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 "libavutil/avassert.h"
26 #include "libavutil/channel_layout.h"
27 #include "libavutil/opt.h"
28 #include "lossless_audiodsp.h"
31 #include "bytestream.h"
38 * Monkey's Audio lossless audio decoder
41 #define MAX_CHANNELS 2
42 #define MAX_BYTESPERSAMPLE 3
44 #define APE_FRAMECODE_MONO_SILENCE 1
45 #define APE_FRAMECODE_STEREO_SILENCE 3
46 #define APE_FRAMECODE_PSEUDO_STEREO 4
48 #define HISTORY_SIZE 512
49 #define PREDICTOR_ORDER 8
50 /** Total size of all predictor histories */
51 #define PREDICTOR_SIZE 50
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)
58 #define YADAPTCOEFFSA 18
59 #define XADAPTCOEFFSA 14
60 #define YADAPTCOEFFSB 10
61 #define XADAPTCOEFFSB 5
64 * Possible compression levels
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
76 #define APE_FILTER_LEVELS 3
78 /** Filter orders depending on compression level */
79 static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
87 /** Filter fraction bits depending on compression level */
88 static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
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
107 typedef struct APERice {
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
119 /** Filter histories */
120 typedef struct APEPredictor {
128 int32_t coeffsA[2][4]; ///< adaption coefficients
129 int32_t coeffsB[2][5]; ///< adaption coefficients
130 int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
132 unsigned int sample_pos;
135 /** Decoder context */
136 typedef struct APEContext {
137 AVClass *class; ///< class for AVOptions
138 AVCodecContext *avctx;
139 BswapDSPContext bdsp;
140 LLAudDSPContext adsp;
142 int samples; ///< samples left to decode in current frame
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
150 uint32_t CRC; ///< frame CRC
151 int frameflags; ///< frame flags
152 APEPredictor predictor; ///< predictor used for final reconstruction
154 int32_t *decoded_buffer;
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
159 int16_t* filterbuf[APE_FILTER_LEVELS]; ///< filter memory
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
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
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);
180 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
181 int32_t *decoded1, int count);
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);
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);
200 static av_cold int ape_decode_close(AVCodecContext *avctx)
202 APEContext *s = avctx->priv_data;
205 for (i = 0; i < APE_FILTER_LEVELS; i++)
206 av_freep(&s->filterbuf[i]);
208 av_freep(&s->decoded_buffer);
210 s->decoded_size = s->data_size = 0;
215 static av_cold int ape_decode_init(AVCodecContext *avctx)
217 APEContext *s = avctx->priv_data;
220 if (avctx->extradata_size != 6) {
221 av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
222 return AVERROR(EINVAL);
224 if (avctx->channels > 2) {
225 av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
226 return AVERROR(EINVAL);
228 s->bps = avctx->bits_per_coded_sample;
231 avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
234 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
237 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
240 avpriv_request_sample(avctx,
241 "%d bits per coded sample", s->bps);
242 return AVERROR_PATCHWELCOME;
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);
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;
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])
263 FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
264 (ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
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;
281 s->entropy_decode_mono = entropy_decode_mono_3990;
282 s->entropy_decode_stereo = entropy_decode_stereo_3990;
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;
292 s->predictor_decode_mono = predictor_decode_mono_3950;
293 s->predictor_decode_stereo = predictor_decode_stereo_3950;
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;
302 ape_decode_close(avctx);
