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.
77194 Commits
32 Branches
326MB
Tree: d8f13e783a
Branches Tags
${ item.name }
Create branch ${ searchTerm }
from 'd8f13e783a'
${ noResults }
FFmpeg/libavcodec/diracdec.c

2128 lines
74KB
Raw Blame History 

  1. /*

  2.  * Copyright (C) 2007 Marco Gerards <marco@gnu.org>

  3.  * Copyright (C) 2009 David Conrad

  4.  * Copyright (C) 2011 Jordi Ortiz

  5.  *

  6.  * This file is part of FFmpeg.

  7.  *

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

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

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

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

  12.  *

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

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

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

  16.  * Lesser General Public License for more details.

  17.  *

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

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

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

  21.  */

  22. 

  23. /**

  24.  * @file

  25.  * Dirac Decoder

  26.  * @author Marco Gerards <marco@gnu.org>, David Conrad, Jordi Ortiz <nenjordi@gmail.com>

  27.  */

  28. 

  29. #include "avcodec.h"

  30. #include "get_bits.h"

  31. #include "bytestream.h"

  32. #include "internal.h"

  33. #include "golomb.h"

  34. #include "dirac_arith.h"

  35. #include "mpeg12data.h"

  36. #include "libavcodec/mpegvideo.h"

  37. #include "mpegvideoencdsp.h"

  38. #include "dirac_dwt.h"

  39. #include "dirac.h"

  40. #include "diracdsp.h"

  41. #include "videodsp.h"

  42. 

  43. /**

  44.  * The spec limits the number of wavelet decompositions to 4 for both

  45.  * level 1 (VC-2) and 128 (long-gop default).

  46.  * 5 decompositions is the maximum before >16-bit buffers are needed.

  47.  * Schroedinger allows this for DD 9,7 and 13,7 wavelets only, limiting

  48.  * the others to 4 decompositions (or 3 for the fidelity filter).

  49.  *

  50.  * We use this instead of MAX_DECOMPOSITIONS to save some memory.

  51.  */

  52. #define MAX_DWT_LEVELS 5

  53. 

  54. /**

  55.  * The spec limits this to 3 for frame coding, but in practice can be as high as 6

  56.  */

  57. #define MAX_REFERENCE_FRAMES 8

  58. #define MAX_DELAY 5         /* limit for main profile for frame coding (TODO: field coding) */

  59. #define MAX_FRAMES (MAX_REFERENCE_FRAMES + MAX_DELAY + 1)

  60. #define MAX_QUANT 255        /* max quant for VC-2 */

  61. #define MAX_BLOCKSIZE 32    /* maximum xblen/yblen we support */

  62. 

  63. /**

  64.  * DiracBlock->ref flags, if set then the block does MC from the given ref

  65.  */

  66. #define DIRAC_REF_MASK_REF1   1

  67. #define DIRAC_REF_MASK_REF2   2

  68. #define DIRAC_REF_MASK_GLOBAL 4

  69. 

  70. /**

  71.  * Value of Picture.reference when Picture is not a reference picture, but

  72.  * is held for delayed output.

  73.  */

  74. #define DELAYED_PIC_REF 4

  75. 

  76. #define CALC_PADDING(size, depth)                       \

  77.     (((size + (1 << depth) - 1) >> depth) << depth)

  78. 

  79. #define DIVRNDUP(a, b) (((a) + (b) - 1) / (b))

  80. 

  81. typedef struct {

  82.     AVFrame *avframe;

  83.     int interpolated[3];    /* 1 if hpel[] is valid */

  84.     uint8_t *hpel[3][4];

  85.     uint8_t *hpel_base[3][4];

  86.     int reference;

  87. } DiracFrame;

  88. 

  89. typedef struct {

  90.     union {

  91.         int16_t mv[2][2];

  92.         int16_t dc[3];

  93.     } u; /* anonymous unions aren't in C99 :( */

  94.     uint8_t ref;

  95. } DiracBlock;

  96. 

  97. typedef struct SubBand {

  98.     int level;

  99.     int orientation;

  100.     int stride; /* in bytes */

  101.     int width;

  102.     int height;

  103.     int pshift;

  104.     int quant;

  105.     uint8_t *ibuf;

  106.     struct SubBand *parent;

  107. 

  108.     /* for low delay */

  109.     unsigned length;

  110.     const uint8_t *coeff_data;

  111. } SubBand;

  112. 

  113. typedef struct Plane {

  114.     int width;

  115.     int height;

  116.     ptrdiff_t stride;

  117. 

  118.     int idwt_width;

  119.     int idwt_height;

  120.     int idwt_stride;

  121.     uint8_t *idwt_buf;

  122.     uint8_t *idwt_buf_base;

  123.     uint8_t *idwt_tmp;

  124. 

  125.     /* block length */

  126.     uint8_t xblen;

  127.     uint8_t yblen;

  128.     /* block separation (block n+1 starts after this many pixels in block n) */

  129.     uint8_t xbsep;

  130.     uint8_t ybsep;

  131.     /* amount of overspill on each edge (half of the overlap between blocks) */

  132.     uint8_t xoffset;

  133.     uint8_t yoffset;

  134. 

  135.     SubBand band[MAX_DWT_LEVELS][4];

  136. } Plane;

  137. 

  138. typedef struct DiracContext {

  139.     AVCodecContext *avctx;

  140.     MpegvideoEncDSPContext mpvencdsp;

  141.     VideoDSPContext vdsp;

  142.     DiracDSPContext diracdsp;

  143.     DiracVersionInfo version;

  144.     GetBitContext gb;

  145.     dirac_source_params source;

  146.     int seen_sequence_header;

  147.     int frame_number;           /* number of the next frame to display       */

  148.     Plane plane[3];

  149.     int chroma_x_shift;

  150.     int chroma_y_shift;

  151. 

  152.     int bit_depth;              /* bit depth                                 */

  153.     int pshift;                 /* pixel shift = bit_depth > 8               */

  154. 

  155.     int zero_res;               /* zero residue flag                         */

  156.     int is_arith;               /* whether coeffs use arith or golomb coding */

  157.     int core_syntax;            /* use core syntax only                      */

  158.     int low_delay;              /* use the low delay syntax                  */

  159.     int hq_picture;             /* high quality picture, enables low_delay   */

  160.     int ld_picture;             /* use low delay picture, turns on low_delay */

  161.     int dc_prediction;          /* has dc prediction                         */

  162.     int globalmc_flag;          /* use global motion compensation            */

  163.     int num_refs;               /* number of reference pictures              */

  164. 

  165.     /* wavelet decoding */

  166.     unsigned wavelet_depth;     /* depth of the IDWT                         */

  167.     unsigned wavelet_idx;

  168. 

  169.     /**

  170.      * schroedinger older than 1.0.8 doesn't store

  171.      * quant delta if only one codebook exists in a band

  172.      */

  173.     unsigned old_delta_quant;

  174.     unsigned codeblock_mode;

  175. 

  176.     unsigned num_x;              /* number of horizontal slices               */

  177.     unsigned num_y;              /* number of vertical slices                 */

  178. 

  179.     struct {

  180.         unsigned width;

  181.         unsigned height;

  182.     } codeblock[MAX_DWT_LEVELS+1];

  183. 

  184.     struct {

  185.         AVRational bytes;       /* average bytes per slice                   */

  186.         uint8_t quant[MAX_DWT_LEVELS][4]; /* [DIRAC_STD] E.1 */

  187.     } lowdelay;

  188. 

  189.     struct {

  190.         int pan_tilt[2];        /* pan/tilt vector                           */

  191.         int zrs[2][2];          /* zoom/rotate/shear matrix                  */

  192.         int perspective[2];     /* perspective vector                        */

  193.         unsigned zrs_exp;

  194.         unsigned perspective_exp;

  195.     } globalmc[2];

  196. 

  197.     /* motion compensation */

  198.     uint8_t mv_precision;       /* [DIRAC_STD] REFS_WT_PRECISION             */

  199.     int16_t weight[2];          /* [DIRAC_STD] REF1_WT and REF2_WT           */

  200.     unsigned weight_log2denom;  /* [DIRAC_STD] REFS_WT_PRECISION             */

  201. 

  202.     int blwidth;                /* number of blocks (horizontally)           */

  203.     int blheight;               /* number of blocks (vertically)             */

  204.     int sbwidth;                /* number of superblocks (horizontally)      */

  205.     int sbheight;               /* number of superblocks (vertically)        */

  206. 

  207.     uint8_t *sbsplit;

  208.     DiracBlock *blmotion;

  209. 

  210.     uint8_t *edge_emu_buffer[4];

  211.     uint8_t *edge_emu_buffer_base;

  212. 

  213.     uint16_t *mctmp;            /* buffer holding the MC data multiplied by OBMC weights */

  214.     uint8_t *mcscratch;

  215.     int buffer_stride;

  216. 

  217.     DECLARE_ALIGNED(16, uint8_t, obmc_weight)[3][MAX_BLOCKSIZE*MAX_BLOCKSIZE];

  218. 

  219.     void (*put_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);

  220.     void (*avg_pixels_tab[4])(uint8_t *dst, const uint8_t *src[5], int stride, int h);

  221.     void (*add_obmc)(uint16_t *dst, const uint8_t *src, int stride, const uint8_t *obmc_weight, int yblen);

  222.     dirac_weight_func weight_func;

  223.     dirac_biweight_func biweight_func;

  224. 

  225.     DiracFrame *current_picture;

  226.     DiracFrame *ref_pics[2];

  227. 

  228.     DiracFrame *ref_frames[MAX_REFERENCE_FRAMES+1];

  229.     DiracFrame *delay_frames[MAX_DELAY+1];

  230.     DiracFrame all_frames[MAX_FRAMES];

  231. } DiracContext;

  232. 

  233. enum dirac_subband {

  234.     subband_ll = 0,

  235.     subband_hl = 1,

  236.     subband_lh = 2,

  237.     subband_hh = 3,

  238.     subband_nb,

  239. };

  240. 

  241. static const uint8_t default_qmat[][4][4] = {

  242.     { { 5,  3,  3,  0}, { 0,  4,  4,  1}, { 0,  5,  5,  2}, { 0,  6,  6,  3} },

  243.     { { 4,  2,  2,  0}, { 0,  4,  4,  2}, { 0,  5,  5,  3}, { 0,  7,  7,  5} },

  244.     { { 5,  3,  3,  0}, { 0,  4,  4,  1}, { 0,  5,  5,  2}, { 0,  6,  6,  3} },

  245.     { { 8,  4,  4,  0}, { 0,  4,  4,  0}, { 0,  4,  4,  0}, { 0,  4,  4,  0} },

  246.     { { 8,  4,  4,  0}, { 0,  4,  4,  0}, { 0,  4,  4,  0}, { 0,  4,  4,  0} },

  247.     { { 0,  4,  4,  8}, { 0,  8,  8, 12}, { 0, 13, 13, 17}, { 0, 17, 17, 21} },

  248.     { { 3,  1,  1,  0}, { 0,  4,  4,  2}, { 0,  6,  6,  5}, { 0,  9,  9,  7} },

  249. };

  250. 

  251. static const int32_t qscale_tab[128] = {

  252.         4,         5,         6,         7,          8,        10,        11,        13,

  253.        16,        19,        23,        27,         32,        38,        45,        54,

  254.        64,        76,        91,       108,        128,       152,       181,       215,

  255.       256,       304,       362,       431,        512,       609,       724,       861,

  256.      1024,      1218,      1448,      1722,       2048,      2435,      2896,      3444,

  257.      4096,      4871,      5793,      6889,       8192,      9742,     11585,     13777,

  258.     16384,     19484,    -13317,     27554,      32768,     -1581,      9853,    -10518,

  259.     65536,     -3164,    -16782,    -21037,     131072,     -6328,      2922,     23552,

  260.    262144,    -12658,      5844,    -18524,     524288,     15232,     11689,     28578,

  261.   1048576,    -10085,    -13110,     -8471,    2097152,    -20170,     10267,    -16943,

  262.   4194304,       208,    -15954,     31741,    8388608,       416,      4579,     -2146,

  263.  16777216,       832,      9158,     -4293,   33554432,      1663,    -18172,     -8587,

  264.  67108864,      3326,       143,    -17175,   134217728,     6653,       285,     31276,

  265. 268435456,     13306,       570,     -3075,   536870912,   -13938,      1140,     -6152,

  266. 1073741824,    12672,      2281,    -12304, -2147483648,   -15205,      4561,    -24610,

  267.          0,    10138,      9122,     16407,           0,   -20274,    -18243,    -32813,

  268. };

  269. 

