path: root/libavcodec/apedec.c (plain)
blob: a6b14b8e247c7ec84f346e4581f9d38146139281

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	#include <inttypes.h>
24
25	#include "libavutil/avassert.h"
26	#include "libavutil/channel_layout.h"
27	#include "libavutil/opt.h"
28	#include "lossless_audiodsp.h"
29	#include "avcodec.h"
30	#include "bswapdsp.h"
31	#include "bytestream.h"
32	#include "internal.h"
33	#include "get_bits.h"
34	#include "unary.h"
35
36	/**
37	* @file
38	* Monkey's Audio lossless audio decoder
39	*/
40
41	#define MAX_CHANNELS 2
42	#define MAX_BYTESPERSAMPLE 3
43
44	#define APE_FRAMECODE_MONO_SILENCE 1
45	#define APE_FRAMECODE_STEREO_SILENCE 3
46	#define APE_FRAMECODE_PSEUDO_STEREO 4
47
48	#define HISTORY_SIZE 512
49	#define PREDICTOR_ORDER 8
50	/** Total size of all predictor histories */
51	#define PREDICTOR_SIZE 50
52
53	#define YDELAYA (18 + PREDICTOR_ORDER*4)
54	#define YDELAYB (18 + PREDICTOR_ORDER*3)
55	#define XDELAYA (18 + PREDICTOR_ORDER*2)
56	#define XDELAYB (18 + PREDICTOR_ORDER)
57
58	#define YADAPTCOEFFSA 18
59	#define XADAPTCOEFFSA 14
60	#define YADAPTCOEFFSB 10
61	#define XADAPTCOEFFSB 5
62
63	/**
64	* Possible compression levels
65	* @{
66	*/
67	enum APECompressionLevel {
68	COMPRESSION_LEVEL_FAST = 1000,
69	COMPRESSION_LEVEL_NORMAL = 2000,
70	COMPRESSION_LEVEL_HIGH = 3000,
71	COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
72	COMPRESSION_LEVEL_INSANE = 5000
73	};
74	/** @} */
75
76	#define APE_FILTER_LEVELS 3
77
78	/** Filter orders depending on compression level */
79	static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
80	{ 0, 0, 0 },
81	{ 16, 0, 0 },
82	{ 64, 0, 0 },
83	{ 32, 256, 0 },
84	{ 16, 256, 1280 }
85	};
86
87	/** Filter fraction bits depending on compression level */
88	static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
89	{ 0, 0, 0 },
90	{ 11, 0, 0 },
91	{ 11, 0, 0 },
92	{ 10, 13, 0 },
93	{ 11, 13, 15 }
94	};
95
96
97	/** Filters applied to the decoded data */
98	typedef struct APEFilter {
99	int16_t *coeffs; ///< actual coefficients used in filtering
100	int16_t *adaptcoeffs; ///< adaptive filter coefficients used for correcting of actual filter coefficients
101	int16_t *historybuffer; ///< filter memory
102	int16_t *delay; ///< filtered values
103
104	int avg;
105	} APEFilter;
106
107	typedef struct APERice {
108	uint32_t k;
109	uint32_t ksum;
110	} APERice;
111
112	typedef struct APERangecoder {
113	uint32_t low; ///< low end of interval
114	uint32_t range; ///< length of interval
115	uint32_t help; ///< bytes_to_follow resp. intermediate value
116	unsigned int buffer; ///< buffer for input/output
117	} APERangecoder;
118
119	/** Filter histories */
120	typedef struct APEPredictor {
121	int32_t *buf;
122
123	int32_t lastA[2];
124
125	int32_t filterA[2];
126	int32_t filterB[2];
127
128	int32_t coeffsA[2][4]; ///< adaption coefficients
129	int32_t coeffsB[2][5]; ///< adaption coefficients
130	int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
131
132	unsigned int sample_pos;
133	} APEPredictor;
134
135	/** Decoder context */
136	typedef struct APEContext {
137	AVClass *class; ///< class for AVOptions
138	AVCodecContext *avctx;
139	BswapDSPContext bdsp;
140	LLAudDSPContext adsp;
141	int channels;
142	int samples; ///< samples left to decode in current frame
143	int bps;
144
145	int fileversion; ///< codec version, very important in decoding process
146	int compression_level; ///< compression levels
147	int fset; ///< which filter set to use (calculated from compression level)
148	int flags; ///< global decoder flags
149
150	uint32_t CRC; ///< frame CRC
151	int frameflags; ///< frame flags
152	APEPredictor predictor; ///< predictor used for final reconstruction
153
154	int32_t *decoded_buffer;
155	int decoded_size;
156	int32_t *decoded[MAX_CHANNELS]; ///< decoded data for each channel
157	int blocks_per_loop; ///< maximum number of samples to decode for each call
158
159	int16_t* filterbuf[APE_FILTER_LEVELS]; ///< filter memory
160
161	APERangecoder rc; ///< rangecoder used to decode actual values
162	APERice riceX; ///< rice code parameters for the second channel
163	APERice riceY; ///< rice code parameters for the first channel
164	APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
165	GetBitContext gb;
166
167	uint8_t *data; ///< current frame data
168	uint8_t *data_end; ///< frame data end
169	int data_size; ///< frame data allocated size
170	const uint8_t *ptr; ///< current position in frame data
171
172	int error;
173
174	void (*entropy_decode_mono)(struct APEContext *ctx, int blockstodecode);
175	void (*entropy_decode_stereo)(struct APEContext *ctx, int blockstodecode);
176	void (*predictor_decode_mono)(struct APEContext *ctx, int count);
177	void (*predictor_decode_stereo)(struct APEContext *ctx, int count);
178	} APEContext;
179
180	static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
181	int32_t *decoded1, int count);
182
183	static void entropy_decode_mono_0000(APEContext *ctx, int blockstodecode);
184	static void entropy_decode_stereo_0000(APEContext *ctx, int blockstodecode);
185	static void entropy_decode_mono_3860(APEContext *ctx, int blockstodecode);
186	static void entropy_decode_stereo_3860(APEContext *ctx, int blockstodecode);
187	static void entropy_decode_mono_3900(APEContext *ctx, int blockstodecode);
