Mercurial > libavcodec.hg
annotate vp56.h @ 12032:572c81b3be19 libavcodec
CMOV-ify vp56 arithcoder
This incantation causes gcc 4.3 to generate cmov on x86, a vastly better option
than a completely unpredictable branch.
Hopefully this carries over to newer versions and other CPUs with conditionals.
~5 cycles saved per call on a Core i7.
author | darkshikari |
---|---|
date | Wed, 30 Jun 2010 23:18:47 +0000 |
parents | 5578dcdf030c |
children | 5de2b84a1fc3 |
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" |
12029 | 31 #include "cabac.h" |
11665 | 32 #include "vp56dsp.h" |
3695 | 33 |
8299 | 34 typedef struct vp56_context VP56Context; |
35 typedef struct vp56_mv VP56mv; | |
3695 | 36 |
8299 | 37 typedef void (*VP56ParseVectorAdjustment)(VP56Context *s, |
8300 | 38 VP56mv *vect); |
8299 | 39 typedef void (*VP56Filter)(VP56Context *s, uint8_t *dst, uint8_t *src, |
8300 | 40 int offset1, int offset2, int stride, |
41 VP56mv mv, int mask, int select, int luma); | |
8299 | 42 typedef void (*VP56ParseCoeff)(VP56Context *s); |
43 typedef void (*VP56DefaultModelsInit)(VP56Context *s); | |
44 typedef void (*VP56ParseVectorModels)(VP56Context *s); | |
45 typedef void (*VP56ParseCoeffModels)(VP56Context *s); | |
8300 | 46 typedef int (*VP56ParseHeader)(VP56Context *s, const uint8_t *buf, |
47 int buf_size, int *golden_frame); | |
3695 | 48 |
49 typedef struct { | |
50 int high; | |
12031 | 51 int bits; /* Stored negated (i.e. negative "bits" is a positive number of bits left) |
52 * in order to eliminate a negate in cache refilling */ | |
6297 | 53 const uint8_t *buffer; |
9919
c7c1c6b35a73
vp56dec: ensure range coder won't read past the end of input buffer
aurel
parents:
9428
diff
changeset
|
54 const uint8_t *end; |
3695 | 55 unsigned long code_word; |
8299 | 56 } VP56RangeCoder; |
3695 | 57 |
58 typedef struct { | |
59 uint8_t not_null_dc; | |
8299 | 60 VP56Frame ref_frame; |
3695 | 61 DCTELEM dc_coeff; |
8299 | 62 } VP56RefDc; |
3695 | 63 |
64 struct vp56_mv { | |
65 int x; | |
66 int y; | |
67 }; | |
68 | |
69 typedef struct { | |
70 uint8_t type; | |
8299 | 71 VP56mv mv; |
72 } VP56Macroblock; | |
3695 | 73 |
5711 | 74 typedef struct { |
75 uint8_t coeff_reorder[64]; /* used in vp6 only */ | |
76 uint8_t coeff_index_to_pos[64]; /* used in vp6 only */ | |
77 uint8_t vector_sig[2]; /* delta sign */ | |
78 uint8_t vector_dct[2]; /* delta coding types */ | |
79 uint8_t vector_pdi[2][2]; /* predefined delta init */ | |
80 uint8_t vector_pdv[2][7]; /* predefined delta values */ | |
81 uint8_t vector_fdv[2][8]; /* 8 bit delta value definition */ | |
82 uint8_t coeff_dccv[2][11]; /* DC coeff value */ | |
83 uint8_t coeff_ract[2][3][6][11]; /* Run/AC coding type and AC coeff value */ | |
84 uint8_t coeff_acct[2][3][3][6][5];/* vp5 only AC coding type for coding group < 3 */ | |
85 uint8_t coeff_dcct[2][36][5]; /* DC coeff coding type */ | |
86 uint8_t coeff_runv[2][14]; /* run value (vp6 only) */ | |
87 uint8_t mb_type[3][10][10]; /* model for decoding MB type */ | |
88 uint8_t mb_types_stats[3][10][2];/* contextual, next MB type stats */ | |
8304 | 89 } VP56Model; |
5711 | 90 |
3695 | 91 struct vp56_context { |
92 AVCodecContext *avctx; | |
93 DSPContext dsp; | |
11665 | 94 VP56DSPContext vp56dsp; |
3695 | 95 ScanTable scantable; |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
96 AVFrame frames[4]; |
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
97 AVFrame *framep[6]; |
3695 | 98 uint8_t *edge_emu_buffer_alloc; |
99 uint8_t *edge_emu_buffer; | |
8299 | 100 VP56RangeCoder c; |
101 VP56RangeCoder cc; | |
102 VP56RangeCoder *ccp; | |
4308 | 103 int sub_version; |
3695 | 104 |
105 /* frame info */ | |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
106 int plane_width[4]; |
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
107 int plane_height[4]; |