303 return AVERROR(ENOMEM);
307 * @name APE range decoding functions
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)
317 /** Start the decoder */
318 static inline void range_start_decoding(APEContext *ctx)
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;
325 /** Perform normalization */
326 static inline void range_dec_normalize(APEContext *ctx)
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;
336 ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF);
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
347 static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
349 range_dec_normalize(ctx);
350 ctx->rc.help = ctx->rc.range / tot_f;
351 return ctx->rc.low / ctx->rc.help;
355 * Decode value with given size in bits
356 * @param ctx decoder context
357 * @param shift number of bits to decode
359 static inline int range_decode_culshift(APEContext *ctx, int shift)
361 range_dec_normalize(ctx);
362 ctx->rc.help = ctx->rc.range >> shift;
363 return ctx->rc.low / ctx->rc.help;
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)
373 static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
375 ctx->rc.low -= ctx->rc.help * lt_f;
376 ctx->rc.range = ctx->rc.help * sy_f;
379 /** Decode n bits (n <= 16) without modelling */
380 static inline int range_decode_bits(APEContext *ctx, int n)
382 int sym = range_decode_culshift(ctx, n);
383 range_decode_update(ctx, 1, sym);
388 #define MODEL_ELEMENTS 64
391 * Fixed probabilities for symbols in Monkey Audio version 3.97
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,
400 * Probability ranges for symbols in Monkey Audio version 3.97
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,
409 * Fixed probabilities for symbols in Monkey Audio version 3.98
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,
418 * Probability ranges for symbols in Monkey Audio version 3.98
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,
428 * @param ctx decoder context
429 * @param counts probability range start position
430 * @param counts_diff probability range widths
432 static inline int range_get_symbol(APEContext *ctx,
433 const uint16_t counts[],
434 const uint16_t counts_diff[])
438 cf = range_decode_culshift(ctx, 16);
441 symbol= cf - 65535 + 63;
442 range_decode_update(ctx, 1, cf);
447 /* figure out the symbol inefficiently; a binary search would be much better */
448 for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
450 range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
454 /** @} */ // group rangecoder
456 static inline void update_rice(APERice *rice, unsigned int x)
458 int lim = rice->k ? (1 << (rice->k + 4)) : 0;
459 rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
461 if (rice->ksum < lim)
463 else if (rice->ksum >= (1 << (rice->k + 5)))
467 static inline int get_rice_ook(GetBitContext *gb, int k)
471 x = get_unary(gb, 1, get_bits_left(gb));
474 x = (x << k) | get_bits(gb, k);
479 static inline int ape_decode_value_3860(APEContext *ctx, GetBitContext *gb,
482 unsigned int x, overflow;
484 overflow = get_unary(gb, 1, get_bits_left(gb));
486 if (ctx->fileversion > 3880) {
487 while (overflow >= 16) {
495 else if(rice->k <= MIN_CACHE_BITS) {
496 x = (overflow << rice->k) + get_bits(gb, rice->k);
498 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", rice->k);
499 return AVERROR_INVALIDDATA;
501 rice->ksum += x - (rice->ksum + 8 >> 4);
502 if (rice->ksum < (rice->k ? 1 << (rice->k + 4) : 0))
504 else if (rice->ksum >= (1 << (rice->k + 5)) && rice->k < 24)
507 /* Convert to signed */
508 return ((x >> 1) ^ ((x & 1) - 1)) + 1;
511 static inline int ape_decode_value_3900(APEContext *ctx, APERice *rice)
513 unsigned int x, overflow;
516 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
518 if (overflow == (MODEL_ELEMENTS - 1)) {
519 tmpk = range_decode_bits(ctx, 5);
522 tmpk = (rice->k < 1) ? 0 : rice->k - 1;
524 if (tmpk <= 16 || ctx->fileversion < 3910) {
526 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
527 return AVERROR_INVALIDDATA;
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);