188	static void entropy_decode_stereo_3900(APEContext *ctx, int blockstodecode);
189	static void entropy_decode_stereo_3930(APEContext *ctx, int blockstodecode);
190	static void entropy_decode_mono_3990(APEContext *ctx, int blockstodecode);
191	static void entropy_decode_stereo_3990(APEContext *ctx, int blockstodecode);
192
193	static void predictor_decode_mono_3800(APEContext *ctx, int count);
194	static void predictor_decode_stereo_3800(APEContext *ctx, int count);
195	static void predictor_decode_mono_3930(APEContext *ctx, int count);
196	static void predictor_decode_stereo_3930(APEContext *ctx, int count);
197	static void predictor_decode_mono_3950(APEContext *ctx, int count);
198	static void predictor_decode_stereo_3950(APEContext *ctx, int count);
199
200	static av_cold int ape_decode_close(AVCodecContext *avctx)
201	{
202	APEContext *s = avctx->priv_data;
203	int i;
204
205	for (i = 0; i < APE_FILTER_LEVELS; i++)
206	av_freep(&s->filterbuf[i]);
207
208	av_freep(&s->decoded_buffer);
209	av_freep(&s->data);
210	s->decoded_size = s->data_size = 0;
211
212	return 0;
213	}
214
215	static av_cold int ape_decode_init(AVCodecContext *avctx)
216	{
217	APEContext *s = avctx->priv_data;
218	int i;
219
220	if (avctx->extradata_size != 6) {
221	av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
222	return AVERROR(EINVAL);
223	}
224	if (avctx->channels > 2) {
225	av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
226	return AVERROR(EINVAL);
227	}
228	s->bps = avctx->bits_per_coded_sample;
229	switch (s->bps) {
230	case 8:
231	avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
232	break;
233	case 16:
234	avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
235	break;
236	case 24:
237	avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
238	break;
239	default:
240	avpriv_request_sample(avctx,
241	"%d bits per coded sample", s->bps);
242	return AVERROR_PATCHWELCOME;
243	}
244	s->avctx = avctx;
245	s->channels = avctx->channels;
246	s->fileversion = AV_RL16(avctx->extradata);
247	s->compression_level = AV_RL16(avctx->extradata + 2);
248	s->flags = AV_RL16(avctx->extradata + 4);
249
250	av_log(avctx, AV_LOG_VERBOSE, "Compression Level: %d - Flags: %d\n",
251	s->compression_level, s->flags);
252	if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE ||
253	!s->compression_level ||
254	(s->fileversion < 3930 && s->compression_level == COMPRESSION_LEVEL_INSANE)) {
255	av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
256	s->compression_level);
257	return AVERROR_INVALIDDATA;
258	}
259	s->fset = s->compression_level / 1000 - 1;
260	for (i = 0; i < APE_FILTER_LEVELS; i++) {
261	if (!ape_filter_orders[s->fset][i])
262	break;
263	FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
264	(ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
265	filter_alloc_fail);
266	}
267
268	if (s->fileversion < 3860) {
269	s->entropy_decode_mono = entropy_decode_mono_0000;
270	s->entropy_decode_stereo = entropy_decode_stereo_0000;
271	} else if (s->fileversion < 3900) {
272	s->entropy_decode_mono = entropy_decode_mono_3860;
273	s->entropy_decode_stereo = entropy_decode_stereo_3860;
274	} else if (s->fileversion < 3930) {
275	s->entropy_decode_mono = entropy_decode_mono_3900;
276	s->entropy_decode_stereo = entropy_decode_stereo_3900;
277	} else if (s->fileversion < 3990) {
278	s->entropy_decode_mono = entropy_decode_mono_3900;
279	s->entropy_decode_stereo = entropy_decode_stereo_3930;
280	} else {
281	s->entropy_decode_mono = entropy_decode_mono_3990;
282	s->entropy_decode_stereo = entropy_decode_stereo_3990;
283	}
284
285	if (s->fileversion < 3930) {
286	s->predictor_decode_mono = predictor_decode_mono_3800;
287	s->predictor_decode_stereo = predictor_decode_stereo_3800;
288	} else if (s->fileversion < 3950) {
289	s->predictor_decode_mono = predictor_decode_mono_3930;
290	s->predictor_decode_stereo = predictor_decode_stereo_3930;
291	} else {
292	s->predictor_decode_mono = predictor_decode_mono_3950;
293	s->predictor_decode_stereo = predictor_decode_stereo_3950;
294	}
295
296	ff_bswapdsp_init(&s->bdsp);
297	ff_llauddsp_init(&s->adsp);
298	avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
299
300	return 0;
301	filter_alloc_fail:
302	ape_decode_close(avctx);
303	return AVERROR(ENOMEM);
304	}
305
306	/**
307	* @name APE range decoding functions
308	* @{
309	*/
310
311	#define CODE_BITS 32
312	#define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1))
313	#define SHIFT_BITS (CODE_BITS - 9)
314	#define EXTRA_BITS ((CODE_BITS-2) % 8 + 1)
315	#define BOTTOM_VALUE (TOP_VALUE >> 8)
316
317	/** Start the decoder */
318	static inline void range_start_decoding(APEContext *ctx)
319	{
320	ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
321	ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS);
322	ctx->rc.range = (uint32_t) 1 << EXTRA_BITS;
323	}
324
325	/** Perform normalization */
326	static inline void range_dec_normalize(APEContext *ctx)
327	{
328	while (ctx->rc.range <= BOTTOM_VALUE) {
329	ctx->rc.buffer <<= 8;
330	if(ctx->ptr < ctx->data_end) {
331	ctx->rc.buffer += *ctx->ptr;
332	ctx->ptr++;
333	} else {
334	ctx->error = 1;
335	}
336	ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF);
337	ctx->rc.range <<= 8;
338	}
339	}
340
341	/**
342	* Calculate cumulative frequency for next symbol. Does NO update!
