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
ape_decode_close(AVCodecContext * avctx)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
ape_decode_init(AVCodecContext * avctx)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_DEBUG, "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 */
range_start_decoding(APEContext * ctx)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 */
range_dec_normalize(APEContext * ctx)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 culmulative 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 culmulative frequency
346 */
range_decode_culfreq(APEContext * ctx,int tot_f)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 */
range_decode_culshift(APEContext * ctx,int shift)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 */
range_decode_update(APEContext * ctx,int sy_f,int lt_f)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 */
range_decode_bits(APEContext * ctx,int n)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 */
range_get_symbol(APEContext * ctx,const uint16_t counts[],const uint16_t counts_diff[])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
update_rice(APERice * rice,unsigned int x)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
get_rice_ook(GetBitContext * gb,int k)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
ape_decode_value_3860(APEContext * ctx,GetBitContext * gb,APERice * rice)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: %d\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 if (x & 1)
509 return (x >> 1) + 1;
510 else
511 return -(x >> 1);
512 }
513
ape_decode_value_3900(APEContext * ctx,APERice * rice)514 static inline int ape_decode_value_3900(APEContext *ctx, APERice *rice)
515 {
516 unsigned int x, overflow;
517 int tmpk;
518
519 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
520
521 if (overflow == (MODEL_ELEMENTS - 1)) {
522 tmpk = range_decode_bits(ctx, 5);
523 overflow = 0;
524 } else
525 tmpk = (rice->k < 1) ? 0 : rice->k - 1;
526
527 if (tmpk <= 16 || ctx->fileversion < 3910) {
528 if (tmpk > 23) {
529 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
530 return AVERROR_INVALIDDATA;
531 }
532 x = range_decode_bits(ctx, tmpk);
533 } else if (tmpk <= 31) {
534 x = range_decode_bits(ctx, 16);
535 x |= (range_decode_bits(ctx, tmpk - 16) << 16);
536 } else {
537 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
538 return AVERROR_INVALIDDATA;
539 }
540 x += overflow << tmpk;
541
542 update_rice(rice, x);
543
544 /* Convert to signed */
545 if (x & 1)
546 return (x >> 1) + 1;
547 else
548 return -(x >> 1);
549 }
550
ape_decode_value_3990(APEContext * ctx,APERice * rice)551 static inline int ape_decode_value_3990(APEContext *ctx, APERice *rice)
552 {
553 unsigned int x, overflow;
554 int base, pivot;
555
556 pivot = rice->ksum >> 5;
557 if (pivot == 0)
558 pivot = 1;
559
560 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
561
562 if (overflow == (MODEL_ELEMENTS - 1)) {
563 overflow = range_decode_bits(ctx, 16) << 16;
564 overflow |= range_decode_bits(ctx, 16);
565 }
566
567 if (pivot < 0x10000) {
568 base = range_decode_culfreq(ctx, pivot);
569 range_decode_update(ctx, 1, base);
570 } else {
571 int base_hi = pivot, base_lo;
572 int bbits = 0;
573
574 while (base_hi & ~0xFFFF) {
575 base_hi >>= 1;
576 bbits++;
577 }
578 base_hi = range_decode_culfreq(ctx, base_hi + 1);
579 range_decode_update(ctx, 1, base_hi);
580 base_lo = range_decode_culfreq(ctx, 1 << bbits);
581 range_decode_update(ctx, 1, base_lo);
582
583 base = (base_hi << bbits) + base_lo;
584 }
585
586 x = base + overflow * pivot;
587
588 update_rice(rice, x);
589
590 /* Convert to signed */
591 if (x & 1)
592 return (x >> 1) + 1;
593 else
594 return -(x >> 1);
595 }
596
decode_array_0000(APEContext * ctx,GetBitContext * gb,int32_t * out,APERice * rice,int blockstodecode)597 static void decode_array_0000(APEContext *ctx, GetBitContext *gb,
598 int32_t *out, APERice *rice, int blockstodecode)
599 {
600 int i;
601 int ksummax, ksummin;
602
603 rice->ksum = 0;
604 for (i = 0; i < 5; i++) {
605 out[i] = get_rice_ook(&ctx->gb, 10);
606 rice->ksum += out[i];
607 }
608 rice->k = av_log2(rice->ksum / 10) + 1;
609 if (rice->k >= 24)
610 return;
611 for (; i < 64; i++) {
612 out[i] = get_rice_ook(&ctx->gb, rice->k);
613 rice->ksum += out[i];
614 rice->k = av_log2(rice->ksum / ((i + 1) * 2)) + 1;
615 if (rice->k >= 24)
616 return;
617 }
618 ksummax = 1 << rice->k + 7;
619 ksummin = rice->k ? (1 << rice->k + 6) : 0;
620 for (; i < blockstodecode; i++) {
