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.
41838 Commits
32 Branches
776MB
Tree: 3058872c29
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '3058872c29'
${ noResults }
FFmpeg/libavcodec/svq3.c

1367 lines
49KB
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. #include "tpeldsp.h"

  58. 

  59. #if CONFIG_ZLIB

  60. #include <zlib.h>

  61. #endif

  62. 

  63. #include "svq1.h"

  64. #include "svq3.h"

  65. 

  66. /**

  67.  * @file

  68.  * svq3 decoder.

  69.  */

  70. 

  71. typedef struct SVQ3Context {

  72.     H264Context h;

  73.     HpelDSPContext hdsp;

  74.     TpelDSPContext tdsp;

  75.     H264Picture *cur_pic;

  76.     H264Picture *next_pic;

  77.     H264Picture *last_pic;

  78.     int halfpel_flag;

  79.     int thirdpel_flag;

  80.     int unknown_flag;

  81.     int next_slice_index;

  82.     uint32_t watermark_key;

  83.     int adaptive_quant;

  84.     int next_p_frame_damaged;

  85.     int h_edge_pos;

  86.     int v_edge_pos;

  87.     int last_frame_output;

  88. } SVQ3Context;

  89. 

  90. #define FULLPEL_MODE  1

  91. #define HALFPEL_MODE  2

  92. #define THIRDPEL_MODE 3

  93. #define PREDICT_MODE  4

  94. 

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

  96.  * o-->o-->o   o

  97.  *         |  /|

  98.  * o   o   o / o

  99.  * | / |   |/  |

  100.  * o   o   o   o

  101.  *   /

  102.  * o-->o-->o-->o

  103.  */

  104. static const uint8_t svq3_scan[16] = {

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

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

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

  108.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  109. };

  110. 

  111. static const uint8_t luma_dc_zigzag_scan[16] = {

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

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

  114.     1 * 16 + 2 * 64, 2 * 16 + 2 * 64, 3 * 16 + 2 * 64, 0 * 16 + 3 * 64,

  115.     3 * 16 + 1 * 64, 1 * 16 + 3 * 64, 2 * 16 + 3 * 64, 3 * 16 + 3 * 64,

  116. };

  117. 

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

  119.     { 0, 0 },

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

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

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

  123.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  124.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  125.     { 2, 4 }, { 3, 3 }, { 4, 2 },

  126.     { 4, 3 }, { 3, 4 },

  127.     { 4, 4 }

  128. };

  129. 

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

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

  132.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

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

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

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

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

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

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

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

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

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

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

  143. };

  144. 

  145. static const struct {

  146.     uint8_t run;

  147.     uint8_t level;

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

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

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

  151.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  152.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  153. };

  154. 

  155. static const uint32_t svq3_dequant_coeff[32] = {

  156.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  157.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  158.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  159.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  160. };

  161. 

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

  163. {

  164.     const int qmul = svq3_dequant_coeff[qp];

  165. #define stride 16

  166.     int i;

  167.     int temp[16];

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

  169. 

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

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

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

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

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

  175. 

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

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

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

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

  180.     }

  181. 

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

  183.         const int offset = x_offset[i];

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

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

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

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

  188. 

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

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

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

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

  193.     }

  194. }

  195. #undef stride

  196. 

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

  198.                         int stride, int qp, int dc)

  199. {

  200.     const int qmul = svq3_dequant_coeff[qp];

  201.     int i;

  202. 

  203.     if (dc) {

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

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

  206.         block[0] = 0;

  207.     }

  208. 

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

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

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

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

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

  214. 

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

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

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

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

  219.     }

  220. 

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

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

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

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

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

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

  227. 

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

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

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

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

  232.     }

  233. 

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

  235. }

  236. 

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

  238.                                     int index, const int type)

  239. {

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

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

  242. 

  243.     int run, level, limit;

  244.     unsigned vlc;

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

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

  247. 

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

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

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

  251.             vlc      = vlc + 1 >> 1;

  252. 

