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.
87358 Commits
32 Branches
292MB
Tree: 2c933c5168
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '2c933c5168'
${ noResults }
FFmpeg/libavcodec/svq3.c

1659 lines
58KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The FFmpeg Project

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

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

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

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

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

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

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

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

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

  31.  *

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

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

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

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

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

  37.  *

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

  39.  * correctly decodes this file:

  40.  *  http://samples.mplayerhq.hu/V-codecs/SVQ3/Vertical400kbit.sorenson3.mov

  41.  */

  42. 

  43. #include <inttypes.h>

  44. 

  45. #include "libavutil/attributes.h"

  46. #include "internal.h"

  47. #include "avcodec.h"

  48. #include "mpegutils.h"

  49. #include "h264dec.h"

  50. #include "h264data.h"

  51. #include "golomb.h"

  52. #include "hpeldsp.h"

  53. #include "mathops.h"

  54. #include "rectangle.h"

  55. #include "tpeldsp.h"

  56. 

  57. #if CONFIG_ZLIB

  58. #include <zlib.h>

  59. #endif

  60. 

  61. #include "svq1.h"

  62. 

  63. /**

  64.  * @file

  65.  * svq3 decoder.

  66.  */

  67. 

  68. typedef struct SVQ3Frame {

  69.     AVFrame *f;

  70. 

  71.     AVBufferRef *motion_val_buf[2];

  72.     int16_t (*motion_val[2])[2];

  73. 

  74.     AVBufferRef *mb_type_buf;

  75.     uint32_t *mb_type;

  76. 

  77. 

  78.     AVBufferRef *ref_index_buf[2];

  79.     int8_t *ref_index[2];

  80. } SVQ3Frame;

  81. 

  82. typedef struct SVQ3Context {

  83.     AVCodecContext *avctx;

  84. 

  85.     H264DSPContext  h264dsp;

  86.     H264PredContext hpc;

  87.     HpelDSPContext hdsp;

  88.     TpelDSPContext tdsp;

  89.     VideoDSPContext vdsp;

  90. 

  91.     SVQ3Frame *cur_pic;

  92.     SVQ3Frame *next_pic;

  93.     SVQ3Frame *last_pic;

  94.     GetBitContext gb;

  95.     GetBitContext gb_slice;

  96.     uint8_t *slice_buf;

  97.     int slice_size;

  98.     int halfpel_flag;

  99.     int thirdpel_flag;

  100.     int has_watermark;

  101.     uint32_t watermark_key;

  102.     uint8_t *buf;

  103.     int buf_size;

  104.     int adaptive_quant;

  105.     int next_p_frame_damaged;

  106.     int h_edge_pos;

  107.     int v_edge_pos;

  108.     int last_frame_output;

  109.     int slice_num;

  110.     int qscale;

  111.     int cbp;

  112.     int frame_num;

  113.     int frame_num_offset;

  114.     int prev_frame_num_offset;

  115.     int prev_frame_num;

  116. 

  117.     enum AVPictureType pict_type;

  118.     enum AVPictureType slice_type;

  119.     int low_delay;

  120. 

  121.     int mb_x, mb_y;

  122.     int mb_xy;

  123.     int mb_width, mb_height;

  124.     int mb_stride, mb_num;

  125.     int b_stride;

  126. 

  127.     uint32_t *mb2br_xy;

  128. 

  129.     int chroma_pred_mode;

  130.     int intra16x16_pred_mode;

  131. 

  132.     int8_t   intra4x4_pred_mode_cache[5 * 8];

  133.     int8_t (*intra4x4_pred_mode);

  134. 

  135.     unsigned int top_samples_available;

  136.     unsigned int topright_samples_available;

  137.     unsigned int left_samples_available;

  138. 

  139.     uint8_t *edge_emu_buffer;

  140. 

  141.     DECLARE_ALIGNED(16, int16_t, mv_cache)[2][5 * 8][2];

  142.     DECLARE_ALIGNED(8,  int8_t, ref_cache)[2][5 * 8];

  143.     DECLARE_ALIGNED(16, int16_t, mb)[16 * 48 * 2];

  144.     DECLARE_ALIGNED(16, int16_t, mb_luma_dc)[3][16 * 2];

  145.     DECLARE_ALIGNED(8, uint8_t, non_zero_count_cache)[15 * 8];

  146.     uint32_t dequant4_coeff[QP_MAX_NUM + 1][16];

  147.     int block_offset[2 * (16 * 3)];

  148. } SVQ3Context;

  149. 

  150. #define FULLPEL_MODE  1

  151. #define HALFPEL_MODE  2

  152. #define THIRDPEL_MODE 3

  153. #define PREDICT_MODE  4

  154. 

  155. /* dual scan (from some older H.264 draft)

  156.  * o-->o-->o   o

  157.  *         |  /|

  158.  * o   o   o / o

  159.  * | / |   |/  |

  160.  * o   o   o   o

  161.  *   /

  162.  * o-->o-->o-->o

  163.  */

  164. static const uint8_t svq3_scan[16] = {

  165.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  166.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

  167.     0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4,

  168.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  169. };

  170. 

  171. static const uint8_t luma_dc_zigzag_scan[16] = {

  172.     0 * 16 + 0 * 64, 1 * 16 + 0 * 64, 2 * 16 + 0 * 64, 0 * 16 + 2 * 64,

  173.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

  174.     1 * 16 + 2 * 64, 2 * 16 + 2 * 64, 3 * 16 + 2 * 64, 0 * 16 + 3 * 64,

  175.     3 * 16 + 1 * 64, 1 * 16 + 3 * 64, 2 * 16 + 3 * 64, 3 * 16 + 3 * 64,

  176. };

  177. 

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

  179.     { 0, 0 },

  180.     { 1, 0 }, { 0, 1 },

  181.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  182.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  183.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  184.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  185.     { 2, 4 }, { 3, 3 }, { 4, 2 },

  186.     { 4, 3 }, { 3, 4 },

  187.     { 4, 4 }

  188. };

  189. 

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

  191.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  192.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

  193.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  4,  3 }, { 0, 1,  2,  4,  3 },

  194.       { 0,  2,  1,  4,  3 }, { 2, 0,  1,  3,  4 }, { 0, 4,  2,  1,  3 } },

  195.     { { 2,  0, -1, -1, -1 }, { 2, 1,  0,  4,  3 }, { 1, 2,  4,  0,  3 },

  196.       { 2,  1,  0,  4,  3 }, { 2, 1,  4,  3,  0 }, { 1, 2,  4,  0,  3 } },

  197.     { { 2,  0, -1, -1, -1 }, { 2, 0,  1,  4,  3 }, { 1, 2,  0,  4,  3 },

  198.       { 2,  1,  0,  4,  3 }, { 2, 1,  3,  4,  0 }, { 2, 4,  1,  0,  3 } },

  199.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  3,  4 }, { 1, 2,  3,  0,  4 },

  200.       { 2,  0,  1,  3,  4 }, { 2, 1,  3,  0,  4 }, { 2, 0,  4,  3,  1 } },

  201.     { { 0,  2, -1, -1, -1 }, { 0, 2,  4,  1,  3 }, { 1, 4,  2,  0,  3 },

  202.       { 4,  2,  0,  1,  3 }, { 2, 0,  1,  4,  3 }, { 4, 2,  1,  0,  3 } },

  203. };

  204. 

  205. static const struct {

  206.     uint8_t run;

  207.     uint8_t level;

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

  209.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  210.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  211.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  212.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  213. };

  214. 

  215. static const uint32_t svq3_dequant_coeff[32] = {

  216.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  217.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  218.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  219.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  220. };

  221. 

  222. static int svq3_decode_end(AVCodecContext *avctx);

  223. 

