Mercurial > libavcodec.hg
annotate apedec.c @ 7209:da8d843b87cb libavcodec
Another use of scalar_product_float()
author | vitor |
---|---|
date | Sun, 06 Jul 2008 10:32:29 +0000 |
parents | 87b1dfb5a98d |
children | 85ab7655ad4d |
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" | |
26 #include "bitstream.h" | |
27 #include "bytestream.h" | |
28 | |
29 /** | |
30 * @file apedec.c | |
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 } | |
174 if (avctx->bits_per_sample != 16) { | |
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); | |
201 return 0; | |
202 } | |
203 | |
6517
48759bfbd073
Apply 'cold' attribute to init/uninit functions in libavcodec
zuxy
parents:
6482
diff
changeset
|
204 static av_cold int ape_decode_close(AVCodecContext * avctx) |
5673 | 205 { |
206 APEContext *s = avctx->priv_data; | |
207 int i; | |
208 | |
209 for (i = 0; i < APE_FILTER_LEVELS; i++) | |
210 av_freep(&s->filterbuf[i]); | |
211 | |
212 return 0; | |
213 } | |
214 | |
215 /** | |
216 * @defgroup rangecoder APE range decoder | |
217 * @{ | |
218 */ | |
219 | |
220 #define CODE_BITS 32 | |
221 #define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1)) | |
222 #define SHIFT_BITS (CODE_BITS - 9) | |
223 #define EXTRA_BITS ((CODE_BITS-2) % 8 + 1) | |
224 #define BOTTOM_VALUE (TOP_VALUE >> 8) | |
225 | |
226 /** Start the decoder */ | |
227 static inline void range_start_decoding(APEContext * ctx) | |
228 { | |
229 ctx->rc.buffer = bytestream_get_byte(&ctx->ptr); | |
230 ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS); | |
231 ctx->rc.range = (uint32_t) 1 << EXTRA_BITS; | |
232 } | |
233 | |
234 /** Perform normalization */ | |
235 static inline void range_dec_normalize(APEContext * ctx) | |
236 { | |
237 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
|
238 ctx->rc.buffer <<= 8; |
28e45bf84973
Prevent segfault due to reading over the end of the input buffer.
michael
parents:
6223
diff
changeset
|
239 if(ctx->ptr < ctx->data_end) |
28e45bf84973
Prevent segfault due to reading over the end of the input buffer.
michael
parents:
6223
diff
changeset
|
240 ctx->rc.buffer += *ctx->ptr; |
28e45bf84973
Prevent segfault due to reading over the end of the input buffer.
michael
parents:
6223
diff
changeset
|
241 ctx->ptr++; |
5673 | 242 ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF); |
243 ctx->rc.range <<= 8; | |
244 } | |
245 } | |
246 | |
247 /** | |
248 * Calculate culmulative frequency for next symbol. Does NO update! | |
249 * @param tot_f is the total frequency or (code_value)1<<shift | |
250 * @return the culmulative frequency | |
251 */ | |
252 static inline int range_decode_culfreq(APEContext * ctx, int tot_f) | |
253 { | |
254 range_dec_normalize(ctx); | |
255 ctx->rc.help = ctx->rc.range / tot_f; | |
256 return ctx->rc.low / ctx->rc.help; | |
257 } | |
258 | |
259 /** | |
260 * Decode value with given size in bits | |
261 * @param shift number of bits to decode | |
262 */ | |
263 static inline int range_decode_culshift(APEContext * ctx, int shift) | |
264 { | |
265 range_dec_normalize(ctx); | |
266 ctx->rc.help = ctx->rc.range >> shift; | |
267 return ctx->rc.low / ctx->rc.help; | |
268 } | |
269 | |
270 | |
271 /** | |
272 * Update decoding state | |
273 * @param sy_f the interval length (frequency of the symbol) | |
274 * @param lt_f the lower end (frequency sum of < symbols) | |
275 */ | |
276 static inline void range_decode_update(APEContext * ctx, int sy_f, int lt_f) | |
277 { | |
