Mercurial > libavcodec.hg
annotate vp56.h @ 12008:98fd80705850 libavcodec
Faster C VP8 normal inner loop filter
author | darkshikari |
---|---|
date | Tue, 29 Jun 2010 04:34:04 +0000 |
parents | f2007d7c3f1d |
children | 934968bd410d |
rev | line source |
---|---|
3695 | 1 /** |
11644
7dd2a45249a9
Remove explicit filename from Doxygen @file commands.
diego
parents:
11369
diff
changeset
|
2 * @file |
3695 | 3 * VP5 and VP6 compatible video decoder (common features) |
4 * | |
5 * Copyright (C) 2006 Aurelien Jacobs <aurel@gnuage.org> | |
6 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3759
diff
changeset
|
7 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3759
diff
changeset
|
8 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3759
diff
changeset
|
9 * FFmpeg is free software; you can redistribute it and/or |
3695 | 10 * modify it under the terms of the GNU Lesser General Public |
11 * License as published by the Free Software Foundation; either | |
12 * version 2.1 of the License, or (at your option) any later version. | |
13 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3759
diff
changeset
|
14 * FFmpeg is distributed in the hope that it will be useful, |
3695 | 15 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
17 * Lesser General Public License for more details. | |
18 * | |
19 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3759
diff
changeset
|
20 * License along with FFmpeg; if not, write to the Free Software |
5215 | 21 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
3695 | 22 */ |
23 | |
7760 | 24 #ifndef AVCODEC_VP56_H |
25 #define AVCODEC_VP56_H | |
3695 | 26 |
27 #include "vp56data.h" | |
28 #include "dsputil.h" | |
9428 | 29 #include "get_bits.h" |
5089 | 30 #include "bytestream.h" |
11665 | 31 #include "vp56dsp.h" |
3695 | 32 |
8299 | 33 typedef struct vp56_context VP56Context; |
34 typedef struct vp56_mv VP56mv; | |
3695 | 35 |
8299 | 36 typedef void (*VP56ParseVectorAdjustment)(VP56Context *s, |
8300 | 37 VP56mv *vect); |
8299 | 38 typedef void (*VP56Filter)(VP56Context *s, uint8_t *dst, uint8_t *src, |
8300 | 39 int offset1, int offset2, int stride, |
40 VP56mv mv, int mask, int select, int luma); | |
8299 | 41 typedef void (*VP56ParseCoeff)(VP56Context *s); |
42 typedef void (*VP56DefaultModelsInit)(VP56Context *s); | |
43 typedef void (*VP56ParseVectorModels)(VP56Context *s); | |
44 typedef void (*VP56ParseCoeffModels)(VP56Context *s); | |
8300 | 45 typedef int (*VP56ParseHeader)(VP56Context *s, const uint8_t *buf, |
46 int buf_size, int *golden_frame); | |
3695 | 47 |
48 typedef struct { | |
49 int high; | |
50 int bits; | |
6297 | 51 const uint8_t *buffer; |
9919
c7c1c6b35a73
vp56dec: ensure range coder won't read past the end of input buffer
aurel
parents:
9428
diff
changeset
|
52 const uint8_t *end; |
3695 | 53 unsigned long code_word; |
8299 | 54 } VP56RangeCoder; |
3695 | 55 |
56 typedef struct { | |
57 uint8_t not_null_dc; | |
8299 | 58 VP56Frame ref_frame; |
3695 | 59 DCTELEM dc_coeff; |
8299 | 60 } VP56RefDc; |
3695 | 61 |
62 struct vp56_mv { | |
63 int x; | |
64 int y; | |
65 }; | |
66 | |
67 typedef struct { | |
68 uint8_t type; | |
8299 | 69 VP56mv mv; |
70 } VP56Macroblock; | |
3695 | 71 |
5711 | 72 typedef struct { |
