libavcodec/vp8.h

   1 /*
   2  * VP8 compatible video decoder
   3  *
   4  * Copyright (C) 2010 David Conrad
   5  * Copyright (C) 2010 Ronald S. Bultje
   6  * Copyright (C) 2010 Fiona Glaser
   7  * Copyright (C) 2012 Daniel Kang
   8  *
   9  * This file is part of Libav.
  10  *
  11  * Libav is free software; you can redistribute it and/or
  12  * modify it under the terms of the GNU Lesser General Public
  13  * License as published by the Free Software Foundation; either
  14  * version 2.1 of the License, or (at your option) any later version.
  15  *
  16  * Libav is distributed in the hope that it will be useful,
  17  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  18  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  19  * Lesser General Public License for more details.
  20  *
  21  * You should have received a copy of the GNU Lesser General Public
  22  * License along with Libav; if not, write to the Free Software
  23  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  24  */
  25
  26 #ifndef AVCODEC_VP8_H
  27 #define AVCODEC_VP8_H
  28
  29 #include "libavutil/buffer.h"
  30
  31 #include "h264pred.h"
  32 #include "thread.h"
  33 #include "vp56.h"
  34 #include "vp8dsp.h"
  35
  36 #if HAVE_PTHREADS
  37 #   include <pthread.h>
  38 #elif HAVE_W32THREADS
  39 #   include "compat/w32pthreads.h"
  40 #endif
  41
  42 #define VP8_MAX_QUANT 127
  43
  44 enum dct_token {
  45     DCT_0,
  46     DCT_1,
  47     DCT_2,
  48     DCT_3,
  49     DCT_4,
  50     DCT_CAT1,
  51     DCT_CAT2,
  52     DCT_CAT3,
  53     DCT_CAT4,
  54     DCT_CAT5,
  55     DCT_CAT6,
  56     DCT_EOB,
  57
  58     NUM_DCT_TOKENS
  59 };
  60
  61 // used to signal 4x4 intra pred in luma MBs
  62 #define MODE_I4x4 4
  63
  64 enum inter_mvmode {
  65     VP8_MVMODE_ZERO = MODE_I4x4 + 1,
  66     VP8_MVMODE_MV,
  67     VP8_MVMODE_SPLIT
  68 };
  69
  70 enum inter_splitmvmode {
  71     VP8_SPLITMVMODE_16x8 = 0,    ///< 2 16x8 blocks (vertical)
  72     VP8_SPLITMVMODE_8x16,        ///< 2 8x16 blocks (horizontal)
  73     VP8_SPLITMVMODE_8x8,         ///< 2x2 blocks of 8x8px each
  74     VP8_SPLITMVMODE_4x4,         ///< 4x4 blocks of 4x4px each
  75     VP8_SPLITMVMODE_NONE,        ///< (only used in prediction) no split MVs
  76 };
  77
  78 typedef struct VP8FilterStrength {
  79     uint8_t filter_level;
  80     uint8_t inner_limit;
  81     uint8_t inner_filter;
  82 } VP8FilterStrength;
  83
  84 typedef struct VP8Macroblock {
  85     uint8_t skip;
  86     // TODO: make it possible to check for at least (i4x4 or split_mv)
  87     // in one op. are others needed?
  88     uint8_t mode;
  89     uint8_t ref_frame;
  90     uint8_t partitioning;
  91     uint8_t chroma_pred_mode;
  92     uint8_t segment;
  93     uint8_t intra4x4_pred_mode_mb[16];
  94     DECLARE_ALIGNED(4, uint8_t, intra4x4_pred_mode_top)[4];
  95     VP56mv mv;
  96     VP56mv bmv[16];
  97 } VP8Macroblock;
  98
  99 typedef struct VP8ThreadData {
 100     DECLARE_ALIGNED(16, int16_t, block)[6][4][16];
 101     DECLARE_ALIGNED(16, int16_t, block_dc)[16];
 102     /**
 103      * This is the index plus one of the last non-zero coeff
 104      * for each of the blocks in the current macroblock.
 105      * So, 0 -> no coeffs
 106      *     1 -> dc-only (special transform)
 107      *     2+-> full transform
 108      */
 109     DECLARE_ALIGNED(16, uint8_t, non_zero_count_cache)[6][4];
 110     /**
 111      * For coeff decode, we need to know whether the above block had non-zero
 112      * coefficients. This means for each macroblock, we need data for 4 luma
 113      * blocks, 2 u blocks, 2 v blocks, and the luma dc block, for a total of 9
 114      * per macroblock. We keep the last row in top_nnz.
