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"
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;
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 int32_t scalarproduct_and_madd_int16_c(int16_t *v1, const int16_t *v2,
223 *v1++ += mul * *v3++;
228 static av_cold int ape_decode_init(AVCodecContext *avctx)
230 APEContext *s = avctx->priv_data;
233 if (avctx->extradata_size != 6) {
234 av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
235 return AVERROR(EINVAL);
237 if (avctx->channels > 2) {
238 av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
239 return AVERROR(EINVAL);
241 s->bps = avctx->bits_per_coded_sample;
244 avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
247 avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
250 avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
253 avpriv_request_sample(avctx,
254 "%d bits per coded sample", s->bps);
255 return AVERROR_PATCHWELCOME;
258 s->channels = avctx->channels;
259 s->fileversion = AV_RL16(avctx->extradata);
260 s->compression_level = AV_RL16(avctx->extradata + 2);
261 s->flags = AV_RL16(avctx->extradata + 4);
263 av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n",
264 s->compression_level, s->flags);
265 if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE ||
266 !s->compression_level ||
267 (s->fileversion < 3930 && s->compression_level == COMPRESSION_LEVEL_INSANE)) {
268 av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
269 s->compression_level);
270 return AVERROR_INVALIDDATA;
272 s->fset = s->compression_level / 1000 - 1;
273 for (i = 0; i < APE_FILTER_LEVELS; i++) {
274 if (!ape_filter_orders[s->fset][i])
276 FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
277 (ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
281 if (s->fileversion < 3860) {
282 s->entropy_decode_mono = entropy_decode_mono_0000;
283 s->entropy_decode_stereo = entropy_decode_stereo_0000;
284 } else if (s->fileversion < 3900) {
285 s->entropy_decode_mono = entropy_decode_mono_3860;
286 s->entropy_decode_stereo = entropy_decode_stereo_3860;
287 } else if (s->fileversion < 3930) {
288 s->entropy_decode_mono = entropy_decode_mono_3900;
289 s->entropy_decode_stereo = entropy_decode_stereo_3900;
290 } else if (s->fileversion < 3990) {
291 s->entropy_decode_mono = entropy_decode_mono_3900;
292 s->entropy_decode_stereo = entropy_decode_stereo_3930;
294 s->entropy_decode_mono = entropy_decode_mono_3990;
295 s->entropy_decode_stereo = entropy_decode_stereo_3990;
298 if (s->fileversion < 3930) {
299 s->predictor_decode_mono = predictor_decode_mono_3800;
300 s->predictor_decode_stereo = predictor_decode_stereo_3800;
301 } else if (s->fileversion < 3950) {
302 s->predictor_decode_mono = predictor_decode_mono_3930;
303 s->predictor_decode_stereo = predictor_decode_stereo_3930;
305 s->predictor_decode_mono = predictor_decode_mono_3950;
306 s->predictor_decode_stereo = predictor_decode_stereo_3950;
309 s->adsp.scalarproduct_and_madd_int16 = scalarproduct_and_madd_int16_c;
312 ff_apedsp_init_arm(&s->adsp);
314 ff_apedsp_init_ppc(&s->adsp);
316 ff_apedsp_init_x86(&s->adsp);
318 ff_dsputil_init(&s->dsp, avctx);
319 avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;
323 ape_decode_close(avctx);
324 return AVERROR(ENOMEM);
328 * @name APE range decoding functions
333 #define TOP_VALUE ((unsigned int)1 << (CODE_BITS-1))
334 #define SHIFT_BITS (CODE_BITS - 9)
335 #define EXTRA_BITS ((CODE_BITS-2) % 8 + 1)
336 #define BOTTOM_VALUE (TOP_VALUE >> 8)
338 /** Start the decoder */
339 static inline void range_start_decoding(APEContext *ctx)
341 ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
342 ctx->rc.low = ctx->rc.buffer >> (8 - EXTRA_BITS);
343 ctx->rc.range = (uint32_t) 1 << EXTRA_BITS;
346 /** Perform normalization */
347 static inline void range_dec_normalize(APEContext *ctx)
349 while (ctx->rc.range <= BOTTOM_VALUE) {
350 ctx->rc.buffer <<= 8;
351 if(ctx->ptr < ctx->data_end) {
352 ctx->rc.buffer += *ctx->ptr;
357 ctx->rc.low = (ctx->rc.low << 8) | ((ctx->rc.buffer >> 1) & 0xFF);
363 * Calculate culmulative frequency for next symbol. Does NO update!
