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 <stdatomic.h> 30 31 #include "libavutil/buffer.h" 32 #include "libavutil/mem_internal.h" 33 #include "libavutil/thread.h" 34 35 #include "h264pred.h" 36 #include "thread.h" 37 #include "vp56.h" 38 #include "vp8dsp.h" 39 40 #define VP8_MAX_QUANT 127 41 42 enum dct_token { 43 DCT_0, 44 DCT_1, 45 DCT_2, 46 DCT_3, 47 DCT_4, 48 DCT_CAT1, 49 DCT_CAT2, 50 DCT_CAT3, 51 DCT_CAT4, 52 DCT_CAT5, 53 DCT_CAT6, 54 DCT_EOB, 55 56 NUM_DCT_TOKENS 57 }; 58 59 // used to signal 4x4 intra pred in luma MBs 60 #define MODE_I4x4 4 61 62 enum inter_mvmode { 63 VP8_MVMODE_ZERO = MODE_I4x4 + 1, 64 VP8_MVMODE_MV, 65 VP8_MVMODE_SPLIT 66 }; 67 68 enum inter_splitmvmode { 69 VP8_SPLITMVMODE_16x8 = 0, ///< 2 16x8 blocks (vertical) 70 VP8_SPLITMVMODE_8x16, ///< 2 8x16 blocks (horizontal) 71 VP8_SPLITMVMODE_8x8, ///< 2x2 blocks of 8x8px each 72 VP8_SPLITMVMODE_4x4, ///< 4x4 blocks of 4x4px each 73 VP8_SPLITMVMODE_NONE, ///< (only used in prediction) no split MVs 74 }; 75 76 typedef struct VP8FilterStrength { 77 uint8_t filter_level; 78 uint8_t inner_limit; 79 uint8_t inner_filter; 80 } VP8FilterStrength; 81 82 typedef struct VP8Macroblock { 83 uint8_t skip; 84 // TODO: make it possible to check for at least (i4x4 or split_mv) 85 // in one op. are others needed? 86 uint8_t mode; 87 uint8_t ref_frame; 88 uint8_t partitioning; 89 uint8_t chroma_pred_mode; 90 uint8_t segment; 91 uint8_t intra4x4_pred_mode_mb[16]; 92 DECLARE_ALIGNED(4, uint8_t, intra4x4_pred_mode_top)[4]; 93 VP56mv mv; 94 VP56mv bmv[16]; 95 } VP8Macroblock; 96 97 typedef struct VP8intmv { 98 int x; 99 int y; 100 } VP8intmv; 101 102 typedef struct VP8mvbounds { 103 VP8intmv mv_min; 104 VP8intmv mv_max; 105 } VP8mvbounds; 106 107 typedef struct VP8ThreadData { 108 DECLARE_ALIGNED(16, int16_t, block)[6][4][16]; 109 DECLARE_ALIGNED(16, int16_t, block_dc)[16]; 110 /** 111 * This is the index plus one of the last non-zero coeff 112 * for each of the blocks in the current macroblock. 113 * So, 0 -> no coeffs 114 * 1 -> dc-only (special transform) 115 * 2+-> full transform 116 */ 117 DECLARE_ALIGNED(16, uint8_t, non_zero_count_cache)[6][4]; 118 /** 119 * For coeff decode, we need to know whether the above block had non-zero 120 * coefficients. This means for each macroblock, we need data for 4 luma 121 * blocks, 2 u blocks, 2 v blocks, and the luma dc block, for a total of 9 122 * per macroblock. We keep the last row in top_nnz. 123 */ 124 DECLARE_ALIGNED(8, uint8_t, left_nnz)[9]; 125 int thread_nr; 126 #if HAVE_THREADS 127 pthread_mutex_t lock; 128 pthread_cond_t cond; 129 #endif 130 atomic_int thread_mb_pos; // (mb_y << 16) | (mb_x & 0xFFFF) 131 atomic_int wait_mb_pos; // What the current thread is waiting on. 132 133 #define EDGE_EMU_LINESIZE 32 134 DECLARE_ALIGNED(16, uint8_t, edge_emu_buffer)[21 * EDGE_EMU_LINESIZE]; 135 VP8FilterStrength *filter_strength; 136 VP8mvbounds mv_bounds; 137 } VP8ThreadData; 138 139 typedef struct VP8Frame { 140 ThreadFrame