621 out[i] = get_rice_ook(&ctx->gb, rice->k);
622 rice->ksum += out[i] - out[i - 64];
623 while (rice->ksum < ksummin) {
624 rice->k--;
625 ksummin = rice->k ? ksummin >> 1 : 0;
626 ksummax >>= 1;
627 }
628 while (rice->ksum >= ksummax) {
629 rice->k++;
630 if (rice->k > 24)
631 return;
632 ksummax <<= 1;
633 ksummin = ksummin ? ksummin << 1 : 128;
634 }
635 }
636
637 for (i = 0; i < blockstodecode; i++) {
638 if (out[i] & 1)
639 out[i] = (out[i] >> 1) + 1;
640 else
641 out[i] = -(out[i] >> 1);
642 }
643 }
644
entropy_decode_mono_0000(APEContext * ctx,int blockstodecode)645 static void entropy_decode_mono_0000(APEContext *ctx, int blockstodecode)
646 {
647 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
648 blockstodecode);
649 }
650
entropy_decode_stereo_0000(APEContext * ctx,int blockstodecode)651 static void entropy_decode_stereo_0000(APEContext *ctx, int blockstodecode)
652 {
653 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
654 blockstodecode);
655 decode_array_0000(ctx, &ctx->gb, ctx->decoded[1], &ctx->riceX,
656 blockstodecode);
657 }
658
entropy_decode_mono_3860(APEContext * ctx,int blockstodecode)659 static void entropy_decode_mono_3860(APEContext *ctx, int blockstodecode)
660 {
661 int32_t *decoded0 = ctx->decoded[0];
662
663 while (blockstodecode--)
664 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
665 }
666
entropy_decode_stereo_3860(APEContext * ctx,int blockstodecode)667 static void entropy_decode_stereo_3860(APEContext *ctx, int blockstodecode)
668 {
669 int32_t *decoded0 = ctx->decoded[0];
670 int32_t *decoded1 = ctx->decoded[1];
671 int blocks = blockstodecode;
672
673 while (blockstodecode--)
674 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
675 while (blocks--)
676 *decoded1++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceX);
677 }
678
entropy_decode_mono_3900(APEContext * ctx,int blockstodecode)679 static void entropy_decode_mono_3900(APEContext *ctx, int blockstodecode)
680 {
681 int32_t *decoded0 = ctx->decoded[0];
682
683 while (blockstodecode--)
684 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
685 }
686
entropy_decode_stereo_3900(APEContext * ctx,int blockstodecode)687 static void entropy_decode_stereo_3900(APEContext *ctx, int blockstodecode)
688 {
689 int32_t *decoded0 = ctx->decoded[0];
690 int32_t *decoded1 = ctx->decoded[1];
691 int blocks = blockstodecode;
692
693 while (blockstodecode--)
694 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
695 range_dec_normalize(ctx);
696 // because of some implementation peculiarities we need to backpedal here
697 ctx->ptr -= 1;
698 range_start_decoding(ctx);
699 while (blocks--)
700 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
701 }
702
entropy_decode_stereo_3930(APEContext * ctx,int blockstodecode)703 static void entropy_decode_stereo_3930(APEContext *ctx, int blockstodecode)
704 {
705 int32_t *decoded0 = ctx->decoded[0];
706 int32_t *decoded1 = ctx->decoded[1];
707
708 while (blockstodecode--) {
709 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
710 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
711 }
712 }
713
entropy_decode_mono_3990(APEContext * ctx,int blockstodecode)714 static void entropy_decode_mono_3990(APEContext *ctx, int blockstodecode)
715 {
716 int32_t *decoded0 = ctx->decoded[0];
717
718 while (blockstodecode--)
719 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
720 }
721
entropy_decode_stereo_3990(APEContext * ctx,int blockstodecode)722 static void entropy_decode_stereo_3990(APEContext *ctx, int blockstodecode)
723 {
724 int32_t *decoded0 = ctx->decoded[0];
725 int32_t *decoded1 = ctx->decoded[1];
726
727 while (blockstodecode--) {
728 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
729 *decoded1++ = ape_decode_value_3990(ctx, &ctx->riceX);
730 }
731 }
732
init_entropy_decoder(APEContext * ctx)733 static int init_entropy_decoder(APEContext *ctx)
734 {
735 /* Read the CRC */
736 if (ctx->fileversion >= 3900) {
737 if (ctx->data_end - ctx->ptr < 6)
738 return AVERROR_INVALIDDATA;
739 ctx->CRC = bytestream_get_be32(&ctx->ptr);
740 } else {
741 ctx->CRC = get_bits_long(&ctx->gb, 32);
742 }
743
744 /* Read the frame flags if they exist */
745 ctx->frameflags = 0;
746 if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
747 ctx->CRC &= ~0x80000000;
748
749 if (ctx->data_end - ctx->ptr < 6)
750 return AVERROR_INVALIDDATA;
751 ctx->frameflags = bytestream_get_be32(&ctx->ptr);
752 }
753
754 /* Initialize the rice structs */
755 ctx->riceX.k = 10;
756 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
