apedec.c

Go to the documentation of this file.

/*
 * Monkey's Audio lossless audio decoder
 * Copyright (c) 2007 Benjamin Zores <ben@geexbox.org>
 *  based upon libdemac from Dave Chapman.
 *
 * This file is part of Libav.
 *
 * Libav is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * Libav is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with Libav; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

#include "libavutil/avassert.h"
#include "libavutil/channel_layout.h"
#include "libavutil/opt.h"
#include "avcodec.h"
#include "dsputil.h"
#include "bytestream.h"
#include "internal.h"

#define MAX_CHANNELS        2
#define MAX_BYTESPERSAMPLE  3

#define APE_FRAMECODE_MONO_SILENCE    1
#define APE_FRAMECODE_STEREO_SILENCE  3
#define APE_FRAMECODE_PSEUDO_STEREO   4

#define HISTORY_SIZE 512
#define PREDICTOR_ORDER 8

#define PREDICTOR_SIZE 50

#define YDELAYA (18 + PREDICTOR_ORDER*4)
#define YDELAYB (18 + PREDICTOR_ORDER*3)
#define XDELAYA (18 + PREDICTOR_ORDER*2)
#define XDELAYB (18 + PREDICTOR_ORDER)

#define YADAPTCOEFFSA 18
#define XADAPTCOEFFSA 14
#define YADAPTCOEFFSB 10
#define XADAPTCOEFFSB 5

enum APECompressionLevel {
    COMPRESSION_LEVEL_FAST       = 1000,
    COMPRESSION_LEVEL_NORMAL     = 2000,
    COMPRESSION_LEVEL_HIGH       = 3000,
    COMPRESSION_LEVEL_EXTRA_HIGH = 4000,
    COMPRESSION_LEVEL_INSANE     = 5000
};
#define APE_FILTER_LEVELS 3

static const uint16_t ape_filter_orders[5][APE_FILTER_LEVELS] = {
    {  0,   0,    0 },
    { 16,   0,    0 },
    { 64,   0,    0 },
    { 32, 256,    0 },
    { 16, 256, 1280 }
};

static const uint8_t ape_filter_fracbits[5][APE_FILTER_LEVELS] = {
    {  0,  0,  0 },
    { 11,  0,  0 },
    { 11,  0,  0 },
    { 10, 13,  0 },
    { 11, 13, 15 }
};


typedef struct APEFilter {
    int16_t *coeffs;        
    int16_t *adaptcoeffs;   
    int16_t *historybuffer; 
    int16_t *delay;         

    int avg;
} APEFilter;

typedef struct APERice {
    uint32_t k;
    uint32_t ksum;
} APERice;

typedef struct APERangecoder {
    uint32_t low;           
    uint32_t range;         
    uint32_t help;          
    unsigned int buffer;    
} APERangecoder;

typedef struct APEPredictor {
    int32_t *buf;

    int32_t lastA[2];

    int32_t filterA[2];
    int32_t filterB[2];

    int32_t coeffsA[2][4];  
    int32_t coeffsB[2][5];  
    int32_t historybuffer[HISTORY_SIZE + PREDICTOR_SIZE];
} APEPredictor;

typedef struct APEContext {
    AVClass *class;                          
    AVCodecContext *avctx;
    AVFrame frame;
    DSPContext dsp;
    int channels;
    int samples;                             
    int bps;

    int fileversion;                         
    int compression_level;                   
    int fset;                                
    int flags;                               

    uint32_t CRC;                            
    int frameflags;                          
    APEPredictor predictor;                  

    int32_t *decoded_buffer;
    int decoded_size;
    int32_t *decoded[MAX_CHANNELS];          
    int blocks_per_loop;                     

    int16_t* filterbuf[APE_FILTER_LEVELS];   

    APERangecoder rc;                        
    APERice riceX;                           
    APERice riceY;                           
    APEFilter filters[APE_FILTER_LEVELS][2]; 

    uint8_t *data;                           
    uint8_t *data_end;                       
    int data_size;                           
    const uint8_t *ptr;                      

    int error;
} APEContext;