  224. static void svq3_luma_dc_dequant_idct_c(int16_t *output, int16_t *input, int qp)

  225. {

  226.     const unsigned qmul = svq3_dequant_coeff[qp];

  227. #define stride 16

  228.     int i;

  229.     int temp[16];

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

  231. 

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

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

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

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

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

  237. 

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

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

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

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

  242.     }

  243. 

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

  245.         const int offset = x_offset[i];

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

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

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

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

  250. 

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

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

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

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

  255.     }

  256. }

  257. #undef stride

  258. 

  259. static void svq3_add_idct_c(uint8_t *dst, int16_t *block,

  260.                             int stride, int qp, int dc)

  261. {

  262.     const int qmul = svq3_dequant_coeff[qp];

  263.     int i;

  264. 

  265.     if (dc) {

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

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

  268.         block[0] = 0;

  269.     }

  270. 

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

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

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

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

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

  276. 

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

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

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

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

  281.     }

  282. 

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

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

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

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

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

  288.         const int rr = (dc + 0x80000u);

  289. 

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

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

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

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

  294.     }

  295. 

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

  297. }

  298. 

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

  300.                                     int index, const int type)

  301. {

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

  303.         luma_dc_zigzag_scan, ff_zigzag_scan, svq3_scan, ff_h264_chroma_dc_scan

  304.     };

  305. 

  306.     int run, level, sign, limit;

  307.     unsigned vlc;

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

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

  310. 

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

  312.         for (; (vlc = get_interleaved_ue_golomb(gb)) != 0; index++) {

  313.             if ((int32_t)vlc < 0)

  314.                 return -1;

  315. 

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

  317.             vlc      = vlc + 1 >> 1;

  318. 

  319.             if (type == 3) {

  320.                 if (vlc < 3) {

  321.                     run   = 0;

  322.                     level = vlc;

  323.                 } else if (vlc < 4) {

  324.                     run   = 1;

  325.                     level = 1;

  326.                 } else {

  327.                     run   = vlc & 0x3;

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

  329.                 }

  330.             } else {

  331.                 if (vlc < 16U) {

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

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

  334.                 } else if (intra) {

  335.                     run   = vlc & 0x7;

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

  337.                 } else {

  338.                     run   = vlc & 0xF;

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

  340.                 }

  341.             }

  342. 

  343. 

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

  345.                 return -1;

  346. 

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

  348.         }

  349. 

  350.         if (type != 2) {

  351.             break;

  352.         }

  353.     }

  354. 

  355.     return 0;

  356. }

  357. 

  358. static av_always_inline int

  359. svq3_fetch_diagonal_mv(const SVQ3Context *s, const int16_t **C,

  360.                        int i, int list, int part_width)

  361. {

  362.     const int topright_ref = s->ref_cache[list][i - 8 + part_width];

  363. 

  364.     if (topright_ref != PART_NOT_AVAILABLE) {

  365.         *C = s->mv_cache[list][i - 8 + part_width];

  366.         return topright_ref;

  367.     } else {

  368.         *C = s->mv_cache[list][i - 8 - 1];

  369.         return s->ref_cache[list][i - 8 - 1];

  370.     }

  371. }

  372. 

  373. /**

  374.  * Get the predicted MV.

  375.  * @param n the block index

  376.  * @param part_width the width of the partition (4, 8,16) -> (1, 2, 4)

  377.  * @param mx the x component of the predicted motion vector

  378.  * @param my the y component of the predicted motion vector

  379.  */

  380. static av_always_inline void svq3_pred_motion(const SVQ3Context *s, int n,

  381.                                               int part_width, int list,

  382.                                               int ref, int *const mx, int *const my)

  383. {

  384.     const int index8       = scan8[n];

  385.     const int top_ref      = s->ref_cache[list][index8 - 8];

  386.     const int left_ref     = s->ref_cache[list][index8 - 1];

  387.     const int16_t *const A = s->mv_cache[list][index8 - 1];

  388.     const int16_t *const B = s->mv_cache[list][index8 - 8];

  389.     const int16_t *C;

  390.     int diagonal_ref, match_count;

  391. 

  392. /* mv_cache

  393.  * B . . A T T T T

  394.  * U . . L . . , .

  395.  * U . . L . . . .

  396.  * U . . L . . , .

  397.  * . . . L . . . .

  398.  */

  399. 

  400.     diagonal_ref = svq3_fetch_diagonal_mv(s, &C, index8, list, part_width);

  401.     match_count  = (diagonal_ref == ref) + (top_ref == ref) + (left_ref == ref);

  402.     if (match_count > 1) { //most common

  403.         *mx = mid_pred(A[0], B[0], C[0]);

  404.         *my = mid_pred(A[1], B[1], C[1]);

  405.     } else if (match_count == 1) {

  406.         if (left_ref == ref) {

  407.             *mx = A[0];

  408.             *my = A[1];

  409.         } else if (top_ref == ref) {

  410.             *mx = B[0];

  411.             *my = B[1];

  412.         } else {

  413.             *mx = C[0];

  414.             *my = C[1];

  415.         }

  416.     } else {

  417.         if (top_ref      == PART_NOT_AVAILABLE &&

  418.             diagonal_ref == PART_NOT_AVAILABLE &&

  419.             left_ref     != PART_NOT_AVAILABLE) {

  420.             *mx = A[0];

  421.             *my = A[1];

  422.         } else {

  423.             *mx = mid_pred(A[0], B[0], C[0]);

  424.             *my = mid_pred(A[1], B[1], C[1]);

  425.         }

  426.     }

  427. }

  428. 

  429. static inline void svq3_mc_dir_part(SVQ3Context *s,

  430.                                     int x, int y, int width, int height,

  431.                                     int mx, int my, int dxy,

  432.                                     int thirdpel, int dir, int avg)

  433. {

  434.     const SVQ3Frame *pic = (dir == 0) ? s->last_pic : s->next_pic;

  435.     uint8_t *src, *dest;

  436.     int i, emu = 0;

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

  438.     int linesize   = s->cur_pic->f->linesize[0];

  439.     int uvlinesize = s->cur_pic->f->linesize[1];

  440. 

  441.     mx += x;

  442.     my += y;

  443. 

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

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

  446.         emu = 1;

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

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

  449.     }

  450. 

  451.     /* form component predictions */

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

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

  454. 

  455.     if (emu) {

  456.         s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,

  457.                                  linesize, linesize,

  458.                                  width + 1, height + 1,

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

  460.         src = s->edge_emu_buffer;

  461.     }

  462.     if (thirdpel)

  463.         (avg ? s->tdsp.avg_tpel_pixels_tab

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

  465.                                                  width, height);

  466.     else

  467.         (avg ? s->hdsp.avg_pixels_tab

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

  469.                                                        height);

  470. 

  471.     if (!(s->avctx->flags & AV_CODEC_FLAG_GRAY)) {

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

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

  474.         width  = width  >> 1;

  475.         height = height >> 1;

  476.         blocksize++;

  477. 

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

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

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

  481. 

  482.             if (emu) {

  483.                 s->vdsp.emulated_edge_mc(s->edge_emu_buffer, src,

  484.                                          uvlinesize, uvlinesize,

  485.                                          width + 1, height + 1,

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

  487.                                          s->v_edge_pos >> 1);

  488.                 src = s->edge_emu_buffer;

  489.             }

  490.             if (thirdpel)

  491.                 (avg ? s->tdsp.avg_tpel_pixels_tab

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

  493.                                                          uvlinesize,

  494.                                                          width, height);

  495.             else

  496.                 (avg ? s->hdsp.avg_pixels_tab

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

  498.                                                                uvlinesize,

  499.                                                                height);

  500.         }

  501.     }

  502. }

  503. 

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

  505.                               int dir, int avg)

  506. {

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

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

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

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

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

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

  513. 

