Commit | Line | Data |
---|---|---|
bf4a1f17 KS |
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 */ | |
82 | static const uint16_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = { | |
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 | |
157 | uint8_t *ptr; ///< current position in frame data | |
158 | uint8_t *last_ptr; ///< position where last 4608-sample block ended | |
159 | } APEContext; | |
160 | ||
161 | // TODO: dsputilize | |
162 | static inline void vector_add(int16_t * v1, int16_t * v2, int order) | |
163 | { | |
164 | while (order--) | |
165 | *v1++ += *v2++; | |
166 | } | |
167 | ||
168 | // TODO: dsputilize | |
169 | static inline void vector_sub(int16_t * v1, int16_t * v2, int order) | |
170 | { | |
171 | while (order--) | |
172 | *v1++ -= *v2++; | |
173 | } | |
174 | ||
175 | // TODO: dsputilize | |
176 | static inline int32_t scalarproduct(int16_t * v1, int16_t * v2, int order) | |
177 | { | |
178 | int res = 0; | |
179 | ||
180 | while (order--) | |
181 | res += *v1++ * *v2++; | |
182 | ||
183 | return res; | |
184 | } | |
185 | ||
186 | static int ape_decode_init(AVCodecContext * avctx) | |
187 | { | |
188 | APEContext *s = avctx->priv_data; | |
189 | int i; | |
190 | ||
191 | if (avctx->extradata_size != 6) { | |
192 | av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n"); | |
193 | return -1; | |
194 | } | |
195 | if (avctx->bits_per_sample != 16) { | |
196 | av_log(avctx, AV_LOG_ERROR, "Only 16-bit samples are supported\n"); | |
197 | return -1; | |
198 | } | |
199 | if (avctx->channels > 2) { | |
200 | av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n"); | |
201 | return -1; | |
202 | } | |
203 | s->avctx = avctx; | |
204 | s->channels = avctx->channels; | |
205 | s->fileversion = AV_RL16(avctx->extradata); | |
206 | s->compression_level = AV_RL16(avctx->extradata + 2); | |
207 | s->flags = AV_RL16(avctx->extradata + 4); | |
208 | ||
209 | av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n", s->compression_level, s->flags); | |
210 | if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) { | |
211 | av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n", s->compression_level); | |
212 | return -1; | |
213 | } | |
214 | s->fset = s->compression_level / 1000 - 1; | |
215 | for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
216 | if (!ape_filter_orders[s->fset][i]) | |
217 | break; | |
218 | s->filterbuf[i] = av_malloc((ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4); | |
219 | } | |
220 | ||
221 | dsputil_init(&s->dsp, avctx); | |
222 | return 0; | |
223 | } | |
224 | ||
225 | static int ape_decode_close(AVCodecContext * avctx) | |
226 | { | |
227 | APEContext *s = avctx->priv_data; | |
228 | int i; | |
229 | ||
230 | for (i = 0; i < APE_FILTER_LEVELS; i++) | |
231 | av_freep(&s->filterbuf[i]); | |
232 | ||
233 | return 0; | |
234 | } | |
235 | ||
236 | /** | |
237 | * @defgroup rangecoder APE range decoder | |
238 | * @{ | |
239 | */ | |
240 | ||
241 | #define CODE_BITS 32 | |
242 | #define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1)) | |
243 | #define SHIFT_BITS (CODE_BITS - 9) | |
244 | #define EXTRA_BITS ((CODE_BITS-2) % 8 + 1) | |
245 | #define BOTTOM_VALUE (TOP_VALUE >> 8) | |
246 | ||
247 | /** Start the decoder */ | |
248 | static inline void range_start_decoding(APEContext * ctx) | |
249 | { | |
250 | ctx->rc.buffer = bytestream_get_byte(&ctx->ptr); | |