  253.             if (type == 3) {

  254.                 if (vlc < 3) {

  255.                     run   = 0;

  256.                     level = vlc;

  257.                 } else if (vlc < 4) {

  258.                     run   = 1;

  259.                     level = 1;

  260.                 } else {

  261.                     run   = vlc & 0x3;

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

  263.                 }

  264.             } else {

  265.                 if (vlc < 16) {

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

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

  268.                 } else if (intra) {

  269.                     run   = vlc & 0x7;

  270.                     level = (vlc >> 3) +

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

  272.                 } else {

  273.                     run   = vlc & 0xF;

  274.                     level = (vlc >> 4) +

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

  276.                 }

  277.             }

  278. 

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

  280.                 return -1;

  281. 

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

  283.         }

  284. 

  285.         if (type != 2) {

  286.             break;

  287.         }

  288.     }

  289. 

  290.     return 0;

  291. }

  292. 

  293. static inline void svq3_mc_dir_part(SVQ3Context *s,

  294.                                     int x, int y, int width, int height,

  295.                                     int mx, int my, int dxy,

  296.                                     int thirdpel, int dir, int avg)

  297. {

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

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

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

  301.     uint8_t *src, *dest;

  302.     int i, emu = 0;

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

  304. 

  305.     mx += x;

  306.     my += y;

  307. 

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

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

  310.         emu = 1;

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

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

  313.     }

  314. 

  315.     /* form component predictions */

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

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

  318. 

  319.     if (emu) {

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

  321.                                  sl->linesize, sl->linesize,

  322.                                  width + 1, height + 1,

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

  324.         src = sl->edge_emu_buffer;

  325.     }

  326.     if (thirdpel)

  327.         (avg ? s->tdsp.avg_tpel_pixels_tab

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

  329.                                                  width, height);

  330.     else

  331.         (avg ? s->hdsp.avg_pixels_tab

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

  333.                                                        height);

  334. 

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

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

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

  338.         width  = width  >> 1;

  339.         height = height >> 1;

  340.         blocksize++;

  341. 

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

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

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

  345. 

  346.             if (emu) {

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

  348.                                          sl->uvlinesize, sl->uvlinesize,

  349.                                          width + 1, height + 1,

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

  351.                                          s->v_edge_pos >> 1);

  352.                 src = sl->edge_emu_buffer;

  353.             }

  354.             if (thirdpel)

  355.                 (avg ? s->tdsp.avg_tpel_pixels_tab

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

  357.                                                          sl->uvlinesize,

  358.                                                          width, height);

  359.             else

  360.                 (avg ? s->hdsp.avg_pixels_tab

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

  362.                                                                sl->uvlinesize,

  363.                                                                height);

  364.         }

  365.     }

  366. }

  367. 

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

  369.                               int dir, int avg)

  370. {

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

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

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

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

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

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

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

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

  379. 

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

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

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

  383.                              (4 * sl->mb_y + (i >> 2)) * h->b_stride;

  384.             int dxy;

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

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

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

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

  389. 

  390.             if (mode != PREDICT_MODE) {

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

  392.             } else {

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

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

  395. 

  396.                 if (dir == 0) {

  397.                     mx = mx * h->frame_num_offset /

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

  399.                     my = my * h->frame_num_offset /

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

  401.                 } else {

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

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

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

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

  406.                 }

  407.             }

  408. 

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

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

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

  412. 

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

  414.             if (mode == PREDICT_MODE) {

  415.                 dx = dy = 0;

  416.             } else {

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

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

  419. 

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

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

  422.                     return -1;

  423.                 }

  424.             }

  425. 

  426.             /* compute motion vector */

  427.             if (mode == THIRDPEL_MODE) {

  428.                 int fx, fy;

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

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

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

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

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

  434. 

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

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

  437.                 mx += mx;

  438.                 my += my;

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

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

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

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

  443. 

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

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

  446.                 mx *= 3;

  447.                 my *= 3;

  448.             } else {

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

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

  451. 

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

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

  454.                 mx *= 6;

  455.                 my *= 6;

  456.             }

  457. 

  458.             /* update mv_cache */

  459.             if (mode != PREDICT_MODE) {

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

  461. 

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

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

  464. 

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

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

  467.                 }

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

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

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

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

  472.             }

  473. 

