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.
79732 Commits
32 Branches
380MB
Tree: 5db920bcb6
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from '5db920bcb6'
${ noResults }
FFmpeg/libavcodec/svq3.c

1675 lines
59KB
Raw Blame History 

  1. /*

  2.  * Copyright (c) 2003 The FFmpeg Project

  3.  *

  4.  * This file is part of FFmpeg.

  5.  *

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

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

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

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

  10.  *

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

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

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

  14.  * Lesser General Public License for more details.

  15.  *

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

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

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

  19.  */

  20. 

  21. /*

  22.  * How to use this decoder:

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

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

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

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

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

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

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

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

  31.  *

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

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

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

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

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

  37.  *

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

  39.  * correctly decodes this file:

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

  41.  */

  42. 

  43. #include <inttypes.h>

  44. 

  45. #include "libavutil/attributes.h"

  46. #include "internal.h"

  47. #include "avcodec.h"

  48. #include "mpegutils.h"

  49. #include "h264.h"

  50. #include "h264_mvpred.h"

  51. #include "h264data.h"

  52. #include "golomb.h"

  53. #include "hpeldsp.h"

  54. #include "mathops.h"

  55. #include "rectangle.h"

  56. #include "tpeldsp.h"

  57. #include "vdpau_internal.h"

  58. 

  59. #if CONFIG_ZLIB

  60. #include <zlib.h>

  61. #endif

  62. 

  63. #include "svq1.h"

  64. 

  65. /**

  66.  * @file

  67.  * svq3 decoder.

  68.  */

  69. 

  70. typedef struct SVQ3Context {

  71.     H264Context h;

  72. 

  73.     H264DSPContext  h264dsp;

  74.     H264PredContext hpc;

  75.     HpelDSPContext hdsp;

  76.     TpelDSPContext tdsp;

  77.     VideoDSPContext vdsp;

  78. 

  79.     H264Picture *cur_pic;

  80.     H264Picture *next_pic;

  81.     H264Picture *last_pic;

  82.     GetBitContext gb;

  83.     uint8_t *slice_buf;

  84.     int slice_size;

  85.     int halfpel_flag;

  86.     int thirdpel_flag;

  87.     int has_watermark;

  88.     uint32_t watermark_key;

  89.     uint8_t *buf;

  90.     int buf_size;

  91.     int adaptive_quant;

  92.     int next_p_frame_damaged;

  93.     int h_edge_pos;

  94.     int v_edge_pos;

  95.     int last_frame_output;

  96. 

  97.     enum AVPictureType pict_type;

  98. 

  99.     int mb_x, mb_y;

  100.     int mb_xy;

  101.     int mb_width, mb_height;

  102.     int mb_stride, mb_num;

  103.     int b_stride;

  104. 

  105.     uint32_t *mb2br_xy;

  106. 

  107.     int chroma_pred_mode;

  108.     int intra16x16_pred_mode;

  109. 

  110.     int8_t   intra4x4_pred_mode_cache[5 * 8];

  111.     int8_t (*intra4x4_pred_mode);

  112. 

  113.     unsigned int top_samples_available;

  114.     unsigned int topright_samples_available;

  115.     unsigned int left_samples_available;

  116. 

  117.     uint8_t *edge_emu_buffer;

  118. 

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

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

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

  122. } SVQ3Context;

  123. 

  124. #define FULLPEL_MODE  1

  125. #define HALFPEL_MODE  2

  126. #define THIRDPEL_MODE 3

  127. #define PREDICT_MODE  4

  128. 

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

  130.  * o-->o-->o   o

  131.  *         |  /|

  132.  * o   o   o / o

  133.  * | / |   |/  |

  134.  * o   o   o   o

  135.  *   /

  136.  * o-->o-->o-->o

  137.  */

  138. static const uint8_t svq3_scan[16] = {

  139.     0 + 0 * 4, 1 + 0 * 4, 2 + 0 * 4, 2 + 1 * 4,

  140.     2 + 2 * 4, 3 + 0 * 4, 3 + 1 * 4, 3 + 2 * 4,

  141.     0 + 1 * 4, 0 + 2 * 4, 1 + 1 * 4, 1 + 2 * 4,

  142.     0 + 3 * 4, 1 + 3 * 4, 2 + 3 * 4, 3 + 3 * 4,

  143. };

  144. 

  145. static const uint8_t luma_dc_zigzag_scan[16] = {

  146.     0 * 16 + 0 * 64, 1 * 16 + 0 * 64, 2 * 16 + 0 * 64, 0 * 16 + 2 * 64,

  147.     3 * 16 + 0 * 64, 0 * 16 + 1 * 64, 1 * 16 + 1 * 64, 2 * 16 + 1 * 64,

  148.     1 * 16 + 2 * 64, 2 * 16 + 2 * 64, 3 * 16 + 2 * 64, 0 * 16 + 3 * 64,

  149.     3 * 16 + 1 * 64, 1 * 16 + 3 * 64, 2 * 16 + 3 * 64, 3 * 16 + 3 * 64,

  150. };

  151. 

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

  153.     { 0, 0 },

  154.     { 1, 0 }, { 0, 1 },

  155.     { 0, 2 }, { 1, 1 }, { 2, 0 },

  156.     { 3, 0 }, { 2, 1 }, { 1, 2 }, { 0, 3 },

  157.     { 0, 4 }, { 1, 3 }, { 2, 2 }, { 3, 1 }, { 4, 0 },

  158.     { 4, 1 }, { 3, 2 }, { 2, 3 }, { 1, 4 },

  159.     { 2, 4 }, { 3, 3 }, { 4, 2 },

  160.     { 4, 3 }, { 3, 4 },

  161.     { 4, 4 }

  162. };

  163. 

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

  165.     { { 2, -1, -1, -1, -1 }, { 2, 1, -1, -1, -1 }, { 1, 2, -1, -1, -1 },

  166.       { 2,  1, -1, -1, -1 }, { 1, 2, -1, -1, -1 }, { 1, 2, -1, -1, -1 } },

  167.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  4,  3 }, { 0, 1,  2,  4,  3 },

  168.       { 0,  2,  1,  4,  3 }, { 2, 0,  1,  3,  4 }, { 0, 4,  2,  1,  3 } },

  169.     { { 2,  0, -1, -1, -1 }, { 2, 1,  0,  4,  3 }, { 1, 2,  4,  0,  3 },

  170.       { 2,  1,  0,  4,  3 }, { 2, 1,  4,  3,  0 }, { 1, 2,  4,  0,  3 } },

  171.     { { 2,  0, -1, -1, -1 }, { 2, 0,  1,  4,  3 }, { 1, 2,  0,  4,  3 },

  172.       { 2,  1,  0,  4,  3 }, { 2, 1,  3,  4,  0 }, { 2, 4,  1,  0,  3 } },

  173.     { { 0,  2, -1, -1, -1 }, { 0, 2,  1,  3,  4 }, { 1, 2,  3,  0,  4 },

  174.       { 2,  0,  1,  3,  4 }, { 2, 1,  3,  0,  4 }, { 2, 0,  4,  3,  1 } },

  175.     { { 0,  2, -1, -1, -1 }, { 0, 2,  4,  1,  3 }, { 1, 4,  2,  0,  3 },

  176.       { 4,  2,  0,  1,  3 }, { 2, 0,  1,  4,  3 }, { 4, 2,  1,  0,  3 } },

  177. };

  178. 