343	* @param ctx decoder context
344	* @param tot_f is the total frequency or (code_value)1<<shift
345	* @return the cumulative frequency
346	*/
347	static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
348	{
349	range_dec_normalize(ctx);
350	ctx->rc.help = ctx->rc.range / tot_f;
351	return ctx->rc.low / ctx->rc.help;
352	}
353
354	/**
355	* Decode value with given size in bits
356	* @param ctx decoder context
357	* @param shift number of bits to decode
358	*/
359	static inline int range_decode_culshift(APEContext *ctx, int shift)
360	{
361	range_dec_normalize(ctx);
362	ctx->rc.help = ctx->rc.range >> shift;
363	return ctx->rc.low / ctx->rc.help;
364	}
365
366
367	/**
368	* Update decoding state
369	* @param ctx decoder context
370	* @param sy_f the interval length (frequency of the symbol)
371	* @param lt_f the lower end (frequency sum of < symbols)
372	*/
373	static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
374	{
375	ctx->rc.low -= ctx->rc.help * lt_f;
376	ctx->rc.range = ctx->rc.help * sy_f;
377	}
378
379	/** Decode n bits (n <= 16) without modelling */
380	static inline int range_decode_bits(APEContext *ctx, int n)
381	{
382	int sym = range_decode_culshift(ctx, n);
383	range_decode_update(ctx, 1, sym);
384	return sym;
385	}
386
387
388	#define MODEL_ELEMENTS 64
389
390	/**
391	* Fixed probabilities for symbols in Monkey Audio version 3.97
392	*/
393	static const uint16_t counts_3970[22] = {
394	0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
395	62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
396	65450, 65469, 65480, 65487, 65491, 65493,
397	};
398
399	/**
400	* Probability ranges for symbols in Monkey Audio version 3.97
401	*/
402	static const uint16_t counts_diff_3970[21] = {
403	14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
404	1104, 677, 415, 248, 150, 89, 54, 31,
405	19, 11, 7, 4, 2,
406	};
407
408	/**
409	* Fixed probabilities for symbols in Monkey Audio version 3.98
410	*/
411	static const uint16_t counts_3980[22] = {
412	0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
413	64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
414	65485, 65488, 65490, 65491, 65492, 65493,
415	};
416
417	/**
418	* Probability ranges for symbols in Monkey Audio version 3.98
419	*/
420	static const uint16_t counts_diff_3980[21] = {
421	19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
422	261, 119, 65, 31, 19, 10, 6, 3,
423	3, 2, 1, 1, 1,
424	};
425
426	/**
427	* Decode symbol
428	* @param ctx decoder context
429	* @param counts probability range start position
430	* @param counts_diff probability range widths
431	*/
432	static inline int range_get_symbol(APEContext *ctx,
433	const uint16_t counts[],
434	const uint16_t counts_diff[])
435	{
436	int symbol, cf;
437
438	cf = range_decode_culshift(ctx, 16);
439
440	if(cf > 65492){
441	symbol= cf - 65535 + 63;
442	range_decode_update(ctx, 1, cf);
443	if(cf > 65535)
444	ctx->error=1;
445	return symbol;
446	}
447	/* figure out the symbol inefficiently; a binary search would be much better */
448	for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
449
450	range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
451
452	return symbol;
453	}
454	/** @} */ // group rangecoder
455
456	static inline void update_rice(APERice *rice, unsigned int x)
457	{
458	int lim = rice->k ? (1 << (rice->k + 4)) : 0;
459	rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
460
461	if (rice->ksum < lim)
462	rice->k--;
463	else if (rice->ksum >= (1 << (rice->k + 5)))
464	rice->k++;
465	}
466
467	static inline int get_rice_ook(GetBitContext *gb, int k)
468	{
469	unsigned int x;
470
471	x = get_unary(gb, 1, get_bits_left(gb));
472
473	if (k)
474	x = (x << k) | get_bits(gb, k);
475
476	return x;
477	}
478
479	static inline int ape_decode_value_3860(APEContext *ctx, GetBitContext *gb,
480	APERice *rice)
481	{
482	unsigned int x, overflow;
483
484	overflow = get_unary(gb, 1, get_bits_left(gb));
485
486	if (ctx->fileversion > 3880) {
487	while (overflow >= 16) {
488	overflow -= 16;
489	rice->k += 4;
490	}
491	}
492
493	if (!rice->k)
494	x = overflow;
495	else if(rice->k <= MIN_CACHE_BITS) {
496	x = (overflow << rice->k) + get_bits(gb, rice->k);
497	} else {
498	av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %"PRIu32"\n", rice->k);
499	return AVERROR_INVALIDDATA;
500	}
501	rice->ksum += x - (rice->ksum + 8 >> 4);
502	if (rice->ksum < (rice->k ? 1 << (rice->k + 4) : 0))
503	rice->k--;
504	else if (rice->ksum >= (1 << (rice->k + 5)) && rice->k < 24)
505	rice->k++;
506
507	/* Convert to signed */
508	return ((x >> 1) ^ ((x & 1) - 1)) + 1;
509	}
510
511	static inline int ape_decode_value_3900(APEContext *ctx, APERice *rice)
512	{
513	unsigned int x, overflow;
514	int tmpk;
515
516	overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
517
518	if (overflow == (MODEL_ELEMENTS - 1)) {
519	tmpk = range_decode_bits(ctx, 5);
520	overflow = 0;
521	} else
522	tmpk = (rice->k < 1) ? 0 : rice->k - 1;
523
524	if (tmpk <= 16 || ctx->fileversion < 3910) {
525	if (tmpk > 23) {
526	av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
527	return AVERROR_INVALIDDATA;
528	}
529	x = range_decode_bits(ctx, tmpk);
530	} else if (tmpk <= 31) {
531	x = range_decode_bits(ctx, 16);
532	x |= (range_decode_bits(ctx, tmpk - 16) << 16);
533	} else {
534	av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
535	return AVERROR_INVALIDDATA;
536	}
537	x += overflow << tmpk;
538
539	update_rice(rice, x);
540
541	/* Convert to signed */
542	return ((x >> 1) ^ ((x & 1) - 1)) + 1;
543	}
544
545	static inline int ape_decode_value_3990(APEContext *ctx, APERice *rice)
546	{
547	unsigned int x, overflow;
548	int base, pivot;
549
550	pivot = rice->ksum >> 5;
551	if (pivot == 0)
552	pivot = 1;
553
554	overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
555
556	if (overflow == (MODEL_ELEMENTS - 1)) {
557	overflow = range_decode_bits(ctx, 16) << 16;
558	overflow |= range_decode_bits(ctx, 16);
559	}
560
561	if (pivot < 0x10000) {
562	base = range_decode_culfreq(ctx, pivot);
563	range_decode_update(ctx, 1, base);
564	} else {
565	int base_hi = pivot, base_lo;
566	int bbits = 0;
567
568	while (base_hi & ~0xFFFF) {
569	base_hi >>= 1;
570	bbits++;
571	}