3695 | 108 int mb_width; /* number of horizontal MB */ |
109 int mb_height; /* number of vertical MB */ | |
110 int block_offset[6]; | |
111 | |
112 int quantizer; | |
113 uint16_t dequant_dc; | |
114 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
|
115 int8_t *qscale_table; |
3695 | 116 |
117 /* DC predictors management */ | |
8299 | 118 VP56RefDc *above_blocks; |
119 VP56RefDc left_block[4]; | |
3695 | 120 int above_block_idx[6]; |
121 DCTELEM prev_dc[3][3]; /* [plan][ref_frame] */ | |
122 | |
123 /* blocks / macroblock */ | |
8299 | 124 VP56mb mb_type; |
125 VP56Macroblock *macroblocks; | |
11369 | 126 DECLARE_ALIGNED(16, DCTELEM, block_coeff)[6][64]; |
3695 | 127 |
128 /* motion vectors */ | |
8299 | 129 VP56mv mv[6]; /* vectors for each block in MB */ |
130 VP56mv vector_candidate[2]; | |
3695 | 131 int vector_candidate_pos; |
132 | |
133 /* filtering hints */ | |
4348 | 134 int filter_header; /* used in vp6 only */ |
3695 | 135 int deblock_filtering; |
136 int filter_selection; | |
137 int filter_mode; | |
138 int max_vector_length; | |
139 int sample_variance_threshold; | |
140 | |
141 uint8_t coeff_ctx[4][64]; /* used in vp5 only */ | |
142 uint8_t coeff_ctx_last[4]; /* used in vp5 only */ | |
143 | |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
144 int has_alpha; |
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
145 |
3695 | 146 /* upside-down flipping hints */ |
147 int flip; /* are we flipping ? */ | |
148 int frbi; /* first row block index in MB */ | |
149 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
|
150 int stride[4]; /* stride for each plan */ |
3695 | 151 |
152 const uint8_t *vp56_coord_div; | |
8299 | 153 VP56ParseVectorAdjustment parse_vector_adjustment; |
154 VP56Filter filter; | |
155 VP56ParseCoeff parse_coeff; | |
156 VP56DefaultModelsInit default_models_init; | |
157 VP56ParseVectorModels parse_vector_models; | |
158 VP56ParseCoeffModels parse_coeff_models; | |
159 VP56ParseHeader parse_header; | |
5711 | 160 |
8304 | 161 VP56Model *modelp; |
162 VP56Model models[2]; | |
5821 | 163 |
164 /* huffman decoding */ | |
165 int use_huffman; | |
166 GetBitContext gb; | |
167 VLC dccv_vlc[2]; | |
168 VLC runv_vlc[2]; | |
169 VLC ract_vlc[2][3][6]; | |
170 unsigned int nb_null[2][2]; /* number of consecutive NULL DC/AC */ | |
3695 | 171 }; |
172 | |
173 | |
5714
314be1cfdcb0
add a new vp6a codec (add alpha plan support to vp6)
aurel
parents:
5711
diff
changeset
|
174 void vp56_init(AVCodecContext *avctx, int flip, int has_alpha); |
3695 | 175 int vp56_free(AVCodecContext *avctx); |
8299 | 176 void vp56_init_dequant(VP56Context *s, int quantizer); |
3695 | 177 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
|
178 AVPacket *avpkt); |
3695 | 179 |
180 | |
181 /** | |
182 * vp56 specific range coder implementation | |
183 */ | |
184 | |
8299 | 185 static inline void vp56_init_range_decoder(VP56RangeCoder *c, |
6297 | 186 const uint8_t *buf, int buf_size) |
3695 | 187 { |
188 c->high = 255; | |
12031 | 189 c->bits = -8; |
3695 | 190 c->buffer = buf; |
9919
c7c1c6b35a73
vp56dec: ensure range coder won't read past the end of input buffer
aurel
parents:
9428
diff
changeset
|
191 c->end = buf + buf_size; |
5089 | 192 c->code_word = bytestream_get_be16(&c->buffer); |
3695 | 193 } |
194 | |
8299 | 195 static inline int vp56_rac_get_prob(VP56RangeCoder *c, uint8_t prob) |
3695 | 196 { |
11920 | 197 unsigned int low = 1 + (((c->high - 1) * prob) >> 8); |
3695 | 198 unsigned int low_shift = low << 8; |
199 int bit = c->code_word >= low_shift; | |
12029 | 200 int shift; |
3695 | 201 |
12032 | 202 c->high = bit ? c->high - low : low; |
203 c->code_word = bit ? c->code_word - low_shift : c->code_word; | |
3695 | 204 |
205 /* normalize */ | |
12029 | 206 shift = ff_h264_norm_shift[c->high] - 1; |
207 c->high <<= shift; | |
208 c->code_word <<= shift; | |