757 ctx->riceY.k = 10;
758 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
759
760 if (ctx->fileversion >= 3900) {
761 /* The first 8 bits of input are ignored. */
762 ctx->ptr++;
763
764 range_start_decoding(ctx);
765 }
766
767 return 0;
768 }
769
770 static const int32_t initial_coeffs_fast_3320[1] = {
771 375,
772 };
773
774 static const int32_t initial_coeffs_a_3800[3] = {
775 64, 115, 64,
776 };
777
778 static const int32_t initial_coeffs_b_3800[2] = {
779 740, 0
780 };
781
782 static const int32_t initial_coeffs_3930[4] = {
783 360, 317, -109, 98
784 };
785
init_predictor_decoder(APEContext * ctx)786 static void init_predictor_decoder(APEContext *ctx)
787 {
788 APEPredictor *p = &ctx->predictor;
789
790 /* Zero the history buffers */
791 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(*p->historybuffer));
792 p->buf = p->historybuffer;
793
794 /* Initialize and zero the coefficients */
795 if (ctx->fileversion < 3930) {
796 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
797 memcpy(p->coeffsA[0], initial_coeffs_fast_3320,
798 sizeof(initial_coeffs_fast_3320));
799 memcpy(p->coeffsA[1], initial_coeffs_fast_3320,
800 sizeof(initial_coeffs_fast_3320));
801 } else {
802 memcpy(p->coeffsA[0], initial_coeffs_a_3800,
803 sizeof(initial_coeffs_a_3800));
804 memcpy(p->coeffsA[1], initial_coeffs_a_3800,
805 sizeof(initial_coeffs_a_3800));
806 }
807 } else {
808 memcpy(p->coeffsA[0], initial_coeffs_3930, sizeof(initial_coeffs_3930));
809 memcpy(p->coeffsA[1], initial_coeffs_3930, sizeof(initial_coeffs_3930));
810 }
811 memset(p->coeffsB, 0, sizeof(p->coeffsB));
812 if (ctx->fileversion < 3930) {
813 memcpy(p->coeffsB[0], initial_coeffs_b_3800,
814 sizeof(initial_coeffs_b_3800));
815 memcpy(p->coeffsB[1], initial_coeffs_b_3800,
816 sizeof(initial_coeffs_b_3800));
817 }
818
819 p->filterA[0] = p->filterA[1] = 0;
820 p->filterB[0] = p->filterB[1] = 0;
821 p->lastA[0] = p->lastA[1] = 0;
822
823 p->sample_pos = 0;
824 }
825
826 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
APESIGN(int32_t x)827 static inline int APESIGN(int32_t x) {
828 return (x < 0) - (x > 0);
829 }
830
filter_fast_3320(APEPredictor * p,const int decoded,const int filter,const int delayA)831 static av_always_inline int filter_fast_3320(APEPredictor *p,
832 const int decoded, const int filter,
833 const int delayA)
834 {
835 int32_t predictionA;
836
837 p->buf[delayA] = p->lastA[filter];
838 if (p->sample_pos < 3) {
839 p->lastA[filter] = decoded;
840 p->filterA[filter] = decoded;
841 return decoded;
842 }
843
844 predictionA = p->buf[delayA] * 2 - p->buf[delayA - 1];
845 p->lastA[filter] = decoded + (predictionA * p->coeffsA[filter][0] >> 9);
846
847 if ((decoded ^ predictionA) > 0)
848 p->coeffsA[filter][0]++;
849 else
850 p->coeffsA[filter][0]--;
851
852 p->filterA[filter] += p->lastA[filter];
853
854 return p->filterA[filter];
855 }
856
filter_3800(APEPredictor * p,const int decoded,const int filter,const int delayA,const int delayB,const int start,const int shift)857 static av_always_inline int filter_3800(APEPredictor *p,
858 const int decoded, const int filter,
859 const int delayA, const int delayB,
860 const int start, const int shift)
861 {
862 int32_t predictionA, predictionB, sign;
863 int32_t d0, d1, d2, d3, d4;
864
865 p->buf[delayA] = p->lastA[filter];
866 p->buf[delayB] = p->filterB[filter];
867 if (p->sample_pos < start) {
868 predictionA = decoded + p->filterA[filter];
869 p->lastA[filter] = decoded;
870 p->filterB[filter] = decoded;
871 p->filterA[filter] = predictionA;
872 return predictionA;
873 }
874 d2 = p->buf[delayA];
875 d1 = (p->buf[delayA] - p->buf[delayA - 1]) << 1;
876 d0 = p->buf[delayA] + ((p->buf[delayA - 2] - p->buf[delayA - 1]) << 3);
877 d3 = p->buf[delayB] * 2 - p->buf[delayB - 1];
878 d4 = p->buf[delayB];
879
880 predictionA = d0 * p->coeffsA[filter][0] +
881 d1 * p->coeffsA[filter][1] +
882 d2 * p->coeffsA[filter][2];
883
884 sign = APESIGN(decoded);
885 p->coeffsA[filter][0] += (((d0 >> 30) & 2) - 1) * sign;
886 p->coeffsA[filter][1] += (((d1 >> 28) & 8) - 4) * sign;
887 p->coeffsA[filter][2] += (((d2 >> 28) & 8) - 4) * sign;
888
889 predictionB = d3 * p->coeffsB[filter][0] -
890 d4 * p->coeffsB[filter][1];
891 p->lastA[filter] = decoded + (predictionA >> 11);
892 sign = APESIGN(p->lastA[filter]);
893 p->coeffsB[filter][0] += (((d3 >> 29) & 4) - 2) * sign;
894 p->coeffsB[filter][1] -= (((d4 >> 30) & 2) - 1) * sign;
895
896 p->filterB[filter] = p->lastA[filter] + (predictionB >> shift);
897 p->filterA[filter] = p->filterB[filter] + ((p->filterA[filter] * 31) >> 5);