572	base_hi = range_decode_culfreq(ctx, base_hi + 1);
573	range_decode_update(ctx, 1, base_hi);
574	base_lo = range_decode_culfreq(ctx, 1 << bbits);
575	range_decode_update(ctx, 1, base_lo);
576
577	base = (base_hi << bbits) + base_lo;
578	}
579
580	x = base + overflow * pivot;
581
582	update_rice(rice, x);
583
584	/* Convert to signed */
585	return ((x >> 1) ^ ((x & 1) - 1)) + 1;
586	}
587
588	static void decode_array_0000(APEContext *ctx, GetBitContext *gb,
589	int32_t *out, APERice *rice, int blockstodecode)
590	{
591	int i;
592	int ksummax, ksummin;
593
594	rice->ksum = 0;
595	for (i = 0; i < FFMIN(blockstodecode, 5); i++) {
596	out[i] = get_rice_ook(&ctx->gb, 10);
597	rice->ksum += out[i];
598	}
599	rice->k = av_log2(rice->ksum / 10) + 1;
600	if (rice->k >= 24)
601	return;
602	for (; i < FFMIN(blockstodecode, 64); i++) {
603	out[i] = get_rice_ook(&ctx->gb, rice->k);
604	rice->ksum += out[i];
605	rice->k = av_log2(rice->ksum / ((i + 1) * 2)) + 1;
606	if (rice->k >= 24)
607	return;
608	}
609	ksummax = 1 << rice->k + 7;
610	ksummin = rice->k ? (1 << rice->k + 6) : 0;
611	for (; i < blockstodecode; i++) {
612	out[i] = get_rice_ook(&ctx->gb, rice->k);
613	rice->ksum += out[i] - out[i - 64];
614	while (rice->ksum < ksummin) {
615	rice->k--;
616	ksummin = rice->k ? ksummin >> 1 : 0;
617	ksummax >>= 1;
618	}
619	while (rice->ksum >= ksummax) {
620	rice->k++;
621	if (rice->k > 24)
622	return;
623	ksummax <<= 1;
624	ksummin = ksummin ? ksummin << 1 : 128;
625	}
626	}
627
628	for (i = 0; i < blockstodecode; i++)
629	out[i] = ((out[i] >> 1) ^ ((out[i] & 1) - 1)) + 1;
630	}
631
632	static void entropy_decode_mono_0000(APEContext *ctx, int blockstodecode)
633	{
634	decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
635	blockstodecode);
636	}
637
638	static void entropy_decode_stereo_0000(APEContext *ctx, int blockstodecode)
639	{
640	decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
641	blockstodecode);
642	decode_array_0000(ctx, &ctx->gb, ctx->decoded[1], &ctx->riceX,
643	blockstodecode);
644	}
645
646	static void entropy_decode_mono_3860(APEContext *ctx, int blockstodecode)
647	{
648	int32_t *decoded0 = ctx->decoded[0];
649
650	while (blockstodecode--)
651	*decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
652	}
653
654	static void entropy_decode_stereo_3860(APEContext *ctx, int blockstodecode)
655	{
656	int32_t *decoded0 = ctx->decoded[0];
657	int32_t *decoded1 = ctx->decoded[1];
658	int blocks = blockstodecode;
659
660	while (blockstodecode--)
661	*decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
662	while (blocks--)
663	*decoded1++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceX);
664	}
665
666	static void entropy_decode_mono_3900(APEContext *ctx, int blockstodecode)
667	{
668	int32_t *decoded0 = ctx->decoded[0];
669
670	while (blockstodecode--)
671	*decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
672	}
673
674	static void entropy_decode_stereo_3900(APEContext *ctx, int blockstodecode)
675	{
676	int32_t *decoded0 = ctx->decoded[0];
677	int32_t *decoded1 = ctx->decoded[1];
678	int blocks = blockstodecode;
679
680	while (blockstodecode--)
681	*decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
682	range_dec_normalize(ctx);
683	// because of some implementation peculiarities we need to backpedal here
684	ctx->ptr -= 1;
685	range_start_decoding(ctx);
686	while (blocks--)
687	*decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
688	}
689
690	static void entropy_decode_stereo_3930(APEContext *ctx, int blockstodecode)
691	{
692	int32_t *decoded0 = ctx->decoded[0];
693	int32_t *decoded1 = ctx->decoded[1];
694
695	while (blockstodecode--) {
696	*decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
697	*decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
698	}
699	}
700
701	static void entropy_decode_mono_3990(APEContext *ctx, int blockstodecode)
702	{
703	int32_t *decoded0 = ctx->decoded[0];
704
705	while (blockstodecode--)
706	*decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
707	}
708
709	static void entropy_decode_stereo_3990(APEContext *ctx, int blockstodecode)
710	{
711	int32_t *decoded0 = ctx->decoded[0];
712	int32_t *decoded1 = ctx->decoded[1];
713
714	while (blockstodecode--) {
715	*decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
716	*decoded1++ = ape_decode_value_3990(ctx, &ctx->riceX);
717	}
718	}
719
720	static int init_entropy_decoder(APEContext *ctx)
721	{
722	/* Read the CRC */
723	if (ctx->fileversion >= 3900) {
724	if (ctx->data_end - ctx->ptr < 6)
725	return AVERROR_INVALIDDATA;
726	ctx->CRC = bytestream_get_be32(&ctx->ptr);
727	} else {
728	ctx->CRC = get_bits_long(&ctx->gb, 32);
729	}
730
731	/* Read the frame flags if they exist */
732	ctx->frameflags = 0;
733	if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
734	ctx->CRC &= ~0x80000000;
735
736	if (ctx->data_end - ctx->ptr < 6)
737	return AVERROR_INVALIDDATA;
738	ctx->frameflags = bytestream_get_be32(&ctx->ptr);
739	}
740
741	/* Initialize the rice structs */
742	ctx->riceX.k = 10;
743	ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
744	ctx->riceY.k = 10;
745	ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
746
747	if (ctx->fileversion >= 3900) {
748	/* The first 8 bits of input are ignored. */
749	ctx->ptr++;
750
751	range_start_decoding(ctx);
752	}
753
754	return 0;
755	}
756
757	static const int32_t initial_coeffs_fast_3320[1] = {
758	375,
759	};
760
761	static const int32_t initial_coeffs_a_3800[3] = {
762	64, 115, 64,
763	};
764
765	static const int32_t initial_coeffs_b_3800[2] = {
766	740, 0
767	};
768
769	static const int32_t initial_coeffs_3930[4] = {
770	360, 317, -109, 98
771	};
772
773	static void init_predictor_decoder(APEContext *ctx)
774	{
775	APEPredictor *p = &ctx->predictor;
776
777	/* Zero the history buffers */
778	memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(*p->historybuffer));
779	p->buf = p->historybuffer;
780
781	/* Initialize and zero the coefficients */
782	if (ctx->fileversion < 3930) {
783	if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
784	memcpy(p->coeffsA[0], initial_coeffs_fast_3320,
785	sizeof(initial_coeffs_fast_3320));
786	memcpy(p->coeffsA[1], initial_coeffs_fast_3320,
787	sizeof(initial_coeffs_fast_3320));