251 | ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS); | |
252 | ctx->rc.range = (uint32_t) 1 << EXTRA_BITS; | |
253 | } | |
254 | ||
255 | /** Perform normalization */ | |
256 | static inline void range_dec_normalize(APEContext * ctx) | |
257 | { | |
258 | while (ctx->rc.range <= BOTTOM_VALUE) { | |
259 | ctx->rc.buffer = (ctx->rc.buffer << 8) | bytestream_get_byte(&ctx->ptr); | |
260 | ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF); | |
261 | ctx->rc.range <<= 8; | |
262 | } | |
263 | } | |
264 | ||
265 | /** | |
266 | * Calculate culmulative frequency for next symbol. Does NO update! | |
267 | * @param tot_f is the total frequency or (code_value)1<<shift | |
268 | * @return the culmulative frequency | |
269 | */ | |
270 | static inline int range_decode_culfreq(APEContext * ctx, int tot_f) | |
271 | { | |
272 | range_dec_normalize(ctx); | |
273 | ctx->rc.help = ctx->rc.range / tot_f; | |
274 | return ctx->rc.low / ctx->rc.help; | |
275 | } | |
276 | ||
277 | /** | |
278 | * Decode value with given size in bits | |
279 | * @param shift number of bits to decode | |
280 | */ | |
281 | static inline int range_decode_culshift(APEContext * ctx, int shift) | |
282 | { | |
283 | range_dec_normalize(ctx); | |
284 | ctx->rc.help = ctx->rc.range >> shift; | |
285 | return ctx->rc.low / ctx->rc.help; | |
286 | } | |
287 | ||
288 | ||
289 | /** | |
290 | * Update decoding state | |
291 | * @param sy_f the interval length (frequency of the symbol) | |
292 | * @param lt_f the lower end (frequency sum of < symbols) | |
293 | */ | |
294 | static inline void range_decode_update(APEContext * ctx, int sy_f, int lt_f) | |
295 | { | |
296 | ctx->rc.low -= ctx->rc.help * lt_f; | |
297 | ctx->rc.range = ctx->rc.help * sy_f; | |
298 | } | |
299 | ||
300 | /** Decode n bits (n <= 16) without modelling */ | |
301 | static inline int range_decode_bits(APEContext * ctx, int n) | |
302 | { | |
303 | int sym = range_decode_culshift(ctx, n); | |
304 | range_decode_update(ctx, 1, sym); | |
305 | return sym; | |
306 | } | |
307 | ||
308 | ||
309 | #define MODEL_ELEMENTS 64 | |
310 | ||
311 | /** | |
312 | * Fixed probabilities for symbols in Monkey Audio version 3.97 | |
313 | */ | |
314 | static const uint32_t counts_3970[65] = { | |
315 | 0, 14824, 28224, 39348, 47855, 53994, 58171, 60926, | |
316 | 62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419, | |
317 | 65450, 65469, 65480, 65487, 65491, 65493, 65494, 65495, | |
318 | 65496, 65497, 65498, 65499, 65500, 65501, 65502, 65503, | |
319 | 65504, 65505, 65506, 65507, 65508, 65509, 65510, 65511, | |
320 | 65512, 65513, 65514, 65515, 65516, 65517, 65518, 65519, | |
321 | 65520, 65521, 65522, 65523, 65524, 65525, 65526, 65527, | |
322 | 65528, 65529, 65530, 65531, 65532, 65533, 65534, 65535, | |
323 | 65536 | |
324 | }; | |
325 | ||
326 | /** | |
327 | * Probability ranges for symbols in Monkey Audio version 3.97 | |
328 | */ | |
329 | static const uint16_t counts_diff_3970[64] = { | |
330 | 14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756, | |
331 | 1104, 677, 415, 248, 150, 89, 54, 31, | |
332 | 19, 11, 7, 4, 2, 1, 1, 1, | |
333 | 1, 1, 1, 1, 1, 1, 1, 1, | |
334 | 1, 1, 1, 1, 1, 1, 1, 1, | |
335 | 1, 1, 1, 1, 1, 1, 1, 1, | |
336 | 1, 1, 1, 1, 1, 1, 1, 1, | |
337 | 1, 1, 1, 1, 1, 1, 1, 1 | |
338 | }; | |
339 | ||
340 | /** | |
341 | * Fixed probabilities for symbols in Monkey Audio version 3.98 | |
342 | */ | |