 115      */
 116     DECLARE_ALIGNED(8, uint8_t, left_nnz)[9];
 117     int thread_nr;
 118 #if HAVE_THREADS
 119     pthread_mutex_t lock;
 120     pthread_cond_t cond;
 121 #endif
 122     int thread_mb_pos; // (mb_y << 16) | (mb_x & 0xFFFF)
 123     int wait_mb_pos; // What the current thread is waiting on.
 124
 125 #define EDGE_EMU_LINESIZE 32
 126     DECLARE_ALIGNED(16, uint8_t, edge_emu_buffer)[21 * EDGE_EMU_LINESIZE];
 127     VP8FilterStrength *filter_strength;
 128 } VP8ThreadData;
 129
 130 typedef struct VP8Frame {
 131     ThreadFrame tf;
 132     AVBufferRef *seg_map;
 133
 134     AVBufferRef *hwaccel_priv_buf;
 135     void *hwaccel_picture_private;
 136 } VP8Frame;
 137
 138 #define MAX_THREADS 8
 139 typedef struct VP8Context {
 140     VP8ThreadData *thread_data;
 141     AVCodecContext *avctx;
 142     enum AVPixelFormat pix_fmt;
 143     int actually_webp;
 144
 145     VP8Frame *framep[4];
 146     VP8Frame *next_framep[4];
 147     VP8Frame *curframe;
 148     VP8Frame *prev_frame;
 149
 150     uint16_t mb_width;   /* number of horizontal MB */
 151     uint16_t mb_height;  /* number of vertical MB */
 152     ptrdiff_t linesize;
 153     ptrdiff_t uvlinesize;
 154
 155     uint8_t keyframe;
 156     uint8_t deblock_filter;
 157     uint8_t mbskip_enabled;
 158     uint8_t profile;
 159     VP56mv mv_min;
 160     VP56mv mv_max;
 161
 162     int8_t sign_bias[4]; ///< one state [0, 1] per ref frame type
 163     int ref_count[3];
 164
 165     /**
 166      * Base parameters for segmentation, i.e. per-macroblock parameters.
 167      * These must be kept unchanged even if segmentation is not used for
 168      * a frame, since the values persist between interframes.
 169      */
 170     struct {
 171         uint8_t enabled;
 172         uint8_t absolute_vals;
 173         uint8_t update_map;
 174         uint8_t update_feature_data;
 175         int8_t base_quant[4];
 176         int8_t filter_level[4];     ///< base loop filter level
 177     } segmentation;
 178
 179     struct {
 180         uint8_t simple;
 181         uint8_t level;
 182         uint8_t sharpness;
 183     } filter;
 184
 185     VP8Macroblock *macroblocks;
 186
 187     uint8_t *intra4x4_pred_mode_top;
 188     uint8_t intra4x4_pred_mode_left[4];
 189
 190     /**
 191      * Macroblocks can have one of 4 different quants in a frame when
 192      * segmentation is enabled.
 193      * If segmentation is disabled, only the first segment's values are used.
 194      */
 195     struct {
 196         // [0] - DC qmul  [1] - AC qmul
 197         int16_t luma_qmul[2];
 198         int16_t luma_dc_qmul[2];    ///< luma dc-only block quant
 199         int16_t chroma_qmul[2];
 200     } qmat[4];
 201
 202     // Raw quantisation values, which may be needed by hwaccel decode.