278 ctx->rc.low -= ctx->rc.help * lt_f; | |
279 ctx->rc.range = ctx->rc.help * sy_f; | |
280 } | |
281 | |
282 /** Decode n bits (n <= 16) without modelling */ | |
283 static inline int range_decode_bits(APEContext * ctx, int n) | |
284 { | |
285 int sym = range_decode_culshift(ctx, n); | |
286 range_decode_update(ctx, 1, sym); | |
287 return sym; | |
288 } | |
289 | |
290 | |
291 #define MODEL_ELEMENTS 64 | |
292 | |
293 /** | |
294 * Fixed probabilities for symbols in Monkey Audio version 3.97 | |
295 */ | |
6482 | 296 static const uint16_t counts_3970[22] = { |
5673 | 297 0, 14824, 28224, 39348, 47855, 53994, 58171, 60926, |
298 62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419, | |
6444 | 299 65450, 65469, 65480, 65487, 65491, 65493, |
5673 | 300 }; |
301 | |
302 /** | |
303 * Probability ranges for symbols in Monkey Audio version 3.97 | |
304 */ | |
6444 | 305 static const uint16_t counts_diff_3970[21] = { |
5673 | 306 14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756, |
307 1104, 677, 415, 248, 150, 89, 54, 31, | |
6444 | 308 19, 11, 7, 4, 2, |
5673 | 309 }; |
310 | |
311 /** | |
312 * Fixed probabilities for symbols in Monkey Audio version 3.98 | |
313 */ | |
6482 | 314 static const uint16_t counts_3980[22] = { |
5673 | 315 0, 19578, 36160, 48417, 56323, 60899, 63265, 64435, |
316 64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482, | |
6444 | 317 65485, 65488, 65490, 65491, 65492, 65493, |
5673 | 318 }; |
319 | |
320 /** | |
321 * Probability ranges for symbols in Monkey Audio version 3.98 | |
322 */ | |
6444 | 323 static const uint16_t counts_diff_3980[21] = { |
5673 | 324 19578, 16582, 12257, 7906, 4576, 2366, 1170, 536, |
325 261, 119, 65, 31, 19, 10, 6, 3, | |
6444 | 326 3, 2, 1, 1, 1, |
5673 | 327 }; |
328 | |
329 /** | |
330 * Decode symbol | |
331 * @param counts probability range start position | |
332 * @param count_diffs probability range widths | |
333 */ | |
334 static inline int range_get_symbol(APEContext * ctx, | |
6482 | 335 const uint16_t counts[], |
5673 | 336 const uint16_t counts_diff[]) |
337 { | |
338 int symbol, cf; | |
339 | |
340 cf = range_decode_culshift(ctx, 16); | |
341 | |
6443
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
342 if(cf > 65492){ |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
343 symbol= cf - 65535 + 63; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
344 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
|
345 if(cf > 65535) |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
346 ctx->error=1; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
347 return symbol; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
348 } |
5673 | 349 /* figure out the symbol inefficiently; a binary search would be much better */ |
350 for (symbol = 0; counts[symbol + 1] <= cf; symbol++); | |
351 | |
352 range_decode_update(ctx, counts_diff[symbol], counts[symbol]); | |
353 | |
354 return symbol; | |
355 } | |
356 /** @} */ // group rangecoder | |
357 | |
358 static inline void update_rice(APERice *rice, int x) | |
359 { | |
360 rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5); | |
361 | |
362 if (rice->k == 0) | |
363 rice->k = 1; | |
364 else if (rice->ksum < (1 << (rice->k + 4))) | |
365 rice->k--; | |
366 else if (rice->ksum >= (1 << (rice->k + 5))) | |
367 rice->k++; | |
368 } | |
369 | |
370 static inline int ape_decode_value(APEContext * ctx, APERice *rice) | |
371 { | |
372 int x, overflow; | |
373 | |