  270. static const int32_t qoffset_intra_tab[128] = {

  271.         1,         2,         3,         4,         4,         5,         6,         7,

  272.         8,        10,        12,        14,          16,        19,        23,        27,

  273.        32,        38,        46,        54,          64,        76,        91,       108,

  274.       128,       152,       181,       216,         256,       305,       362,       431,

  275.       512,       609,       724,       861,        1024,      1218,      1448,      1722,

  276.      2048,      2436,      2897,      3445,        4096,      4871,      5793,      6889,

  277.      8192,      9742,     -6658,     13777,       16384,      -790,      4927,     -5258,

  278.     32768,     -1581,     -8390,    -10518,       65536,     -3163,      1461,     11776,

  279.    131072,     -6328,      2922,     -9261,      262144,      7616,      5845,     14289,

  280.    524288,     -5042,     -6554,     -4235,     1048576,    -10084,      5134,     -8471,

  281.   2097152,       104,     -7976,     15871,     4194304,       208,      2290,     -1072,

  282.   8388608,       416,      4579,     -2146,    16777216,       832,     -9085,     -4293,

  283.  33554432,      1663,        72,     -8587,    67108864,      3327,       143,     15638,

  284. 134217728,      6653,       285,     -1537,   268435456,     -6968,       570,     -3075,

  285. 536870912,      6336,      1141,     -6151, -1073741823,     -7602,      2281,    -12304,

  286.         0,      5069,      4561,      8204,           0,    -10136,     -9121,    -16406,

  287. };

  288. 

  289. static const int qoffset_inter_tab[MAX_QUANT+1] = {

  290.     1,     2,     2,     3,     3,     4,     4,     5,

  291.     6,     7,     9,    10,    12,    14,    17,    20,

  292.     24,    29,    34,    41,    48,    57,    68,    81,

  293.     96,   114,   136,   162,   192,   228,   272,   323,

  294.     384,   457,   543,   646,   768,   913,  1086,  1292,

  295.     1536,  1827,  2172,  2583,  3072,  3653,  4344,  5166,

  296.     6144,  7307,  8689, 10333, 12288, 14613, 17378, 20666,

  297.     24576, 29226

  298. };

  299. 

  300. /* magic number division by 3 from schroedinger */

  301. static inline int divide3(int x)

  302. {

  303.     return ((x+1)*21845 + 10922) >> 16;

  304. }

  305. 

  306. static DiracFrame *remove_frame(DiracFrame *framelist[], int picnum)

  307. {

  308.     DiracFrame *remove_pic = NULL;

  309.     int i, remove_idx = -1;

  310. 

  311.     for (i = 0; framelist[i]; i++)

  312.         if (framelist[i]->avframe->display_picture_number == picnum) {

  313.             remove_pic = framelist[i];

  314.             remove_idx = i;

  315.         }

  316. 

  317.     if (remove_pic)

  318.         for (i = remove_idx; framelist[i]; i++)

  319.             framelist[i] = framelist[i+1];

  320. 

  321.     return remove_pic;

  322. }

  323. 

  324. static int add_frame(DiracFrame *framelist[], int maxframes, DiracFrame *frame)

  325. {

  326.     int i;

  327.     for (i = 0; i < maxframes; i++)

  328.         if (!framelist[i]) {

  329.             framelist[i] = frame;

  330.             return 0;

  331.         }

  332.     return -1;

  333. }

  334. 

  335. static int alloc_sequence_buffers(DiracContext *s)

  336. {

  337.     int sbwidth  = DIVRNDUP(s->source.width,  4);

  338.     int sbheight = DIVRNDUP(s->source.height, 4);

  339.     int i, w, h, top_padding;

  340. 

  341.     /* todo: think more about this / use or set Plane here */

  342.     for (i = 0; i < 3; i++) {

  343.         int max_xblen = MAX_BLOCKSIZE >> (i ? s->chroma_x_shift : 0);

  344.         int max_yblen = MAX_BLOCKSIZE >> (i ? s->chroma_y_shift : 0);

  345.         w = s->source.width  >> (i ? s->chroma_x_shift : 0);

  346.         h = s->source.height >> (i ? s->chroma_y_shift : 0);

  347. 

  348.         /* we allocate the max we support here since num decompositions can

  349.          * change from frame to frame. Stride is aligned to 16 for SIMD, and

  350.          * 1<<MAX_DWT_LEVELS top padding to avoid if(y>0) in arith decoding

  351.          * MAX_BLOCKSIZE padding for MC: blocks can spill up to half of that

  352.          * on each side */

  353.         top_padding = FFMAX(1<<MAX_DWT_LEVELS, max_yblen/2);

  354.         w = FFALIGN(CALC_PADDING(w, MAX_DWT_LEVELS), 8); /* FIXME: Should this be 16 for SSE??? */

  355.         h = top_padding + CALC_PADDING(h, MAX_DWT_LEVELS) + max_yblen/2;

  356. 

  357.         s->plane[i].idwt_buf_base = av_mallocz_array((w+max_xblen), h * (2 << s->pshift));

  358.         s->plane[i].idwt_tmp      = av_malloc_array((w+16), 2 << s->pshift);

  359.         s->plane[i].idwt_buf      = s->plane[i].idwt_buf_base + (top_padding*w)*(2 << s->pshift);

  360.         if (!s->plane[i].idwt_buf_base || !s->plane[i].idwt_tmp)

  361.             return AVERROR(ENOMEM);

  362.     }

  363. 

  364.     /* fixme: allocate using real stride here */

  365.     s->sbsplit  = av_malloc_array(sbwidth, sbheight);

  366.     s->blmotion = av_malloc_array(sbwidth, sbheight * 16 * sizeof(*s->blmotion));

  367. 

  368.     if (!s->sbsplit || !s->blmotion)

  369.         return AVERROR(ENOMEM);

  370.     return 0;

  371. }

  372. 

  373. static int alloc_buffers(DiracContext *s, int stride)

  374. {

  375.     int w = s->source.width;

  376.     int h = s->source.height;

  377. 

  378.     av_assert0(stride >= w);

  379.     stride += 64;

  380. 

  381.     if (s->buffer_stride >= stride)

  382.         return 0;

  383.     s->buffer_stride = 0;

  384. 

  385.     av_freep(&s->edge_emu_buffer_base);

  386.     memset(s->edge_emu_buffer, 0, sizeof(s->edge_emu_buffer));

  387.     av_freep(&s->mctmp);

  388.     av_freep(&s->mcscratch);

  389. 

  390.     s->edge_emu_buffer_base = av_malloc_array(stride, MAX_BLOCKSIZE);

  391. 

  392.     s->mctmp     = av_malloc_array((stride+MAX_BLOCKSIZE), (h+MAX_BLOCKSIZE) * sizeof(*s->mctmp));

  393.     s->mcscratch = av_malloc_array(stride, MAX_BLOCKSIZE);

  394. 

  395.     if (!s->edge_emu_buffer_base || !s->mctmp || !s->mcscratch)

  396.         return AVERROR(ENOMEM);

  397. 

  398.     s->buffer_stride = stride;

  399.     return 0;

  400. }

  401. 

  402. static void free_sequence_buffers(DiracContext *s)

  403. {

  404.     int i, j, k;

  405. 

  406.     for (i = 0; i < MAX_FRAMES; i++) {

  407.         if (s->all_frames[i].avframe->data[0]) {

  408.             av_frame_unref(s->all_frames[i].avframe);

  409.             memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));

  410.         }

  411. 

  412.         for (j = 0; j < 3; j++)

  413.             for (k = 1; k < 4; k++)

  414.                 av_freep(&s->all_frames[i].hpel_base[j][k]);

  415.     }

  416. 

  417.     memset(s->ref_frames, 0, sizeof(s->ref_frames));

  418.     memset(s->delay_frames, 0, sizeof(s->delay_frames));

  419. 

  420.     for (i = 0; i < 3; i++) {

  421.         av_freep(&s->plane[i].idwt_buf_base);

  422.         av_freep(&s->plane[i].idwt_tmp);

  423.     }

  424. 

  425.     s->buffer_stride = 0;

  426.     av_freep(&s->sbsplit);

  427.     av_freep(&s->blmotion);

  428.     av_freep(&s->edge_emu_buffer_base);

  429. 

  430.     av_freep(&s->mctmp);

  431.     av_freep(&s->mcscratch);

  432. }

  433. 

  434. static av_cold int dirac_decode_init(AVCodecContext *avctx)

  435. {

  436.     DiracContext *s = avctx->priv_data;

  437.     int i;

  438. 

  439.     s->avctx = avctx;

  440.     s->frame_number = -1;

  441. 

  442.     ff_diracdsp_init(&s->diracdsp);

  443.     ff_mpegvideoencdsp_init(&s->mpvencdsp, avctx);

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

  445. 

  446.     for (i = 0; i < MAX_FRAMES; i++) {

  447.         s->all_frames[i].avframe = av_frame_alloc();

  448.         if (!s->all_frames[i].avframe) {

  449.             while (i > 0)

  450.                 av_frame_free(&s->all_frames[--i].avframe);

  451.             return AVERROR(ENOMEM);

  452.         }

  453.     }

  454. 

  455.     return 0;

  456. }

  457. 

  458. static void dirac_decode_flush(AVCodecContext *avctx)

  459. {

  460.     DiracContext *s = avctx->priv_data;

  461.     free_sequence_buffers(s);

  462.     s->seen_sequence_header = 0;

  463.     s->frame_number = -1;

  464. }

  465. 

  466. static av_cold int dirac_decode_end(AVCodecContext *avctx)

  467. {

  468.     DiracContext *s = avctx->priv_data;

  469.     int i;

  470. 

  471.     dirac_decode_flush(avctx);

  472.     for (i = 0; i < MAX_FRAMES; i++)

  473.         av_frame_free(&s->all_frames[i].avframe);

  474. 

  475.     return 0;

  476. }

  477. 

  478. #define SIGN_CTX(x) (CTX_SIGN_ZERO + ((x) > 0) - ((x) < 0))

  479. 

  480. static inline int coeff_unpack_golomb(GetBitContext *gb, int qfactor, int qoffset)

  481. {

  482.     int sign, coeff;

  483. 

  484.     coeff = svq3_get_ue_golomb(gb);

  485.     if (coeff) {

  486.         coeff = (coeff * qfactor + qoffset + 2) >> 2;

  487.         sign  = get_bits1(gb);

  488.         coeff = (coeff ^ -sign) + sign;

  489.     }

  490.     return coeff;

  491. }

  492. 

  493. #define UNPACK_ARITH(n, type) \

  494.     static inline void coeff_unpack_arith_##n(DiracArith *c, int qfactor, int qoffset, \

  495.                                               SubBand *b, type *buf, int x, int y) \

  496.     { \

  497.         int coeff, sign, sign_pred = 0, pred_ctx = CTX_ZPZN_F1; \

  498.         const int mstride = -(b->stride >> (1+b->pshift)); \

  499.         if (b->parent) { \

  500.             const type *pbuf = (type *)b->parent->ibuf; \

  501.             const int stride = b->parent->stride >> (1+b->parent->pshift); \

  502.             pred_ctx += !!pbuf[stride * (y>>1) + (x>>1)] << 1; \

  503.         } \

  504.         if (b->orientation == subband_hl) \

  505.             sign_pred = buf[mstride]; \

  506.         if (x) { \

  507.             pred_ctx += !(buf[-1] | buf[mstride] | buf[-1 + mstride]); \

  508.             if (b->orientation == subband_lh) \

  509.                 sign_pred = buf[-1]; \

  510.         } else { \

  511.             pred_ctx += !buf[mstride]; \

  512.         } \

  513.         coeff = dirac_get_arith_uint(c, pred_ctx, CTX_COEFF_DATA); \

  514.         if (coeff) { \

  515.             coeff = (coeff * qfactor + qoffset + 2) >> 2; \

  516.             sign  = dirac_get_arith_bit(c, SIGN_CTX(sign_pred)); \

  517.             coeff = (coeff ^ -sign) + sign; \

  518.         } \

  519.         *buf = coeff; \

  520.     } \

  521. 

  522. UNPACK_ARITH(8, int16_t)

  523. UNPACK_ARITH(10, int32_t)

  524. 

  525. /**

  526.  * Decode the coeffs in the rectangle defined by left, right, top, bottom

  527.  * [DIRAC_STD] 13.4.3.2 Codeblock unpacking loop. codeblock()

  528.  */

  529. static inline void codeblock(DiracContext *s, SubBand *b,

  530.                              GetBitContext *gb, DiracArith *c,

  531.                              int left, int right, int top, int bottom,

  532.                              int blockcnt_one, int is_arith)

  533. {

  534.     int x, y, zero_block;

  535.     int qoffset, qfactor;

  536.     uint8_t *buf;

  537. 

  538.     /* check for any coded coefficients in this codeblock */

  539.     if (!blockcnt_one) {

  540.         if (is_arith)

  541.             zero_block = dirac_get_arith_bit(c, CTX_ZERO_BLOCK);

  542.         else

  543.             zero_block = get_bits1(gb);

  544. 

  545.         if (zero_block)

  546.             return;

  547.     }

  548. 

  549.     if (s->codeblock_mode && !(s->old_delta_quant && blockcnt_one)) {

  550.         int quant = b->quant;

  551.         if (is_arith)

  552.             quant += dirac_get_arith_int(c, CTX_DELTA_Q_F, CTX_DELTA_Q_DATA);

  553.         else

  554.             quant += dirac_get_se_golomb(gb);

  555.         if (quant < 0) {

  556.             av_log(s->avctx, AV_LOG_ERROR, "Invalid quant\n");

  557.             return;

  558.         }

  559.         b->quant = quant;

  560.     }

  561. 