343 | static const uint32_t counts_3980[65] = { | |
344 | 0, 19578, 36160, 48417, 56323, 60899, 63265, 64435, | |
345 | 64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482, | |
346 | 65485, 65488, 65490, 65491, 65492, 65493, 65494, 65495, | |
347 | 65496, 65497, 65498, 65499, 65500, 65501, 65502, 65503, | |
348 | 65504, 65505, 65506, 65507, 65508, 65509, 65510, 65511, | |
349 | 65512, 65513, 65514, 65515, 65516, 65517, 65518, 65519, | |
350 | 65520, 65521, 65522, 65523, 65524, 65525, 65526, 65527, | |
351 | 65528, 65529, 65530, 65531, 65532, 65533, 65534, 65535, | |
352 | 65536 | |
353 | }; | |
354 | ||
355 | /** | |
356 | * Probability ranges for symbols in Monkey Audio version 3.98 | |
357 | */ | |
358 | static const uint16_t counts_diff_3980[64] = { | |
359 | 19578, 16582, 12257, 7906, 4576, 2366, 1170, 536, | |
360 | 261, 119, 65, 31, 19, 10, 6, 3, | |
361 | 3, 2, 1, 1, 1, 1, 1, 1, | |
362 | 1, 1, 1, 1, 1, 1, 1, 1, | |
363 | 1, 1, 1, 1, 1, 1, 1, 1, | |
364 | 1, 1, 1, 1, 1, 1, 1, 1, | |
365 | 1, 1, 1, 1, 1, 1, 1, 1, | |
366 | 1, 1, 1, 1, 1, 1, 1, 1 | |
367 | }; | |
368 | ||
369 | /** | |
370 | * Decode symbol | |
371 | * @param counts probability range start position | |
372 | * @param count_diffs probability range widths | |
373 | */ | |
374 | static inline int range_get_symbol(APEContext * ctx, | |
375 | const uint32_t counts[], | |
376 | const uint16_t counts_diff[]) | |
377 | { | |
378 | int symbol, cf; | |
379 | ||
380 | cf = range_decode_culshift(ctx, 16); | |
381 | ||
382 | /* figure out the symbol inefficiently; a binary search would be much better */ | |
383 | for (symbol = 0; counts[symbol + 1] <= cf; symbol++); | |
384 | ||
385 | range_decode_update(ctx, counts_diff[symbol], counts[symbol]); | |
386 | ||
387 | return symbol; | |
388 | } | |
389 | /** @} */ // group rangecoder | |
390 | ||
391 | static inline void update_rice(APERice *rice, int x) | |
392 | { | |
393 | rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5); | |
394 | ||
395 | if (rice->k == 0) | |
396 | rice->k = 1; | |
397 | else if (rice->ksum < (1 << (rice->k + 4))) | |
398 | rice->k--; | |
399 | else if (rice->ksum >= (1 << (rice->k + 5))) | |
400 | rice->k++; | |
401 | } | |
402 | ||
403 | static inline int ape_decode_value(APEContext * ctx, APERice *rice) | |
404 | { | |
405 | int x, overflow; | |
406 | ||
407 | if (ctx->fileversion < 3980) { | |
408 | int tmpk; | |
409 | ||
410 | overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970); | |
411 | ||
412 | if (overflow == (MODEL_ELEMENTS - 1)) { | |
413 | tmpk = range_decode_bits(ctx, 5); | |
414 | overflow = 0; | |
415 | } else | |
416 | tmpk = (rice->k < 1) ? 0 : rice->k - 1; | |
417 | ||
418 | if (tmpk <= 16) | |
419 | x = range_decode_bits(ctx, tmpk); | |
420 | else { | |
421 | x = range_decode_bits(ctx, 16); | |
422 | x |= (range_decode_bits(ctx, tmpk - 16) << 16); | |
423 | } | |
424 | x += overflow << tmpk; | |
425 | } else { | |
426 | int base, pivot; | |
427 | ||
428 | pivot = rice->ksum >> 5; | |
429 | if (pivot == 0) | |
430 | pivot = 1; | |
431 | ||
432 | overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980); | |
433 | ||
434 | if (overflow == (MODEL_ELEMENTS - 1)) { | |
435 | overflow = range_decode_bits(ctx, 16) << 16; | |
436 | overflow |= range_decode_bits(ctx, 16); | |
437 | } | |
438 | ||
439 | base = range_decode_culfreq(ctx, pivot); | |
440 | range_decode_update(ctx, 1, base); | |
441 | ||
