Mercurial > libavcodec.hg
annotate apedec.c @ 11577:e4a69ce0256f libavcodec
Optimize bidir_refine a bit.
compiled code is less tham 1/4 the size.
a tiny bit faster
author | michael |
---|---|
date | Fri, 02 Apr 2010 14:04:15 +0000 |
parents | 8a4984c5cacc |
children | 7dd2a45249a9 |
rev | line source |
---|---|
5673 | 1 /* |
2 * Monkey's Audio lossless audio decoder | |
3 * Copyright (c) 2007 Benjamin Zores <ben@geexbox.org> | |
4 * based upon libdemac from Dave Chapman. | |
5 * | |
6 * This file is part of FFmpeg. | |
7 * | |
8 * FFmpeg is free software; you can redistribute it and/or | |
9 * modify it under the terms of the GNU Lesser General Public | |
10 * License as published by the Free Software Foundation; either | |
11 * version 2.1 of the License, or (at your option) any later version. | |
12 * | |
13 * FFmpeg is distributed in the hope that it will be useful, | |
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
16 * Lesser General Public License for more details. | |
17 * | |
18 * You should have received a copy of the GNU Lesser General Public | |
19 * License along with FFmpeg; if not, write to the Free Software | |
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA | |
21 */ | |
22 | |
23 #define ALT_BITSTREAM_READER_LE | |
24 #include "avcodec.h" | |
25 #include "dsputil.h" | |
9428 | 26 #include "get_bits.h" |
5673 | 27 #include "bytestream.h" |
28 | |
29 /** | |
8718
e9d9d946f213
Use full internal pathname in doxygen @file directives.
diego
parents:
8683
diff
changeset
|
30 * @file libavcodec/apedec.c |
5673 | 31 * Monkey's Audio lossless audio decoder |
32 */ | |
33 | |
34 #define BLOCKS_PER_LOOP 4608 | |
35 #define MAX_CHANNELS 2 | |
36 #define MAX_BYTESPERSAMPLE 3 | |
37 | |
38 #define APE_FRAMECODE_MONO_SILENCE 1 | |
39 #define APE_FRAMECODE_STEREO_SILENCE 3 | |
40 #define APE_FRAMECODE_PSEUDO_STEREO 4 | |
41 | |
42 #define HISTORY_SIZE 512 | |
43 #define PREDICTOR_ORDER 8 | |
44 /** Total size of all predictor histories */ | |
45 #define PREDICTOR_SIZE 50 | |
46 | |
47 #define YDELAYA (18 + PREDICTOR_ORDER*4) | |
48 #define YDELAYB (18 + PREDICTOR_ORDER*3) | |
49 #define XDELAYA (18 + PREDICTOR_ORDER*2) | |
50 #define XDELAYB (18 + PREDICTOR_ORDER) | |
51 | |
52 #define YADAPTCOEFFSA 18 | |
53 #define XADAPTCOEFFSA 14 | |
54 #define YADAPTCOEFFSB 10 | |
55 #define XADAPTCOEFFSB 5 | |
56 | |
57 /** | |
58 * Possible compression levels | |
59 * @{ | |
60 */ | |
61 enum APECompressionLevel { | |
62 COMPRESSION_LEVEL_FAST = 1000, | |
63 COMPRESSION_LEVEL_NORMAL = 2000, | |
64 COMPRESSION_LEVEL_HIGH = 3000, | |
65 COMPRESSION_LEVEL_EXTRA_HIGH = 4000, | |
66 COMPRESSION_LEVEL_INSANE = 5000 | |
67 }; | |
68 /** @} */ | |
69 | |
70 #define APE_FILTER_LEVELS 3 | |
71 | |
72 /** Filter orders depending on compression level */ | |
73 static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = { | |
74 { 0, 0, 0 }, | |
75 { 16, 0, 0 }, | |
76 { 64, 0, 0 }, | |
77 { 32, 256, 0 }, | |
78 { 16, 256, 1280 } | |
79 }; | |
80 | |
81 /** Filter fraction bits depending on compression level */ | |
6482 | 82 static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = { |
5673 | 83 { 0, 0, 0 }, |
84 { 11, 0, 0 }, | |
85 { 11, 0, 0 }, | |
86 { 10, 13, 0 }, | |
87 { 11, 13, 15 } | |
88 }; | |
89 | |
90 | |
91 /** Filters applied to the decoded data */ | |
92 typedef struct APEFilter { | |
93 int16_t *coeffs; ///< actual coefficients used in filtering | |
94 int16_t *adaptcoeffs; ///< adaptive filter coefficients used for correcting of actual filter coefficients | |
95 int16_t *historybuffer; ///< filter memory | |
96 int16_t *delay; ///< filtered values | |
97 | |
98 int avg; | |
99 } APEFilter; | |
100 | |
101 typedef struct APERice { | |
102 uint32_t k; | |
103 uint32_t ksum; | |
104 } APERice; | |
105 | |
106 typedef struct APERangecoder { | |
107 uint32_t low; ///< low end of interval | |
108 uint32_t range; ///< length of interval | |
109 uint32_t help; ///< bytes_to_follow resp. intermediate value | |
110 unsigned int buffer; ///< buffer for input/output | |