534 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
535 return AVERROR_INVALIDDATA;
537 x += overflow << tmpk;
539 update_rice(rice, x);
541 /* Convert to signed */
542 return ((x >> 1) ^ ((x & 1) - 1)) + 1;
545 static inline int ape_decode_value_3990(APEContext *ctx, APERice *rice)
547 unsigned int x, overflow;
550 pivot = rice->ksum >> 5;
554 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
556 if (overflow == (MODEL_ELEMENTS - 1)) {
557 overflow = range_decode_bits(ctx, 16) << 16;
558 overflow |= range_decode_bits(ctx, 16);
561 if (pivot < 0x10000) {
562 base = range_decode_culfreq(ctx, pivot);
563 range_decode_update(ctx, 1, base);
565 int base_hi = pivot, base_lo;
568 while (base_hi & ~0xFFFF) {
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);
577 base = (base_hi << bbits) + base_lo;
580 x = base + overflow * pivot;
582 update_rice(rice, x);
584 /* Convert to signed */
585 return ((x >> 1) ^ ((x & 1) - 1)) + 1;
588 static void decode_array_0000(APEContext *ctx, GetBitContext *gb,
589 int32_t *out, APERice *rice, int blockstodecode)
592 int ksummax, ksummin;
595 for (i = 0; i < FFMIN(blockstodecode, 5); i++) {
596 out[i] = get_rice_ook(&ctx->gb, 10);
597 rice->ksum += out[i];
599 rice->k = av_log2(rice->ksum / 10) + 1;
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;
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) {
616 ksummin = rice->k ? ksummin >> 1 : 0;
619 while (rice->ksum >= ksummax) {
624 ksummin = ksummin ? ksummin << 1 : 128;
628 for (i = 0; i < blockstodecode; i++)
629 out[i] = ((out[i] >> 1) ^ ((out[i] & 1) - 1)) + 1;
632 static void entropy_decode_mono_0000(APEContext *ctx, int blockstodecode)
634 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
638 static void entropy_decode_stereo_0000(APEContext *ctx, int blockstodecode)
640 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
642 decode_array_0000(ctx, &ctx->gb, ctx->decoded[1], &ctx->riceX,
646 static void entropy_decode_mono_3860(APEContext *ctx, int blockstodecode)
648 int32_t *decoded0 = ctx->decoded[0];
650 while (blockstodecode--)
651 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
654 static void entropy_decode_stereo_3860(APEContext *ctx, int blockstodecode)
656 int32_t *decoded0 = ctx->decoded[0];
657 int32_t *decoded1 = ctx->decoded[1];
658 int blocks = blockstodecode;
660 while (blockstodecode--)
661 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
663 *decoded1++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceX);
666 static void entropy_decode_mono_3900(APEContext *ctx, int blockstodecode)
668 int32_t *decoded0 = ctx->decoded[0];
670 while (blockstodecode--)
671 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
674 static void entropy_decode_stereo_3900(APEContext *ctx, int blockstodecode)
676 int32_t *decoded0 = ctx->decoded[0];
677 int32_t *decoded1 = ctx->decoded[1];
678 int blocks = blockstodecode;
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
685 range_start_decoding(ctx);
687 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
690 static void entropy_decode_stereo_3930(APEContext *ctx, int blockstodecode)
692 int32_t *decoded0 = ctx->decoded[0];
693 int32_t *decoded1 = ctx->decoded[1];
695 while (blockstodecode--) {
696 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
697 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
701 static void entropy_decode_mono_3990(APEContext *ctx, int blockstodecode)
703 int32_t *decoded0 = ctx->decoded[0];
705 while (blockstodecode--)
706 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
709 static void entropy_decode_stereo_3990(APEContext *ctx, int blockstodecode)
711 int32_t *decoded0 = ctx->decoded[0];
712 int32_t *decoded1 = ctx->decoded[1];
714 while (blockstodecode--) {
715 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
716 *decoded1++ = ape_decode_value_3990(ctx, &ctx->riceX);
720 static int init_entropy_decoder(APEContext *ctx)
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);
728 ctx->CRC = get_bits_long(&ctx->gb, 32);
731 /* Read the frame flags if they exist */