442 | x = base + overflow * pivot; | |
443 | } | |
444 | ||
445 | update_rice(rice, x); | |
446 | ||
447 | /* Convert to signed */ | |
448 | if (x & 1) | |
449 | return (x >> 1) + 1; | |
450 | else | |
451 | return -(x >> 1); | |
452 | } | |
453 | ||
454 | static void entropy_decode(APEContext * ctx, int blockstodecode, int stereo) | |
455 | { | |
456 | int32_t *decoded0 = ctx->decoded0; | |
457 | int32_t *decoded1 = ctx->decoded1; | |
458 | ||
459 | ctx->blocksdecoded = blockstodecode; | |
460 | ||
461 | if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
462 | /* We are pure silence, just memset the output buffer. */ | |
463 | memset(decoded0, 0, blockstodecode * sizeof(int32_t)); | |
464 | memset(decoded1, 0, blockstodecode * sizeof(int32_t)); | |
465 | } else { | |
466 | while (blockstodecode--) { | |
467 | *decoded0++ = ape_decode_value(ctx, &ctx->riceY); | |
468 | if (stereo) | |
469 | *decoded1++ = ape_decode_value(ctx, &ctx->riceX); | |
470 | } | |
471 | } | |
472 | ||
473 | if (ctx->blocksdecoded == ctx->currentframeblocks) | |
474 | range_dec_normalize(ctx); /* normalize to use up all bytes */ | |
475 | } | |
476 | ||
477 | static void init_entropy_decoder(APEContext * ctx) | |
478 | { | |
479 | /* Read the CRC */ | |
480 | ctx->CRC = bytestream_get_be32(&ctx->ptr); | |
481 | ||
482 | /* Read the frame flags if they exist */ | |
483 | ctx->frameflags = 0; | |
484 | if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) { | |
485 | ctx->CRC &= ~0x80000000; | |
486 | ||
487 | ctx->frameflags = bytestream_get_be32(&ctx->ptr); | |
488 | } | |
489 | ||
490 | /* Keep a count of the blocks decoded in this frame */ | |
491 | ctx->blocksdecoded = 0; | |
492 | ||
52b541ad | 493 | /* Initialize the rice structs */ |
bf4a1f17 KS |
494 | ctx->riceX.k = 10; |
495 | ctx->riceX.ksum = (1 << ctx->riceX.k) * 16; | |
496 | ctx->riceY.k = 10; | |
497 | ctx->riceY.ksum = (1 << ctx->riceY.k) * 16; | |
498 | ||
499 | /* The first 8 bits of input are ignored. */ | |
500 | ctx->ptr++; | |
501 | ||
502 | range_start_decoding(ctx); | |
503 | } | |
504 | ||
505 | static const int32_t initial_coeffs[4] = { | |
506 | 360, 317, -109, 98 | |
507 | }; | |
508 | ||
509 | static void init_predictor_decoder(APEContext * ctx) | |
510 | { | |
511 | APEPredictor *p = &ctx->predictor; | |
512 | ||
513 | /* Zero the history buffers */ | |
514 | memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t)); | |
515 | p->buf = p->historybuffer; | |
516 | ||
52b541ad | 517 | /* Initialize and zero the co-efficients */ |
bf4a1f17 KS |
518 | memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs)); |
519 | memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs)); | |
520 | memset(p->coeffsB, 0, sizeof(p->coeffsB)); | |
521 | ||
522 | p->filterA[0] = p->filterA[1] = 0; | |
523 | p->filterB[0] = p->filterB[1] = 0; | |
524 | p->lastA[0] = p->lastA[1] = 0; | |
525 | } | |
526 | ||
527 | /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */ | |
528 | static inline int APESIGN(int32_t x) { | |
529 | return (x < 0) - (x > 0); | |
530 | } | |
531 | ||
532 | 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) | |
533 | { | |
534 | int32_t predictionA, predictionB; | |
535 | ||
536 | p->buf[delayA] = p->lastA[filter]; | |
537 | p->buf[adaptA] = APESIGN(p->buf[delayA]); | |
538 | p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1]; | |
539 | p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]); | |