898
899 return p->filterA[filter];
900 }
901
long_filter_high_3800(int32_t * buffer,int order,int shift,int32_t * coeffs,int32_t * delay,int length)902 static void long_filter_high_3800(int32_t *buffer, int order, int shift,
903 int32_t *coeffs, int32_t *delay, int length)
904 {
905 int i, j;
906 int32_t dotprod, sign;
907
908 memset(coeffs, 0, order * sizeof(*coeffs));
909 for (i = 0; i < order; i++)
910 delay[i] = buffer[i];
911 for (i = order; i < length; i++) {
912 dotprod = 0;
913 sign = APESIGN(buffer[i]);
914 for (j = 0; j < order; j++) {
915 dotprod += delay[j] * coeffs[j];
916 coeffs[j] += ((delay[j] >> 31) | 1) * sign;
917 }
918 buffer[i] -= dotprod >> shift;
919 for (j = 0; j < order - 1; j++)
920 delay[j] = delay[j + 1];
921 delay[order - 1] = buffer[i];
922 }
923 }
924
long_filter_ehigh_3830(int32_t * buffer,int length)925 static void long_filter_ehigh_3830(int32_t *buffer, int length)
926 {
927 int i, j;
928 int32_t dotprod, sign;
929 int32_t coeffs[8] = { 0 }, delay[8] = { 0 };
930
931 for (i = 0; i < length; i++) {
932 dotprod = 0;
933 sign = APESIGN(buffer[i]);
934 for (j = 7; j >= 0; j--) {
935 dotprod += delay[j] * coeffs[j];
936 coeffs[j] += ((delay[j] >> 31) | 1) * sign;
937 }
938 for (j = 7; j > 0; j--)
939 delay[j] = delay[j - 1];
940 delay[0] = buffer[i];
941 buffer[i] -= dotprod >> 9;
942 }
943 }
944
predictor_decode_stereo_3800(APEContext * ctx,int count)945 static void predictor_decode_stereo_3800(APEContext *ctx, int count)
946 {
947 APEPredictor *p = &ctx->predictor;
948 int32_t *decoded0 = ctx->decoded[0];
949 int32_t *decoded1 = ctx->decoded[1];
950 int32_t coeffs[256], delay[256];
951 int start = 4, shift = 10;
952
953 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
954 start = 16;
955 long_filter_high_3800(decoded0, 16, 9, coeffs, delay, count);
956 long_filter_high_3800(decoded1, 16, 9, coeffs, delay, count);
957 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
958 int order = 128, shift2 = 11;
959
960 if (ctx->fileversion >= 3830) {
961 order <<= 1;
962 shift++;
963 shift2++;
964 long_filter_ehigh_3830(decoded0 + order, count - order);
965 long_filter_ehigh_3830(decoded1 + order, count - order);
966 }
967 start = order;
968 long_filter_high_3800(decoded0, order, shift2, coeffs, delay, count);
969 long_filter_high_3800(decoded1, order, shift2, coeffs, delay, count);
970 }
971
972 while (count--) {
973 int X = *decoded0, Y = *decoded1;
974 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
975 *decoded0 = filter_fast_3320(p, Y, 0, YDELAYA);
976 decoded0++;
977 *decoded1 = filter_fast_3320(p, X, 1, XDELAYA);
978 decoded1++;
979 } else {
980 *decoded0 = filter_3800(p, Y, 0, YDELAYA, YDELAYB,
981 start, shift);
982 decoded0++;
983 *decoded1 = filter_3800(p, X, 1, XDELAYA, XDELAYB,
984 start, shift);
985 decoded1++;
986 }
987
988 /* Combined */
989 p->buf++;
990 p->sample_pos++;
991
992 /* Have we filled the history buffer? */
993 if (p->buf == p->historybuffer + HISTORY_SIZE) {
994 memmove(p->historybuffer, p->buf,
995 PREDICTOR_SIZE * sizeof(*p->historybuffer));
996 p->buf = p->historybuffer;
997 }
998 }
999 }
1000
predictor_decode_mono_3800(APEContext * ctx,int count)1001 static void predictor_decode_mono_3800(APEContext *ctx, int count)
1002 {
1003 APEPredictor *p = &ctx->predictor;
1004 int32_t *decoded0 = ctx->decoded[0];
1005 int32_t coeffs[256], delay[256];
1006 int start = 4, shift = 10;
1007
1008 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
1009 start = 16;
1010 long_filter_high_3800(decoded0, 16, 9, coeffs, delay, count);
1011 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
1012 int order = 128, shift2 = 11;
1013
1014 if (ctx->fileversion >= 3830) {
1015 order <<= 1;
1016 shift++;
1017 shift2++;
1018 long_filter_ehigh_3830(decoded0 + order, count - order);
1019 }
1020 start = order;
1021 long_filter_high_3800(decoded0, order, shift2, coeffs, delay, count);
1022 }
1023
1024 while (count--) {
1025 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
1026 *decoded0 = filter_fast_3320(p, *decoded0, 0, YDELAYA);
1027 decoded0++;
1028 } else {
1029 *decoded0 = filter_3800(p, *decoded0, 0, YDELAYA, YDELAYB,
1030 start, shift);
1031 decoded0++;
1032 }
1033
1034 /* Combined */
1035 p->buf++;
1036 p->sample_pos++;
1037
1038 /* Have we filled the history buffer? */
1039 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1040 memmove(p->historybuffer, p->buf,
1041 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1042 p->buf = p->historybuffer;
1043 }
1044 }
1045 }
1046
