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.
70890 Commits
32 Branches
981MB
Tree: c27f53b9fb
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'c27f53b9fb'
${ noResults }
FFmpeg/libavcodec/svq3.c

1391 lines
49KB
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. 

  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. #include "tpeldsp.h"

  58. #include "vdpau_internal.h"

  59. 

  60. #if CONFIG_ZLIB

  61. #include <zlib.h>

  62. #endif

  63. 

  64. #include "svq1.h"

  65. #include "svq3.h"

  66. 

  67. /**

  68.  * @file

  69.  * svq3 decoder.

  70.  */

  71. 

  72. typedef struct SVQ3Context {

  73.     H264Context h;

  74.     HpelDSPContext hdsp;

  75.     TpelDSPContext tdsp;

  76.     H264Picture *cur_pic;

  77.     H264Picture *next_pic;

  78.     H264Picture *last_pic;

  79.     int halfpel_flag;

  80.     int thirdpel_flag;

  81.     int unknown_flag;

  82.     int next_slice_index;

  83.     uint32_t watermark_key;

  84.     uint8_t *buf;

  85.     int buf_size;

  86.     int adaptive_quant;

  87.     int next_p_frame_damaged;

  88.     int h_edge_pos;

  89.     int v_edge_pos;

  90.     int last_frame_output;

  91. } SVQ3Context;

  92. 

  93. #define FULLPEL_MODE  1

  94. #define HALFPEL_MODE  2

  95. #define THIRDPEL_MODE 3

  96. #define PREDICT_MODE  4

  97. 

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

  99.  * o-->o-->o   o

  100.  *         |  /|

  101.  * o   o   o / o

  102.  * | / |   |/  |

  103.  * o   o   o   o

  104.  *   /

  105.  * o-->o-->o-->o

  106.  */

  107. static const uint8_t svq3_scan[16] = {

  108.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  109.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

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

  111.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  112. };

  113. 

  114. static const uint8_t luma_dc_zigzag_scan[16] = {

  115.     0 * 16 + 0 * 64, 1 * 16 + 0 * 64, 2 * 16 + 0 * 64, 0 * 16 + 2 * 64,

  116.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

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

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

  119. };

  120. 

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

  122.     { 0, 0 },

  123.     { 1, 0 }, { 0, 1 },

  124.     { 0, 2 }, { 1, 1 }, { 2, 0 },

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

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

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

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

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

  130.     { 4, 4 }

  131. };

  132. 

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

  134.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

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

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

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

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

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

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

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

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

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

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

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

  146. };

  147. 

  148. static const struct {

  149.     uint8_t run;

  150.     uint8_t level;

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

  152.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  153.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  154.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  155.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  156. };

  157. 

  158. static const uint32_t svq3_dequant_coeff[32] = {

  159.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  160.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  161.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  162.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  163. };

  164. 

  165. static int svq3_decode_end(AVCodecContext *avctx);

  166. 

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

  168. {

  169.     const int qmul = svq3_dequant_coeff[qp];

  170. #define stride 16

  171.     int i;

  172.     int temp[16];

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

  174. 

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

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

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

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

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

  180. 

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

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

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

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

  185.     }

  186. 

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

  188.         const int offset = x_offset[i];

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

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

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

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

  193. 

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

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

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

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

  198.     }

  199. }

  200. #undef stride

  201. 

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

  203.                         int stride, int qp, int dc)

  204. {

  205.     const int qmul = svq3_dequant_coeff[qp];

  206.     int i;

  207. 

  208.     if (dc) {

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

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

  211.         block[0] = 0;

  212.     }

  213. 

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

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

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

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

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

  219. 

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

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

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

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

  224.     }

  225. 

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

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

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

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

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

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

  232. 

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

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

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

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

  237.     }

  238. 

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

  240. }

  241. 

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

  243.                                     int index, const int type)

  244. {

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

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

  247. 

  248.     int run, level, sign, limit;

  249.     unsigned vlc;

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

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

  252. 