  179. static const struct {

  180.     uint8_t run;

  181.     uint8_t level;

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

  183.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 2, 1 }, { 0, 2 }, { 3, 1 }, { 4, 1 }, { 5, 1 },

  184.       { 0, 3 }, { 1, 2 }, { 2, 2 }, { 6, 1 }, { 7, 1 }, { 8, 1 }, { 9, 1 }, { 0, 4 } },

  185.     { { 0, 0 }, { 0, 1 }, { 1, 1 }, { 0, 2 }, { 2, 1 }, { 0, 3 }, { 0, 4 }, { 0, 5 },

  186.       { 3, 1 }, { 4, 1 }, { 1, 2 }, { 1, 3 }, { 0, 6 }, { 0, 7 }, { 0, 8 }, { 0, 9 } }

  187. };

  188. 

  189. static const uint32_t svq3_dequant_coeff[32] = {

  190.      3881,  4351,  4890,  5481,   6154,   6914,   7761,   8718,

  191.      9781, 10987, 12339, 13828,  15523,  17435,  19561,  21873,

  192.     24552, 27656, 30847, 34870,  38807,  43747,  49103,  54683,

  193.     61694, 68745, 77615, 89113, 100253, 109366, 126635, 141533

  194. };

  195. 

  196. static int svq3_decode_end(AVCodecContext *avctx);

  197. 

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

  199. {

  200.     const int qmul = svq3_dequant_coeff[qp];

  201. #define stride 16

  202.     int i;

  203.     int temp[16];

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

  205. 

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

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

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

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

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

  211. 

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

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

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

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

  216.     }

  217. 

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

  219.         const int offset = x_offset[i];

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

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

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

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

  224. 

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

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

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

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

  229.     }

  230. }

  231. #undef stride

  232. 

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

  234.                             int stride, int qp, int dc)

  235. {

  236.     const int qmul = svq3_dequant_coeff[qp];

  237.     int i;

  238. 

  239.     if (dc) {

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

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

  242.         block[0] = 0;

  243.     }

  244. 

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

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

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

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

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

  250. 

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

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

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

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

  255.     }

  256. 

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

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

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

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

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

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

  263. 

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

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

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

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

  268.     }

  269. 

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

  271. }

  272. 

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

  274.                                     int index, const int type)

  275. {

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

  277.         luma_dc_zigzag_scan, ff_zigzag_scan, svq3_scan, ff_h264_chroma_dc_scan

  278.     };

  279. 

  280.     int run, level, sign, limit;

  281.     unsigned vlc;

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

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

  284. 

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

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

  287.             if ((int32_t)vlc < 0)

  288.                 return -1;

  289. 

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

  291.             vlc      = vlc + 1 >> 1;

  292. 

  293.             if (type == 3) {

  294.                 if (vlc < 3) {

  295.                     run   = 0;

  296.                     level = vlc;

  297.                 } else if (vlc < 4) {

  298.                     run   = 1;

  299.                     level = 1;

  300.                 } else {

  301.                     run   = vlc & 0x3;

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

  303.                 }

  304.             } else {

  305.                 if (vlc < 16U) {

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

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

  308.                 } else if (intra) {

  309.                     run   = vlc & 0x7;

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

  311.                 } else {

  312.                     run   = vlc & 0xF;

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

  314.                 }

  315.             }

  316. 

  317. 

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

  319.                 return -1;

  320. 

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

  322.         }

  323. 

  324.         if (type != 2) {

  325.             break;

  326.         }

  327.     }

  328. 

  329.     return 0;

  330. }

  331. 

  332. static av_always_inline int

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

  334.                        int i, int list, int part_width)

  335. {

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

  337. 

  338.     if (topright_ref != PART_NOT_AVAILABLE) {

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

  340.         return topright_ref;

  341.     } else {

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

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

  344.     }

  345. }

  346. 

  347. /**

  348.  * Get the predicted MV.

  349.  * @param n the block index

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

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

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

  353.  */

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

  355.                                               int part_width, int list,

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

  357. {

  358.     const int index8       = scan8[n];

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

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

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

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

  363.     const int16_t *C;

  364.     int diagonal_ref, match_count;

  365. 

  366. /* mv_cache

  367.  * B . . A T T T T

  368.  * U . . L . . , .

  369.  * U . . L . . . .

  370.  * U . . L . . , .

  371.  * . . . L . . . .

  372.  */

  373. 

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

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

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

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

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

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

  380.         if (left_ref == ref) {

  381.             *mx = A[0];

  382.             *my = A[1];

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

  384.             *mx = B[0];

  385.             *my = B[1];

  386.         } else {

  387.             *mx = C[0];

  388.             *my = C[1];

  389.         }

  390.     } else {

  391.         if (top_ref      == PART_NOT_AVAILABLE &&

  392.             diagonal_ref == PART_NOT_AVAILABLE &&

  393.             left_ref     != PART_NOT_AVAILABLE) {

  394.             *mx = A[0];

  395.             *my = A[1];

  396.         } else {

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

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

  399.         }

  400.     }

  401. }

  402. 

  403. static inline void svq3_mc_dir_part(SVQ3Context *s,

  404.                                     int x, int y, int width, int height,

  405.                                     int mx, int my, int dxy,

  406.                                     int thirdpel, int dir, int avg)

  407. {

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

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

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

  411.     uint8_t *src, *dest;

  412.     int i, emu = 0;

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

  414. 

  415.     mx += x;

  416.     my += y;

  417. 

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

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

  420.         emu = 1;

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

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

  423.     }

  424. 

  425.     /* form component predictions */

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

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

  428. 

  429.     if (emu) {

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

  431.                                  sl->linesize, sl->linesize,

  432.                                  width + 1, height + 1,

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

  434.         src = s->edge_emu_buffer;

  435.     }

  436.     if (thirdpel)

  437.         (avg ? s->tdsp.avg_tpel_pixels_tab

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

  439.                                                  width, height);

  440.     else

  441.         (avg ? s->hdsp.avg_pixels_tab

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

  443.                                                        height);

  444. 

  445.     if (!(h->flags & AV_CODEC_FLAG_GRAY)) {

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

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

  448.         width  = width  >> 1;

  449.         height = height >> 1;

  450.         blocksize++;

  451. 

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

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

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

  455. 

  456.             if (emu) {

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

  458.                                          sl->uvlinesize, sl->uvlinesize,

  459.                                          width + 1, height + 1,

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

  461.                                          s->v_edge_pos >> 1);

  462.                 src = s->edge_emu_buffer;

  463.             }

  464.             if (thirdpel)

  465.                 (avg ? s->tdsp.avg_tpel_pixels_tab

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

  467.                                                          sl->uvlinesize,

  468.                                                          width, height);

  469.             else

  470.                 (avg ? s->hdsp.avg_pixels_tab

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

  472.                                                                sl->uvlinesize,

  473.                                                                height);

  474.         }

  475.     }

  476. }

  477. 

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

  479.                               int dir, int avg)

  480. {

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

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

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

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

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

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

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

  488. 

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

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

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

  492.                              (4 * s->mb_y + (i >> 2)) * s->b_stride;

  493.             int dxy;

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

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

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

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

  498. 

  499.             if (mode != PREDICT_MODE) {

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

  501.             } else {

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

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

  504. 