  562.     b->quant = FFMIN(b->quant, MAX_QUANT);

  563. 

  564.     qfactor = qscale_tab[b->quant];

  565.     /* TODO: context pointer? */

  566.     if (!s->num_refs)

  567.         qoffset = qoffset_intra_tab[b->quant];

  568.     else

  569.         qoffset = qoffset_inter_tab[b->quant];

  570. 

  571.     buf = b->ibuf + top * b->stride;

  572.     if (is_arith) {

  573.         for (y = top; y < bottom; y++) {

  574.             for (x = left; x < right; x++) {

  575.                 if (b->pshift) {

  576.                     coeff_unpack_arith_10(c, qfactor, qoffset, b, (int32_t*)(buf)+x, x, y);

  577.                 } else {

  578.                     coeff_unpack_arith_8(c, qfactor, qoffset, b, (int16_t*)(buf)+x, x, y);

  579.                 }

  580.             }

  581.             buf += b->stride;

  582.         }

  583.     } else {

  584.         for (y = top; y < bottom; y++) {

  585.             for (x = left; x < right; x++) {

  586.                 int val = coeff_unpack_golomb(gb, qfactor, qoffset);

  587.                 if (b->pshift) {

  588.                     AV_WN32(&buf[4*x], val);

  589.                 } else {

  590.                     AV_WN16(&buf[2*x], val);

  591.                 }

  592.             }

  593.             buf += b->stride;

  594.          }

  595.      }

  596. }

  597. 

  598. /**

  599.  * Dirac Specification ->

  600.  * 13.3 intra_dc_prediction(band)

  601.  */

  602. #define INTRA_DC_PRED(n, type) \

  603.     static inline void intra_dc_prediction_##n(SubBand *b) \

  604.     { \

  605.         type *buf = (type*)b->ibuf; \

  606.         int x, y; \

  607.         \

  608.         for (x = 1; x < b->width; x++) \

  609.             buf[x] += buf[x-1]; \

  610.         buf += (b->stride >> (1+b->pshift)); \

  611.         \

  612.         for (y = 1; y < b->height; y++) { \

  613.             buf[0] += buf[-(b->stride >> (1+b->pshift))]; \

  614.             \

  615.             for (x = 1; x < b->width; x++) { \

  616.                 int pred = buf[x - 1] + buf[x - (b->stride >> (1+b->pshift))] + buf[x - (b->stride >> (1+b->pshift))-1]; \

  617.                 buf[x]  += divide3(pred); \

  618.             } \

  619.             buf += (b->stride >> (1+b->pshift)); \

  620.         } \

  621.     } \

  622. 

  623. INTRA_DC_PRED(8, int16_t)

  624. INTRA_DC_PRED(10, int32_t)

  625. 

  626. /**

  627.  * Dirac Specification ->

  628.  * 13.4.2 Non-skipped subbands.  subband_coeffs()

  629.  */

  630. static av_always_inline void decode_subband_internal(DiracContext *s, SubBand *b, int is_arith)

  631. {

  632.     int cb_x, cb_y, left, right, top, bottom;

  633.     DiracArith c;

  634.     GetBitContext gb;

  635.     int cb_width  = s->codeblock[b->level + (b->orientation != subband_ll)].width;

  636.     int cb_height = s->codeblock[b->level + (b->orientation != subband_ll)].height;

  637.     int blockcnt_one = (cb_width + cb_height) == 2;

  638. 

  639.     if (!b->length)

  640.         return;

  641. 

  642.     init_get_bits8(&gb, b->coeff_data, b->length);

  643. 

  644.     if (is_arith)

  645.         ff_dirac_init_arith_decoder(&c, &gb, b->length);

  646. 

  647.     top = 0;

  648.     for (cb_y = 0; cb_y < cb_height; cb_y++) {

  649.         bottom = (b->height * (cb_y+1LL)) / cb_height;

  650.         left = 0;

  651.         for (cb_x = 0; cb_x < cb_width; cb_x++) {

  652.             right = (b->width * (cb_x+1LL)) / cb_width;

  653.             codeblock(s, b, &gb, &c, left, right, top, bottom, blockcnt_one, is_arith);

  654.             left = right;

  655.         }

  656.         top = bottom;

  657.     }

  658. 

  659.     if (b->orientation == subband_ll && s->num_refs == 0) {

  660.         if (s->pshift) {

  661.             intra_dc_prediction_10(b);

  662.         } else {

  663.             intra_dc_prediction_8(b);

  664.         }

  665.     }

  666. }

  667. 

  668. static int decode_subband_arith(AVCodecContext *avctx, void *b)

  669. {

  670.     DiracContext *s = avctx->priv_data;

  671.     decode_subband_internal(s, b, 1);

  672.     return 0;

  673. }

  674. 

  675. static int decode_subband_golomb(AVCodecContext *avctx, void *arg)

  676. {

  677.     DiracContext *s = avctx->priv_data;

  678.     SubBand **b     = arg;

  679.     decode_subband_internal(s, *b, 0);

  680.     return 0;

  681. }

  682. 

  683. /**

  684.  * Dirac Specification ->

  685.  * [DIRAC_STD] 13.4.1 core_transform_data()

  686.  */

  687. static void decode_component(DiracContext *s, int comp)

  688. {

  689.     AVCodecContext *avctx = s->avctx;

  690.     SubBand *bands[3*MAX_DWT_LEVELS+1];

  691.     enum dirac_subband orientation;

  692.     int level, num_bands = 0;

  693. 

  694.     /* Unpack all subbands at all levels. */

  695.     for (level = 0; level < s->wavelet_depth; level++) {

  696.         for (orientation = !!level; orientation < 4; orientation++) {

  697.             SubBand *b = &s->plane[comp].band[level][orientation];

  698.             bands[num_bands++] = b;

  699. 

  700.             align_get_bits(&s->gb);

  701.             /* [DIRAC_STD] 13.4.2 subband() */

  702.             b->length = svq3_get_ue_golomb(&s->gb);

  703.             if (b->length) {

  704.                 b->quant = svq3_get_ue_golomb(&s->gb);

  705.                 align_get_bits(&s->gb);

  706.                 b->coeff_data = s->gb.buffer + get_bits_count(&s->gb)/8;

  707.                 b->length = FFMIN(b->length, FFMAX(get_bits_left(&s->gb)/8, 0));

  708.                 skip_bits_long(&s->gb, b->length*8);

  709.             }

  710.         }

  711.         /* arithmetic coding has inter-level dependencies, so we can only execute one level at a time */

  712.         if (s->is_arith)

  713.             avctx->execute(avctx, decode_subband_arith, &s->plane[comp].band[level][!!level],

  714.                            NULL, 4-!!level, sizeof(SubBand));

  715.     }

  716.     /* golomb coding has no inter-level dependencies, so we can execute all subbands in parallel */

  717.     if (!s->is_arith)

  718.         avctx->execute(avctx, decode_subband_golomb, bands, NULL, num_bands, sizeof(SubBand*));

  719. }

  720. 

  721. #define PARSE_VALUES(type, x, gb, ebits, buf1, buf2) \

  722.     type *buf = (type *)buf1; \

  723.     buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \

  724.     if (get_bits_count(gb) >= ebits) \

  725.         return; \

  726.     if (buf2) { \

  727.         buf = (type *)buf2; \

  728.         buf[x] = coeff_unpack_golomb(gb, qfactor, qoffset); \

  729.         if (get_bits_count(gb) >= ebits) \

  730.             return; \

  731.     } \

  732. 

  733. static void decode_subband(DiracContext *s, GetBitContext *gb, int quant,

  734.                            int slice_x, int slice_y, int bits_end,

  735.                            SubBand *b1, SubBand *b2)

  736. {

  737.     int left   = b1->width  * slice_x    / s->num_x;

  738.     int right  = b1->width  *(slice_x+1) / s->num_x;

  739.     int top    = b1->height * slice_y    / s->num_y;

  740.     int bottom = b1->height *(slice_y+1) / s->num_y;

  741. 

  742.     int qfactor = qscale_tab[quant & 0x7f];

  743.     int qoffset = qoffset_intra_tab[quant & 0x7f];

  744. 

  745.     uint8_t *buf1 =      b1->ibuf + top * b1->stride;

  746.     uint8_t *buf2 = b2 ? b2->ibuf + top * b2->stride: NULL;

  747.     int x, y;

  748.     /* we have to constantly check for overread since the spec explicitly

  749.        requires this, with the meaning that all remaining coeffs are set to 0 */

  750.     if (get_bits_count(gb) >= bits_end)

  751.         return;

  752. 

  753.     if (s->pshift) {

  754.         for (y = top; y < bottom; y++) {

  755.             for (x = left; x < right; x++) {

  756.                 PARSE_VALUES(int32_t, x, gb, bits_end, buf1, buf2);

  757.             }

  758.             buf1 += b1->stride;

  759.             if (buf2)

  760.                 buf2 += b2->stride;

  761.         }

  762.     }

  763.     else {

  764.         for (y = top; y < bottom; y++) {

  765.             for (x = left; x < right; x++) {

  766.                 PARSE_VALUES(int16_t, x, gb, bits_end, buf1, buf2);

  767.             }

  768.             buf1 += b1->stride;

  769.             if (buf2)

  770.                 buf2 += b2->stride;

  771.         }

  772.     }

  773. }

  774. 

  775. /* Used by Low Delay and High Quality profiles */

  776. typedef struct DiracSlice {

  777.     GetBitContext gb;

  778.     int slice_x;

  779.     int slice_y;

  780.     int bytes;

  781. } DiracSlice;

  782. 

  783. 

  784. /**

  785.  * Dirac Specification ->

  786.  * 13.5.2 Slices. slice(sx,sy)

  787.  */

  788. static int decode_lowdelay_slice(AVCodecContext *avctx, void *arg)

  789. {

  790.     DiracContext *s = avctx->priv_data;

  791.     DiracSlice *slice = arg;

  792.     GetBitContext *gb = &slice->gb;

  793.     enum dirac_subband orientation;

  794.     int level, quant, chroma_bits, chroma_end;

  795. 

  796.     int quant_base  = get_bits(gb, 7); /*[DIRAC_STD] qindex */

  797.     int length_bits = av_log2(8 * slice->bytes)+1;

  798.     int luma_bits   = get_bits_long(gb, length_bits);

  799.     int luma_end    = get_bits_count(gb) + FFMIN(luma_bits, get_bits_left(gb));

  800. 

  801.     /* [DIRAC_STD] 13.5.5.2 luma_slice_band */

  802.     for (level = 0; level < s->wavelet_depth; level++)

  803.         for (orientation = !!level; orientation < 4; orientation++) {

  804.             quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);

  805.             decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, luma_end,

  806.                            &s->plane[0].band[level][orientation], NULL);

  807.         }

  808. 

  809.     /* consume any unused bits from luma */

  810.     skip_bits_long(gb, get_bits_count(gb) - luma_end);

  811. 

  812.     chroma_bits = 8*slice->bytes - 7 - length_bits - luma_bits;

  813.     chroma_end  = get_bits_count(gb) + FFMIN(chroma_bits, get_bits_left(gb));

  814.     /* [DIRAC_STD] 13.5.5.3 chroma_slice_band */

  815.     for (level = 0; level < s->wavelet_depth; level++)

  816.         for (orientation = !!level; orientation < 4; orientation++) {

  817.             quant = FFMAX(quant_base - s->lowdelay.quant[level][orientation], 0);

  818.             decode_subband(s, gb, quant, slice->slice_x, slice->slice_y, chroma_end,

  819.                            &s->plane[1].band[level][orientation],

  820.                            &s->plane[2].band[level][orientation]);

  821.         }

  822. 

  823.     return 0;

  824. }

  825. 

  826. /**

  827.  * Dirac Specification ->

  828.  * 13.5.1 low_delay_transform_data()

  829.  */

  830. static int decode_lowdelay(DiracContext *s)

  831. {

  832.     AVCodecContext *avctx = s->avctx;

  833.     int slice_x, slice_y, bytes, bufsize;

  834.     const uint8_t *buf;

  835.     DiracSlice *slices;

  836.     int slice_num = 0;

  837. 

  838.     slices = av_mallocz_array(s->num_x, s->num_y * sizeof(DiracSlice));

  839.     if (!slices)

  840.         return AVERROR(ENOMEM);

  841. 

  842.     align_get_bits(&s->gb);

  843.     /*[DIRAC_STD] 13.5.2 Slices. slice(sx,sy) */

  844.     buf = s->gb.buffer + get_bits_count(&s->gb)/8;

  845.     bufsize = get_bits_left(&s->gb);

  846. 

  847.     for (slice_y = 0; bufsize > 0 && slice_y < s->num_y; slice_y++) {

  848.         for (slice_x = 0; bufsize > 0 && slice_x < s->num_x; slice_x++) {

  849.             bytes = (slice_num+1) * s->lowdelay.bytes.num / s->lowdelay.bytes.den

  850.                 - slice_num    * s->lowdelay.bytes.num / s->lowdelay.bytes.den;

  851.             slices[slice_num].bytes   = bytes;

  852.             slices[slice_num].slice_x = slice_x;

  853.             slices[slice_num].slice_y = slice_y;

  854.             init_get_bits(&slices[slice_num].gb, buf, bufsize);

  855.             slice_num++;

  856. 