788	} else {
789	memcpy(p->coeffsA[0], initial_coeffs_a_3800,
790	sizeof(initial_coeffs_a_3800));
791	memcpy(p->coeffsA[1], initial_coeffs_a_3800,
792	sizeof(initial_coeffs_a_3800));
793	}
794	} else {
795	memcpy(p->coeffsA[0], initial_coeffs_3930, sizeof(initial_coeffs_3930));
796	memcpy(p->coeffsA[1], initial_coeffs_3930, sizeof(initial_coeffs_3930));
797	}
798	memset(p->coeffsB, 0, sizeof(p->coeffsB));
799	if (ctx->fileversion < 3930) {
800	memcpy(p->coeffsB[0], initial_coeffs_b_3800,
801	sizeof(initial_coeffs_b_3800));
802	memcpy(p->coeffsB[1], initial_coeffs_b_3800,
803	sizeof(initial_coeffs_b_3800));
804	}
805
806	p->filterA[0] = p->filterA[1] = 0;
807	p->filterB[0] = p->filterB[1] = 0;
808	p->lastA[0] = p->lastA[1] = 0;
809
810	p->sample_pos = 0;
811	}
812
813	/** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
814	static inline int APESIGN(int32_t x) {
815	return (x < 0) - (x > 0);
816	}
817
818	static av_always_inline int filter_fast_3320(APEPredictor *p,
819	const int decoded, const int filter,
820	const int delayA)
821	{
822	int32_t predictionA;
823
824	p->buf[delayA] = p->lastA[filter];
825	if (p->sample_pos < 3) {
826	p->lastA[filter] = decoded;
827	p->filterA[filter] = decoded;
828	return decoded;
829	}
830
831	predictionA = p->buf[delayA] * 2 - p->buf[delayA - 1];
832	p->lastA[filter] = decoded + (predictionA * p->coeffsA[filter][0] >> 9);
833
834	if ((decoded ^ predictionA) > 0)
835	p->coeffsA[filter][0]++;
836	else
837	p->coeffsA[filter][0]--;
838
839	p->filterA[filter] += p->lastA[filter];
840
841	return p->filterA[filter];
842	}
843
844	static av_always_inline int filter_3800(APEPredictor *p,
845	const int decoded, const int filter,
846	const int delayA, const int delayB,
847	const int start, const int shift)
848	{
849	int32_t predictionA, predictionB, sign;
850	int32_t d0, d1, d2, d3, d4;
851
852	p->buf[delayA] = p->lastA[filter];
853	p->buf[delayB] = p->filterB[filter];
854	if (p->sample_pos < start) {
855	predictionA = decoded + p->filterA[filter];
856	p->lastA[filter] = decoded;
857	p->filterB[filter] = decoded;
858	p->filterA[filter] = predictionA;
859	return predictionA;
860	}
861	d2 = p->buf[delayA];
862	d1 = (p->buf[delayA] - p->buf[delayA - 1]) << 1;
863	d0 = p->buf[delayA] + ((p->buf[delayA - 2] - p->buf[delayA - 1]) << 3);
864	d3 = p->buf[delayB] * 2 - p->buf[delayB - 1];
865	d4 = p->buf[delayB];
866
867	predictionA = d0 * p->coeffsA[filter][0] +
868	d1 * p->coeffsA[filter][1] +
869	d2 * p->coeffsA[filter][2];
870
871	sign = APESIGN(decoded);
872	p->coeffsA[filter][0] += (((d0 >> 30) & 2) - 1) * sign;
873	p->coeffsA[filter][1] += (((d1 >> 28) & 8) - 4) * sign;
874	p->coeffsA[filter][2] += (((d2 >> 28) & 8) - 4) * sign;
875
876	predictionB = d3 * p->coeffsB[filter][0] -
877	d4 * p->coeffsB[filter][1];
878	p->lastA[filter] = decoded + (predictionA >> 11);
879	sign = APESIGN(p->lastA[filter]);
880	p->coeffsB[filter][0] += (((d3 >> 29) & 4) - 2) * sign;
881	p->coeffsB[filter][1] -= (((d4 >> 30) & 2) - 1) * sign;
882
883	p->filterB[filter] = p->lastA[filter] + (predictionB >> shift);
884	p->filterA[filter] = p->filterB[filter] + ((p->filterA[filter] * 31) >> 5);
885
886	return p->filterA[filter];
887	}
888
889	static void long_filter_high_3800(int32_t *buffer, int order, int shift, int length)
890	{
891	int i, j;
892	int32_t dotprod, sign;
893	int32_t coeffs[256], delay[256];
894
895	if (order >= length)
896	return;
897
898	memset(coeffs, 0, order * sizeof(*coeffs));
899	for (i = 0; i < order; i++)
900	delay[i] = buffer[i];
901	for (i = order; i < length; i++) {
902	dotprod = 0;
903	sign = APESIGN(buffer[i]);
904	for (j = 0; j < order; j++) {
905	dotprod += delay[j] * coeffs[j];
906	coeffs[j] += ((delay[j] >> 31) | 1) * sign;
907	}
908	buffer[i] -= dotprod >> shift;
909	for (j = 0; j < order - 1; j++)
910	delay[j] = delay[j + 1];
911	delay[order - 1] = buffer[i];
912	}
913	}
914
915	static void long_filter_ehigh_3830(int32_t *buffer, int length)
916	{
917	int i, j;
918	int32_t dotprod, sign;
919	int32_t coeffs[8] = { 0 }, delay[8] = { 0 };
920
921	for (i = 0; i < length; i++) {
922	dotprod = 0;
923	sign = APESIGN(buffer[i]);
924	for (j = 7; j >= 0; j--) {
925	dotprod += delay[j] * coeffs[j];
926	coeffs[j] += ((delay[j] >> 31) | 1) * sign;
927	}
928	for (j = 7; j > 0; j--)
929	delay[j] = delay[j - 1];
930	delay[0] = buffer[i];
931	buffer[i] -= dotprod >> 9;
932	}
933	}
934
935	static void predictor_decode_stereo_3800(APEContext *ctx, int count)
936	{
937	APEPredictor *p = &ctx->predictor;
938	int32_t *decoded0 = ctx->decoded[0];
939	int32_t *decoded1 = ctx->decoded[1];
940	int start = 4, shift = 10;
941
942	if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
943	start = 16;
944	long_filter_high_3800(decoded0, 16, 9, count);
945	long_filter_high_3800(decoded1, 16, 9, count);
946	} else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
947	int order = 128, shift2 = 11;
948
949	if (ctx->fileversion >= 3830) {
950	order <<= 1;
951	shift++;
952	shift2++;
953	long_filter_ehigh_3830(decoded0 + order, count - order);
954	long_filter_ehigh_3830(decoded1 + order, count - order);
955	}
956	start = order;
957	long_filter_high_3800(decoded0, order, shift2, count);
958	long_filter_high_3800(decoded1, order, shift2, count);
959	}
960
961	while (count--) {
962	int X = *decoded0, Y = *decoded1;
963	if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
964	*decoded0 = filter_fast_3320(p, Y, 0, YDELAYA);
965	decoded0++;
966	*decoded1 = filter_fast_3320(p, X, 1, XDELAYA);
967	decoded1++;
968	} else {
969	*decoded0 = filter_3800(p, Y, 0, YDELAYA, YDELAYB,
970	start, shift);
971	decoded0++;
972	*decoded1 = filter_3800(p, X, 1, XDELAYA, XDELAYB,
973	start, shift);
974	decoded1++;
975	}
976
977	/* Combined */
978	p->buf++;
979	p->sample_pos++;
980
981	/* Have we filled the history buffer? */
982	if (p->buf == p->historybuffer + HISTORY_SIZE) {
983	memmove(p->historybuffer, p->buf,
984	PREDICTOR_SIZE * sizeof(*p->historybuffer));
985	p->buf = p->historybuffer;
986	}
987	}
988	}
989
990	static void predictor_decode_mono_3800(APEContext *ctx, int count)
991	{
992	APEPredictor *p = &ctx->predictor;
993	int32_t *decoded0 = ctx->decoded[0];
994	int start = 4, shift = 10;
995