111 } APERangecoder; | |
112 | |
113 /** Filter histories */ | |
114 typedef struct APEPredictor { | |
115 int32_t *buf; | |
116 | |
117 int32_t lastA[2]; | |
118 | |
119 int32_t filterA[2]; | |
120 int32_t filterB[2]; | |
121 | |
122 int32_t coeffsA[2][4]; ///< adaption coefficients | |
123 int32_t coeffsB[2][5]; ///< adaption coefficients | |
124 int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE]; | |
125 } APEPredictor; | |
126 | |
127 /** Decoder context */ | |
128 typedef struct APEContext { | |
129 AVCodecContext *avctx; | |
130 DSPContext dsp; | |
131 int channels; | |
132 int samples; ///< samples left to decode in current frame | |
133 | |
134 int fileversion; ///< codec version, very important in decoding process | |
135 int compression_level; ///< compression levels | |
136 int fset; ///< which filter set to use (calculated from compression level) | |
137 int flags; ///< global decoder flags | |
138 | |
139 uint32_t CRC; ///< frame CRC | |
140 int frameflags; ///< frame flags | |
141 int currentframeblocks; ///< samples (per channel) in current frame | |
142 int blocksdecoded; ///< count of decoded samples in current frame | |
143 APEPredictor predictor; ///< predictor used for final reconstruction | |
144 | |
145 int32_t decoded0[BLOCKS_PER_LOOP]; ///< decoded data for the first channel | |
146 int32_t decoded1[BLOCKS_PER_LOOP]; ///< decoded data for the second channel | |
147 | |
148 int16_t* filterbuf[APE_FILTER_LEVELS]; ///< filter memory | |
149 | |
150 APERangecoder rc; ///< rangecoder used to decode actual values | |
151 APERice riceX; ///< rice code parameters for the second channel | |
152 APERice riceY; ///< rice code parameters for the first channel | |
153 APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction | |
154 | |
155 uint8_t *data; ///< current frame data | |
156 uint8_t *data_end; ///< frame data end | |
6223 | 157 const uint8_t *ptr; ///< current position in frame data |
158 const uint8_t *last_ptr; ///< position where last 4608-sample block ended | |
6443
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
159 |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
160 int error; |
5673 | 161 } APEContext; |
162 | |
163 // TODO: dsputilize | |
164 | |
6517
48759bfbd073
Apply 'cold' attribute to init/uninit functions in libavcodec
zuxy
parents:
6482
diff
changeset
|
165 static av_cold int ape_decode_init(AVCodecContext * avctx) |
5673 | 166 { |
167 APEContext *s = avctx->priv_data; | |
168 int i; | |
169 | |
170 if (avctx->extradata_size != 6) { | |
171 av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n"); | |
172 return -1; | |
173 } | |
7823
4525dcd81357
Bump Major version, this commit is almost just renaming bits_per_sample to
michael
parents:
7723
diff
changeset
|
174 if (avctx->bits_per_coded_sample != 16) { |
5673 | 175 av_log(avctx, AV_LOG_ERROR, "Only 16-bit samples are supported\n"); |
176 return -1; | |
177 } | |
178 if (avctx->channels > 2) { | |
179 av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n"); | |
180 return -1; | |
181 } | |
182 s->avctx = avctx; | |
183 s->channels = avctx->channels; | |
184 s->fileversion = AV_RL16(avctx->extradata); | |
185 s->compression_level = AV_RL16(avctx->extradata + 2); | |
186 s->flags = AV_RL16(avctx->extradata + 4); | |
187 | |
188 av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n", s->compression_level, s->flags); | |
189 if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) { | |
190 av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n", s->compression_level); | |
191 return -1; | |
192 } | |
193 s->fset = s->compression_level / 1000 - 1; | |
194 for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
195 if (!ape_filter_orders[s->fset][i]) | |
196 break; | |
197 s->filterbuf[i] = av_malloc((ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4); | |
198 } | |
199 | |
200 dsputil_init(&s->dsp, avctx); | |
7451
85ab7655ad4d
Modify all codecs to report their supported input and output sample format(s).