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

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

  255.             if ((int32_t)vlc < 0)

  256.                 return -1;

  257. 

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

  259.             vlc      = vlc + 1 >> 1;

  260. 

  261.             if (type == 3) {

  262.                 if (vlc < 3) {

  263.                     run   = 0;

  264.                     level = vlc;

  265.                 } else if (vlc < 4) {

  266.                     run   = 1;

  267.                     level = 1;

  268.                 } else {

  269.                     run   = vlc & 0x3;

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

  271.                 }

  272.             } else {

  273.                 if (vlc < 16U) {

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

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

  276.                 } else if (intra) {

  277.                     run   = vlc & 0x7;

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

  279.                 } else {

  280.                     run   = vlc & 0xF;

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

  282.                 }

  283.             }

  284. 

  285. 

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

  287.                 return -1;

  288. 

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

  290.         }

  291. 

  292.         if (type != 2) {

  293.             break;

  294.         }

  295.     }

  296. 

  297.     return 0;

  298. }

  299. 

  300. static inline void svq3_mc_dir_part(SVQ3Context *s,

  301.                                     int x, int y, int width, int height,

  302.                                     int mx, int my, int dxy,

  303.                                     int thirdpel, int dir, int avg)

  304. {

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

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

  307.     uint8_t *src, *dest;

  308.     int i, emu = 0;

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

  310. 

  311.     mx += x;

  312.     my += y;

  313. 

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

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

  316.         emu = 1;

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

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

  319.     }

  320. 

  321.     /* form component predictions */

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

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

  324. 

  325.     if (emu) {

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

  327.                                  h->linesize, h->linesize,

  328.                                  width + 1, height + 1,

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

  330.         src = h->edge_emu_buffer;

  331.     }

  332.     if (thirdpel)

  333.         (avg ? s->tdsp.avg_tpel_pixels_tab

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

  335.                                                  width, height);

  336.     else

  337.         (avg ? s->hdsp.avg_pixels_tab

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

  339.                                                        height);

  340. 

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

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

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

  344.         width  = width  >> 1;

  345.         height = height >> 1;

  346.         blocksize++;

  347. 

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

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

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

  351. 

  352.             if (emu) {

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

  354.                                          h->uvlinesize, h->uvlinesize,

  355.                                          width + 1, height + 1,

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

  357.                                          s->v_edge_pos >> 1);

  358.                 src = h->edge_emu_buffer;

  359.             }

  360.             if (thirdpel)

  361.                 (avg ? s->tdsp.avg_tpel_pixels_tab

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

  363.                                                          h->uvlinesize,

  364.                                                          width, height);

  365.             else

  366.                 (avg ? s->hdsp.avg_pixels_tab

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

  368.                                                                h->uvlinesize,

  369.                                                                height);

  370.         }

  371.     }

  372. }

  373. 

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

  375.                               int dir, int avg)

  376. {

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

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

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

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

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

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

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

  384. 

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

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

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

  388.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  389.             int dxy;

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

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

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

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

  394. 

  395.             if (mode != PREDICT_MODE) {

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

  397.             } else {

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

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

  400. 

  401.                 if (dir == 0) {

  402.                     mx = mx * h->frame_num_offset /

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

  404.                     my = my * h->frame_num_offset /

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

  406.                 } else {

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

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

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

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

  411.                 }

  412.             }

  413. 

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

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

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

  417. 

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

  419.             if (mode == PREDICT_MODE) {

  420.                 dx = dy = 0;

  421.             } else {

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

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

  424. 

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

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

  427.                     return -1;

  428.                 }

  429.             }

  430. 

  431.             /* compute motion vector */

  432.             if (mode == THIRDPEL_MODE) {

  433.                 int fx, fy;

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

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

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

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

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

  439. 

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

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

  442.                 mx += mx;

  443.                 my += my;

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

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

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

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

  448. 

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

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

  451.                 mx *= 3;

  452.                 my *= 3;

  453.             } else {

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

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

  456. 