  474.             /* write back motion vectors */

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

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

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

  478.         }

  479. 

  480.     return 0;

  481. }

  482. 

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

  484. {

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

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

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

  488.     int cbp = 0;

  489.     uint32_t vlc;

  490.     int8_t *top, *left;

  491.     const int mb_xy         = sl->mb_xy;

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

  493. 

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

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

  496.     sl->topright_samples_available = 0xFFFF;

  497. 

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

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

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

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

  502.                              0, 0, 0, 0, 0, 0);

  503. 

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

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

  506.                                  0, 0, 0, 0, 1, 1);

  507. 

  508.             mb_type = MB_TYPE_SKIP;

  509.         } else {

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

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

  512.                 return -1;

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

  514.                 return -1;

  515. 

  516.             mb_type = MB_TYPE_16x16;

  517.         }

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

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

  520.             mode = THIRDPEL_MODE;

  521.         else if (s->halfpel_flag &&

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

  523.             mode = HALFPEL_MODE;

  524.         else

  525.             mode = FULLPEL_MODE;

  526. 

  527.         /* fill caches */

  528.         /* note ref_cache should contain here:

  529.          *  ????????

  530.          *  ???11111

  531.          *  N??11111

  532.          *  N??11111

  533.          *  N??11111

  534.          */

  535. 

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

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

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

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

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

  541.             } else {

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

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

  544.             }

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

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

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

  548.                        4 * 2 * sizeof(int16_t));

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

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

  551. 

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

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

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

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

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

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

  558.                 } else

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

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

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

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

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

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

  565.                 } else

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

  567.             } else

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

  569.                        PART_NOT_AVAILABLE, 8);

  570. 

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

  572.                 break;

  573.         }

  574. 

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

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

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

  578.                 return -1;

  579.         } else {        /* AV_PICTURE_TYPE_B */

  580.             if (mb_type != 2) {

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

  582.                     return -1;

  583.             } else {

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

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

  586.                            0, 4 * 2 * sizeof(int16_t));

  587.             }

  588.             if (mb_type != 1) {

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

  590.                     return -1;

  591.             } else {

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

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

  594.                            0, 4 * 2 * sizeof(int16_t));

  595.             }

  596.         }

  597. 

  598.         mb_type = MB_TYPE_16x16;

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

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

  601. 

  602.         if (mb_type == 8) {

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

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

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

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

  607.                     sl->left_samples_available = 0x5F5F;

  608.             }

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

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

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

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

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

  614. 

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

  616.                     sl->top_samples_available = 0x33FF;

  617.             }

  618. 

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

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

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

  622. 

  623.                 if (vlc >= 25) {

  624.                     av_log(h->avctx, AV_LOG_ERROR,

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

  626.                     return -1;

  627.                 }

  628. 

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

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

  631. 

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

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

  634. 

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

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

  637.                     return -1;

  638.                 }

  639.             }

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

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

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

  643.         }

  644. 

  645.         write_back_intra_pred_mode(h, sl);

  646. 

  647.         if (mb_type == 8) {

  648.             ff_h264_check_intra4x4_pred_mode(h, sl);

  649. 

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

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

  652.         } else {

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

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

  655. 

  656.             sl->top_samples_available  = 0x33FF;

  657.             sl->left_samples_available = 0x5F5F;

  658.         }

  659. 

  660.         mb_type = MB_TYPE_INTRA4x4;

  661.     } else {                      /* INTRA16x16 */

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

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

  664. 

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

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

  667.             return sl->intra16x16_pred_mode;

  668.         }

  669. 

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

  671.         mb_type = MB_TYPE_INTRA16x16;

  672.     }

  673. 

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

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

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

  677.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  681.                        0, 4 * 2 * sizeof(int16_t));

  682.         }

  683.     }

  684.     if (!IS_INTRA4x4(mb_type)) {

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

  686.     }

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

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

  689.     }

  690. 

  691.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  695.             return -1;

  696.         }

  697. 

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

  699.                                 : golomb_to_inter_cbp[vlc];

  700.     }

  701.     if (IS_INTRA16x16(mb_type) ||

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

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

  704. 

