apedec: fix handling of packet sizes that are not a multiple of 4 bytes
[libav.git] / libavcodec / apedec.c
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 Libav.
7 *
8 * Libav 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 * Libav 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 Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 #define BITSTREAM_READER_LE
24 #include "avcodec.h"
25 #include "dsputil.h"
26 #include "get_bits.h"
27 #include "bytestream.h"
28 #include "libavutil/audioconvert.h"
29 #include "libavutil/avassert.h"
30
31 /**
32 * @file
33 * Monkey's Audio lossless audio decoder
34 */
35
36 #define BLOCKS_PER_LOOP 4608
37 #define MAX_CHANNELS 2
38 #define MAX_BYTESPERSAMPLE 3
39
40 #define APE_FRAMECODE_MONO_SILENCE 1
41 #define APE_FRAMECODE_STEREO_SILENCE 3
42 #define APE_FRAMECODE_PSEUDO_STEREO 4
43
44 #define HISTORY_SIZE 512
45 #define PREDICTOR_ORDER 8
46 /** Total size of all predictor histories */
47 #define PREDICTOR_SIZE 50
48
49 #define YDELAYA (18 + PREDICTOR_ORDER*4)
50 #define YDELAYB (18 + PREDICTOR_ORDER*3)
51 #define XDELAYA (18 + PREDICTOR_ORDER*2)
52 #define XDELAYB (18 + PREDICTOR_ORDER)
53
54 #define YADAPTCOEFFSA 18
55 #define XADAPTCOEFFSA 14
56 #define YADAPTCOEFFSB 10
57 #define XADAPTCOEFFSB 5
58
59 /**
60 * Possible compression levels
61 * @{
62 */
63 enum APECompressionLevel {
64 COMPRESSION_LEVEL_FAST = 1000,
65 COMPRESSION_LEVEL_NORMAL = 2000,
66 COMPRESSION_LEVEL_HIGH = 3000,
67 COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
68 COMPRESSION_LEVEL_INSANE = 5000
69 };
70 /** @} */
71
72 #define APE_FILTER_LEVELS 3
73
74 /** Filter orders depending on compression level */
75 static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
76 { 0, 0, 0 },
77 { 16, 0, 0 },
78 { 64, 0, 0 },
79 { 32, 256, 0 },
80 { 16, 256, 1280 }
81 };
82
83 /** Filter fraction bits depending on compression level */
84 static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
85 { 0, 0, 0 },
86 { 11, 0, 0 },
87 { 11, 0, 0 },
88 { 10, 13, 0 },
89 { 11, 13, 15 }
90 };
91
92
93 /** Filters applied to the decoded data */
94 typedef struct APEFilter {
95 int16_t *coeffs; ///< actual coefficients used in filtering
96 int16_t *adaptcoeffs; ///< adaptive filter coefficients used for correcting of actual filter coefficients
97 int16_t *historybuffer; ///< filter memory
98 int16_t *delay; ///< filtered values
99
100 int avg;
101 } APEFilter;
102
103 typedef struct APERice {
104 uint32_t k;
105 uint32_t ksum;
106 } APERice;
107
108 typedef struct APERangecoder {
109 uint32_t low; ///< low end of interval
110 uint32_t range; ///< length of interval
111 uint32_t help; ///< bytes_to_follow resp. intermediate value
112 unsigned int buffer; ///< buffer for input/output
113 } APERangecoder;
114
115 /** Filter histories */
116 typedef struct APEPredictor {
117 int32_t *buf;
118
119 int32_t lastA[2];
120
121 int32_t filterA[2];
122 int32_t filterB[2];
123
124 int32_t coeffsA[2][4]; ///< adaption coefficients
125 int32_t coeffsB[2][5]; ///< adaption coefficients
126 int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
127 } APEPredictor;
128
129 /** Decoder context */
130 typedef struct APEContext {
131 AVCodecContext *avctx;
132 AVFrame frame;
133 DSPContext dsp;
134 int channels;
135 int samples; ///< samples left to decode in current frame
136
137 int fileversion; ///< codec version, very important in decoding process
138 int compression_level; ///< compression levels
139 int fset; ///< which filter set to use (calculated from compression level)
140 int flags; ///< global decoder flags
141
142 uint32_t CRC; ///< frame CRC
143 int frameflags; ///< frame flags
144 APEPredictor predictor; ///< predictor used for final reconstruction