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

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

  516.             const int b_xy = (4 * s->mb_x + (j >> 2)) +

  517.                              (4 * s->mb_y + (i >> 2)) * s->b_stride;

  518.             int dxy;

  519.             x = 16 * s->mb_x + j;

  520.             y = 16 * s->mb_y + i;

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

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

  523. 

  524.             if (mode != PREDICT_MODE) {

  525.                 svq3_pred_motion(s, k, part_width >> 2, dir, 1, &mx, &my);

  526.             } else {

  527.                 mx = s->next_pic->motion_val[0][b_xy][0] * 2;

  528.                 my = s->next_pic->motion_val[0][b_xy][1] * 2;

  529. 

  530.                 if (dir == 0) {

  531.                     mx = mx * s->frame_num_offset /

  532.                          s->prev_frame_num_offset + 1 >> 1;

  533.                     my = my * s->frame_num_offset /

  534.                          s->prev_frame_num_offset + 1 >> 1;

  535.                 } else {

  536.                     mx = mx * (s->frame_num_offset - s->prev_frame_num_offset) /

  537.                          s->prev_frame_num_offset + 1 >> 1;

  538.                     my = my * (s->frame_num_offset - s->prev_frame_num_offset) /

  539.                          s->prev_frame_num_offset + 1 >> 1;

  540.                 }

  541.             }

  542. 

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

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

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

  546. 

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

  548.             if (mode == PREDICT_MODE) {

  549.                 dx = dy = 0;

  550.             } else {

  551.                 dy = get_interleaved_se_golomb(&s->gb_slice);

  552.                 dx = get_interleaved_se_golomb(&s->gb_slice);

  553. 

  554.                 if (dx != (int16_t)dx || dy != (int16_t)dy) {

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

  556.                     return -1;

  557.                 }

  558.             }

  559. 

  560.             /* compute motion vector */

  561.             if (mode == THIRDPEL_MODE) {

  562.                 int fx, fy;

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

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

  565.                 fx  = (unsigned)(mx + 0x30000) / 3 - 0x10000;

  566.                 fy  = (unsigned)(my + 0x30000) / 3 - 0x10000;

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

  568. 

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

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

  571.                 mx += mx;

  572.                 my += my;

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

  574.                 mx  = (unsigned)(mx + 1 + 0x30000) / 3 + dx - 0x10000;

  575.                 my  = (unsigned)(my + 1 + 0x30000) / 3 + dy - 0x10000;

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

  577. 

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

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

  580.                 mx *= 3;

  581.                 my *= 3;

  582.             } else {

  583.                 mx = (unsigned)(mx + 3 + 0x60000) / 6 + dx - 0x10000;

  584.                 my = (unsigned)(my + 3 + 0x60000) / 6 + dy - 0x10000;

  585. 

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

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

  588.                 mx *= 6;

  589.                 my *= 6;

  590.             }

  591. 

  592.             /* update mv_cache */

  593.             if (mode != PREDICT_MODE) {

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

  595. 

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

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

  598. 

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

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

  601.                 }

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

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

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

  605.                     AV_WN32A(s->mv_cache[dir][scan8[k]], mv);

  606.             }

  607. 

  608.             /* write back motion vectors */

  609.             fill_rectangle(s->cur_pic->motion_val[dir][b_xy],

  610.                            part_width >> 2, part_height >> 2, s->b_stride,

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

  612.         }

  613. 

  614.     return 0;

  615. }

  616. 

  617. static av_always_inline void hl_decode_mb_idct_luma(SVQ3Context *s,

  618.                                                     int mb_type, const int *block_offset,

  619.                                                     int linesize, uint8_t *dest_y)

  620. {

  621.     int i;

  622.     if (!IS_INTRA4x4(mb_type)) {

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

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

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

  626.                 svq3_add_idct_c(ptr, s->mb + i * 16, linesize,

  627.                                 s->qscale, IS_INTRA(mb_type) ? 1 : 0);

  628.             }

  629.     }

  630. }

  631. 

  632. static av_always_inline void hl_decode_mb_predict_luma(SVQ3Context *s,

  633.                                                        int mb_type,

  634.                                                        const int *block_offset,

  635.                                                        int linesize,

  636.                                                        uint8_t *dest_y)

  637. {

  638.     int i;

  639.     int qscale = s->qscale;

  640. 

  641.     if (IS_INTRA4x4(mb_type)) {

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

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

  644.             const int dir      = s->intra4x4_pred_mode_cache[scan8[i]];

  645. 

  646.             uint8_t *topright;

  647.             int nnz, tr;

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

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

  650.                 av_assert2(s->mb_y || linesize <= block_offset[i]);

  651.                 if (!topright_avail) {

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

  653.                     topright = (uint8_t *)&tr;

  654.                 } else

  655.                     topright = ptr + 4 - linesize;

  656.             } else

  657.                 topright = NULL;

  658. 

  659.             s->hpc.pred4x4[dir](ptr, topright, linesize);

  660.             nnz = s->non_zero_count_cache[scan8[i]];

  661.             if (nnz) {

  662.                 svq3_add_idct_c(ptr, s->mb + i * 16, linesize, qscale, 0);

  663.             }

  664.         }

  665.     } else {

  666.         s->hpc.pred16x16[s->intra16x16_pred_mode](dest_y, linesize);

  667.         svq3_luma_dc_dequant_idct_c(s->mb, s->mb_luma_dc[0], qscale);

  668.     }

  669. }

  670. 

  671. static void hl_decode_mb(SVQ3Context *s)

  672. {

  673.     const int mb_x    = s->mb_x;

  674.     const int mb_y    = s->mb_y;

  675.     const int mb_xy   = s->mb_xy;

  676.     const int mb_type = s->cur_pic->mb_type[mb_xy];

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

  678.     int linesize, uvlinesize;

  679.     int i, j;

  680.     const int *block_offset = &s->block_offset[0];

  681.     const int block_h   = 16 >> 1;

  682. 

  683.     linesize   = s->cur_pic->f->linesize[0];

  684.     uvlinesize = s->cur_pic->f->linesize[1];

  685. 

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

  687.     dest_cb = s->cur_pic->f->data[1] +  mb_x * 8 + mb_y * uvlinesize * block_h;

  688.     dest_cr = s->cur_pic->f->data[2] +  mb_x * 8 + mb_y * uvlinesize * block_h;

  689. 

  690.     s->vdsp.prefetch(dest_y  + (s->mb_x & 3) * 4 * linesize   + 64, linesize,      4);

  691.     s->vdsp.prefetch(dest_cb + (s->mb_x & 7)     * uvlinesize + 64, dest_cr - dest_cb, 2);

  692. 

  693.     if (IS_INTRA(mb_type)) {

  694.         s->hpc.pred8x8[s->chroma_pred_mode](dest_cb, uvlinesize);

  695.         s->hpc.pred8x8[s->chroma_pred_mode](dest_cr, uvlinesize);

  696. 

  697.         hl_decode_mb_predict_luma(s, mb_type, block_offset, linesize, dest_y);

  698.     }

  699. 

  700.     hl_decode_mb_idct_luma(s, mb_type, block_offset, linesize, dest_y);

  701. 

  702.     if (s->cbp & 0x30) {

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

  704.         s->h264dsp.h264_chroma_dc_dequant_idct(s->mb + 16 * 16 * 1,

  705.                                                s->dequant4_coeff[4][0]);

  706.         s->h264dsp.h264_chroma_dc_dequant_idct(s->mb + 16 * 16 * 2,

  707.                                                s->dequant4_coeff[4][0]);

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

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

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

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

  712.                     svq3_add_idct_c(ptr, s->mb + i * 16,

  713.                                     uvlinesize, ff_h264_chroma_qp[0][s->qscale + 12] - 12, 2);

  714.                 }

  715.         }

  716.     }

  717. }

  718. 

