2 * Monkey's Audio lossless audio decoder
3 * Copyright (c) 2007 Benjamin Zores <ben@geexbox.org>
4 * based upon libdemac from Dave Chapman.
6 * This file is part of FFmpeg.
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.
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.
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
25 #include "bytestream.h"
26 #include "libavutil/audioconvert.h"
27 #include "libavutil/avassert.h"
28 #include "libavutil/opt.h"
32 * Monkey's Audio lossless audio decoder
35 #define MAX_CHANNELS 2
36 #define MAX_BYTESPERSAMPLE 3
38 #define APE_FRAMECODE_MONO_SILENCE 1
39 #define APE_FRAMECODE_STEREO_SILENCE 3
40 #define APE_FRAMECODE_PSEUDO_STEREO 4
42 #define HISTORY_SIZE 512
43 #define PREDICTOR_ORDER 8
44 /** Total size of all predictor histories */
45 #define PREDICTOR_SIZE 50
47 #define YDELAYA (18 + PREDICTOR_ORDER*4)
48 #define YDELAYB (18 + PREDICTOR_ORDER*3)
49 #define XDELAYA (18 + PREDICTOR_ORDER*2)
50 #define XDELAYB (18 + PREDICTOR_ORDER)
52 #define YADAPTCOEFFSA 18
53 #define XADAPTCOEFFSA 14
54 #define YADAPTCOEFFSB 10
55 #define XADAPTCOEFFSB 5
58 * Possible compression levels
61 enum APECompressionLevel {
62 COMPRESSION_LEVEL_FAST = 1000,
63 COMPRESSION_LEVEL_NORMAL = 2000,
64 COMPRESSION_LEVEL_HIGH = 3000,
65 COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
66 COMPRESSION_LEVEL_INSANE = 5000
70 #define APE_FILTER_LEVELS 3
72 /** Filter orders depending on compression level */
73 static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
81 /** Filter fraction bits depending on compression level */
82 static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
91 /** Filters applied to the decoded data */
92 typedef struct APEFilter {
93 int16_t *coeffs; ///< actual coefficients used in filtering
94 int16_t *adaptcoeffs; ///< adaptive filter coefficients used for correcting of actual filter coefficients
95 int16_t *historybuffer; ///< filter memory
96 int16_t *delay; ///< filtered values
101 typedef struct APERice {
106 typedef struct APERangecoder {
107 uint32_t low; ///< low end of interval
108 uint32_t range; ///< length of interval
109 uint32_t help; ///< bytes_to_follow resp. intermediate value
110 unsigned int buffer; ///< buffer for input/output
113 /** Filter histories */
114 typedef struct APEPredictor {
122 int32_t coeffsA[2][4]; ///< adaption coefficients
123 int32_t coeffsB[2][5]; ///< adaption coefficients
124 int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
127 /** Decoder context */
128 typedef struct APEContext {
129 AVClass *class; ///< class for AVOptions
130 AVCodecContext *avctx;
134 int samples; ///< samples left to decode in current frame
137 int fileversion; ///< codec version, very important in decoding process
138 int compression_level; ///< compression levels
139 int fset; ///< which filter set to use (calculated from compression level)
140 int flags; ///< global decoder flags
142 uint32_t CRC; ///< frame CRC
143 int frameflags; ///< frame flags
144 APEPredictor predictor; ///< predictor used for final reconstruction
146 int32_t *decoded_buffer;
148 int32_t *decoded[MAX_CHANNELS]; ///< decoded data for each channel
149 int blocks_per_loop; ///< maximum number of samples to decode for each call