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

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

  707.             return -1;

  708.         }

  709.     }

  710.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  714.             av_log(h->avctx, AV_LOG_ERROR,

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

  716.             return -1;

  717.         }

  718.     }

  719. 

  720.     if (cbp) {

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

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

  723. 

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

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

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

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

  728.                                  2 * (j & 2) + 4 * (i & 2))

  729.                               : (4 * i + j);

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

  731. 

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

  733.                         av_log(h->avctx, AV_LOG_ERROR,

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

  735.                         return -1;

  736.                     }

  737.                 }

  738.             }

  739. 

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

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

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

  743.                     av_log(h->avctx, AV_LOG_ERROR,

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

  745.                     return -1;

  746.                 }

  747. 

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

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

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

  751.                         k                                 = 16 * i + j;

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

  753. 

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

  755.                             av_log(h->avctx, AV_LOG_ERROR,

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

  757.                             return -1;

  758.                         }

  759.                     }

  760.                 }

  761.             }

  762.         }

  763.     }

  764. 

  765.     sl->cbp                   = cbp;

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

  767. 

  768.     if (IS_INTRA(mb_type))

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

  770. 

  771.     return 0;

  772. }

  773. 

  774. static int svq3_decode_slice_header(AVCodecContext *avctx)

  775. {

  776.     SVQ3Context *s = avctx->priv_data;

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

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

  779.     const int mb_xy   = sl->mb_xy;

  780.     int i, header;

  781.     unsigned slice_id;

  782. 

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

  784. 

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

  786.         /* TODO: what? */

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

  788.         return -1;

  789.     } else {

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

  791. 

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

  793.                               8 * show_bits(&h->gb, 8 * length) +

  794.                               8 * length;

  795. 

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

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

  798.             return -1;

  799.         }

  800. 

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

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

  803. 

  804.         if (s->watermark_key) {

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

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

  807.                     header ^ s->watermark_key);

  808.         }

  809.         if (length > 0) {

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

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

  812.         }

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

  814.     }

  815. 

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

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

  818.         return -1;

  819.     }

  820. 

  821.     sl->slice_type = golomb_to_pict_type[slice_id];

  822. 

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

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

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

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

  827.     } else {

  828.         skip_bits1(&h->gb);

  829.         sl->mb_skip_run = 0;

  830.     }

  831. 

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

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

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

  835. 

  836.     /* unknown fields */

  837.     skip_bits1(&h->gb);

  838. 

  839.     if (s->unknown_flag)

  840.         skip_bits1(&h->gb);

  841. 

  842.     skip_bits1(&h->gb);

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

  844. 

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

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

  847. 

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

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

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

  851.                -1, 4 * sizeof(int8_t));

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

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

  854.     }

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

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

  857.                -1, 8 * sizeof(int8_t) * (h->mb_width - sl->mb_x));

  858. 

  859.         if (sl->mb_x > 0)

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

  861.     }

  862. 

  863.     return 0;

  864. }

  865. 

  866. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  867. {

  868.     SVQ3Context *s = avctx->priv_data;

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

  870.     H264SliceContext *sl;

  871.     int m;

  872.     unsigned char *extradata;

  873.     unsigned char *extradata_end;

  874.     unsigned int size;

  875.     int marker_found = 0;

  876. 

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

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

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

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

  881.         av_freep(&s->cur_pic);

  882.         av_freep(&s->last_pic);

  883.         av_freep(&s->next_pic);

  884.         return AVERROR(ENOMEM);

  885.     }

  886. 

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

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

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

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

  891.         return AVERROR(ENOMEM);

  892. 

  893.     if (ff_h264_decode_init(avctx) < 0)

  894.         return -1;

  895. 

  896.     // we will overwrite it later during decoding

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

  898. 

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

  900.     ff_h264chroma_init(&h->h264chroma, 8);

  901.     ff_h264qpel_init(&h->h264qpel, 8);

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

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

  904. 

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

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

  907. 

  908.     h->sps.bit_depth_luma = 8;

  909.     h->chroma_format_idc = 1;

  910. 

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

  912.     ff_tpeldsp_init(&s->tdsp);

  913. 