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

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

  459.                 mx *= 6;

  460.                 my *= 6;

  461.             }

  462. 

  463.             /* update mv_cache */

  464.             if (mode != PREDICT_MODE) {

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

  466. 

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

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

  469. 

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

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

  472.                 }

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

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

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

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

  477.             }

  478. 

  479.             /* write back motion vectors */

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

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

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

  483.         }

  484. 

  485.     return 0;

  486. }

  487. 

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

  489. {

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

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

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

  493.     int cbp = 0;

  494.     uint32_t vlc;

  495.     int8_t *top, *left;

  496.     const int mb_xy         = h->mb_xy;

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

  498. 

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

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

  501.     h->topright_samples_available = 0xFFFF;

  502. 

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

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

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

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

  507.                              0, 0, 0, 0, 0, 0);

  508. 

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

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

  511.                                  0, 0, 0, 0, 1, 1);

  512. 

  513.             mb_type = MB_TYPE_SKIP;

  514.         } else {

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

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

  517.                 return -1;

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

  519.                 return -1;

  520. 

  521.             mb_type = MB_TYPE_16x16;

  522.         }

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

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

  525.             mode = THIRDPEL_MODE;

  526.         else if (s->halfpel_flag &&

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

  528.             mode = HALFPEL_MODE;

  529.         else

  530.             mode = FULLPEL_MODE;

  531. 

  532.         /* fill caches */

  533.         /* note ref_cache should contain here:

  534.          *  ????????

  535.          *  ???11111

  536.          *  N??11111

  537.          *  N??11111

  538.          *  N??11111

  539.          */

  540. 

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

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

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

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

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

  546.             } else {

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

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

  549.             }

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

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

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

  553.                        4 * 2 * sizeof(int16_t));

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

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

  556. 

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

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

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

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

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

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

  563.                 } else

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

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

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

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

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

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

  570.                 } else

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

  572.             } else

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

  574.                        PART_NOT_AVAILABLE, 8);

  575. 

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

  577.                 break;

  578.         }

  579. 

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

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

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

  583.                 return -1;

  584.         } else {        /* AV_PICTURE_TYPE_B */

  585.             if (mb_type != 2) {

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

  587.                     return -1;

  588.             } else {

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

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

  591.                            0, 4 * 2 * sizeof(int16_t));

  592.             }

  593.             if (mb_type != 1) {

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

  595.                     return -1;

  596.             } else {

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

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

  599.                            0, 4 * 2 * sizeof(int16_t));

  600.             }

  601.         }

  602. 

  603.         mb_type = MB_TYPE_16x16;

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

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

  606. 

  607.         if (mb_type == 8) {

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

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

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

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

  612.                     h->left_samples_available = 0x5F5F;

  613.             }

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

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

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

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

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

  619. 

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

  621.                     h->top_samples_available = 0x33FF;

  622.             }

  623. 

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

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

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

  627. 

  628.                 if (vlc >= 25U) {

  629.                     av_log(h->avctx, AV_LOG_ERROR,

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

  631.                     return -1;

  632.                 }

  633. 

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

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

  636. 

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

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

  639. 

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

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

  642.                     return -1;

  643.                 }

  644.             }

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

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

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

  648.         }

  649. 

  650.         write_back_intra_pred_mode(h);

  651. 

  652.         if (mb_type == 8) {

  653.             ff_h264_check_intra4x4_pred_mode(h);

  654. 

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

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

  657.         } else {

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

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

  660. 

  661.             h->top_samples_available  = 0x33FF;

  662.             h->left_samples_available = 0x5F5F;

  663.         }

  664. 

  665.         mb_type = MB_TYPE_INTRA4x4;

  666.     } else {                      /* INTRA16x16 */

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

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

  669. 

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

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

  672.             return h->intra16x16_pred_mode;

  673.         }

  674. 

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

  676.         mb_type = MB_TYPE_INTRA16x16;

  677.     }

  678. 