tf; 141 AVBufferRef *seg_map; 142 143 AVBufferRef *hwaccel_priv_buf; 144 void *hwaccel_picture_private; 145 } VP8Frame; 146 147 #define MAX_THREADS 8 148 typedef struct VP8Context { 149 VP8ThreadData *thread_data; 150 AVCodecContext *avctx; 151 enum AVPixelFormat pix_fmt; 152 int actually_webp; 153 154 VP8Frame *framep[4]; 155 VP8Frame *next_framep[4]; 156 VP8Frame *curframe; 157 VP8Frame *prev_frame; 158 159 uint16_t mb_width; /* number of horizontal MB */ 160 uint16_t mb_height; /* number of vertical MB */ 161 ptrdiff_t linesize; 162 ptrdiff_t uvlinesize; 163 164 uint8_t keyframe; 165 uint8_t deblock_filter; 166 uint8_t mbskip_enabled; 167 uint8_t profile; 168 VP8mvbounds mv_bounds; 169 170 int8_t sign_bias[4]; ///< one state [0, 1] per ref frame type 171 int ref_count[3]; 172 173 /** 174 * Base parameters for segmentation, i.e. per-macroblock parameters. 175 * These must be kept unchanged even if segmentation is not used for 176 * a frame, since the values persist between interframes. 177 */ 178 struct { 179 uint8_t enabled; 180 uint8_t absolute_vals; 181 uint8_t update_map; 182 uint8_t update_feature_data; 183 int8_t base_quant[4]; 184 int8_t filter_level[4]; ///< base loop filter level 185 } segmentation; 186 187 struct { 188 uint8_t simple; 189 uint8_t level; 190 uint8_t sharpness; 191 } filter; 192 193 VP8Macroblock *macroblocks; 194 195 uint8_t *intra4x4_pred_mode_top; 196 uint8_t intra4x4_pred_mode_left[4]; 197 198 /** 199 * Macroblocks can have one of 4 different quants in a frame when 200 * segmentation is enabled. 201 * If segmentation is disabled, only the first segment's values are used. 202 */ 203 struct { 204 // [0] - DC qmul [1] - AC qmul 205 int16_t luma_qmul[2]; 206 int16_t luma_dc_qmul[2]; ///< luma dc-only block quant 207 int16_t chroma_qmul[2]; 208 } qmat[4]; 209 210 // Raw quantisation values, which may be needed by hwaccel decode. 211 struct { 212 int yac_qi; 213 int ydc_delta; 214 int y2dc_delta; 215 int y2ac_delta; 216 int uvdc_delta; 217 int uvac_delta; 218 } quant; 219 220 struct { 221 uint8_t enabled; ///< whether each mb can have a different strength based on mode/ref 222 uint8_t update; 223 224 /** 225 * filter strength adjustment for the following macroblock modes: 226 * [0-3] - i16x16 (always zero) 227 * [4] - i4x4 228 * [5] - zero mv 229 * [6] - inter modes except for zero or split mv 230 * [7] - split mv 231 * i16x16 modes never have any adjustment 232 */ 233 int8_t mode[VP8_MVMODE_SPLIT + 1]; 234 235 /** 236 * filter strength adjustment for macroblocks that reference: 237 * [0] - intra / VP56_FRAME_CURRENT 238 * [1] - VP56_FRAME_PREVIOUS 239 * [2] - VP56_FRAME_GOLDEN 240 * [3] - altref / VP56_FRAME_GOLDEN2 241 */ 242 int8_t ref[4]; 243 } lf_delta; 244 245 uint8_t (*top_border)[16 + 8 + 8]; 246 uint8_t (*top_nnz)[9]; 247 248 VP56RangeCoder c; ///< header context, includes mb modes and motion vectors 249 250 /* This contains the entropy coder state at the end of the header 251 * block, in the form specified by the standard. For use by 252 * hwaccels, so that a hardware decoder has the information to 253 * start decoding at the macroblock layer. 