364 * @param ctx decoder context
365 * @param tot_f is the total frequency or (code_value)1<<shift
366 * @return the culmulative frequency
368 static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
370 range_dec_normalize(ctx);
371 ctx->rc.help = ctx->rc.range / tot_f;
372 return ctx->rc.low / ctx->rc.help;
376 * Decode value with given size in bits
377 * @param ctx decoder context
378 * @param shift number of bits to decode
380 static inline int range_decode_culshift(APEContext *ctx, int shift)
382 range_dec_normalize(ctx);
383 ctx->rc.help = ctx->rc.range >> shift;
384 return ctx->rc.low / ctx->rc.help;
389 * Update decoding state
390 * @param ctx decoder context
391 * @param sy_f the interval length (frequency of the symbol)
392 * @param lt_f the lower end (frequency sum of < symbols)
394 static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
396 ctx->rc.low -= ctx->rc.help * lt_f;
397 ctx->rc.range = ctx->rc.help * sy_f;
400 /** Decode n bits (n <= 16) without modelling */
401 static inline int range_decode_bits(APEContext *ctx, int n)
403 int sym = range_decode_culshift(ctx, n);
404 range_decode_update(ctx, 1, sym);
409 #define MODEL_ELEMENTS 64
412 * Fixed probabilities for symbols in Monkey Audio version 3.97
414 static const uint16_t counts_3970[22] = {
415 0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
416 62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
417 65450, 65469, 65480, 65487, 65491, 65493,
421 * Probability ranges for symbols in Monkey Audio version 3.97
423 static const uint16_t counts_diff_3970[21] = {
424 14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
425 1104, 677, 415, 248, 150, 89, 54, 31,
430 * Fixed probabilities for symbols in Monkey Audio version 3.98
432 static const uint16_t counts_3980[22] = {
433 0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
434 64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
435 65485, 65488, 65490, 65491, 65492, 65493,
439 * Probability ranges for symbols in Monkey Audio version 3.98
441 static const uint16_t counts_diff_3980[21] = {
442 19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
443 261, 119, 65, 31, 19, 10, 6, 3,
449 * @param ctx decoder context
450 * @param counts probability range start position
451 * @param counts_diff probability range widths
453 static inline int range_get_symbol(APEContext *ctx,
454 const uint16_t counts[],
455 const uint16_t counts_diff[])
459 cf = range_decode_culshift(ctx, 16);
462 symbol= cf - 65535 + 63;
463 range_decode_update(ctx, 1, cf);
468 /* figure out the symbol inefficiently; a binary search would be much better */
469 for (symbol = 0; counts[symbol + 1] <= cf; symbol++);
471 range_decode_update(ctx, counts_diff[symbol], counts[symbol]);
475 /** @} */ // group rangecoder
477 static inline void update_rice(APERice *rice, unsigned int x)
479 int lim = rice->k ? (1 << (rice->k + 4)) : 0;
480 rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);
482 if (rice->ksum < lim)
484 else if (rice->ksum >= (1 << (rice->k + 5)))
488 static inline int get_rice_ook(GetBitContext *gb, int k)
492 x = get_unary(gb, 1, get_bits_left(gb));
495 x = (x << k) | get_bits(gb, k);
500 static inline int ape_decode_value_3860(APEContext *ctx, GetBitContext *gb,
503 unsigned int x, overflow;
505 overflow = get_unary(gb, 1, get_bits_left(gb));
507 if (ctx->fileversion > 3880) {
508 while (overflow >= 16) {
516 else if(rice->k <= MIN_CACHE_BITS) {
517 x = (overflow << rice->k) + get_bits(gb, rice->k);
519 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", rice->k);
520 return AVERROR_INVALIDDATA;
522 rice->ksum += x - (rice->ksum + 8 >> 4);
523 if (rice->ksum < (rice->k ? 1 << (rice->k + 4) : 0))
525 else if (rice->ksum >= (1 << (rice->k + 5)) && rice->k < 24)
528 /* Convert to signed */
535 static inline int ape_decode_value_3900(APEContext *ctx, APERice *rice)
537 unsigned int x, overflow;
540 overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);
542 if (overflow == (MODEL_ELEMENTS - 1)) {
543 tmpk = range_decode_bits(ctx, 5);
546 tmpk = (rice->k < 1) ? 0 : rice->k - 1;
548 if (tmpk <= 16 || ctx->fileversion < 3910) {
550 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
551 return AVERROR_INVALIDDATA;
553 x = range_decode_bits(ctx, tmpk);
554 } else if (tmpk <= 31) {
555 x = range_decode_bits(ctx, 16);
556 x |= (range_decode_bits(ctx, tmpk - 16) << 16);
558 av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
559 return AVERROR_INVALIDDATA;