predictor_update_3930(APEPredictor * p,const int decoded,const int filter,const int delayA)1047 static av_always_inline int predictor_update_3930(APEPredictor *p,
1048 const int decoded, const int filter,
1049 const int delayA)
1050 {
1051 int32_t predictionA, sign;
1052 int32_t d0, d1, d2, d3;
1053
1054 p->buf[delayA] = p->lastA[filter];
1055 d0 = p->buf[delayA ];
1056 d1 = p->buf[delayA ] - p->buf[delayA - 1];
1057 d2 = p->buf[delayA - 1] - p->buf[delayA - 2];
1058 d3 = p->buf[delayA - 2] - p->buf[delayA - 3];
1059
1060 predictionA = d0 * p->coeffsA[filter][0] +
1061 d1 * p->coeffsA[filter][1] +
1062 d2 * p->coeffsA[filter][2] +
1063 d3 * p->coeffsA[filter][3];
1064
1065 p->lastA[filter] = decoded + (predictionA >> 9);
1066 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1067
1068 sign = APESIGN(decoded);
1069 p->coeffsA[filter][0] += ((d0 < 0) * 2 - 1) * sign;
1070 p->coeffsA[filter][1] += ((d1 < 0) * 2 - 1) * sign;
1071 p->coeffsA[filter][2] += ((d2 < 0) * 2 - 1) * sign;
1072 p->coeffsA[filter][3] += ((d3 < 0) * 2 - 1) * sign;
1073
1074 return p->filterA[filter];
1075 }
1076
predictor_decode_stereo_3930(APEContext * ctx,int count)1077 static void predictor_decode_stereo_3930(APEContext *ctx, int count)
1078 {
1079 APEPredictor *p = &ctx->predictor;
1080 int32_t *decoded0 = ctx->decoded[0];
1081 int32_t *decoded1 = ctx->decoded[1];
1082
1083 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1084
1085 while (count--) {
1086 /* Predictor Y */
1087 int Y = *decoded1, X = *decoded0;
1088 *decoded0 = predictor_update_3930(p, Y, 0, YDELAYA);
1089 decoded0++;
1090 *decoded1 = predictor_update_3930(p, X, 1, XDELAYA);
1091 decoded1++;
1092
1093 /* Combined */
1094 p->buf++;
1095
1096 /* Have we filled the history buffer? */
1097 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1098 memmove(p->historybuffer, p->buf,
1099 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1100 p->buf = p->historybuffer;
1101 }
1102 }
1103 }
1104
predictor_decode_mono_3930(APEContext * ctx,int count)1105 static void predictor_decode_mono_3930(APEContext *ctx, int count)
1106 {
1107 APEPredictor *p = &ctx->predictor;
1108 int32_t *decoded0 = ctx->decoded[0];
1109
1110 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1111
1112 while (count--) {
1113 *decoded0 = predictor_update_3930(p, *decoded0, 0, YDELAYA);
1114 decoded0++;
1115
1116 p->buf++;
1117
1118 /* Have we filled the history buffer? */
1119 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1120 memmove(p->historybuffer, p->buf,
1121 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1122 p->buf = p->historybuffer;
1123 }
1124 }
1125 }
1126
predictor_update_filter(APEPredictor * p,const int decoded,const int filter,const int delayA,const int delayB,const int adaptA,const int adaptB)1127 static av_always_inline int predictor_update_filter(APEPredictor *p,
1128 const int decoded, const int filter,
1129 const int delayA, const int delayB,
1130 const int adaptA, const int adaptB)
1131 {
1132 int32_t predictionA, predictionB, sign;
1133
1134 p->buf[delayA] = p->lastA[filter];
1135 p->buf[adaptA] = APESIGN(p->buf[delayA]);
1136 p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
1137 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
1138
1139 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] +
1140 p->buf[delayA - 1] * p->coeffsA[filter][1] +
1141 p->buf[delayA - 2] * p->coeffsA[filter][2] +
1142 p->buf[delayA - 3] * p->coeffsA[filter][3];
1143
1144 /* Apply a scaled first-order filter compression */
1145 p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
1146 p->buf[adaptB] = APESIGN(p->buf[delayB]);
1147 p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
1148 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
1149 p->filterB[filter] = p->filterA[filter ^ 1];
1150
1151 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] +
1152 p->buf[delayB - 1] * p->coeffsB[filter][1] +
1153 p->buf[delayB - 2] * p->coeffsB[filter][2] +
1154 p->buf[delayB - 3] * p->coeffsB[filter][3] +
1155 p->buf[delayB - 4] * p->coeffsB[filter][4];
1156
1157 p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
1158 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1159
1160 sign = APESIGN(decoded);
1161 p->coeffsA[filter][0] += p->buf[adaptA ] * sign;
1162 p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
1163 p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
1164 p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
1165 p->coeffsB[filter][0] += p->buf[adaptB ] * sign;
1166 p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