73 uint8_t coeff_reorder[64]; /* used in vp6 only */ | |
74 uint8_t coeff_index_to_pos[64]; /* used in vp6 only */ | |
75 uint8_t vector_sig[2]; /* delta sign */ | |
76 uint8_t vector_dct[2]; /* delta coding types */ | |
77 uint8_t vector_pdi[2][2]; /* predefined delta init */ | |
78 uint8_t vector_pdv[2][7]; /* predefined delta values */ | |
79 uint8_t vector_fdv[2][8]; /* 8 bit delta value definition */ | |
80 uint8_t coeff_dccv[2][11]; /* DC coeff value */ | |
81 uint8_t coeff_ract[2][3][6][11]; /* Run/AC coding type and AC coeff value */ | |
82 uint8_t coeff_acct[2][3][3][6][5];/* vp5 only AC coding type for coding group < 3 */ | |
83 uint8_t coeff_dcct[2][36][5]; /* DC coeff coding type */ | |
84 uint8_t coeff_runv[2][14]; /* run value (vp6 only) */ | |
85 uint8_t mb_type[3][10][10]; /* model for decoding MB type */ | |
86 uint8_t mb_types_stats[3][10][2];/* contextual, next MB type stats */ | |
8304 | 87 } VP56Model; |
5711 | 88 |
3695 | 89 struct vp56_context { |
90 AVCodecContext *avctx; | |
91 DSPContext dsp; | |
11665 | 92 VP56DSPContext vp56dsp; |
3695 | 93 ScanTable scantable; |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
94 AVFrame frames[4]; |
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
95 AVFrame *framep[6]; |
3695 | 96 uint8_t *edge_emu_buffer_alloc; |
97 uint8_t *edge_emu_buffer; | |
8299 | 98 VP56RangeCoder c; |
99 VP56RangeCoder cc; | |
100 VP56RangeCoder *ccp; | |
4308 | 101 int sub_version; |
3695 | 102 |
103 /* frame info */ | |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
104 int plane_width[4]; |
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
105 int plane_height[4]; |
3695 | 106 int mb_width; /* number of horizontal MB */ |
107 int mb_height; /* number of vertical MB */ | |
108 int block_offset[6]; | |
109 | |
110 int quantizer; | |
111 uint16_t dequant_dc; | |
112 uint16_t dequant_ac; | |
11053
c57e72227d7d
Make VP5 and VP6 decoders output a qscale table to allow for more automatic
reimar
parents:
10961
diff
changeset
|
113 int8_t *qscale_table; |
3695 | 114 |
115 /* DC predictors management */ | |
8299 | 116 VP56RefDc *above_blocks; |
117 VP56RefDc left_block[4]; | |
3695 | 118 int above_block_idx[6]; |
119 DCTELEM prev_dc[3][3]; /* [plan][ref_frame] */ | |
120 | |
121 /* blocks / macroblock */ | |
8299 | 122 VP56mb mb_type; |
123 VP56Macroblock *macroblocks; | |
11369 | 124 DECLARE_ALIGNED(16, DCTELEM, block_coeff)[6][64]; |
3695 | 125 |
126 /* motion vectors */ | |
8299 | 127 VP56mv mv[6]; /* vectors for each block in MB */ |
128 VP56mv vector_candidate[2]; | |
3695 | 129 int vector_candidate_pos; |
130 | |
131 /* filtering hints */ | |
4348 | 132 int filter_header; /* used in vp6 only */ |
3695 | 133 int deblock_filtering; |
134 int filter_selection; | |
135 int filter_mode; | |
136 int max_vector_length; | |
137 int sample_variance_threshold; | |
138 | |
139 uint8_t coeff_ctx[4][64]; /* used in vp5 only */ | |
140 uint8_t coeff_ctx_last[4]; /* used in vp5 only */ | |
141 | |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
142 int has_alpha; |
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
143 |
3695 | 144 /* upside-down flipping hints */ |
145 int flip; /* are we flipping ? */ | |