  857.             buf     += bytes;

  858.             if (bufsize/8 >= bytes)

  859.                 bufsize -= bytes*8;

  860.             else

  861.                 bufsize = 0;

  862.         }

  863.     }

  864. 

  865.     avctx->execute(avctx, decode_lowdelay_slice, slices, NULL, slice_num,

  866.                    sizeof(struct DiracSlice)); /* [DIRAC_STD] 13.5.2 Slices */

  867.     if (s->dc_prediction) {

  868.         if (s->pshift) {

  869.             intra_dc_prediction_10(&s->plane[0].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */

  870.             intra_dc_prediction_10(&s->plane[1].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */

  871.             intra_dc_prediction_10(&s->plane[2].band[0][0]); /* [DIRAC_STD] 13.3 intra_dc_prediction() */

  872.         } else {

  873.             intra_dc_prediction_8(&s->plane[0].band[0][0]);

  874.             intra_dc_prediction_8(&s->plane[1].band[0][0]);

  875.             intra_dc_prediction_8(&s->plane[2].band[0][0]);

  876.         }

  877.     }

  878.     av_free(slices);

  879.     return 0;

  880. }

  881. 

  882. static void init_planes(DiracContext *s)

  883. {

  884.     int i, w, h, level, orientation;

  885. 

  886.     for (i = 0; i < 3; i++) {

  887.         Plane *p = &s->plane[i];

  888. 

  889.         p->width       = s->source.width  >> (i ? s->chroma_x_shift : 0);

  890.         p->height      = s->source.height >> (i ? s->chroma_y_shift : 0);

  891.         p->idwt_width  = w = CALC_PADDING(p->width , s->wavelet_depth);

  892.         p->idwt_height = h = CALC_PADDING(p->height, s->wavelet_depth);

  893.         p->idwt_stride = FFALIGN(p->idwt_width << (1 + s->pshift), 8);

  894. 

  895.         for (level = s->wavelet_depth-1; level >= 0; level--) {

  896.             w = w>>1;

  897.             h = h>>1;

  898.             for (orientation = !!level; orientation < 4; orientation++) {

  899.                 SubBand *b = &p->band[level][orientation];

  900. 

  901.                 b->pshift = s->pshift;

  902.                 b->ibuf   = p->idwt_buf;

  903.                 b->level  = level;

  904.                 b->stride = p->idwt_stride << (s->wavelet_depth - level);

  905.                 b->width  = w;

  906.                 b->height = h;

  907.                 b->orientation = orientation;

  908. 

  909.                 if (orientation & 1)

  910.                     b->ibuf += w << (1+b->pshift);

  911.                 if (orientation > 1)

  912.                     b->ibuf += (b->stride>>1);

  913. 

  914.                 if (level)

  915.                     b->parent = &p->band[level-1][orientation];

  916.             }

  917.         }

  918. 

  919.         if (i > 0) {

  920.             p->xblen = s->plane[0].xblen >> s->chroma_x_shift;

  921.             p->yblen = s->plane[0].yblen >> s->chroma_y_shift;

  922.             p->xbsep = s->plane[0].xbsep >> s->chroma_x_shift;

  923.             p->ybsep = s->plane[0].ybsep >> s->chroma_y_shift;

  924.         }

  925. 

  926.         p->xoffset = (p->xblen - p->xbsep)/2;

  927.         p->yoffset = (p->yblen - p->ybsep)/2;

  928.     }

  929. }

  930. 

  931. /**

  932.  * Unpack the motion compensation parameters

  933.  * Dirac Specification ->

  934.  * 11.2 Picture prediction data. picture_prediction()

  935.  */

  936. static int dirac_unpack_prediction_parameters(DiracContext *s)

  937. {

  938.     static const uint8_t default_blen[] = { 4, 12, 16, 24 };

  939. 

  940.     GetBitContext *gb = &s->gb;

  941.     unsigned idx, ref;

  942. 

  943.     align_get_bits(gb);

  944.     /* [DIRAC_STD] 11.2.2 Block parameters. block_parameters() */

  945.     /* Luma and Chroma are equal. 11.2.3 */

  946.     idx = svq3_get_ue_golomb(gb); /* [DIRAC_STD] index */

  947. 

  948.     if (idx > 4) {

  949.         av_log(s->avctx, AV_LOG_ERROR, "Block prediction index too high\n");

  950.         return AVERROR_INVALIDDATA;

  951.     }

  952. 

  953.     if (idx == 0) {

  954.         s->plane[0].xblen = svq3_get_ue_golomb(gb);

  955.         s->plane[0].yblen = svq3_get_ue_golomb(gb);

  956.         s->plane[0].xbsep = svq3_get_ue_golomb(gb);

  957.         s->plane[0].ybsep = svq3_get_ue_golomb(gb);

  958.     } else {

  959.         /*[DIRAC_STD] preset_block_params(index). Table 11.1 */

  960.         s->plane[0].xblen = default_blen[idx-1];

  961.         s->plane[0].yblen = default_blen[idx-1];

  962.         s->plane[0].xbsep = 4 * idx;

  963.         s->plane[0].ybsep = 4 * idx;

  964.     }

  965.     /*[DIRAC_STD] 11.2.4 motion_data_dimensions()

  966.       Calculated in function dirac_unpack_block_motion_data */

  967. 

  968.     if (s->plane[0].xblen % (1 << s->chroma_x_shift) != 0 ||

  969.         s->plane[0].yblen % (1 << s->chroma_y_shift) != 0 ||

  970.         !s->plane[0].xblen || !s->plane[0].yblen) {

  971.         av_log(s->avctx, AV_LOG_ERROR,

  972.                "invalid x/y block length (%d/%d) for x/y chroma shift (%d/%d)\n",

  973.                s->plane[0].xblen, s->plane[0].yblen, s->chroma_x_shift, s->chroma_y_shift);

  974.         return AVERROR_INVALIDDATA;

  975.     }

  976.     if (!s->plane[0].xbsep || !s->plane[0].ybsep || s->plane[0].xbsep < s->plane[0].xblen/2 || s->plane[0].ybsep < s->plane[0].yblen/2) {

  977.         av_log(s->avctx, AV_LOG_ERROR, "Block separation too small\n");

  978.         return AVERROR_INVALIDDATA;

  979.     }

  980.     if (s->plane[0].xbsep > s->plane[0].xblen || s->plane[0].ybsep > s->plane[0].yblen) {

  981.         av_log(s->avctx, AV_LOG_ERROR, "Block separation greater than size\n");

  982.         return AVERROR_INVALIDDATA;

  983.     }

  984.     if (FFMAX(s->plane[0].xblen, s->plane[0].yblen) > MAX_BLOCKSIZE) {

  985.         av_log(s->avctx, AV_LOG_ERROR, "Unsupported large block size\n");

  986.         return AVERROR_PATCHWELCOME;

  987.     }

  988. 

  989.     /*[DIRAC_STD] 11.2.5 Motion vector precision. motion_vector_precision()

  990.       Read motion vector precision */

  991.     s->mv_precision = svq3_get_ue_golomb(gb);

  992.     if (s->mv_precision > 3) {

  993.         av_log(s->avctx, AV_LOG_ERROR, "MV precision finer than eighth-pel\n");

  994.         return AVERROR_INVALIDDATA;

  995.     }

  996. 

  997.     /*[DIRAC_STD] 11.2.6 Global motion. global_motion()

  998.       Read the global motion compensation parameters */

  999.     s->globalmc_flag = get_bits1(gb);

  1000.     if (s->globalmc_flag) {

  1001.         memset(s->globalmc, 0, sizeof(s->globalmc));

  1002.         /* [DIRAC_STD] pan_tilt(gparams) */

  1003.         for (ref = 0; ref < s->num_refs; ref++) {

  1004.             if (get_bits1(gb)) {

  1005.                 s->globalmc[ref].pan_tilt[0] = dirac_get_se_golomb(gb);

  1006.                 s->globalmc[ref].pan_tilt[1] = dirac_get_se_golomb(gb);

  1007.             }

  1008.             /* [DIRAC_STD] zoom_rotate_shear(gparams)

  1009.                zoom/rotation/shear parameters */

  1010.             if (get_bits1(gb)) {

  1011.                 s->globalmc[ref].zrs_exp   = svq3_get_ue_golomb(gb);

  1012.                 s->globalmc[ref].zrs[0][0] = dirac_get_se_golomb(gb);

  1013.                 s->globalmc[ref].zrs[0][1] = dirac_get_se_golomb(gb);

  1014.                 s->globalmc[ref].zrs[1][0] = dirac_get_se_golomb(gb);

  1015.                 s->globalmc[ref].zrs[1][1] = dirac_get_se_golomb(gb);

  1016.             } else {

  1017.                 s->globalmc[ref].zrs[0][0] = 1;

  1018.                 s->globalmc[ref].zrs[1][1] = 1;

  1019.             }

  1020.             /* [DIRAC_STD] perspective(gparams) */

  1021.             if (get_bits1(gb)) {

  1022.                 s->globalmc[ref].perspective_exp = svq3_get_ue_golomb(gb);

  1023.                 s->globalmc[ref].perspective[0]  = dirac_get_se_golomb(gb);

  1024.                 s->globalmc[ref].perspective[1]  = dirac_get_se_golomb(gb);

  1025.             }

  1026.         }

  1027.     }

  1028. 

  1029.     /*[DIRAC_STD] 11.2.7 Picture prediction mode. prediction_mode()

  1030.       Picture prediction mode, not currently used. */

  1031.     if (svq3_get_ue_golomb(gb)) {

  1032.         av_log(s->avctx, AV_LOG_ERROR, "Unknown picture prediction mode\n");

  1033.         return AVERROR_INVALIDDATA;

  1034.     }

  1035. 

  1036.     /* [DIRAC_STD] 11.2.8 Reference picture weight. reference_picture_weights()

  1037.        just data read, weight calculation will be done later on. */

  1038.     s->weight_log2denom = 1;

  1039.     s->weight[0]        = 1;

  1040.     s->weight[1]        = 1;

  1041. 

  1042.     if (get_bits1(gb)) {

  1043.         s->weight_log2denom = svq3_get_ue_golomb(gb);

  1044.         s->weight[0] = dirac_get_se_golomb(gb);

  1045.         if (s->num_refs == 2)

  1046.             s->weight[1] = dirac_get_se_golomb(gb);

  1047.     }

  1048.     return 0;

  1049. }

  1050. 

  1051. /**

  1052.  * Dirac Specification ->

  1053.  * 11.3 Wavelet transform data. wavelet_transform()

  1054.  */

  1055. static int dirac_unpack_idwt_params(DiracContext *s)

  1056. {

  1057.     GetBitContext *gb = &s->gb;

  1058.     int i, level;

  1059.     unsigned tmp;

  1060. 

  1061. #define CHECKEDREAD(dst, cond, errmsg) \

  1062.     tmp = svq3_get_ue_golomb(gb); \

  1063.     if (cond) { \

  1064.         av_log(s->avctx, AV_LOG_ERROR, errmsg); \

  1065.         return AVERROR_INVALIDDATA; \

  1066.     }\

  1067.     dst = tmp;

  1068. 

  1069.     align_get_bits(gb);

  1070. 

  1071.     s->zero_res = s->num_refs ? get_bits1(gb) : 0;

  1072.     if (s->zero_res)

  1073.         return 0;

  1074. 

  1075.     /*[DIRAC_STD] 11.3.1 Transform parameters. transform_parameters() */

  1076.     CHECKEDREAD(s->wavelet_idx, tmp > 6, "wavelet_idx is too big\n")

  1077. 

  1078.     CHECKEDREAD(s->wavelet_depth, tmp > MAX_DWT_LEVELS || tmp < 1, "invalid number of DWT decompositions\n")

  1079. 

  1080.     if (!s->low_delay) {

  1081.         /* Codeblock parameters (core syntax only) */

  1082.         if (get_bits1(gb)) {

  1083.             for (i = 0; i <= s->wavelet_depth; i++) {

  1084.                 CHECKEDREAD(s->codeblock[i].width , tmp < 1 || tmp > (s->avctx->width >>s->wavelet_depth-i), "codeblock width invalid\n")

  1085.                 CHECKEDREAD(s->codeblock[i].height, tmp < 1 || tmp > (s->avctx->height>>s->wavelet_depth-i), "codeblock height invalid\n")

  1086.             }

  1087. 