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

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

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

  682.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  686.                        0, 4 * 2 * sizeof(int16_t));

  687.         }

  688.     }

  689.     if (!IS_INTRA4x4(mb_type)) {

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

  691.     }

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

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

  694.     }

  695. 

  696.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  700.             return -1;

  701.         }

  702. 

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

  704.                                 : golomb_to_inter_cbp[vlc];

  705.     }

  706.     if (IS_INTRA16x16(mb_type) ||

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

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

  709. 

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

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

  712.             return -1;

  713.         }

  714.     }

  715.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  719.             av_log(h->avctx, AV_LOG_ERROR,

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

  721.             return -1;

  722.         }

  723.     }

  724. 

  725.     if (cbp) {

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

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

  728. 

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

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

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

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

  733.                                  2 * (j & 2) + 4 * (i & 2))

  734.                               : (4 * i + j);

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

  736. 

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

  738.                         av_log(h->avctx, AV_LOG_ERROR,

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

  740.                         return -1;

  741.                     }

  742.                 }

  743.             }

  744. 

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

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

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

  748.                     av_log(h->avctx, AV_LOG_ERROR,

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

  750.                     return -1;

  751.                 }

  752. 

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

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

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

  756.                         k                                 = 16 * i + j;

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

  758. 

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

  760.                             av_log(h->avctx, AV_LOG_ERROR,

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

  762.                             return -1;

  763.                         }

  764.                     }

  765.                 }

  766.             }

  767.         }

  768.     }

  769. 

  770.     h->cbp                              = cbp;

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

  772. 

  773.     if (IS_INTRA(mb_type))

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

  775. 

  776.     return 0;

  777. }

  778. 

  779. static int svq3_decode_slice_header(AVCodecContext *avctx)

  780. {

  781.     SVQ3Context *s = avctx->priv_data;

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

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

  784.     const int mb_xy   = h->mb_xy;

  785.     int i, header;

  786.     unsigned slice_id;

  787. 

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

  789. 

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

  791.         /* TODO: what? */

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

  793.         return -1;

  794.     } else {

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

  796. 

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

  798.                               8 * show_bits(&h->gb, 8 * length) +

  799.                               8 * length;

  800. 

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

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

  803.             return -1;

  804.         }

  805. 

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

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

  808. 

  809.         if (s->watermark_key) {

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

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

  812.                     header ^ s->watermark_key);

  813.         }

  814.         if (length > 0) {

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

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

  817.         }

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

  819.     }

  820. 

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

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

  823.         return -1;

  824.     }

  825. 

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

  827. 

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

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

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

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

  832.     } else {

  833.         skip_bits1(&h->gb);

  834.         h->mb_skip_run = 0;

  835.     }

  836. 

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

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

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

  840. 

  841.     /* unknown fields */

  842.     skip_bits1(&h->gb);

  843. 

  844.     if (s->unknown_flag)

  845.         skip_bits1(&h->gb);

  846. 

  847.     skip_bits1(&h->gb);

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

  849. 

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

  851.         return AVERROR_INVALIDDATA;

  852. 

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

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

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

  856.                -1, 4 * sizeof(int8_t));

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

  858.                -1, 8 * sizeof(int8_t) * h->mb_x);

  859.     }

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

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

  862.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  863. 

  864.         if (h->mb_x > 0)

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

  866.     }

  867. 

  868.     return 0;

  869. }

  870. 

  871. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  872. {

  873.     SVQ3Context *s = avctx->priv_data;

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

  875.     int m;

  876.     unsigned char *extradata;

  877.     unsigned char *extradata_end;

  878.     unsigned int size;

  879.     int marker_found = 0;

  880.     int ret;

  881. 

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

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

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

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

  886.         ret = AVERROR(ENOMEM);

  887.         goto fail;

  888.     }

  889. 

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

  891.         goto fail;

  892. 

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

  894.     ff_tpeldsp_init(&s->tdsp);

  895. 