// TODO: dsputilize

static av_cold int ape_decode_close(AVCodecContext *avctx)
{
    APEContext *s = avctx->priv_data;
    int i;

    for (i = 0; i < APE_FILTER_LEVELS; i++)
        av_freep(&s->filterbuf[i]);

    av_freep(&s->decoded_buffer);
    av_freep(&s->data);
    s->decoded_size = s->data_size = 0;

    return 0;
}

static av_cold int ape_decode_init(AVCodecContext *avctx)
{
    APEContext *s = avctx->priv_data;
    int i;

    if (avctx->extradata_size != 6) {
        av_log(avctx, AV_LOG_ERROR, "Incorrect extradata\n");
        return AVERROR(EINVAL);
    }
    if (avctx->channels > 2) {
        av_log(avctx, AV_LOG_ERROR, "Only mono and stereo is supported\n");
        return AVERROR(EINVAL);
    }
    s->bps = avctx->bits_per_coded_sample;
    switch (s->bps) {
    case 8:
        avctx->sample_fmt = AV_SAMPLE_FMT_U8P;
        break;
    case 16:
        avctx->sample_fmt = AV_SAMPLE_FMT_S16P;
        break;
    case 24:
        avctx->sample_fmt = AV_SAMPLE_FMT_S32P;
        break;
    default:
        av_log_ask_for_sample(avctx, "Unsupported bits per coded sample %d\n",
                              s->bps);
        return AVERROR_PATCHWELCOME;
    }
    s->avctx             = avctx;
    s->channels          = avctx->channels;
    s->fileversion       = AV_RL16(avctx->extradata);
    s->compression_level = AV_RL16(avctx->extradata + 2);
    s->flags             = AV_RL16(avctx->extradata + 4);

    av_log(avctx, AV_LOG_DEBUG, "Compression Level: %d - Flags: %d\n",
           s->compression_level, s->flags);
    if (s->compression_level % 1000 || s->compression_level > COMPRESSION_LEVEL_INSANE) {
        av_log(avctx, AV_LOG_ERROR, "Incorrect compression level %d\n",
               s->compression_level);
        return AVERROR_INVALIDDATA;
    }
    s->fset = s->compression_level / 1000 - 1;
    for (i = 0; i < APE_FILTER_LEVELS; i++) {
        if (!ape_filter_orders[s->fset][i])
            break;
        FF_ALLOC_OR_GOTO(avctx, s->filterbuf[i],
                         (ape_filter_orders[s->fset][i] * 3 + HISTORY_SIZE) * 4,
                         filter_alloc_fail);
    }

    ff_dsputil_init(&s->dsp, avctx);
    avctx->channel_layout = (avctx->channels==2) ? AV_CH_LAYOUT_STEREO : AV_CH_LAYOUT_MONO;

    avcodec_get_frame_defaults(&s->frame);
    avctx->coded_frame = &s->frame;

    return 0;
filter_alloc_fail:
    ape_decode_close(avctx);
    return AVERROR(ENOMEM);
}

#define CODE_BITS    32
#define TOP_VALUE    ((unsigned int)1 << (CODE_BITS-1))
#define SHIFT_BITS   (CODE_BITS - 9)
#define EXTRA_BITS   ((CODE_BITS-2) % 8 + 1)
#define BOTTOM_VALUE (TOP_VALUE >> 8)

static inline void range_start_decoding(APEContext *ctx)
{
    ctx->rc.buffer = bytestream_get_byte(&ctx->ptr);
    ctx->rc.low    = ctx->rc.buffer >> (8 - EXTRA_BITS);
    ctx->rc.range  = (uint32_t) 1 << EXTRA_BITS;
}

static inline void range_dec_normalize(APEContext *ctx)
{
    while (ctx->rc.range <= BOTTOM_VALUE) {
        ctx->rc.buffer <<= 8;
        if(ctx->ptr < ctx->data_end) {
            ctx->rc.buffer += *ctx->ptr;
            ctx->ptr++;
        } else {
            ctx->error = 1;
        }
        ctx->rc.low    = (ctx->rc.low << 8)    | ((ctx->rc.buffer >> 1) & 0xFF);
        ctx->rc.range  <<= 8;
    }
}

static inline int range_decode_culfreq(APEContext *ctx, int tot_f)
{
    range_dec_normalize(ctx);
    ctx->rc.help = ctx->rc.range / tot_f;
    return ctx->rc.low / ctx->rc.help;
}

static inline int range_decode_culshift(APEContext *ctx, int shift)
{
    range_dec_normalize(ctx);
    ctx->rc.help = ctx->rc.range >> shift;
    return ctx->rc.low / ctx->rc.help;
}


static inline void range_decode_update(APEContext *ctx, int sy_f, int lt_f)
{
    ctx->rc.low  -= ctx->rc.help * lt_f;
    ctx->rc.range = ctx->rc.help * sy_f;
}

static inline int range_decode_bits(APEContext *ctx, int n)
{
    int sym = range_decode_culshift(ctx, n);
    range_decode_update(ctx, 1, sym);
    return sym;
}