  1088.             CHECKEDREAD(s->codeblock_mode, tmp > 1, "unknown codeblock mode\n")

  1089.         } else

  1090.             for (i = 0; i <= s->wavelet_depth; i++)

  1091.                 s->codeblock[i].width = s->codeblock[i].height = 1;

  1092.     } else {

  1093.         /* Slice parameters + quantization matrix*/

  1094.         /*[DIRAC_STD] 11.3.4 Slice coding Parameters (low delay syntax only). slice_parameters() */

  1095.         s->num_x     = svq3_get_ue_golomb(gb);

  1096.         s->num_y     = svq3_get_ue_golomb(gb);

  1097.         s->lowdelay.bytes.num = svq3_get_ue_golomb(gb);

  1098.         s->lowdelay.bytes.den = svq3_get_ue_golomb(gb);

  1099. 

  1100.         if (s->lowdelay.bytes.den <= 0) {

  1101.             av_log(s->avctx,AV_LOG_ERROR,"Invalid lowdelay.bytes.den\n");

  1102.             return AVERROR_INVALIDDATA;

  1103.         }

  1104. 

  1105.         /* [DIRAC_STD] 11.3.5 Quantisation matrices (low-delay syntax). quant_matrix() */

  1106.         if (get_bits1(gb)) {

  1107.             av_log(s->avctx,AV_LOG_DEBUG,"Low Delay: Has Custom Quantization Matrix!\n");

  1108.             /* custom quantization matrix */

  1109.             s->lowdelay.quant[0][0] = svq3_get_ue_golomb(gb);

  1110.             for (level = 0; level < s->wavelet_depth; level++) {

  1111.                 s->lowdelay.quant[level][1] = svq3_get_ue_golomb(gb);

  1112.                 s->lowdelay.quant[level][2] = svq3_get_ue_golomb(gb);

  1113.                 s->lowdelay.quant[level][3] = svq3_get_ue_golomb(gb);

  1114.             }

  1115.         } else {

  1116.             if (s->wavelet_depth > 4) {

  1117.                 av_log(s->avctx,AV_LOG_ERROR,"Mandatory custom low delay matrix missing for depth %d\n", s->wavelet_depth);

  1118.                 return AVERROR_INVALIDDATA;

  1119.             }

  1120.             /* default quantization matrix */

  1121.             for (level = 0; level < s->wavelet_depth; level++)

  1122.                 for (i = 0; i < 4; i++) {

  1123.                     s->lowdelay.quant[level][i] = default_qmat[s->wavelet_idx][level][i];

  1124.                     /* haar with no shift differs for different depths */

  1125.                     if (s->wavelet_idx == 3)

  1126.                         s->lowdelay.quant[level][i] += 4*(s->wavelet_depth-1 - level);

  1127.                 }

  1128.         }

  1129.     }

  1130.     return 0;

  1131. }

  1132. 

  1133. static inline int pred_sbsplit(uint8_t *sbsplit, int stride, int x, int y)

  1134. {

  1135.     static const uint8_t avgsplit[7] = { 0, 0, 1, 1, 1, 2, 2 };

  1136. 

  1137.     if (!(x|y))

  1138.         return 0;

  1139.     else if (!y)

  1140.         return sbsplit[-1];

  1141.     else if (!x)

  1142.         return sbsplit[-stride];

  1143. 

  1144.     return avgsplit[sbsplit[-1] + sbsplit[-stride] + sbsplit[-stride-1]];

  1145. }

  1146. 

  1147. static inline int pred_block_mode(DiracBlock *block, int stride, int x, int y, int refmask)

  1148. {

  1149.     int pred;

  1150. 

  1151.     if (!(x|y))

  1152.         return 0;

  1153.     else if (!y)

  1154.         return block[-1].ref & refmask;

  1155.     else if (!x)

  1156.         return block[-stride].ref & refmask;

  1157. 

  1158.     /* return the majority */

  1159.     pred = (block[-1].ref & refmask) + (block[-stride].ref & refmask) + (block[-stride-1].ref & refmask);

  1160.     return (pred >> 1) & refmask;

  1161. }

  1162. 

  1163. static inline void pred_block_dc(DiracBlock *block, int stride, int x, int y)

  1164. {

  1165.     int i, n = 0;

  1166. 

  1167.     memset(block->u.dc, 0, sizeof(block->u.dc));

  1168. 

  1169.     if (x && !(block[-1].ref & 3)) {

  1170.         for (i = 0; i < 3; i++)

  1171.             block->u.dc[i] += block[-1].u.dc[i];

  1172.         n++;

  1173.     }

  1174. 

  1175.     if (y && !(block[-stride].ref & 3)) {

  1176.         for (i = 0; i < 3; i++)

  1177.             block->u.dc[i] += block[-stride].u.dc[i];

  1178.         n++;

  1179.     }

  1180. 

  1181.     if (x && y && !(block[-1-stride].ref & 3)) {

  1182.         for (i = 0; i < 3; i++)

  1183.             block->u.dc[i] += block[-1-stride].u.dc[i];

  1184.         n++;

  1185.     }

  1186. 

  1187.     if (n == 2) {

  1188.         for (i = 0; i < 3; i++)

  1189.             block->u.dc[i] = (block->u.dc[i]+1)>>1;

  1190.     } else if (n == 3) {

  1191.         for (i = 0; i < 3; i++)

  1192.             block->u.dc[i] = divide3(block->u.dc[i]);

  1193.     }

  1194. }

  1195. 

  1196. static inline void pred_mv(DiracBlock *block, int stride, int x, int y, int ref)

  1197. {

  1198.     int16_t *pred[3];

  1199.     int refmask = ref+1;

  1200.     int mask = refmask | DIRAC_REF_MASK_GLOBAL; /*  exclude gmc blocks */

  1201.     int n = 0;

  1202. 

  1203.     if (x && (block[-1].ref & mask) == refmask)

  1204.         pred[n++] = block[-1].u.mv[ref];

  1205. 

  1206.     if (y && (block[-stride].ref & mask) == refmask)

  1207.         pred[n++] = block[-stride].u.mv[ref];

  1208. 

  1209.     if (x && y && (block[-stride-1].ref & mask) == refmask)

  1210.         pred[n++] = block[-stride-1].u.mv[ref];

  1211. 

  1212.     switch (n) {

  1213.     case 0:

  1214.         block->u.mv[ref][0] = 0;

  1215.         block->u.mv[ref][1] = 0;

  1216.         break;

  1217.     case 1:

  1218.         block->u.mv[ref][0] = pred[0][0];

  1219.         block->u.mv[ref][1] = pred[0][1];

  1220.         break;

  1221.     case 2:

  1222.         block->u.mv[ref][0] = (pred[0][0] + pred[1][0] + 1) >> 1;

  1223.         block->u.mv[ref][1] = (pred[0][1] + pred[1][1] + 1) >> 1;

  1224.         break;

  1225.     case 3:

  1226.         block->u.mv[ref][0] = mid_pred(pred[0][0], pred[1][0], pred[2][0]);

  1227.         block->u.mv[ref][1] = mid_pred(pred[0][1], pred[1][1], pred[2][1]);

  1228.         break;

  1229.     }

  1230. }

  1231. 

  1232. static void global_mv(DiracContext *s, DiracBlock *block, int x, int y, int ref)

  1233. {

  1234.     int ez      = s->globalmc[ref].zrs_exp;

  1235.     int ep      = s->globalmc[ref].perspective_exp;

  1236.     int (*A)[2] = s->globalmc[ref].zrs;

  1237.     int *b      = s->globalmc[ref].pan_tilt;

  1238.     int *c      = s->globalmc[ref].perspective;

  1239. 

  1240.     int m       = (1<<ep) - (c[0]*x + c[1]*y);

  1241.     int mx      = m * ((A[0][0] * x + A[0][1]*y) + (1<<ez) * b[0]);

  1242.     int my      = m * ((A[1][0] * x + A[1][1]*y) + (1<<ez) * b[1]);

  1243. 

  1244.     block->u.mv[ref][0] = (mx + (1<<(ez+ep))) >> (ez+ep);

  1245.     block->u.mv[ref][1] = (my + (1<<(ez+ep))) >> (ez+ep);

  1246. }

  1247. 

  1248. static void decode_block_params(DiracContext *s, DiracArith arith[8], DiracBlock *block,

  1249.                                 int stride, int x, int y)

  1250. {

  1251.     int i;

  1252. 

  1253.     block->ref  = pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF1);

  1254.     block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF1);

  1255. 

  1256.     if (s->num_refs == 2) {

  1257.         block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_REF2);

  1258.         block->ref ^= dirac_get_arith_bit(arith, CTX_PMODE_REF2) << 1;

  1259.     }

  1260. 

  1261.     if (!block->ref) {

  1262.         pred_block_dc(block, stride, x, y);

  1263.         for (i = 0; i < 3; i++)

  1264.             block->u.dc[i] += dirac_get_arith_int(arith+1+i, CTX_DC_F1, CTX_DC_DATA);

  1265.         return;

  1266.     }

  1267. 

  1268.     if (s->globalmc_flag) {

  1269.         block->ref |= pred_block_mode(block, stride, x, y, DIRAC_REF_MASK_GLOBAL);

  1270.         block->ref ^= dirac_get_arith_bit(arith, CTX_GLOBAL_BLOCK) << 2;

  1271.     }

  1272. 

  1273.     for (i = 0; i < s->num_refs; i++)

  1274.         if (block->ref & (i+1)) {

  1275.             if (block->ref & DIRAC_REF_MASK_GLOBAL) {

  1276.                 global_mv(s, block, x, y, i);

  1277.             } else {

  1278.                 pred_mv(block, stride, x, y, i);

  1279.                 block->u.mv[i][0] += dirac_get_arith_int(arith + 4 + 2 * i, CTX_MV_F1, CTX_MV_DATA);

  1280.                 block->u.mv[i][1] += dirac_get_arith_int(arith + 5 + 2 * i, CTX_MV_F1, CTX_MV_DATA);

  1281.             }

  1282.         }

  1283. }

  1284. 

  1285. /**

  1286.  * Copies the current block to the other blocks covered by the current superblock split mode

  1287.  */

  1288. static void propagate_block_data(DiracBlock *block, int stride, int size)

  1289. {

  1290.     int x, y;

  1291.     DiracBlock *dst = block;

  1292. 

  1293.     for (x = 1; x < size; x++)

  1294.         dst[x] = *block;

  1295. 

  1296.     for (y = 1; y < size; y++) {

  1297.         dst += stride;

  1298.         for (x = 0; x < size; x++)

  1299.             dst[x] = *block;

  1300.     }

  1301. }

  1302. 

  1303. /**

  1304.  * Dirac Specification ->

  1305.  * 12. Block motion data syntax

  1306.  */

  1307. static int dirac_unpack_block_motion_data(DiracContext *s)

  1308. {

  1309.     GetBitContext *gb = &s->gb;

  1310.     uint8_t *sbsplit = s->sbsplit;

  1311.     int i, x, y, q, p;

  1312.     DiracArith arith[8];

  1313. 

  1314.     align_get_bits(gb);

  1315. 

  1316.     /* [DIRAC_STD] 11.2.4 and 12.2.1 Number of blocks and superblocks */

  1317.     s->sbwidth  = DIVRNDUP(s->source.width,  4*s->plane[0].xbsep);

  1318.     s->sbheight = DIVRNDUP(s->source.height, 4*s->plane[0].ybsep);

  1319.     s->blwidth  = 4 * s->sbwidth;

  1320.     s->blheight = 4 * s->sbheight;

  1321. 

  1322.     /* [DIRAC_STD] 12.3.1 Superblock splitting modes. superblock_split_modes()

  1323.        decode superblock split modes */

  1324.     ff_dirac_init_arith_decoder(arith, gb, svq3_get_ue_golomb(gb));     /* svq3_get_ue_golomb(gb) is the length */

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

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

  1327.             unsigned int split  = dirac_get_arith_uint(arith, CTX_SB_F1, CTX_SB_DATA);

  1328.             if (split > 2)

  1329.                 return AVERROR_INVALIDDATA;

  1330.             sbsplit[x] = (split + pred_sbsplit(sbsplit+x, s->sbwidth, x, y)) % 3;

  1331.         }

  1332.         sbsplit += s->sbwidth;

  1333.     }

  1334. 

  1335.     /* setup arith decoding */

  1336.     ff_dirac_init_arith_decoder(arith, gb, svq3_get_ue_golomb(gb));

  1337.     for (i = 0; i < s->num_refs; i++) {

  1338.         ff_dirac_init_arith_decoder(arith + 4 + 2 * i, gb, svq3_get_ue_golomb(gb));

  1339.         ff_dirac_init_arith_decoder(arith + 5 + 2 * i, gb, svq3_get_ue_golomb(gb));

  1340.     }

  1341.     for (i = 0; i < 3; i++)

  1342.         ff_dirac_init_arith_decoder(arith+1+i, gb, svq3_get_ue_golomb(gb));

  1343. 

  1344.     for (y = 0; y < s->sbheight; y++)

  1345.         for (x = 0; x < s->sbwidth; x++) {

  1346.             int blkcnt = 1 << s->sbsplit[y * s->sbwidth + x];

  1347.             int step   = 4 >> s->sbsplit[y * s->sbwidth + x];

  1348. 