  505.                 if (dir == 0) {

  506.                     mx = mx * h->frame_num_offset /

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

  508.                     my = my * h->frame_num_offset /

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

  510.                 } else {

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

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

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

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

  515.                 }

  516.             }

  517. 

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

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

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

  521. 

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

  523.             if (mode == PREDICT_MODE) {

  524.                 dx = dy = 0;

  525.             } else {

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

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

  528. 

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

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

  531.                     return -1;

  532.                 }

  533.             }

  534. 

  535.             /* compute motion vector */

  536.             if (mode == THIRDPEL_MODE) {

  537.                 int fx, fy;

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

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

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

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

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

  543. 

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

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

  546.                 mx += mx;

  547.                 my += my;

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

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

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

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

  552. 

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

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

  555.                 mx *= 3;

  556.                 my *= 3;

  557.             } else {

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

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

  560. 

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

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

  563.                 mx *= 6;

  564.                 my *= 6;

  565.             }

  566. 

  567.             /* update mv_cache */

  568.             if (mode != PREDICT_MODE) {

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

  570. 

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

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

  573. 

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

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

  576.                 }

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

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

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

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

  581.             }

  582. 

  583.             /* write back motion vectors */

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

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

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

  587.         }

  588. 

  589.     return 0;

  590. }

  591. 

  592. static av_always_inline void hl_decode_mb_idct_luma(const H264Context *h, H264SliceContext *sl,

  593.                                                     int mb_type, const int *block_offset,

  594.                                                     int linesize, uint8_t *dest_y)

  595. {

  596.     int i;

  597.     if (!IS_INTRA4x4(mb_type)) {

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

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

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

  601.                 svq3_add_idct_c(ptr, sl->mb + i * 16, linesize,

  602.                                 sl->qscale, IS_INTRA(mb_type) ? 1 : 0);

  603.             }

  604.     }

  605. }

  606. 

  607. static av_always_inline int dctcoef_get(int16_t *mb, int index)

  608. {

  609.     return AV_RN16A(mb + index);

  610. }

  611. 

  612. static av_always_inline void hl_decode_mb_predict_luma(SVQ3Context *s,

  613.                                                        const H264Context *h,

  614.                                                        H264SliceContext *sl,

  615.                                                        int mb_type,

  616.                                                        const int *block_offset,

  617.                                                        int linesize,

  618.                                                        uint8_t *dest_y)

  619. {

  620.     int i;

  621.     int qscale = sl->qscale;

  622. 

  623.     if (IS_INTRA4x4(mb_type)) {

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

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

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

  627. 

  628.             uint8_t *topright;

  629.             int nnz, tr;

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

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

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

  633.                 if (!topright_avail) {

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

  635.                     topright = (uint8_t *)&tr;

  636.                 } else

  637.                     topright = ptr + 4 - linesize;

  638.             } else

  639.                 topright = NULL;

  640. 

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

  642.             nnz = sl->non_zero_count_cache[scan8[i]];

  643.             if (nnz) {

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

  645.             }

  646.         }

  647.     } else {

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

  649.         svq3_luma_dc_dequant_idct_c(sl->mb, sl->mb_luma_dc[0], qscale);

  650.     }

  651. }

  652. 

  653. static void hl_decode_mb(SVQ3Context *s, const H264Context *h, H264SliceContext *sl)

  654. {

  655.     const int mb_x    = s->mb_x;

  656.     const int mb_y    = s->mb_y;

  657.     const int mb_xy   = s->mb_xy;

  658.     const int mb_type = h->cur_pic.mb_type[mb_xy];

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

  660.     int linesize, uvlinesize;

  661.     int i, j;

  662.     const int *block_offset = &h->block_offset[0];

  663.     const int block_h   = 16 >> h->chroma_y_shift;

  664. 

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

  666.     dest_cb = h->cur_pic.f->data[1] +  mb_x * 8 + mb_y * sl->uvlinesize * block_h;

  667.     dest_cr = h->cur_pic.f->data[2] +  mb_x * 8 + mb_y * sl->uvlinesize * block_h;

  668. 

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

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

  671. 

  672.     linesize   = sl->mb_linesize   = sl->linesize;

  673.     uvlinesize = sl->mb_uvlinesize = sl->uvlinesize;

  674. 

  675.     if (IS_INTRA(mb_type)) {

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

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

  678. 

  679.         hl_decode_mb_predict_luma(s, h, sl, mb_type, block_offset, linesize, dest_y);

  680.     }

  681. 

  682.     hl_decode_mb_idct_luma(h, sl, mb_type, block_offset, linesize, dest_y);

  683. 

  684.     if (sl->cbp & 0x30) {

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

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

  687.                                                s->dequant4_coeff[sl->chroma_qp[0]][0]);

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

  689.                                                s->dequant4_coeff[sl->chroma_qp[1]][0]);

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

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

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

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

  694.                     svq3_add_idct_c(ptr, sl->mb + i * 16,

  695.                                     uvlinesize, ff_h264_chroma_qp[0][sl->qscale + 12] - 12, 2);

  696.                 }

  697.         }

  698.     }

  699. }

  700. 

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

  702. {

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

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

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

  706.     int cbp = 0;

  707.     uint32_t vlc;

  708.     int8_t *top, *left;

  709.     const int mb_xy = s->mb_xy;

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

  711. 

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

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

  714.     s->topright_samples_available = 0xFFFF;

  715. 

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

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

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

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

  720.                              0, 0, 0, 0, 0, 0);

  721. 

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

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

  724.                                  0, 0, 0, 0, 1, 1);

  725. 

  726.             mb_type = MB_TYPE_SKIP;

  727.         } else {

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

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

  730.                 return -1;

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

  732.                 return -1;

  733. 

  734.             mb_type = MB_TYPE_16x16;

  735.         }

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

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

  738.             mode = THIRDPEL_MODE;

  739.         else if (s->halfpel_flag &&

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

  741.             mode = HALFPEL_MODE;

  742.         else

  743.             mode = FULLPEL_MODE;

  744. 

  745.         /* fill caches */

  746.         /* note ref_cache should contain here:

  747.          *  ????????

  748.          *  ???11111

  749.          *  N??11111

  750.          *  N??11111

  751.          *  N??11111

  752.          */

  753. 

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

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

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

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

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

  759.             } else {

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

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

  762.             }

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

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

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

  766.                        4 * 2 * sizeof(int16_t));

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

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

  769. 

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

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

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

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

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

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

  776.                 } else

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

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

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

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

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

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

  783.                 } else

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

  785.             } else

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

  787.                        PART_NOT_AVAILABLE, 8);

  788. 

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

  790.                 break;

  791.         }

  792. 

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

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

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

  796.                 return -1;

  797.         } else {        /* AV_PICTURE_TYPE_B */

  798.             if (mb_type != 2) {

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

  800.                     return -1;

  801.             } else {

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

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

  804.                            0, 4 * 2 * sizeof(int16_t));

  805.             }

  806.             if (mb_type != 1) {

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

  808.                     return -1;

  809.             } else {

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

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

  812.                            0, 4 * 2 * sizeof(int16_t));

  813.             }

  814.         }

  815. 

  816.         mb_type = MB_TYPE_16x16;

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

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

  819.         int8_t *i4x4_cache = s->intra4x4_pred_mode_cache;

  820. 