  914.     sl = h->slice_ctx;

  915. 

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

  917.     sl->is_complex     = 1;

  918.     h->picture_structure = PICT_FRAME;

  919.     avctx->pix_fmt     = AV_PIX_FMT_YUVJ420P;

  920.     avctx->color_range = AVCOL_RANGE_JPEG;

  921. 

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

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

  924. 

  925.     s->halfpel_flag  = 1;

  926.     s->thirdpel_flag = 1;

  927.     s->unknown_flag  = 0;

  928. 

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

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

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

  932.     if (extradata) {

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

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

  935.                 marker_found = 1;

  936.                 break;

  937.             }

  938.             extradata++;

  939.         }

  940.     }

  941. 

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

  943.     if (marker_found) {

  944.         GetBitContext gb;

  945.         int frame_size_code;

  946. 

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

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

  949.             return AVERROR_INVALIDDATA;

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

  951. 

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

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

  954.         switch (frame_size_code) {

  955.         case 0:

  956.             avctx->width  = 160;

  957.             avctx->height = 120;

  958.             break;

  959.         case 1:

  960.             avctx->width  = 128;

  961.             avctx->height =  96;

  962.             break;

  963.         case 2:

  964.             avctx->width  = 176;

  965.             avctx->height = 144;

  966.             break;

  967.         case 3:

  968.             avctx->width  = 352;

  969.             avctx->height = 288;

  970.             break;

  971.         case 4:

  972.             avctx->width  = 704;

  973.             avctx->height = 576;

  974.             break;

  975.         case 5:

  976.             avctx->width  = 240;

  977.             avctx->height = 180;

  978.             break;

  979.         case 6:

  980.             avctx->width  = 320;

  981.             avctx->height = 240;

  982.             break;

  983.         case 7:

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

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

  986.             break;

  987.         }

  988. 

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

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

  991. 

  992.         /* unknown fields */

  993.         skip_bits1(&gb);

  994.         skip_bits1(&gb);

  995.         skip_bits1(&gb);

  996.         skip_bits1(&gb);

  997. 

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

  999. 

  1000.         /* unknown field */

  1001.         skip_bits1(&gb);

  1002. 

  1003.         while (get_bits1(&gb))

  1004.             skip_bits(&gb, 8);

  1005. 

  1006.         s->unknown_flag  = get_bits1(&gb);

  1007.         avctx->has_b_frames = !h->low_delay;

  1008.         if (s->unknown_flag) {

  1009. #if CONFIG_ZLIB

  1010.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  1011.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  1012.             int u1                    = svq3_get_ue_golomb(&gb);

  1013.             int u2                    = get_bits(&gb, 8);

  1014.             int u3                    = get_bits(&gb, 2);

  1015.             int u4                    = svq3_get_ue_golomb(&gb);

  1016.             unsigned long buf_len     = watermark_width *

  1017.                                         watermark_height * 4;

  1018.             int offset                = get_bits_count(&gb) + 7 >> 3;

  1019.             uint8_t *buf;

  1020. 

  1021.             if (watermark_height > 0 &&

  1022.                 (uint64_t)watermark_width * 4 > UINT_MAX / watermark_height)

  1023.                 return -1;

  1024. 

  1025.             buf = av_malloc(buf_len);

  1026.             av_log(avctx, AV_LOG_DEBUG, "watermark size: %ux%u\n",

  1027.                    watermark_width, watermark_height);

  1028.             av_log(avctx, AV_LOG_DEBUG,

  1029.                    "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n",

  1030.                    u1, u2, u3, u4, offset);

  1031.             if (uncompress(buf, &buf_len, extradata + 8 + offset,

  1032.                            size - offset) != Z_OK) {

  1033.                 av_log(avctx, AV_LOG_ERROR,

  1034.                        "could not uncompress watermark logo\n");

  1035.                 av_free(buf);

  1036.                 return -1;

  1037.             }

  1038.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1039.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1040.             av_log(avctx, AV_LOG_DEBUG,

  1041.                    "watermark key %#"PRIx32"\n", s->watermark_key);

  1042.             av_free(buf);

  1043. #else

  1044.             av_log(avctx, AV_LOG_ERROR,

  1045.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1046.             return -1;

  1047. #endif

  1048.         }

  1049.     }

  1050. 