146 int frbi; /* first row block index in MB */ | |
147 int srbi; /* second row block index in MB */ | |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
148 int stride[4]; /* stride for each plan */ |
3695 | 149 |
150 const uint8_t *vp56_coord_div; | |
8299 | 151 VP56ParseVectorAdjustment parse_vector_adjustment; |
152 VP56Filter filter; | |
153 VP56ParseCoeff parse_coeff; | |
154 VP56DefaultModelsInit default_models_init; | |
155 VP56ParseVectorModels parse_vector_models; | |
156 VP56ParseCoeffModels parse_coeff_models; | |
157 VP56ParseHeader parse_header; | |
5711 | 158 |
8304 | 159 VP56Model *modelp; |
160 VP56Model models[2]; | |
5821 | 161 |
162 /* huffman decoding */ | |
163 int use_huffman; | |
164 GetBitContext gb; | |
165 VLC dccv_vlc[2]; | |
166 VLC runv_vlc[2]; | |
167 VLC ract_vlc[2][3][6]; | |
168 unsigned int nb_null[2][2]; /* number of consecutive NULL DC/AC */ | |
3695 | 169 }; |
170 | |
171 | |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
172 void vp56_init(AVCodecContext *avctx, int flip, int has_alpha); |
3695 | 173 int vp56_free(AVCodecContext *avctx); |
8299 | 174 void vp56_init_dequant(VP56Context *s, int quantizer); |
3695 | 175 int vp56_decode_frame(AVCodecContext *avctx, void *data, int *data_size, |
9356
2983bd7deaf5
fix vp5/vp6 decoding by using new prototype for decode function
aurel
parents:
8718
diff
changeset
|
176 AVPacket *avpkt); |
3695 | 177 |
178 | |
179 /** | |
180 * vp56 specific range coder implementation | |
181 */ | |
182 | |
8299 | 183 static inline void vp56_init_range_decoder(VP56RangeCoder *c, |
6297 | 184 const uint8_t *buf, int buf_size) |
3695 | 185 { |
186 c->high = 255; | |
187 c->bits = 8; | |
188 c->buffer = buf; | |
9919
c7c1c6b35a73
vp56dec: ensure range coder won't read past the end of input buffer
aurel
parents:
9428
diff
changeset
|
189 c->end = buf + buf_size; |
5089 | 190 c->code_word = bytestream_get_be16(&c->buffer); |
3695 | 191 } |
192 | |
8299 | 193 static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob) |
3695 | 194 { |
11920 | 195 unsigned int low = 1 + (((c->high - 1) * prob) >> 8); |
3695 | 196 unsigned int low_shift = low << 8; |
197 int bit = c->code_word >= low_shift; | |
198 | |
199 if (bit) { | |
200 c->high -= low; | |
201 c->code_word -= low_shift; | |
202 } else { | |
203 c->high = low; | |
204 } | |
205 | |
206 /* normalize */ | |
207 while (c->high < 128) { | |
208 c->high <<= 1; | |
209 c->code_word <<= 1; | |
9919
c7c1c6b35a73
vp56dec: ensure range coder won't read past the end of input buffer
aurel
parents:
9428
diff
changeset
|
210 if (--c->bits == 0 && c->buffer < c->end) { |
3695 | 211 c->bits = 8; |
212 c->code_word |= *c->buffer++; | |
213 } | |
214 } | |
215 return bit; | |
216 } | |
217 | |
8299 | 218 static inline int vp56_rac_get(VP56RangeCoder *c) |
3695 | 219 { |
220 /* equiprobable */ | |
221 int low = (c->high + 1) >> 1; | |
222 unsigned int low_shift = low << 8; | |
223 int bit = c->code_word >= low_shift; | |
224 if (bit) { | |
225 c->high = (c->high - low) << 1; | |
226 c->code_word -= low_shift; | |
227 } else { | |
228 c->high = low << 1; | |
229 } | |
230 | |
231 /* normalize */ | |
232 c->code_word <<= 1; | |
9919
c7c1c6b35a73
vp56dec: ensure range coder won't read past the end of input buffer
