vp8: implement sliced threading
[libav.git] / 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 Jason Garrett-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 "vp56.h"
30 #include "vp56data.h"
31 #include "vp8dsp.h"
32 #include "h264pred.h"
33
34 #define VP8_MAX_QUANT 127
35
36 enum dct_token {
37 DCT_0,
38 DCT_1,
39 DCT_2,
40 DCT_3,
41 DCT_4,
42 DCT_CAT1,
43 DCT_CAT2,
44 DCT_CAT3,
45 DCT_CAT4,
46 DCT_CAT5,
47 DCT_CAT6,
48 DCT_EOB,
49
50 NUM_DCT_TOKENS
51 };
52
53 // used to signal 4x4 intra pred in luma MBs
54 #define MODE_I4x4 4
55
56 enum inter_mvmode {
57 VP8_MVMODE_ZERO = MODE_I4x4 + 1,
58 VP8_MVMODE_MV,
59 VP8_MVMODE_SPLIT
60 };
61
62 enum inter_splitmvmode {
63 VP8_SPLITMVMODE_16x8 = 0, ///< 2 16x8 blocks (vertical)
64 VP8_SPLITMVMODE_8x16, ///< 2 8x16 blocks (horizontal)
65 VP8_SPLITMVMODE_8x8, ///< 2x2 blocks of 8x8px each
66 VP8_SPLITMVMODE_4x4, ///< 4x4 blocks of 4x4px each
67 VP8_SPLITMVMODE_NONE, ///< (only used in prediction) no split MVs
68 };
69
70 typedef struct {
71 uint8_t filter_level;
72 uint8_t inner_limit;
73 uint8_t inner_filter;
74 } VP8FilterStrength;
75
76 typedef struct {
77 uint8_t skip;
78 // todo: make it possible to check for at least (i4x4 or split_mv)
79 // in one op. are others needed?
80 uint8_t mode;
81 uint8_t ref_frame;
82 uint8_t partitioning;
83 uint8_t chroma_pred_mode;
84 uint8_t segment;
85 uint8_t intra4x4_pred_mode_mb[16];
86 uint8_t intra4x4_pred_mode_top[4];
87 VP56mv mv;
88 VP56mv bmv[16];
89 } VP8Macroblock;
90
91 typedef struct {
92 pthread_mutex_t lock;
93 pthread_cond_t cond;
94 int thread_nr;
95 int thread_mb_pos; // (mb_y << 16) | (mb_x & 0xFFFF)
96 int wait_mb_pos; // What the current thread is waiting on.
97 uint8_t *edge_emu_buffer;
98 /**
99 * For coeff decode, we need to know whether the above block had non-zero
100 * coefficients. This means for each macroblock, we need data for 4 luma
101 * blocks, 2 u blocks, 2 v blocks, and the luma dc block, for a total of 9
102 * per macroblock. We keep the last row in top_nnz.
103 */
104 DECLARE_ALIGNED(8, uint8_t, left_nnz)[9];
105 /**
106 * This is the index plus one of the last non-zero coeff
107 * for each of the blocks in the current macroblock.
108 * So, 0 -> no coeffs
109 * 1 -> dc-only (special transform)
110 * 2+-> full transform
111 */
112 DECLARE_ALIGNED(16, uint8_t, non_zero_count_cache)[6][4];
113 DECLARE_ALIGNED(16, DCTELEM, block)[6][4][16];
114 DECLARE_ALIGNED(16, DCTELEM, block_dc)[16];
115 VP8FilterStrength *filter_strength;
116 } VP8ThreadData;
117
118 #define MAX_THREADS 8
119 typedef struct {
120 VP8ThreadData *thread_data;
121 AVCodecContext *avctx;
122 AVFrame *framep[4];
123 AVFrame *next_framep[4];
124 AVFrame *curframe;
125 AVFrame *prev_frame;
126
127 uint16_t mb_width; /* number of horizontal MB */
128 uint16_t mb_height; /* number of vertical MB */
129 int linesize;
130 int uvlinesize;
131
132 uint8_t keyframe;
133 uint8_t deblock_filter;
134 uint8_t mbskip_enabled;
135 uint8_t profile;
136 VP56mv mv_min;
137 VP56mv mv_max;
138
139 int8_t sign_bias[4]; ///< one state [0, 1] per ref frame type
140 int ref_count[3];
141
142 /**
143 * Base parameters for segmentation, i.e. per-macroblock parameters.