  1051.     h->width  = avctx->width;

  1052.     h->height = avctx->height;

  1053.     h->mb_width  = (h->width + 15) / 16;

  1054.     h->mb_height = (h->height + 15) / 16;

  1055.     h->mb_stride = h->mb_width + 1;

  1056.     h->mb_num    = h->mb_width * h->mb_height;

  1057.     h->b_stride = 4 * h->mb_width;

  1058.     s->h_edge_pos = h->mb_width * 16;

  1059.     s->v_edge_pos = h->mb_height * 16;

  1060. 

  1061.     if (ff_h264_alloc_tables(h) < 0) {

  1062.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1063.         return AVERROR(ENOMEM);

  1064.     }

  1065. 

  1066.     return 0;

  1067. }

  1068. 

  1069. static void free_picture(AVCodecContext *avctx, H264Picture *pic)

  1070. {

  1071.     int i;

  1072.     for (i = 0; i < 2; i++) {

  1073.         av_buffer_unref(&pic->motion_val_buf[i]);

  1074.         av_buffer_unref(&pic->ref_index_buf[i]);

  1075.     }

  1076.     av_buffer_unref(&pic->mb_type_buf);

  1077. 

  1078.     av_frame_unref(pic->f);

  1079. }

  1080. 

  1081. static int get_buffer(AVCodecContext *avctx, H264Picture *pic)

  1082. {

  1083.     SVQ3Context *s = avctx->priv_data;

  1084.     H264Context *h = &s->h;

  1085.     H264SliceContext *sl = &h->slice_ctx[0];

  1086.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1087.     const int mb_array_size = h->mb_stride * h->mb_height;

  1088.     const int b4_stride     = h->mb_width * 4 + 1;

  1089.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1090.     int ret;

  1091. 

  1092.     if (!pic->motion_val_buf[0]) {

  1093.         int i;

  1094. 

  1095.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1096.         if (!pic->mb_type_buf)

  1097.             return AVERROR(ENOMEM);

  1098.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1099. 

  1100.         for (i = 0; i < 2; i++) {

  1101.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1102.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1103.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1104.                 ret = AVERROR(ENOMEM);

  1105.                 goto fail;

  1106.             }

  1107. 

  1108.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1109.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1110.         }

  1111.     }

  1112.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1113. 

  1114.     ret = ff_get_buffer(avctx, pic->f,

  1115.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1116.     if (ret < 0)

  1117.         goto fail;

  1118. 

  1119.     if (!sl->edge_emu_buffer) {

  1120.         sl->edge_emu_buffer = av_mallocz(pic->f->linesize[0] * 17);

  1121.         if (!sl->edge_emu_buffer)

  1122.             return AVERROR(ENOMEM);

  1123.     }

  1124. 

  1125.     sl->linesize   = pic->f->linesize[0];

  1126.     sl->uvlinesize = pic->f->linesize[1];

  1127. 

  1128.     return 0;

  1129. fail:

  1130.     free_picture(avctx, pic);

  1131.     return ret;

  1132. }

  1133. 

  1134. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1135.                              int *got_frame, AVPacket *avpkt)

  1136. {

  1137.     const uint8_t *buf = avpkt->data;

  1138.     SVQ3Context *s     = avctx->priv_data;

  1139.     H264Context *h     = &s->h;

  1140.     H264SliceContext *sl = &h->slice_ctx[0];

  1141.     int buf_size       = avpkt->size;

  1142.     int ret, m, i;

  1143. 

  1144.     /* special case for last picture */

  1145.     if (buf_size == 0) {

  1146.         if (s->next_pic->f->data[0] && !h->low_delay && !s->last_frame_output) {

  1147.             ret = av_frame_ref(data, s->next_pic->f);

  1148.             if (ret < 0)

  1149.                 return ret;

  1150.             s->last_frame_output = 1;

  1151.             *got_frame          = 1;

  1152.         }

  1153.         return 0;

  1154.     }

  1155. 

  1156.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1157. 