151 int16_t* filterbuf[APE_FILTER_LEVELS]; ///< filter memory
153 APERangecoder rc; ///< rangecoder used to decode actual values
154 APERice riceX; ///< rice code parameters for the second channel
155 APERice riceY; ///< rice code parameters for the first channel
156 APEFilter filters[APE_FILTER_LEVELS][2]; ///< filters used for reconstruction
158 uint8_t *data; ///< current frame data
159 uint8_t *data_end; ///< frame data end
160 int data_size; ///< frame data allocated size
161 const uint8_t *ptr; ///< current position in frame data
168 static av_cold int ape_decode_close(AVCodecContext *avctx)
170 APEContext *s = avctx->priv_data;
173 for (i = 0; i < APE_FILTER_LEVELS; i++)
174 av_freep(&s->filterbuf[i]);
176 av_freep(&s->decoded_buffer);
178 s->decoded_size = s->data_size = 0;
183 static av_cold int ape_decode_init(AVCodecContext *avctx)
185 APEContext *s = avctx->priv_data;
188 if (avctx->extradata_size != 6) {
189 av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
190 return AVERROR(EINVAL);
192 if (avctx->channels > 2) {
193 av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
194 return AVERROR(EINVAL);
196 s->bps = avctx->bits_per_coded_sample;
199 avctx->sample_fmt = AV_SAMPLE_FMT_U8;
202 avctx->sample_fmt = AV_SAMPLE_FMT_S16;
205 avctx->sample_fmt = AV_SAMPLE_FMT_S32;
208 av_log_ask_for_sample(avctx, "Unsupported bits per coded sample %d\n",
210 return AVERROR_PATCHWELCOME;
213 s->channels = avctx->channels;
214 s->fileversion = AV_RL16(avctx->extradata);
215 s->compression_level = AV_RL16(avctx->extradata + 2);
216 s->flags = AV_RL16(avctx->extradata + 4);
218 av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n",
219 s->compression_level, s->flags);
220 if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
221 av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
222 s->compression_level);
223 return AVERROR_INVALIDDATA;
225 s->fset = s->compression_level / 1000 - 1;
226 for (i = 0; i < APE_FILTER_LEVELS; i++) {
227 if (!ape_filter_orders[s->fset][i])
229 FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
230 (ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
234 ff_dsputil_init(&s->dsp, avctx);
235 avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
237 avcodec_get_frame_defaults(&s->frame);
238 avctx->coded_frame = &s->frame;
242 ape_decode_close(avctx);
243 return AVERROR(ENOMEM);
247 * @name APE range decoding functions
252 #define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1))
253 #define SHIFT_BITS (CODE_BITS - 9)
254 #define EXTRA_BITS ((CODE_BITS-2) % 8 + 1)
255 #define BOTTOM_VALUE (TOP_VALUE >> 8)
257 /** Start the decoder */
258 static inline void range_start_decoding(APEContext *ctx)
260 ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
261 ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS);
262 ctx->rc.range = (uint32_t) 1 << EXTRA_BITS;
265 /** Perform normalization */
266 static inline void range_dec_normalize(APEContext *ctx)
268 while (ctx->rc.range <= BOTTOM_VALUE) {
269 ctx->rc.buffer <<= 8;
270 if(ctx->ptr < ctx->data_end) {
271 ctx->rc.buffer += *ctx->ptr;
276 ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF);
282 * Calculate culmulative frequency for next symbol. Does NO update!