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

  720. {

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

  722.     int cbp = 0;

  723.     uint32_t vlc;

  724.     int8_t *top, *left;

  725.     const int mb_xy = s->mb_xy;

  726.     const int b_xy  = 4 * s->mb_x + 4 * s->mb_y * s->b_stride;

  727. 

  728.     s->top_samples_available      = (s->mb_y == 0) ? 0x33FF : 0xFFFF;

  729.     s->left_samples_available     = (s->mb_x == 0) ? 0x5F5F : 0xFFFF;

  730.     s->topright_samples_available = 0xFFFF;

  731. 

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

  733.         if (s->pict_type == AV_PICTURE_TYPE_P ||

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

  735.             svq3_mc_dir_part(s, 16 * s->mb_x, 16 * s->mb_y, 16, 16,

  736.                              0, 0, 0, 0, 0, 0);

  737. 

  738.             if (s->pict_type == AV_PICTURE_TYPE_B)

  739.                 svq3_mc_dir_part(s, 16 * s->mb_x, 16 * s->mb_y, 16, 16,

  740.                                  0, 0, 0, 0, 1, 1);

  741. 

  742.             mb_type = MB_TYPE_SKIP;

  743.         } else {

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

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

  746.                 return -1;

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

  748.                 return -1;

  749. 

  750.             mb_type = MB_TYPE_16x16;

  751.         }

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

  753.         if (s->thirdpel_flag && s->halfpel_flag == !get_bits1(&s->gb_slice))

  754.             mode = THIRDPEL_MODE;

  755.         else if (s->halfpel_flag &&

  756.                  s->thirdpel_flag == !get_bits1(&s->gb_slice))

  757.             mode = HALFPEL_MODE;

  758.         else

  759.             mode = FULLPEL_MODE;

  760. 

  761.         /* fill caches */

  762.         /* note ref_cache should contain here:

  763.          *  ????????

  764.          *  ???11111

  765.          *  N??11111

  766.          *  N??11111

  767.          *  N??11111

  768.          */

  769. 

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

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

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

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

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

  775.             } else {

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

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

  778.             }

  779.             if (s->mb_y > 0) {

  780.                 memcpy(s->mv_cache[m][scan8[0] - 1 * 8],

  781.                        s->cur_pic->motion_val[m][b_xy - s->b_stride],

  782.                        4 * 2 * sizeof(int16_t));

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

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

  785. 

  786.                 if (s->mb_x < s->mb_width - 1) {

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

  788.                               s->cur_pic->motion_val[m][b_xy - s->b_stride + 4]);

  789.                     s->ref_cache[m][scan8[0] + 4 - 1 * 8] =

  790.                         (s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride + 1] + 6] == -1 ||

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

  792.                 } else

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

  794.                 if (s->mb_x > 0) {

  795.                     AV_COPY32(s->mv_cache[m][scan8[0] - 1 - 1 * 8],

  796.                               s->cur_pic->motion_val[m][b_xy - s->b_stride - 1]);

  797.                     s->ref_cache[m][scan8[0] - 1 - 1 * 8] =

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

  799.                 } else

  800.                     s->ref_cache[m][scan8[0] - 1 - 1 * 8] = PART_NOT_AVAILABLE;

  801.             } else

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

  803.                        PART_NOT_AVAILABLE, 8);

  804. 

  805.             if (s->pict_type != AV_PICTURE_TYPE_B)

  806.                 break;

  807.         }

  808. 

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

  810.         if (s->pict_type == AV_PICTURE_TYPE_P) {

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

  812.                 return -1;

  813.         } else {        /* AV_PICTURE_TYPE_B */

  814.             if (mb_type != 2) {

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

  816.                     return -1;

  817.             } else {

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

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

  820.                            0, 4 * 2 * sizeof(int16_t));

  821.             }

  822.             if (mb_type != 1) {

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

  824.                     return -1;

  825.             } else {

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

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

  828.                            0, 4 * 2 * sizeof(int16_t));

  829.             }

  830.         }

  831. 

  832.         mb_type = MB_TYPE_16x16;

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

  834.         int8_t *i4x4       = s->intra4x4_pred_mode + s->mb2br_xy[s->mb_xy];

  835.         int8_t *i4x4_cache = s->intra4x4_pred_mode_cache;

  836. 

  837.         memset(s->intra4x4_pred_mode_cache, -1, 8 * 5 * sizeof(int8_t));

  838. 

  839.         if (mb_type == 8) {

  840.             if (s->mb_x > 0) {

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

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

  843.                 if (s->intra4x4_pred_mode_cache[scan8[0] - 1] == -1)

  844.                     s->left_samples_available = 0x5F5F;

  845.             }

  846.             if (s->mb_y > 0) {

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

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

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

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

  851. 

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

  853.                     s->top_samples_available = 0x33FF;

  854.             }

  855. 

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

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

  858.                 vlc = get_interleaved_ue_golomb(&s->gb_slice);

  859. 

  860.                 if (vlc >= 25U) {

  861.                     av_log(s->avctx, AV_LOG_ERROR,

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

  863.                     return -1;

  864.                 }

  865. 

  866.                 left = &s->intra4x4_pred_mode_cache[scan8[i] - 1];

  867.                 top  = &s->intra4x4_pred_mode_cache[scan8[i] - 8];

  868. 

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

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

  871. 

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

  873.                     av_log(s->avctx, AV_LOG_ERROR, "weird prediction\n");

  874.                     return -1;

  875.                 }

  876.             }

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

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

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

  880.         }

  881. 

  882.         AV_COPY32(i4x4, i4x4_cache + 4 + 8 * 4);

  883.         i4x4[4] = i4x4_cache[7 + 8 * 3];

  884.         i4x4[5] = i4x4_cache[7 + 8 * 2];

  885.         i4x4[6] = i4x4_cache[7 + 8 * 1];

  886. 

  887.         if (mb_type == 8) {

  888.             ff_h264_check_intra4x4_pred_mode(s->intra4x4_pred_mode_cache,

  889.                                              s->avctx, s->top_samples_available,

  890.                                              s->left_samples_available);

  891. 

  892.             s->top_samples_available  = (s->mb_y == 0) ? 0x33FF : 0xFFFF;

  893.             s->left_samples_available = (s->mb_x == 0) ? 0x5F5F : 0xFFFF;

  894.         } else {

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

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

  897. 

  898.             s->top_samples_available  = 0x33FF;

  899.             s->left_samples_available = 0x5F5F;

  900.         }

  901. 

  902.         mb_type = MB_TYPE_INTRA4x4;

  903.     } else {                      /* INTRA16x16 */

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

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

  906. 

  907.         if ((s->intra16x16_pred_mode = ff_h264_check_intra_pred_mode(s->avctx, s->top_samples_available,

  908.                                                                      s->left_samples_available, dir, 0)) < 0) {

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

  910.             return s->intra16x16_pred_mode;

  911.         }

  912. 

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

  914.         mb_type = MB_TYPE_INTRA16x16;

  915.     }

  916. 

  917.     if (!IS_INTER(mb_type) && s->pict_type != AV_PICTURE_TYPE_I) {

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

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

  920.                    0, 4 * 2 * sizeof(int16_t));

  921.         if (s->pict_type == AV_PICTURE_TYPE_B) {

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

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

  924.                        0, 4 * 2 * sizeof(int16_t));

  925.         }

  926.     }

  927.     if (!IS_INTRA4x4(mb_type)) {

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

  929.     }

  930.     if (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B) {

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

  932.     }

  933. 