145
146 int32_t decoded0[BLOCKS_PER_LOOP]; ///< decoded data for the first channel
147 int32_t decoded1[BLOCKS_PER_LOOP]; ///< decoded data for the second channel
148
149 int16_t* filterbuf[APE_FILTER_LEVELS]; ///< filter memory
150
151 APERangecoder rc; ///< rangecoder used to decode actual values
152 APERice riceX; ///< rice code parameters for the second channel
153 APERice riceY; ///< rice code parameters for the first channel
154 APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
155
156 uint8_t *data; ///< current frame data
157 uint8_t *data_end; ///< frame data end
158 const uint8_t *ptr; ///< current position in frame data
159
160 int error;
161 } APEContext;
162
163 // TODO: dsputilize
164
165 static av_cold int ape_decode_close(AVCodecContext *avctx)
166 {
167 APEContext *s = avctx->priv_data;
168 int i;
169
170 for (i = 0; i < APE_FILTER_LEVELS; i++)
171 av_freep(&s->filterbuf[i]);
172
173 av_freep(&s->data);
174 return 0;
175 }
176
177 static av_cold int ape_decode_init(AVCodecContext *avctx)
178 {
179 APEContext *s = avctx->priv_data;
180 int i;
181
182 if (avctx->extradata_size != 6) {
183 av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
184 return AVERROR(EINVAL);
185 }
186 if (avctx->bits_per_coded_sample != 16) {
187 av_log(avctx, AV_LOG_ERROR, "Only 16-bit samples are supported\n");
188 return AVERROR(EINVAL);
189 }
190 if (avctx->channels > 2) {
191 av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
192 return AVERROR(EINVAL);
193 }
194 s->avctx = avctx;
195 s->channels = avctx->channels;
196 s->fileversion = AV_RL16(avctx->extradata);
197 s->compression_level = AV_RL16(avctx->extradata + 2);
198 s->flags = AV_RL16(avctx->extradata + 4);
199
200 av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n",
201 s->compression_level, s->flags);
202 if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
203 av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
204 s->compression_level);
205 return AVERROR_INVALIDDATA;
206 }
207 s->fset = s->compression_level / 1000 - 1;
208 for (i = 0; i < APE_FILTER_LEVELS; i++) {
209 if (!ape_filter_orders[s->fset][i])
210 break;
211 FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
212 (ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
213 filter_alloc_fail);
214 }
215
216 dsputil_init(&s->dsp, avctx);
217 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
218 avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
219
220 avcodec_get_frame_defaults(&s->frame);
221 avctx->coded_frame = &s->frame;
222
223 return 0;
224 filter_alloc_fail:
225 ape_decode_close(avctx);
226 return AVERROR(ENOMEM);
227 }
228
229 /**
230 * @name APE range decoding functions
231 * @{
232 */
233
234 #define CODE_BITS 32
235 #define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1))
236 #define SHIFT_BITS (CODE_BITS - 9)
237 #define EXTRA_BITS ((CODE_BITS-2) % 8 + 1)
238 #define BOTTOM_VALUE (TOP_VALUE >> 8)
239
240 /** Start the decoder */
241 static inline void range_start_decoding(APEContext *ctx)
242 {
243 ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
244 ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS);
245 ctx->rc.range = (uint32_t) 1 << EXTRA_BITS;
246 }
247
248 /** Perform normalization */
249 static inline void range_dec_normalize(APEContext *ctx)
250 {
251 while (ctx->rc.range <= BOTTOM_VALUE) {
252 ctx->rc.buffer <<= 8;
253 if(ctx->ptr < ctx->data_end) {
254 ctx->rc.buffer += *ctx->ptr;
255 ctx->ptr++;
256 } else {
257 ctx->error = 1;
258 }
259 ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF);
260 ctx->rc.range <<= 8;
261 }
262 }
263
264 /**
265 * Calculate culmulative frequency for next symbol. Does NO update!