540 | ||
541 | predictionA = p->buf[delayA ] * p->coeffsA[filter][0] + | |
542 | p->buf[delayA - 1] * p->coeffsA[filter][1] + | |
543 | p->buf[delayA - 2] * p->coeffsA[filter][2] + | |
544 | p->buf[delayA - 3] * p->coeffsA[filter][3]; | |
545 | ||
546 | /* Apply a scaled first-order filter compression */ | |
547 | p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5); | |
548 | p->buf[adaptB] = APESIGN(p->buf[delayB]); | |
549 | p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1]; | |
550 | p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]); | |
551 | p->filterB[filter] = p->filterA[filter ^ 1]; | |
552 | ||
553 | predictionB = p->buf[delayB ] * p->coeffsB[filter][0] + | |
554 | p->buf[delayB - 1] * p->coeffsB[filter][1] + | |
555 | p->buf[delayB - 2] * p->coeffsB[filter][2] + | |
556 | p->buf[delayB - 3] * p->coeffsB[filter][3] + | |
557 | p->buf[delayB - 4] * p->coeffsB[filter][4]; | |
558 | ||
559 | p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10); | |
560 | p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5); | |
561 | ||
562 | if (!decoded) // no need updating filter coefficients | |
563 | return p->filterA[filter]; | |
564 | ||
565 | if (decoded > 0) { | |
566 | p->coeffsA[filter][0] -= p->buf[adaptA ]; | |
567 | p->coeffsA[filter][1] -= p->buf[adaptA - 1]; | |
568 | p->coeffsA[filter][2] -= p->buf[adaptA - 2]; | |
569 | p->coeffsA[filter][3] -= p->buf[adaptA - 3]; | |
570 | ||
571 | p->coeffsB[filter][0] -= p->buf[adaptB ]; | |
572 | p->coeffsB[filter][1] -= p->buf[adaptB - 1]; | |
573 | p->coeffsB[filter][2] -= p->buf[adaptB - 2]; | |
574 | p->coeffsB[filter][3] -= p->buf[adaptB - 3]; | |
575 | p->coeffsB[filter][4] -= p->buf[adaptB - 4]; | |
576 | } else { | |
577 | p->coeffsA[filter][0] += p->buf[adaptA ]; | |
578 | p->coeffsA[filter][1] += p->buf[adaptA - 1]; | |
579 | p->coeffsA[filter][2] += p->buf[adaptA - 2]; | |
580 | p->coeffsA[filter][3] += p->buf[adaptA - 3]; | |
581 | ||
582 | p->coeffsB[filter][0] += p->buf[adaptB ]; | |
583 | p->coeffsB[filter][1] += p->buf[adaptB - 1]; | |
584 | p->coeffsB[filter][2] += p->buf[adaptB - 2]; | |
585 | p->coeffsB[filter][3] += p->buf[adaptB - 3]; | |
586 | p->coeffsB[filter][4] += p->buf[adaptB - 4]; | |
587 | } | |
588 | return p->filterA[filter]; | |
589 | } | |
590 | ||
591 | static void predictor_decode_stereo(APEContext * ctx, int count) | |
592 | { | |
593 | int32_t predictionA, predictionB; | |
594 | APEPredictor *p = &ctx->predictor; | |
595 | int32_t *decoded0 = ctx->decoded0; | |
596 | int32_t *decoded1 = ctx->decoded1; | |
597 | ||
598 | while (count--) { | |
599 | /* Predictor Y */ | |
600 | predictionA = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB, YADAPTCOEFFSA, YADAPTCOEFFSB); | |
601 | predictionB = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB, XADAPTCOEFFSA, XADAPTCOEFFSB); | |
602 | *(decoded0++) = predictionA; | |
603 | *(decoded1++) = predictionB; | |
604 | ||
605 | /* Combined */ | |
606 | p->buf++; | |
607 | ||
608 | /* Have we filled the history buffer? */ | |
609 | if (p->buf == p->historybuffer + HISTORY_SIZE) { | |
610 | memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t)); | |
611 | p->buf = p->historybuffer; | |
612 | } | |
613 | } | |
614 | } | |
615 | ||
616 | static void predictor_decode_mono(APEContext * ctx, int count) | |
617 | { | |
618 | APEPredictor *p = &ctx->predictor; | |
619 | int32_t *decoded0 = ctx->decoded0; | |