996	if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
997	start = 16;
998	long_filter_high_3800(decoded0, 16, 9, count);
999	} else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
1000	int order = 128, shift2 = 11;
1001
1002	if (ctx->fileversion >= 3830) {
1003	order <<= 1;
1004	shift++;
1005	shift2++;
1006	long_filter_ehigh_3830(decoded0 + order, count - order);
1007	}
1008	start = order;
1009	long_filter_high_3800(decoded0, order, shift2, count);
1010	}
1011
1012	while (count--) {
1013	if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
1014	*decoded0 = filter_fast_3320(p, *decoded0, 0, YDELAYA);
1015	decoded0++;
1016	} else {
1017	*decoded0 = filter_3800(p, *decoded0, 0, YDELAYA, YDELAYB,
1018	start, shift);
1019	decoded0++;
1020	}
1021
1022	/* Combined */
1023	p->buf++;
1024	p->sample_pos++;
1025
1026	/* Have we filled the history buffer? */
1027	if (p->buf == p->historybuffer + HISTORY_SIZE) {
1028	memmove(p->historybuffer, p->buf,
1029	PREDICTOR_SIZE * sizeof(*p->historybuffer));
1030	p->buf = p->historybuffer;
1031	}
1032	}
1033	}
1034
1035	static av_always_inline int predictor_update_3930(APEPredictor *p,
1036	const int decoded, const int filter,
1037	const int delayA)
1038	{
1039	int32_t predictionA, sign;
1040	int32_t d0, d1, d2, d3;
1041
1042	p->buf[delayA] = p->lastA[filter];
1043	d0 = p->buf[delayA ];
1044	d1 = p->buf[delayA ] - p->buf[delayA - 1];
1045	d2 = p->buf[delayA - 1] - p->buf[delayA - 2];
1046	d3 = p->buf[delayA - 2] - p->buf[delayA - 3];
1047
1048	predictionA = d0 * p->coeffsA[filter][0] +
1049	d1 * p->coeffsA[filter][1] +
1050	d2 * p->coeffsA[filter][2] +
1051	d3 * p->coeffsA[filter][3];
1052
1053	p->lastA[filter] = decoded + (predictionA >> 9);
1054	p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1055
1056	sign = APESIGN(decoded);
1057	p->coeffsA[filter][0] += ((d0 < 0) * 2 - 1) * sign;
1058	p->coeffsA[filter][1] += ((d1 < 0) * 2 - 1) * sign;
1059	p->coeffsA[filter][2] += ((d2 < 0) * 2 - 1) * sign;
1060	p->coeffsA[filter][3] += ((d3 < 0) * 2 - 1) * sign;
1061
1062	return p->filterA[filter];
1063	}
1064
1065	static void predictor_decode_stereo_3930(APEContext *ctx, int count)
1066	{
1067	APEPredictor *p = &ctx->predictor;
1068	int32_t *decoded0 = ctx->decoded[0];
1069	int32_t *decoded1 = ctx->decoded[1];
1070
1071	ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1072
1073	while (count--) {
1074	/* Predictor Y */
1075	int Y = *decoded1, X = *decoded0;
1076	*decoded0 = predictor_update_3930(p, Y, 0, YDELAYA);
1077	decoded0++;
1078	*decoded1 = predictor_update_3930(p, X, 1, XDELAYA);
1079	decoded1++;
1080
1081	/* Combined */
1082	p->buf++;
1083
1084	/* Have we filled the history buffer? */
1085	if (p->buf == p->historybuffer + HISTORY_SIZE) {
1086	memmove(p->historybuffer, p->buf,
1087	PREDICTOR_SIZE * sizeof(*p->historybuffer));
1088	p->buf = p->historybuffer;
1089	}
1090	}
1091	}
1092
1093	static void predictor_decode_mono_3930(APEContext *ctx, int count)
1094	{
1095	APEPredictor *p = &ctx->predictor;
1096	int32_t *decoded0 = ctx->decoded[0];
1097
1098	ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1099
1100	while (count--) {
1101	*decoded0 = predictor_update_3930(p, *decoded0, 0, YDELAYA);
1102	decoded0++;
1103
1104	p->buf++;
1105
1106	/* Have we filled the history buffer? */
1107	if (p->buf == p->historybuffer + HISTORY_SIZE) {
1108	memmove(p->historybuffer, p->buf,
1109	PREDICTOR_SIZE * sizeof(*p->historybuffer));
1110	p->buf = p->historybuffer;
1111	}
1112	}
1113	}
1114
1115	static av_always_inline int predictor_update_filter(APEPredictor *p,
1116	const int decoded, const int filter,
1117	const int delayA, const int delayB,
1118	const int adaptA, const int adaptB)
1119	{
1120	int32_t predictionA, predictionB, sign;
1121
1122	p->buf[delayA] = p->lastA[filter];
1123	p->buf[adaptA] = APESIGN(p->buf[delayA]);
1124	p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
1125	p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
1126
1127	predictionA = p->buf[delayA ] * p->coeffsA[filter][0] +
1128	p->buf[delayA - 1] * p->coeffsA[filter][1] +
1129	p->buf[delayA - 2] * p->coeffsA[filter][2] +
1130	p->buf[delayA - 3] * p->coeffsA[filter][3];
1131
1132	/* Apply a scaled first-order filter compression */
1133	p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
1134	p->buf[adaptB] = APESIGN(p->buf[delayB]);
1135	p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
1136	p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
1137	p->filterB[filter] = p->filterA[filter ^ 1];
1138
1139	predictionB = p->buf[delayB ] * p->coeffsB[filter][0] +
1140	p->buf[delayB - 1] * p->coeffsB[filter][1] +
1141	p->buf[delayB - 2] * p->coeffsB[filter][2] +
1142	p->buf[delayB - 3] * p->coeffsB[filter][3] +
1143	p->buf[delayB - 4] * p->coeffsB[filter][4];
1144
1145	p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
1146	p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1147
1148	sign = APESIGN(decoded);
1149	p->coeffsA[filter][0] += p->buf[adaptA ] * sign;
1150	p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
1151	p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
1152	p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
1153	p->coeffsB[filter][0] += p->buf[adaptB ] * sign;
1154	p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
1155	p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
1156	p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
1157	p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
1158
1159	return p->filterA[filter];
1160	}
1161
1162	static void predictor_decode_stereo_3950(APEContext *ctx, int count)
1163	{
1164	APEPredictor *p = &ctx->predictor;
1165	int32_t *decoded0 = ctx->decoded[0];
1166	int32_t *decoded1 = ctx->decoded[1];
1167
1168	ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1169
1170	while (count--) {
1171	/* Predictor Y */
1172	*decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
1173	YADAPTCOEFFSA, YADAPTCOEFFSB);
1174	decoded0++;
1175	*decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
1176	XADAPTCOEFFSA, XADAPTCOEFFSB);
1177	decoded1++;
1178
1179	/* Combined */
1180	p->buf++;
1181
1182	/* Have we filled the history buffer? */
1183	if (p->buf == p->historybuffer + HISTORY_SIZE) {