1167 p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
1168 p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
1169 p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
1170
1171 return p->filterA[filter];
1172 }
1173
predictor_decode_stereo_3950(APEContext * ctx,int count)1174 static void predictor_decode_stereo_3950(APEContext *ctx, int count)
1175 {
1176 APEPredictor *p = &ctx->predictor;
1177 int32_t *decoded0 = ctx->decoded[0];
1178 int32_t *decoded1 = ctx->decoded[1];
1179
1180 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1181
1182 while (count--) {
1183 /* Predictor Y */
1184 *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
1185 YADAPTCOEFFSA, YADAPTCOEFFSB);
1186 decoded0++;
1187 *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
1188 XADAPTCOEFFSA, XADAPTCOEFFSB);
1189 decoded1++;
1190
1191 /* Combined */
1192 p->buf++;
1193
1194 /* Have we filled the history buffer? */
1195 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1196 memmove(p->historybuffer, p->buf,
1197 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1198 p->buf = p->historybuffer;
1199 }
1200 }
1201 }
1202
predictor_decode_mono_3950(APEContext * ctx,int count)1203 static void predictor_decode_mono_3950(APEContext *ctx, int count)
1204 {
1205 APEPredictor *p = &ctx->predictor;
1206 int32_t *decoded0 = ctx->decoded[0];
1207 int32_t predictionA, currentA, A, sign;
1208
1209 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1210
1211 currentA = p->lastA[0];
1212
1213 while (count--) {
1214 A = *decoded0;
1215
1216 p->buf[YDELAYA] = currentA;
1217 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
1218
1219 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] +
1220 p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
1221 p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
1222 p->buf[YDELAYA - 3] * p->coeffsA[0][3];
1223
1224 currentA = A + (predictionA >> 10);
1225
1226 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]);
1227 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
1228
1229 sign = APESIGN(A);
1230 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign;
1231 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
1232 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
1233 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
1234
1235 p->buf++;
1236
1237 /* Have we filled the history buffer? */
1238 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1239 memmove(p->historybuffer, p->buf,
1240 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1241 p->buf = p->historybuffer;
1242 }
1243
1244 p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
1245 *(decoded0++) = p->filterA[0];
1246 }
1247
1248 p->lastA[0] = currentA;
1249 }
1250
do_init_filter(APEFilter * f,int16_t * buf,int order)1251 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
1252 {
1253 f->coeffs = buf;
1254 f->historybuffer = buf + order;
1255 f->delay = f->historybuffer + order * 2;
1256 f->adaptcoeffs = f->historybuffer + order;
1257
1258 memset(f->historybuffer, 0, (order * 2) * sizeof(*f->historybuffer));
1259 memset(f->coeffs, 0, order * sizeof(*f->coeffs));
1260 f->avg = 0;
1261 }
1262
init_filter(APEContext * ctx,APEFilter * f,int16_t * buf,int order)1263 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
1264 {
1265 do_init_filter(&f[0], buf, order);
1266 do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
1267 }
1268
do_apply_filter(APEContext * ctx,int version,APEFilter * f,int32_t * data,int count,int order,int fracbits)1269 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
1270 int32_t *data, int count, int order, int fracbits)
1271 {
1272 int res;
1273 int absres;
1274
1275 while (count--) {
1276 /* round fixedpoint scalar product */
1277 res = ctx->adsp.scalarproduct_and_madd_int16(f->coeffs,
1278 f->delay - order,
1279 f->adaptcoeffs - order,
1280 order, APESIGN(*data));
1281 res = (res + (1 << (fracbits - 1))) >> fracbits;
1282 res += *data;
1283 *data++ = res;
1284
1285 /* Update the output history */
1286 *f->delay++ = av_clip_int16(res);
1287
1288 if (version < 3980) {
1289 /* Version ??? to < 3.98 files (untested) */
1290 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
1291 f->adaptcoeffs[-4] >>= 1;
1292 f->adaptcoeffs[-8] >>= 1;
1293 } else {
1294 /* Version 3.98 and later files */
1295
1296 /* Update the adaption coefficients */
1297 absres = FFABS(res);
1298 if (absres)
1299 *f->adaptcoeffs = ((res & (-1<<31)) ^ (-1<<30)) >>
1300 (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