pross
parents:
7203
diff
changeset
|
201 avctx->sample_fmt = SAMPLE_FMT_S16; |
8174
f11197441364
Add channel layout to several audio decoders I maintain
kostya
parents:
7914
diff
changeset
|
202 avctx->channel_layout = (avctx->channels==2) ? CH_LAYOUT_STEREO : CH_LAYOUT_MONO; |
5673 | 203 return 0; |
204 } | |
205 | |
6517
48759bfbd073
Apply 'cold' attribute to init/uninit functions in libavcodec
zuxy
parents:
6482
diff
changeset
|
206 static av_cold int ape_decode_close(AVCodecContext * avctx) |
5673 | 207 { |
208 APEContext *s = avctx->priv_data; | |
209 int i; | |
210 | |
211 for (i = 0; i < APE_FILTER_LEVELS; i++) | |
212 av_freep(&s->filterbuf[i]); | |
213 | |
11220
5518cf6b6f2e
16l trocadero: don't forget to free frame data buffer in APE decoder
kostya
parents:
10644
diff
changeset
|
214 av_freep(&s->data); |
5673 | 215 return 0; |
216 } | |
217 | |
218 /** | |
219 * @defgroup rangecoder APE range decoder | |
220 * @{ | |
221 */ | |
222 | |
223 #define CODE_BITS 32 | |
224 #define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1)) | |
225 #define SHIFT_BITS (CODE_BITS - 9) | |
226 #define EXTRA_BITS ((CODE_BITS-2) % 8 + 1) | |
227 #define BOTTOM_VALUE (TOP_VALUE >> 8) | |
228 | |
229 /** Start the decoder */ | |
230 static inline void range_start_decoding(APEContext * ctx) | |
231 { | |
232 ctx->rc.buffer = bytestream_get_byte(&ctx->ptr); | |
233 ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS); | |
234 ctx->rc.range = (uint32_t) 1 << EXTRA_BITS; | |
235 } | |
236 | |
237 /** Perform normalization */ | |
238 static inline void range_dec_normalize(APEContext * ctx) | |
239 { | |
240 while (ctx->rc.range <= BOTTOM_VALUE) { | |
6442
28e45bf84973
Prevent segfault due to reading over the end of the input buffer.
michael
parents:
6223
diff
changeset
|
241 ctx->rc.buffer <<= 8; |
28e45bf84973
Prevent segfault due to reading over the end of the input buffer.
michael
parents:
6223
diff
changeset
|
242 if(ctx->ptr < ctx->data_end) |
28e45bf84973
Prevent segfault due to reading over the end of the input buffer.
michael
parents:
6223
diff
changeset
|
243 ctx->rc.buffer += *ctx->ptr; |
28e45bf84973
Prevent segfault due to reading over the end of the input buffer.
michael
parents:
6223
diff
changeset
|
244 ctx->ptr++; |
5673 | 245 ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF); |
246 ctx->rc.range <<= 8; | |
247 } | |
248 } | |
249 | |
250 /** | |
251 * Calculate culmulative frequency for next symbol. Does NO update! | |
8683 | 252 * @param ctx decoder context |
5673 | 253 * @param tot_f is the total frequency or (code_value)1<<shift |
254 * @return the culmulative frequency | |
255 */ | |
256 static inline int range_decode_culfreq(APEContext * ctx, int tot_f) | |
257 { | |
258 range_dec_normalize(ctx); | |
259 ctx->rc.help = ctx->rc.range / tot_f; | |
260 return ctx->rc.low / ctx->rc.help; | |
261 } | |
262 | |
263 /** | |
264 * Decode value with given size in bits | |
8683 | 265 * @param ctx decoder context |
5673 | 266 * @param shift number of bits to decode |
267 */ | |
268 static inline int range_decode_culshift(APEContext * ctx, int shift) | |
269 { | |
270 range_dec_normalize(ctx); | |
271 ctx->rc.help = ctx->rc.range >> shift; | |
272 return ctx->rc.low / ctx->rc.help; | |
273 } | |
274 | |
275 | |
276 /** | |
277 * Update decoding state | |
8683 | 278 * @param ctx decoder context |
5673 | 279 * @param sy_f the interval length (frequency of the symbol) |
280 * @param lt_f the lower end (frequency sum of < symbols) | |
281 */ | |
282 static inline void range_decode_update(APEContext * ctx, int sy_f, int lt_f) | |
283 { | |
284 ctx->rc.low -= ctx->rc.help * lt_f; | |
285 ctx->rc.range = ctx->rc.help * sy_f; | |
286 } | |
287 | |
288 /** Decode n bits (n <= 16) without modelling */ | |
289 static inline int range_decode_bits(APEContext * ctx, int n) | |
290 { | |
291 int sym = range_decode_culshift(ctx, n); | |
292 range_decode_update(ctx, 1, sym); | |
293 return sym; | |
294 } | |
295 | |
296 | |
297 #define MODEL_ELEMENTS 64 | |
298 | |
299 /** | |
300 * Fixed probabilities for symbols in Monkey Audio version 3.97 | |
301 */ | |
6482 | 302 static const uint16_t counts_3970[22] = { |
5673 | 303 0, 14824, 28224, 39348, 47855, 53994, 58171, 60926, |
304 62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419, | |
6444 | 305 65450, 65469, 65480, 65487, 65491, 65493, |
5673 | 306 }; |
307 | |
308 /** | |
309 * Probability ranges for symbols in Monkey Audio version 3.97 | |
310 */ | |
6444 | 311 static const uint16_t counts_diff_3970[21] = { |