1184	memmove(p->historybuffer, p->buf,
1185	PREDICTOR_SIZE * sizeof(*p->historybuffer));
1186	p->buf = p->historybuffer;
1187	}
1188	}
1189	}
1190
1191	static void predictor_decode_mono_3950(APEContext *ctx, int count)
1192	{
1193	APEPredictor *p = &ctx->predictor;
1194	int32_t *decoded0 = ctx->decoded[0];
1195	int32_t predictionA, currentA, A, sign;
1196
1197	ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1198
1199	currentA = p->lastA[0];
1200
1201	while (count--) {
1202	A = *decoded0;
1203
1204	p->buf[YDELAYA] = currentA;
1205	p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
1206
1207	predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] +
1208	p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
1209	p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
1210	p->buf[YDELAYA - 3] * p->coeffsA[0][3];
1211
1212	currentA = A + (predictionA >> 10);
1213
1214	p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]);
1215	p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
1216
1217	sign = APESIGN(A);
1218	p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign;
1219	p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
1220	p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
1221	p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
1222
1223	p->buf++;
1224
1225	/* Have we filled the history buffer? */
1226	if (p->buf == p->historybuffer + HISTORY_SIZE) {
1227	memmove(p->historybuffer, p->buf,
1228	PREDICTOR_SIZE * sizeof(*p->historybuffer));
1229	p->buf = p->historybuffer;
1230	}
1231
1232	p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
1233	*(decoded0++) = p->filterA[0];
1234	}
1235
1236	p->lastA[0] = currentA;
1237	}
1238
1239	static void do_init_filter(APEFilter *f, int16_t *buf, int order)
1240	{
1241	f->coeffs = buf;
1242	f->historybuffer = buf + order;
1243	f->delay = f->historybuffer + order * 2;
1244	f->adaptcoeffs = f->historybuffer + order;
1245
1246	memset(f->historybuffer, 0, (order * 2) * sizeof(*f->historybuffer));
1247	memset(f->coeffs, 0, order * sizeof(*f->coeffs));
1248	f->avg = 0;
1249	}
1250
1251	static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
1252	{
1253	do_init_filter(&f[0], buf, order);
1254	do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
1255	}
1256
1257	static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
1258	int32_t *data, int count, int order, int fracbits)
1259	{
1260	int res;
1261	int absres;
1262
1263	while (count--) {
1264	/* round fixedpoint scalar product */
1265	res = ctx->adsp.scalarproduct_and_madd_int16(f->coeffs,
1266	f->delay - order,
1267	f->adaptcoeffs - order,
1268	order, APESIGN(*data));
1269	res = (res + (1 << (fracbits - 1))) >> fracbits;
1270	res += *data;
1271	*data++ = res;
1272
1273	/* Update the output history */
1274	*f->delay++ = av_clip_int16(res);
1275
1276	if (version < 3980) {
1277	/* Version ??? to < 3.98 files (untested) */
1278	f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
1279	f->adaptcoeffs[-4] >>= 1;
1280	f->adaptcoeffs[-8] >>= 1;
1281	} else {
1282	/* Version 3.98 and later files */
1283
1284	/* Update the adaption coefficients */
1285	absres = FFABS(res);
1286	if (absres)
1287	*f->adaptcoeffs = APESIGN(res) *
1288	(8 << ((absres > f->avg * 3) + (absres > f->avg * 4 / 3)));
1289	/* equivalent to the following code
1290	if (absres <= f->avg * 4 / 3)
1291	*f->adaptcoeffs = APESIGN(res) * 8;
1292	else if (absres <= f->avg * 3)
1293	*f->adaptcoeffs = APESIGN(res) * 16;
1294	else
1295	*f->adaptcoeffs = APESIGN(res) * 32;
1296	*/
1297	else
1298	*f->adaptcoeffs = 0;
1299
1300	f->avg += (absres - f->avg) / 16;
1301
1302	f->adaptcoeffs[-1] >>= 1;
1303	f->adaptcoeffs[-2] >>= 1;
1304	f->adaptcoeffs[-8] >>= 1;
1305	}
1306
1307	f->adaptcoeffs++;
1308
1309	/* Have we filled the history buffer? */
1310	if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
1311	memmove(f->historybuffer, f->delay - (order * 2),
1312	(order * 2) * sizeof(*f->historybuffer));
1313	f->delay = f->historybuffer + order * 2;
1314	f->adaptcoeffs = f->historybuffer + order;
1315	}
1316	}
1317	}
1318
1319	static void apply_filter(APEContext *ctx, APEFilter *f,
1320	int32_t *data0, int32_t *data1,
1321	int count, int order, int fracbits)
1322	{
1323	do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
1324	if (data1)
1325	do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
1326	}
1327
1328	static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
1329	int32_t *decoded1, int count)
1330	{
1331	int i;
1332
1333	for (i = 0; i < APE_FILTER_LEVELS; i++) {
1334	if (!ape_filter_orders[ctx->fset][i])
1335	break;
1336	apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
1337	ape_filter_orders[ctx->fset][i],
1338	ape_filter_fracbits[ctx->fset][i]);
1339	}
1340	}
1341
1342	static int init_frame_decoder(APEContext *ctx)
1343	{
1344	int i, ret;
1345	if ((ret = init_entropy_decoder(ctx)) < 0)
1346	return ret;
1347	init_predictor_decoder(ctx);
1348
1349	for (i = 0; i < APE_FILTER_LEVELS; i++) {
1350	if (!ape_filter_orders[ctx->fset][i])
1351	break;
1352	init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
1353	ape_filter_orders[ctx->fset][i]);
1354	}
1355	return 0;
1356	}
1357
1358	static void ape_unpack_mono(APEContext *ctx, int count)
1359	{
1360	if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
1361	/* We are pure silence, so we're done. */
1362	av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
1363	return;
1364	}
1365
1366	ctx->entropy_decode_mono(ctx, count);
1367
1368	/* Now apply the predictor decoding */
1369	ctx->predictor_decode_mono(ctx, count);
1370
1371	/* Pseudo-stereo - just copy left channel to right channel */
1372	if (ctx->channels == 2) {
1373	memcpy(ctx->decoded[1], ctx->decoded[0], count * sizeof(*ctx->decoded[1]));
1374	}
1375	}
1376
1377	static void ape_unpack_stereo(APEContext *ctx, int count)
1378	{
1379	int32_t left, right;
1380	int32_t *decoded0 = ctx->decoded[0];
1381	int32_t *decoded1 = ctx->decoded[1];
1382
1383	if ((ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) == APE_FRAMECODE_STEREO_SILENCE) {
1384	/* We are pure silence, so we're done. */
1385	av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
1386	return;
1387	}
1388
1389	ctx->entropy_decode_stereo(ctx, count);