1301 else
1302 *f->adaptcoeffs = 0;
1303
1304 f->avg += (absres - f->avg) / 16;
1305
1306 f->adaptcoeffs[-1] >>= 1;
1307 f->adaptcoeffs[-2] >>= 1;
1308 f->adaptcoeffs[-8] >>= 1;
1309 }
1310
1311 f->adaptcoeffs++;
1312
1313 /* Have we filled the history buffer? */
1314 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
1315 memmove(f->historybuffer, f->delay - (order * 2),
1316 (order * 2) * sizeof(*f->historybuffer));
1317 f->delay = f->historybuffer + order * 2;
1318 f->adaptcoeffs = f->historybuffer + order;
1319 }
1320 }
1321 }
1322
apply_filter(APEContext * ctx,APEFilter * f,int32_t * data0,int32_t * data1,int count,int order,int fracbits)1323 static void apply_filter(APEContext *ctx, APEFilter *f,
1324 int32_t *data0, int32_t *data1,
1325 int count, int order, int fracbits)
1326 {
1327 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
1328 if (data1)
1329 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
1330 }
1331
ape_apply_filters(APEContext * ctx,int32_t * decoded0,int32_t * decoded1,int count)1332 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
1333 int32_t *decoded1, int count)
1334 {
1335 int i;
1336
1337 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1338 if (!ape_filter_orders[ctx->fset][i])
1339 break;
1340 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
1341 ape_filter_orders[ctx->fset][i],
1342 ape_filter_fracbits[ctx->fset][i]);
1343 }
1344 }
1345
init_frame_decoder(APEContext * ctx)1346 static int init_frame_decoder(APEContext *ctx)
1347 {
1348 int i, ret;
1349 if ((ret = init_entropy_decoder(ctx)) < 0)
1350 return ret;
1351 init_predictor_decoder(ctx);
1352
1353 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1354 if (!ape_filter_orders[ctx->fset][i])
1355 break;
1356 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
1357 ape_filter_orders[ctx->fset][i]);
1358 }
1359 return 0;
1360 }
1361
ape_unpack_mono(APEContext * ctx,int count)1362 static void ape_unpack_mono(APEContext *ctx, int count)
1363 {
1364 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
1365 /* We are pure silence, so we're done. */
1366 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
1367 return;
1368 }
1369
1370 ctx->entropy_decode_mono(ctx, count);
1371
1372 /* Now apply the predictor decoding */
1373 ctx->predictor_decode_mono(ctx, count);
1374
1375 /* Pseudo-stereo - just copy left channel to right channel */
1376 if (ctx->channels == 2) {
1377 memcpy(ctx->decoded[1], ctx->decoded[0], count * sizeof(*ctx->decoded[1]));
1378 }
1379 }
1380
ape_unpack_stereo(APEContext * ctx,int count)1381 static void ape_unpack_stereo(APEContext *ctx, int count)
1382 {
1383 int32_t left, right;
1384 int32_t *decoded0 = ctx->decoded[0];
1385 int32_t *decoded1 = ctx->decoded[1];
1386
1387 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
1388 /* We are pure silence, so we're done. */
1389 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
1390 return;
1391 }
1392
1393 ctx->entropy_decode_stereo(ctx, count);
1394
1395 /* Now apply the predictor decoding */
1396 ctx->predictor_decode_stereo(ctx, count);
1397
1398 /* Decorrelate and scale to output depth */
1399 while (count--) {
1400 left = *decoded1 - (*decoded0 / 2);
1401 right = left + *decoded0;
1402
1403 *(decoded0++) = left;
1404 *(decoded1++) = right;
1405 }
1406 }
1407
ape_decode_frame(AVCodecContext * avctx,void * data,int * got_frame_ptr,AVPacket * avpkt)1408 static int ape_decode_frame(AVCodecContext *avctx, void *data,
1409 int *got_frame_ptr, AVPacket *avpkt)
1410 {
1411 AVFrame *frame = data;
1412 const uint8_t *buf = avpkt->data;
1413 APEContext *s = avctx->priv_data;
1414 uint8_t *sample8;
1415 int16_t *sample16;
1416 int32_t *sample24;
1417 int i, ch, ret;
1418 int blockstodecode;
1419
1420 /* this should never be negative, but bad things will happen if it is, so
1421 check it just to make sure. */
1422 av_assert0(s->samples >= 0);
1423
1424 if(!s->samples){
1425 uint32_t nblocks, offset;
1426 int buf_size;
1427
1428 if (!avpkt->size) {
1429 *got_frame_ptr = 0;
1430 return 0;
1431 }
1432 if (avpkt->size < 8) {
1433 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1434 return AVERROR_INVALIDDATA;
1435 }
1436 buf_size = avpkt->size & ~3;
1437 if (buf_size != avpkt->size) {
1438 av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
1439 "extra bytes at the end will be skipped.\n");
1440 }
1441 if (s->fileversion < 3950) // previous versions overread two bytes
1442 buf_size += 2;
1443 av_fast_padded_malloc(&s->data, &s->data_size, buf_size);
1444 if (!s->data)