374 if (ctx->fileversion < 3980) { | |
375 int tmpk; | |
376 | |
377 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970); | |
378 | |
379 if (overflow == (MODEL_ELEMENTS - 1)) { | |
380 tmpk = range_decode_bits(ctx, 5); | |
381 overflow = 0; | |
382 } else | |
383 tmpk = (rice->k < 1) ? 0 : rice->k - 1; | |
384 | |
385 if (tmpk <= 16) | |
386 x = range_decode_bits(ctx, tmpk); | |
387 else { | |
388 x = range_decode_bits(ctx, 16); | |
389 x |= (range_decode_bits(ctx, tmpk - 16) << 16); | |
390 } | |
391 x += overflow << tmpk; | |
392 } else { | |
393 int base, pivot; | |
394 | |
395 pivot = rice->ksum >> 5; | |
396 if (pivot == 0) | |
397 pivot = 1; | |
398 | |
399 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980); | |
400 | |
401 if (overflow == (MODEL_ELEMENTS - 1)) { | |
402 overflow = range_decode_bits(ctx, 16) << 16; | |
403 overflow |= range_decode_bits(ctx, 16); | |
404 } | |
405 | |
406 base = range_decode_culfreq(ctx, pivot); | |
407 range_decode_update(ctx, 1, base); | |
408 | |
409 x = base + overflow * pivot; | |
410 } | |
411 | |
412 update_rice(rice, x); | |
413 | |
414 /* Convert to signed */ | |
415 if (x & 1) | |
416 return (x >> 1) + 1; | |
417 else | |
418 return -(x >> 1); | |
419 } | |
420 | |
421 static void entropy_decode(APEContext * ctx, int blockstodecode, int stereo) | |
422 { | |
423 int32_t *decoded0 = ctx->decoded0; | |
424 int32_t *decoded1 = ctx->decoded1; | |
425 | |
426 ctx->blocksdecoded = blockstodecode; | |
427 | |
428 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
429 /* We are pure silence, just memset the output buffer. */ | |
430 memset(decoded0, 0, blockstodecode * sizeof(int32_t)); | |
431 memset(decoded1, 0, blockstodecode * sizeof(int32_t)); | |
432 } else { | |
433 while (blockstodecode--) { | |
434 *decoded0++ = ape_decode_value(ctx, &ctx->riceY); | |
435 if (stereo) | |
436 *decoded1++ = ape_decode_value(ctx, &ctx->riceX); | |
437 } | |
438 } | |
439 | |
440 if (ctx->blocksdecoded == ctx->currentframeblocks) | |
441 range_dec_normalize(ctx); /* normalize to use up all bytes */ | |
442 } | |
443 | |
444 static void init_entropy_decoder(APEContext * ctx) | |
445 { | |
446 /* Read the CRC */ | |
447 ctx->CRC = bytestream_get_be32(&ctx->ptr); | |
448 | |
449 /* Read the frame flags if they exist */ | |
450 ctx->frameflags = 0; | |
451 if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) { | |
452 ctx->CRC &= ~0x80000000; | |
453 | |
454 ctx->frameflags = bytestream_get_be32(&ctx->ptr); | |
455 } | |
456 | |
457 /* Keep a count of the blocks decoded in this frame */ | |
458 ctx->blocksdecoded = 0; | |
459 | |
5963 | 460 /* Initialize the rice structs */ |
5673 | 461 ctx->riceX.k = 10; |
462 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16; | |
463 ctx->riceY.k = 10; | |
464 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16; | |
465 | |
466 /* The first 8 bits of input are ignored. */ | |
467 ctx->ptr++; | |
468 | |
469 range_start_decoding(ctx); | |
470 } | |
471 | |
472 static const int32_t initial_coeffs[4] = { | |
473 360, 317, -109, 98 | |
474 }; | |
475 | |
476 static void init_predictor_decoder(APEContext * ctx) | |
477 { | |
478 APEPredictor *p = &ctx->predictor; | |
479 | |
480 /* Zero the history buffers */ | |
481 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t)); | |
482 p->buf = p->historybuffer; | |
483 | |
5966 | 484 /* Initialize and zero the coefficients */ |
5673 | 485 memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs)); |