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

  822. 

  823.         if (mb_type == 8) {

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

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

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

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

  828.                     s->left_samples_available = 0x5F5F;

  829.             }

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

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

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

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

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

  835. 

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

  837.                     s->top_samples_available = 0x33FF;

  838.             }

  839. 

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

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

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

  843. 

  844.                 if (vlc >= 25U) {

  845.                     av_log(h->avctx, AV_LOG_ERROR,

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

  847.                     return -1;

  848.                 }

  849. 

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

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

  852. 

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

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

  855. 

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

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

  858.                     return -1;

  859.                 }

  860.             }

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

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

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

  864.         }

  865. 

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

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

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

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

  870. 

  871.         if (mb_type == 8) {

  872.             ff_h264_check_intra4x4_pred_mode(s->intra4x4_pred_mode_cache,

  873.                                              h->avctx, s->top_samples_available,

  874.                                              s->left_samples_available);

  875. 

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

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

  878.         } else {

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

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

  881. 

  882.             s->top_samples_available  = 0x33FF;

  883.             s->left_samples_available = 0x5F5F;

  884.         }

  885. 

  886.         mb_type = MB_TYPE_INTRA4x4;

  887.     } else {                      /* INTRA16x16 */

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

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

  890. 

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

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

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

  894.             return s->intra16x16_pred_mode;

  895.         }

  896. 

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

  898.         mb_type = MB_TYPE_INTRA16x16;

  899.     }

  900. 

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

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

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

  904.                    0, 4 * 2 * sizeof(int16_t));

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

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

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

  908.                        0, 4 * 2 * sizeof(int16_t));

  909.         }

  910.     }

  911.     if (!IS_INTRA4x4(mb_type)) {

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

  913.     }

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

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

  916.     }

  917. 

  918.     if (!IS_INTRA16x16(mb_type) &&

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

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

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

  922.             return -1;

  923.         }

  924. 

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

  926.                                 : ff_h264_golomb_to_inter_cbp[vlc];

  927.     }

  928.     if (IS_INTRA16x16(mb_type) ||

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

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

  931. 

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

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

  934.             return -1;

  935.         }

  936.     }

  937.     if (IS_INTRA16x16(mb_type)) {

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

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

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

  941.             av_log(h->avctx, AV_LOG_ERROR,

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

  943.             return -1;

  944.         }

  945.     }

  946. 

  947.     if (cbp) {

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

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

  950. 

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

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

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

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

  955.                                  2 * (j & 2) + 4 * (i & 2))

  956.                               : (4 * i + j);

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

  958. 

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

  960.                         av_log(h->avctx, AV_LOG_ERROR,

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

  962.                         return -1;

  963.                     }

  964.                 }

  965.             }

  966. 

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

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

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

  970.                     av_log(h->avctx, AV_LOG_ERROR,

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

  972.                     return -1;

  973.                 }

  974. 

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

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

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

  978.                         k                                 = 16 * i + j;

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

  980. 

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

  982.                             av_log(h->avctx, AV_LOG_ERROR,

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

  984.                             return -1;

  985.                         }

  986.                     }

  987.                 }

  988.             }

  989.         }

  990.     }

  991. 

  992.     sl->cbp                   = cbp;

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

  994. 

  995.     if (IS_INTRA(mb_type))

  996.         s->chroma_pred_mode = ff_h264_check_intra_pred_mode(h->avctx, s->top_samples_available,

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

  998. 

  999.     return 0;

  1000. }

  1001. 

  1002. static int svq3_decode_slice_header(AVCodecContext *avctx)

  1003. {

  1004.     SVQ3Context *s = avctx->priv_data;

  1005.     H264Context *h    = &s->h;

  1006.     H264SliceContext *sl = &h->slice_ctx[0];

  1007.     const int mb_xy   = s->mb_xy;

  1008.     int i, header;

  1009.     unsigned slice_id;

  1010. 

  1011.     header = get_bits(&s->gb, 8);

  1012. 

  1013.     if (((header & 0x9F) != 1 && (header & 0x9F) != 2) || (header & 0x60) == 0) {

  1014.         /* TODO: what? */

  1015.         av_log(avctx, AV_LOG_ERROR, "unsupported slice header (%02X)\n", header);

  1016.         return -1;

  1017.     } else {

  1018.         int slice_bits, slice_bytes, slice_length;

  1019.         int length = header >> 5 & 3;

  1020. 

  1021.         slice_length = show_bits(&s->gb, 8 * length);

  1022.         slice_bits   = slice_length * 8;

  1023.         slice_bytes  = slice_length + length - 1;

  1024. 

  1025.         if (slice_bytes > get_bits_left(&s->gb)) {

  1026.             av_log(avctx, AV_LOG_ERROR, "slice after bitstream end\n");

  1027.             return -1;

  1028.         }

  1029. 

  1030.         skip_bits(&s->gb, 8);

  1031. 

  1032.         av_fast_malloc(&s->slice_buf, &s->slice_size, slice_bytes + AV_INPUT_BUFFER_PADDING_SIZE);

  1033.         if (!s->slice_buf)

  1034.             return AVERROR(ENOMEM);

  1035. 

  1036.         memcpy(s->slice_buf, s->gb.buffer + s->gb.index / 8, slice_bytes);

  1037. 

  1038.         init_get_bits(&h->gb, s->slice_buf, slice_bits);

  1039. 

  1040.         if (s->watermark_key) {

  1041.             uint32_t header = AV_RL32(&h->gb.buffer[1]);

  1042.             AV_WL32(&h->gb.buffer[1], header ^ s->watermark_key);

  1043.         }

  1044.         if (length > 0) {

  1045.             memmove(s->slice_buf, &s->slice_buf[slice_length], length - 1);

  1046.         }

  1047.         skip_bits_long(&s->gb, slice_bytes * 8);

  1048.     }

  1049. 

  1050.     if ((slice_id = svq3_get_ue_golomb(&h->gb)) >= 3) {

  1051.         av_log(h->avctx, AV_LOG_ERROR, "illegal slice type %u \n", slice_id);

  1052.         return -1;

  1053.     }

  1054. 

  1055.     s->pict_type = ff_h264_golomb_to_pict_type[slice_id];

  1056. 

  1057.     if ((header & 0x9F) == 2) {

  1058.         i              = (s->mb_num < 64) ? 6 : (1 + av_log2(s->mb_num - 1));

  1059.         sl->mb_skip_run = get_bits(&h->gb, i) -

  1060.                          (s->mb_y * s->mb_width + s->mb_x);

  1061.     } else {

  1062.         skip_bits1(&h->gb);

  1063.         sl->mb_skip_run = 0;

  1064.     }

  1065. 

  1066.     sl->slice_num     = get_bits(&h->gb, 8);

  1067.     sl->qscale        = get_bits(&h->gb, 5);

  1068.     s->adaptive_quant = get_bits1(&h->gb);

  1069. 

  1070.     /* unknown fields */

  1071.     skip_bits1(&h->gb);

  1072. 

  1073.     if (s->has_watermark)

  1074.         skip_bits1(&h->gb);

  1075. 

  1076.     skip_bits1(&h->gb);

  1077.     skip_bits(&h->gb, 2);

  1078. 

  1079.     if (skip_1stop_8data_bits(&h->gb) < 0)

  1080.         return AVERROR_INVALIDDATA;

  1081. 