1445 return AVERROR(ENOMEM);
1446 s->bdsp.bswap_buf((uint32_t *) s->data, (const uint32_t *) buf,
1447 buf_size >> 2);
1448 memset(s->data + (buf_size & ~3), 0, buf_size & 3);
1449 s->ptr = s->data;
1450 s->data_end = s->data + buf_size;
1451
1452 nblocks = bytestream_get_be32(&s->ptr);
1453 offset = bytestream_get_be32(&s->ptr);
1454 if (s->fileversion >= 3900) {
1455 if (offset > 3) {
1456 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
1457 s->data = NULL;
1458 return AVERROR_INVALIDDATA;
1459 }
1460 if (s->data_end - s->ptr < offset) {
1461 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1462 return AVERROR_INVALIDDATA;
1463 }
1464 s->ptr += offset;
1465 } else {
1466 if ((ret = init_get_bits8(&s->gb, s->ptr, s->data_end - s->ptr)) < 0)
1467 return ret;
1468 if (s->fileversion > 3800)
1469 skip_bits_long(&s->gb, offset * 8);
1470 else
1471 skip_bits_long(&s->gb, offset);
1472 }
1473
1474 if (!nblocks || nblocks > INT_MAX) {
1475 av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %"PRIu32".\n",
1476 nblocks);
1477 return AVERROR_INVALIDDATA;
1478 }
1479 s->samples = nblocks;
1480
1481 /* Initialize the frame decoder */
1482 if (init_frame_decoder(s) < 0) {
1483 av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
1484 return AVERROR_INVALIDDATA;
1485 }
1486 }
1487
1488 if (!s->data) {
1489 *got_frame_ptr = 0;
1490 return avpkt->size;
1491 }
1492
1493 blockstodecode = FFMIN(s->blocks_per_loop, s->samples);
1494 // for old files coefficients were not interleaved,
1495 // so we need to decode all of them at once
1496 if (s->fileversion < 3930)
1497 blockstodecode = s->samples;
1498
1499 /* reallocate decoded sample buffer if needed */
1500 av_fast_malloc(&s->decoded_buffer, &s->decoded_size,
1501 2 * FFALIGN(blockstodecode, 8) * sizeof(*s->decoded_buffer));
1502 if (!s->decoded_buffer)
1503 return AVERROR(ENOMEM);
1504 memset(s->decoded_buffer, 0, s->decoded_size);
1505 s->decoded[0] = s->decoded_buffer;
1506 s->decoded[1] = s->decoded_buffer + FFALIGN(blockstodecode, 8);
1507
1508 /* get output buffer */
1509 frame->nb_samples = blockstodecode;
1510 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1511 return ret;
1512
1513 s->error=0;
1514
1515 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
1516 ape_unpack_mono(s, blockstodecode);
1517 else
1518 ape_unpack_stereo(s, blockstodecode);
1519 emms_c();
1520
1521 if (s->error) {
1522 s->samples=0;
1523 av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
1524 return AVERROR_INVALIDDATA;
1525 }
1526
1527 switch (s->bps) {
1528 case 8:
1529 for (ch = 0; ch < s->channels; ch++) {
1530 sample8 = (uint8_t *)frame->data[ch];
1531 for (i = 0; i < blockstodecode; i++)
1532 *sample8++ = (s->decoded[ch][i] + 0x80) & 0xff;
1533 }
1534 break;
1535 case 16:
1536 for (ch = 0; ch < s->channels; ch++) {
1537 sample16 = (int16_t *)frame->data[ch];
1538 for (i = 0; i < blockstodecode; i++)
1539 *sample16++ = s->decoded[ch][i];
1540 }
1541 break;
1542 case 24:
1543 for (ch = 0; ch < s->channels; ch++) {
1544 sample24 = (int32_t *)frame->data[ch];
1545 for (i = 0; i < blockstodecode; i++)
1546 *sample24++ = s->decoded[ch][i] << 8;
1547 }
1548 break;
1549 }
1550
1551 s->samples -= blockstodecode;
1552
1553 *got_frame_ptr = 1;
1554
1555 return !s->samples ? avpkt->size : 0;
1556 }
1557
ape_flush(AVCodecContext * avctx)1558 static void ape_flush(AVCodecContext *avctx)
1559 {
1560 APEContext *s = avctx->priv_data;
1561 s->samples= 0;
1562 }
1563
1564 #define OFFSET(x) offsetof(APEContext, x)
1565 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1566 static const AVOption options[] = {
1567 { "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" },
1568 { "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" },
1569 { NULL},
1570 };
1571
1572 static const AVClass ape_decoder_class = {
1573 .class_name = "APE decoder",
1574 .item_name = av_default_item_name,
1575 .option = options,
1576 .version = LIBAVUTIL_VERSION_INT,
1577 };
1578
1579 AVCodec ff_ape_decoder = {
1580 .name = "ape",
1581 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
1582 .type = AVMEDIA_TYPE_AUDIO,
1583 .id = AV_CODEC_ID_APE,
1584 .priv_data_size = sizeof(APEContext),
1585 .init = ape_decode_init,
1586 .close = ape_decode_close,
1587 .decode = ape_decode_frame,
1588 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
1589 .flush = ape_flush,
1590 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
1591 AV_SAMPLE_FMT_S16P,
1592 AV_SAMPLE_FMT_S32P,
1593 AV_SAMPLE_FMT_NONE },
1594 .priv_class = &ape_decoder_class,
1595 };
1596