266 * @param ctx decoder context
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 ctx decoder context
280 * @param shift number of bits to decode
281 */
282 static inline int range_decode_culshift(APEContext *ctx, int shift)
283 {
284 range_dec_normalize(ctx);
285 ctx->rc.help = ctx->rc.range >> shift;
286 return ctx->rc.low / ctx->rc.help;
287 }
288
289
290 /**
291 * Update decoding state
292 * @param ctx decoder context
293 * @param sy_f the interval length (frequency of the symbol)
294 * @param lt_f the lower end (frequency sum of < symbols)
295 */
296 static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
297 {
298 ctx->rc.low -= ctx->rc.help * lt_f;
299 ctx->rc.range = ctx->rc.help * sy_f;
300 }
301
302 /** Decode n bits (n <= 16) without modelling */
303 static inline int range_decode_bits(APEContext *ctx, int n)
304 {
305 int sym = range_decode_culshift(ctx, n);
306 range_decode_update(ctx, 1, sym);
307 return sym;
308 }
309
310
311 #define MODEL_ELEMENTS 64
312
313 /**
314 * Fixed probabilities for symbols in Monkey Audio version 3.97
315 */
316 static const uint16_t counts_3970[22] = {
317 0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
318 62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
319 65450, 65469, 65480, 65487, 65491, 65493,
320 };
321
322 /**
323 * Probability ranges for symbols in Monkey Audio version 3.97
324 */
325 static const uint16_t counts_diff_3970[21] = {
326 14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
327 1104, 677, 415, 248, 150, 89, 54, 31,
328 19, 11, 7, 4, 2,
329 };
330
331 /**
332 * Fixed probabilities for symbols in Monkey Audio version 3.98
333 */
334 static const uint16_t counts_3980[22] = {
335 0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
336 64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
337 65485, 65488, 65490, 65491, 65492, 65493,
338 };
339
340 /**
341 * Probability ranges for symbols in Monkey Audio version 3.98
342 */
343 static const uint16_t counts_diff_3980[21] = {
344 19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
345 261, 119, 65, 31, 19, 10, 6, 3,
346 3, 2, 1, 1, 1,
347 };
348
349 /**
350 * Decode symbol
351 * @param ctx decoder context
352 * @param counts probability range start position
353 * @param counts_diff probability range widths
354 */
355 static inline int range_get_symbol(APEContext *ctx,
356 const uint16_t counts[],
357 const uint16_t counts_diff[])
358 {
359 int symbol, cf;
360
361 cf = range_decode_culshift(ctx, 16);
362
363 if(cf > 65492){
364 symbol= cf - 65535 + 63;
365 range_decode_update(ctx, 1, cf);
366 if(cf > 65535)
367 ctx->error=1;
368 return symbol;
369 }
370 /* figure out the symbol inefficiently; a binary search would be much better */
371 for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
372
373 range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
374
375 return symbol;
376 }
377 /** @} */ // group rangecoder
378
379 static inline void update_rice(APERice *rice, int x)
380 {
381 int lim = rice->k ? (1 << (rice->k + 4)) : 0;
382 rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
383
384 if (rice->ksum < lim)
385 rice->k--;
386 else if (rice->ksum >= (1 << (rice->k + 5)))
387 rice->k++;
388 }
389
390 static inline int ape_decode_value(APEContext *ctx, APERice *rice)
391 {
392 int x, overflow;
393
394 if (ctx->fileversion < 3990) {
395 int tmpk;
396
397 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
398
399 if (overflow == (MODEL_ELEMENTS - 1)) {
400 tmpk = range_decode_bits(ctx, 5);
401 overflow = 0;
402 } else
403 tmpk = (rice->k < 1) ? 0 : rice->k - 1;
404
405 if (tmpk <= 16)
406 x = range_decode_bits(ctx, tmpk);
407 else {
408 x = range_decode_bits(ctx, 16);
409 x |= (range_decode_bits(ctx, tmpk - 16) << 16);
410 }
411 x += overflow << tmpk;
412 } else {
413 int base, pivot;