  1349.             for (q = 0; q < blkcnt; q++)

  1350.                 for (p = 0; p < blkcnt; p++) {

  1351.                     int bx = 4 * x + p*step;

  1352.                     int by = 4 * y + q*step;

  1353.                     DiracBlock *block = &s->blmotion[by*s->blwidth + bx];

  1354.                     decode_block_params(s, arith, block, s->blwidth, bx, by);

  1355.                     propagate_block_data(block, s->blwidth, step);

  1356.                 }

  1357.         }

  1358. 

  1359.     return 0;

  1360. }

  1361. 

  1362. static int weight(int i, int blen, int offset)

  1363. {

  1364. #define ROLLOFF(i) offset == 1 ? ((i) ? 5 : 3) :        \

  1365.     (1 + (6*(i) + offset - 1) / (2*offset - 1))

  1366. 

  1367.     if (i < 2*offset)

  1368.         return ROLLOFF(i);

  1369.     else if (i > blen-1 - 2*offset)

  1370.         return ROLLOFF(blen-1 - i);

  1371.     return 8;

  1372. }

  1373. 

  1374. static void init_obmc_weight_row(Plane *p, uint8_t *obmc_weight, int stride,

  1375.                                  int left, int right, int wy)

  1376. {

  1377.     int x;

  1378.     for (x = 0; left && x < p->xblen >> 1; x++)

  1379.         obmc_weight[x] = wy*8;

  1380.     for (; x < p->xblen >> right; x++)

  1381.         obmc_weight[x] = wy*weight(x, p->xblen, p->xoffset);

  1382.     for (; x < p->xblen; x++)

  1383.         obmc_weight[x] = wy*8;

  1384.     for (; x < stride; x++)

  1385.         obmc_weight[x] = 0;

  1386. }

  1387. 

  1388. static void init_obmc_weight(Plane *p, uint8_t *obmc_weight, int stride,

  1389.                              int left, int right, int top, int bottom)

  1390. {

  1391.     int y;

  1392.     for (y = 0; top && y < p->yblen >> 1; y++) {

  1393.         init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);

  1394.         obmc_weight += stride;

  1395.     }

  1396.     for (; y < p->yblen >> bottom; y++) {

  1397.         int wy = weight(y, p->yblen, p->yoffset);

  1398.         init_obmc_weight_row(p, obmc_weight, stride, left, right, wy);

  1399.         obmc_weight += stride;

  1400.     }

  1401.     for (; y < p->yblen; y++) {

  1402.         init_obmc_weight_row(p, obmc_weight, stride, left, right, 8);

  1403.         obmc_weight += stride;

  1404.     }

  1405. }

  1406. 

  1407. static void init_obmc_weights(DiracContext *s, Plane *p, int by)

  1408. {

  1409.     int top = !by;

  1410.     int bottom = by == s->blheight-1;

  1411. 

  1412.     /* don't bother re-initing for rows 2 to blheight-2, the weights don't change */

  1413.     if (top || bottom || by == 1) {

  1414.         init_obmc_weight(p, s->obmc_weight[0], MAX_BLOCKSIZE, 1, 0, top, bottom);

  1415.         init_obmc_weight(p, s->obmc_weight[1], MAX_BLOCKSIZE, 0, 0, top, bottom);

  1416.         init_obmc_weight(p, s->obmc_weight[2], MAX_BLOCKSIZE, 0, 1, top, bottom);

  1417.     }

  1418. }

  1419. 

  1420. static const uint8_t epel_weights[4][4][4] = {

  1421.     {{ 16,  0,  0,  0 },

  1422.      { 12,  4,  0,  0 },

  1423.      {  8,  8,  0,  0 },

  1424.      {  4, 12,  0,  0 }},

  1425.     {{ 12,  0,  4,  0 },

  1426.      {  9,  3,  3,  1 },

  1427.      {  6,  6,  2,  2 },

  1428.      {  3,  9,  1,  3 }},

  1429.     {{  8,  0,  8,  0 },

  1430.      {  6,  2,  6,  2 },

  1431.      {  4,  4,  4,  4 },

  1432.      {  2,  6,  2,  6 }},

  1433.     {{  4,  0, 12,  0 },

  1434.      {  3,  1,  9,  3 },

  1435.      {  2,  2,  6,  6 },

  1436.      {  1,  3,  3,  9 }}

  1437. };

  1438. 

  1439. /**

  1440.  * For block x,y, determine which of the hpel planes to do bilinear

  1441.  * interpolation from and set src[] to the location in each hpel plane

  1442.  * to MC from.

  1443.  *

  1444.  * @return the index of the put_dirac_pixels_tab function to use

  1445.  *  0 for 1 plane (fpel,hpel), 1 for 2 planes (qpel), 2 for 4 planes (qpel), and 3 for epel

  1446.  */

  1447. static int mc_subpel(DiracContext *s, DiracBlock *block, const uint8_t *src[5],

  1448.                      int x, int y, int ref, int plane)

  1449. {

  1450.     Plane *p = &s->plane[plane];

  1451.     uint8_t **ref_hpel = s->ref_pics[ref]->hpel[plane];

  1452.     int motion_x = block->u.mv[ref][0];

  1453.     int motion_y = block->u.mv[ref][1];

  1454.     int mx, my, i, epel, nplanes = 0;

  1455. 

  1456.     if (plane) {

  1457.         motion_x >>= s->chroma_x_shift;

  1458.         motion_y >>= s->chroma_y_shift;

  1459.     }

  1460. 

  1461.     mx         = motion_x & ~(-1U << s->mv_precision);

  1462.     my         = motion_y & ~(-1U << s->mv_precision);

  1463.     motion_x >>= s->mv_precision;

  1464.     motion_y >>= s->mv_precision;

  1465.     /* normalize subpel coordinates to epel */

  1466.     /* TODO: template this function? */

  1467.     mx      <<= 3 - s->mv_precision;

  1468.     my      <<= 3 - s->mv_precision;

  1469. 

  1470.     x += motion_x;

  1471.     y += motion_y;

  1472.     epel = (mx|my)&1;

  1473. 

  1474.     /* hpel position */

  1475.     if (!((mx|my)&3)) {

  1476.         nplanes = 1;

  1477.         src[0] = ref_hpel[(my>>1)+(mx>>2)] + y*p->stride + x;

  1478.     } else {

  1479.         /* qpel or epel */

  1480.         nplanes = 4;

  1481.         for (i = 0; i < 4; i++)

  1482.             src[i] = ref_hpel[i] + y*p->stride + x;

  1483. 

  1484.         /* if we're interpolating in the right/bottom halves, adjust the planes as needed

  1485.            we increment x/y because the edge changes for half of the pixels */

  1486.         if (mx > 4) {

  1487.             src[0] += 1;

  1488.             src[2] += 1;

  1489.             x++;

  1490.         }

  1491.         if (my > 4) {

  1492.             src[0] += p->stride;

  1493.             src[1] += p->stride;

  1494.             y++;

  1495.         }

  1496. 

  1497.         /* hpel planes are:

  1498.            [0]: F  [1]: H

  1499.            [2]: V  [3]: C */

  1500.         if (!epel) {

  1501.             /* check if we really only need 2 planes since either mx or my is

  1502.                a hpel position. (epel weights of 0 handle this there) */

  1503.             if (!(mx&3)) {

  1504.                 /* mx == 0: average [0] and [2]

  1505.                    mx == 4: average [1] and [3] */

  1506.                 src[!mx] = src[2 + !!mx];

  1507.                 nplanes = 2;

  1508.             } else if (!(my&3)) {

  1509.                 src[0] = src[(my>>1)  ];

  1510.                 src[1] = src[(my>>1)+1];

  1511.                 nplanes = 2;

  1512.             }

  1513.         } else {

  1514.             /* adjust the ordering if needed so the weights work */

  1515.             if (mx > 4) {

  1516.                 FFSWAP(const uint8_t *, src[0], src[1]);

  1517.                 FFSWAP(const uint8_t *, src[2], src[3]);

  1518.             }

  1519.             if (my > 4) {

  1520.                 FFSWAP(const uint8_t *, src[0], src[2]);

  1521.                 FFSWAP(const uint8_t *, src[1], src[3]);

  1522.             }

  1523.             src[4] = epel_weights[my&3][mx&3];

  1524.         }

  1525.     }

  1526. 

  1527.     /* fixme: v/h _edge_pos */

  1528.     if (x + p->xblen > p->width +EDGE_WIDTH/2 ||

  1529.         y + p->yblen > p->height+EDGE_WIDTH/2 ||

  1530.         x < 0 || y < 0) {

  1531.         for (i = 0; i < nplanes; i++) {

  1532.             s->vdsp.emulated_edge_mc(s->edge_emu_buffer[i], src[i],

  1533.                                      p->stride, p->stride,

  1534.                                      p->xblen, p->yblen, x, y,

  1535.                                      p->width+EDGE_WIDTH/2, p->height+EDGE_WIDTH/2);

  1536.             src[i] = s->edge_emu_buffer[i];

  1537.         }

  1538.     }

  1539.     return (nplanes>>1) + epel;

  1540. }

  1541. 

  1542. static void add_dc(uint16_t *dst, int dc, int stride,

  1543.                    uint8_t *obmc_weight, int xblen, int yblen)

  1544. {

  1545.     int x, y;

  1546.     dc += 128;

  1547. 

  1548.     for (y = 0; y < yblen; y++) {

  1549.         for (x = 0; x < xblen; x += 2) {

  1550.             dst[x  ] += dc * obmc_weight[x  ];

  1551.             dst[x+1] += dc * obmc_weight[x+1];

  1552.         }

  1553.         dst          += stride;

  1554.         obmc_weight  += MAX_BLOCKSIZE;

  1555.     }

  1556. }

  1557. 

  1558. static void block_mc(DiracContext *s, DiracBlock *block,

  1559.                      uint16_t *mctmp, uint8_t *obmc_weight,

  1560.                      int plane, int dstx, int dsty)

  1561. {

  1562.     Plane *p = &s->plane[plane];

  1563.     const uint8_t *src[5];

  1564.     int idx;

  1565. 

  1566.     switch (block->ref&3) {

  1567.     case 0: /* DC */

  1568.         add_dc(mctmp, block->u.dc[plane], p->stride, obmc_weight, p->xblen, p->yblen);

  1569.         return;

  1570.     case 1:

  1571.     case 2:

  1572.         idx = mc_subpel(s, block, src, dstx, dsty, (block->ref&3)-1, plane);

  1573.         s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);

  1574.         if (s->weight_func)

  1575.             s->weight_func(s->mcscratch, p->stride, s->weight_log2denom,

  1576.                            s->weight[0] + s->weight[1], p->yblen);

  1577.         break;

  1578.     case 3:

  1579.         idx = mc_subpel(s, block, src, dstx, dsty, 0, plane);

  1580.         s->put_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);

  1581.         idx = mc_subpel(s, block, src, dstx, dsty, 1, plane);

  1582.         if (s->biweight_func) {

  1583.             /* fixme: +32 is a quick hack */

  1584.             s->put_pixels_tab[idx](s->mcscratch + 32, src, p->stride, p->yblen);

  1585.             s->biweight_func(s->mcscratch, s->mcscratch+32, p->stride, s->weight_log2denom,

  1586.                              s->weight[0], s->weight[1], p->yblen);

  1587.         } else

  1588.             s->avg_pixels_tab[idx](s->mcscratch, src, p->stride, p->yblen);

  1589.         break;

  1590.     }

  1591.     s->add_obmc(mctmp, s->mcscratch, p->stride, obmc_weight, p->yblen);

  1592. }

  1593. 

  1594. static void mc_row(DiracContext *s, DiracBlock *block, uint16_t *mctmp, int plane, int dsty)

  1595. {

  1596.     Plane *p = &s->plane[plane];

  1597.     int x, dstx = p->xbsep - p->xoffset;

  1598. 

  1599.     block_mc(s, block, mctmp, s->obmc_weight[0], plane, -p->xoffset, dsty);

  1600.     mctmp += p->xbsep;

  1601. 

  1602.     for (x = 1; x < s->blwidth-1; x++) {

  1603.         block_mc(s, block+x, mctmp, s->obmc_weight[1], plane, dstx, dsty);

  1604.         dstx  += p->xbsep;

  1605.         mctmp += p->xbsep;

  1606.     }

  1607.     block_mc(s, block+x, mctmp, s->obmc_weight[2], plane, dstx, dsty);

  1608. }

  1609. 

  1610. static void select_dsp_funcs(DiracContext *s, int width, int height, int xblen, int yblen)

  1611. {

  1612.     int idx = 0;

  1613.     if (xblen > 8)

  1614.         idx = 1;

  1615.     if (xblen > 16)

  1616.         idx = 2;

  1617. 

  1618.     memcpy(s->put_pixels_tab, s->diracdsp.put_dirac_pixels_tab[idx], sizeof(s->put_pixels_tab));

  1619.     memcpy(s->avg_pixels_tab, s->diracdsp.avg_dirac_pixels_tab[idx], sizeof(s->avg_pixels_tab));

  1620.     s->add_obmc = s->diracdsp.add_dirac_obmc[idx];

  1621.     if (s->weight_log2denom > 1 || s->weight[0] != 1 || s->weight[1] != 1) {

  1622.         s->weight_func   = s->diracdsp.weight_dirac_pixels_tab[idx];

  1623.         s->biweight_func = s->diracdsp.biweight_dirac_pixels_tab[idx];

  1624.     } else {

  1625.         s->weight_func   = NULL;

  1626.         s->biweight_func = NULL;

  1627.     }

  1628. }

  1629. 