486 memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs)); | |
487 memset(p->coeffsB, 0, sizeof(p->coeffsB)); | |
488 | |
489 p->filterA[0] = p->filterA[1] = 0; | |
490 p->filterB[0] = p->filterB[1] = 0; | |
491 p->lastA[0] = p->lastA[1] = 0; | |
492 } | |
493 | |
494 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */ | |
495 static inline int APESIGN(int32_t x) { | |
496 return (x < 0) - (x > 0); | |
497 } | |
498 | |
499 static int predictor_update_filter(APEPredictor *p, const int decoded, const int filter, const int delayA, const int delayB, const int adaptA, const int adaptB) | |
500 { | |
501 int32_t predictionA, predictionB; | |
502 | |
503 p->buf[delayA] = p->lastA[filter]; | |
504 p->buf[adaptA] = APESIGN(p->buf[delayA]); | |
505 p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1]; | |
506 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]); | |
507 | |
508 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] + | |
509 p->buf[delayA - 1] * p->coeffsA[filter][1] + | |
510 p->buf[delayA - 2] * p->coeffsA[filter][2] + | |
511 p->buf[delayA - 3] * p->coeffsA[filter][3]; | |
512 | |
513 /* Apply a scaled first-order filter compression */ | |
514 p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5); | |
515 p->buf[adaptB] = APESIGN(p->buf[delayB]); | |
516 p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1]; | |
517 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]); | |
518 p->filterB[filter] = p->filterA[filter ^ 1]; | |
519 | |
520 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] + | |
521 p->buf[delayB - 1] * p->coeffsB[filter][1] + | |
522 p->buf[delayB - 2] * p->coeffsB[filter][2] + | |
523 p->buf[delayB - 3] * p->coeffsB[filter][3] + | |
524 p->buf[delayB - 4] * p->coeffsB[filter][4]; | |
525 | |
526 p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10); | |
527 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5); | |
528 | |
529 if (!decoded) // no need updating filter coefficients | |
530 return p->filterA[filter]; | |
531 | |
532 if (decoded > 0) { | |
533 p->coeffsA[filter][0] -= p->buf[adaptA ]; | |
534 p->coeffsA[filter][1] -= p->buf[adaptA - 1]; | |
535 p->coeffsA[filter][2] -= p->buf[adaptA - 2]; | |
536 p->coeffsA[filter][3] -= p->buf[adaptA - 3]; | |
537 | |
538 p->coeffsB[filter][0] -= p->buf[adaptB ]; | |
539 p->coeffsB[filter][1] -= p->buf[adaptB - 1]; | |
540 p->coeffsB[filter][2] -= p->buf[adaptB - 2]; | |
541 p->coeffsB[filter][3] -= p->buf[adaptB - 3]; | |
542 p->coeffsB[filter][4] -= p->buf[adaptB - 4]; | |
543 } else { | |
544 p->coeffsA[filter][0] += p->buf[adaptA ]; | |
545 p->coeffsA[filter][1] += p->buf[adaptA - 1]; | |
546 p->coeffsA[filter][2] += p->buf[adaptA - 2]; | |
547 p->coeffsA[filter][3] += p->buf[adaptA - 3]; | |
548 | |
549 p->coeffsB[filter][0] += p->buf[adaptB ]; | |
550 p->coeffsB[filter][1] += p->buf[adaptB - 1]; | |
551 p->coeffsB[filter][2] += p->buf[adaptB - 2]; | |
552 p->coeffsB[filter][3] += p->buf[adaptB - 3]; | |
553 p->coeffsB[filter][4] += p->buf[adaptB - 4]; | |
554 } | |
555 return p->filterA[filter]; | |
556 } | |
557 | |
558 static void predictor_decode_stereo(APEContext * ctx, int count) | |
559 { | |
560 int32_t predictionA, predictionB; | |
561 APEPredictor *p = &ctx->predictor; | |
562 int32_t *decoded0 = ctx->decoded0; | |
563 int32_t *decoded1 = ctx->decoded1; | |
564 | |
565 while (count--) { | |
566 /* Predictor Y */ | |
567 predictionA = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB, YADAPTCOEFFSA, YADAPTCOEFFSB); | |