414
415 pivot = rice->ksum >> 5;
416 if (pivot == 0)
417 pivot = 1;
418
419 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
420
421 if (overflow == (MODEL_ELEMENTS - 1)) {
422 overflow = range_decode_bits(ctx, 16) << 16;
423 overflow |= range_decode_bits(ctx, 16);
424 }
425
426 if (pivot < 0x10000) {
427 base = range_decode_culfreq(ctx, pivot);
428 range_decode_update(ctx, 1, base);
429 } else {
430 int base_hi = pivot, base_lo;
431 int bbits = 0;
432
433 while (base_hi & ~0xFFFF) {
434 base_hi >>= 1;
435 bbits++;
436 }
437 base_hi = range_decode_culfreq(ctx, base_hi + 1);
438 range_decode_update(ctx, 1, base_hi);
439 base_lo = range_decode_culfreq(ctx, 1 << bbits);
440 range_decode_update(ctx, 1, base_lo);
441
442 base = (base_hi << bbits) + base_lo;
443 }
444
445 x = base + overflow * pivot;
446 }
447
448 update_rice(rice, x);
449
450 /* Convert to signed */
451 if (x & 1)
452 return (x >> 1) + 1;
453 else
454 return -(x >> 1);
455 }
456
457 static void entropy_decode(APEContext *ctx, int blockstodecode, int stereo)
458 {
459 int32_t *decoded0 = ctx->decoded0;
460 int32_t *decoded1 = ctx->decoded1;
461
462 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
463 /* We are pure silence, just memset the output buffer. */
464 memset(decoded0, 0, blockstodecode * sizeof(int32_t));
465 memset(decoded1, 0, blockstodecode * sizeof(int32_t));
466 } else {
467 while (blockstodecode--) {
468 *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
469 if (stereo)
470 *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
471 }
472 }
473 }
474
475 static int init_entropy_decoder(APEContext *ctx)
476 {
477 /* Read the CRC */
478 if (ctx->data_end - ctx->ptr < 6)
479 return AVERROR_INVALIDDATA;
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 if (ctx->data_end - ctx->ptr < 6)
488 return AVERROR_INVALIDDATA;
489 ctx->frameflags = bytestream_get_be32(&ctx->ptr);
490 }
491
492 /* Initialize the rice structs */
493 ctx->riceX.k = 10;
494 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
495 ctx->riceY.k = 10;
496 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
497
498 /* The first 8 bits of input are ignored. */
499 ctx->ptr++;
500
501 range_start_decoding(ctx);
502
503 return 0;
504 }
505
506 static const int32_t initial_coeffs[4] = {
507 360, 317, -109, 98
508 };
509
510 static void init_predictor_decoder(APEContext *ctx)
511 {
512 APEPredictor *p = &ctx->predictor;
513
514 /* Zero the history buffers */
515 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(int32_t));
516 p->buf = p->historybuffer;
517
518 /* Initialize and zero the coefficients */
519 memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
520 memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
521 memset(p->coeffsB, 0, sizeof(p->coeffsB));
522
523 p->filterA[0] = p->filterA[1] = 0;
524 p->filterB[0] = p->filterB[1] = 0;
525 p->lastA[0] = p->lastA[1] = 0;
526 }
527
528 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
529 static inline int APESIGN(int32_t x) {
530 return (x < 0) - (x > 0);
531 }
532
533 static av_always_inline int predictor_update_filter(APEPredictor *p,
534 const int decoded, const int filter,
535 const int delayA, const int delayB,
536 const int adaptA, const int adaptB)
537 {
538 int32_t predictionA, predictionB, sign;
539
540 p->buf[delayA] = p->lastA[filter];
541 p->buf[adaptA] = APESIGN(p->buf[delayA]);
542 p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
543 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
544
545 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] +
546 p->buf[delayA - 1] * p->coeffsA[filter][1] +
547 p->buf[delayA - 2] * p->coeffsA[filter][2] +
548 p->buf[delayA - 3] * p->coeffsA[filter][3];
549
550 /* Apply a scaled first-order filter compression */