aurel
parents:
9428
diff
changeset
|
233 if (--c->bits == 0 && c->buffer < c->end) { |
3695 | 234 c->bits = 8; |
235 c->code_word |= *c->buffer++; | |
236 } | |
237 return bit; | |
238 } | |
239 | |
11921 | 240 // rounding is different than vp56_rac_get, is vp56_rac_get wrong? |
241 static inline int vp8_rac_get(VP56RangeCoder *c) | |
242 { | |
243 return vp56_rac_get_prob(c, 128); | |
244 } | |
245 | |
8299 | 246 static inline int vp56_rac_gets(VP56RangeCoder *c, int bits) |
3695 | 247 { |
248 int value = 0; | |
249 | |
250 while (bits--) { | |
251 value = (value << 1) | vp56_rac_get(c); | |
252 } | |
253 | |
254 return value; | |
255 } | |
256 | |
11921 | 257 static inline int vp8_rac_get_uint(VP56RangeCoder *c, int bits) |
258 { | |
259 int value = 0; | |
260 | |
261 while (bits--) { | |
262 value = (value << 1) | vp8_rac_get(c); | |
263 } | |
264 | |
265 return value; | |
266 } | |
267 | |
268 // fixme: add 1 bit to all the calls to this? | |
269 static inline int vp8_rac_get_sint(VP56RangeCoder *c, int bits) | |
270 { | |
271 int v; | |
272 | |
273 if (!vp8_rac_get(c)) | |
274 return 0; | |
275 | |
276 v = vp8_rac_get_uint(c, bits); | |
277 | |
278 if (vp8_rac_get(c)) | |
279 v = -v; | |
280 | |
281 return v; | |
282 } | |
283 | |
284 // P(7) | |
8299 | 285 static inline int vp56_rac_gets_nn(VP56RangeCoder *c, int bits) |
3695 | 286 { |
287 int v = vp56_rac_gets(c, 7) << 1; | |
288 return v + !v; | |
289 } | |
290 | |
11921 | 291 static inline int vp8_rac_get_nn(VP56RangeCoder *c) |
292 { | |
293 int v = vp8_rac_get_uint(c, 7) << 1; | |
294 return v + !v; | |
295 } | |
296 | |
8299 | 297 static inline int vp56_rac_get_tree(VP56RangeCoder *c, |
298 const VP56Tree *tree, | |
3695 | 299 const uint8_t *probs) |
300 { | |
301 while (tree->val > 0) { | |
302 if (vp56_rac_get_prob(c, probs[tree->prob_idx])) | |
303 tree += tree->val; | |
304 else | |
305 tree++; | |
306 } | |
307 return -tree->val; | |
308 } | |
309 | |
11921 | 310 /** |
311 * This is identical to vp8_rac_get_tree except for the possibility of starting | |
312 * on a node other than the root node, needed for coeff decode where this is | |
313 * used to save a bit after a 0 token (by disallowing EOB to immediately follow.) | |
314 */ | |
315 static inline int vp8_rac_get_tree_with_offset(VP56RangeCoder *c, const int8_t (*tree)[2], | |
316 const uint8_t *probs, int i) | |
317 { | |
318 do { | |
319 i = tree[i][vp56_rac_get_prob(c, probs[i])]; | |
320 } while (i > 0); | |
321 | |
322 return -i; | |
323 } | |
324 | |
325 // how probabilities are associated with decisions is different I think | |
326 // well, the new scheme fits in the old but this way has one fewer branches per decision | |
327 static inline int vp8_rac_get_tree(VP56RangeCoder *c, const int8_t (*tree)[2], | |
328 const uint8_t *probs) | |
329 { | |
330 return vp8_rac_get_tree_with_offset(c, tree, probs, 0); | |
331 } | |
332 | |
333 // DCTextra | |
334 static inline int vp8_rac_get_coeff(VP56RangeCoder *c, const uint8_t *prob) | |
335 { | |
336 int v = 0; | |
337 | |
338 do { | |
339 v = (v<<1) + vp56_rac_get_prob(c, *prob++); | |
340 } while (*prob); | |
341 | |
342 return v; | |
343 } | |
344 | |
7760 | 345 #endif /* AVCODEC_VP56_H */ |