  934.     if (!IS_INTRA16x16(mb_type) &&

  935.         (!IS_SKIP(mb_type) || s->pict_type == AV_PICTURE_TYPE_B)) {

  936.         if ((vlc = get_interleaved_ue_golomb(&s->gb_slice)) >= 48U){

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

  938.             return -1;

  939.         }

  940. 

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

  942.                                 : ff_h264_golomb_to_inter_cbp[vlc];

  943.     }

  944.     if (IS_INTRA16x16(mb_type) ||

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

  946.         s->qscale += get_interleaved_se_golomb(&s->gb_slice);

  947. 

  948.         if (s->qscale > 31u) {

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

  950.             return -1;

  951.         }

  952.     }

  953.     if (IS_INTRA16x16(mb_type)) {

  954.         AV_ZERO128(s->mb_luma_dc[0] + 0);

  955.         AV_ZERO128(s->mb_luma_dc[0] + 8);

  956.         if (svq3_decode_block(&s->gb_slice, s->mb_luma_dc[0], 0, 1)) {

  957.             av_log(s->avctx, AV_LOG_ERROR,

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

  959.             return -1;

  960.         }

  961.     }

  962. 

  963.     if (cbp) {

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

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

  966. 

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

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

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

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

  971.                                  2 * (j & 2) + 4 * (i & 2))

  972.                               : (4 * i + j);

  973.                     s->non_zero_count_cache[scan8[k]] = 1;

  974. 

  975.                     if (svq3_decode_block(&s->gb_slice, &s->mb[16 * k], index, type)) {

  976.                         av_log(s->avctx, AV_LOG_ERROR,

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

  978.                         return -1;

  979.                     }

  980.                 }

  981.             }

  982. 

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

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

  985.                 if (svq3_decode_block(&s->gb_slice, &s->mb[16 * 16 * i], 0, 3)) {

  986.                     av_log(s->avctx, AV_LOG_ERROR,

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

  988.                     return -1;

  989.                 }

  990. 

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

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

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

  994.                         k                                 = 16 * i + j;

  995.                         s->non_zero_count_cache[scan8[k]] = 1;

  996. 

  997.                         if (svq3_decode_block(&s->gb_slice, &s->mb[16 * k], 1, 1)) {

  998.                             av_log(s->avctx, AV_LOG_ERROR,

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

  1000.                             return -1;

  1001.                         }

  1002.                     }

  1003.                 }

  1004.             }

  1005.         }

  1006.     }

  1007. 

  1008.     s->cbp                     = cbp;

  1009.     s->cur_pic->mb_type[mb_xy] = mb_type;

  1010. 

  1011.     if (IS_INTRA(mb_type))

  1012.         s->chroma_pred_mode = ff_h264_check_intra_pred_mode(s->avctx, s->top_samples_available,

  1013.                                                             s->left_samples_available, DC_PRED8x8, 1);

  1014. 

  1015.     return 0;

  1016. }

  1017. 

  1018. static int svq3_decode_slice_header(AVCodecContext *avctx)

  1019. {

  1020.     SVQ3Context *s = avctx->priv_data;

  1021.     const int mb_xy   = s->mb_xy;

  1022.     int i, header;

  1023.     unsigned slice_id;

  1024. 

  1025.     header = get_bits(&s->gb, 8);

  1026. 

  1027.     if (((header & 0x9F) != 1 && (header & 0x9F) != 2) || (header & 0x60) == 0) {

  1028.         /* TODO: what? */

  1029.         av_log(avctx, AV_LOG_ERROR, "unsupported slice header (%02X)\n", header);

  1030.         return -1;

  1031.     } else {

  1032.         int slice_bits, slice_bytes, slice_length;

  1033.         int length = header >> 5 & 3;

  1034. 

  1035.         slice_length = show_bits(&s->gb, 8 * length);

  1036.         slice_bits   = slice_length * 8;

  1037.         slice_bytes  = slice_length + length - 1;

  1038. 

  1039.         if (8LL*slice_bytes > get_bits_left(&s->gb)) {

  1040.             av_log(avctx, AV_LOG_ERROR, "slice after bitstream end\n");

  1041.             return -1;

  1042.         }

  1043. 

  1044.         skip_bits(&s->gb, 8);

  1045. 

  1046.         av_fast_malloc(&s->slice_buf, &s->slice_size, slice_bytes + AV_INPUT_BUFFER_PADDING_SIZE);

  1047.         if (!s->slice_buf)

  1048.             return AVERROR(ENOMEM);

  1049. 

  1050.         memcpy(s->slice_buf, s->gb.buffer + s->gb.index / 8, slice_bytes);

  1051. 

  1052.         init_get_bits(&s->gb_slice, s->slice_buf, slice_bits);

  1053. 

  1054.         if (s->watermark_key) {

  1055.             uint32_t header = AV_RL32(&s->gb_slice.buffer[1]);

  1056.             AV_WL32(&s->gb_slice.buffer[1], header ^ s->watermark_key);

  1057.         }

  1058.         if (length > 0) {

  1059.             memmove(s->slice_buf, &s->slice_buf[slice_length], length - 1);

  1060.         }

  1061.         skip_bits_long(&s->gb, slice_bytes * 8);

  1062.     }

  1063. 

  1064.     if ((slice_id = get_interleaved_ue_golomb(&s->gb_slice)) >= 3) {

  1065.         av_log(s->avctx, AV_LOG_ERROR, "illegal slice type %u \n", slice_id);

  1066.         return -1;

  1067.     }

  1068.     if (get_bits1(&s->gb_slice)) {

  1069.         avpriv_report_missing_feature(s->avctx, "Media key encryption");

  1070.         return AVERROR_PATCHWELCOME;

  1071.     }

  1072. 

  1073.     s->slice_type = ff_h264_golomb_to_pict_type[slice_id];

  1074. 

  1075.     if ((header & 0x9F) == 2) {

  1076.         i = (s->mb_num < 64) ? 5 : av_log2(s->mb_num - 1);

  1077.         get_bits(&s->gb_slice, i);

  1078.     }

  1079. 

  1080.     s->slice_num      = get_bits(&s->gb_slice, 8);

  1081.     s->qscale         = get_bits(&s->gb_slice, 5);

  1082.     s->adaptive_quant = get_bits1(&s->gb_slice);

  1083. 

  1084.     /* unknown fields */

  1085.     skip_bits1(&s->gb_slice);

  1086. 

  1087.     if (s->has_watermark)

  1088.         skip_bits1(&s->gb_slice);

  1089. 

  1090.     skip_bits1(&s->gb_slice);

  1091.     skip_bits(&s->gb_slice, 2);

  1092. 

  1093.     if (skip_1stop_8data_bits(&s->gb_slice) < 0)

  1094.         return AVERROR_INVALIDDATA;

  1095. 

  1096.     /* reset intra predictors and invalidate motion vector references */

  1097.     if (s->mb_x > 0) {

  1098.         memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - 1] + 3,

  1099.                -1, 4 * sizeof(int8_t));

  1100.         memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - s->mb_x],

  1101.                -1, 8 * sizeof(int8_t) * s->mb_x);

  1102.     }

  1103.     if (s->mb_y > 0) {

  1104.         memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - s->mb_stride],

  1105.                -1, 8 * sizeof(int8_t) * (s->mb_width - s->mb_x));

  1106. 

  1107.         if (s->mb_x > 0)

  1108.             s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride - 1] + 3] = -1;

  1109.     }

  1110. 

  1111.     return 0;

  1112. }

  1113. 

  1114. static void init_dequant4_coeff_table(SVQ3Context *s)

  1115. {

  1116.     int q, x;

  1117.     const int max_qp = 51;

  1118. 