551 p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
552 p->buf[adaptB] = APESIGN(p->buf[delayB]);
553 p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
554 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
555 p->filterB[filter] = p->filterA[filter ^ 1];
556
557 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] +
558 p->buf[delayB - 1] * p->coeffsB[filter][1] +
559 p->buf[delayB - 2] * p->coeffsB[filter][2] +
560 p->buf[delayB - 3] * p->coeffsB[filter][3] +
561 p->buf[delayB - 4] * p->coeffsB[filter][4];
562
563 p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
564 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
565
566 sign = APESIGN(decoded);
567 p->coeffsA[filter][0] += p->buf[adaptA ] * sign;
568 p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
569 p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
570 p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
571 p->coeffsB[filter][0] += p->buf[adaptB ] * sign;
572 p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
573 p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
574 p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
575 p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
576
577 return p->filterA[filter];
578 }
579
580 static void predictor_decode_stereo(APEContext *ctx, int count)
581 {
582 APEPredictor *p = &ctx->predictor;
583 int32_t *decoded0 = ctx->decoded0;
584 int32_t *decoded1 = ctx->decoded1;
585
586 while (count--) {
587 /* Predictor Y */
588 *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
589 YADAPTCOEFFSA, YADAPTCOEFFSB);
590 decoded0++;
591 *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
592 XADAPTCOEFFSA, XADAPTCOEFFSB);
593 decoded1++;
594
595 /* Combined */
596 p->buf++;
597
598 /* Have we filled the history buffer? */
599 if (p->buf == p->historybuffer + HISTORY_SIZE) {
600 memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
601 p->buf = p->historybuffer;
602 }
603 }
604 }
605
606 static void predictor_decode_mono(APEContext *ctx, int count)
607 {
608 APEPredictor *p = &ctx->predictor;
609 int32_t *decoded0 = ctx->decoded0;
610 int32_t predictionA, currentA, A, sign;
611
612 currentA = p->lastA[0];
613
614 while (count--) {
615 A = *decoded0;
616
617 p->buf[YDELAYA] = currentA;
618 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
619
620 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] +
621 p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
622 p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
623 p->buf[YDELAYA - 3] * p->coeffsA[0][3];
624
625 currentA = A + (predictionA >> 10);
626
627 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]);
628 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
629
630 sign = APESIGN(A);
631 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign;
632 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
633 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
634 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
635
636 p->buf++;
637
638 /* Have we filled the history buffer? */
639 if (p->buf == p->historybuffer + HISTORY_SIZE) {
640 memmove(p->historybuffer, p->buf, PREDICTOR_SIZE * sizeof(int32_t));
641 p->buf = p->historybuffer;
642 }
643
644 p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
645 *(decoded0++) = p->filterA[0];
646 }
647
648 p->lastA[0] = currentA;
649 }
650
651 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
652 {
653 f->coeffs = buf;
654 f->historybuffer = buf + order;
655 f->delay = f->historybuffer + order * 2;
656 f->adaptcoeffs = f->historybuffer + order;
657
658 memset(f->historybuffer, 0, (order * 2) * sizeof(int16_t));
659 memset(f->coeffs, 0, order * sizeof(int16_t));
660 f->avg = 0;
661 }
662
663 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
664 {
665 do_init_filter(&f[0], buf, order);