  1158.     sl->mb_x = sl->mb_y = sl->mb_xy = 0;

  1159. 

  1160.     if (svq3_decode_slice_header(avctx))

  1161.         return -1;

  1162. 

  1163.     h->pict_type = sl->slice_type;

  1164. 

  1165.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1166.         FFSWAP(H264Picture*, s->next_pic, s->last_pic);

  1167. 

  1168.     av_frame_unref(s->cur_pic->f);

  1169. 

  1170.     /* for skipping the frame */

  1171.     s->cur_pic->f->pict_type = h->pict_type;

  1172.     s->cur_pic->f->key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1173. 

  1174.     ret = get_buffer(avctx, s->cur_pic);

  1175.     if (ret < 0)

  1176.         return ret;

  1177. 

  1178.     h->cur_pic_ptr = s->cur_pic;

  1179.     h->cur_pic     = *s->cur_pic;

  1180. 

  1181.     for (i = 0; i < 16; i++) {

  1182.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);

  1183.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);

  1184.     }

  1185.     for (i = 0; i < 16; i++) {

  1186.         h->block_offset[16 + i]      =

  1187.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1188.         h->block_offset[48 + 16 + i] =

  1189.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1190.     }

  1191. 

  1192.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1193.         if (!s->last_pic->f->data[0]) {

  1194.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1195.             ret = get_buffer(avctx, s->last_pic);

  1196.             if (ret < 0)

  1197.                 return ret;

  1198.             memset(s->last_pic->f->data[0], 0, avctx->height * s->last_pic->f->linesize[0]);

  1199.             memset(s->last_pic->f->data[1], 0x80, (avctx->height / 2) *

  1200.                    s->last_pic->f->linesize[1]);

  1201.             memset(s->last_pic->f->data[2], 0x80, (avctx->height / 2) *

  1202.                    s->last_pic->f->linesize[2]);

  1203.         }

  1204. 

  1205.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f->data[0]) {

  1206.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1207.             ret = get_buffer(avctx, s->next_pic);

  1208.             if (ret < 0)

  1209.                 return ret;

  1210.             memset(s->next_pic->f->data[0], 0, avctx->height * s->next_pic->f->linesize[0]);

  1211.             memset(s->next_pic->f->data[1], 0x80, (avctx->height / 2) *

  1212.                    s->next_pic->f->linesize[1]);

  1213.             memset(s->next_pic->f->data[2], 0x80, (avctx->height / 2) *

  1214.                    s->next_pic->f->linesize[2]);

  1215.         }

  1216.     }

  1217. 

  1218.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1219.         av_log(h->avctx, AV_LOG_DEBUG,

  1220.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1221.                av_get_picture_type_char(h->pict_type),

  1222.                s->halfpel_flag, s->thirdpel_flag,

  1223.                s->adaptive_quant, h->slice_ctx[0].qscale, sl->slice_num);

  1224. 

  1225.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1226.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1227.         avctx->skip_frame >= AVDISCARD_ALL)

  1228.         return 0;

  1229. 

  1230.     if (s->next_p_frame_damaged) {

  1231.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1232.             return 0;

  1233.         else

  1234.             s->next_p_frame_damaged = 0;

  1235.     }

  1236. 

  1237.     if (h->pict_type == AV_PICTURE_TYPE_B) {

  1238.         h->frame_num_offset = sl->slice_num - h->prev_frame_num;

  1239. 

  1240.         if (h->frame_num_offset < 0)

  1241.             h->frame_num_offset += 256;

  1242.         if (h->frame_num_offset == 0 ||

  1243.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1244.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1245.             return -1;

  1246.         }

  1247.     } else {

  1248.         h->prev_frame_num        = h->frame_num;

  1249.         h->frame_num             = sl->slice_num;

  1250.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1251. 

  1252.         if (h->prev_frame_num_offset < 0)

  1253.             h->prev_frame_num_offset += 256;

  1254.     }

  1255. 

  1256.     for (m = 0; m < 2; m++) {

  1257.         int i;

  1258.         for (i = 0; i < 4; i++) {

  1259.             int j;

  1260.             for (j = -1; j < 4; j++)

  1261.                 sl->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1262.             if (i < 3)

  1263.                 sl->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1264.         }

  1265.     }

  1266. 