#define MODEL_ELEMENTS 64

static const uint16_t counts_3970[22] = {
        0, 14824, 28224, 39348, 47855, 53994, 58171, 60926,
    62682, 63786, 64463, 64878, 65126, 65276, 65365, 65419,
    65450, 65469, 65480, 65487, 65491, 65493,
};

static const uint16_t counts_diff_3970[21] = {
    14824, 13400, 11124, 8507, 6139, 4177, 2755, 1756,
    1104, 677, 415, 248, 150, 89, 54, 31,
    19, 11, 7, 4, 2,
};

static const uint16_t counts_3980[22] = {
        0, 19578, 36160, 48417, 56323, 60899, 63265, 64435,
    64971, 65232, 65351, 65416, 65447, 65466, 65476, 65482,
    65485, 65488, 65490, 65491, 65492, 65493,
};

static const uint16_t counts_diff_3980[21] = {
    19578, 16582, 12257, 7906, 4576, 2366, 1170, 536,
    261, 119, 65, 31, 19, 10, 6, 3,
    3, 2, 1, 1, 1,
};

static inline int range_get_symbol(APEContext *ctx,
                                   const uint16_t counts[],
                                   const uint16_t counts_diff[])
{
    int symbol, cf;

    cf = range_decode_culshift(ctx, 16);

    if(cf > 65492){
        symbol= cf - 65535 + 63;
        range_decode_update(ctx, 1, cf);
        if(cf > 65535)
            ctx->error=1;
        return symbol;
    }
    /* figure out the symbol inefficiently; a binary search would be much better */
    for (symbol = 0; counts[symbol + 1] <= cf; symbol++);

    range_decode_update(ctx, counts_diff[symbol], counts[symbol]);

    return symbol;
} // group rangecoder

static inline void update_rice(APERice *rice, unsigned int x)
{
    int lim = rice->k ? (1 << (rice->k + 4)) : 0;
    rice->ksum += ((x + 1) / 2) - ((rice->ksum + 16) >> 5);

    if (rice->ksum < lim)
        rice->k--;
    else if (rice->ksum >= (1 << (rice->k + 5)))
        rice->k++;
}

static inline int ape_decode_value(APEContext *ctx, APERice *rice)
{
    unsigned int x, overflow;

    if (ctx->fileversion < 3990) {
        int tmpk;

        overflow = range_get_symbol(ctx, counts_3970, counts_diff_3970);

        if (overflow == (MODEL_ELEMENTS - 1)) {
            tmpk = range_decode_bits(ctx, 5);
            overflow = 0;
        } else
            tmpk = (rice->k < 1) ? 0 : rice->k - 1;

        if (tmpk <= 16)
            x = range_decode_bits(ctx, tmpk);
        else if (tmpk <= 32) {
            x = range_decode_bits(ctx, 16);
            x |= (range_decode_bits(ctx, tmpk - 16) << 16);
        } else {
            av_log(ctx->avctx, AV_LOG_ERROR, "Too many bits: %d\n", tmpk);
            return AVERROR_INVALIDDATA;
        }
        x += overflow << tmpk;
    } else {
        int base, pivot;

        pivot = rice->ksum >> 5;
        if (pivot == 0)
            pivot = 1;

        overflow = range_get_symbol(ctx, counts_3980, counts_diff_3980);

        if (overflow == (MODEL_ELEMENTS - 1)) {
            overflow  = range_decode_bits(ctx, 16) << 16;
            overflow |= range_decode_bits(ctx, 16);
        }

        if (pivot < 0x10000) {
            base = range_decode_culfreq(ctx, pivot);
            range_decode_update(ctx, 1, base);
        } else {
            int base_hi = pivot, base_lo;
            int bbits = 0;

            while (base_hi & ~0xFFFF) {
                base_hi >>= 1;
                bbits++;
            }
            base_hi = range_decode_culfreq(ctx, base_hi + 1);
            range_decode_update(ctx, 1, base_hi);
            base_lo = range_decode_culfreq(ctx, 1 << bbits);
            range_decode_update(ctx, 1, base_lo);

            base = (base_hi << bbits) + base_lo;
        }

        x = base + overflow * pivot;
    }

    update_rice(rice, x);