283 * @param ctx decoder context
284 * @param tot_f is the total frequency or (code_value)1<<shift
285 * @return the culmulative frequency
287 static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
289 range_dec_normalize(ctx);
290 ctx->rc.help = ctx->rc.range / tot_f;
291 return ctx->rc.low / ctx->rc.help;
295 * Decode value with given size in bits
296 * @param ctx decoder context
297 * @param shift number of bits to decode
299 static inline int range_decode_culshift(APEContext *ctx, int shift)
301 range_dec_normalize(ctx);
302 ctx->rc.help = ctx->rc.range >> shift;
303 return ctx->rc.low / ctx->rc.help;
308 * Update decoding state
309 * @param ctx decoder context
310 * @param sy_f the interval length (frequency of the symbol)
311 * @param lt_f the lower end (frequency sum of < symbols)
313 static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
315 ctx->rc.low -= ctx->rc.help * lt_f;
316 ctx->rc.range = ctx->rc.help * sy_f;
319 /** Decode n bits (n <= 16) without modelling */
320 static inline int range_decode_bits(APEContext *ctx, int n)
322 int sym = range_decode_culshift(ctx, n);
323 range_decode_update(ctx, 1, sym);
328 #define MODEL_ELEMENTS 64
331 * Fixed probabilities for symbols in Monkey Audio version 3.97
333 static const uint16_t counts_3970[22] = {
334 0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
335 62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
336 65450, 65469, 65480, 65487, 65491, 65493,
340 * Probability ranges for symbols in Monkey Audio version 3.97
342 static const uint16_t counts_diff_3970[21] = {
343 14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
344 1104, 677, 415, 248, 150, 89, 54, 31,
349 * Fixed probabilities for symbols in Monkey Audio version 3.98
351 static const uint16_t counts_3980[22] = {
352 0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
353 64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
354 65485, 65488, 65490, 65491, 65492, 65493,
358 * Probability ranges for symbols in Monkey Audio version 3.98
360 static const uint16_t counts_diff_3980[21] = {
361 19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
362 261, 119, 65, 31, 19, 10, 6, 3,
368 * @param ctx decoder context
369 * @param counts probability range start position
370 * @param counts_diff probability range widths
372 static inline int range_get_symbol(APEContext *ctx,
373 const uint16_t counts[],
374 const uint16_t counts_diff[])
378 cf = range_decode_culshift(ctx, 16);
381 symbol= cf - 65535 + 63;
382 range_decode_update(ctx, 1, cf);
387 /* figure out the symbol inefficiently; a binary search would be much better */
388 for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
390 range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
394 /** @} */ // group rangecoder
396 static inline void update_rice(APERice *rice, int x)
398 int lim = rice->k ? (1 << (rice->k + 4)) : 0;
399 rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
401 if (rice->ksum < lim)
403 else if (rice->ksum >= (1 << (rice->k + 5)))
407 static inline int ape_decode_value(APEContext *ctx, APERice *rice)
411 if (ctx->fileversion < 3990) {
414 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
416 if (overflow == (MODEL_ELEMENTS - 1)) {
417 tmpk = range_decode_bits(ctx, 5);
420 tmpk = (rice->k < 1) ? 0 : rice->k - 1;
423 x = range_decode_bits(ctx, tmpk);
425 x = range_decode_bits(ctx, 16);
426 x |= (range_decode_bits(ctx, tmpk - 16) << 16);
428 x += overflow << tmpk;
432 pivot = rice->ksum >> 5;
436 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
438 if (overflow == (MODEL_ELEMENTS - 1)) {
439 overflow = range_decode_bits(ctx, 16) << 16;
440 overflow |= range_decode_bits(ctx, 16);
443 if (pivot < 0x10000) {
444 base = range_decode_culfreq(ctx, pivot);
445 range_decode_update(ctx, 1, base);
447 int base_hi = pivot, base_lo;
450 while (base_hi & ~0xFFFF) {
454 base_hi = range_decode_culfreq(ctx, base_hi + 1);
455 range_decode_update(ctx, 1, base_hi);
456 base_lo = range_decode_culfreq(ctx, 1 << bbits);
457 range_decode_update(ctx, 1, base_lo);
459 base = (base_hi << bbits) + base_lo;
462 x = base + overflow * pivot;
465 update_rice(rice, x);
467 /* Convert to signed */
474 static void entropy_decode(APEContext *ctx, int blockstodecode, int stereo)
476 int32_t *decoded0 = ctx->decoded[0];
477 int32_t *decoded1 = ctx->decoded[1];
479 while (blockstodecode--) {