144 * These must be kept unchanged even if segmentation is not used for
145 * a frame, since the values persist between interframes.
146 */
147 struct {
148 uint8_t enabled;
149 uint8_t absolute_vals;
150 uint8_t update_map;
151 int8_t base_quant[4];
152 int8_t filter_level[4]; ///< base loop filter level
153 } segmentation;
154
155 struct {
156 uint8_t simple;
157 uint8_t level;
158 uint8_t sharpness;
159 } filter;
160
161 VP8Macroblock *macroblocks;
162
163 uint8_t *intra4x4_pred_mode_top;
164 uint8_t intra4x4_pred_mode_left[4];
165
166 /**
167 * Macroblocks can have one of 4 different quants in a frame when
168 * segmentation is enabled.
169 * If segmentation is disabled, only the first segment's values are used.
170 */
171 struct {
172 // [0] - DC qmul [1] - AC qmul
173 int16_t luma_qmul[2];
174 int16_t luma_dc_qmul[2]; ///< luma dc-only block quant
175 int16_t chroma_qmul[2];
176 } qmat[4];
177
178 struct {
179 uint8_t enabled; ///< whether each mb can have a different strength based on mode/ref
180
181 /**
182 * filter strength adjustment for the following macroblock modes:
183 * [0-3] - i16x16 (always zero)
184 * [4] - i4x4
185 * [5] - zero mv
186 * [6] - inter modes except for zero or split mv
187 * [7] - split mv
188 * i16x16 modes never have any adjustment
189 */
190 int8_t mode[VP8_MVMODE_SPLIT+1];
191
192 /**
193 * filter strength adjustment for macroblocks that reference:
194 * [0] - intra / VP56_FRAME_CURRENT
195 * [1] - VP56_FRAME_PREVIOUS
196 * [2] - VP56_FRAME_GOLDEN
197 * [3] - altref / VP56_FRAME_GOLDEN2
198 */
199 int8_t ref[4];
200 } lf_delta;
201
202 uint8_t (*top_border)[16+8+8];
203 uint8_t (*top_nnz)[9];
204
205 VP56RangeCoder c; ///< header context, includes mb modes and motion vectors
206
207 /**
208 * These are all of the updatable probabilities for binary decisions.
209 * They are only implictly reset on keyframes, making it quite likely
210 * for an interframe to desync if a prior frame's header was corrupt
211 * or missing outright!
212 */
213 struct {
214 uint8_t segmentid[3];
215 uint8_t mbskip;
216 uint8_t intra;
217 uint8_t last;
218 uint8_t golden;
219 uint8_t pred16x16[4];
220 uint8_t pred8x8c[3];
221 uint8_t token[4][16][3][NUM_DCT_TOKENS-1];
222 uint8_t mvc[2][19];
223 } prob[2];
224
225 VP8Macroblock *macroblocks_base;
226 int invisible;
227 int update_last; ///< update VP56_FRAME_PREVIOUS with the current one
228 int update_golden; ///< VP56_FRAME_NONE if not updated, or which frame to copy if so
229 int update_altref;
230
231 /**
232 * If this flag is not set, all the probability updates
233 * are discarded after this frame is decoded.
234 */
235 int update_probabilities;
236
237 /**
238 * All coefficients are contained in separate arith coding contexts.
239 * There can be 1, 2, 4, or 8 of these after the header context.
240 */
241 int num_coeff_partitions;
242 VP56RangeCoder coeff_partition[8];
243 DSPContext dsp;
244 VP8DSPContext vp8dsp;
245 H264PredContext hpc;
246 vp8_mc_func put_pixels_tab[3][3][3];
247 AVFrame frames[5];
248
249 /**
250 * A list of segmentation_map buffers that are to be free()'ed in
251 * the next decoding iteration. We can't free() them right away
252 * because the map may still be used by subsequent decoding threads.
253 * Unused if frame threading is off.
254 */
255 uint8_t *segmentation_maps[5];
256 int num_maps_to_be_freed;
257 int maps_are_invalid;
258 int num_jobs;
259 /**
260 * This describes the macroblock memory layout.
261 * 0 -> Only width+height*2+1 macroblocks allocated (frame/single thread).
262 * 1 -> Macroblocks for entire frame alloced (sliced thread).
263 */
264 int mb_layout;
265 } VP8Context;
266
267 #endif /* AVCODEC_VP8_H */