5673 | 312 14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756, |
313 1104, 677, 415, 248, 150, 89, 54, 31, | |
6444 | 314 19, 11, 7, 4, 2, |
5673 | 315 }; |
316 | |
317 /** | |
318 * Fixed probabilities for symbols in Monkey Audio version 3.98 | |
319 */ | |
6482 | 320 static const uint16_t counts_3980[22] = { |
5673 | 321 0, 19578, 36160, 48417, 56323, 60899, 63265, 64435, |
322 64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482, | |
6444 | 323 65485, 65488, 65490, 65491, 65492, 65493, |
5673 | 324 }; |
325 | |
326 /** | |
327 * Probability ranges for symbols in Monkey Audio version 3.98 | |
328 */ | |
6444 | 329 static const uint16_t counts_diff_3980[21] = { |
5673 | 330 19578, 16582, 12257, 7906, 4576, 2366, 1170, 536, |
331 261, 119, 65, 31, 19, 10, 6, 3, | |
6444 | 332 3, 2, 1, 1, 1, |
5673 | 333 }; |
334 | |
335 /** | |
336 * Decode symbol | |
8683 | 337 * @param ctx decoder context |
5673 | 338 * @param counts probability range start position |
8683 | 339 * @param counts_diff probability range widths |
5673 | 340 */ |
341 static inline int range_get_symbol(APEContext * ctx, | |
6482 | 342 const uint16_t counts[], |
5673 | 343 const uint16_t counts_diff[]) |
344 { | |
345 int symbol, cf; | |
346 | |
347 cf = range_decode_culshift(ctx, 16); | |
348 | |
6443
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
349 if(cf > 65492){ |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
350 symbol= cf - 65535 + 63; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
351 range_decode_update(ctx, 1, cf); |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
352 if(cf > 65535) |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
353 ctx->error=1; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
354 return symbol; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
355 } |
5673 | 356 /* figure out the symbol inefficiently; a binary search would be much better */ |
357 for (symbol = 0; counts[symbol + 1] <= cf; symbol++); | |
358 | |
359 range_decode_update(ctx, counts_diff[symbol], counts[symbol]); | |
360 | |
361 return symbol; | |
362 } | |
363 /** @} */ // group rangecoder | |
364 | |
365 static inline void update_rice(APERice *rice, int x) | |
366 { | |
7914
c15b1e83d27d
Correct wrong lower limit and condition used in APE decoder
kostya
parents:
7823
diff
changeset
|
367 int lim = rice->k ? (1 << (rice->k + 4)) : 0; |
5673 | 368 rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5); |
369 | |
7914
c15b1e83d27d
Correct wrong lower limit and condition used in APE decoder
kostya
parents:
7823
diff
changeset
|
370 if (rice->ksum < lim) |
5673 | 371 rice->k--; |
372 else if (rice->ksum >= (1 << (rice->k + 5))) | |
373 rice->k++; | |
374 } | |
375 | |
376 static inline int ape_decode_value(APEContext * ctx, APERice *rice) | |
377 { | |
378 int x, overflow; | |
379 | |
7723
20b105281e87
While APE changed container format in 3.98, frequency tables for range coding
kostya
parents:
7451
diff
changeset
|
380 if (ctx->fileversion < 3990) { |
5673 | 381 int tmpk; |
382 | |
383 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970); | |
384 | |
385 if (overflow == (MODEL_ELEMENTS - 1)) { | |
386 tmpk = range_decode_bits(ctx, 5); | |
387 overflow = 0; | |
388 } else | |
389 tmpk = (rice->k < 1) ? 0 : rice->k - 1; | |
390 | |
391 if (tmpk <= 16) | |
392 x = range_decode_bits(ctx, tmpk); | |
393 else { | |
394 x = range_decode_bits(ctx, 16); | |
395 x |= (range_decode_bits(ctx, tmpk - 16) << 16); | |
396 } | |
397 x += overflow << tmpk; | |
398 } else { | |
399 int base, pivot; | |
400 | |
401 pivot = rice->ksum >> 5; | |
402 if (pivot == 0) | |
403 pivot = 1; | |
404 | |
405 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980); | |
406 | |
407 if (overflow == (MODEL_ELEMENTS - 1)) { | |
408 overflow = range_decode_bits(ctx, 16) << 16; | |
409 overflow |= range_decode_bits(ctx, 16); | |
410 } | |
411 | |
10545
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
412 if (pivot < 0x10000) { |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
413 base = range_decode_culfreq(ctx, pivot); |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
414 range_decode_update(ctx, 1, base); |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
415 } else { |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
416 int base_hi = pivot, base_lo; |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
417 int bbits = 0; |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
418 |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
419 while (base_hi & ~0xFFFF) { |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