    /* Convert to signed */
    if (x & 1)
        return (x >> 1) + 1;
    else
        return -(x >> 1);
}

static void entropy_decode(APEContext *ctx, int blockstodecode, int stereo)
{
    int32_t *decoded0 = ctx->decoded[0];
    int32_t *decoded1 = ctx->decoded[1];

    while (blockstodecode--) {
        *decoded0++ = ape_decode_value(ctx, &ctx->riceY);
        if (stereo)
            *decoded1++ = ape_decode_value(ctx, &ctx->riceX);
    }
}

static int init_entropy_decoder(APEContext *ctx)
{
    /* Read the CRC */
    if (ctx->data_end - ctx->ptr < 6)
        return AVERROR_INVALIDDATA;
    ctx->CRC = bytestream_get_be32(&ctx->ptr);

    /* Read the frame flags if they exist */
    ctx->frameflags = 0;
    if ((ctx->fileversion > 3820) && (ctx->CRC & 0x80000000)) {
        ctx->CRC &= ~0x80000000;

        if (ctx->data_end - ctx->ptr < 6)
            return AVERROR_INVALIDDATA;
        ctx->frameflags = bytestream_get_be32(&ctx->ptr);
    }

    /* Initialize the rice structs */
    ctx->riceX.k = 10;
    ctx->riceX.ksum = (1 << ctx->riceX.k) * 16;
    ctx->riceY.k = 10;
    ctx->riceY.ksum = (1 << ctx->riceY.k) * 16;

    /* The first 8 bits of input are ignored. */
    ctx->ptr++;

    range_start_decoding(ctx);

    return 0;
}

static const int32_t initial_coeffs[4] = {
    360, 317, -109, 98
};

static void init_predictor_decoder(APEContext *ctx)
{
    APEPredictor *p = &ctx->predictor;

    /* Zero the history buffers */
    memset(p->historybuffer, 0, PREDICTOR_SIZE * sizeof(*p->historybuffer));
    p->buf = p->historybuffer;

    /* Initialize and zero the coefficients */
    memcpy(p->coeffsA[0], initial_coeffs, sizeof(initial_coeffs));
    memcpy(p->coeffsA[1], initial_coeffs, sizeof(initial_coeffs));
    memset(p->coeffsB, 0, sizeof(p->coeffsB));

    p->filterA[0] = p->filterA[1] = 0;
    p->filterB[0] = p->filterB[1] = 0;
    p->lastA[0]   = p->lastA[1]   = 0;
}

static inline int APESIGN(int32_t x) {
    return (x < 0) - (x > 0);
}

static av_always_inline int predictor_update_filter(APEPredictor *p,
                                                    const int decoded, const int filter,
                                                    const int delayA,  const int delayB,
                                                    const int adaptA,  const int adaptB)
{
    int32_t predictionA, predictionB, sign;

    p->buf[delayA]     = p->lastA[filter];
    p->buf[adaptA]     = APESIGN(p->buf[delayA]);
    p->buf[delayA - 1] = p->buf[delayA] - p->buf[delayA - 1];
    p->buf[adaptA - 1] = APESIGN(p->buf[delayA - 1]);

    predictionA = p->buf[delayA    ] * p->coeffsA[filter][0] +
                  p->buf[delayA - 1] * p->coeffsA[filter][1] +
                  p->buf[delayA - 2] * p->coeffsA[filter][2] +
                  p->buf[delayA - 3] * p->coeffsA[filter][3];

