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.
38308 Commits
32 Branches
283MB
Tree: ce1e8045e0
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'ce1e8045e0'
${ noResults }
FFmpeg/libavcodec/svq3.c

1327 lines
47KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The Libav Project

  3.  *

  4.  * This file is part of Libav.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

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

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

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

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

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

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

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

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

  31.  *

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

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

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

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

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

  37.  *

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

  39.  * correctly decodes this file:

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

  41.  */

  42. 

  43. #include "libavutil/attributes.h"

  44. #include "internal.h"

  45. #include "avcodec.h"

  46. #include "mpegvideo.h"

  47. #include "h264.h"

  48. 

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

  50. 

  51. #include "h264_mvpred.h"

  52. #include "golomb.h"

  53. #include "hpeldsp.h"

  54. #include "rectangle.h"

  55. 

  56. #if CONFIG_ZLIB

  57. #include <zlib.h>

  58. #endif

  59. 

  60. #include "svq1.h"

  61. #include "svq3.h"

  62. 

  63. /**

  64.  * @file

  65.  * svq3 decoder.

  66.  */

  67. 

  68. typedef struct {

  69.     H264Context h;

  70.     HpelDSPContext hdsp;

  71.     Picture *cur_pic;

  72.     Picture *next_pic;

  73.     Picture *last_pic;

  74.     int halfpel_flag;

  75.     int thirdpel_flag;

  76.     int unknown_flag;

  77.     int next_slice_index;

  78.     uint32_t watermark_key;

  79.     int adaptive_quant;

  80.     int next_p_frame_damaged;

  81.     int h_edge_pos;

  82.     int v_edge_pos;

  83.     int last_frame_output;

  84. } SVQ3Context;

  85. 

  86. #define FULLPEL_MODE  1

  87. #define HALFPEL_MODE  2

  88. #define THIRDPEL_MODE 3

  89. #define PREDICT_MODE  4

  90. 

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

  92.  * o-->o-->o   o

  93.  *         |  /|

  94.  * o   o   o / o

  95.  * | / |   |/  |

  96.  * o   o   o   o

  97.  *   /

  98.  * o-->o-->o-->o

  99.  */

  100. static const uint8_t svq3_scan[16] = {

  101.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  102.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

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

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

  105. };

  106. 

  107. static const uint8_t luma_dc_zigzag_scan[16] = {

  108.     0 * 16 + 0 * 64, 1 * 16 + 0 * 64, 2 * 16 + 0 * 64, 0 * 16 + 2 * 64,

  109.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

  110.     1 * 16 + 2 * 64, 2 * 16 + 2 * 64, 3 * 16 + 2 * 64, 0 * 16 + 3 * 64,

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

  112. };

  113. 

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

  115.     { 0, 0 },

  116.     { 1, 0 }, { 0, 1 },

  117.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  118.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  119.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  120.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

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

  122.     { 4, 3 }, { 3, 4 },

  123.     { 4, 4 }

  124. };

  125. 

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

  127.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  128.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

  129.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  4,  3 }, { 0, 1,  2,  4,  3 },

  130.       { 0,  2,  1,  4,  3 }, { 2, 0,  1,  3,  4 }, { 0, 4,  2,  1,  3 } },

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

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

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

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

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

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

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

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

  139. };

  140. 

  141. static const struct {

  142.     uint8_t run;

  143.     uint8_t level;

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

  145.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  146.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

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

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

  149. };

  150. 

  151. static const uint32_t svq3_dequant_coeff[32] = {

  152.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  153.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  154.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  155.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  156. };

  157. 

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

  159. {

  160.     const int qmul = svq3_dequant_coeff[qp];

  161. #define stride 16

  162.     int i;

  163.     int temp[16];

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

  165. 

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

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

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

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

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

  171. 

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

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

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

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

  176.     }

  177. 

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

  179.         const int offset = x_offset[i];

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

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

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

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

  184. 

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

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

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

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

  189.     }

  190. }

  191. #undef stride

  192. 

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

  194.                         int stride, int qp, int dc)

  195. {

  196.     const int qmul = svq3_dequant_coeff[qp];

  197.     int i;

  198. 