480 *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
482 *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
486 static int init_entropy_decoder(APEContext *ctx)
489 if (ctx->data_end - ctx->ptr < 6)
490 return AVERROR_INVALIDDATA;
491 ctx->CRC = bytestream_get_be32(&ctx->ptr);
493 /* Read the frame flags if they exist */
495 if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
496 ctx->CRC &= ~0x80000000;
498 if (ctx->data_end - ctx->ptr < 6)
499 return AVERROR_INVALIDDATA;
500 ctx->frameflags = bytestream_get_be32(&ctx->ptr);
503 /* Initialize the rice structs */
505 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
507 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
509 /* The first 8 bits of input are ignored. */
512 range_start_decoding(ctx);
517 static const int32_t initial_coeffs[4] = {
521 static void init_predictor_decoder(APEContext *ctx)
523 APEPredictor *p = &ctx->predictor;
525 /* Zero the history buffers */
526 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(*p->historybuffer));
527 p->buf = p->historybuffer;
529 /* Initialize and zero the coefficients */
530 memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
531 memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
532 memset(p->coeffsB, 0, sizeof(p->coeffsB));
534 p->filterA[0] = p->filterA[1] = 0;
535 p->filterB[0] = p->filterB[1] = 0;
536 p->lastA[0] = p->lastA[1] = 0;
539 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
540 static inline int APESIGN(int32_t x) {
541 return (x < 0) - (x > 0);
544 static av_always_inline int predictor_update_filter(APEPredictor *p,
545 const int decoded, const int filter,
546 const int delayA, const int delayB,
547 const int adaptA, const int adaptB)
549 int32_t predictionA, predictionB, sign;
551 p->buf[delayA] = p->lastA[filter];
552 p->buf[adaptA] = APESIGN(p->buf[delayA]);
553 p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
554 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
556 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] +
557 p->buf[delayA - 1] * p->coeffsA[filter][1] +
558 p->buf[delayA - 2] * p->coeffsA[filter][2] +
559 p->buf[delayA - 3] * p->coeffsA[filter][3];
561 /* Apply a scaled first-order filter compression */
562 p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
563 p->buf[adaptB] = APESIGN(p->buf[delayB]);
564 p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
565 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
566 p->filterB[filter] = p->filterA[filter ^ 1];
568 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] +
569 p->buf[delayB - 1] * p->coeffsB[filter][1] +
570 p->buf[delayB - 2] * p->coeffsB[filter][2] +
571 p->buf[delayB - 3] * p->coeffsB[filter][3] +
572 p->buf[delayB - 4] * p->coeffsB[filter][4];
574 p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
575 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
577 sign = APESIGN(decoded);
578 p->coeffsA[filter][0] += p->buf[adaptA ] * sign;
579 p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
580 p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
581 p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
582 p->coeffsB[filter][0] += p->buf[adaptB ] * sign;
583 p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
584 p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
585 p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
586 p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
588 return p->filterA[filter];
591 static void predictor_decode_stereo(APEContext *ctx, int count)
593 APEPredictor *p = &ctx->predictor;
594 int32_t *decoded0 = ctx->decoded[0];
595 int32_t *decoded1 = ctx->decoded[1];
599 *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
600 YADAPTCOEFFSA, YADAPTCOEFFSB);
602 *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
603 XADAPTCOEFFSA, XADAPTCOEFFSB);
609 /* Have we filled the history buffer? */
610 if (p->buf == p->historybuffer + HISTORY_SIZE) {
611 memmove(p->historybuffer, p->buf,
612 PREDICTOR_SIZE * sizeof(*p->historybuffer));
613 p->buf = p->historybuffer;
618 static void predictor_decode_mono(APEContext *ctx, int count)
620 APEPredictor *p = &ctx->predictor;
621 int32_t *decoded0 = ctx->decoded[0];
622 int32_t predictionA, currentA, A, sign;
624 currentA = p->lastA[0];
629 p->buf[YDELAYA] = currentA;
630 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
632 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] +