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

  897.     h->is_complex      = 1;

  898.     h->sps.chroma_format_idc = 1;

  899.     h->picture_structure = PICT_FRAME;

  900.     avctx->pix_fmt     = AV_PIX_FMT_YUVJ420P;

  901.     avctx->color_range = AVCOL_RANGE_JPEG;

  902. 

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

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

  905. 

  906.     s->halfpel_flag  = 1;

  907.     s->thirdpel_flag = 1;

  908.     s->unknown_flag  = 0;

  909. 

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

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

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

  913.     if (extradata) {

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

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

  916.                 marker_found = 1;

  917.                 break;

  918.             }

  919.             extradata++;

  920.         }

  921.     }

  922. 

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

  924.     if (marker_found) {

  925.         GetBitContext gb;

  926.         int frame_size_code;

  927. 

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

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

  930.             ret = AVERROR_INVALIDDATA;

  931.             goto fail;

  932.         }

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

  934. 

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

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

  937.         switch (frame_size_code) {

  938.         case 0:

  939.             avctx->width  = 160;

  940.             avctx->height = 120;

  941.             break;

  942.         case 1:

  943.             avctx->width  = 128;

  944.             avctx->height =  96;

  945.             break;

  946.         case 2:

  947.             avctx->width  = 176;

  948.             avctx->height = 144;

  949.             break;

  950.         case 3:

  951.             avctx->width  = 352;

  952.             avctx->height = 288;

  953.             break;

  954.         case 4:

  955.             avctx->width  = 704;

  956.             avctx->height = 576;

  957.             break;

  958.         case 5:

  959.             avctx->width  = 240;

  960.             avctx->height = 180;

  961.             break;

  962.         case 6:

  963.             avctx->width  = 320;

  964.             avctx->height = 240;

  965.             break;

  966.         case 7:

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

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

  969.             break;

  970.         }

  971. 

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

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

  974. 

  975.         /* unknown fields */

  976.         skip_bits1(&gb);

  977.         skip_bits1(&gb);

  978.         skip_bits1(&gb);

  979.         skip_bits1(&gb);

  980. 

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

  982. 

  983.         /* unknown field */

  984.         skip_bits1(&gb);

  985. 

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

  987.             ret = AVERROR_INVALIDDATA;

  988.             goto fail;

  989.         }

  990. 

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

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

  993.         if (s->unknown_flag) {

  994. #if CONFIG_ZLIB

  995.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  996.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  997.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  1000.             int u4                    = svq3_get_ue_golomb(&gb);

  1001.             unsigned long buf_len     = watermark_width *

  1002.                                         watermark_height * 4;

  1003.             int offset                = get_bits_count(&gb) + 7 >> 3;

  1004.             uint8_t *buf;

  1005. 

  1006.             if (watermark_height <= 0 ||

  1007.                 (uint64_t)watermark_width * 4 > UINT_MAX / watermark_height) {

  1008.                 ret = -1;

  1009.                 goto fail;

  1010.             }

  1011. 

  1012.             buf = av_malloc(buf_len);

  1013.             if (!buf) {

  1014.                 ret = AVERROR(ENOMEM);

  1015.                 goto fail;

  1016.             }

  1017.             av_log(avctx, AV_LOG_DEBUG, "watermark size: %ux%u\n",

  1018.                    watermark_width, watermark_height);

  1019.             av_log(avctx, AV_LOG_DEBUG,

  1020.                    "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n",

  1021.                    u1, u2, u3, u4, offset);

  1022.             if (uncompress(buf, &buf_len, extradata + 8 + offset,

  1023.                            size - offset) != Z_OK) {

  1024.                 av_log(avctx, AV_LOG_ERROR,

  1025.                        "could not uncompress watermark logo\n");

  1026.                 av_free(buf);

  1027.                 ret = -1;

  1028.                 goto fail;

  1029.             }

  1030.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1031.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1032.             av_log(avctx, AV_LOG_DEBUG,

  1033.                    "watermark key %#"PRIx32"\n", s->watermark_key);

  1034.             av_free(buf);

  1035. #else

  1036.             av_log(avctx, AV_LOG_ERROR,

  1037.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1038.             ret = -1;

  1039.             goto fail;

  1040. #endif

  1041.         }

  1042.     }

  1043. 