  199.     if (dc) {

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

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

  202.         block[0] = 0;

  203.     }

  204. 

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

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

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

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

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

  210. 

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

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

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

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

  215.     }

  216. 

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

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

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

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

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

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

  223. 

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

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

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

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

  228.     }

  229. 

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

  231. }

  232. 

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

  234.                                     int index, const int type)

  235. {

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

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

  238. 

  239.     int run, level, limit;

  240.     unsigned vlc;

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

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

  243. 

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

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

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

  247.             vlc      = vlc + 1 >> 1;

  248. 

  249.             if (type == 3) {

  250.                 if (vlc < 3) {

  251.                     run   = 0;

  252.                     level = vlc;

  253.                 } else if (vlc < 4) {

  254.                     run   = 1;

  255.                     level = 1;

  256.                 } else {

  257.                     run   = vlc & 0x3;

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

  259.                 }

  260.             } else {

  261.                 if (vlc < 16) {

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

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

  264.                 } else if (intra) {

  265.                     run   = vlc & 0x7;

  266.                     level = (vlc >> 3) +

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

  268.                 } else {

  269.                     run   = vlc & 0xF;

  270.                     level = (vlc >> 4) +

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

  272.                 }

  273.             }

  274. 

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

  276.                 return -1;

  277. 

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

  279.         }

  280. 

  281.         if (type != 2) {

  282.             break;

  283.         }

  284.     }

  285. 

  286.     return 0;

  287. }

  288. 

  289. static inline void svq3_mc_dir_part(SVQ3Context *s,

  290.                                     int x, int y, int width, int height,

  291.                                     int mx, int my, int dxy,

  292.                                     int thirdpel, int dir, int avg)

  293. {

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

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

  296.     uint8_t *src, *dest;

  297.     int i, emu = 0;

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

  299. 

  300.     mx += x;

  301.     my += y;

  302. 

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

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

  305.         emu = 1;

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

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

  308.     }

  309. 

  310.     /* form component predictions */

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

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

  313. 

  314.     if (emu) {

  315.         h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->linesize,

  316.                                  width + 1, height + 1,

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

  318.         src = h->edge_emu_buffer;

  319.     }

  320.     if (thirdpel)

  321.         (avg ? h->dsp.avg_tpel_pixels_tab

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

  323.                                                 width, height);

  324.     else

  325.         (avg ? s->hdsp.avg_pixels_tab

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

  327.                                                        height);

  328. 

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

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

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

  332.         width  = width  >> 1;

  333.         height = height >> 1;

  334.         blocksize++;

  335. 

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

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

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

  339. 

  340.             if (emu) {

  341.                 h->vdsp.emulated_edge_mc(h->edge_emu_buffer, src, h->uvlinesize,

  342.                                          width + 1, height + 1,

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

  344.                                          s->v_edge_pos >> 1);

  345.                 src = h->edge_emu_buffer;

  346.             }

  347.             if (thirdpel)

  348.                 (avg ? h->dsp.avg_tpel_pixels_tab

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

  350.                                                         h->uvlinesize,

  351.                                                         width, height);

  352.             else

  353.                 (avg ? s->hdsp.avg_pixels_tab

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

  355.                                                                h->uvlinesize,

  356.                                                                height);

  357.         }

  358.     }

  359. }

  360. 

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

  362.                               int dir, int avg)

  363. {

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

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

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

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

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

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

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

  371. 

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

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

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

  375.                              (4 * h->mb_y + (i >> 2)) * h->b_stride;

  376.             int dxy;

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

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

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

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

  381. 

  382.             if (mode != PREDICT_MODE) {

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

  384.             } else {

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

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

  387. 

  388.                 if (dir == 0) {

  389.                     mx = mx * h->frame_num_offset /

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

  391.                     my = my * h->frame_num_offset /

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

  393.                 } else {

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

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

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

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

  398.                 }

  399.             }

  400. 

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

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

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

  404. 

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

  406.             if (mode == PREDICT_MODE) {

  407.                 dx = dy = 0;

  408.             } else {

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

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

  411. 

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

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

  414.                     return -1;

  415.                 }

  416.             }

  417. 

  418.             /* compute motion vector */

  419.             if (mode == THIRDPEL_MODE) {

  420.                 int fx, fy;

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

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

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

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

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

  426. 