561 x += overflow << tmpk;
563 update_rice(rice, x);
565 /* Convert to signed */
572 static inline int ape_decode_value_3990(APEContext *ctx, APERice *rice)
574 unsigned int x, overflow;
577 pivot = rice->ksum >> 5;
581 overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);
583 if (overflow == (MODEL_ELEMENTS - 1)) {
584 overflow = range_decode_bits(ctx, 16) << 16;
585 overflow |= range_decode_bits(ctx, 16);
588 if (pivot < 0x10000) {
589 base = range_decode_culfreq(ctx, pivot);
590 range_decode_update(ctx, 1, base);
592 int base_hi = pivot, base_lo;
595 while (base_hi & ~0xFFFF) {
599 base_hi = range_decode_culfreq(ctx, base_hi + 1);
600 range_decode_update(ctx, 1, base_hi);
601 base_lo = range_decode_culfreq(ctx, 1 << bbits);
602 range_decode_update(ctx, 1, base_lo);
604 base = (base_hi << bbits) + base_lo;
607 x = base + overflow * pivot;
609 update_rice(rice, x);
611 /* Convert to signed */
618 static void decode_array_0000(APEContext *ctx, GetBitContext *gb,
619 int32_t *out, APERice *rice, int blockstodecode)
622 int ksummax, ksummin;
625 for (i = 0; i < 5; i++) {
626 out[i] = get_rice_ook(&ctx->gb, 10);
627 rice->ksum += out[i];
629 rice->k = av_log2(rice->ksum / 10) + 1;
632 for (; i < 64; i++) {
633 out[i] = get_rice_ook(&ctx->gb, rice->k);
634 rice->ksum += out[i];
635 rice->k = av_log2(rice->ksum / ((i + 1) * 2)) + 1;
639 ksummax = 1 << rice->k + 7;
640 ksummin = rice->k ? (1 << rice->k + 6) : 0;
641 for (; i < blockstodecode; i++) {
642 out[i] = get_rice_ook(&ctx->gb, rice->k);
643 rice->ksum += out[i] - out[i - 64];
644 while (rice->ksum < ksummin) {
646 ksummin = rice->k ? ksummin >> 1 : 0;
649 while (rice->ksum >= ksummax) {
654 ksummin = ksummin ? ksummin << 1 : 128;
658 for (i = 0; i < blockstodecode; i++) {
660 out[i] = (out[i] >> 1) + 1;
662 out[i] = -(out[i] >> 1);
666 static void entropy_decode_mono_0000(APEContext *ctx, int blockstodecode)
668 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
672 static void entropy_decode_stereo_0000(APEContext *ctx, int blockstodecode)
674 decode_array_0000(ctx, &ctx->gb, ctx->decoded[0], &ctx->riceY,
676 decode_array_0000(ctx, &ctx->gb, ctx->decoded[1], &ctx->riceX,
680 static void entropy_decode_mono_3860(APEContext *ctx, int blockstodecode)
682 int32_t *decoded0 = ctx->decoded[0];
684 while (blockstodecode--)
685 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
688 static void entropy_decode_stereo_3860(APEContext *ctx, int blockstodecode)
690 int32_t *decoded0 = ctx->decoded[0];
691 int32_t *decoded1 = ctx->decoded[1];
692 int blocks = blockstodecode;
694 while (blockstodecode--)
695 *decoded0++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceY);
697 *decoded1++ = ape_decode_value_3860(ctx, &ctx->gb, &ctx->riceX);
700 static void entropy_decode_mono_3900(APEContext *ctx, int blockstodecode)
702 int32_t *decoded0 = ctx->decoded[0];
704 while (blockstodecode--)
705 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
708 static void entropy_decode_stereo_3900(APEContext *ctx, int blockstodecode)
710 int32_t *decoded0 = ctx->decoded[0];
711 int32_t *decoded1 = ctx->decoded[1];
712 int blocks = blockstodecode;
714 while (blockstodecode--)
715 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
716 range_dec_normalize(ctx);
717 // because of some implementation peculiarities we need to backpedal here
719 range_start_decoding(ctx);
721 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
724 static void entropy_decode_stereo_3930(APEContext *ctx, int blockstodecode)
726 int32_t *decoded0 = ctx->decoded[0];
727 int32_t *decoded1 = ctx->decoded[1];
729 while (blockstodecode--) {
730 *decoded0++ = ape_decode_value_3900(ctx, &ctx->riceY);
731 *decoded1++ = ape_decode_value_3900(ctx, &ctx->riceX);
735 static void entropy_decode_mono_3990(APEContext *ctx, int blockstodecode)
737 int32_t *decoded0 = ctx->decoded[0];
739 while (blockstodecode--)
740 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
743 static void entropy_decode_stereo_3990(APEContext *ctx, int blockstodecode)
745 int32_t *decoded0 = ctx->decoded[0];
746 int32_t *decoded1 = ctx->decoded[1];
748 while (blockstodecode--) {
749 *decoded0++ = ape_decode_value_3990(ctx, &ctx->riceY);
750 *decoded1++ = ape_decode_value_3990(ctx, &ctx->riceX);
754 static int init_entropy_decoder(APEContext *ctx)
757 if (ctx->fileversion >= 3900) {
758 if (ctx->data_end - ctx->ptr < 6)
759 return AVERROR_INVALIDDATA;
760 ctx->CRC = bytestream_get_be32(&ctx->ptr);