    /*  Apply a scaled first-order filter compression */
    p->buf[delayB]     = p->filterA[filter ^ 1] - ((p->filterB[filter] * 31) >> 5);
    p->buf[adaptB]     = APESIGN(p->buf[delayB]);
    p->buf[delayB - 1] = p->buf[delayB] - p->buf[delayB - 1];
    p->buf[adaptB - 1] = APESIGN(p->buf[delayB - 1]);
    p->filterB[filter] = p->filterA[filter ^ 1];

    predictionB = p->buf[delayB    ] * p->coeffsB[filter][0] +
                  p->buf[delayB - 1] * p->coeffsB[filter][1] +
                  p->buf[delayB - 2] * p->coeffsB[filter][2] +
                  p->buf[delayB - 3] * p->coeffsB[filter][3] +
                  p->buf[delayB - 4] * p->coeffsB[filter][4];

    p->lastA[filter] = decoded + ((predictionA + (predictionB >> 1)) >> 10);
    p->filterA[filter] = p->lastA[filter] + ((p->filterA[filter] * 31) >> 5);

    sign = APESIGN(decoded);
    p->coeffsA[filter][0] += p->buf[adaptA    ] * sign;
    p->coeffsA[filter][1] += p->buf[adaptA - 1] * sign;
    p->coeffsA[filter][2] += p->buf[adaptA - 2] * sign;
    p->coeffsA[filter][3] += p->buf[adaptA - 3] * sign;
    p->coeffsB[filter][0] += p->buf[adaptB    ] * sign;
    p->coeffsB[filter][1] += p->buf[adaptB - 1] * sign;
    p->coeffsB[filter][2] += p->buf[adaptB - 2] * sign;
    p->coeffsB[filter][3] += p->buf[adaptB - 3] * sign;
    p->coeffsB[filter][4] += p->buf[adaptB - 4] * sign;

    return p->filterA[filter];
}

static void predictor_decode_stereo(APEContext *ctx, int count)
{
    APEPredictor *p = &ctx->predictor;
    int32_t *decoded0 = ctx->decoded[0];
    int32_t *decoded1 = ctx->decoded[1];

    while (count--) {
        /* Predictor Y */
        *decoded0 = predictor_update_filter(p, *decoded0, 0, YDELAYA, YDELAYB,
                                            YADAPTCOEFFSA, YADAPTCOEFFSB);
        decoded0++;
        *decoded1 = predictor_update_filter(p, *decoded1, 1, XDELAYA, XDELAYB,
                                            XADAPTCOEFFSA, XADAPTCOEFFSB);
        decoded1++;

        /* Combined */
        p->buf++;

        /* Have we filled the history buffer? */
        if (p->buf == p->historybuffer + HISTORY_SIZE) {
            memmove(p->historybuffer, p->buf,
                    PREDICTOR_SIZE * sizeof(*p->historybuffer));
            p->buf = p->historybuffer;
        }
    }
}

static void predictor_decode_mono(APEContext *ctx, int count)
{
    APEPredictor *p = &ctx->predictor;
    int32_t *decoded0 = ctx->decoded[0];
    int32_t predictionA, currentA, A, sign;

    currentA = p->lastA[0];

    while (count--) {
        A = *decoded0;

        p->buf[YDELAYA] = currentA;
        p->buf[YDELAYA - 1] = p->buf[YDELAYA] - p->buf[YDELAYA - 1];

        predictionA = p->buf[YDELAYA    ] * p->coeffsA[0][0] +
                      p->buf[YDELAYA - 1] * p->coeffsA[0][1] +
                      p->buf[YDELAYA - 2] * p->coeffsA[0][2] +
                      p->buf[YDELAYA - 3] * p->coeffsA[0][3];

        currentA = A + (predictionA >> 10);

        p->buf[YADAPTCOEFFSA]     = APESIGN(p->buf[YDELAYA    ]);
        p->buf[YADAPTCOEFFSA - 1] = APESIGN(p->buf[YDELAYA - 1]);

        sign = APESIGN(A);
        p->coeffsA[0][0] += p->buf[YADAPTCOEFFSA    ] * sign;
        p->coeffsA[0][1] += p->buf[YADAPTCOEFFSA - 1] * sign;
        p->coeffsA[0][2] += p->buf[YADAPTCOEFFSA - 2] * sign;
        p->coeffsA[0][3] += p->buf[YADAPTCOEFFSA - 3] * sign;

        p->buf++;