  1082.     /* reset intra predictors and invalidate motion vector references */

  1083.     if (s->mb_x > 0) {

  1084.         memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - 1] + 3,

  1085.                -1, 4 * sizeof(int8_t));

  1086.         memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - s->mb_x],

  1087.                -1, 8 * sizeof(int8_t) * s->mb_x);

  1088.     }

  1089.     if (s->mb_y > 0) {

  1090.         memset(s->intra4x4_pred_mode + s->mb2br_xy[mb_xy - s->mb_stride],

  1091.                -1, 8 * sizeof(int8_t) * (s->mb_width - s->mb_x));

  1092. 

  1093.         if (s->mb_x > 0)

  1094.             s->intra4x4_pred_mode[s->mb2br_xy[mb_xy - s->mb_stride - 1] + 3] = -1;

  1095.     }

  1096. 

  1097.     return 0;

  1098. }

  1099. 

  1100. static void init_dequant4_coeff_table(SVQ3Context *s)

  1101. {

  1102.     int q, x;

  1103.     const int max_qp = 51;

  1104. 

  1105.     for (q = 0; q < max_qp + 1; q++) {

  1106.         int shift = ff_h264_quant_div6[q] + 2;

  1107.         int idx   = ff_h264_quant_rem6[q];

  1108.         for (x = 0; x < 16; x++)

  1109.             s->dequant4_coeff[q][(x >> 2) | ((x << 2) & 0xF)] =

  1110.                 ((uint32_t)ff_h264_dequant4_coeff_init[idx][(x & 1) + ((x >> 2) & 1)] * 16) << shift;

  1111.     }

  1112. }

  1113. 

  1114. static av_cold int svq3_decode_init(AVCodecContext *avctx)

  1115. {

  1116.     SVQ3Context *s = avctx->priv_data;

  1117.     H264Context *h = &s->h;

  1118.     H264SliceContext *sl;

  1119.     int m, x, y;

  1120.     unsigned char *extradata;

  1121.     unsigned char *extradata_end;

  1122.     unsigned int size;

  1123.     int marker_found = 0;

  1124.     int ret;

  1125. 

  1126.     s->cur_pic  = av_mallocz(sizeof(*s->cur_pic));

  1127.     s->last_pic = av_mallocz(sizeof(*s->last_pic));

  1128.     s->next_pic = av_mallocz(sizeof(*s->next_pic));

  1129.     if (!s->next_pic || !s->last_pic || !s->cur_pic) {

  1130.         ret = AVERROR(ENOMEM);

  1131.         goto fail;

  1132.     }

  1133. 

  1134.     s->cur_pic->f  = av_frame_alloc();

  1135.     s->last_pic->f = av_frame_alloc();

  1136.     s->next_pic->f = av_frame_alloc();

  1137.     if (!s->cur_pic->f || !s->last_pic->f || !s->next_pic->f)

  1138.         return AVERROR(ENOMEM);

  1139. 

  1140.     if ((ret = ff_h264_decode_init(avctx)) < 0)

  1141.         goto fail;

  1142. 

  1143.     // we will overwrite it later during decoding

  1144.     av_frame_free(&h->cur_pic.f);

  1145. 

  1146.     av_frame_free(&h->last_pic_for_ec.f);

  1147. 

  1148.     ff_h264dsp_init(&s->h264dsp, 8, 1);

  1149.     av_assert0(h->sps.bit_depth_chroma == 0);

  1150.     ff_h264_pred_init(&s->hpc, AV_CODEC_ID_SVQ3, 8, 1);

  1151.     ff_videodsp_init(&s->vdsp, 8);

  1152. 

  1153. 

  1154.     avctx->bits_per_raw_sample = 8;

  1155.     h->sps.bit_depth_luma = 8;

  1156.     h->chroma_format_idc = 1;

  1157. 

  1158.     ff_hpeldsp_init(&s->hdsp, avctx->flags);

  1159.     ff_tpeldsp_init(&s->tdsp);

  1160. 

  1161.     sl = h->slice_ctx;

  1162. 

  1163.     h->flags           = avctx->flags;

  1164.     sl->is_complex     = 1;

  1165.     h->sps.chroma_format_idc = 1;

  1166.     h->picture_structure = PICT_FRAME;

  1167.     avctx->pix_fmt     = AV_PIX_FMT_YUVJ420P;

  1168.     avctx->color_range = AVCOL_RANGE_JPEG;

  1169. 

  1170.     h->slice_ctx[0].chroma_qp[0] = h->slice_ctx[0].chroma_qp[1] = 4;

  1171.     h->chroma_x_shift = h->chroma_y_shift = 1;

  1172. 

  1173.     s->halfpel_flag  = 1;

  1174.     s->thirdpel_flag = 1;

  1175.     s->has_watermark = 0;

  1176. 

  1177.     /* prowl for the "SEQH" marker in the extradata */

  1178.     extradata     = (unsigned char *)avctx->extradata;

  1179.     extradata_end = avctx->extradata + avctx->extradata_size;

  1180.     if (extradata) {

  1181.         for (m = 0; m + 8 < avctx->extradata_size; m++) {

  1182.             if (!memcmp(extradata, "SEQH", 4)) {

  1183.                 marker_found = 1;

  1184.                 break;

  1185.             }

  1186.             extradata++;

  1187.         }

  1188.     }

  1189. 

  1190.     /* if a match was found, parse the extra data */

  1191.     if (marker_found) {

  1192.         GetBitContext gb;

  1193.         int frame_size_code;

  1194.         int unk0, unk1, unk2, unk3, unk4;

  1195. 

  1196.         size = AV_RB32(&extradata[4]);

  1197.         if (size > extradata_end - extradata - 8) {

  1198.             ret = AVERROR_INVALIDDATA;

  1199.             goto fail;

  1200.         }

  1201.         init_get_bits(&gb, extradata + 8, size * 8);

  1202. 

  1203.         /* 'frame size code' and optional 'width, height' */

  1204.         frame_size_code = get_bits(&gb, 3);

  1205.         switch (frame_size_code) {

  1206.         case 0:

  1207.             avctx->width  = 160;

  1208.             avctx->height = 120;

  1209.             break;

  1210.         case 1:

  1211.             avctx->width  = 128;

  1212.             avctx->height =  96;

  1213.             break;

  1214.         case 2:

  1215.             avctx->width  = 176;

  1216.             avctx->height = 144;

  1217.             break;

  1218.         case 3:

  1219.             avctx->width  = 352;

  1220.             avctx->height = 288;

  1221.             break;

  1222.         case 4:

  1223.             avctx->width  = 704;

  1224.             avctx->height = 576;

  1225.             break;

  1226.         case 5:

  1227.             avctx->width  = 240;

  1228.             avctx->height = 180;

  1229.             break;

  1230.         case 6:

  1231.             avctx->width  = 320;

  1232.             avctx->height = 240;

  1233.             break;

  1234.         case 7:

  1235.             avctx->width  = get_bits(&gb, 12);

  1236.             avctx->height = get_bits(&gb, 12);

  1237.             break;

  1238.         }

  1239. 

  1240.         s->halfpel_flag  = get_bits1(&gb);

  1241.         s->thirdpel_flag = get_bits1(&gb);

  1242. 