420 base_hi >>= 1; |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
421 bbits++; |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
422 } |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
423 base_hi = range_decode_culfreq(ctx, base_hi + 1); |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
424 range_decode_update(ctx, 1, base_hi); |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
425 base_lo = range_decode_culfreq(ctx, 1 << bbits); |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
426 range_decode_update(ctx, 1, base_lo); |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
427 |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
428 base = (base_hi << bbits) + base_lo; |
e5a8a6b9a717
Implement missing case for decoding samples with large pivot value in APE
kostya
parents:
10181
diff
changeset
|
429 } |
5673 | 430 |
431 x = base + overflow * pivot; | |
432 } | |
433 | |
434 update_rice(rice, x); | |
435 | |
436 /* Convert to signed */ | |
437 if (x & 1) | |
438 return (x >> 1) + 1; | |
439 else | |
440 return -(x >> 1); | |
441 } | |
442 | |
443 static void entropy_decode(APEContext * ctx, int blockstodecode, int stereo) | |
444 { | |
445 int32_t *decoded0 = ctx->decoded0; | |
446 int32_t *decoded1 = ctx->decoded1; | |
447 | |
448 ctx->blocksdecoded = blockstodecode; | |
449 | |
450 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
451 /* We are pure silence, just memset the output buffer. */ | |
452 memset(decoded0, 0, blockstodecode * sizeof(int32_t)); | |
453 memset(decoded1, 0, blockstodecode * sizeof(int32_t)); | |
454 } else { | |
455 while (blockstodecode--) { | |
456 *decoded0++ = ape_decode_value(ctx, &ctx->riceY); | |
457 if (stereo) | |
458 *decoded1++ = ape_decode_value(ctx, &ctx->riceX); | |
459 } | |
460 } | |
461 | |
462 if (ctx->blocksdecoded == ctx->currentframeblocks) | |
463 range_dec_normalize(ctx); /* normalize to use up all bytes */ | |
464 } | |
465 | |
466 static void init_entropy_decoder(APEContext * ctx) | |
467 { | |
468 /* Read the CRC */ | |
469 ctx->CRC = bytestream_get_be32(&ctx->ptr); | |
470 | |
471 /* Read the frame flags if they exist */ | |
472 ctx->frameflags = 0; | |
473 if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) { | |
474 ctx->CRC &= ~0x80000000; | |
475 | |
476 ctx->frameflags = bytestream_get_be32(&ctx->ptr); | |
477 } | |
478 | |
479 /* Keep a count of the blocks decoded in this frame */ | |
480 ctx->blocksdecoded = 0; | |
481 | |
5963 | 482 /* Initialize the rice structs */ |
5673 | 483 ctx->riceX.k = 10; |
484 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16; | |
485 ctx->riceY.k = 10; | |
486 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16; | |
487 | |
488 /* The first 8 bits of input are ignored. */ | |
489 ctx->ptr++; | |
490 | |
491 range_start_decoding(ctx); | |
492 } | |
493 | |
494 static const int32_t initial_coeffs[4] = { | |
495 360, 317, -109, 98 | |
496 }; | |
497 | |
498 static void init_predictor_decoder(APEContext * ctx) | |
499 { | |
500 APEPredictor *p = &ctx->predictor; | |
501 | |
502 /* Zero the history buffers */ | |
503 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t)); | |
504 p->buf = p->historybuffer; | |
505 | |
5966 | 506 /* Initialize and zero the coefficients */ |
5673 | 507 memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs)); |
508 memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs)); | |
509 memset(p->coeffsB, 0, sizeof(p->coeffsB)); | |
510 | |
511 p->filterA[0] = p->filterA[1] = 0; | |
512 p->filterB[0] = p->filterB[1] = 0; | |
513 p->lastA[0] = p->lastA[1] = 0; | |
514 } | |
515 | |
516 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */ | |
517 static inline int APESIGN(int32_t x) { | |
518 return (x < 0) - (x > 0); | |
519 } | |
520 | |
10631 | 521 static av_always_inline int predictor_update_filter(APEPredictor *p, const int decoded, const int filter, const int delayA, const int delayB, const int adaptA, const int adaptB) |
5673 | 522 { |
10632 | 523 int32_t predictionA, predictionB, sign; |
5673 | 524 |
525 p->buf[delayA] = p->lastA[filter]; | |
526 p->buf[adaptA] = APESIGN(p->buf[delayA]); | |
527 p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1]; | |
528 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]); | |
529 | |
530 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] + | |
531 p->buf[delayA - 1] * p->coeffsA[filter][1] + | |
532 p->buf[delayA - 2] * p->coeffsA[filter][2] + | |
533 p->buf[delayA - 3] * p->coeffsA[filter][3]; | |
534 | |
535 /* Apply a scaled first-order filter compression */ | |