620 | int32_t predictionA, currentA, A; | |
621 | ||
622 | currentA = p->lastA[0]; | |
623 | ||
624 | while (count--) { | |
625 | A = *decoded0; | |
626 | ||
627 | p->buf[YDELAYA] = currentA; | |
628 | p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1]; | |
629 | ||
630 | predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] + | |
631 | p->buf[YDELAYA - 1] * p->coeffsA[0][1] + | |
632 | p->buf[YDELAYA - 2] * p->coeffsA[0][2] + | |
633 | p->buf[YDELAYA - 3] * p->coeffsA[0][3]; | |
634 | ||
635 | currentA = A + (predictionA >> 10); | |
636 | ||
637 | p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]); | |
638 | p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]); | |
639 | ||
640 | if (A > 0) { | |
641 | p->coeffsA[0][0] -= p->buf[YADAPTCOEFFSA ]; | |
642 | p->coeffsA[0][1] -= p->buf[YADAPTCOEFFSA - 1]; | |
643 | p->coeffsA[0][2] -= p->buf[YADAPTCOEFFSA - 2]; | |
644 | p->coeffsA[0][3] -= p->buf[YADAPTCOEFFSA - 3]; | |
645 | } else if (A < 0) { | |
646 | p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ]; | |
647 | p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1]; | |
648 | p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2]; | |
649 | p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3]; | |
650 | } | |
651 | ||
652 | p->buf++; | |
653 | ||
654 | /* Have we filled the history buffer? */ | |
655 | if (p->buf == p->historybuffer + HISTORY_SIZE) { | |
656 | memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t)); | |
657 | p->buf = p->historybuffer; | |
658 | } | |
659 | ||
660 | p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5); | |
661 | *(decoded0++) = p->filterA[0]; | |
662 | } | |
663 | ||
664 | p->lastA[0] = currentA; | |
665 | } | |
666 | ||
667 | static void do_init_filter(APEFilter *f, int16_t * buf, int order) | |
668 | { | |
669 | f->coeffs = buf; | |
670 | f->historybuffer = buf + order; | |
671 | f->delay = f->historybuffer + order * 2; | |
672 | f->adaptcoeffs = f->historybuffer + order; | |
673 | ||
674 | memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t)); | |
675 | memset(f->coeffs, 0, order * sizeof(int16_t)); | |
676 | f->avg = 0; | |
677 | } | |
678 | ||
679 | static void init_filter(APEContext * ctx, APEFilter *f, int16_t * buf, int order) | |
680 | { | |
681 | do_init_filter(&f[0], buf, order); | |
682 | do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order); | |
683 | } | |
684 | ||
685 | static inline void do_apply_filter(int version, APEFilter *f, int32_t *data, int count, int order, int fracbits) | |
686 | { | |
687 | int res; | |
688 | int absres; | |
689 | ||
690 | while (count--) { | |
691 | /* round fixedpoint scalar product */ | |
692 | res = (scalarproduct(f->delay - order, f->coeffs, order) + (1 << (fracbits - 1))) >> fracbits; | |
693 | ||
694 | if (*data < 0) | |
695 | vector_add(f->coeffs, f->adaptcoeffs - order, order); | |
696 | else if (*data > 0) | |
697 | vector_sub(f->coeffs, f->adaptcoeffs - order, order); | |
698 | ||
699 | res += *data; | |
700 | ||
701 | *data++ = res; | |
702 | ||
703 | /* Update the output history */ | |
704 | *f->delay++ = av_clip_int16(res); | |
705 | ||
706 | if (version < 3980) { | |
707 | /* Version ??? to < 3.98 files (untested) */ | |
708 | f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4; | |
709 | f->adaptcoeffs[-4] >>= 1; | |
710 | f->adaptcoeffs[-8] >>= 1; | |
711 | } else { | |