733 if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
734 ctx->CRC &= ~0x80000000;
736 if (ctx->data_end - ctx->ptr < 6)
737 return AVERROR_INVALIDDATA;
738 ctx->frameflags = bytestream_get_be32(&ctx->ptr);
741 /* Initialize the rice structs */
743 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
745 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
747 if (ctx->fileversion >= 3900) {
748 /* The first 8 bits of input are ignored. */
751 range_start_decoding(ctx);
757 static const int32_t initial_coeffs_fast_3320[1] = {
761 static const int32_t initial_coeffs_a_3800[3] = {
765 static const int32_t initial_coeffs_b_3800[2] = {
769 static const int32_t initial_coeffs_3930[4] = {
773 static void init_predictor_decoder(APEContext *ctx)
775 APEPredictor *p = &ctx->predictor;
777 /* Zero the history buffers */
778 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(*p->historybuffer));
779 p->buf = p->historybuffer;
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));
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));
795 memcpy(p->coeffsA[0], initial_coeffs_3930, sizeof(initial_coeffs_3930));
796 memcpy(p->coeffsA[1], initial_coeffs_3930, sizeof(initial_coeffs_3930));
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));
806 p->filterA[0] = p->filterA[1] = 0;
807 p->filterB[0] = p->filterB[1] = 0;
808 p->lastA[0] = p->lastA[1] = 0;
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);
818 static av_always_inline int filter_fast_3320(APEPredictor *p,
819 const int decoded, const int filter,
824 p->buf[delayA] = p->lastA[filter];
825 if (p->sample_pos < 3) {
826 p->lastA[filter] = decoded;
827 p->filterA[filter] = decoded;
831 predictionA = p->buf[delayA] * 2 - p->buf[delayA - 1];
832 p->lastA[filter] = decoded + (predictionA * p->coeffsA[filter][0] >> 9);
834 if ((decoded ^ predictionA) > 0)
835 p->coeffsA[filter][0]++;
837 p->coeffsA[filter][0]--;
839 p->filterA[filter] += p->lastA[filter];
841 return p->filterA[filter];
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)
849 int32_t predictionA, predictionB, sign;
850 int32_t d0, d1, d2, d3, d4;
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;
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];
867 predictionA = d0 * p->coeffsA[filter][0] +
868 d1 * p->coeffsA[filter][1] +
869 d2 * p->coeffsA[filter][2];
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;
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;
883 p->filterB[filter] = p->lastA[filter] + (predictionB >> shift);
884 p->filterA[filter] = p->filterB[filter] + ((p->filterA[filter] * 31) >> 5);
886 return p->filterA[filter];
889 static void long_filter_high_3800(int32_t *buffer, int order, int shift, int length)
892 int32_t dotprod, sign;
893 int32_t coeffs[256], delay[256];
895 memset(coeffs, 0, order * sizeof(*coeffs));
896 for (i = 0; i < order; i++)
897 delay[i] = buffer[i];
898 for (i = order; i < length; i++) {
900 sign = APESIGN(buffer[i]);
901 for (j = 0; j < order; j++) {
902 dotprod += delay[j] * coeffs[j];
903 coeffs[j] += ((delay[j] >> 31) | 1) * sign;
905 buffer[i] -= dotprod >> shift;
906 for (j = 0; j < order - 1; j++)
907 delay[j] = delay[j + 1];
908 delay[order - 1] = buffer[i];
912 static void long_filter_ehigh_3830(int32_t *buffer, int length)
915 int32_t dotprod, sign;
916 int32_t coeffs[8] = { 0 }, delay[8] = { 0 };
918 for (i = 0; i < length; i++) {
920 sign = APESIGN(buffer[i]);
921 for (j = 7; j >= 0; j--) {
922 dotprod += delay[j] * coeffs[j];
923 coeffs[j] += ((delay[j] >> 31) | 1) * sign;
925 for (j = 7; j > 0; j--)
926 delay[j] = delay[j - 1];
927 delay[0] = buffer[i];
928 buffer[i] -= dotprod >> 9;
932 static void predictor_decode_stereo_3800(APEContext *ctx, int count)
934 APEPredictor *p = &ctx->predictor;
935 int32_t *decoded0 = ctx->decoded[0];
936 int32_t *decoded1 = ctx->decoded[1];
937 int start = 4, shift = 10;
939 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
941 long_filter_high_3800(decoded0, 16, 9, count);
942 long_filter_high_3800(decoded1, 16, 9, count);
943 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