  1243.         /* unknown fields */

  1244.         unk0 = get_bits1(&gb);

  1245.         unk1 = get_bits1(&gb);

  1246.         unk2 = get_bits1(&gb);

  1247.         unk3 = get_bits1(&gb);

  1248. 

  1249.         h->low_delay = get_bits1(&gb);

  1250. 

  1251.         /* unknown field */

  1252.         unk4 = get_bits1(&gb);

  1253. 

  1254.         av_log(avctx, AV_LOG_DEBUG, "Unknown fields %d %d %d %d %d\n",

  1255.                unk0, unk1, unk2, unk3, unk4);

  1256. 

  1257.         if (skip_1stop_8data_bits(&gb) < 0) {

  1258.             ret = AVERROR_INVALIDDATA;

  1259.             goto fail;

  1260.         }

  1261. 

  1262.         s->has_watermark  = get_bits1(&gb);

  1263.         avctx->has_b_frames = !h->low_delay;

  1264.         if (s->has_watermark) {

  1265. #if CONFIG_ZLIB

  1266.             unsigned watermark_width  = svq3_get_ue_golomb(&gb);

  1267.             unsigned watermark_height = svq3_get_ue_golomb(&gb);

  1268.             int u1                    = svq3_get_ue_golomb(&gb);

  1269.             int u2                    = get_bits(&gb, 8);

  1270.             int u3                    = get_bits(&gb, 2);

  1271.             int u4                    = svq3_get_ue_golomb(&gb);

  1272.             unsigned long buf_len     = watermark_width *

  1273.                                         watermark_height * 4;

  1274.             int offset                = get_bits_count(&gb) + 7 >> 3;

  1275.             uint8_t *buf;

  1276. 

  1277.             if (watermark_height <= 0 ||

  1278.                 (uint64_t)watermark_width * 4 > UINT_MAX / watermark_height) {

  1279.                 ret = -1;

  1280.                 goto fail;

  1281.             }

  1282. 

  1283.             buf = av_malloc(buf_len);

  1284.             if (!buf) {

  1285.                 ret = AVERROR(ENOMEM);

  1286.                 goto fail;

  1287.             }

  1288.             av_log(avctx, AV_LOG_DEBUG, "watermark size: %ux%u\n",

  1289.                    watermark_width, watermark_height);

  1290.             av_log(avctx, AV_LOG_DEBUG,

  1291.                    "u1: %x u2: %x u3: %x compressed data size: %d offset: %d\n",

  1292.                    u1, u2, u3, u4, offset);

  1293.             if (uncompress(buf, &buf_len, extradata + 8 + offset,

  1294.                            size - offset) != Z_OK) {

  1295.                 av_log(avctx, AV_LOG_ERROR,

  1296.                        "could not uncompress watermark logo\n");

  1297.                 av_free(buf);

  1298.                 ret = -1;

  1299.                 goto fail;

  1300.             }

  1301.             s->watermark_key = ff_svq1_packet_checksum(buf, buf_len, 0);

  1302.             s->watermark_key = s->watermark_key << 16 | s->watermark_key;

  1303.             av_log(avctx, AV_LOG_DEBUG,

  1304.                    "watermark key %#"PRIx32"\n", s->watermark_key);

  1305.             av_free(buf);

  1306. #else

  1307.             av_log(avctx, AV_LOG_ERROR,

  1308.                    "this svq3 file contains watermark which need zlib support compiled in\n");

  1309.             ret = -1;

  1310.             goto fail;

  1311. #endif

  1312.         }

  1313.     }

  1314. 

  1315.     s->mb_width   = (avctx->width + 15) / 16;

  1316.     s->mb_height  = (avctx->height + 15) / 16;

  1317.     s->mb_stride  = s->mb_width + 1;

  1318.     s->mb_num     = s->mb_width * s->mb_height;

  1319.     s->b_stride   = 4 * s->mb_width;

  1320.     s->h_edge_pos = s->mb_width * 16;

  1321.     s->v_edge_pos = s->mb_height * 16;

  1322. 

  1323.     s->intra4x4_pred_mode = av_mallocz(s->mb_stride * 2 * 8);

  1324.     if (!s->intra4x4_pred_mode)

  1325.         return AVERROR(ENOMEM);

  1326. 

  1327.     s->mb2br_xy = av_mallocz(s->mb_stride * (s->mb_height + 1) *

  1328.                              sizeof(*s->mb2br_xy));

  1329.     if (!s->mb2br_xy)

  1330.         return AVERROR(ENOMEM);

  1331. 

  1332.     for (y = 0; y < s->mb_height; y++)

  1333.         for (x = 0; x < s->mb_width; x++) {

  1334.             const int mb_xy = x + y * s->mb_stride;

  1335. 

  1336.             s->mb2br_xy[mb_xy] = 8 * (mb_xy % (2 * s->mb_stride));

  1337.         }

  1338. 

  1339.     init_dequant4_coeff_table(s);

  1340. 

  1341.     return 0;

  1342. fail:

  1343.     svq3_decode_end(avctx);

  1344.     return ret;

  1345. }

  1346. 

  1347. static void free_picture(AVCodecContext *avctx, H264Picture *pic)

  1348. {

  1349.     int i;

  1350.     for (i = 0; i < 2; i++) {

  1351.         av_buffer_unref(&pic->motion_val_buf[i]);

  1352.         av_buffer_unref(&pic->ref_index_buf[i]);

  1353.     }

  1354.     av_buffer_unref(&pic->mb_type_buf);

  1355. 

  1356.     av_frame_unref(pic->f);

  1357. }

  1358. 

  1359. static int get_buffer(AVCodecContext *avctx, H264Picture *pic)

  1360. {

  1361.     SVQ3Context *s = avctx->priv_data;

  1362.     H264Context *h = &s->h;

  1363.     H264SliceContext *sl = &h->slice_ctx[0];

  1364.     const int big_mb_num    = s->mb_stride * (s->mb_height + 1) + 1;

  1365.     const int mb_array_size = s->mb_stride * s->mb_height;

  1366.     const int b4_stride     = s->mb_width * 4 + 1;

  1367.     const int b4_array_size = b4_stride * s->mb_height * 4;

  1368.     int ret;

  1369. 

  1370.     if (!pic->motion_val_buf[0]) {

  1371.         int i;

  1372. 

  1373.         pic->mb_type_buf = av_buffer_allocz((big_mb_num + s->mb_stride) * sizeof(uint32_t));

  1374.         if (!pic->mb_type_buf)

  1375.             return AVERROR(ENOMEM);

  1376.         pic->mb_type = (uint32_t*)pic->mb_type_buf->data + 2 * s->mb_stride + 1;

  1377. 

  1378.         for (i = 0; i < 2; i++) {

  1379.             pic->motion_val_buf[i] = av_buffer_allocz(2 * (b4_array_size + 4) * sizeof(int16_t));

  1380.             pic->ref_index_buf[i]  = av_buffer_allocz(4 * mb_array_size);

  1381.             if (!pic->motion_val_buf[i] || !pic->ref_index_buf[i]) {

  1382.                 ret = AVERROR(ENOMEM);

  1383.                 goto fail;

  1384.             }

  1385. 

  1386.             pic->motion_val[i] = (int16_t (*)[2])pic->motion_val_buf[i]->data + 4;

  1387.             pic->ref_index[i]  = pic->ref_index_buf[i]->data;

  1388.         }

  1389.     }

  1390.     pic->reference = !(s->pict_type == AV_PICTURE_TYPE_B);

  1391. 