12031 | 209 c->bits += shift; |
210 if(c->bits >= 0 && c->buffer < c->end) { | |
211 c->code_word |= *c->buffer++ << c->bits; | |
212 c->bits -= 8; | |
3695 | 213 } |
214 return bit; | |
215 } | |
216 | |
8299 | 217 static inline int vp56_rac_get(VP56RangeCoder *c) |
3695 | 218 { |
219 /* equiprobable */ | |
220 int low = (c->high + 1) >> 1; | |
221 unsigned int low_shift = low << 8; | |
222 int bit = c->code_word >= low_shift; | |
223 if (bit) { | |
224 c->high = (c->high - low) << 1; | |
225 c->code_word -= low_shift; | |
226 } else { | |
227 c->high = low << 1; | |
228 } | |
229 | |
230 /* normalize */ | |
231 c->code_word <<= 1; | |
12031 | 232 if (++c->bits == 0 && c->buffer < c->end) { |
233 c->bits = -8; | |
3695 | 234 c->code_word |= *c->buffer++; |
235 } | |
236 return bit; | |
237 } | |
238 | |
11921 | 239 // rounding is different than vp56_rac_get, is vp56_rac_get wrong? |
240 static inline int vp8_rac_get(VP56RangeCoder *c) | |
241 { | |
242 return vp56_rac_get_prob(c, 128); | |
243 } | |
244 | |
8299 | 245 static inline int vp56_rac_gets(VP56RangeCoder *c, int bits) |
3695 | 246 { |
247 int value = 0; | |
248 | |
249 while (bits--) { | |
250 value = (value << 1) | vp56_rac_get(c); | |
251 } | |
252 | |
253 return value; | |
254 } | |
255 | |
11921 | 256 static inline int vp8_rac_get_uint(VP56RangeCoder *c, int bits) |
257 { | |
258 int value = 0; | |
259 | |
260 while (bits--) { | |
261 value = (value << 1) | vp8_rac_get(c); | |
262 } | |
263 | |
264 return value; | |
265 } | |
266 | |
267 // fixme: add 1 bit to all the calls to this? | |
268 static inline int vp8_rac_get_sint(VP56RangeCoder *c, int bits) | |
269 { | |
270 int v; | |
271 | |
272 if (!vp8_rac_get(c)) | |
273 return 0; | |
274 | |
275 v = vp8_rac_get_uint(c, bits); | |
276 | |
277 if (vp8_rac_get(c)) | |
278 v = -v; | |
279 | |
280 return v; | |
281 } | |
282 | |
283 // P(7) | |
8299 | 284 static inline int vp56_rac_gets_nn(VP56RangeCoder *c, int bits) |
3695 | 285 { |
286 int v = vp56_rac_gets(c, 7) << 1; | |
287 return v + !v; | |
288 } | |
289 | |
11921 | 290 static inline int vp8_rac_get_nn(VP56RangeCoder *c) |
291 { | |
292 int v = vp8_rac_get_uint(c, 7) << 1; | |
293 return v + !v; | |
294 } | |
295 | |
8299 | 296 static inline int vp56_rac_get_tree(VP56RangeCoder *c, |
297 const VP56Tree *tree, | |
3695 | 298 const uint8_t *probs) |
299 { | |
300 while (tree->val > 0) { | |
301 if (vp56_rac_get_prob(c, probs[tree->prob_idx])) | |
302 tree += tree->val; | |
303 else | |
304 tree++; | |
305 } | |
306 return -tree->val; | |
307 } | |
308 | |
11921 | 309 /** |
310 * This is identical to vp8_rac_get_tree except for the possibility of starting | |
311 * on a node other than the root node, needed for coeff decode where this is | |
312 * used to save a bit after a 0 token (by disallowing EOB to immediately follow.) | |
313 */ | |
314 static inline int vp8_rac_get_tree_with_offset(VP56RangeCoder *c, const int8_t (*tree)[2], | |
315 const uint8_t *probs, int i) | |
316 { | |
317 do { | |
318 i = tree[i][vp56_rac_get_prob(c, probs[i])]; | |
319 } while (i > 0); | |
320 | |
321 return -i; | |
322 } | |
323 | |
324 // how probabilities are associated with decisions is different I think | |
325 // well, the new scheme fits in the old but this way has one fewer branches per decision | |
326 static inline int vp8_rac_get_tree(VP56RangeCoder *c, const int8_t (*tree)[2], | |
327 const uint8_t *probs) | |
328 { | |
329 return vp8_rac_get_tree_with_offset(c, tree, probs, 0); | |
330 } | |
331 | |
332 // DCTextra | |
333 static inline int vp8_rac_get_coeff(VP56RangeCoder *c, const uint8_t *prob) | |
334 { | |
335 int v = 0; | |
336 | |
337 do { | |
338 v = (v<<1) + vp56_rac_get_prob(c, *prob++); | |
339 } while (*prob); | |
340 | |
341 return v; | |
342 } | |
343 | |
7760 | 344 #endif /* AVCODEC_VP56_H */ |