944 int order = 128, shift2 = 11;
946 if (ctx->fileversion >= 3830) {
950 long_filter_ehigh_3830(decoded0 + order, count - order);
951 long_filter_ehigh_3830(decoded1 + order, count - order);
954 long_filter_high_3800(decoded0, order, shift2, count);
955 long_filter_high_3800(decoded1, order, shift2, count);
959 int X = *decoded0, Y = *decoded1;
960 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
961 *decoded0 = filter_fast_3320(p, Y, 0, YDELAYA);
963 *decoded1 = filter_fast_3320(p, X, 1, XDELAYA);
966 *decoded0 = filter_3800(p, Y, 0, YDELAYA, YDELAYB,
969 *decoded1 = filter_3800(p, X, 1, XDELAYA, XDELAYB,
978 /* Have we filled the history buffer? */
979 if (p->buf == p->historybuffer + HISTORY_SIZE) {
980 memmove(p->historybuffer, p->buf,
981 PREDICTOR_SIZE * sizeof(*p->historybuffer));
982 p->buf = p->historybuffer;
987 static void predictor_decode_mono_3800(APEContext *ctx, int count)
989 APEPredictor *p = &ctx->predictor;
990 int32_t *decoded0 = ctx->decoded[0];
991 int start = 4, shift = 10;
993 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
995 long_filter_high_3800(decoded0, 16, 9, count);
996 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
997 int order = 128, shift2 = 11;
999 if (ctx->fileversion >= 3830) {
1003 long_filter_ehigh_3830(decoded0 + order, count - order);
1006 long_filter_high_3800(decoded0, order, shift2, count);
1010 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
1011 *decoded0 = filter_fast_3320(p, *decoded0, 0, YDELAYA);
1014 *decoded0 = filter_3800(p, *decoded0, 0, YDELAYA, YDELAYB,
1023 /* Have we filled the history buffer? */
1024 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1025 memmove(p->historybuffer, p->buf,
1026 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1027 p->buf = p->historybuffer;
1032 static av_always_inline int predictor_update_3930(APEPredictor *p,
1033 const int decoded, const int filter,
1036 int32_t predictionA, sign;
1037 int32_t d0, d1, d2, d3;
1039 p->buf[delayA] = p->lastA[filter];
1040 d0 = p->buf[delayA ];
1041 d1 = p->buf[delayA ] - p->buf[delayA - 1];
1042 d2 = p->buf[delayA - 1] - p->buf[delayA - 2];
1043 d3 = p->buf[delayA - 2] - p->buf[delayA - 3];
1045 predictionA = d0 * p->coeffsA[filter][0] +
1046 d1 * p->coeffsA[filter][1] +
1047 d2 * p->coeffsA[filter][2] +
1048 d3 * p->coeffsA[filter][3];
1050 p->lastA[filter] = decoded + (predictionA >> 9);
1051 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1053 sign = APESIGN(decoded);
1054 p->coeffsA[filter][0] += ((d0 < 0) * 2 - 1) * sign;
1055 p->coeffsA[filter][1] += ((d1 < 0) * 2 - 1) * sign;
1056 p->coeffsA[filter][2] += ((d2 < 0) * 2 - 1) * sign;
1057 p->coeffsA[filter][3] += ((d3 < 0) * 2 - 1) * sign;
1059 return p->filterA[filter];
1062 static void predictor_decode_stereo_3930(APEContext *ctx, int count)
1064 APEPredictor *p = &ctx->predictor;
1065 int32_t *decoded0 = ctx->decoded[0];
1066 int32_t *decoded1 = ctx->decoded[1];
1068 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1072 int Y = *decoded1, X = *decoded0;
1073 *decoded0 = predictor_update_3930(p, Y, 0, YDELAYA);
1075 *decoded1 = predictor_update_3930(p, X, 1, XDELAYA);
1081 /* Have we filled the history buffer? */
1082 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1083 memmove(p->historybuffer, p->buf,
1084 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1085 p->buf = p->historybuffer;
1090 static void predictor_decode_mono_3930(APEContext *ctx, int count)
1092 APEPredictor *p = &ctx->predictor;
1093 int32_t *decoded0 = ctx->decoded[0];
1095 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1098 *decoded0 = predictor_update_3930(p, *decoded0, 0, YDELAYA);
1103 /* Have we filled the history buffer? */
1104 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1105 memmove(p->historybuffer, p->buf,
1106 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1107 p->buf = p->historybuffer;
1112 static av_always_inline int predictor_update_filter(APEPredictor *p,
1113 const int decoded, const int filter,
1114 const int delayA, const int delayB,
1115 const int adaptA, const int adaptB)