  1392.     ret = ff_get_buffer(avctx, pic->f,

  1393.                         pic->reference ? AV_GET_BUFFER_FLAG_REF : 0);

  1394.     if (ret < 0)

  1395.         goto fail;

  1396. 

  1397.     if (!s->edge_emu_buffer) {

  1398.         s->edge_emu_buffer = av_mallocz_array(pic->f->linesize[0], 17);

  1399.         if (!s->edge_emu_buffer)

  1400.             return AVERROR(ENOMEM);

  1401.     }

  1402. 

  1403.     sl->linesize   = pic->f->linesize[0];

  1404.     sl->uvlinesize = pic->f->linesize[1];

  1405. 

  1406.     return 0;

  1407. fail:

  1408.     free_picture(avctx, pic);

  1409.     return ret;

  1410. }

  1411. 

  1412. static int svq3_decode_frame(AVCodecContext *avctx, void *data,

  1413.                              int *got_frame, AVPacket *avpkt)

  1414. {

  1415.     SVQ3Context *s     = avctx->priv_data;

  1416.     H264Context *h     = &s->h;

  1417.     H264SliceContext *sl = &h->slice_ctx[0];

  1418.     int buf_size       = avpkt->size;

  1419.     int left;

  1420.     uint8_t *buf;

  1421.     int ret, m, i;

  1422. 

  1423.     /* special case for last picture */

  1424.     if (buf_size == 0) {

  1425.         if (s->next_pic->f->data[0] && !h->low_delay && !s->last_frame_output) {

  1426.             ret = av_frame_ref(data, s->next_pic->f);

  1427.             if (ret < 0)

  1428.                 return ret;

  1429.             s->last_frame_output = 1;

  1430.             *got_frame          = 1;

  1431.         }

  1432.         return 0;

  1433.     }

  1434. 

  1435.     sl->mb_x = sl->mb_y = sl->mb_xy = 0;

  1436. 

  1437.     if (s->watermark_key) {

  1438.         av_fast_padded_malloc(&s->buf, &s->buf_size, buf_size);

  1439.         if (!s->buf)

  1440.             return AVERROR(ENOMEM);

  1441.         memcpy(s->buf, avpkt->data, buf_size);

  1442.         buf = s->buf;

  1443.     } else {

  1444.         buf = avpkt->data;

  1445.     }

  1446. 

  1447.     ret = init_get_bits(&s->gb, buf, 8 * buf_size);

  1448.     if (ret < 0)

  1449.         return ret;

  1450. 

  1451.     if (svq3_decode_slice_header(avctx))

  1452.         return -1;

  1453. 

  1454.     if (s->pict_type != AV_PICTURE_TYPE_B)

  1455.         FFSWAP(H264Picture*, s->next_pic, s->last_pic);

  1456. 

  1457.     av_frame_unref(s->cur_pic->f);

  1458. 

  1459.     /* for skipping the frame */

  1460.     s->cur_pic->f->pict_type = s->pict_type;

  1461.     s->cur_pic->f->key_frame = (s->pict_type == AV_PICTURE_TYPE_I);

  1462. 

  1463.     ret = get_buffer(avctx, s->cur_pic);

  1464.     if (ret < 0)

  1465.         return ret;

  1466. 

  1467.     h->cur_pic_ptr = s->cur_pic;

  1468.     h->cur_pic     = *s->cur_pic;

  1469. 

  1470.     for (i = 0; i < 16; i++) {

  1471.         h->block_offset[i]           = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);

  1472.         h->block_offset[48 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->linesize * ((scan8[i] - scan8[0]) >> 3);

  1473.     }

  1474.     for (i = 0; i < 16; i++) {

  1475.         h->block_offset[16 + i]      =

  1476.         h->block_offset[32 + i]      = (4 * ((scan8[i] - scan8[0]) & 7)) + 4 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1477.         h->block_offset[48 + 16 + i] =

  1478.         h->block_offset[48 + 32 + i] = (4 * ((scan8[i] - scan8[0]) & 7)) + 8 * sl->uvlinesize * ((scan8[i] - scan8[0]) >> 3);

  1479.     }

  1480. 

  1481.     if (s->pict_type != AV_PICTURE_TYPE_I) {

  1482.         if (!s->last_pic->f->data[0]) {

  1483.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1484.             av_frame_unref(s->last_pic->f);

  1485.             ret = get_buffer(avctx, s->last_pic);

  1486.             if (ret < 0)

  1487.                 return ret;

  1488.             memset(s->last_pic->f->data[0], 0, avctx->height * s->last_pic->f->linesize[0]);

  1489.             memset(s->last_pic->f->data[1], 0x80, (avctx->height / 2) *

  1490.                    s->last_pic->f->linesize[1]);

  1491.             memset(s->last_pic->f->data[2], 0x80, (avctx->height / 2) *

  1492.                    s->last_pic->f->linesize[2]);

  1493.         }

  1494. 

  1495.         if (s->pict_type == AV_PICTURE_TYPE_B && !s->next_pic->f->data[0]) {

  1496.             av_log(avctx, AV_LOG_ERROR, "Missing reference frame.\n");

  1497.             av_frame_unref(s->next_pic->f);

  1498.             ret = get_buffer(avctx, s->next_pic);

  1499.             if (ret < 0)

  1500.                 return ret;

  1501.             memset(s->next_pic->f->data[0], 0, avctx->height * s->next_pic->f->linesize[0]);

  1502.             memset(s->next_pic->f->data[1], 0x80, (avctx->height / 2) *

  1503.                    s->next_pic->f->linesize[1]);

  1504.             memset(s->next_pic->f->data[2], 0x80, (avctx->height / 2) *

  1505.                    s->next_pic->f->linesize[2]);

  1506.         }

  1507.     }

  1508. 

  1509.     if (avctx->debug & FF_DEBUG_PICT_INFO)

  1510.         av_log(h->avctx, AV_LOG_DEBUG,

  1511.                "%c hpel:%d, tpel:%d aqp:%d qp:%d, slice_num:%02X\n",

  1512.                av_get_picture_type_char(s->pict_type),

  1513.                s->halfpel_flag, s->thirdpel_flag,

  1514.                s->adaptive_quant, h->slice_ctx[0].qscale, sl->slice_num);

  1515. 

  1516.     if (avctx->skip_frame >= AVDISCARD_NONREF && s->pict_type == AV_PICTURE_TYPE_B ||

  1517.         avctx->skip_frame >= AVDISCARD_NONKEY && s->pict_type != AV_PICTURE_TYPE_I ||

  1518.         avctx->skip_frame >= AVDISCARD_ALL)

  1519.         return 0;

  1520. 

  1521.     if (s->next_p_frame_damaged) {

  1522.         if (s->pict_type == AV_PICTURE_TYPE_B)

  1523.             return 0;

  1524.         else

  1525.             s->next_p_frame_damaged = 0;

  1526.     }

  1527. 

  1528.     if (s->pict_type == AV_PICTURE_TYPE_B) {

  1529.         h->frame_num_offset = sl->slice_num - h->prev_frame_num;

  1530. 

