| 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 FFmpeg. |
| 10 | * |
| 11 | * FFmpeg 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 | * FFmpeg 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 FFmpeg; 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_OS2THREADS |
| 39 | # include "compat/os2threads.h" |
| 40 | #elif HAVE_W32THREADS |
| 41 | # include "compat/w32pthreads.h" |
| 42 | #endif |
| 43 | |
| 44 | #define VP8_MAX_QUANT 127 |
| 45 | |
| 46 | enum dct_token { |
| 47 | DCT_0, |
| 48 | DCT_1, |
| 49 | DCT_2, |
| 50 | DCT_3, |
| 51 | DCT_4, |
| 52 | DCT_CAT1, |
| 53 | DCT_CAT2, |
| 54 | DCT_CAT3, |
| 55 | DCT_CAT4, |
| 56 | DCT_CAT5, |
| 57 | DCT_CAT6, |
| 58 | DCT_EOB, |
| 59 | |
| 60 | NUM_DCT_TOKENS |
| 61 | }; |
| 62 | |
| 63 | // used to signal 4x4 intra pred in luma MBs |
| 64 | #define MODE_I4x4 4 |
| 65 | |
| 66 | enum inter_mvmode { |
| 67 | VP8_MVMODE_ZERO = MODE_I4x4 + 1, |
| 68 | VP8_MVMODE_MV, |
| 69 | VP8_MVMODE_SPLIT |
| 70 | }; |
| 71 | |
| 72 | enum inter_splitmvmode { |
| 73 | VP8_SPLITMVMODE_16x8 = 0, ///< 2 16x8 blocks (vertical) |
| 74 | VP8_SPLITMVMODE_8x16, ///< 2 8x16 blocks (horizontal) |
| 75 | VP8_SPLITMVMODE_8x8, ///< 2x2 blocks of 8x8px each |
| 76 | VP8_SPLITMVMODE_4x4, ///< 4x4 blocks of 4x4px each |
| 77 | VP8_SPLITMVMODE_NONE, ///< (only used in prediction) no split MVs |
| 78 | }; |
| 79 | |
| 80 | typedef struct VP8FilterStrength { |
| 81 | uint8_t filter_level; |
| 82 | uint8_t inner_limit; |
| 83 | uint8_t inner_filter; |
| 84 | } VP8FilterStrength; |
| 85 | |
| 86 | typedef struct VP8Macroblock { |
| 87 | uint8_t skip; |
| 88 | // TODO: make it possible to check for at least (i4x4 or split_mv) |
| 89 | // in one op. are others needed? |
| 90 | uint8_t mode; |
| 91 | uint8_t ref_frame; |
| 92 | uint8_t partitioning; |
| 93 | uint8_t chroma_pred_mode; |
| 94 | uint8_t segment; |
| 95 | uint8_t intra4x4_pred_mode_mb[16]; |
| 96 | DECLARE_ALIGNED(4, uint8_t, intra4x4_pred_mode_top)[4]; |
| 97 | VP56mv mv; |
| 98 | VP56mv bmv[16]; |
| 99 | } VP8Macroblock; |
| 100 | |
| 101 | typedef struct VP8ThreadData { |
| 102 | DECLARE_ALIGNED(16, int16_t, block)[6][4][16]; |
| 103 | DECLARE_ALIGNED(16, int16_t, block_dc)[16]; |
| 104 | /** |
| 105 | * This is the index plus one of the last non-zero coeff |
| 106 | * for each of the blocks in the current macroblock. |
| 107 | * So, 0 -> no coeffs |
| 108 | * 1 -> dc-only (special transform) |
| 109 | * 2+-> full transform |
| 110 | */ |
| 111 | DECLARE_ALIGNED(16, uint8_t, non_zero_count_cache)[6][4]; |
| 112 | /** |
| 113 | * For coeff decode, we need to know whether the above block had non-zero |
| 114 | * coefficients. This means for each macroblock, we need data for 4 luma |
| 115 | * blocks, 2 u blocks, 2 v blocks, and the luma dc block, for a total of 9 |
| 116 | * per macroblock. We keep the last row in top_nnz. |
| 117 | */ |
| 118 | DECLARE_ALIGNED(8, uint8_t, left_nnz)[9]; |
| 119 | int thread_nr; |
| 120 | #if HAVE_THREADS |
| 121 | pthread_mutex_t lock; |
| 122 | pthread_cond_t cond; |
| 123 | #endif |
| 124 | int thread_mb_pos; // (mb_y << 16) | (mb_x & 0xFFFF) |