  1630. static int interpolate_refplane(DiracContext *s, DiracFrame *ref, int plane, int width, int height)

  1631. {

  1632.     /* chroma allocates an edge of 8 when subsampled

  1633.        which for 4:2:2 means an h edge of 16 and v edge of 8

  1634.        just use 8 for everything for the moment */

  1635.     int i, edge = EDGE_WIDTH/2;

  1636. 

  1637.     ref->hpel[plane][0] = ref->avframe->data[plane];

  1638.     s->mpvencdsp.draw_edges(ref->hpel[plane][0], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM); /* EDGE_TOP | EDGE_BOTTOM values just copied to make it build, this needs to be ensured */

  1639. 

  1640.     /* no need for hpel if we only have fpel vectors */

  1641.     if (!s->mv_precision)

  1642.         return 0;

  1643. 

  1644.     for (i = 1; i < 4; i++) {

  1645.         if (!ref->hpel_base[plane][i])

  1646.             ref->hpel_base[plane][i] = av_malloc((height+2*edge) * ref->avframe->linesize[plane] + 32);

  1647.         if (!ref->hpel_base[plane][i]) {

  1648.             return AVERROR(ENOMEM);

  1649.         }

  1650.         /* we need to be 16-byte aligned even for chroma */

  1651.         ref->hpel[plane][i] = ref->hpel_base[plane][i] + edge*ref->avframe->linesize[plane] + 16;

  1652.     }

  1653. 

  1654.     if (!ref->interpolated[plane]) {

  1655.         s->diracdsp.dirac_hpel_filter(ref->hpel[plane][1], ref->hpel[plane][2],

  1656.                                       ref->hpel[plane][3], ref->hpel[plane][0],

  1657.                                       ref->avframe->linesize[plane], width, height);

  1658.         s->mpvencdsp.draw_edges(ref->hpel[plane][1], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);

  1659.         s->mpvencdsp.draw_edges(ref->hpel[plane][2], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);

  1660.         s->mpvencdsp.draw_edges(ref->hpel[plane][3], ref->avframe->linesize[plane], width, height, edge, edge, EDGE_TOP | EDGE_BOTTOM);

  1661.     }

  1662.     ref->interpolated[plane] = 1;

  1663. 

  1664.     return 0;

  1665. }

  1666. 

  1667. /**

  1668.  * Dirac Specification ->

  1669.  * 13.0 Transform data syntax. transform_data()

  1670.  */

  1671. static int dirac_decode_frame_internal(DiracContext *s)

  1672. {

  1673.     DWTContext d;

  1674.     int y, i, comp, dsty;

  1675.     int ret;

  1676. 

  1677.     if (s->low_delay) {

  1678.         /* [DIRAC_STD] 13.5.1 low_delay_transform_data() */

  1679.         for (comp = 0; comp < 3; comp++) {

  1680.             Plane *p = &s->plane[comp];

  1681.             memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height);

  1682.         }

  1683.         if (!s->zero_res) {

  1684.             if ((ret = decode_lowdelay(s)) < 0)

  1685.                 return ret;

  1686.         }

  1687.     }

  1688. 

  1689.     for (comp = 0; comp < 3; comp++) {

  1690.         Plane *p       = &s->plane[comp];

  1691.         uint8_t *frame = s->current_picture->avframe->data[comp];

  1692. 

  1693.         /* FIXME: small resolutions */

  1694.         for (i = 0; i < 4; i++)

  1695.             s->edge_emu_buffer[i] = s->edge_emu_buffer_base + i*FFALIGN(p->width, 16);

  1696. 

  1697.         if (!s->zero_res && !s->low_delay)

  1698.         {

  1699.             memset(p->idwt_buf, 0, p->idwt_stride * p->idwt_height);

  1700.             decode_component(s, comp); /* [DIRAC_STD] 13.4.1 core_transform_data() */

  1701.         }

  1702.         ret = ff_spatial_idwt_init2(&d, p->idwt_buf, p->idwt_width, p->idwt_height, p->idwt_stride,

  1703.                                     s->wavelet_idx+2, s->wavelet_depth, p->idwt_tmp, s->bit_depth);

  1704.         if (ret < 0)

  1705.             return ret;

  1706. 

  1707.         if (!s->num_refs) { /* intra */

  1708.             for (y = 0; y < p->height; y += 16) {

  1709.                 ff_spatial_idwt_slice2(&d, y+16); /* decode */

  1710.                 s->diracdsp.put_signed_rect_clamped[s->pshift](frame + y*p->stride, p->stride,

  1711.                                                                p->idwt_buf + y*p->idwt_stride, p->idwt_stride, p->width, 16);

  1712.             }

  1713.         } else { /* inter */

  1714.             int rowheight = p->ybsep*p->stride;

  1715. 

  1716.             select_dsp_funcs(s, p->width, p->height, p->xblen, p->yblen);

  1717. 

  1718.             for (i = 0; i < s->num_refs; i++) {

  1719.                 int ret = interpolate_refplane(s, s->ref_pics[i], comp, p->width, p->height);

  1720.                 if (ret < 0)

  1721.                     return ret;

  1722.             }

  1723. 

  1724.             memset(s->mctmp, 0, 4*p->yoffset*p->stride);

  1725. 

  1726.             dsty = -p->yoffset;

  1727.             for (y = 0; y < s->blheight; y++) {

  1728.                 int h     = 0,

  1729.                     start = FFMAX(dsty, 0);

  1730.                 uint16_t *mctmp    = s->mctmp + y*rowheight;

  1731.                 DiracBlock *blocks = s->blmotion + y*s->blwidth;

  1732. 

  1733.                 init_obmc_weights(s, p, y);

  1734. 

  1735.                 if (y == s->blheight-1 || start+p->ybsep > p->height)

  1736.                     h = p->height - start;

  1737.                 else

  1738.                     h = p->ybsep - (start - dsty);

  1739.                 if (h < 0)

  1740.                     break;

  1741. 

  1742.                 memset(mctmp+2*p->yoffset*p->stride, 0, 2*rowheight);

  1743.                 mc_row(s, blocks, mctmp, comp, dsty);

  1744. 

  1745.                 mctmp += (start - dsty)*p->stride + p->xoffset;

  1746.                 ff_spatial_idwt_slice2(&d, start + h); /* decode */

  1747.                 /* NOTE: add_rect_clamped hasn't been templated hence the shifts.

  1748.                  * idwt_stride is passed as pixels, not in bytes as in the rest of the decoder */

  1749.                 s->diracdsp.add_rect_clamped(frame + start*p->stride, mctmp, p->stride,

  1750.                                              (int16_t*)(p->idwt_buf) + start*(p->idwt_stride >> 1), (p->idwt_stride >> 1), p->width, h);

  1751. 

  1752.                 dsty += p->ybsep;

  1753.             }

  1754.         }

  1755.     }

  1756. 

  1757. 

  1758.     return 0;

  1759. }

  1760. 

  1761. static int get_buffer_with_edge(AVCodecContext *avctx, AVFrame *f, int flags)

  1762. {

  1763.     int ret, i;

  1764.     int chroma_x_shift, chroma_y_shift;

  1765.     avcodec_get_chroma_sub_sample(avctx->pix_fmt, &chroma_x_shift, &chroma_y_shift);

  1766. 

  1767.     f->width  = avctx->width  + 2 * EDGE_WIDTH;

  1768.     f->height = avctx->height + 2 * EDGE_WIDTH + 2;

  1769.     ret = ff_get_buffer(avctx, f, flags);

  1770.     if (ret < 0)

  1771.         return ret;

  1772. 

  1773.     for (i = 0; f->data[i]; i++) {

  1774.         int offset = (EDGE_WIDTH >> (i && i<3 ? chroma_y_shift : 0)) *

  1775.                      f->linesize[i] + 32;

  1776.         f->data[i] += offset;

  1777.     }

  1778.     f->width  = avctx->width;

  1779.     f->height = avctx->height;

  1780. 

  1781.     return 0;

  1782. }

  1783. 

  1784. /**

  1785.  * Dirac Specification ->

  1786.  * 11.1.1 Picture Header. picture_header()

  1787.  */

  1788. static int dirac_decode_picture_header(DiracContext *s)

  1789. {

  1790.     unsigned retire, picnum;

  1791.     int i, j, ret;

  1792.     int64_t refdist, refnum;

  1793.     GetBitContext *gb = &s->gb;

  1794. 

  1795.     /* [DIRAC_STD] 11.1.1 Picture Header. picture_header() PICTURE_NUM */

  1796.     picnum = s->current_picture->avframe->display_picture_number = get_bits_long(gb, 32);

  1797. 

  1798. 

  1799.     av_log(s->avctx,AV_LOG_DEBUG,"PICTURE_NUM: %d\n",picnum);

  1800. 

  1801.     /* if this is the first keyframe after a sequence header, start our

  1802.        reordering from here */

  1803.     if (s->frame_number < 0)

  1804.         s->frame_number = picnum;

  1805. 

  1806.     s->ref_pics[0] = s->ref_pics[1] = NULL;

  1807.     for (i = 0; i < s->num_refs; i++) {

  1808.         refnum = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;

  1809.         refdist = INT64_MAX;

  1810. 

  1811.         /* find the closest reference to the one we want */

  1812.         /* Jordi: this is needed if the referenced picture hasn't yet arrived */

  1813.         for (j = 0; j < MAX_REFERENCE_FRAMES && refdist; j++)

  1814.             if (s->ref_frames[j]

  1815.                 && FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum) < refdist) {

  1816.                 s->ref_pics[i] = s->ref_frames[j];

  1817.                 refdist = FFABS(s->ref_frames[j]->avframe->display_picture_number - refnum);

  1818.             }

  1819. 

  1820.         if (!s->ref_pics[i] || refdist)

  1821.             av_log(s->avctx, AV_LOG_DEBUG, "Reference not found\n");

  1822. 

  1823.         /* if there were no references at all, allocate one */

  1824.         if (!s->ref_pics[i])

  1825.             for (j = 0; j < MAX_FRAMES; j++)

  1826.                 if (!s->all_frames[j].avframe->data[0]) {

  1827.                     s->ref_pics[i] = &s->all_frames[j];

  1828.                     get_buffer_with_edge(s->avctx, s->ref_pics[i]->avframe, AV_GET_BUFFER_FLAG_REF);

  1829.                     break;

  1830.                 }

  1831. 

  1832.         if (!s->ref_pics[i]) {

  1833.             av_log(s->avctx, AV_LOG_ERROR, "Reference could not be allocated\n");

  1834.             return AVERROR_INVALIDDATA;

  1835.         }

  1836. 

  1837.     }

  1838. 

  1839.     /* retire the reference frames that are not used anymore */

  1840.     if (s->current_picture->reference) {

  1841.         retire = (picnum + dirac_get_se_golomb(gb)) & 0xFFFFFFFF;

  1842.         if (retire != picnum) {

  1843.             DiracFrame *retire_pic = remove_frame(s->ref_frames, retire);

  1844. 

  1845.             if (retire_pic)

  1846.                 retire_pic->reference &= DELAYED_PIC_REF;

  1847.             else

  1848.                 av_log(s->avctx, AV_LOG_DEBUG, "Frame to retire not found\n");

  1849.         }

  1850. 

  1851.         /* if reference array is full, remove the oldest as per the spec */

  1852.         while (add_frame(s->ref_frames, MAX_REFERENCE_FRAMES, s->current_picture)) {

  1853.             av_log(s->avctx, AV_LOG_ERROR, "Reference frame overflow\n");

  1854.             remove_frame(s->ref_frames, s->ref_frames[0]->avframe->display_picture_number)->reference &= DELAYED_PIC_REF;

  1855.         }

  1856.     }

  1857. 

  1858.     if (s->num_refs) {

  1859.         ret = dirac_unpack_prediction_parameters(s);  /* [DIRAC_STD] 11.2 Picture Prediction Data. picture_prediction() */

  1860.         if (ret < 0)

  1861.             return ret;

  1862.         ret = dirac_unpack_block_motion_data(s);      /* [DIRAC_STD] 12. Block motion data syntax                       */

  1863.         if (ret < 0)

  1864.             return ret;

  1865.     }

  1866.     ret = dirac_unpack_idwt_params(s);                /* [DIRAC_STD] 11.3 Wavelet transform data                        */

  1867.     if (ret < 0)

  1868.         return ret;

  1869. 

  1870.     init_planes(s);

  1871.     return 0;

  1872. }

  1873. 

  1874. static int get_delayed_pic(DiracContext *s, AVFrame *picture, int *got_frame)

  1875. {

  1876.     DiracFrame *out = s->delay_frames[0];

  1877.     int i, out_idx  = 0;

  1878.     int ret;

  1879. 