568 predictionB = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB, XADAPTCOEFFSA, XADAPTCOEFFSB); | |
569 *(decoded0++) = predictionA; | |
570 *(decoded1++) = predictionB; | |
571 | |
572 /* Combined */ | |
573 p->buf++; | |
574 | |
575 /* Have we filled the history buffer? */ | |
576 if (p->buf == p->historybuffer + HISTORY_SIZE) { | |
577 memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t)); | |
578 p->buf = p->historybuffer; | |
579 } | |
580 } | |
581 } | |
582 | |
583 static void predictor_decode_mono(APEContext * ctx, int count) | |
584 { | |
585 APEPredictor *p = &ctx->predictor; | |
586 int32_t *decoded0 = ctx->decoded0; | |
587 int32_t predictionA, currentA, A; | |
588 | |
589 currentA = p->lastA[0]; | |
590 | |
591 while (count--) { | |
592 A = *decoded0; | |
593 | |
594 p->buf[YDELAYA] = currentA; | |
595 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1]; | |
596 | |
597 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] + | |
598 p->buf[YDELAYA - 1] * p->coeffsA[0][1] + | |
599 p->buf[YDELAYA - 2] * p->coeffsA[0][2] + | |
600 p->buf[YDELAYA - 3] * p->coeffsA[0][3]; | |
601 | |
602 currentA = A + (predictionA >> 10); | |
603 | |
604 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]); | |
605 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]); | |
606 | |
607 if (A > 0) { | |
608 p->coeffsA[0][0] -= p->buf[YADAPTCOEFFSA ]; | |
609 p->coeffsA[0][1] -= p->buf[YADAPTCOEFFSA - 1]; | |
610 p->coeffsA[0][2] -= p->buf[YADAPTCOEFFSA - 2]; | |
611 p->coeffsA[0][3] -= p->buf[YADAPTCOEFFSA - 3]; | |
612 } else if (A < 0) { | |
613 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ]; | |
614 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1]; | |
615 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2]; | |
616 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3]; | |
617 } | |
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 | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
652 static inline 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 */ | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
659 res = (ctx->dsp.scalarproduct_int16(f->delay - order, f->coeffs, order, 0) + (1 << (fracbits - 1))) >> fracbits; |
5673 | 660 |
661 if (*data < 0) | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
662 ctx->dsp.add_int16(f->coeffs, f->adaptcoeffs - order, order); |
5673 | 663 else if (*data > 0) |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
664 ctx->dsp.sub_int16(f->coeffs, f->adaptcoeffs - order, order); |
5673 | 665 |
666 res += *data; | |
667 | |
668 *data++ = res; | |
669 | |
670 /* Update the output history */ | |
671 *f->delay++ = av_clip_int16(res); | |
672 | |
673 if (version < 3980) { | |
674 /* Version ??? to < 3.98 files (untested) */ | |
675 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4; | |
676 f->adaptcoeffs[-4] >>= 1; | |
677 f->adaptcoeffs[-8] >>= 1; | |
678 } else { | |
679 /* Version 3.98 and later files */ | |
680 | |
681 /* Update the adaption coefficients */ | |
682 absres = (res < 0 ? -res : res); | |
683 | |
684 if (absres > (f->avg * 3)) | |
685 *f->adaptcoeffs = ((res >> 25) & 64) - 32; | |
686 else if (absres > (f->avg * 4) / 3) | |
687 *f->adaptcoeffs = ((res >> 26) & 32) - 16; | |
688 else if (absres > 0) | |
689 *f->adaptcoeffs = ((res >> 27) & 16) - 8; | |
690 else | |
691 *f->adaptcoeffs = 0; | |
692 | |
693 f->avg += (absres - f->avg) / 16; | |
694 | |