  1119.     for (q = 0; q < max_qp + 1; q++) {

  1120.         int shift = ff_h264_quant_div6[q] + 2;

  1121.         int idx   = ff_h264_quant_rem6[q];

  1122.         for (x = 0; x < 16; x++)

  1123.             s->dequant4_coeff[q][(x >> 2) | ((x << 2) & 0xF)] =

  1124.                 ((uint32_t)ff_h264_dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] * 16) << shift;

  1125.     }

  1126. }

  1127. 

  1128. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  1129. {

  1130.     SVQ3Context *s = avctx->priv_data;

  1131.     int m, x, y;

  1132.     unsigned char *extradata;

  1133.     unsigned char *extradata_end;

  1134.     unsigned int size;

  1135.     int marker_found = 0;

  1136.     int ret;

  1137. 

  1138.     s->cur_pic  = av_mallocz(sizeof(*s->cur_pic));

  1139.     s->last_pic = av_mallocz(sizeof(*s->last_pic));

  1140.     s->next_pic = av_mallocz(sizeof(*s->next_pic));

  1141.     if (!s->next_pic || !s->last_pic || !s->cur_pic) {

  1142.         ret = AVERROR(ENOMEM);

  1143.         goto fail;

  1144.     }

  1145. 

  1146.     s->cur_pic->f  = av_frame_alloc();

  1147.     s->last_pic->f = av_frame_alloc();

  1148.     s->next_pic->f = av_frame_alloc();

  1149.     if (!s->cur_pic->f || !s->last_pic->f || !s->next_pic->f)

  1150.         return AVERROR(ENOMEM);

  1151. 

  1152.     ff_h264dsp_init(&s->h264dsp, 8, 1);

  1153.     ff_h264_pred_init(&s->hpc, AV_CODEC_ID_SVQ3, 8, 1);

  1154.     ff_videodsp_init(&s->vdsp, 8);

  1155. 

  1156. 

  1157.     avctx->bits_per_raw_sample = 8;

  1158. 

  1159.     ff_hpeldsp_init(&s->hdsp, avctx->flags);

  1160.     ff_tpeldsp_init(&s->tdsp);

  1161. 

  1162.     avctx->pix_fmt     = AV_PIX_FMT_YUVJ420P;

  1163.     avctx->color_range = AVCOL_RANGE_JPEG;

  1164. 

  1165.     s->avctx         = avctx;

  1166.     s->halfpel_flag  = 1;

  1167.     s->thirdpel_flag = 1;

  1168.     s->has_watermark = 0;

  1169. 

  1170.     /* prowl for the "SEQH" marker in the extradata */

  1171.     extradata     = (unsigned char *)avctx->extradata;

  1172.     extradata_end = avctx->extradata + avctx->extradata_size;

  1173.     if (extradata) {

  1174.         for (m = 0; m + 8 < avctx->extradata_size; m++) {

  1175.             if (!memcmp(extradata, "SEQH", 4)) {

  1176.                 marker_found = 1;

  1177.                 break;

  1178.             }

  1179.             extradata++;

  1180.         }

  1181.     }

  1182. 

  1183.     /* if a match was found, parse the extra data */

  1184.     if (marker_found) {

  1185.         GetBitContext gb;

  1186.         int frame_size_code;

  1187.         int unk0, unk1, unk2, unk3, unk4;

  1188. 

  1189.         size = AV_RB32(&extradata[4]);

  1190.         if (size > extradata_end - extradata - 8) {

  1191.             ret = AVERROR_INVALIDDATA;

  1192.             goto fail;

  1193.         }

  1194.         init_get_bits(&gb, extradata + 8, size * 8);

  1195. 

  1196.         /* 'frame size code' and optional 'width, height' */

  1197.         frame_size_code = get_bits(&gb, 3);

  1198.         switch (frame_size_code) {

  1199.         case 0:

  1200.             avctx->width  = 160;

  1201.             avctx->height = 120;

  1202.             break;

  1203.         case 1:

  1204.             avctx->width  = 128;

  1205.             avctx->height =  96;

  1206.             break;

  1207.         case 2:

  1208.             avctx->width  = 176;

  1209.             avctx->height = 144;

  1210.             break;

  1211.         case 3:

  1212.             avctx->width  = 352;

  1213.             avctx->height = 288;

  1214.             break;

  1215.         case 4:

  1216.             avctx->width  = 704;

  1217.             avctx->height = 576;

  1218.             break;

  1219.         case 5:

  1220.             avctx->width  = 240;

  1221.             avctx->height = 180;

  1222.             break;

  1223.         case 6:

  1224.             avctx->width  = 320;

  1225.             avctx->height = 240;

  1226.             break;

  1227.         case 7:

  1228.             avctx->width  = get_bits(&gb, 12);

  1229.             avctx->height = get_bits(&gb, 12);

  1230.             break;

  1231.         }

  1232. 

  1233.         s->halfpel_flag  = get_bits1(&gb);

  1234.         s->thirdpel_flag = get_bits1(&gb);

  1235. 

  1236.         /* unknown fields */

  1237.         unk0 = get_bits1(&gb);

  1238.         unk1 = get_bits1(&gb);

  1239.         unk2 = get_bits1(&gb);

  1240.         unk3 = get_bits1(&gb);

  1241. 

  1242.         s->low_delay = get_bits1(&gb);

  1243. 

  1244.         /* unknown field */

  1245.         unk4 = get_bits1(&gb);

  1246. 

  1247.         av_log(avctx, AV_LOG_DEBUG, "Unknown fields %d %d %d %d %d\n",

  1248.                unk0, unk1, unk2, unk3, unk4);

  1249. 

  1250.         if (skip_1stop_8data_bits(&gb) < 0) {

  1251.             ret = AVERROR_INVALIDDATA;

  1252.             goto fail;

  1253.         }

  1254. 

  1255.         s->has_watermark  = get_bits1(&gb);

  1256.         avctx->has_b_frames = !s->low_delay;

  1257.         if (s->has_watermark) {

  1258. #if CONFIG_ZLIB

  1259.             unsigned watermark_width  = get_interleaved_ue_golomb(&gb);

  1260.             unsigned watermark_height = get_interleaved_ue_golomb(&gb);

  1261.             int u1                    = get_interleaved_ue_golomb(&gb);

  1262.             int u2                    = get_bits(&gb, 8);

  1263.             int u3                    = get_bits(&gb, 2);

  1264.             int u4                    = get_interleaved_ue_golomb(&gb);

  1265.             unsigned long buf_len     = watermark_width *

  1266.                                         watermark_height * 4;

  1267.             int offset                = get_bits_count(&gb) + 7 >> 3;

  1268.             uint8_t *buf;

  1269. 

  1270.             if (watermark_height <= 0 ||

  1271.                 (uint64_t)watermark_width * 4 > UINT_MAX / watermark_height) {

  1272.                 ret = -1;

  1273.                 goto fail;

  1274.             }

  1275. 

  1276.             buf = av_malloc(buf_len);

  1277.             if (!buf) {

  1278.                 ret = AVERROR(ENOMEM);

  1279.                 goto fail;

  1280.             }

  1281.             av_log(avctx, AV_LOG_DEBUG, "watermark size: %ux%u\n",

  1282.                    watermark_width, watermark_height);

  1283.             av_log(avctx, AV_LOG_DEBUG,

  1284.                    "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n",

  1285.                    u1, u2, u3, u4, offset);

  1286.             if (uncompress(buf, &buf_len, extradata + 8 + offset,

  1287.                            size - offset) != Z_OK) {

  1288.                 av_log(avctx, AV_LOG_ERROR,

  1289.                        "could not uncompress watermark logo\n");

  1290.                 av_free(buf);

  1291.                 ret = -1;

  1292.                 goto fail;

  1293.             }

  1294.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1295.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1296.             av_log(avctx, AV_LOG_DEBUG,

  1297.                    "watermark key %#"PRIx32"\n", s->watermark_key);

  1298.             av_free(buf);

  1299. #else

  1300.             av_log(avctx, AV_LOG_ERROR,

  1301.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1302.             ret = -1;

  1303.             goto fail;

  1304. #endif

  1305.         }

  1306.     }

  1307. 