762 ctx->CRC = get_bits_long(&ctx->gb, 32);
765 /* Read the frame flags if they exist */
767 if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
768 ctx->CRC &= ~0x80000000;
770 if (ctx->data_end - ctx->ptr < 6)
771 return AVERROR_INVALIDDATA;
772 ctx->frameflags = bytestream_get_be32(&ctx->ptr);
775 /* Initialize the rice structs */
777 ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
779 ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;
781 if (ctx->fileversion >= 3900) {
782 /* The first 8 bits of input are ignored. */
785 range_start_decoding(ctx);
791 static const int32_t initial_coeffs_fast_3320[1] = {
795 static const int32_t initial_coeffs_a_3800[3] = {
799 static const int32_t initial_coeffs_b_3800[2] = {
803 static const int32_t initial_coeffs_3930[4] = {
807 static void init_predictor_decoder(APEContext *ctx)
809 APEPredictor *p = &ctx->predictor;
811 /* Zero the history buffers */
812 memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(*p->historybuffer));
813 p->buf = p->historybuffer;
815 /* Initialize and zero the coefficients */
816 if (ctx->fileversion < 3930) {
817 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
818 memcpy(p->coeffsA[0], initial_coeffs_fast_3320,
819 sizeof(initial_coeffs_fast_3320));
820 memcpy(p->coeffsA[1], initial_coeffs_fast_3320,
821 sizeof(initial_coeffs_fast_3320));
823 memcpy(p->coeffsA[0], initial_coeffs_a_3800,
824 sizeof(initial_coeffs_a_3800));
825 memcpy(p->coeffsA[1], initial_coeffs_a_3800,
826 sizeof(initial_coeffs_a_3800));
829 memcpy(p->coeffsA[0], initial_coeffs_3930, sizeof(initial_coeffs_3930));
830 memcpy(p->coeffsA[1], initial_coeffs_3930, sizeof(initial_coeffs_3930));
832 memset(p->coeffsB, 0, sizeof(p->coeffsB));
833 if (ctx->fileversion < 3930) {
834 memcpy(p->coeffsB[0], initial_coeffs_b_3800,
835 sizeof(initial_coeffs_b_3800));
836 memcpy(p->coeffsB[1], initial_coeffs_b_3800,
837 sizeof(initial_coeffs_b_3800));
840 p->filterA[0] = p->filterA[1] = 0;
841 p->filterB[0] = p->filterB[1] = 0;
842 p->lastA[0] = p->lastA[1] = 0;
847 /** Get inverse sign of integer (-1 for positive, 1 for negative and 0 for zero) */
848 static inline int APESIGN(int32_t x) {
849 return (x < 0) - (x > 0);
852 static av_always_inline int filter_fast_3320(APEPredictor *p,
853 const int decoded, const int filter,
858 p->buf[delayA] = p->lastA[filter];
859 if (p->sample_pos < 3) {
860 p->lastA[filter] = decoded;
861 p->filterA[filter] = decoded;
865 predictionA = p->buf[delayA] * 2 - p->buf[delayA - 1];
866 p->lastA[filter] = decoded + (predictionA * p->coeffsA[filter][0] >> 9);
868 if ((decoded ^ predictionA) > 0)
869 p->coeffsA[filter][0]++;
871 p->coeffsA[filter][0]--;
873 p->filterA[filter] += p->lastA[filter];
875 return p->filterA[filter];
878 static av_always_inline int filter_3800(APEPredictor *p,
879 const int decoded, const int filter,
880 const int delayA, const int delayB,
881 const int start, const int shift)
883 int32_t predictionA, predictionB, sign;
884 int32_t d0, d1, d2, d3, d4;
886 p->buf[delayA] = p->lastA[filter];
887 p->buf[delayB] = p->filterB[filter];
888 if (p->sample_pos < start) {
889 predictionA = decoded + p->filterA[filter];
890 p->lastA[filter] = decoded;
891 p->filterB[filter] = decoded;
892 p->filterA[filter] = predictionA;
896 d1 = (p->buf[delayA] - p->buf[delayA - 1]) << 1;
897 d0 = p->buf[delayA] + ((p->buf[delayA - 2] - p->buf[delayA - 1]) << 3);
898 d3 = p->buf[delayB] * 2 - p->buf[delayB - 1];
901 predictionA = d0 * p->coeffsA[filter][0] +
902 d1 * p->coeffsA[filter][1] +
903 d2 * p->coeffsA[filter][2];
905 sign = APESIGN(decoded);
906 p->coeffsA[filter][0] += (((d0 >> 30) & 2) - 1) * sign;
907 p->coeffsA[filter][1] += (((d1 >> 28) & 8) - 4) * sign;
908 p->coeffsA[filter][2] += (((d2 >> 28) & 8) - 4) * sign;
910 predictionB = d3 * p->coeffsB[filter][0] -
911 d4 * p->coeffsB[filter][1];
912 p->lastA[filter] = decoded + (predictionA >> 11);
913 sign = APESIGN(p->lastA[filter]);
914 p->coeffsB[filter][0] += (((d3 >> 29) & 4) - 2) * sign;
915 p->coeffsB[filter][1] -= (((d4 >> 30) & 2) - 1) * sign;
917 p->filterB[filter] = p->lastA[filter] + (predictionB >> shift);
918 p->filterA[filter] = p->filterB[filter] + ((p->filterA[filter] * 31) >> 5);
920 return p->filterA[filter];
923 static void long_filter_high_3800(int32_t *buffer, int order, int shift,
924 int32_t *coeffs, int32_t *delay, int length)
927 int32_t dotprod, sign;
929 memset(coeffs, 0, order * sizeof(*coeffs));