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

  1045.     h->height = avctx->height;

  1046.     h->mb_width  = (h->width + 15) / 16;

  1047.     h->mb_height = (h->height + 15) / 16;

  1048.     h->mb_stride = h->mb_width + 1;

  1049.     h->mb_num    = h->mb_width * h->mb_height;

  1050.     h->b_stride = 4 * h->mb_width;

  1051.     s->h_edge_pos = h->mb_width * 16;

  1052.     s->v_edge_pos = h->mb_height * 16;

  1053. 

  1054.     if ((ret = ff_h264_alloc_tables(h)) < 0) {

  1055.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1056.         goto fail;

  1057.     }

  1058. 

  1059.     return 0;

  1060. fail:

  1061.     svq3_decode_end(avctx);

  1062.     return ret;

  1063. }

  1064. 

  1065. static void free_picture(AVCodecContext *avctx, H264Picture *pic)

  1066. {

  1067.     int i;

  1068.     for (i = 0; i < 2; i++) {

  1069.         av_buffer_unref(&pic->motion_val_buf[i]);

  1070.         av_buffer_unref(&pic->ref_index_buf[i]);

  1071.     }

  1072.     av_buffer_unref(&pic->mb_type_buf);

  1073. 

  1074.     av_frame_unref(&pic->f);

  1075. }

  1076. 

  1077. static int get_buffer(AVCodecContext *avctx, H264Picture *pic)

  1078. {

  1079.     SVQ3Context *s = avctx->priv_data;

  1080.     H264Context *h = &s->h;

  1081.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1082.     const int mb_array_size = h->mb_stride * h->mb_height;

  1083.     const int b4_stride     = h->mb_width * 4 + 1;

  1084.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1085.     int ret;

  1086. 

  1087.     if (!pic->motion_val_buf[0]) {

  1088.         int i;

  1089. 

  1090.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1091.         if (!pic->mb_type_buf)

  1092.             return AVERROR(ENOMEM);

  1093.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1094. 

  1095.         for (i = 0; i < 2; i++) {

  1096.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1097.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1098.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1099.                 ret = AVERROR(ENOMEM);

  1100.                 goto fail;

  1101.             }

  1102. 

  1103.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1104.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1105.         }

  1106.     }

  1107.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1108. 

  1109.     ret = ff_get_buffer(avctx, &pic->f,

  1110.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1111.     if (ret < 0)

  1112.         goto fail;

  1113. 

  1114.     if (!h->edge_emu_buffer) {

  1115.         h->edge_emu_buffer = av_mallocz_array(pic->f.linesize[0], 17);

  1116.         if (!h->edge_emu_buffer)

  1117.             return AVERROR(ENOMEM);

  1118.     }

  1119. 

  1120.     h->linesize   = pic->f.linesize[0];

  1121.     h->uvlinesize = pic->f.linesize[1];

  1122. 

  1123.     return 0;

  1124. fail:

  1125.     free_picture(avctx, pic);

  1126.     return ret;

  1127. }

  1128. 

  1129. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1130.                              int *got_frame, AVPacket *avpkt)

  1131. {

  1132.     SVQ3Context *s     = avctx->priv_data;

  1133.     H264Context *h     = &s->h;

  1134.     int buf_size       = avpkt->size;

  1135.     int left;

  1136.     uint8_t *buf;

  1137.     int ret, m, i;

  1138. 

  1139.     /* special case for last picture */

  1140.     if (buf_size == 0) {

  1141.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1142.             ret = av_frame_ref(data, &s->next_pic->f);

  1143.             if (ret < 0)

  1144.                 return ret;

  1145.             s->last_frame_output = 1;

  1146.             *got_frame          = 1;

  1147.         }

  1148.         return 0;

  1149.     }

  1150. 

  1151.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1152. 

