_decode_slice_header(avctx))

  1249.                     return -1;

  1250. 

  1251.                 /* TODO: support s->mb_skip_run */

  1252.             }

  1253. 

  1254.             mb_type = svq3_get_ue_golomb(&h->gb);

  1255. 

  1256.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1257.                 mb_type += 8;

  1258.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1259.                 mb_type += 4;

  1260.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1261.                 av_log(h->avctx, AV_LOG_ERROR,

  1262.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1263.                 return -1;

  1264.             }

  1265. 

  1266.             if (mb_type != 0)

  1267.                 ff_h264_hl_decode_mb(h);

  1268. 

  1269.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1270.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1271.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1272.         }

  1273. 

  1274.         ff_draw_horiz_band(avctx, &s->cur_pic->f,

  1275.                            s->last_pic->f.data[0] ? &s->last_pic->f : NULL,

  1276.                            16 * h->mb_y, 16, h->picture_structure, 0,

  1277.                            h->low_delay);

  1278.     }

  1279. 

  1280.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1281.         ret = av_frame_ref(data, &s->cur_pic->f);

  1282.     else if (s->last_pic->f.data[0])

  1283.         ret = av_frame_ref(data, &s->last_pic->f);

  1284.     if (ret < 0)

  1285.         return ret;

  1286. 

  1287.     /* Do not output the last pic after seeking. */

  1288.     if (s->last_pic->f.data[0] || h->low_delay)

  1289.         *got_frame = 1;

  1290. 

  1291.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1292.         FFSWAP(H264Picture*, s->cur_pic, s->next_pic);

  1293.     } else {

  1294.         av_frame_unref(&s->cur_pic->f);

  1295.     }

  1296. 

  1297.     return buf_size;

  1298. }

  1299. 

  1300. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1301. {

  1302.     SVQ3Context *s = avctx->priv_data;

  1303.     H264Context *h = &s->h;

  1304. 

  1305.     free_picture(avctx, s->cur_pic);

  1306.     free_picture(avctx, s->next_pic);

  1307.     free_picture(avctx, s->last_pic);

  1308.     av_freep(&s->cur_pic);

  1309.     av_freep(&s->next_pic);

  1310.     av_freep(&s->last_pic);

  1311. 

  1312.     av_frame_unref(&h->cur_pic.f);

  1313. 

  1314.     ff_h264_free_context(h);

  1315. 

  1316.     return 0;

  1317. }

  1318. 

  1319. AVCodec ff_svq3_decoder = {

  1320.     .name           = "svq3",

  1321.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1322.     .type           = AVMEDIA_TYPE_VIDEO,

  1323.     .id             = AV_CODEC_ID_SVQ3,

  1324.     .priv_data_size = sizeof(SVQ3Context),

  1325.     .init           = svq3_decode_init,

  1326.     .close          = svq3_decode_end,

  1327.     .decode         = svq3_decode_frame,

  1328.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1329.                       CODEC_CAP_DR1             |

  1330.                       CODEC_CAP_DELAY,

  1331.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1332.                                                      AV_PIX_FMT_NONE},

  1333. };