666 do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
667 }
668
669 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
670 int32_t *data, int count, int order, int fracbits)
671 {
672 int res;
673 int absres;
674
675 while (count--) {
676 /* round fixedpoint scalar product */
677 res = ctx->dsp.scalarproduct_and_madd_int16(f->coeffs, f->delay - order,
678 f->adaptcoeffs - order,
679 order, APESIGN(*data));
680 res = (res + (1 << (fracbits - 1))) >> fracbits;
681 res += *data;
682 *data++ = res;
683
684 /* Update the output history */
685 *f->delay++ = av_clip_int16(res);
686
687 if (version < 3980) {
688 /* Version ??? to < 3.98 files (untested) */
689 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
690 f->adaptcoeffs[-4] >>= 1;
691 f->adaptcoeffs[-8] >>= 1;
692 } else {
693 /* Version 3.98 and later files */
694
695 /* Update the adaption coefficients */
696 absres = FFABS(res);
697 if (absres)
698 *f->adaptcoeffs = ((res & (-1<<31)) ^ (-1<<30)) >>
699 (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
700 else
701 *f->adaptcoeffs = 0;
702
703 f->avg += (absres - f->avg) / 16;
704
705 f->adaptcoeffs[-1] >>= 1;
706 f->adaptcoeffs[-2] >>= 1;
707 f->adaptcoeffs[-8] >>= 1;
708 }
709
710 f->adaptcoeffs++;
711
712 /* Have we filled the history buffer? */
713 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
714 memmove(f->historybuffer, f->delay - (order * 2),
715 (order * 2) * sizeof(int16_t));
716 f->delay = f->historybuffer + order * 2;
717 f->adaptcoeffs = f->historybuffer + order;
718 }
719 }
720 }
721
722 static void apply_filter(APEContext *ctx, APEFilter *f,
723 int32_t *data0, int32_t *data1,
724 int count, int order, int fracbits)
725 {
726 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
727 if (data1)
728 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
729 }
730
731 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
732 int32_t *decoded1, int count)
733 {
734 int i;
735
736 for (i = 0; i < APE_FILTER_LEVELS; i++) {
737 if (!ape_filter_orders[ctx->fset][i])
738 break;
739 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
740 ape_filter_orders[ctx->fset][i],
741 ape_filter_fracbits[ctx->fset][i]);
742 }
743 }
744
745 static int init_frame_decoder(APEContext *ctx)
746 {
747 int i, ret;
748 if ((ret = init_entropy_decoder(ctx)) < 0)
749 return ret;
750 init_predictor_decoder(ctx);
751
752 for (i = 0; i < APE_FILTER_LEVELS; i++) {
753 if (!ape_filter_orders[ctx->fset][i])
754 break;
755 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
756 ape_filter_orders[ctx->fset][i]);
757 }
758 return 0;
759 }
760
761 static void ape_unpack_mono(APEContext *ctx, int count)
762 {
763 int32_t *decoded0 = ctx->decoded0;
764 int32_t *decoded1 = ctx->decoded1;
765
766 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
767 entropy_decode(ctx, count, 0);
768 /* We are pure silence, so we're done. */
769 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
770 return;
771 }
772
773 entropy_decode(ctx, count, 0);
774 ape_apply_filters(ctx, decoded0, NULL, count);
775
776 /* Now apply the predictor decoding */
777 predictor_decode_mono(ctx, count);
778
779 /* Pseudo-stereo - just copy left channel to right channel */
780 if (ctx->channels == 2) {
781 memcpy(decoded1, decoded0, count * sizeof(*decoded1));
782 }
783 }
784
785 static void ape_unpack_stereo(APEContext *ctx, int count)
786 {
787 int32_t left, right;
788 int32_t *decoded0 = ctx->decoded0;
789 int32_t *decoded1 = ctx->decoded1;
790
791 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
792 /* We are pure silence, so we're done. */
793 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
794 return;
795 }
796
797 entropy_decode(ctx, count, 1);
798 ape_apply_filters(ctx, decoded0, decoded1, count);
799
800 /* Now apply the predictor decoding */