| 125 | int wait_mb_pos; // What the current thread is waiting on. |
| 126 | |
| 127 | #define EDGE_EMU_LINESIZE 32 |
| 128 | DECLARE_ALIGNED(16, uint8_t, edge_emu_buffer)[21 * EDGE_EMU_LINESIZE]; |
| 129 | VP8FilterStrength *filter_strength; |
| 130 | } VP8ThreadData; |
| 131 | |
| 132 | typedef struct VP8Frame { |
| 133 | ThreadFrame tf; |
| 134 | AVBufferRef *seg_map; |
| 135 | } VP8Frame; |
| 136 | |
| 137 | #define MAX_THREADS 8 |
| 138 | typedef struct VP8Context { |
| 139 | VP8ThreadData *thread_data; |
| 140 | AVCodecContext *avctx; |
| 141 | VP8Frame *framep[4]; |
| 142 | VP8Frame *next_framep[4]; |
| 143 | VP8Frame *curframe; |
| 144 | VP8Frame *prev_frame; |
| 145 | |
| 146 | uint16_t mb_width; /* number of horizontal MB */ |
| 147 | uint16_t mb_height; /* number of vertical MB */ |
| 148 | int linesize; |
| 149 | int uvlinesize; |
| 150 | |
| 151 | uint8_t keyframe; |
| 152 | uint8_t deblock_filter; |
| 153 | uint8_t mbskip_enabled; |
| 154 | uint8_t profile; |
| 155 | VP56mv mv_min; |
| 156 | VP56mv mv_max; |
| 157 | |
| 158 | int8_t sign_bias[4]; ///< one state [0, 1] per ref frame type |
| 159 | int ref_count[3]; |
| 160 | |
| 161 | /** |
| 162 | * Base parameters for segmentation, i.e. per-macroblock parameters. |
| 163 | * These must be kept unchanged even if segmentation is not used for |
| 164 | * a frame, since the values persist between interframes. |
| 165 | */ |
| 166 | struct { |
| 167 | uint8_t enabled; |
| 168 | uint8_t absolute_vals; |
| 169 | uint8_t update_map; |
| 170 | int8_t base_quant[4]; |
| 171 | int8_t filter_level[4]; ///< base loop filter level |
| 172 | } segmentation; |
| 173 | |
| 174 | struct { |
| 175 | uint8_t simple; |
| 176 | uint8_t level; |
| 177 | uint8_t sharpness; |
| 178 | } filter; |
| 179 | |
| 180 | VP8Macroblock *macroblocks; |
| 181 | |
| 182 | uint8_t *intra4x4_pred_mode_top; |
| 183 | uint8_t intra4x4_pred_mode_left[4]; |
| 184 | |
| 185 | /** |
| 186 | * Macroblocks can have one of 4 different quants in a frame when |
| 187 | * segmentation is enabled. |
| 188 | * If segmentation is disabled, only the first segment's values are used. |
| 189 | */ |
| 190 | struct { |
| 191 | // [0] - DC qmul [1] - AC qmul |
| 192 | int16_t luma_qmul[2]; |
| 193 | int16_t luma_dc_qmul[2]; ///< luma dc-only block quant |
| 194 | int16_t chroma_qmul[2]; |
| 195 | } qmat[4]; |
| 196 | |
| 197 | struct { |
| 198 | uint8_t enabled; ///< whether each mb can have a different strength based on mode/ref |
| 199 | |
| 200 | /** |
| 201 | * filter strength adjustment for the following macroblock modes: |
| 202 | * [0-3] - i16x16 (always zero) |
| 203 | * [4] - i4x4 |
| 204 | * [5] - zero mv |
| 205 | * [6] - inter modes except for zero or split mv |
| 206 | * [7] - split mv |
| 207 | * i16x16 modes never have any adjustment |
| 208 | */ |
| 209 | int8_t mode[VP8_MVMODE_SPLIT + 1]; |
| 210 | |
| 211 | /** |
| 212 | * filter strength adjustment for macroblocks that reference: |
| 213 | * [0] - intra / VP56_FRAME_CURRENT |
| 214 | * [1] - VP56_FRAME_PREVIOUS |
| 215 | * [2] - VP56_FRAME_GOLDEN |
| 216 | * [3] - altref / VP56_FRAME_GOLDEN2 |
| 217 | */ |
| 218 | int8_t ref[4]; |