1117 int32_t predictionA, predictionB, sign;
1119 p->buf[delayA] = p->lastA[filter];
1120 p->buf[adaptA] = APESIGN(p->buf[delayA]);
1121 p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
1122 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
1124 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] +
1125 p->buf[delayA - 1] * p->coeffsA[filter][1] +
1126 p->buf[delayA - 2] * p->coeffsA[filter][2] +
1127 p->buf[delayA - 3] * p->coeffsA[filter][3];
1129 /* Apply a scaled first-order filter compression */
1130 p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
1131 p->buf[adaptB] = APESIGN(p->buf[delayB]);
1132 p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
1133 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
1134 p->filterB[filter] = p->filterA[filter ^ 1];
1136 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] +
1137 p->buf[delayB - 1] * p->coeffsB[filter][1] +
1138 p->buf[delayB - 2] * p->coeffsB[filter][2] +
1139 p->buf[delayB - 3] * p->coeffsB[filter][3] +
1140 p->buf[delayB - 4] * p->coeffsB[filter][4];
1142 p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
1143 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1145 sign = APESIGN(decoded);
1146 p->coeffsA[filter][0] += p->buf[adaptA ] * sign;
1147 p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
1148 p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
1149 p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
1150 p->coeffsB[filter][0] += p->buf[adaptB ] * sign;
1151 p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
1152 p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
1153 p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
1154 p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
1156 return p->filterA[filter];
1159 static void predictor_decode_stereo_3950(APEContext *ctx, int count)
1161 APEPredictor *p = &ctx->predictor;
1162 int32_t *decoded0 = ctx->decoded[0];
1163 int32_t *decoded1 = ctx->decoded[1];
1165 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1169 *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
1170 YADAPTCOEFFSA, YADAPTCOEFFSB);
1172 *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
1173 XADAPTCOEFFSA, XADAPTCOEFFSB);
1179 /* Have we filled the history buffer? */
1180 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1181 memmove(p->historybuffer, p->buf,
1182 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1183 p->buf = p->historybuffer;
1188 static void predictor_decode_mono_3950(APEContext *ctx, int count)
1190 APEPredictor *p = &ctx->predictor;
1191 int32_t *decoded0 = ctx->decoded[0];
1192 int32_t predictionA, currentA, A, sign;
1194 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1196 currentA = p->lastA[0];
1201 p->buf[YDELAYA] = currentA;
1202 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
1204 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] +
1205 p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
1206 p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
1207 p->buf[YDELAYA - 3] * p->coeffsA[0][3];
1209 currentA = A + (predictionA >> 10);
1211 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]);
1212 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
1215 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign;
1216 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
1217 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
1218 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
1222 /* Have we filled the history buffer? */
1223 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1224 memmove(p->historybuffer, p->buf,
1225 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1226 p->buf = p->historybuffer;
1229 p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
1230 *(decoded0++) = p->filterA[0];
1233 p->lastA[0] = currentA;
1236 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
1239 f->historybuffer = buf + order;
1240 f->delay = f->historybuffer + order * 2;
1241 f->adaptcoeffs = f->historybuffer + order;
1243 memset(f->historybuffer, 0, (order * 2) * sizeof(*f->historybuffer));