  1153.     if (s->watermark_key) {

  1154.         av_fast_padded_malloc(&s->buf, &s->buf_size, buf_size);

  1155.         if (!s->buf)

  1156.             return AVERROR(ENOMEM);

  1157.         memcpy(s->buf, avpkt->data, buf_size);

  1158.         buf = s->buf;

  1159.     } else {

  1160.         buf = avpkt->data;

  1161.     }

  1162. 

  1163.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1164. 

  1165.     if (svq3_decode_slice_header(avctx))

  1166.         return -1;

  1167. 

  1168.     h->pict_type = h->slice_type;

  1169. 

  1170.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1171.         FFSWAP(H264Picture*, s->next_pic, s->last_pic);

  1172. 

  1173.     av_frame_unref(&s->cur_pic->f);

  1174. 

  1175.     /* for skipping the frame */

  1176.     s->cur_pic->f.pict_type = h->pict_type;

  1177.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1178. 

  1179.     ret = get_buffer(avctx, s->cur_pic);

  1180.     if (ret < 0)

  1181.         return ret;

  1182. 

  1183.     h->cur_pic_ptr = s->cur_pic;

  1184.     av_frame_unref(&h->cur_pic.f);

  1185.     memcpy(&h->cur_pic.tf, &s->cur_pic->tf, sizeof(h->cur_pic) - offsetof(H264Picture, tf));

  1186.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1187.     if (ret < 0)

  1188.         return ret;

  1189. 

  1190.     for (i = 0; i < 16; i++) {

  1191.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1192.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1193.     }

  1194.     for (i = 0; i < 16; i++) {

  1195.         h->block_offset[16 + i]      =

  1196.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1197.         h->block_offset[48 + 16 + i] =

  1198.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1199.     }

  1200. 

  1201.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1202.         if (!s->last_pic->f.data[0]) {

  1203.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1204.             av_frame_unref(&s->last_pic->f);

  1205.             ret = get_buffer(avctx, s->last_pic);

  1206.             if (ret < 0)

  1207.                 return ret;

  1208.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1209.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1210.                    s->last_pic->f.linesize[1]);

  1211.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1212.                    s->last_pic->f.linesize[2]);

  1213.         }

  1214. 

  1215.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1216.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1217.             av_frame_unref(&s->next_pic->f);

  1218.             ret = get_buffer(avctx, s->next_pic);

  1219.             if (ret < 0)

  1220.                 return ret;

  1221.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1222.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1223.                    s->next_pic->f.linesize[1]);

  1224.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1225.                    s->next_pic->f.linesize[2]);

  1226.         }

  1227.     }

  1228. 

  1229.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1230.         av_log(h->avctx, AV_LOG_DEBUG,

  1231.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1232.                av_get_picture_type_char(h->pict_type),

  1233.                s->halfpel_flag, s->thirdpel_flag,

  1234.                s->adaptive_quant, h->slice_ctx[0].qscale, h->slice_num);

  1235. 

  1236.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1237.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1238.         avctx->skip_frame >= AVDISCARD_ALL)

  1239.         return 0;

  1240. 

  1241.     if (s->next_p_frame_damaged) {

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

  1243.             return 0;

  1244.         else

  1245.             s->next_p_frame_damaged = 0;

  1246.     }

  1247. 

  1248.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1249.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1250. 

  1251.         if (h->frame_num_offset < 0)

  1252.             h->frame_num_offset += 256;

  1253.         if (h->frame_num_offset == 0 ||

  1254.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1255.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1256.             return -1;

  1257.         }

  1258.     } else {

  1259.         h->prev_frame_num        = h->frame_num;

  1260.         h->frame_num             = h->slice_num;

  1261.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1262. 

  1263.         if (h->prev_frame_num_offset < 0)

  1264.             h->prev_frame_num_offset += 256;

  1265.     }

  1266. 

  1267.     for (m = 0; m < 2; m++) {

  1268.         int i;

  1269.         for (i = 0; i < 4; i++) {

  1270.             int j;

  1271.             for (j = -1; j < 4; j++)

  1272.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1273.             if (i < 3)

  1274.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1275.         }

  1276.     }

  1277. 