  1308.     s->mb_width   = (avctx->width + 15) / 16;

  1309.     s->mb_height  = (avctx->height + 15) / 16;

  1310.     s->mb_stride  = s->mb_width + 1;

  1311.     s->mb_num     = s->mb_width * s->mb_height;

  1312.     s->b_stride   = 4 * s->mb_width;

  1313.     s->h_edge_pos = s->mb_width * 16;

  1314.     s->v_edge_pos = s->mb_height * 16;

  1315. 

  1316.     s->intra4x4_pred_mode = av_mallocz(s->mb_stride * 2 * 8);

  1317.     if (!s->intra4x4_pred_mode)

  1318.         return AVERROR(ENOMEM);

  1319. 

  1320.     s->mb2br_xy = av_mallocz(s->mb_stride * (s->mb_height + 1) *

  1321.                              sizeof(*s->mb2br_xy));

  1322.     if (!s->mb2br_xy)

  1323.         return AVERROR(ENOMEM);

  1324. 

  1325.     for (y = 0; y < s->mb_height; y++)

  1326.         for (x = 0; x < s->mb_width; x++) {

  1327.             const int mb_xy = x + y * s->mb_stride;

  1328. 

  1329.             s->mb2br_xy[mb_xy] = 8 * (mb_xy % (2 * s->mb_stride));

  1330.         }

  1331. 

  1332.     init_dequant4_coeff_table(s);

  1333. 

  1334.     return 0;

  1335. fail:

  1336.     svq3_decode_end(avctx);

  1337.     return ret;

  1338. }

  1339. 

  1340. static void free_picture(AVCodecContext *avctx, SVQ3Frame *pic)

  1341. {

  1342.     int i;

  1343.     for (i = 0; i < 2; i++) {

  1344.         av_buffer_unref(&pic->motion_val_buf[i]);

  1345.         av_buffer_unref(&pic->ref_index_buf[i]);

  1346.     }

  1347.     av_buffer_unref(&pic->mb_type_buf);

  1348. 

  1349.     av_frame_unref(pic->f);

  1350. }

  1351. 

  1352. static int get_buffer(AVCodecContext *avctx, SVQ3Frame *pic)

  1353. {

  1354.     SVQ3Context *s = avctx->priv_data;

  1355.     const int big_mb_num    = s->mb_stride * (s->mb_height + 1) + 1;

  1356.     const int mb_array_size = s->mb_stride * s->mb_height;

  1357.     const int b4_stride     = s->mb_width * 4 + 1;

  1358.     const int b4_array_size = b4_stride * s->mb_height * 4;

  1359.     int ret;

  1360. 

  1361.     if (!pic->motion_val_buf[0]) {

  1362.         int i;

  1363. 

  1364.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + s->mb_stride) * sizeof(uint32_t));

  1365.         if (!pic->mb_type_buf)

  1366.             return AVERROR(ENOMEM);

  1367.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * s->mb_stride + 1;

  1368. 

  1369.         for (i = 0; i < 2; i++) {

  1370.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1371.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1372.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1373.                 ret = AVERROR(ENOMEM);

  1374.                 goto fail;

  1375.             }

  1376. 

  1377.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1378.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1379.         }

  1380.     }

  1381. 

  1382.     ret = ff_get_buffer(avctx, pic->f,

  1383.                         (s->pict_type != AV_PICTURE_TYPE_B) ?

  1384.                          AV_GET_BUFFER_FLAG_REF : 0);

  1385.     if (ret < 0)

  1386.         goto fail;

  1387. 

  1388.     if (!s->edge_emu_buffer) {

  1389.         s->edge_emu_buffer = av_mallocz_array(pic->f->linesize[0], 17);

  1390.         if (!s->edge_emu_buffer)

  1391.             return AVERROR(ENOMEM);

  1392.     }

  1393. 

  1394.     return 0;

  1395. fail:

  1396.     free_picture(avctx, pic);

  1397.     return ret;

  1398. }

  1399. 

  1400. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1401.                              int *got_frame, AVPacket *avpkt)

  1402. {

  1403.     SVQ3Context *s     = avctx->priv_data;

  1404.     int buf_size       = avpkt->size;

  1405.     int left;

  1406.     uint8_t *buf;

  1407.     int ret, m, i;

  1408. 

  1409.     /* special case for last picture */

  1410.     if (buf_size == 0) {

  1411.         if (s->next_pic->f->data[0] && !s->low_delay && !s->last_frame_output) {

  1412.             ret = av_frame_ref(data, s->next_pic->f);

  1413.             if (ret < 0)

  1414.                 return ret;

  1415.             s->last_frame_output = 1;

  1416.             *got_frame          = 1;

  1417.         }

  1418.         return 0;

  1419.     }

  1420. 

  1421.     s->mb_x = s->mb_y = s->mb_xy = 0;

  1422. 

  1423.     if (s->watermark_key) {

  1424.         av_fast_padded_malloc(&s->buf, &s->buf_size, buf_size);

  1425.         if (!s->buf)

  1426.             return AVERROR(ENOMEM);

  1427.         memcpy(s->buf, avpkt->data, buf_size);

  1428.         buf = s->buf;

  1429.     } else {

  1430.         buf = avpkt->data;

  1431.     }

  1432. 

  1433.     ret = init_get_bits(&s->gb, buf, 8 * buf_size);

  1434.     if (ret < 0)

  1435.         return ret;

  1436. 

  1437.     if (svq3_decode_slice_header(avctx))

  1438.         return -1;

  1439. 

  1440.     s->pict_type = s->slice_type;

  1441. 

  1442.     if (s->pict_type != AV_PICTURE_TYPE_B)

  1443.         FFSWAP(SVQ3Frame*, s->next_pic, s->last_pic);

  1444. 

  1445.     av_frame_unref(s->cur_pic->f);

  1446. 

  1447.     /* for skipping the frame */

  1448.     s->cur_pic->f->pict_type = s->pict_type;

  1449.     s->cur_pic->f->key_frame = (s->pict_type == AV_PICTURE_TYPE_I);

  1450. 

  1451.     ret = get_buffer(avctx, s->cur_pic);

  1452.     if (ret < 0)

  1453.         return ret;

  1454. 

  1455.     for (i = 0; i < 16; i++) {

  1456.         s->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * s->cur_pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);

  1457.         s->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * s->cur_pic->f->linesize[0] * ((scan8[i] - scan8[0]) >> 3);

  1458.     }

  1459.     for (i = 0; i < 16; i++) {

  1460.         s->block_offset[16 + i]      =

  1461.         s->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * s->cur_pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);

  1462.         s->block_offset[48 + 16 + i] =

  1463.         s->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * s->cur_pic->f->linesize[1] * ((scan8[i] - scan8[0]) >> 3);

  1464.     }

  1465. 

  1466.     if (s->pict_type != AV_PICTURE_TYPE_I) {

  1467.         if (!s->last_pic->f->data[0]) {

  1468.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1469.             av_frame_unref(s->last_pic->f);

  1470.             ret = get_buffer(avctx, s->last_pic);

  1471.             if (ret < 0)

  1472.                 return ret;

  1473.             memset(s->last_pic->f->data[0], 0, avctx->height * s->last_pic->f->linesize[0]);

  1474.             memset(s->last_pic->f->data[1], 0x80, (avctx->height / 2) *

  1475.                    s->last_pic->f->linesize[1]);

  1476.             memset(s->last_pic->f->data[2], 0x80, (avctx->height / 2) *

  1477.                    s->last_pic->f->linesize[2]);

  1478.         }

  1479. 