1244 memset(f->coeffs, 0, order * sizeof(*f->coeffs));
1248 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
1250 do_init_filter(&f[0], buf, order);
1251 do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
1254 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
1255 int32_t *data, int count, int order, int fracbits)
1261 /* round fixedpoint scalar product */
1262 res = ctx->adsp.scalarproduct_and_madd_int16(f->coeffs,
1264 f->adaptcoeffs - order,
1265 order, APESIGN(*data));
1266 res = (res + (1 << (fracbits - 1))) >> fracbits;
1270 /* Update the output history */
1271 *f->delay++ = av_clip_int16(res);
1273 if (version < 3980) {
1274 /* Version ??? to < 3.98 files (untested) */
1275 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
1276 f->adaptcoeffs[-4] >>= 1;
1277 f->adaptcoeffs[-8] >>= 1;
1279 /* Version 3.98 and later files */
1281 /* Update the adaption coefficients */
1282 absres = FFABS(res);
1284 *f->adaptcoeffs = ((res & INT32_MIN) ^ (-(1<<30))) >>
1285 (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
1287 *f->adaptcoeffs = 0;
1289 f->avg += (absres - f->avg) / 16;
1291 f->adaptcoeffs[-1] >>= 1;
1292 f->adaptcoeffs[-2] >>= 1;
1293 f->adaptcoeffs[-8] >>= 1;
1298 /* Have we filled the history buffer? */
1299 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
1300 memmove(f->historybuffer, f->delay - (order * 2),
1301 (order * 2) * sizeof(*f->historybuffer));
1302 f->delay = f->historybuffer + order * 2;
1303 f->adaptcoeffs = f->historybuffer + order;
1308 static void apply_filter(APEContext *ctx, APEFilter *f,
1309 int32_t *data0, int32_t *data1,
1310 int count, int order, int fracbits)
1312 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
1314 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
1317 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
1318 int32_t *decoded1, int count)
1322 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1323 if (!ape_filter_orders[ctx->fset][i])
1325 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
1326 ape_filter_orders[ctx->fset][i],
1327 ape_filter_fracbits[ctx->fset][i]);
1331 static int init_frame_decoder(APEContext *ctx)
1334 if ((ret = init_entropy_decoder(ctx)) < 0)
1336 init_predictor_decoder(ctx);
1338 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1339 if (!ape_filter_orders[ctx->fset][i])
1341 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
1342 ape_filter_orders[ctx->fset][i]);
1347 static void ape_unpack_mono(APEContext *ctx, int count)
1349 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
1350 /* We are pure silence, so we're done. */
1351 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
1355 ctx->entropy_decode_mono(ctx, count);
1357 /* Now apply the predictor decoding */
1358 ctx->predictor_decode_mono(ctx, count);
1360 /* Pseudo-stereo - just copy left channel to right channel */
1361 if (ctx->channels == 2) {
1362 memcpy(ctx->decoded[1], ctx->decoded[0], count * sizeof(*ctx->decoded[1]));
1366 static void ape_unpack_stereo(APEContext *ctx, int count)
1368 int32_t left, right;
1369 int32_t *decoded0 = ctx->decoded[0];
1370 int32_t *decoded1 = ctx->decoded[1];
1372 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
1373 /* We are pure silence, so we're done. */
1374 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
1378 ctx->entropy_decode_stereo(ctx, count);
1380 /* Now apply the predictor decoding */
1381 ctx->predictor_decode_stereo(ctx, count);
1383 /* Decorrelate and scale to output depth */
1385 left = *decoded1 - (*decoded0 / 2);
1386 right = left + *decoded0;
1388 *(decoded0++) = left;
1389 *(decoded1++) = right;
1393 static int ape_decode_frame(AVCodecContext *avctx, void *data,
1394 int *got_frame_ptr, AVPacket *avpkt)
1396 AVFrame *frame = data;
1397 const uint8_t *buf = avpkt->data;
1398 APEContext *s = avctx->priv_data;
1405 /* this should never be negative, but bad things will happen if it is, so
1406 check it just to make sure. */
1407 av_assert0(s->samples >= 0);
1410 uint32_t nblocks, offset;
1417 if (avpkt->size < 8) {