  1278.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1279.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1280.             unsigned mb_type;

  1281.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1282. 

  1283.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1284.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1285.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1286.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1287.                 h->gb.size_in_bits = 8 * buf_size;

  1288. 

  1289.                 if (svq3_decode_slice_header(avctx))

  1290.                     return -1;

  1291. 

  1292.                 /* TODO: support s->mb_skip_run */

  1293.             }

  1294. 

  1295.             mb_type = svq3_get_ue_golomb(&h->gb);

  1296. 

  1297.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1298.                 mb_type += 8;

  1299.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1300.                 mb_type += 4;

  1301.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1302.                 av_log(h->avctx, AV_LOG_ERROR,

  1303.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1304.                 return -1;

  1305.             }

  1306. 

  1307.             if (mb_type != 0 || h->cbp)

  1308.                 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);

  1309. 

  1310.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1311.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1312.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1313.         }

  1314. 

  1315.         ff_draw_horiz_band(avctx, &s->cur_pic->f,

  1316.                            s->last_pic->f.data[0] ? &s->last_pic->f : NULL,

  1317.                            16 * h->mb_y, 16, h->picture_structure, 0,

  1318.                            h->low_delay);

  1319.     }

  1320. 

  1321.     left = buf_size*8 - get_bits_count(&h->gb);

  1322. 

  1323.     if (h->mb_y != h->mb_height || h->mb_x != h->mb_width) {

  1324.         av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, h->mb_y, h->mb_x, left);

  1325.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1326.     }

  1327. 

  1328.     if (left < 0) {

  1329.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1330.         return -1;

  1331.     }

  1332. 

  1333.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1334.         ret = av_frame_ref(data, &s->cur_pic->f);

  1335.     else if (s->last_pic->f.data[0])

  1336.         ret = av_frame_ref(data, &s->last_pic->f);

  1337.     if (ret < 0)

  1338.         return ret;

  1339. 

  1340.     /* Do not output the last pic after seeking. */

  1341.     if (s->last_pic->f.data[0] || h->low_delay)

  1342.         *got_frame = 1;

  1343. 

  1344.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1345.         FFSWAP(H264Picture*, s->cur_pic, s->next_pic);

  1346.     } else {

  1347.         av_frame_unref(&s->cur_pic->f);

  1348.     }

  1349. 

  1350.     return buf_size;

  1351. }

  1352. 

  1353. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1354. {

  1355.     SVQ3Context *s = avctx->priv_data;

  1356.     H264Context *h = &s->h;

  1357. 

  1358.     free_picture(avctx, s->cur_pic);

  1359.     free_picture(avctx, s->next_pic);

  1360.     free_picture(avctx, s->last_pic);

  1361.     av_freep(&s->cur_pic);

  1362.     av_freep(&s->next_pic);

  1363.     av_freep(&s->last_pic);

  1364. 

  1365.     av_frame_unref(&h->cur_pic.f);

  1366. 

  1367.     ff_h264_free_context(h);

  1368. 

  1369.     av_freep(&s->buf);

  1370.     s->buf_size = 0;

  1371.     av_freep(&h->edge_emu_buffer);

  1372. 

  1373.     return 0;

  1374. }

  1375. 

  1376. AVCodec ff_svq3_decoder = {

  1377.     .name           = "svq3",

  1378.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1379.     .type           = AVMEDIA_TYPE_VIDEO,

  1380.     .id             = AV_CODEC_ID_SVQ3,

  1381.     .priv_data_size = sizeof(SVQ3Context),

  1382.     .init           = svq3_decode_init,

  1383.     .close          = svq3_decode_end,

  1384.     .decode         = svq3_decode_frame,

  1385.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1386.                       CODEC_CAP_DR1             |

  1387.                       CODEC_CAP_DELAY,

  1388.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1389.                                                      AV_PIX_FMT_NONE},

  1390. };