695 f->adaptcoeffs[-1] >>= 1; | |
696 f->adaptcoeffs[-2] >>= 1; | |
697 f->adaptcoeffs[-8] >>= 1; | |
698 } | |
699 | |
700 f->adaptcoeffs++; | |
701 | |
702 /* Have we filled the history buffer? */ | |
703 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) { | |
704 memmove(f->historybuffer, f->delay - (order * 2), | |
705 (order * 2) * sizeof(int16_t)); | |
706 f->delay = f->historybuffer + order * 2; | |
707 f->adaptcoeffs = f->historybuffer + order; | |
708 } | |
709 } | |
710 } | |
711 | |
712 static void apply_filter(APEContext * ctx, APEFilter *f, | |
713 int32_t * data0, int32_t * data1, | |
714 int count, int order, int fracbits) | |
715 { | |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
716 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits); |
5673 | 717 if (data1) |
7203
87b1dfb5a98d
Add several vector functions used by Monkey's Audio decoder to dsputil
kostya
parents:
7040
diff
changeset
|
718 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits); |
5673 | 719 } |
720 | |
721 static void ape_apply_filters(APEContext * ctx, int32_t * decoded0, | |
722 int32_t * decoded1, int count) | |
723 { | |
724 int i; | |
725 | |
726 for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
727 if (!ape_filter_orders[ctx->fset][i]) | |
728 break; | |
729 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count, ape_filter_orders[ctx->fset][i], ape_filter_fracbits[ctx->fset][i]); | |
730 } | |
731 } | |
732 | |
733 static void init_frame_decoder(APEContext * ctx) | |
734 { | |
735 int i; | |
736 init_entropy_decoder(ctx); | |
737 init_predictor_decoder(ctx); | |
738 | |
739 for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
740 if (!ape_filter_orders[ctx->fset][i]) | |
741 break; | |
742 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i], ape_filter_orders[ctx->fset][i]); | |
743 } | |
744 } | |
745 | |
746 static void ape_unpack_mono(APEContext * ctx, int count) | |
747 { | |
748 int32_t left; | |
749 int32_t *decoded0 = ctx->decoded0; | |
750 int32_t *decoded1 = ctx->decoded1; | |
751 | |
752 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
753 entropy_decode(ctx, count, 0); | |
754 /* We are pure silence, so we're done. */ | |
755 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n"); | |
756 return; | |
757 } | |
758 | |
759 entropy_decode(ctx, count, 0); | |
760 ape_apply_filters(ctx, decoded0, NULL, count); | |
761 | |
762 /* Now apply the predictor decoding */ | |
763 predictor_decode_mono(ctx, count); | |
764 | |
765 /* Pseudo-stereo - just copy left channel to right channel */ | |
766 if (ctx->channels == 2) { | |
767 while (count--) { | |
768 left = *decoded0; | |
769 *(decoded1++) = *(decoded0++) = left; | |
770 } | |
771 } | |
772 } | |
773 | |
774 static void ape_unpack_stereo(APEContext * ctx, int count) | |
775 { | |
776 int32_t left, right; | |
777 int32_t *decoded0 = ctx->decoded0; | |
778 int32_t *decoded1 = ctx->decoded1; | |
779 | |
780 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
781 /* We are pure silence, so we're done. */ | |
782 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n"); | |
783 return; | |
784 } | |
785 | |
786 entropy_decode(ctx, count, 1); | |
787 ape_apply_filters(ctx, decoded0, decoded1, count); | |
788 | |
789 /* Now apply the predictor decoding */ | |
790 predictor_decode_stereo(ctx, count); | |
791 | |
792 /* Decorrelate and scale to output depth */ | |
793 while (count--) { | |
794 left = *decoded1 - (*decoded0 / 2); | |