 203     struct {
 204         int yac_qi;
 205         int ydc_delta;
 206         int y2dc_delta;
 207         int y2ac_delta;
 208         int uvdc_delta;
 209         int uvac_delta;
 210     } quant;
 211
 212     struct {
 213         uint8_t enabled;    ///< whether each mb can have a different strength based on mode/ref
 214         uint8_t update;
 215
 216         /**
 217          * filter strength adjustment for the following macroblock modes:
 218          * [0-3] - i16x16 (always zero)
 219          * [4]   - i4x4
 220          * [5]   - zero mv
 221          * [6]   - inter modes except for zero or split mv
 222          * [7]   - split mv
 223          *  i16x16 modes never have any adjustment
 224          */
 225         int8_t mode[VP8_MVMODE_SPLIT + 1];
 226
 227         /**
 228          * filter strength adjustment for macroblocks that reference:
 229          * [0] - intra / VP56_FRAME_CURRENT
 230          * [1] - VP56_FRAME_PREVIOUS
 231          * [2] - VP56_FRAME_GOLDEN
 232          * [3] - altref / VP56_FRAME_GOLDEN2
 233          */
 234         int8_t ref[4];
 235     } lf_delta;
 236
 237     uint8_t (*top_border)[16 + 8 + 8];
 238     uint8_t (*top_nnz)[9];
 239
 240     VP56RangeCoder c;   ///< header context, includes mb modes and motion vectors
 241
 242     /* This contains the entropy coder state at the end of the header
 243      * block, in the form specified by the standard.  For use by
 244      * hwaccels, so that a hardware decoder has the information to
 245      * start decoding at the macroblock layer.
 246      */
 247     struct {
 248         const uint8_t *input;
 249         uint32_t range;
 250         uint32_t value;
 251         int bit_count;
 252     } coder_state_at_header_end;
 253
 254     int header_partition_size;
 255
 256     /**
 257      * These are all of the updatable probabilities for binary decisions.
 258      * They are only implicitly reset on keyframes, making it quite likely
 259      * for an interframe to desync if a prior frame's header was corrupt
 260      * or missing outright!
 261      */
 262     struct {
 263         uint8_t segmentid[3];
 264         uint8_t mbskip;
 265         uint8_t intra;
 266         uint8_t last;
 267         uint8_t golden;
 268         uint8_t pred16x16[4];
 269         uint8_t pred8x8c[3];
 270         uint8_t token[4][16][3][NUM_DCT_TOKENS - 1];
 271         uint8_t mvc[2][19];
 272         uint8_t scan[16];
 273     } prob[2];
 274
 275     VP8Macroblock *macroblocks_base;
 276     int invisible;
 277     int update_last;    ///< update VP56_FRAME_PREVIOUS with the current one
 278     int update_golden;  ///< VP56_FRAME_NONE if not updated, or which frame to copy if so
 279     int update_altref;
 280
 281     /**
 282      * If this flag is not set, all the probability updates
 283      * are discarded after this frame is decoded.
 284      */
 285     int update_probabilities;
 286
 287     /**
 288      * All coefficients are contained in separate arith coding contexts.
 289      * There can be 1, 2, 4, or 8 of these after the header context.
 290      */
 291     int num_coeff_partitions;
 292     VP56RangeCoder coeff_partition[8];
 293     int coeff_partition_size[8];
 294     VideoDSPContext vdsp;
 295     VP8DSPContext vp8dsp;
 296     H264PredContext hpc;
 297     vp8_mc_func put_pixels_tab[3][3][3];
 298     VP8Frame frames[5];
 299
 300     uint8_t colorspace; ///< 0 is the only value allowed (meaning bt601)
 301     uint8_t fullrange;  ///< whether we can skip clamping in dsp functions
 302
 303     int num_jobs;
 304     /**
 305      * This describes the macroblock memory layout.
 306      * 0 -> Only width+height*2+1 macroblocks allocated (frame/single thread).
 307      * 1 -> Macroblocks for entire frame allocated (sliced thread).
 308      */
 309     int mb_layout;
 310
 311     /**
 312      * Fade bit present in bitstream (VP7)
 313      */
 314     int fade_present;
 315
 316     /**
 317      * Interframe DC prediction (VP7)
 318      * [0] VP56_FRAME_PREVIOUS
 319      * [1] VP56_FRAME_GOLDEN
 320      */
 321     uint16_t inter_dc_pred[2][2];
 322
 323     /**
 324      * Macroblock features (VP7)
 325      */
 326     uint8_t feature_enabled[4];
 327     uint8_t feature_present_prob[4];
 328     uint8_t feature_index_prob[4][3];
 329     uint8_t feature_value[4][4];
 330 } VP8Context;
 331
 332 int ff_vp8_decode_init(AVCodecContext *avctx);
 333
 334 int ff_vp8_decode_frame(AVCodecContext *avctx, void *data, int *got_frame,
 335                         AVPacket *avpkt);
 336
 337 int ff_vp8_decode_free(AVCodecContext *avctx);
 338
 339 #endif /* AVCODEC_VP8_H */