930 for (i = 0; i < order; i++)
931 delay[i] = buffer[i];
932 for (i = order; i < length; i++) {
934 sign = APESIGN(buffer[i]);
935 for (j = 0; j < order; j++) {
936 dotprod += delay[j] * coeffs[j];
937 coeffs[j] -= (((delay[j] >> 30) & 2) - 1) * sign;
939 buffer[i] -= dotprod >> shift;
940 for (j = 0; j < order - 1; j++)
941 delay[j] = delay[j + 1];
942 delay[order - 1] = buffer[i];
946 static void long_filter_ehigh_3830(int32_t *buffer, int length)
949 int32_t dotprod, sign;
950 int32_t coeffs[8], delay[8];
952 memset(coeffs, 0, sizeof(coeffs));
953 memset(delay, 0, sizeof(delay));
954 for (i = 0; i < length; i++) {
956 sign = APESIGN(buffer[i]);
957 for (j = 7; j >= 0; j--) {
958 dotprod += delay[j] * coeffs[j];
959 coeffs[j] -= (((delay[j] >> 30) & 2) - 1) * sign;
961 for (j = 7; j > 0; j--)
962 delay[j] = delay[j - 1];
963 delay[0] = buffer[i];
964 buffer[i] -= dotprod >> 9;
968 static void predictor_decode_stereo_3800(APEContext *ctx, int count)
970 APEPredictor *p = &ctx->predictor;
971 int32_t *decoded0 = ctx->decoded[0];
972 int32_t *decoded1 = ctx->decoded[1];
973 int32_t coeffs[256], delay[256];
974 int start = 4, shift = 10;
976 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
978 long_filter_high_3800(decoded0, 16, 9, coeffs, delay, count);
979 long_filter_high_3800(decoded1, 16, 9, coeffs, delay, count);
980 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
981 int order = 128, shift2 = 11;
983 if (ctx->fileversion >= 3830) {
987 long_filter_ehigh_3830(decoded0 + order, count - order);
988 long_filter_ehigh_3830(decoded1 + order, count - order);
991 long_filter_high_3800(decoded0, order, shift2, coeffs, delay, count);
992 long_filter_high_3800(decoded1, order, shift2, coeffs, delay, count);
996 int X = *decoded0, Y = *decoded1;
997 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
998 *decoded0 = filter_fast_3320(p, Y, 0, YDELAYA);
1000 *decoded1 = filter_fast_3320(p, X, 1, XDELAYA);
1003 *decoded0 = filter_3800(p, Y, 0, YDELAYA, YDELAYB,
1006 *decoded1 = filter_3800(p, X, 1, XDELAYA, XDELAYB,
1015 /* Have we filled the history buffer? */
1016 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1017 memmove(p->historybuffer, p->buf,
1018 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1019 p->buf = p->historybuffer;
1024 static void predictor_decode_mono_3800(APEContext *ctx, int count)
1026 APEPredictor *p = &ctx->predictor;
1027 int32_t *decoded0 = ctx->decoded[0];
1028 int32_t coeffs[256], delay[256];
1029 int start = 4, shift = 10;
1031 if (ctx->compression_level == COMPRESSION_LEVEL_HIGH) {
1033 long_filter_high_3800(decoded0, 16, 9, coeffs, delay, count);
1034 } else if (ctx->compression_level == COMPRESSION_LEVEL_EXTRA_HIGH) {
1035 int order = 128, shift2 = 11;
1037 if (ctx->fileversion >= 3830) {
1041 long_filter_ehigh_3830(decoded0 + order, count - order);
1044 long_filter_high_3800(decoded0, order, shift2, coeffs, delay, count);
1048 if (ctx->compression_level == COMPRESSION_LEVEL_FAST) {
1049 *decoded0 = filter_fast_3320(p, *decoded0, 0, YDELAYA);
1052 *decoded0 = filter_3800(p, *decoded0, 0, YDELAYA, YDELAYB,
1061 /* Have we filled the history buffer? */
1062 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1063 memmove(p->historybuffer, p->buf,
1064 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1065 p->buf = p->historybuffer;
1070 static av_always_inline int predictor_update_3930(APEPredictor *p,
1071 const int decoded, const int filter,
1074 int32_t predictionA, sign;
1075 int32_t d0, d1, d2, d3;
1077 p->buf[delayA] = p->lastA[filter];
1078 d0 = p->buf[delayA ];
1079 d1 = p->buf[delayA ] - p->buf[delayA - 1];
1080 d2 = p->buf[delayA - 1] - p->buf[delayA - 2];
1081 d3 = p->buf[delayA - 2] - p->buf[delayA - 3];
1083 predictionA = d0 * p->coeffsA[filter][0] +
1084 d1 * p->coeffsA[filter][1] +
1085 d2 * p->coeffsA[filter][2] +
1086 d3 * p->coeffsA[filter][3];
1088 p->lastA[filter] = decoded + (predictionA >> 9);
1089 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1091 sign = APESIGN(decoded);
1092 p->coeffsA[filter][0] += ((d0 < 0) * 2 - 1) * sign;
1093 p->coeffsA[filter][1] += ((d1 < 0) * 2 - 1) * sign;
1094 p->coeffsA[filter][2] += ((d2 < 0) * 2 - 1) * sign;
1095 p->coeffsA[filter][3] += ((d3 < 0) * 2 - 1) * sign;
1097 return p->filterA[filter];
1100 static void predictor_decode_stereo_3930(APEContext *ctx, int count)
1102 APEPredictor *p = &ctx->predictor;
1103 int32_t *decoded0 = ctx->decoded[0];
1104 int32_t *decoded1 = ctx->decoded[1];