1418 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1419 return AVERROR_INVALIDDATA;
1421 buf_size = avpkt->size & ~3;
1422 if (buf_size != avpkt->size) {
1423 av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
1424 "extra bytes at the end will be skipped.\n");
1426 if (s->fileversion < 3950) // previous versions overread two bytes
1428 av_fast_padded_malloc(&s->data, &s->data_size, buf_size);
1430 return AVERROR(ENOMEM);
1431 s->bdsp.bswap_buf((uint32_t *) s->data, (const uint32_t *) buf,
1433 memset(s->data + (buf_size & ~3), 0, buf_size & 3);
1435 s->data_end = s->data + buf_size;
1437 nblocks = bytestream_get_be32(&s->ptr);
1438 offset = bytestream_get_be32(&s->ptr);
1439 if (s->fileversion >= 3900) {
1441 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
1443 return AVERROR_INVALIDDATA;
1445 if (s->data_end - s->ptr < offset) {
1446 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1447 return AVERROR_INVALIDDATA;
1451 if ((ret = init_get_bits8(&s->gb, s->ptr, s->data_end - s->ptr)) < 0)
1453 if (s->fileversion > 3800)
1454 skip_bits_long(&s->gb, offset * 8);
1456 skip_bits_long(&s->gb, offset);
1459 if (!nblocks || nblocks > INT_MAX) {
1460 av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %"PRIu32".\n",
1462 return AVERROR_INVALIDDATA;
1465 /* Initialize the frame decoder */
1466 if (init_frame_decoder(s) < 0) {
1467 av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
1468 return AVERROR_INVALIDDATA;
1470 s->samples = nblocks;
1478 blockstodecode = FFMIN(s->blocks_per_loop, s->samples);
1479 // for old files coefficients were not interleaved,
1480 // so we need to decode all of them at once
1481 if (s->fileversion < 3930)
1482 blockstodecode = s->samples;
1484 /* reallocate decoded sample buffer if needed */
1485 av_fast_malloc(&s->decoded_buffer, &s->decoded_size,
1486 2 * FFALIGN(blockstodecode, 8) * sizeof(*s->decoded_buffer));
1487 if (!s->decoded_buffer)
1488 return AVERROR(ENOMEM);
1489 memset(s->decoded_buffer, 0, s->decoded_size);
1490 s->decoded[0] = s->decoded_buffer;
1491 s->decoded[1] = s->decoded_buffer + FFALIGN(blockstodecode, 8);
1493 /* get output buffer */
1494 frame->nb_samples = blockstodecode;
1495 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1500 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
1501 ape_unpack_mono(s, blockstodecode);
1503 ape_unpack_stereo(s, blockstodecode);
1508 av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
1509 return AVERROR_INVALIDDATA;
1514 for (ch = 0; ch < s->channels; ch++) {
1515 sample8 = (uint8_t *)frame->data[ch];
1516 for (i = 0; i < blockstodecode; i++)
1517 *sample8++ = (s->decoded[ch][i] + 0x80) & 0xff;
1521 for (ch = 0; ch < s->channels; ch++) {
1522 sample16 = (int16_t *)frame->data[ch];
1523 for (i = 0; i < blockstodecode; i++)
1524 *sample16++ = s->decoded[ch][i];
1528 for (ch = 0; ch < s->channels; ch++) {
1529 sample24 = (int32_t *)frame->data[ch];
1530 for (i = 0; i < blockstodecode; i++)
1531 *sample24++ = s->decoded[ch][i] << 8;
1536 s->samples -= blockstodecode;
1540 return !s->samples ? avpkt->size : 0;
1543 static void ape_flush(AVCodecContext *avctx)
1545 APEContext *s = avctx->priv_data;
1549 #define OFFSET(x) offsetof(APEContext, x)
1550 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1551 static const AVOption options[] = {
1552 { "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" },
1553 { "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" },
1557 static const AVClass ape_decoder_class = {
1558 .class_name = "APE decoder",
1559 .item_name = av_default_item_name,
1561 .version = LIBAVUTIL_VERSION_INT,
1564 AVCodec ff_ape_decoder = {
1566 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
1567 .type = AVMEDIA_TYPE_AUDIO,
1568 .id = AV_CODEC_ID_APE,
1569 .priv_data_size = sizeof(APEContext),
1570 .init = ape_decode_init,
1571 .close = ape_decode_close,
1572 .decode = ape_decode_frame,
1573 .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DELAY |
1576 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
1579 AV_SAMPLE_FMT_NONE },
1580 .priv_class = &ape_decoder_class,