  1267.     for (sl->mb_y = 0; sl->mb_y < h->mb_height; sl->mb_y++) {

  1268.         for (sl->mb_x = 0; sl->mb_x < h->mb_width; sl->mb_x++) {

  1269.             unsigned mb_type;

  1270.             sl->mb_xy = sl->mb_x + sl->mb_y * h->mb_stride;

  1271. 

  1272.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1273.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1274.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1275.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1276.                 h->gb.size_in_bits = 8 * buf_size;

  1277. 

  1278.                 if (svq3_decode_slice_header(avctx))

  1279.                     return -1;

  1280. 

  1281.                 /* TODO: support s->mb_skip_run */

  1282.             }

  1283. 

  1284.             mb_type = svq3_get_ue_golomb(&h->gb);

  1285. 

  1286.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1287.                 mb_type += 8;

  1288.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1289.                 mb_type += 4;

  1290.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1291.                 av_log(h->avctx, AV_LOG_ERROR,

  1292.                        "error while decoding MB %d %d\n", sl->mb_x, sl->mb_y);

  1293.                 return -1;

  1294.             }

  1295. 

  1296.             if (mb_type != 0)

  1297.                 ff_h264_hl_decode_mb(h, &h->slice_ctx[0]);

  1298. 

  1299.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1300.                 h->cur_pic.mb_type[sl->mb_x + sl->mb_y * h->mb_stride] =

  1301.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1302.         }

  1303. 

  1304.         ff_draw_horiz_band(avctx, s->cur_pic->f,

  1305.                            s->last_pic->f->data[0] ? s->last_pic->f : NULL,

  1306.                            16 * sl->mb_y, 16, h->picture_structure, 0,

  1307.                            h->low_delay);

  1308.     }

  1309. 

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

  1311.         ret = av_frame_ref(data, s->cur_pic->f);

  1312.     else if (s->last_pic->f->data[0])

  1313.         ret = av_frame_ref(data, s->last_pic->f);

  1314.     if (ret < 0)

  1315.         return ret;

  1316. 

  1317.     /* Do not output the last pic after seeking. */

  1318.     if (s->last_pic->f->data[0] || h->low_delay)

  1319.         *got_frame = 1;

  1320. 

  1321.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1322.         FFSWAP(H264Picture*, s->cur_pic, s->next_pic);

  1323.     } else {

  1324.         av_frame_unref(s->cur_pic->f);

  1325.     }

  1326. 

  1327.     return buf_size;

  1328. }

  1329. 

  1330. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1331. {

  1332.     SVQ3Context *s = avctx->priv_data;

  1333.     H264Context *h = &s->h;

  1334. 

  1335.     free_picture(avctx, s->cur_pic);

  1336.     free_picture(avctx, s->next_pic);

  1337.     free_picture(avctx, s->last_pic);

  1338.     av_frame_free(&s->cur_pic->f);

  1339.     av_frame_free(&s->next_pic->f);

  1340.     av_frame_free(&s->last_pic->f);

  1341.     av_freep(&s->cur_pic);

  1342.     av_freep(&s->next_pic);

  1343.     av_freep(&s->last_pic);

  1344. 

  1345.     memset(&h->cur_pic, 0, sizeof(h->cur_pic));

  1346. 

  1347.     ff_h264_free_context(h);

  1348. 

  1349.     return 0;

  1350. }

  1351. 

  1352. AVCodec ff_svq3_decoder = {

  1353.     .name           = "svq3",

  1354.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1355.     .type           = AVMEDIA_TYPE_VIDEO,

  1356.     .id             = AV_CODEC_ID_SVQ3,

  1357.     .priv_data_size = sizeof(SVQ3Context),

  1358.     .init           = svq3_decode_init,

  1359.     .close          = svq3_decode_end,

  1360.     .decode         = svq3_decode_frame,

  1361.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1362.                       CODEC_CAP_DR1             |

  1363.                       CODEC_CAP_DELAY,

  1364.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1365.                                                      AV_PIX_FMT_NONE},

  1366. };