1106 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1110 int Y = *decoded1, X = *decoded0;
1111 *decoded0 = predictor_update_3930(p, Y, 0, YDELAYA);
1113 *decoded1 = predictor_update_3930(p, X, 1, XDELAYA);
1119 /* Have we filled the history buffer? */
1120 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1121 memmove(p->historybuffer, p->buf,
1122 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1123 p->buf = p->historybuffer;
1128 static void predictor_decode_mono_3930(APEContext *ctx, int count)
1130 APEPredictor *p = &ctx->predictor;
1131 int32_t *decoded0 = ctx->decoded[0];
1133 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1136 *decoded0 = predictor_update_3930(p, *decoded0, 0, YDELAYA);
1141 /* Have we filled the history buffer? */
1142 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1143 memmove(p->historybuffer, p->buf,
1144 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1145 p->buf = p->historybuffer;
1150 static av_always_inline int predictor_update_filter(APEPredictor *p,
1151 const int decoded, const int filter,
1152 const int delayA, const int delayB,
1153 const int adaptA, const int adaptB)
1155 int32_t predictionA, predictionB, sign;
1157 p->buf[delayA] = p->lastA[filter];
1158 p->buf[adaptA] = APESIGN(p->buf[delayA]);
1159 p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
1160 p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);
1162 predictionA = p->buf[delayA ] * p->coeffsA[filter][0] +
1163 p->buf[delayA - 1] * p->coeffsA[filter][1] +
1164 p->buf[delayA - 2] * p->coeffsA[filter][2] +
1165 p->buf[delayA - 3] * p->coeffsA[filter][3];
1167 /* Apply a scaled first-order filter compression */
1168 p->buf[delayB] = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
1169 p->buf[adaptB] = APESIGN(p->buf[delayB]);
1170 p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
1171 p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
1172 p->filterB[filter] = p->filterA[filter ^ 1];
1174 predictionB = p->buf[delayB ] * p->coeffsB[filter][0] +
1175 p->buf[delayB - 1] * p->coeffsB[filter][1] +
1176 p->buf[delayB - 2] * p->coeffsB[filter][2] +
1177 p->buf[delayB - 3] * p->coeffsB[filter][3] +
1178 p->buf[delayB - 4] * p->coeffsB[filter][4];
1180 p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
1181 p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);
1183 sign = APESIGN(decoded);
1184 p->coeffsA[filter][0] += p->buf[adaptA ] * sign;
1185 p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
1186 p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
1187 p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
1188 p->coeffsB[filter][0] += p->buf[adaptB ] * sign;
1189 p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
1190 p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
1191 p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
1192 p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;
1194 return p->filterA[filter];
1197 static void predictor_decode_stereo_3950(APEContext *ctx, int count)
1199 APEPredictor *p = &ctx->predictor;
1200 int32_t *decoded0 = ctx->decoded[0];
1201 int32_t *decoded1 = ctx->decoded[1];
1203 ape_apply_filters(ctx, ctx->decoded[0], ctx->decoded[1], count);
1207 *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
1208 YADAPTCOEFFSA, YADAPTCOEFFSB);
1210 *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
1211 XADAPTCOEFFSA, XADAPTCOEFFSB);
1217 /* Have we filled the history buffer? */
1218 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1219 memmove(p->historybuffer, p->buf,
1220 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1221 p->buf = p->historybuffer;
1226 static void predictor_decode_mono_3950(APEContext *ctx, int count)
1228 APEPredictor *p = &ctx->predictor;
1229 int32_t *decoded0 = ctx->decoded[0];
1230 int32_t predictionA, currentA, A, sign;
1232 ape_apply_filters(ctx, ctx->decoded[0], NULL, count);
1234 currentA = p->lastA[0];
1239 p->buf[YDELAYA] = currentA;
1240 p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];
1242 predictionA = p->buf[YDELAYA ] * p->coeffsA[0][0] +
1243 p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
1244 p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
1245 p->buf[YDELAYA - 3] * p->coeffsA[0][3];
1247 currentA = A + (predictionA >> 10);
1249 p->buf[YADAPTCOEFFSA] = APESIGN(p->buf[YDELAYA ]);
1250 p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);
1253 p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA ] * sign;