254 */ 255 struct { 256 const uint8_t *input; 257 uint32_t range; 258 uint32_t value; 259 int bit_count; 260 } coder_state_at_header_end; 261 262 int header_partition_size; 263 264 /** 265 * These are all of the updatable probabilities for binary decisions. 266 * They are only implicitly reset on keyframes, making it quite likely 267 * for an interframe to desync if a prior frame's header was corrupt 268 * or missing outright! 269 */ 270 struct { 271 uint8_t segmentid[3]; 272 uint8_t mbskip; 273 uint8_t intra; 274 uint8_t last; 275 uint8_t golden; 276 uint8_t pred16x16[4]; 277 uint8_t pred8x8c[3]; 278 uint8_t token[4][16][3][NUM_DCT_TOKENS - 1]; 279 uint8_t mvc[2][19]; 280 uint8_t scan[16]; 281 } prob[2]; 282 283 VP8Macroblock *macroblocks_base; 284 int invisible; 285 int update_last; ///< update VP56_FRAME_PREVIOUS with the current one 286 int update_golden; ///< VP56_FRAME_NONE if not updated, or which frame to copy if so 287 int update_altref; 288 289 /** 290 * If this flag is not set, all the probability updates 291 * are discarded after this frame is decoded. 292 */ 293 int update_probabilities; 294 295 /** 296 * All coefficients are contained in separate arith coding contexts. 297 * There can be 1, 2, 4, or 8 of these after the header context. 298 */ 299 int num_coeff_partitions; 300 VP56RangeCoder coeff_partition[8]; 301 int coeff_partition_size[8]; 302 VideoDSPContext vdsp; 303 VP8DSPContext vp8dsp; 304 H264PredContext hpc; 305 vp8_mc_func put_pixels_tab[3][3][3]; 306 VP8Frame frames[5]; 307 308 uint8_t colorspace; ///< 0 is the only value allowed (meaning bt601) 309 uint8_t fullrange; ///< whether we can skip clamping in dsp functions 310 311 int num_jobs; 312 /** 313 * This describes the macroblock memory layout. 314 * 0 -> Only width+height*2+1 macroblocks allocated (frame/single thread). 315 * 1 -> Macroblocks for entire frame allocated (sliced thread). 316 */ 317 int mb_layout; 318 319 int (*decode_mb_row_no_filter)(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr); 320 void (*filter_mb_row)(AVCodecContext *avctx, void *tdata, int jobnr, int threadnr); 321 322 int vp7; 323 324 /** 325 * Fade bit present in bitstream (VP7) 326 */ 327 int fade_present; 328 329 /** 330 * Interframe DC prediction (VP7) 331 * [0] VP56_FRAME_PREVIOUS 332 * [1] VP56_FRAME_GOLDEN 333 */ 334 uint16_t inter_dc_pred[2][2]; 335 336 /** 337 * Macroblock features (VP7) 338 */ 339 uint8_t feature_enabled[4]; 340 uint8_t feature_present_prob[4]; 341 uint8_t feature_index_prob[4][3]; 342 uint8_t feature_value[4][4]; 343 } VP8Context; 344 345 int ff_vp8_decode_init(AVCodecContext *avctx); 346 347 int ff_vp8_decode_frame(AVCodecContext *avctx, void *data, int *got_frame, 348 AVPacket *avpkt); 349 350 int ff_vp8_decode_free(AVCodecContext *avctx); 351 352 #endif /* AVCODEC_VP8_H */ 353