795 right = left + *decoded0; | |
796 | |
797 *(decoded0++) = left; | |
798 *(decoded1++) = right; | |
799 } | |
800 } | |
801 | |
802 static int ape_decode_frame(AVCodecContext * avctx, | |
803 void *data, int *data_size, | |
6223 | 804 const uint8_t * buf, int buf_size) |
5673 | 805 { |
806 APEContext *s = avctx->priv_data; | |
807 int16_t *samples = data; | |
808 int nblocks; | |
809 int i, n; | |
810 int blockstodecode; | |
811 int bytes_used; | |
812 | |
813 if (buf_size == 0 && !s->samples) { | |
814 *data_size = 0; | |
815 return 0; | |
816 } | |
817 | |
818 /* should not happen but who knows */ | |
819 if (BLOCKS_PER_LOOP * 2 * avctx->channels > *data_size) { | |
820 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); | |
821 return -1; | |
822 } | |
823 | |
824 if(!s->samples){ | |
825 s->data = av_realloc(s->data, (buf_size + 3) & ~3); | |
6223 | 826 s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2); |
5673 | 827 s->ptr = s->last_ptr = s->data; |
828 s->data_end = s->data + buf_size; | |
829 | |
830 nblocks = s->samples = bytestream_get_be32(&s->ptr); | |
831 n = bytestream_get_be32(&s->ptr); | |
832 if(n < 0 || n > 3){ | |
833 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n"); | |
834 s->data = NULL; | |
835 return -1; | |
836 } | |
837 s->ptr += n; | |
838 | |
839 s->currentframeblocks = nblocks; | |
840 buf += 4; | |
841 if (s->samples <= 0) { | |
842 *data_size = 0; | |
843 return buf_size; | |
844 } | |
845 | |
846 memset(s->decoded0, 0, sizeof(s->decoded0)); | |
847 memset(s->decoded1, 0, sizeof(s->decoded1)); | |
848 | |
849 /* Initialize the frame decoder */ | |
850 init_frame_decoder(s); | |
851 } | |
852 | |
853 if (!s->data) { | |
854 *data_size = 0; | |
855 return buf_size; | |
856 } | |
857 | |
858 nblocks = s->samples; | |
859 blockstodecode = FFMIN(BLOCKS_PER_LOOP, nblocks); | |
860 | |
6443
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
861 s->error=0; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
862 |
5673 | 863 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO)) |
864 ape_unpack_mono(s, blockstodecode); | |
865 else | |
866 ape_unpack_stereo(s, blockstodecode); | |
867 | |
6443
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
868 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
|
869 s->samples=0; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
870 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
|
871 return -1; |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
872 } |
e3adb7e96812
Detect and prevent reading over the end of counts_*. We pass the error
michael
parents:
6442
diff
changeset
|
873 |
5673 | 874 for (i = 0; i < blockstodecode; i++) { |
875 *samples++ = s->decoded0[i]; | |
876 if(s->channels == 2) | |
877 *samples++ = s->decoded1[i]; | |
878 } | |
879 | |
880 s->samples -= blockstodecode; | |
881 | |
882 *data_size = blockstodecode * 2 * s->channels; | |
883 bytes_used = s->samples ? s->ptr - s->last_ptr : buf_size; | |
884 s->last_ptr = s->ptr; | |
885 return bytes_used; | |
886 } | |
887 | |
888 AVCodec ape_decoder = { | |
889 "ape", | |
890 CODEC_TYPE_AUDIO, | |
891 CODEC_ID_APE, | |
892 sizeof(APEContext), | |
893 ape_decode_init, | |
894 NULL, | |
895 ape_decode_close, | |
896 ape_decode_frame, | |
7040
e943e1409077
Make AVCodec long_names definition conditional depending on CONFIG_SMALL.
stefano
parents:
6712
diff
changeset
|
897 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"), |
5673 | 898 }; |