712 | /* Version 3.98 and later files */ | |
713 | ||
714 | /* Update the adaption coefficients */ | |
715 | absres = (res < 0 ? -res : res); | |
716 | ||
717 | if (absres > (f->avg * 3)) | |
718 | *f->adaptcoeffs = ((res >> 25) & 64) - 32; | |
719 | else if (absres > (f->avg * 4) / 3) | |
720 | *f->adaptcoeffs = ((res >> 26) & 32) - 16; | |
721 | else if (absres > 0) | |
722 | *f->adaptcoeffs = ((res >> 27) & 16) - 8; | |
723 | else | |
724 | *f->adaptcoeffs = 0; | |
725 | ||
726 | f->avg += (absres - f->avg) / 16; | |
727 | ||
728 | f->adaptcoeffs[-1] >>= 1; | |
729 | f->adaptcoeffs[-2] >>= 1; | |
730 | f->adaptcoeffs[-8] >>= 1; | |
731 | } | |
732 | ||
733 | f->adaptcoeffs++; | |
734 | ||
735 | /* Have we filled the history buffer? */ | |
736 | if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) { | |
737 | memmove(f->historybuffer, f->delay - (order * 2), | |
738 | (order * 2) * sizeof(int16_t)); | |
739 | f->delay = f->historybuffer + order * 2; | |
740 | f->adaptcoeffs = f->historybuffer + order; | |
741 | } | |
742 | } | |
743 | } | |
744 | ||
745 | static void apply_filter(APEContext * ctx, APEFilter *f, | |
746 | int32_t * data0, int32_t * data1, | |
747 | int count, int order, int fracbits) | |
748 | { | |
749 | do_apply_filter(ctx->fileversion, &f[0], data0, count, order, fracbits); | |
750 | if (data1) | |
751 | do_apply_filter(ctx->fileversion, &f[1], data1, count, order, fracbits); | |
752 | } | |
753 | ||
754 | static void ape_apply_filters(APEContext * ctx, int32_t * decoded0, | |
755 | int32_t * decoded1, int count) | |
756 | { | |
757 | int i; | |
758 | ||
759 | for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
760 | if (!ape_filter_orders[ctx->fset][i]) | |
761 | break; | |
762 | apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count, ape_filter_orders[ctx->fset][i], ape_filter_fracbits[ctx->fset][i]); | |
763 | } | |
764 | } | |
765 | ||
766 | static void init_frame_decoder(APEContext * ctx) | |
767 | { | |
768 | int i; | |
769 | init_entropy_decoder(ctx); | |
770 | init_predictor_decoder(ctx); | |
771 | ||
772 | for (i = 0; i < APE_FILTER_LEVELS; i++) { | |
773 | if (!ape_filter_orders[ctx->fset][i]) | |
774 | break; | |
775 | init_filter(ctx, ctx->filters[i], ctx->filterbuf[i], ape_filter_orders[ctx->fset][i]); | |
776 | } | |
777 | } | |
778 | ||
779 | static void ape_unpack_mono(APEContext * ctx, int count) | |
780 | { | |
781 | int32_t left; | |
782 | int32_t *decoded0 = ctx->decoded0; | |
783 | int32_t *decoded1 = ctx->decoded1; | |
784 | ||
785 | if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
786 | entropy_decode(ctx, count, 0); | |
787 | /* We are pure silence, so we're done. */ | |
788 | av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n"); | |
789 | return; | |
790 | } | |
791 | ||
792 | entropy_decode(ctx, count, 0); | |
793 | ape_apply_filters(ctx, decoded0, NULL, count); | |
794 | ||
795 | /* Now apply the predictor decoding */ | |
796 | predictor_decode_mono(ctx, count); | |
797 | ||
798 | /* Pseudo-stereo - just copy left channel to right channel */ | |
799 | if (ctx->channels == 2) { | |
800 | while (count--) { | |
801 | left = *decoded0; | |
802 | *(decoded1++) = *(decoded0++) = left; | |
803 | } | |
804 | } | |
805 | } | |
806 | ||
807 | static void ape_unpack_stereo(APEContext * ctx, int count) | |
808 | { | |
809 | int32_t left, right; | |
810 | int32_t *decoded0 = ctx->decoded0; | |