1254 p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
1255 p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
1256 p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;
1260 /* Have we filled the history buffer? */
1261 if (p->buf == p->historybuffer + HISTORY_SIZE) {
1262 memmove(p->historybuffer, p->buf,
1263 PREDICTOR_SIZE * sizeof(*p->historybuffer));
1264 p->buf = p->historybuffer;
1267 p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
1268 *(decoded0++) = p->filterA[0];
1271 p->lastA[0] = currentA;
1274 static void do_init_filter(APEFilter *f, int16_t *buf, int order)
1277 f->historybuffer = buf + order;
1278 f->delay = f->historybuffer + order * 2;
1279 f->adaptcoeffs = f->historybuffer + order;
1281 memset(f->historybuffer, 0, (order * 2) * sizeof(*f->historybuffer));
1282 memset(f->coeffs, 0, order * sizeof(*f->coeffs));
1286 static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
1288 do_init_filter(&f[0], buf, order);
1289 do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
1292 static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
1293 int32_t *data, int count, int order, int fracbits)
1299 /* round fixedpoint scalar product */
1300 res = ctx->adsp.scalarproduct_and_madd_int16(f->coeffs,
1302 f->adaptcoeffs - order,
1303 order, APESIGN(*data));
1304 res = (res + (1 << (fracbits - 1))) >> fracbits;
1308 /* Update the output history */
1309 *f->delay++ = av_clip_int16(res);
1311 if (version < 3980) {
1312 /* Version ??? to < 3.98 files (untested) */
1313 f->adaptcoeffs[0] = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
1314 f->adaptcoeffs[-4] >>= 1;
1315 f->adaptcoeffs[-8] >>= 1;
1317 /* Version 3.98 and later files */
1319 /* Update the adaption coefficients */
1320 absres = FFABS(res);
1322 *f->adaptcoeffs = ((res & (-1<<31)) ^ (-1<<30)) >>
1323 (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
1325 *f->adaptcoeffs = 0;
1327 f->avg += (absres - f->avg) / 16;
1329 f->adaptcoeffs[-1] >>= 1;
1330 f->adaptcoeffs[-2] >>= 1;
1331 f->adaptcoeffs[-8] >>= 1;
1336 /* Have we filled the history buffer? */
1337 if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
1338 memmove(f->historybuffer, f->delay - (order * 2),
1339 (order * 2) * sizeof(*f->historybuffer));
1340 f->delay = f->historybuffer + order * 2;
1341 f->adaptcoeffs = f->historybuffer + order;
1346 static void apply_filter(APEContext *ctx, APEFilter *f,
1347 int32_t *data0, int32_t *data1,
1348 int count, int order, int fracbits)
1350 do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
1352 do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
1355 static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
1356 int32_t *decoded1, int count)
1360 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1361 if (!ape_filter_orders[ctx->fset][i])
1363 apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
1364 ape_filter_orders[ctx->fset][i],
1365 ape_filter_fracbits[ctx->fset][i]);
1369 static int init_frame_decoder(APEContext *ctx)
1372 if ((ret = init_entropy_decoder(ctx)) < 0)
1374 init_predictor_decoder(ctx);
1376 for (i = 0; i < APE_FILTER_LEVELS; i++) {
1377 if (!ape_filter_orders[ctx->fset][i])
1379 init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
1380 ape_filter_orders[ctx->fset][i]);
1385 static void ape_unpack_mono(APEContext *ctx, int count)
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 mono\n");
1393 ctx->entropy_decode_mono(ctx, count);
1395 /* Now apply the predictor decoding */
1396 ctx->predictor_decode_mono(ctx, count);
1398 /* Pseudo-stereo - just copy left channel to right channel */
1399 if (ctx->channels == 2) {
1400 memcpy(ctx->decoded[1], ctx->decoded[0], count * sizeof(*ctx->decoded[1]));
1404 static void ape_unpack_stereo(APEContext *ctx, int count)
1406 int32_t left, right;
1407 int32_t *decoded0 = ctx->decoded[0];
1408 int32_t *decoded1 = ctx->decoded[1];
1410 if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
1411 /* We are pure silence, so we're done. */
1412 av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
1416 ctx->entropy_decode_stereo(ctx, count);
1418 /* Now apply the predictor decoding */
1419 ctx->predictor_decode_stereo(ctx, count);
1421 /* Decorrelate and scale to output depth */
1423 left = *decoded1 - (*decoded0 / 2);
1424 right = left + *decoded0;
1426 *(decoded0++) = left;
1427 *(decoded1++) = right;
1431 static int ape_decode_frame(AVCodecContext *avctx, void *data,
1432 int *got_frame_ptr, AVPacket *avpkt)