1390
1391	/* Now apply the predictor decoding */
1392	ctx->predictor_decode_stereo(ctx, count);
1393
1394	/* Decorrelate and scale to output depth */
1395	while (count--) {
1396	left = *decoded1 - (*decoded0 / 2);
1397	right = left + *decoded0;
1398
1399	*(decoded0++) = left;
1400	*(decoded1++) = right;
1401	}
1402	}
1403
1404	static int ape_decode_frame(AVCodecContext *avctx, void *data,
1405	int *got_frame_ptr, AVPacket *avpkt)
1406	{
1407	AVFrame *frame = data;
1408	const uint8_t *buf = avpkt->data;
1409	APEContext *s = avctx->priv_data;
1410	uint8_t *sample8;
1411	int16_t *sample16;
1412	int32_t *sample24;
1413	int i, ch, ret;
1414	int blockstodecode;
1415
1416	/* this should never be negative, but bad things will happen if it is, so
1417	check it just to make sure. */
1418	av_assert0(s->samples >= 0);
1419
1420	if(!s->samples){
1421	uint32_t nblocks, offset;
1422	int buf_size;
1423
1424	if (!avpkt->size) {
1425	*got_frame_ptr = 0;
1426	return 0;
1427	}
1428	if (avpkt->size < 8) {
1429	av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1430	return AVERROR_INVALIDDATA;
1431	}
1432	buf_size = avpkt->size & ~3;
1433	if (buf_size != avpkt->size) {
1434	av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
1435	"extra bytes at the end will be skipped.\n");
1436	}
1437	if (s->fileversion < 3950) // previous versions overread two bytes
1438	buf_size += 2;
1439	av_fast_padded_malloc(&s->data, &s->data_size, buf_size);
1440	if (!s->data)
1441	return AVERROR(ENOMEM);
1442	s->bdsp.bswap_buf((uint32_t *) s->data, (const uint32_t *) buf,
1443	buf_size >> 2);
1444	memset(s->data + (buf_size & ~3), 0, buf_size & 3);
1445	s->ptr = s->data;
1446	s->data_end = s->data + buf_size;
1447
1448	nblocks = bytestream_get_be32(&s->ptr);
1449	offset = bytestream_get_be32(&s->ptr);
1450	if (s->fileversion >= 3900) {
1451	if (offset > 3) {
1452	av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
1453	s->data = NULL;
1454	return AVERROR_INVALIDDATA;
1455	}
1456	if (s->data_end - s->ptr < offset) {
1457	av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1458	return AVERROR_INVALIDDATA;
1459	}
1460	s->ptr += offset;
1461	} else {
1462	if ((ret = init_get_bits8(&s->gb, s->ptr, s->data_end - s->ptr)) < 0)
1463	return ret;
1464	if (s->fileversion > 3800)
1465	skip_bits_long(&s->gb, offset * 8);
1466	else
1467	skip_bits_long(&s->gb, offset);
1468	}
1469
1470	if (!nblocks || nblocks > INT_MAX) {
1471	av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %"PRIu32".\n",
1472	nblocks);
1473	return AVERROR_INVALIDDATA;
1474	}
1475
1476	/* Initialize the frame decoder */
1477	if (init_frame_decoder(s) < 0) {
1478	av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
1479	return AVERROR_INVALIDDATA;
1480	}
1481	s->samples = nblocks;
1482	}
1483
1484	if (!s->data) {
1485	*got_frame_ptr = 0;
1486	return avpkt->size;
1487	}
1488
1489	blockstodecode = FFMIN(s->blocks_per_loop, s->samples);
1490	// for old files coefficients were not interleaved,
1491	// so we need to decode all of them at once
1492	if (s->fileversion < 3930)
1493	blockstodecode = s->samples;
1494
1495	/* reallocate decoded sample buffer if needed */
1496	av_fast_malloc(&s->decoded_buffer, &s->decoded_size,
1497	2 * FFALIGN(blockstodecode, 8) * sizeof(*s->decoded_buffer));
1498	if (!s->decoded_buffer)
1499	return AVERROR(ENOMEM);
1500	memset(s->decoded_buffer, 0, s->decoded_size);
1501	s->decoded[0] = s->decoded_buffer;
1502	s->decoded[1] = s->decoded_buffer + FFALIGN(blockstodecode, 8);
1503
1504	/* get output buffer */
1505	frame->nb_samples = blockstodecode;
1506	if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1507	return ret;
1508
1509	s->error=0;
1510
1511	if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
1512	ape_unpack_mono(s, blockstodecode);
1513	else
1514	ape_unpack_stereo(s, blockstodecode);
1515	emms_c();
1516
1517	if (s->error) {
1518	s->samples=0;
1519	av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
1520	return AVERROR_INVALIDDATA;
1521	}
1522
1523	switch (s->bps) {
1524	case 8:
1525	for (ch = 0; ch < s->channels; ch++) {
1526	sample8 = (uint8_t *)frame->data[ch];
1527	for (i = 0; i < blockstodecode; i++)
1528	*sample8++ = (s->decoded[ch][i] + 0x80) & 0xff;
1529	}
1530	break;
1531	case 16:
1532	for (ch = 0; ch < s->channels; ch++) {
1533	sample16 = (int16_t *)frame->data[ch];
1534	for (i = 0; i < blockstodecode; i++)
1535	*sample16++ = s->decoded[ch][i];
1536	}
1537	break;
1538	case 24:
1539	for (ch = 0; ch < s->channels; ch++) {
1540	sample24 = (int32_t *)frame->data[ch];
1541	for (i = 0; i < blockstodecode; i++)
1542	*sample24++ = s->decoded[ch][i] << 8;
1543	}
1544	break;
1545	}
1546
1547	s->samples -= blockstodecode;
1548
1549	*got_frame_ptr = 1;
1550
1551	return !s->samples ? avpkt->size : 0;
1552	}
1553
1554	static void ape_flush(AVCodecContext *avctx)
1555	{
1556	APEContext *s = avctx->priv_data;
1557	s->samples= 0;
1558	}
1559
1560	#define OFFSET(x) offsetof(APEContext, x)
1561	#define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1562	static const AVOption options[] = {
1563	{ "max_samples", "maximum number of samples decoded per call", OFFSET(blocks_per_loop), AV_OPT_TYPE_INT, { .i64 = 4608 }, 1, INT_MAX, PAR, "max_samples" },
1564	{ "all", "no maximum. decode all samples for each packet at once", 0, AV_OPT_TYPE_CONST, { .i64 = INT_MAX }, INT_MIN, INT_MAX, PAR, "max_samples" },
1565	{ NULL},
1566	};
1567
1568	static const AVClass ape_decoder_class = {
1569	.class_name = "APE decoder",
1570	.item_name = av_default_item_name,
1571	.option = options,
1572	.version = LIBAVUTIL_VERSION_INT,
1573	};
1574
1575	AVCodec ff_ape_decoder = {
1576	.name = "ape",
1577	.long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
1578	.type = AVMEDIA_TYPE_AUDIO,
1579	.id = AV_CODEC_ID_APE,
1580	.priv_data_size = sizeof(APEContext),
1581	.init = ape_decode_init,
1582	.close = ape_decode_close,
1583	.decode = ape_decode_frame,
1584	.capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DELAY |
1585	AV_CODEC_CAP_DR1,
1586	.flush = ape_flush,
1587	.sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
1588	AV_SAMPLE_FMT_S16P,
1589	AV_SAMPLE_FMT_S32P,
1590	AV_SAMPLE_FMT_NONE },
1591	.priv_class = &ape_decoder_class,
1592	};
1593