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

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

  429.                 mx += mx;

  430.                 my += my;

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

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

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

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

  435. 

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

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

  438.                 mx *= 3;

  439.                 my *= 3;

  440.             } else {

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

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

  443. 

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

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

  446.                 mx *= 6;

  447.                 my *= 6;

  448.             }

  449. 

  450.             /* update mv_cache */

  451.             if (mode != PREDICT_MODE) {

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

  453. 

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

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

  456. 

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

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

  459.                 }

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

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

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

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

  464.             }

  465. 

  466.             /* write back motion vectors */

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

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

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

  470.         }

  471. 

  472.     return 0;

  473. }

  474. 

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

  476. {

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

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

  479.     int cbp = 0;

  480.     uint32_t vlc;

  481.     int8_t *top, *left;

  482.     const int mb_xy         = h->mb_xy;

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

  484. 

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

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

  487.     h->topright_samples_available = 0xFFFF;

  488. 

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

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

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

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

  493.                              0, 0, 0, 0, 0, 0);

  494. 

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

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

  497.                                  0, 0, 0, 0, 1, 1);

  498. 

  499.             mb_type = MB_TYPE_SKIP;

  500.         } else {

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

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

  503.                 return -1;

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

  505.                 return -1;

  506. 

  507.             mb_type = MB_TYPE_16x16;

  508.         }

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

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

  511.             mode = THIRDPEL_MODE;

  512.         else if (s->halfpel_flag &&

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

  514.             mode = HALFPEL_MODE;

  515.         else

  516.             mode = FULLPEL_MODE;

  517. 

  518.         /* fill caches */

  519.         /* note ref_cache should contain here:

  520.          *  ????????

  521.          *  ???11111

  522.          *  N??11111

  523.          *  N??11111

  524.          *  N??11111

  525.          */

  526. 

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

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

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

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

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

  532.             } else {

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

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

  535.             }

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

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

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

  539.                        4 * 2 * sizeof(int16_t));

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

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

  542. 

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

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

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

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

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

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

  549.                 } else

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

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

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

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

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

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

  556.                 } else

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

  558.             } else

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

  560.                        PART_NOT_AVAILABLE, 8);

  561. 

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

  563.                 break;

  564.         }

  565. 

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

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

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

  569.                 return -1;

  570.         } else {        /* AV_PICTURE_TYPE_B */

  571.             if (mb_type != 2) {

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

  573.                     return -1;

  574.             } else {

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

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

  577.                            0, 4 * 2 * sizeof(int16_t));

  578.             }

  579.             if (mb_type != 1) {

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

  581.                     return -1;

  582.             } else {

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

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

  585.                            0, 4 * 2 * sizeof(int16_t));

  586.             }

  587.         }

  588. 

  589.         mb_type = MB_TYPE_16x16;

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

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

  592. 

  593.         if (mb_type == 8) {

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

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

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

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

  598.                     h->left_samples_available = 0x5F5F;

  599.             }

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

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

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

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

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

  605. 

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

  607.                     h->top_samples_available = 0x33FF;

  608.             }

  609. 

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

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

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

  613. 

  614.                 if (vlc >= 25) {

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

  616.                     return -1;

  617.                 }

  618. 

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

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

  621. 

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

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

  624. 

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

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

  627.                     return -1;

  628.                 }

  629.             }

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

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

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

  633.         }

  634. 

  635.         write_back_intra_pred_mode(h);

  636. 

  637.         if (mb_type == 8) {

  638.             ff_h264_check_intra4x4_pred_mode(h);

  639. 

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

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

  642.         } else {

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

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

  645. 

  646.             h->top_samples_available  = 0x33FF;

  647.             h->left_samples_available = 0x5F5F;

  648.         }

  649. 

  650.         mb_type = MB_TYPE_INTRA4x4;

  651.     } else {                      /* INTRA16x16 */

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

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

  654. 

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

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

  657.             return h->intra16x16_pred_mode;

  658.         }

  659. 

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

  661.         mb_type = MB_TYPE_INTRA16x16;

  662.     }

  663. 