536 p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5); | |
537 p->buf[adaptB] = APESIGN(p->buf[delayB]); | |
538 p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1]; | |
539 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]); | |
540 p->filterB[filter] = p->filterA[filter ^ 1]; | |
541 | |
542 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] + | |
543 p->buf[delayB - 1] * p->coeffsB[filter][1] + | |
544 p->buf[delayB - 2] * p->coeffsB[filter][2] + | |
545 p->buf[delayB - 3] * p->coeffsB[filter][3] + | |
546 p->buf[delayB - 4] * p->coeffsB[filter][4]; | |
547 | |
548 p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10); | |
549 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5); | |
550 | |
10632 | 551 sign = APESIGN(decoded); |
552 p->coeffsA[filter][0] += p->buf[adaptA ] * sign; | |
553 p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign; | |
554 p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign; | |
555 p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign; | |
556 p->coeffsB[filter][0] += p->buf[adaptB ] * sign; | |
557 p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign; | |
558 p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign; | |
559 p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign; | |
560 p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign; | |
5673 | 561 |
562 return p->filterA[filter]; | |
563 } | |
564 | |
565 static void predictor_decode_stereo(APEContext * ctx, int count) | |
566 { | |
567 APEPredictor *p = &ctx->predictor; | |
568 int32_t *decoded0 = ctx->decoded0; | |
569 int32_t *decoded1 = ctx->decoded1; | |
570 | |
571 while (count--) { | |
572 /* Predictor Y */ | |
10631 | 573 *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB, YADAPTCOEFFSA, YADAPTCOEFFSB); |
574 decoded0++; | |
575 *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB, XADAPTCOEFFSA, XADAPTCOEFFSB); | |
576 decoded1++; | |
5673 | 577 |
578 /* Combined */ | |
579 p->buf++; | |
580 | |
581 /* Have we filled the history buffer? */ | |
582 if (p->buf == p->historybuffer + HISTORY_SIZE) { | |
583 memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t)); | |
584 p->buf = p->historybuffer; | |
585 } | |
586 } | |
587 } | |
588 | |
589 static void predictor_decode_mono(APEContext * ctx, int count) | |
590 { | |
591 APEPredictor *p = &ctx->predictor; | |
592 int32_t *decoded0 = ctx->decoded0; | |
10632 | 593 int32_t predictionA, currentA, A, sign; |
5673 | 594 |
595 currentA = p->lastA[0]; | |
596 | |
597 while (count--) { | |
598 A = *decoded0; | |
599 | |
600 p->buf[YDELAYA] = currentA; | |
601 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1]; | |
602 | |
603 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] + | |
604 p->buf[YDELAYA - 1] * p->coeffsA[0][1] + | |
605 p->buf[YDELAYA - 2] * p->coeffsA[0][2] + | |
606 p->buf[YDELAYA - 3] * p->coeffsA[0][3]; | |
607 | |
608 currentA = A + (predictionA >> 10); | |
609 | |
610 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]); | |
611 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]); | |
612 | |
10632 | 613 sign = APESIGN(A); |
614 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign; | |
615 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign; | |
616 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign; | |
617 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign; | |
5673 | 618 |
619 p->buf++; | |
620 | |
621 /* Have we filled the history buffer? */ | |
622 if (p->buf == p->historybuffer + HISTORY_SIZE) { | |
623 memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t)); | |
624 p->buf = p->historybuffer; | |
625 } | |
626 | |
627 p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5); | |
628 *(decoded0++) = p->filterA[0]; | |
629 } | |
630 | |
631 p->lastA[0] = currentA; | |
632 } | |
633 | |
634 static void do_init_filter(APEFilter *f, int16_t * buf, int order) | |
635 { | |
636 f->coeffs = buf; | |
637 f->historybuffer = buf + order; | |
638 f->delay = f->historybuffer + order * 2; | |
639 f->adaptcoeffs = f->historybuffer + order; | |
640 | |
641 memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t)); | |
642 memset(f->coeffs, 0, order * sizeof(int16_t)); | |
643 f->avg = 0; | |
644 } | |
645 | |
646 static void init_filter(APEContext * ctx, APEFilter *f, int16_t * buf, int order) | |
647 { | |
648 do_init_filter(&f[0], buf, order); | |
649 do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order); | |
650 } | |
651 | |
10644 | 652 static void do_apply_filter(APEContext * ctx, int version, APEFilter *f, int32_t *data, int count, int order, int fracbits) |