  1880.     /* find frame with lowest picture number */

  1881.     for (i = 1; s->delay_frames[i]; i++)

  1882.         if (s->delay_frames[i]->avframe->display_picture_number < out->avframe->display_picture_number) {

  1883.             out     = s->delay_frames[i];

  1884.             out_idx = i;

  1885.         }

  1886. 

  1887.     for (i = out_idx; s->delay_frames[i]; i++)

  1888.         s->delay_frames[i] = s->delay_frames[i+1];

  1889. 

  1890.     if (out) {

  1891.         out->reference ^= DELAYED_PIC_REF;

  1892.         *got_frame = 1;

  1893.         if((ret = av_frame_ref(picture, out->avframe)) < 0)

  1894.             return ret;

  1895.     }

  1896. 

  1897.     return 0;

  1898. }

  1899. 

  1900. /**

  1901.  * Dirac Specification ->

  1902.  * 9.6 Parse Info Header Syntax. parse_info()

  1903.  * 4 byte start code + byte parse code + 4 byte size + 4 byte previous size

  1904.  */

  1905. #define DATA_UNIT_HEADER_SIZE 13

  1906. 

  1907. /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3

  1908.    inside the function parse_sequence() */

  1909. static int dirac_decode_data_unit(AVCodecContext *avctx, const uint8_t *buf, int size)

  1910. {

  1911.     DiracContext *s   = avctx->priv_data;

  1912.     DiracFrame *pic   = NULL;

  1913.     int ret, i;

  1914.     uint8_t parse_code;

  1915.     unsigned tmp;

  1916. 

  1917.     if (size < DATA_UNIT_HEADER_SIZE)

  1918.         return AVERROR_INVALIDDATA;

  1919. 

  1920.     parse_code = buf[4];

  1921. 

  1922.     init_get_bits(&s->gb, &buf[13], 8*(size - DATA_UNIT_HEADER_SIZE));

  1923. 

  1924.     if (parse_code == DIRAC_PCODE_SEQ_HEADER) {

  1925.         if (s->seen_sequence_header)

  1926.             return 0;

  1927. 

  1928.         /* [DIRAC_STD] 10. Sequence header */

  1929.         ret = avpriv_dirac_parse_sequence_header(avctx, &s->gb, &s->source,

  1930.                                                  &s->version, &s->bit_depth);

  1931.         if (ret < 0)

  1932.             return ret;

  1933. 

  1934.         s->pshift = s->bit_depth > 8;

  1935. 

  1936.         avcodec_get_chroma_sub_sample(avctx->pix_fmt, &s->chroma_x_shift, &s->chroma_y_shift);

  1937. 

  1938.         ret = alloc_sequence_buffers(s);

  1939.         if (ret < 0)

  1940.             return ret;

  1941. 

  1942.         s->seen_sequence_header = 1;

  1943.     } else if (parse_code == DIRAC_PCODE_END_SEQ) { /* [DIRAC_STD] End of Sequence */

  1944.         free_sequence_buffers(s);

  1945.         s->seen_sequence_header = 0;

  1946.     } else if (parse_code == DIRAC_PCODE_AUX) {

  1947.         if (buf[13] == 1) {     /* encoder implementation/version */

  1948.             int ver[3];

  1949.             /* versions older than 1.0.8 don't store quant delta for

  1950.                subbands with only one codeblock */

  1951.             if (sscanf(buf+14, "Schroedinger %d.%d.%d", ver, ver+1, ver+2) == 3)

  1952.                 if (ver[0] == 1 && ver[1] == 0 && ver[2] <= 7)

  1953.                     s->old_delta_quant = 1;

  1954.         }

  1955.     } else if (parse_code & 0x8) {  /* picture data unit */

  1956.         if (!s->seen_sequence_header) {

  1957.             av_log(avctx, AV_LOG_DEBUG, "Dropping frame without sequence header\n");

  1958.             return AVERROR_INVALIDDATA;

  1959.         }

  1960. 

  1961.         /* find an unused frame */

  1962.         for (i = 0; i < MAX_FRAMES; i++)

  1963.             if (s->all_frames[i].avframe->data[0] == NULL)

  1964.                 pic = &s->all_frames[i];

  1965.         if (!pic) {

  1966.             av_log(avctx, AV_LOG_ERROR, "framelist full\n");

  1967.             return AVERROR_INVALIDDATA;

  1968.         }

  1969. 

  1970.         av_frame_unref(pic->avframe);

  1971. 

  1972.         /* [DIRAC_STD] Defined in 9.6.1 ... */

  1973.         tmp            =  parse_code & 0x03;                   /* [DIRAC_STD] num_refs()      */

  1974.         if (tmp > 2) {

  1975.             av_log(avctx, AV_LOG_ERROR, "num_refs of 3\n");

  1976.             return AVERROR_INVALIDDATA;

  1977.         }

  1978.         s->num_refs      = tmp;

  1979.         s->is_arith      = (parse_code & 0x48) == 0x08;          /* [DIRAC_STD] using_ac()            */

  1980.         s->low_delay     = (parse_code & 0x88) == 0x88;          /* [DIRAC_STD] is_low_delay()        */

  1981.         s->core_syntax   = (parse_code & 0x88) == 0x08;          /* [DIRAC_STD] is_core_syntax()      */

  1982.         s->ld_picture    = (parse_code & 0xF8) == 0xC8;          /* [DIRAC_STD] is_ld_picture()       */

  1983.         s->hq_picture    = (parse_code & 0xF8) == 0xE8;          /* [DIRAC_STD] is_hq_picture()       */

  1984.         s->dc_prediction = (parse_code & 0x28) == 0x08;          /* [DIRAC_STD] using_dc_prediction() */

  1985.         pic->reference   = (parse_code & 0x0C) == 0x0C;          /* [DIRAC_STD] is_reference()        */

  1986.         pic->avframe->key_frame = s->num_refs == 0;              /* [DIRAC_STD] is_intra()            */

  1987.         pic->avframe->pict_type = s->num_refs + 1;               /* Definition of AVPictureType in avutil.h */

  1988. 

  1989.         if (s->version.minor == 2 && parse_code == 0x88)

  1990.             s->ld_picture = 1;

  1991. 

  1992.         if (s->low_delay && !(s->ld_picture || s->hq_picture) ) {

  1993.             av_log(avctx, AV_LOG_ERROR, "Invalid low delay flag\n");

  1994.             return AVERROR_INVALIDDATA;

  1995.         }

  1996. 

  1997.         if ((ret = get_buffer_with_edge(avctx, pic->avframe, (parse_code & 0x0C) == 0x0C ? AV_GET_BUFFER_FLAG_REF : 0)) < 0)

  1998.             return ret;

  1999.         s->current_picture = pic;

  2000.         s->plane[0].stride = pic->avframe->linesize[0];

  2001.         s->plane[1].stride = pic->avframe->linesize[1];

  2002.         s->plane[2].stride = pic->avframe->linesize[2];

  2003. 

  2004.         if (alloc_buffers(s, FFMAX3(FFABS(s->plane[0].stride), FFABS(s->plane[1].stride), FFABS(s->plane[2].stride))) < 0)

  2005.             return AVERROR(ENOMEM);

  2006. 

  2007.         /* [DIRAC_STD] 11.1 Picture parse. picture_parse() */

  2008.         ret = dirac_decode_picture_header(s);

  2009.         if (ret < 0)

  2010.             return ret;

  2011. 

  2012.         /* [DIRAC_STD] 13.0 Transform data syntax. transform_data() */

  2013.         ret = dirac_decode_frame_internal(s);

  2014.         if (ret < 0)

  2015.             return ret;

  2016.     }

  2017.     return 0;

  2018. }

  2019. 

  2020. static int dirac_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, AVPacket *pkt)

  2021. {

  2022.     DiracContext *s     = avctx->priv_data;

  2023.     AVFrame *picture    = data;

  2024.     uint8_t *buf        = pkt->data;

  2025.     int buf_size        = pkt->size;

  2026.     int i, buf_idx      = 0;

  2027.     int ret;

  2028.     unsigned data_unit_size;

  2029. 

  2030.     /* release unused frames */

  2031.     for (i = 0; i < MAX_FRAMES; i++)

  2032.         if (s->all_frames[i].avframe->data[0] && !s->all_frames[i].reference) {

  2033.             av_frame_unref(s->all_frames[i].avframe);

  2034.             memset(s->all_frames[i].interpolated, 0, sizeof(s->all_frames[i].interpolated));

  2035.         }

  2036. 

  2037.     s->current_picture = NULL;

  2038.     *got_frame = 0;

  2039. 

  2040.     /* end of stream, so flush delayed pics */

  2041.     if (buf_size == 0)

  2042.         return get_delayed_pic(s, (AVFrame *)data, got_frame);

  2043. 

  2044.     for (;;) {

  2045.         /*[DIRAC_STD] Here starts the code from parse_info() defined in 9.6

  2046.           [DIRAC_STD] PARSE_INFO_PREFIX = "BBCD" as defined in ISO/IEC 646

  2047.           BBCD start code search */

  2048.         for (; buf_idx + DATA_UNIT_HEADER_SIZE < buf_size; buf_idx++) {

  2049.             if (buf[buf_idx  ] == 'B' && buf[buf_idx+1] == 'B' &&

  2050.                 buf[buf_idx+2] == 'C' && buf[buf_idx+3] == 'D')

  2051.                 break;

  2052.         }

  2053.         /* BBCD found or end of data */

  2054.         if (buf_idx + DATA_UNIT_HEADER_SIZE >= buf_size)

  2055.             break;

  2056. 

  2057.         data_unit_size = AV_RB32(buf+buf_idx+5);

  2058.         if (data_unit_size > buf_size - buf_idx || !data_unit_size) {

  2059.             if(data_unit_size > buf_size - buf_idx)

  2060.             av_log(s->avctx, AV_LOG_ERROR,

  2061.                    "Data unit with size %d is larger than input buffer, discarding\n",

  2062.                    data_unit_size);

  2063.             buf_idx += 4;

  2064.             continue;

  2065.         }

  2066.         /* [DIRAC_STD] dirac_decode_data_unit makes reference to the while defined in 9.3 inside the function parse_sequence() */

  2067.         ret = dirac_decode_data_unit(avctx, buf+buf_idx, data_unit_size);

  2068.         if (ret < 0)

  2069.         {

  2070.             av_log(s->avctx, AV_LOG_ERROR,"Error in dirac_decode_data_unit\n");

  2071.             return ret;

  2072.         }

  2073.         buf_idx += data_unit_size;

  2074.     }

  2075. 

  2076.     if (!s->current_picture)

  2077.         return buf_size;

  2078. 

  2079.     if (s->current_picture->avframe->display_picture_number > s->frame_number) {

  2080.         DiracFrame *delayed_frame = remove_frame(s->delay_frames, s->frame_number);

  2081. 

  2082.         s->current_picture->reference |= DELAYED_PIC_REF;

  2083. 

  2084.         if (add_frame(s->delay_frames, MAX_DELAY, s->current_picture)) {

  2085.             int min_num = s->delay_frames[0]->avframe->display_picture_number;

  2086.             /* Too many delayed frames, so we display the frame with the lowest pts */

  2087.             av_log(avctx, AV_LOG_ERROR, "Delay frame overflow\n");

  2088. 

  2089.             for (i = 1; s->delay_frames[i]; i++)

  2090.                 if (s->delay_frames[i]->avframe->display_picture_number < min_num)

  2091.                     min_num = s->delay_frames[i]->avframe->display_picture_number;

  2092. 

  2093.             delayed_frame = remove_frame(s->delay_frames, min_num);

  2094.             add_frame(s->delay_frames, MAX_DELAY, s->current_picture);

  2095.         }

  2096. 

  2097.         if (delayed_frame) {

  2098.             delayed_frame->reference ^= DELAYED_PIC_REF;

  2099.             if((ret=av_frame_ref(data, delayed_frame->avframe)) < 0)

  2100.                 return ret;

  2101.             *got_frame = 1;

  2102.         }

  2103.     } else if (s->current_picture->avframe->display_picture_number == s->frame_number) {

  2104.         /* The right frame at the right time :-) */

  2105.         if((ret=av_frame_ref(data, s->current_picture->avframe)) < 0)

  2106.             return ret;

  2107.         *got_frame = 1;

  2108.     }

  2109. 

  2110.     if (*got_frame)

  2111.         s->frame_number = picture->display_picture_number + 1;

  2112. 

  2113.     return buf_idx;

  2114. }

  2115. 

  2116. AVCodec ff_dirac_decoder = {

  2117.     .name           = "dirac",

  2118.     .long_name      = NULL_IF_CONFIG_SMALL("BBC Dirac VC-2"),

  2119.     .type           = AVMEDIA_TYPE_VIDEO,

  2120.     .id             = AV_CODEC_ID_DIRAC,

  2121.     .priv_data_size = sizeof(DiracContext),

  2122.     .init           = dirac_decode_init,

  2123.     .close          = dirac_decode_end,

  2124.     .decode         = dirac_decode_frame,

  2125.     .capabilities   = AV_CODEC_CAP_DELAY,

  2126.     .flush          = dirac_decode_flush,

  2127. };