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

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

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

  667.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  671.                        0, 4 * 2 * sizeof(int16_t));

  672.         }

  673.     }

  674.     if (!IS_INTRA4x4(mb_type)) {

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

  676.     }

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

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

  679.     }

  680. 

  681.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  685.             return -1;

  686.         }

  687. 

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

  689.                                 : golomb_to_inter_cbp[vlc];

  690.     }

  691.     if (IS_INTRA16x16(mb_type) ||

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

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

  694. 

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

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

  697.             return -1;

  698.         }

  699.     }

  700.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  704.             av_log(h->avctx, AV_LOG_ERROR,

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

  706.             return -1;

  707.         }

  708.     }

  709. 

  710.     if (cbp) {

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

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

  713. 

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

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

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

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

  718.                                  2 * (j & 2) + 4 * (i & 2))

  719.                               : (4 * i + j);

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

  721. 

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

  723.                         av_log(h->avctx, AV_LOG_ERROR,

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

  725.                         return -1;

  726.                     }

  727.                 }

  728.             }

  729. 

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

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

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

  733.                     av_log(h->avctx, AV_LOG_ERROR,

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

  735.                     return -1;

  736.                 }

  737. 

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

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

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

  741.                         k                                 = 16 * i + j;

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

  743. 

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

  745.                             av_log(h->avctx, AV_LOG_ERROR,

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

  747.                             return -1;

  748.                         }

  749.                     }

  750.                 }

  751.             }

  752.         }

  753.     }

  754. 

  755.     h->cbp                              = cbp;

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

  757. 

  758.     if (IS_INTRA(mb_type))

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

  760. 

  761.     return 0;

  762. }

  763. 

  764. static int svq3_decode_slice_header(AVCodecContext *avctx)

  765. {

  766.     SVQ3Context *s = avctx->priv_data;

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

  768.     const int mb_xy   = h->mb_xy;

  769.     int i, header;

  770.     unsigned slice_id;

  771. 

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

  773. 

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

  775.         /* TODO: what? */

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

  777.         return -1;

  778.     } else {

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

  780. 

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

  782.                               8 * show_bits(&h->gb, 8 * length) +

  783.                               8 * length;

  784. 

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

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

  787.             return -1;

  788.         }

  789. 

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

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

  792. 

  793.         if (s->watermark_key) {

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

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

  796.                     header ^ s->watermark_key);

  797.         }

  798.         if (length > 0) {

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

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

  801.         }

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

  803.     }

  804. 

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

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

  807.         return -1;

  808.     }

  809. 

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

  811. 

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

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

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

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

  816.     } else {

  817.         skip_bits1(&h->gb);

  818.         h->mb_skip_run = 0;

  819.     }

  820. 

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

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

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

  824. 

  825.     /* unknown fields */

  826.     skip_bits1(&h->gb);

  827. 

  828.     if (s->unknown_flag)

  829.         skip_bits1(&h->gb);

  830. 

  831.     skip_bits1(&h->gb);

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

  833. 

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

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

  836. 

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

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

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

  840.                -1, 4 * sizeof(int8_t));

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

  842.                -1, 8 * sizeof(int8_t) * h->mb_x);

  843.     }

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

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

  846.                -1, 8 * sizeof(int8_t) * (h->mb_width - h->mb_x));

  847. 

  848.         if (h->mb_x > 0)

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

  850.     }

  851. 

  852.     return 0;

  853. }

  854. 

  855. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  856. {

  857.     SVQ3Context *s = avctx->priv_data;

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

  859.     int m;

  860.     unsigned char *extradata;

  861.     unsigned char *extradata_end;

  862.     unsigned int size;

  863.     int marker_found = 0;

  864. 

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

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

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

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

  869.         av_freep(&s->cur_pic);

  870.         av_freep(&s->last_pic);

  871.         av_freep(&s->next_pic);

  872.         return AVERROR(ENOMEM);

  873.     }

  874. 

  875.     if (ff_h264_decode_init(avctx) < 0)

  876.         return -1;

  877. 

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

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

  880.     h->is_complex      = 1;

  881.     h->picture_structure = PICT_FRAME;

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

  883. 

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

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

  886. 