5673 | 653 { |
654 int res; | |
655 int absres; | |
656 | |
657 while (count--) { | |
658 /* round fixedpoint scalar product */ | |
10644 | 659 res = ctx->dsp.scalarproduct_and_madd_int16(f->coeffs, f->delay - order, f->adaptcoeffs - order, order, APESIGN(*data)); |
660 res = (res + (1 << (fracbits - 1))) >> fracbits; | |
5673 | 661 res += *data; |
662 *data++ = res; | |
663 | |
664 /* Update the output history */ | |
665 *f->delay++ = av_clip_int16(res); | |
666 | |
667 if (version < 3980) { | |
668 /* Version ??? to < 3.98 files (untested) */ | |
669 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4; | |
670 f->adaptcoeffs[-4] >>= 1; | |
671 f->adaptcoeffs[-8] >>= 1; | |
672 } else { | |
673 /* Version 3.98 and later files */ | |
674 | |
675 /* Update the adaption coefficients */ | |
10639
4e0b0c0cb915
1-13% faster apply_filter, 1-3% faster ape decoding on core2
lorenm
parents:
10638
diff
changeset
|
676 absres = FFABS(res); |
4e0b0c0cb915
1-13% faster apply_filter, 1-3% faster ape decoding on core2
lorenm
parents:
10638
diff
changeset
|
677 if (absres) |
4e0b0c0cb915
1-13% faster apply_filter, 1-3% faster ape decoding on core2
lorenm
parents:
10638
diff
changeset
|
678 *f->adaptcoeffs = ((res & (1<<31)) - (1<<30)) >> (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3)); |
5673 | 679 else |
680 *f->adaptcoeffs = 0; | |
681 | |
682 f->avg += (absres - f->avg) / 16; | |
683 | |
684 f->adaptcoeffs[-1] >>= 1; | |
685 f->adaptcoeffs[-2] >>= 1; | |
686 f->adaptcoeffs[-8] >>= 1; | |
687 } | |
688 | |
689 f->adaptcoeffs++; | |
690 | |
691 /* Have we filled the history buffer? */ | |
692 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) { | |
693 memmove(f->historybuffer, f->delay - (order * 2), | |
694 (order * 2) * sizeof(int16_t)); | |
695 f->delay = f->historybuffer + order * 2; | |
696 f->adaptcoeffs = f->historybuffer + order; | |
697 } | |
698 } | |
699 } | |
700 | |
701 static void apply_filter(APEContext * ctx, APEFilter *f, | |
702 int32_t * data0, int32_t * data1, | |
703 int count, int order, int fracbits) | |
704 { | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
705 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits); |
5673 | 706 if (data1) |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
707 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits); |
5673 | 708 } |
709 | |
710 static void ape_apply_filters(APEContext * ctx, int32_t * decoded0, | |
711 int32_t * decoded1, int count) | |
712 { | |
713 int i; | |
714 | |
715 for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
716 if (!ape_filter_orders[ctx->fset][i]) | |
717 break; | |
718 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count, ape_filter_orders[ctx->fset][i], ape_filter_fracbits[ctx->fset][i]); | |
719 } | |
720 } | |
721 | |
722 static void init_frame_decoder(APEContext * ctx) | |
723 { | |
724 int i; | |
725 init_entropy_decoder(ctx); | |
726 init_predictor_decoder(ctx); | |
727 | |
728 for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
729 if (!ape_filter_orders[ctx->fset][i]) | |
730 break; | |
731 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i], ape_filter_orders[ctx->fset][i]); | |
732 } | |
733 } | |
734 | |
735 static void ape_unpack_mono(APEContext * ctx, int count) | |
736 { | |
737 int32_t left; | |
738 int32_t *decoded0 = ctx->decoded0; | |
739 int32_t *decoded1 = ctx->decoded1; | |
740 | |
741 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
742 entropy_decode(ctx, count, 0); | |
743 /* We are pure silence, so we're done. */ | |
744 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n"); | |
745 return; | |
746 } | |
747 | |
748 entropy_decode(ctx, count, 0); | |
749 ape_apply_filters(ctx, decoded0, NULL, count); | |
750 | |
751 /* Now apply the predictor decoding */ | |
752 predictor_decode_mono(ctx, count); | |
753 | |
754 /* Pseudo-stereo - just copy left channel to right channel */ | |
755 if (ctx->channels == 2) { | |
756 while (count--) { | |
757 left = *decoded0; | |
758 *(decoded1++) = *(decoded0++) = left; | |
759 } | |
760 } | |
761 } | |
762 | |
763 static void ape_unpack_stereo(APEContext * ctx, int count) | |
764 { | |
765 int32_t left, right; | |
766 int32_t *decoded0 = ctx->decoded0; | |
767 int32_t *decoded1 = ctx->decoded1; | |
768 | |
769 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
770 /* We are pure silence, so we're done. */ | |
771 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n"); | |
772 return; | |
773 } | |
774 | |
775 entropy_decode(ctx, count, 1); | |