| 219 | } lf_delta; |
| 220 | |
| 221 | uint8_t (*top_border)[16 + 8 + 8]; |
| 222 | uint8_t (*top_nnz)[9]; |
| 223 | |
| 224 | VP56RangeCoder c; ///< header context, includes mb modes and motion vectors |
| 225 | |
| 226 | /** |
| 227 | * These are all of the updatable probabilities for binary decisions. |
| 228 | * They are only implictly reset on keyframes, making it quite likely |
| 229 | * for an interframe to desync if a prior frame's header was corrupt |
| 230 | * or missing outright! |
| 231 | */ |
| 232 | struct { |
| 233 | uint8_t segmentid[3]; |
| 234 | uint8_t mbskip; |
| 235 | uint8_t intra; |
| 236 | uint8_t last; |
| 237 | uint8_t golden; |
| 238 | uint8_t pred16x16[4]; |
| 239 | uint8_t pred8x8c[3]; |
| 240 | uint8_t token[4][16][3][NUM_DCT_TOKENS - 1]; |
| 241 | uint8_t mvc[2][19]; |
| 242 | uint8_t scan[16]; |
| 243 | } prob[2]; |
| 244 | |
| 245 | VP8Macroblock *macroblocks_base; |
| 246 | int invisible; |
| 247 | int update_last; ///< update VP56_FRAME_PREVIOUS with the current one |
| 248 | int update_golden; ///< VP56_FRAME_NONE if not updated, or which frame to copy if so |
| 249 | int update_altref; |
| 250 | |
| 251 | /** |
| 252 | * If this flag is not set, all the probability updates |
| 253 | * are discarded after this frame is decoded. |
| 254 | */ |
| 255 | int update_probabilities; |
| 256 | |
| 257 | /** |
| 258 | * All coefficients are contained in separate arith coding contexts. |
| 259 | * There can be 1, 2, 4, or 8 of these after the header context. |
| 260 | */ |
| 261 | int num_coeff_partitions; |
| 262 | VP56RangeCoder coeff_partition[8]; |
| 263 | VideoDSPContext vdsp; |
| 264 | VP8DSPContext vp8dsp; |
| 265 | H264PredContext hpc; |
| 266 | vp8_mc_func put_pixels_tab[3][3][3]; |
| 267 | VP8Frame frames[5]; |
| 268 | |
| 269 | int num_jobs; |
| 270 | /** |
| 271 | * This describes the macroblock memory layout. |
| 272 | * 0 -> Only width+height*2+1 macroblocks allocated (frame/single thread). |
| 273 | * 1 -> Macroblocks for entire frame alloced (sliced thread). |
| 274 | */ |
| 275 | int mb_layout; |
| 276 | |
| 277 | void (*decode_mb_row_no_filter)(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr); |
| 278 | void (*filter_mb_row)(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr); |
| 279 | |
| 280 | int vp7; |
| 281 | |
| 282 | /** |
| 283 | * Fade bit present in bitstream (VP7) |
| 284 | */ |
| 285 | int fade_present; |
| 286 | |
| 287 | /** |
| 288 | * Interframe DC prediction (VP7) |
| 289 | * [0] VP56_FRAME_PREVIOUS |
| 290 | * [1] VP56_FRAME_GOLDEN |
| 291 | */ |
| 292 | uint16_t inter_dc_pred[2][2]; |
| 293 | |
| 294 | /** |
| 295 | * Macroblock features (VP7) |
| 296 | */ |
| 297 | uint8_t feature_enabled[4]; |
| 298 | uint8_t feature_present_prob[4]; |
| 299 | uint8_t feature_index_prob[4][3]; |
| 300 | uint8_t feature_value[4][4]; |
| 301 | } VP8Context; |
| 302 | |
| 303 | int ff_vp8_decode_init(AVCodecContext *avctx); |
| 304 | |
| 305 | int ff_vp8_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, |
| 306 | AVPacket *avpkt); |
| 307 | |
| 308 | int ff_vp8_decode_free(AVCodecContext *avctx); |
| 309 | |
| 310 | #endif /* AVCODEC_VP8_H */ |