  887.     s->halfpel_flag  = 1;

  888.     s->thirdpel_flag = 1;

  889.     s->unknown_flag  = 0;

  890. 

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

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

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

  894.     if (extradata) {

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

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

  897.                 marker_found = 1;

  898.                 break;

  899.             }

  900.             extradata++;

  901.         }

  902.     }

  903. 

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

  905.     if (marker_found) {

  906.         GetBitContext gb;

  907.         int frame_size_code;

  908. 

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

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

  911.             return AVERROR_INVALIDDATA;

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

  913. 

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

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

  916.         switch (frame_size_code) {

  917.         case 0:

  918.             avctx->width  = 160;

  919.             avctx->height = 120;

  920.             break;

  921.         case 1:

  922.             avctx->width  = 128;

  923.             avctx->height =  96;

  924.             break;

  925.         case 2:

  926.             avctx->width  = 176;

  927.             avctx->height = 144;

  928.             break;

  929.         case 3:

  930.             avctx->width  = 352;

  931.             avctx->height = 288;

  932.             break;

  933.         case 4:

  934.             avctx->width  = 704;

  935.             avctx->height = 576;

  936.             break;

  937.         case 5:

  938.             avctx->width  = 240;

  939.             avctx->height = 180;

  940.             break;

  941.         case 6:

  942.             avctx->width  = 320;

  943.             avctx->height = 240;

  944.             break;

  945.         case 7:

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

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

  948.             break;

  949.         }

  950. 

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

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

  953. 

  954.         /* unknown fields */

  955.         skip_bits1(&gb);

  956.         skip_bits1(&gb);

  957.         skip_bits1(&gb);

  958.         skip_bits1(&gb);

  959. 

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

  961. 

  962.         /* unknown field */

  963.         skip_bits1(&gb);

  964. 

  965.         while (get_bits1(&gb))

  966.             skip_bits(&gb, 8);

  967. 

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

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

  970.         if (s->unknown_flag) {

  971. #if CONFIG_ZLIB

  972.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  973.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  974.             int u1                    = svq3_get_ue_golomb(&gb);

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

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

  977.             int u4                    = svq3_get_ue_golomb(&gb);

  978.             unsigned long buf_len     = watermark_width *

  979.                                         watermark_height * 4;

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

  981.             uint8_t *buf;

  982. 

  983.             if (watermark_height > 0 &&

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

  985.                 return -1;

  986. 

  987.             buf = av_malloc(buf_len);

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

  989.                    watermark_width, watermark_height);

  990.             av_log(avctx, AV_LOG_DEBUG,

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

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

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

  994.                            size - offset) != Z_OK) {

  995.                 av_log(avctx, AV_LOG_ERROR,

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

  997.                 av_free(buf);

  998.                 return -1;

  999.             }

  1000.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1001.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1002.             av_log(avctx, AV_LOG_DEBUG,

  1003.                    "watermark key %#x\n", s->watermark_key);

  1004.             av_free(buf);

  1005. #else

  1006.             av_log(avctx, AV_LOG_ERROR,

  1007.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1008.             return -1;

  1009. #endif

  1010.         }

  1011.     }

  1012. 

  1013.     h->width  = avctx->width;

  1014.     h->height = avctx->height;

  1015.     h->mb_width  = (h->width + 15) / 16;

  1016.     h->mb_height = (h->height + 15) / 16;

  1017.     h->mb_stride = h->mb_width + 1;

  1018.     h->mb_num    = h->mb_width * h->mb_height;

  1019.     h->b_stride = 4 * h->mb_width;

  1020.     s->h_edge_pos = h->mb_width * 16;

  1021.     s->v_edge_pos = h->mb_height * 16;

  1022. 

  1023.     if (ff_h264_alloc_tables(h) < 0) {

  1024.         av_log(avctx, AV_LOG_ERROR, "svq3 memory allocation failed\n");

  1025.         return AVERROR(ENOMEM);

  1026.     }

  1027. 

  1028.     return 0;

  1029. }

  1030. 

  1031. static void free_picture(AVCodecContext *avctx, Picture *pic)

  1032. {

  1033.     int i;

  1034.     for (i = 0; i < 2; i++) {

  1035.         av_buffer_unref(&pic->motion_val_buf[i]);

  1036.         av_buffer_unref(&pic->ref_index_buf[i]);

  1037.     }

  1038.     av_buffer_unref(&pic->mb_type_buf);

  1039. 