  1531.         if (h->frame_num_offset < 0)

  1532.             h->frame_num_offset += 256;

  1533.         if (h->frame_num_offset == 0 ||

  1534.             h->frame_num_offset >= h->prev_frame_num_offset) {

  1535.             av_log(h->avctx, AV_LOG_ERROR, "error in B-frame picture id\n");

  1536.             return -1;

  1537.         }

  1538.     } else {

  1539.         h->prev_frame_num        = h->frame_num;

  1540.         h->frame_num             = sl->slice_num;

  1541.         h->prev_frame_num_offset = h->frame_num - h->prev_frame_num;

  1542. 

  1543.         if (h->prev_frame_num_offset < 0)

  1544.             h->prev_frame_num_offset += 256;

  1545.     }

  1546. 

  1547.     for (m = 0; m < 2; m++) {

  1548.         int i;

  1549.         for (i = 0; i < 4; i++) {

  1550.             int j;

  1551.             for (j = -1; j < 4; j++)

  1552.                 s->ref_cache[m][scan8[0] + 8 * i + j] = 1;

  1553.             if (i < 3)

  1554.                 s->ref_cache[m][scan8[0] + 8 * i + j] = PART_NOT_AVAILABLE;

  1555.         }

  1556.     }

  1557. 

  1558.     for (s->mb_y = 0; s->mb_y < s->mb_height; s->mb_y++) {

  1559.         for (s->mb_x = 0; s->mb_x < s->mb_width; s->mb_x++) {

  1560.             unsigned mb_type;

  1561.             s->mb_xy = s->mb_x + s->mb_y * s->mb_stride;

  1562. 

  1563.             if ((get_bits_left(&h->gb)) <= 7) {

  1564.                 if (((get_bits_count(&h->gb) & 7) == 0 ||

  1565.                     show_bits(&h->gb, get_bits_left(&h->gb) & 7) == 0)) {

  1566. 

  1567.                     if (svq3_decode_slice_header(avctx))

  1568.                         return -1;

  1569.                 }

  1570.                 /* TODO: support s->mb_skip_run */

  1571.             }

  1572. 

  1573.             mb_type = svq3_get_ue_golomb(&h->gb);

  1574. 

  1575.             if (s->pict_type == AV_PICTURE_TYPE_I)

  1576.                 mb_type += 8;

  1577.             else if (s->pict_type == AV_PICTURE_TYPE_B && mb_type >= 4)

  1578.                 mb_type += 4;

  1579.             if (mb_type > 33 || svq3_decode_mb(s, mb_type)) {

  1580.                 av_log(h->avctx, AV_LOG_ERROR,

  1581.                        "error while decoding MB %d %d\n", s->mb_x, s->mb_y);

  1582.                 return -1;

  1583.             }

  1584. 

  1585.             if (mb_type != 0 || sl->cbp)

  1586.                 hl_decode_mb(s, h, &h->slice_ctx[0]);

  1587. 

  1588.             if (s->pict_type != AV_PICTURE_TYPE_B && !h->low_delay)

  1589.                 h->cur_pic.mb_type[s->mb_x + s->mb_y * s->mb_stride] =

  1590.                     (s->pict_type == AV_PICTURE_TYPE_P && mb_type < 8) ? (mb_type - 1) : -1;

  1591.         }

  1592. 

  1593.         ff_draw_horiz_band(avctx, s->cur_pic->f,

  1594.                            s->last_pic->f->data[0] ? s->last_pic->f : NULL,

  1595.                            16 * s->mb_y, 16, h->picture_structure, 0,

  1596.                            h->low_delay);

  1597.     }

  1598. 

  1599.     left = buf_size*8 - get_bits_count(&h->gb);

  1600. 

  1601.     if (sl->mb_y != h->mb_height || sl->mb_x != h->mb_width) {

  1602.         av_log(avctx, AV_LOG_INFO, "frame num %d incomplete pic x %d y %d left %d\n", avctx->frame_number, sl->mb_y, sl->mb_x, left);

  1603.         //av_hex_dump(stderr, buf+buf_size-8, 8);

  1604.     }

  1605. 

  1606.     if (left < 0) {

  1607.         av_log(avctx, AV_LOG_ERROR, "frame num %d left %d\n", avctx->frame_number, left);

  1608.         return -1;

  1609.     }

  1610. 

  1611.     if (s->pict_type == AV_PICTURE_TYPE_B || h->low_delay)

  1612.         ret = av_frame_ref(data, s->cur_pic->f);

  1613.     else if (s->last_pic->f->data[0])

  1614.         ret = av_frame_ref(data, s->last_pic->f);

  1615.     if (ret < 0)

  1616.         return ret;

  1617. 

  1618.     /* Do not output the last pic after seeking. */

  1619.     if (s->last_pic->f->data[0] || h->low_delay)

  1620.         *got_frame = 1;

  1621. 

  1622.     if (s->pict_type != AV_PICTURE_TYPE_B) {

  1623.         FFSWAP(H264Picture*, s->cur_pic, s->next_pic);

  1624.     } else {

  1625.         av_frame_unref(s->cur_pic->f);

  1626.     }

  1627. 

  1628.     return buf_size;

  1629. }

  1630. 

  1631. static av_cold int svq3_decode_end(AVCodecContext *avctx)

  1632. {

  1633.     SVQ3Context *s = avctx->priv_data;

  1634.     H264Context *h = &s->h;

  1635. 

  1636.     free_picture(avctx, s->cur_pic);

  1637.     free_picture(avctx, s->next_pic);

  1638.     free_picture(avctx, s->last_pic);

  1639.     av_frame_free(&s->cur_pic->f);

  1640.     av_frame_free(&s->next_pic->f);

  1641.     av_frame_free(&s->last_pic->f);

  1642.     av_freep(&s->cur_pic);

  1643.     av_freep(&s->next_pic);

  1644.     av_freep(&s->last_pic);

  1645.     av_freep(&s->slice_buf);

  1646.     av_freep(&s->intra4x4_pred_mode);

  1647.     av_freep(&s->edge_emu_buffer);

  1648.     av_freep(&s->mb2br_xy);

  1649. 

  1650.     memset(&h->cur_pic, 0, sizeof(h->cur_pic));

  1651. 

  1652.     ff_h264_free_context(h);

  1653. 

  1654.     av_freep(&s->buf);

  1655.     s->buf_size = 0;

  1656. 

  1657.     return 0;

  1658. }

  1659. 

  1660. AVCodec ff_svq3_decoder = {

  1661.     .name           = "svq3",

  1662.     .long_name      = NULL_IF_CONFIG_SMALL("Sorenson Vector Quantizer 3 / Sorenson Video 3 / SVQ3"),

  1663.     .type           = AVMEDIA_TYPE_VIDEO,

  1664.     .id             = AV_CODEC_ID_SVQ3,

  1665.     .priv_data_size = sizeof(SVQ3Context),

  1666.     .init           = svq3_decode_init,

  1667.     .close          = svq3_decode_end,

  1668.     .decode         = svq3_decode_frame,

  1669.     .capabilities   = AV_CODEC_CAP_DRAW_HORIZ_BAND |

  1670.                       AV_CODEC_CAP_DR1             |

  1671.                       AV_CODEC_CAP_DELAY,

  1672.     .pix_fmts       = (const enum AVPixelFormat[]) { AV_PIX_FMT_YUVJ420P,

  1673.                                                      AV_PIX_FMT_NONE},

  1674. };