811 | int32_t *decoded1 = ctx->decoded1; | |
812 | ||
813 | if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) { | |
814 | /* We are pure silence, so we're done. */ | |
815 | av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n"); | |
816 | return; | |
817 | } | |
818 | ||
819 | entropy_decode(ctx, count, 1); | |
820 | ape_apply_filters(ctx, decoded0, decoded1, count); | |
821 | ||
822 | /* Now apply the predictor decoding */ | |
823 | predictor_decode_stereo(ctx, count); | |
824 | ||
825 | /* Decorrelate and scale to output depth */ | |
826 | while (count--) { | |
827 | left = *decoded1 - (*decoded0 / 2); | |
828 | right = left + *decoded0; | |
829 | ||
830 | *(decoded0++) = left; | |
831 | *(decoded1++) = right; | |
832 | } | |
833 | } | |
834 | ||
835 | static int ape_decode_frame(AVCodecContext * avctx, | |
836 | void *data, int *data_size, | |
837 | uint8_t * buf, int buf_size) | |
838 | { | |
839 | APEContext *s = avctx->priv_data; | |
840 | int16_t *samples = data; | |
841 | int nblocks; | |
842 | int i, n; | |
843 | int blockstodecode; | |
844 | int bytes_used; | |
845 | ||
846 | if (buf_size == 0 && !s->samples) { | |
847 | *data_size = 0; | |
848 | return 0; | |
849 | } | |
850 | ||
851 | /* should not happen but who knows */ | |
852 | if (BLOCKS_PER_LOOP * 2 * avctx->channels > *data_size) { | |
853 | 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); | |
854 | return -1; | |
855 | } | |
856 | ||
857 | if(!s->samples){ | |
858 | s->data = av_realloc(s->data, (buf_size + 3) & ~3); | |
859 | s->dsp.bswap_buf(s->data, buf, buf_size >> 2); | |
860 | s->ptr = s->last_ptr = s->data; | |
861 | s->data_end = s->data + buf_size; | |
862 | ||
863 | nblocks = s->samples = bytestream_get_be32(&s->ptr); | |
864 | n = bytestream_get_be32(&s->ptr); | |
865 | if(n < 0 || n > 3){ | |
866 | av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n"); | |
867 | s->data = NULL; | |
868 | return -1; | |
869 | } | |
870 | s->ptr += n; | |
871 | ||
872 | s->currentframeblocks = nblocks; | |
873 | buf += 4; | |
874 | if (s->samples <= 0) { | |
875 | *data_size = 0; | |
876 | return buf_size; | |
877 | } | |
878 | ||
879 | memset(s->decoded0, 0, sizeof(s->decoded0)); | |
880 | memset(s->decoded1, 0, sizeof(s->decoded1)); | |
881 | ||
882 | /* Initialize the frame decoder */ | |
883 | init_frame_decoder(s); | |
884 | } | |
885 | ||
886 | if (!s->data) { | |
887 | *data_size = 0; | |
888 | return buf_size; | |
889 | } | |
890 | ||
891 | nblocks = s->samples; | |
892 | blockstodecode = FFMIN(BLOCKS_PER_LOOP, nblocks); | |
893 | ||
894 | if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO)) | |
895 | ape_unpack_mono(s, blockstodecode); | |
896 | else | |
897 | ape_unpack_stereo(s, blockstodecode); | |
898 | ||
899 | for (i = 0; i < blockstodecode; i++) { | |
900 | *samples++ = s->decoded0[i]; | |
901 | if(s->channels == 2) | |
902 | *samples++ = s->decoded1[i]; | |
903 | } | |
904 | ||
905 | s->samples -= blockstodecode; | |
906 | ||
907 | *data_size = blockstodecode * 2 * s->channels; | |
908 | bytes_used = s->samples ? s->ptr - s->last_ptr : buf_size; | |
909 | s->last_ptr = s->ptr; | |
910 | return bytes_used; | |
911 | } | |
912 | ||
913 | AVCodec ape_decoder = { | |
914 | "ape", | |
915 | CODEC_TYPE_AUDIO, | |
916 | CODEC_ID_APE, | |
917 | sizeof(APEContext), | |
918 | ape_decode_init, | |
919 | NULL, | |
920 | ape_decode_close, | |
921 | ape_decode_frame, | |
922 | }; |