  1040.     av_frame_unref(&pic->f);

  1041. }

  1042. 

  1043. static int get_buffer(AVCodecContext *avctx, Picture *pic)

  1044. {

  1045.     SVQ3Context *s = avctx->priv_data;

  1046.     H264Context *h = &s->h;

  1047.     const int big_mb_num    = h->mb_stride * (h->mb_height + 1) + 1;

  1048.     const int mb_array_size = h->mb_stride * h->mb_height;

  1049.     const int b4_stride     = h->mb_width * 4 + 1;

  1050.     const int b4_array_size = b4_stride * h->mb_height * 4;

  1051.     int ret;

  1052. 

  1053.     if (!pic->motion_val_buf[0]) {

  1054.         int i;

  1055. 

  1056.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + h->mb_stride) * sizeof(uint32_t));

  1057.         if (!pic->mb_type_buf)

  1058.             return AVERROR(ENOMEM);

  1059.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * h->mb_stride + 1;

  1060. 

  1061.         for (i = 0; i < 2; i++) {

  1062.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1063.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1064.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1065.                 ret = AVERROR(ENOMEM);

  1066.                 goto fail;

  1067.             }

  1068. 

  1069.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1070.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1071.         }

  1072.     }

  1073.     pic->reference = !(h->pict_type == AV_PICTURE_TYPE_B);

  1074. 

  1075.     ret = ff_get_buffer(avctx, &pic->f,

  1076.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1077.     if (ret < 0)

  1078.         goto fail;

  1079. 

  1080.     if (!h->edge_emu_buffer) {

  1081.         h->edge_emu_buffer = av_mallocz(pic->f.linesize[0] * 17);

  1082.         if (!h->edge_emu_buffer)

  1083.             return AVERROR(ENOMEM);

  1084.     }

  1085. 

  1086.     h->linesize   = pic->f.linesize[0];

  1087.     h->uvlinesize = pic->f.linesize[1];

  1088. 

  1089.     return 0;

  1090. fail:

  1091.     free_picture(avctx, pic);

  1092.     return ret;

  1093. }

  1094. 

  1095. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1096.                              int *got_frame, AVPacket *avpkt)

  1097. {

  1098.     const uint8_t *buf = avpkt->data;

  1099.     SVQ3Context *s     = avctx->priv_data;

  1100.     H264Context *h     = &s->h;

  1101.     int buf_size       = avpkt->size;

  1102.     int ret, m, i;

  1103. 

  1104.     /* special case for last picture */

  1105.     if (buf_size == 0) {

  1106.         if (s->next_pic->f.data[0] && !h->low_delay && !s->last_frame_output) {

  1107.             ret = av_frame_ref(data, &s->next_pic->f);

  1108.             if (ret < 0)

  1109.                 return ret;

  1110.             s->last_frame_output = 1;

  1111.             *got_frame          = 1;

  1112.         }

  1113.         return 0;

  1114.     }

  1115. 

  1116.     init_get_bits(&h->gb, buf, 8 * buf_size);

  1117. 

  1118.     h->mb_x = h->mb_y = h->mb_xy = 0;

  1119. 

  1120.     if (svq3_decode_slice_header(avctx))

  1121.         return -1;

  1122. 

  1123.     h->pict_type = h->slice_type;

  1124. 

  1125.     if (h->pict_type != AV_PICTURE_TYPE_B)

  1126.         FFSWAP(Picture*, s->next_pic, s->last_pic);

  1127. 

  1128.     av_frame_unref(&s->cur_pic->f);

  1129. 

  1130.     /* for skipping the frame */

  1131.     s->cur_pic->f.pict_type = h->pict_type;

  1132.     s->cur_pic->f.key_frame = (h->pict_type == AV_PICTURE_TYPE_I);

  1133. 

  1134.     ret = get_buffer(avctx, s->cur_pic);

  1135.     if (ret < 0)

  1136.         return ret;

  1137. 