776 ape_apply_filters(ctx, decoded0, decoded1, count); | |
777 | |
778 /* Now apply the predictor decoding */ | |
779 predictor_decode_stereo(ctx, count); | |
780 | |
781 /* Decorrelate and scale to output depth */ | |
782 while (count--) { | |
783 left = *decoded1 - (*decoded0 / 2); | |
784 right = left + *decoded0; | |
785 | |
786 *(decoded0++) = left; | |
787 *(decoded1++) = right; | |
788 } | |
789 } | |
790 | |
791 static int ape_decode_frame(AVCodecContext * avctx, | |
792 void *data, int *data_size, | |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
793 AVPacket *avpkt) |
5673 | 794 { |
9355
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
795 const uint8_t *buf = avpkt->data; |
54bc8a2727b0
Implement avcodec_decode_video2(), _audio3() and _subtitle2() which takes an
rbultje
parents:
8718
diff
changeset
|
796 int buf_size = avpkt->size; |
5673 | 797 APEContext *s = avctx->priv_data; |
798 int16_t *samples = data; | |
799 int nblocks; | |
800 int i, n; | |
801 int blockstodecode; | |
802 int bytes_used; | |
803 | |
804 if (buf_size == 0 && !s->samples) { | |
805 *data_size = 0; | |
806 return 0; | |
807 } | |
808 | |
809 /* should not happen but who knows */ | |
810 if (BLOCKS_PER_LOOP * 2 * avctx->channels > *data_size) { | |
811 av_log (avctx, AV_LOG_ERROR, "Packet size is too big to be handled in lavc! (max is %d where you have %d)\n", *data_size, s->samples * 2 * avctx->channels); | |
812 return -1; | |
813 } | |
814 | |
815 if(!s->samples){ | |
816 s->data = av_realloc(s->data, (buf_size + 3) & ~3); | |
6223 | 817 s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2); |
5673 | 818 s->ptr = s->last_ptr = s->data; |
819 s->data_end = s->data + buf_size; | |
820 | |
821 nblocks = s->samples = bytestream_get_be32(&s->ptr); | |
822 n = bytestream_get_be32(&s->ptr); | |
823 if(n < 0 || n > 3){ | |
824 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n"); | |
825 s->data = NULL; | |
826 return -1; | |
827 } | |
828 s->ptr += n; | |
829 | |
830 s->currentframeblocks = nblocks; | |
831 buf += 4; | |
832 if (s->samples <= 0) { | |
833 *data_size = 0; | |
834 return buf_size; | |
835 } | |
836 | |
837 memset(s->decoded0, 0, sizeof(s->decoded0)); | |
838 memset(s->decoded1, 0, sizeof(s->decoded1)); | |
839 | |
840 /* Initialize the frame decoder */ | |
841 init_frame_decoder(s); | |
842 } | |
843 | |
844 if (!s->data) { | |
845 *data_size = 0; | |
846 return buf_size; | |
847 } | |
848 | |
849 nblocks = s->samples; | |
850 blockstodecode = FFMIN(BLOCKS_PER_LOOP, nblocks); | |
851 | |
6443
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
852 s->error=0; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
853 |
5673 | 854 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO)) |
855 ape_unpack_mono(s, blockstodecode); | |
856 else | |
857 ape_unpack_stereo(s, blockstodecode); | |
10638
40ff9c7958da
100l trocadero: call emms_c() after doing decoding with SIMD in APE decoder
kostya
parents:
10632
diff
changeset
|
858 emms_c(); |
5673 | 859 |
6443
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
860 if(s->error || s->ptr > s->data_end){ |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
861 s->samples=0; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
862 av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n"); |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
863 return -1; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
864 } |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
865 |
5673 | 866 for (i = 0; i < blockstodecode; i++) { |
867 *samples++ = s->decoded0[i]; | |
868 if(s->channels == 2) | |
869 *samples++ = s->decoded1[i]; | |
870 } | |
871 | |
872 s->samples -= blockstodecode; | |
873 | |
874 *data_size = blockstodecode * 2 * s->channels; | |
875 bytes_used = s->samples ? s->ptr - s->last_ptr : buf_size; | |
876 s->last_ptr = s->ptr; | |
877 return bytes_used; | |
878 } | |
879 | |
880 AVCodec ape_decoder = { | |
881 "ape", | |
11560
8a4984c5cacc
Define AVMediaType enum, and use it instead of enum CodecType, which
stefano
parents:
11220
diff
changeset
|
882 AVMEDIA_TYPE_AUDIO, |
5673 | 883 CODEC_ID_APE, |
884 sizeof(APEContext), | |
885 ape_decode_init, | |
886 NULL, | |
887 ape_decode_close, | |
888 ape_decode_frame, | |
10181
b33c75c58ffe
Monkey's Audio uses subframes too, declare that in its capabilities
kostya
parents:
9428
diff
changeset
|
889 .capabilities = CODEC_CAP_SUBFRAMES, |
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6712
diff
changeset
|
890 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"), |
5673 | 891 }; |