633 p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
634 p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
635 p->buf[YDELAYA - 3] * p->coeffsA[0][3];
637 currentA = A + (predictionA >> 10);
639 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]);
640 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
643 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign;
644 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
645 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
646 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
650 /* Have we filled the history buffer? */
651 if (p->buf == p->historybuffer + HISTORY_SIZE) {
652 memmove(p->historybuffer, p->buf,
653 PREDICTOR_SIZE * sizeof(*p->historybuffer));
654 p->buf = p->historybuffer;
657 p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
658 *(decoded0++) = p->filterA[0];
661 p->lastA[0] = currentA;
664 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
667 f->historybuffer = buf + order;
668 f->delay = f->historybuffer + order * 2;
669 f->adaptcoeffs = f->historybuffer + order;
671 memset(f->historybuffer, 0, (order * 2) * sizeof(*f->historybuffer));
672 memset(f->coeffs, 0, order * sizeof(*f->coeffs));
676 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
678 do_init_filter(&f[0], buf, order);
679 do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
682 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
683 int32_t *data, int count, int order, int fracbits)
689 /* round fixedpoint scalar product */
690 res = ctx->dsp.scalarproduct_and_madd_int16(f->coeffs, f->delay - order,
691 f->adaptcoeffs - order,
692 order, APESIGN(*data));
693 res = (res + (1 << (fracbits - 1))) >> fracbits;
697 /* Update the output history */
698 *f->delay++ = av_clip_int16(res);
700 if (version < 3980) {
701 /* Version ??? to < 3.98 files (untested) */
702 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
703 f->adaptcoeffs[-4] >>= 1;
704 f->adaptcoeffs[-8] >>= 1;
706 /* Version 3.98 and later files */
708 /* Update the adaption coefficients */
711 *f->adaptcoeffs = ((res & (-1<<31)) ^ (-1<<30)) >>
712 (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
716 f->avg += (absres - f->avg) / 16;
718 f->adaptcoeffs[-1] >>= 1;
719 f->adaptcoeffs[-2] >>= 1;
720 f->adaptcoeffs[-8] >>= 1;
725 /* Have we filled the history buffer? */
726 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
727 memmove(f->historybuffer, f->delay - (order * 2),
728 (order * 2) * sizeof(*f->historybuffer));
729 f->delay = f->historybuffer + order * 2;
730 f->adaptcoeffs = f->historybuffer + order;
735 static void apply_filter(APEContext *ctx, APEFilter *f,
736 int32_t *data0, int32_t *data1,
737 int count, int order, int fracbits)
739 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
741 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
744 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
745 int32_t *decoded1, int count)
749 for (i = 0; i < APE_FILTER_LEVELS; i++) {
750 if (!ape_filter_orders[ctx->fset][i])
752 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
753 ape_filter_orders[ctx->fset][i],
754 ape_filter_fracbits[ctx->fset][i]);
758 static int init_frame_decoder(APEContext *ctx)
761 if ((ret = init_entropy_decoder(ctx)) < 0)
763 init_predictor_decoder(ctx);
765 for (i = 0; i < APE_FILTER_LEVELS; i++) {
766 if (!ape_filter_orders[ctx->fset][i])
768 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
769 ape_filter_orders[ctx->fset][i]);
774 static void ape_unpack_mono(APEContext *ctx, int count)
776 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
777 /* We are pure silence, so we're done. */
778 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
782 entropy_decode(ctx, count, 0);
783 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
785 /* Now apply the predictor decoding */
786 predictor_decode_mono(ctx, count);
788 /* Pseudo-stereo - just copy left channel to right channel */
789 if (ctx->channels == 2) {
790 memcpy(ctx->decoded[1], ctx->decoded[0], count * sizeof(*ctx->decoded[1]));
794 static void ape_unpack_stereo(APEContext *ctx, int count)
797 int32_t *decoded0 = ctx->decoded[0];
798 int32_t *decoded1 = ctx->decoded[1];
800 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
801 /* We are pure silence, so we're done. */
802 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
806 entropy_decode(ctx, count, 1);