  1138.     h->cur_pic_ptr = s->cur_pic;

  1139.     av_frame_unref(&h->cur_pic.f);

  1140.     h->cur_pic     = *s->cur_pic;

  1141.     ret = av_frame_ref(&h->cur_pic.f, &s->cur_pic->f);

  1142.     if (ret < 0)

  1143.         return ret;

  1144. 

  1145.     for (i = 0; i < 16; i++) {

  1146.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1147.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->linesize * ((scan8[i] - scan8[0]) >> 3);

  1148.     }

  1149.     for (i = 0; i < 16; i++) {

  1150.         h->block_offset[16 + i]      =

  1151.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1152.         h->block_offset[48 + 16 + i] =

  1153.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * h->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1154.     }

  1155. 

  1156.     if (h->pict_type != AV_PICTURE_TYPE_I) {

  1157.         if (!s->last_pic->f.data[0]) {

  1158.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1159.             ret = get_buffer(avctx, s->last_pic);

  1160.             if (ret < 0)

  1161.                 return ret;

  1162.             memset(s->last_pic->f.data[0], 0, avctx->height * s->last_pic->f.linesize[0]);

  1163.             memset(s->last_pic->f.data[1], 0x80, (avctx->height / 2) *

  1164.                    s->last_pic->f.linesize[1]);

  1165.             memset(s->last_pic->f.data[2], 0x80, (avctx->height / 2) *

  1166.                    s->last_pic->f.linesize[2]);

  1167.         }

  1168. 

  1169.         if (h->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f.data[0]) {

  1170.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1171.             ret = get_buffer(avctx, s->next_pic);

  1172.             if (ret < 0)

  1173.                 return ret;

  1174.             memset(s->next_pic->f.data[0], 0, avctx->height * s->next_pic->f.linesize[0]);

  1175.             memset(s->next_pic->f.data[1], 0x80, (avctx->height / 2) *

  1176.                    s->next_pic->f.linesize[1]);

  1177.             memset(s->next_pic->f.data[2], 0x80, (avctx->height / 2) *

  1178.                    s->next_pic->f.linesize[2]);

  1179.         }

  1180.     }

  1181. 

  1182.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1183.         av_log(h->avctx, AV_LOG_DEBUG,

  1184.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1185.                av_get_picture_type_char(h->pict_type),

  1186.                s->halfpel_flag, s->thirdpel_flag,

  1187.                s->adaptive_quant, h->qscale, h->slice_num);

  1188. 

  1189.     if (avctx->skip_frame >= AVDISCARD_NONREF && h->pict_type == AV_PICTURE_TYPE_B ||

  1190.         avctx->skip_frame >= AVDISCARD_NONKEY && h->pict_type != AV_PICTURE_TYPE_I ||

  1191.         avctx->skip_frame >= AVDISCARD_ALL)

  1192.         return 0;

  1193. 

  1194.     if (s->next_p_frame_damaged) {

  1195.         if (h->pict_type == AV_PICTURE_TYPE_B)

  1196.             return 0;

  1197.         else

  1198.             s->next_p_frame_damaged = 0;

  1199.     }

  1200. 

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

  1202.         h->frame_num_offset = h->slice_num - h->prev_frame_num;

  1203. 

  1204.         if (h->frame_num_offset < 0)

  1205.             h->frame_num_offset += 256;

  1206.         if (h->frame_num_offset == 0 ||

  1207.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1208.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1209.             return -1;

  1210.         }

  1211.     } else {

  1212.         h->prev_frame_num        = h->frame_num;

  1213.         h->frame_num             = h->slice_num;

  1214.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1215. 

  1216.         if (h->prev_frame_num_offset < 0)

  1217.             h->prev_frame_num_offset += 256;

  1218.     }

  1219. 

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

  1221.         int i;

  1222.         for (i = 0; i < 4; i++) {

  1223.             int j;

  1224.             for (j = -1; j < 4; j++)

  1225.                 h->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1226.             if (i < 3)

  1227.                 h->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1228.         }

  1229.     }

  1230. 

  1231.     for (h->mb_y = 0; h->mb_y < h->mb_height; h->mb_y++) {

  1232.         for (h->mb_x = 0; h->mb_x < h->mb_width; h->mb_x++) {

  1233.             unsigned mb_type;

  1234.             h->mb_xy = h->mb_x + h->mb_y * h->mb_stride;

  1235. 