1434 AVFrame *frame = data;
1435 const uint8_t *buf = avpkt->data;
1436 APEContext *s = avctx->priv_data;
1443 /* this should never be negative, but bad things will happen if it is, so
1444 check it just to make sure. */
1445 av_assert0(s->samples >= 0);
1448 uint32_t nblocks, offset;
1455 if (avpkt->size < 8) {
1456 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1457 return AVERROR_INVALIDDATA;
1459 buf_size = avpkt->size & ~3;
1460 if (buf_size != avpkt->size) {
1461 av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
1462 "extra bytes at the end will be skipped.\n");
1464 if (s->fileversion < 3950) // previous versions overread two bytes
1466 av_fast_padded_malloc(&s->data, &s->data_size, buf_size);
1468 return AVERROR(ENOMEM);
1469 s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
1470 memset(s->data + (buf_size & ~3), 0, buf_size & 3);
1472 s->data_end = s->data + buf_size;
1474 nblocks = bytestream_get_be32(&s->ptr);
1475 offset = bytestream_get_be32(&s->ptr);
1476 if (s->fileversion >= 3900) {
1478 av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
1480 return AVERROR_INVALIDDATA;
1482 if (s->data_end - s->ptr < offset) {
1483 av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
1484 return AVERROR_INVALIDDATA;
1488 if ((ret = init_get_bits8(&s->gb, s->ptr, s->data_end - s->ptr)) < 0)
1490 if (s->fileversion > 3800)
1491 skip_bits_long(&s->gb, offset * 8);
1493 skip_bits_long(&s->gb, offset);
1496 if (!nblocks || nblocks > INT_MAX) {
1497 av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %"PRIu32".\n",
1499 return AVERROR_INVALIDDATA;
1501 s->samples = nblocks;
1503 /* Initialize the frame decoder */
1504 if (init_frame_decoder(s) < 0) {
1505 av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
1506 return AVERROR_INVALIDDATA;
1515 blockstodecode = FFMIN(s->blocks_per_loop, s->samples);
1516 // for old files coefficients were not interleaved,
1517 // so we need to decode all of them at once
1518 if (s->fileversion < 3930)
1519 blockstodecode = s->samples;
1521 /* reallocate decoded sample buffer if needed */
1522 av_fast_malloc(&s->decoded_buffer, &s->decoded_size,
1523 2 * FFALIGN(blockstodecode, 8) * sizeof(*s->decoded_buffer));
1524 if (!s->decoded_buffer)
1525 return AVERROR(ENOMEM);
1526 memset(s->decoded_buffer, 0, s->decoded_size);
1527 s->decoded[0] = s->decoded_buffer;
1528 s->decoded[1] = s->decoded_buffer + FFALIGN(blockstodecode, 8);
1530 /* get output buffer */
1531 frame->nb_samples = blockstodecode;
1532 if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
1537 if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
1538 ape_unpack_mono(s, blockstodecode);
1540 ape_unpack_stereo(s, blockstodecode);
1545 av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
1546 return AVERROR_INVALIDDATA;
1551 for (ch = 0; ch < s->channels; ch++) {
1552 sample8 = (uint8_t *)frame->data[ch];
1553 for (i = 0; i < blockstodecode; i++)
1554 *sample8++ = (s->decoded[ch][i] + 0x80) & 0xff;
1558 for (ch = 0; ch < s->channels; ch++) {
1559 sample16 = (int16_t *)frame->data[ch];
1560 for (i = 0; i < blockstodecode; i++)
1561 *sample16++ = s->decoded[ch][i];
1565 for (ch = 0; ch < s->channels; ch++) {
1566 sample24 = (int32_t *)frame->data[ch];
1567 for (i = 0; i < blockstodecode; i++)
1568 *sample24++ = s->decoded[ch][i] << 8;
1573 s->samples -= blockstodecode;
1577 return !s->samples ? avpkt->size : 0;
1580 static void ape_flush(AVCodecContext *avctx)
1582 APEContext *s = avctx->priv_data;
1586 #define OFFSET(x) offsetof(APEContext, x)
1587 #define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
1588 static const AVOption options[] = {
1589 { "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" },
1590 { "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" },
1594 static const AVClass ape_decoder_class = {
1595 .class_name = "APE decoder",
1596 .item_name = av_default_item_name,
1598 .version = LIBAVUTIL_VERSION_INT,
1601 AVCodec ff_ape_decoder = {
1603 .long_name = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
1604 .type = AVMEDIA_TYPE_AUDIO,
1605 .id = AV_CODEC_ID_APE,
1606 .priv_data_size = sizeof(APEContext),
1607 .init = ape_decode_init,
1608 .close = ape_decode_close,
1609 .decode = ape_decode_frame,
1610 .capabilities = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
1612 .sample_fmts = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
1615 AV_SAMPLE_FMT_NONE },
1616 .priv_class = &ape_decoder_class,