  1480.         if (s->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f->data[0]) {

  1481.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1482.             av_frame_unref(s->next_pic->f);

  1483.             ret = get_buffer(avctx, s->next_pic);

  1484.             if (ret < 0)

  1485.                 return ret;

  1486.             memset(s->next_pic->f->data[0], 0, avctx->height * s->next_pic->f->linesize[0]);

  1487.             memset(s->next_pic->f->data[1], 0x80, (avctx->height / 2) *

  1488.                    s->next_pic->f->linesize[1]);

  1489.             memset(s->next_pic->f->data[2], 0x80, (avctx->height / 2) *

  1490.                    s->next_pic->f->linesize[2]);

  1491.         }

  1492.     }

  1493. 

  1494.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1495.         av_log(s->avctx, AV_LOG_DEBUG,

  1496.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1497.                av_get_picture_type_char(s->pict_type),

  1498.                s->halfpel_flag, s->thirdpel_flag,

  1499.                s->adaptive_quant, s->qscale, s->slice_num);

  1500. 

  1501.     if (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B ||

  1502.         avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I ||

  1503.         avctx->skip_frame >= AVDISCARD_ALL)

  1504.         return 0;

  1505. 

  1506.     if (s->next_p_frame_damaged) {

  1507.         if (s->pict_type == AV_PICTURE_TYPE_B)

  1508.             return 0;

  1509.         else

  1510.             s->next_p_frame_damaged = 0;

  1511.     }

  1512. 

  1513.     if (s->pict_type == AV_PICTURE_TYPE_B) {

  1514.         s->frame_num_offset = s->slice_num - s->prev_frame_num;

  1515. 

  1516.         if (s->frame_num_offset < 0)

  1517.             s->frame_num_offset += 256;

  1518.         if (s->frame_num_offset == 0 ||

  1519.             s->frame_num_offset >= s->prev_frame_num_offset) {

  1520.             av_log(s->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1521.             return -1;

  1522.         }

  1523.     } else {

  1524.         s->prev_frame_num        = s->frame_num;

  1525.         s->frame_num             = s->slice_num;

  1526.         s->prev_frame_num_offset = s->frame_num - s->prev_frame_num;

  1527. 

  1528.         if (s->prev_frame_num_offset < 0)

  1529.             s->prev_frame_num_offset += 256;

  1530.     }

  1531. 

  1532.     for (m = 0; m < 2; m++) {

  1533.         int i;

  1534.         for (i = 0; i < 4; i++) {

  1535.             int j;

  1536.             for (j = -1; j < 4; j++)

  1537.                 s->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1538.             if (i < 3)

  1539.                 s->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1540.         }

  1541.     }

  1542. 

  1543.     for (s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {

  1544.         for (s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {

  1545.             unsigned mb_type;

  1546.             s->mb_xy = s->mb_x + s->mb_y * s->mb_stride;

  1547. 

  1548.             if ((get_bits_left(&s->gb_slice)) <= 7) {

  1549.                 if (((get_bits_count(&s->gb_slice) & 7) == 0 ||

  1550.                     show_bits(&s->gb_slice, get_bits_left(&s->gb_slice) & 7) == 0)) {

  1551. 

  1552.                     if (svq3_decode_slice_header(avctx))

  1553.                         return -1;

  1554.                 }

  1555.                 if (s->slice_type != s->pict_type) {

  1556.                     avpriv_request_sample(avctx, "non constant slice type");

  1557.                 }

  1558.                 /* TODO: support s->mb_skip_run */

  1559.             }

  1560. 

  1561.             mb_type = get_interleaved_ue_golomb(&s->gb_slice);

  1562. 

  1563.             if (s->pict_type == AV_PICTURE_TYPE_I)

  1564.                 mb_type += 8;

  1565.             else if (s->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1566.                 mb_type += 4;

  1567.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1568.                 av_log(s->avctx, AV_LOG_ERROR,

  1569.                        "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  1570.                 return -1;

  1571.             }

  1572. 

  1573.             if (mb_type != 0 || s->cbp)

  1574.                 hl_decode_mb(s);

  1575. 

  1576.             if (s->pict_type != AV_PICTURE_TYPE_B && !s->low_delay)

  1577.                 s->cur_pic->mb_type[s->mb_x + s->mb_y * s->mb_stride] =

  1578.                     (s->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1579.         }

  1580. 

  1581.         ff_draw_horiz_band(avctx, s->cur_pic->f,

  1582.                            s->last_pic->f->data[0] ? s->last_pic->f : NULL,

  1583.                            16 * s->mb_y, 16, PICT_FRAME, 0,

  1584.                            s->low_delay);

  1585.     }

  1586. 

  1587.     left = buf_size*8 - get_bits_count(&s->gb_slice);

  1588. 

  1589.     if (s->mb_y != s->mb_height || s->mb_x != s->mb_width) {

  1590.         av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, s->mb_y, s->mb_x, left);

  1591.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1592.     }

  1593. 

  1594.     if (left < 0) {

  1595.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1596.         return -1;

  1597.     }

  1598. 

  1599.     if (s->pict_type == AV_PICTURE_TYPE_B || s->low_delay)

  1600.         ret = av_frame_ref(data, s->cur_pic->f);

  1601.     else if (s->last_pic->f->data[0])

  1602.         ret = av_frame_ref(data, s->last_pic->f);

  1603.     if (ret < 0)

  1604.         return ret;

  1605. 

  1606.     /* Do not output the last pic after seeking. */

  1607.     if (s->last_pic->f->data[0] || s->low_delay)

  1608.         *got_frame = 1;

  1609. 

  1610.     if (s->pict_type != AV_PICTURE_TYPE_B) {

  1611.         FFSWAP(SVQ3Frame*, s->cur_pic, s->next_pic);

  1612.     } else {

  1613.         av_frame_unref(s->cur_pic->f);

  1614.     }

  1615. 

  1616.     return buf_size;

  1617. }

  1618. 

  1619. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1620. {

  1621.     SVQ3Context *s = avctx->priv_data;

  1622. 

  1623.     free_picture(avctx, s->cur_pic);

  1624.     free_picture(avctx, s->next_pic);

  1625.     free_picture(avctx, s->last_pic);

  1626.     av_frame_free(&s->cur_pic->f);

  1627.     av_frame_free(&s->next_pic->f);

  1628.     av_frame_free(&s->last_pic->f);

  1629.     av_freep(&s->cur_pic);

  1630.     av_freep(&s->next_pic);

  1631.     av_freep(&s->last_pic);

  1632.     av_freep(&s->slice_buf);

  1633.     av_freep(&s->intra4x4_pred_mode);

  1634.     av_freep(&s->edge_emu_buffer);

  1635.     av_freep(&s->mb2br_xy);

  1636. 

  1637. 

  1638.     av_freep(&s->buf);

  1639.     s->buf_size = 0;

  1640. 

  1641.     return 0;

  1642. }

  1643. 

  1644. AVCodec ff_svq3_decoder = {

  1645.     .name           = "svq3",

  1646.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1647.     .type           = AVMEDIA_TYPE_VIDEO,

  1648.     .id             = AV_CODEC_ID_SVQ3,

  1649.     .priv_data_size = sizeof(SVQ3Context),

  1650.     .init           = svq3_decode_init,

  1651.     .close          = svq3_decode_end,

  1652.     .decode         = svq3_decode_frame,

  1653.     .capabilities   = AV_CODEC_CAP_DRAW_HORIZ_BAND |

  1654.                       AV_CODEC_CAP_DR1             |

  1655.                       AV_CODEC_CAP_DELAY,

  1656.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1657.                                                      AV_PIX_FMT_NONE},

  1658. };