  1236.             if ((get_bits_count(&h->gb) + 7) >= h->gb.size_in_bits &&

  1237.                 ((get_bits_count(&h->gb) & 7) == 0 ||

  1238.                  show_bits(&h->gb, -get_bits_count(&h->gb) & 7) == 0)) {

  1239.                 skip_bits(&h->gb, s->next_slice_index - get_bits_count(&h->gb));

  1240.                 h->gb.size_in_bits = 8 * buf_size;

  1241. 

  1242.                 if (svq3_decode_slice_header(avctx))

  1243.                     return -1;

  1244. 

  1245.                 /* TODO: support s->mb_skip_run */

  1246.             }

  1247. 

  1248.             mb_type = svq3_get_ue_golomb(&h->gb);

  1249. 

  1250.             if (h->pict_type == AV_PICTURE_TYPE_I)

  1251.                 mb_type += 8;

  1252.             else if (h->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1253.                 mb_type += 4;

  1254.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1255.                 av_log(h->avctx, AV_LOG_ERROR,

  1256.                        "error while decoding MB %d %d\n", h->mb_x, h->mb_y);

  1257.                 return -1;

  1258.             }

  1259. 

  1260.             if (mb_type != 0)

  1261.                 ff_h264_hl_decode_mb(h);

  1262. 

  1263.             if (h->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1264.                 h->cur_pic.mb_type[h->mb_x + h->mb_y * h->mb_stride] =

  1265.                     (h->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1266.         }

  1267. 

  1268.         ff_draw_horiz_band(avctx, NULL, s->cur_pic, s->last_pic->f.data[0] ? s->last_pic : NULL,

  1269.                            16 * h->mb_y, 16, h->picture_structure, 0, 0,

  1270.                            h->low_delay, h->mb_height * 16, h->mb_width * 16);

  1271.     }

  1272. 

  1273.     if (h->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1274.         ret = av_frame_ref(data, &s->cur_pic->f);

  1275.     else if (s->last_pic->f.data[0])

  1276.         ret = av_frame_ref(data, &s->last_pic->f);

  1277.     if (ret < 0)

  1278.         return ret;

  1279. 

  1280.     /* Do not output the last pic after seeking. */

  1281.     if (s->last_pic->f.data[0] || h->low_delay)

  1282.         *got_frame = 1;

  1283. 

  1284.     if (h->pict_type != AV_PICTURE_TYPE_B) {

  1285.         FFSWAP(Picture*, s->cur_pic, s->next_pic);

  1286.     } else {

  1287.         av_frame_unref(&s->cur_pic->f);

  1288.     }

  1289. 

  1290.     return buf_size;

  1291. }

  1292. 

  1293. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1294. {

  1295.     SVQ3Context *s = avctx->priv_data;

  1296.     H264Context *h = &s->h;

  1297. 

  1298.     free_picture(avctx, s->cur_pic);

  1299.     free_picture(avctx, s->next_pic);

  1300.     free_picture(avctx, s->last_pic);

  1301.     av_freep(&s->cur_pic);

  1302.     av_freep(&s->next_pic);

  1303.     av_freep(&s->last_pic);

  1304. 

  1305.     av_frame_unref(&h->cur_pic.f);

  1306. 

  1307.     ff_h264_free_context(h);

  1308. 

  1309.     return 0;

  1310. }

  1311. 

  1312. AVCodec ff_svq3_decoder = {

  1313.     .name           = "svq3",

  1314.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1315.     .type           = AVMEDIA_TYPE_VIDEO,

  1316.     .id             = AV_CODEC_ID_SVQ3,

  1317.     .priv_data_size = sizeof(SVQ3Context),

  1318.     .init           = svq3_decode_init,

  1319.     .close          = svq3_decode_end,

  1320.     .decode         = svq3_decode_frame,

  1321.     .capabilities   = CODEC_CAP_DRAW_HORIZ_BAND |

  1322.                       CODEC_CAP_DR1             |

  1323.                       CODEC_CAP_DELAY,

  1324.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1325.                                                      AV_PIX_FMT_NONE},

  1326. };