801 predictor_decode_stereo(ctx, count);
802
803 /* Decorrelate and scale to output depth */
804 while (count--) {
805 left = *decoded1 - (*decoded0 / 2);
806 right = left + *decoded0;
807
808 *(decoded0++) = left;
809 *(decoded1++) = right;
810 }
811 }
812
813 static int ape_decode_frame(AVCodecContext *avctx, void *data,
814 int *got_frame_ptr, AVPacket *avpkt)
815 {
816 const uint8_t *buf = avpkt->data;
817 APEContext *s = avctx->priv_data;
818 int16_t *samples;
819 int i, ret;
820 int blockstodecode;
821 int bytes_used = 0;
822
823 /* this should never be negative, but bad things will happen if it is, so
824 check it just to make sure. */
825 av_assert0(s->samples >= 0);
826
827 if(!s->samples){
828 uint32_t nblocks, offset;
829 void *tmp_data;
830 int buf_size;
831
832 if (!avpkt->size) {
833 *got_frame_ptr = 0;
834 return 0;
835 }
836 if (avpkt->size < 8) {
837 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
838 return AVERROR_INVALIDDATA;
839 }
840 buf_size = avpkt->size & ~3;
841 if (buf_size != avpkt->size) {
842 av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
843 "extra bytes at the end will be skipped.\n");
844 }
845
846 tmp_data = av_realloc(s->data, buf_size);
847 if (!tmp_data)
848 return AVERROR(ENOMEM);
849 s->data = tmp_data;
850 s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
851 s->ptr = s->data;
852 s->data_end = s->data + buf_size;
853
854 nblocks = bytestream_get_be32(&s->ptr);
855 offset = bytestream_get_be32(&s->ptr);
856 if (offset > 3) {
857 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
858 s->data = NULL;
859 return AVERROR_INVALIDDATA;
860 }
861 if (s->data_end - s->ptr < offset) {
862 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
863 return AVERROR_INVALIDDATA;
864 }
865 s->ptr += offset;
866
867 if (!nblocks || nblocks > INT_MAX) {
868 av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %u.\n", nblocks);
869 return AVERROR_INVALIDDATA;
870 }
871 s->samples = nblocks;
872
873 memset(s->decoded0, 0, sizeof(s->decoded0));
874 memset(s->decoded1, 0, sizeof(s->decoded1));
875
876 /* Initialize the frame decoder */
877 if (init_frame_decoder(s) < 0) {
878 av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
879 return AVERROR_INVALIDDATA;
880 }
881
882 bytes_used = avpkt->size;
883 }
884
885 if (!s->data) {
886 *got_frame_ptr = 0;
887 return avpkt->size;
888 }
889
890 blockstodecode = FFMIN(BLOCKS_PER_LOOP, s->samples);
891
892 /* get output buffer */
893 s->frame.nb_samples = blockstodecode;
894 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
895 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
896 return ret;
897 }
898 samples = (int16_t *)s->frame.data[0];
899
900 s->error=0;
901
902 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
903 ape_unpack_mono(s, blockstodecode);
904 else
905 ape_unpack_stereo(s, blockstodecode);
906 emms_c();
907
908 if (s->error) {
909 s->samples=0;
910 av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
911 return AVERROR_INVALIDDATA;
912 }
913
914 for (i = 0; i < blockstodecode; i++) {
915 *samples++ = s->decoded0[i];
916 if(s->channels == 2)
917 *samples++ = s->decoded1[i];
918 }
919
920 s->samples -= blockstodecode;
921
922 *got_frame_ptr = 1;
923 *(AVFrame *)data = s->frame;
924
925 return bytes_used;
926 }
927
928 static void ape_flush(AVCodecContext *avctx)
929 {
930 APEContext *s = avctx->priv_data;
931 s->samples= 0;
932 }
933
934 AVCodec ff_ape_decoder = {
935 .name = "ape",
936 .type = AVMEDIA_TYPE_AUDIO,
937 .id = CODEC_ID_APE,
938 .priv_data_size = sizeof(APEContext),
939 .init = ape_decode_init,
940 .close = ape_decode_close,
941 .decode = ape_decode_frame,
942 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
943 .flush = ape_flush,
944 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
945 };