807 ape_apply_filters(ctx, decoded0, decoded1, count);
809 /* Now apply the predictor decoding */
810 predictor_decode_stereo(ctx, count);
812 /* Decorrelate and scale to output depth */
814 left = *decoded1 - (*decoded0 / 2);
815 right = left + *decoded0;
817 *(decoded0++) = left;
818 *(decoded1++) = right;
822 static int ape_decode_frame(AVCodecContext *avctx, void *data,
823 int *got_frame_ptr, AVPacket *avpkt)
825 const uint8_t *buf = avpkt->data;
826 APEContext *s = avctx->priv_data;
834 /* this should never be negative, but bad things will happen if it is, so
835 check it just to make sure. */
836 av_assert0(s->samples >= 0);
839 uint32_t nblocks, offset;
846 if (avpkt->size < 8) {
847 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
848 return AVERROR_INVALIDDATA;
850 buf_size = avpkt->size & ~3;
851 if (buf_size != avpkt->size) {
852 av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
853 "extra bytes at the end will be skipped.\n");
856 av_fast_malloc(&s->data, &s->data_size, buf_size);
858 return AVERROR(ENOMEM);
859 s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
861 s->data_end = s->data + buf_size;
863 nblocks = bytestream_get_be32(&s->ptr);
864 offset = bytestream_get_be32(&s->ptr);
866 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
868 return AVERROR_INVALIDDATA;
870 if (s->data_end - s->ptr < offset) {
871 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
872 return AVERROR_INVALIDDATA;
876 if (!nblocks || nblocks > INT_MAX) {
877 av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %u.\n", nblocks);
878 return AVERROR_INVALIDDATA;
880 s->samples = nblocks;
882 /* Initialize the frame decoder */
883 if (init_frame_decoder(s) < 0) {
884 av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
885 return AVERROR_INVALIDDATA;
888 bytes_used = avpkt->size;
896 blockstodecode = FFMIN(s->blocks_per_loop, s->samples);
898 /* reallocate decoded sample buffer if needed */
899 av_fast_malloc(&s->decoded_buffer, &s->decoded_size,
900 2 * FFALIGN(blockstodecode, 8) * sizeof(*s->decoded_buffer));
901 if (!s->decoded_buffer)
902 return AVERROR(ENOMEM);
903 memset(s->decoded_buffer, 0, s->decoded_size);
904 s->decoded[0] = s->decoded_buffer;
905 s->decoded[1] = s->decoded_buffer + FFALIGN(blockstodecode, 8);
907 /* get output buffer */
908 s->frame.nb_samples = blockstodecode;
909 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) {
910 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
916 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
917 ape_unpack_mono(s, blockstodecode);
919 ape_unpack_stereo(s, blockstodecode);
924 av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
925 return AVERROR_INVALIDDATA;
930 sample8 = (uint8_t *)s->frame.data[0];
931 for (i = 0; i < blockstodecode; i++) {
932 *sample8++ = (s->decoded[0][i] + 0x80) & 0xff;
933 if (s->channels == 2)
934 *sample8++ = (s->decoded[1][i] + 0x80) & 0xff;
938 sample16 = (int16_t *)s->frame.data[0];
939 for (i = 0; i < blockstodecode; i++) {
940 *sample16++ = s->decoded[0][i];
941 if (s->channels == 2)
942 *sample16++ = s->decoded[1][i];
946 sample24 = (int32_t *)s->frame.data[0];
947 for (i = 0; i < blockstodecode; i++) {
948 *sample24++ = s->decoded[0][i] << 8;
949 if (s->channels == 2)
950 *sample24++ = s->decoded[1][i] << 8;
955 s->samples -= blockstodecode;
958 *(AVFrame *)data = s->frame;
963 static void ape_flush(AVCodecContext *avctx)
965 APEContext *s = avctx->priv_data;
969 #define OFFSET(x) offsetof(APEContext, x)
970 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
971 static const AVOption options[] = {
972 { "max_samples", "maximum number of samples decoded per call", OFFSET(blocks_per_loop), AV_OPT_TYPE_INT, { 4608 }, 1, INT_MAX, PAR, "max_samples" },
973 { "all", "no maximum. decode all samples for each packet at once", 0, AV_OPT_TYPE_CONST, { INT_MAX }, INT_MIN, INT_MAX, PAR, "max_samples" },
977 static const AVClass ape_decoder_class = {
978 .class_name = "APE decoder",
979 .item_name = av_default_item_name,
981 .version = LIBAVUTIL_VERSION_INT,
984 AVCodec ff_ape_decoder = {
986 .type = AVMEDIA_TYPE_AUDIO,
988 .priv_data_size = sizeof(APEContext),
989 .init = ape_decode_init,
990 .close = ape_decode_close,
991 .decode = ape_decode_frame,
992 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
994 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
995 .priv_class = &ape_decoder_class,