        /* Have we filled the history buffer? */
        if (p->buf == p->historybuffer + HISTORY_SIZE) {
            memmove(p->historybuffer, p->buf,
                    PREDICTOR_SIZE * sizeof(*p->historybuffer));
            p->buf = p->historybuffer;
        }

        p->filterA[0] = currentA + ((p->filterA[0] * 31) >> 5);
        *(decoded0++) = p->filterA[0];
    }

    p->lastA[0] = currentA;
}

static void do_init_filter(APEFilter *f, int16_t *buf, int order)
{
    f->coeffs = buf;
    f->historybuffer = buf + order;
    f->delay       = f->historybuffer + order * 2;
    f->adaptcoeffs = f->historybuffer + order;

    memset(f->historybuffer, 0, (order * 2) * sizeof(*f->historybuffer));
    memset(f->coeffs, 0, order * sizeof(*f->coeffs));
    f->avg = 0;
}

static void init_filter(APEContext *ctx, APEFilter *f, int16_t *buf, int order)
{
    do_init_filter(&f[0], buf, order);
    do_init_filter(&f[1], buf + order * 3 + HISTORY_SIZE, order);
}

static void do_apply_filter(APEContext *ctx, int version, APEFilter *f,
                            int32_t *data, int count, int order, int fracbits)
{
    int res;
    int absres;

    while (count--) {
        /* round fixedpoint scalar product */
        res = ctx->dsp.scalarproduct_and_madd_int16(f->coeffs, f->delay - order,
                                                    f->adaptcoeffs - order,
                                                    order, APESIGN(*data));
        res = (res + (1 << (fracbits - 1))) >> fracbits;
        res += *data;
        *data++ = res;

        /* Update the output history */
        *f->delay++ = av_clip_int16(res);

        if (version < 3980) {
            /* Version ??? to < 3.98 files (untested) */
            f->adaptcoeffs[0]  = (res == 0) ? 0 : ((res >> 28) & 8) - 4;
            f->adaptcoeffs[-4] >>= 1;
            f->adaptcoeffs[-8] >>= 1;
        } else {
            /* Version 3.98 and later files */

            /* Update the adaption coefficients */
            absres = FFABS(res);
            if (absres)
                *f->adaptcoeffs = ((res & (-1<<31)) ^ (-1<<30)) >>
                                  (25 + (absres <= f->avg*3) + (absres <= f->avg*4/3));
            else
                *f->adaptcoeffs = 0;

            f->avg += (absres - f->avg) / 16;

            f->adaptcoeffs[-1] >>= 1;
            f->adaptcoeffs[-2] >>= 1;
            f->adaptcoeffs[-8] >>= 1;
        }

        f->adaptcoeffs++;

        /* Have we filled the history buffer? */
        if (f->delay == f->historybuffer + HISTORY_SIZE + (order * 2)) {
            memmove(f->historybuffer, f->delay - (order * 2),
                    (order * 2) * sizeof(*f->historybuffer));
            f->delay = f->historybuffer + order * 2;
            f->adaptcoeffs = f->historybuffer + order;
        }
    }
}

static void apply_filter(APEContext *ctx, APEFilter *f,
                         int32_t *data0, int32_t *data1,
                         int count, int order, int fracbits)
{
    do_apply_filter(ctx, ctx->fileversion, &f[0], data0, count, order, fracbits);
    if (data1)
        do_apply_filter(ctx, ctx->fileversion, &f[1], data1, count, order, fracbits);
}

static void ape_apply_filters(APEContext *ctx, int32_t *decoded0,
                              int32_t *decoded1, int count)
{
    int i;

    for (i = 0; i < APE_FILTER_LEVELS; i++) {
        if (!ape_filter_orders[ctx->fset][i])
            break;
        apply_filter(ctx, ctx->filters[i], decoded0, decoded1, count,
                     ape_filter_orders[ctx->fset][i],
                     ape_filter_fracbits[ctx->fset][i]);
    }
}

static int init_frame_decoder(APEContext *ctx)
{
    int i, ret;
    if ((ret = init_entropy_decoder(ctx)) < 0)
        return ret;
    init_predictor_decoder(ctx);

    for (i = 0; i < APE_FILTER_LEVELS; i++) {
        if (!ape_filter_orders[ctx->fset][i])
            break;
        init_filter(ctx, ctx->filters[i], ctx->filterbuf[i],
                    ape_filter_orders[ctx->fset][i]);
    }
    return 0;
}

static void ape_unpack_mono(APEContext *ctx, int count)
{
    if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
        /* We are pure silence, so we're done. */
        av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence mono\n");
        return;
    }

    entropy_decode(ctx, count, 0);
    ape_apply_filters(ctx, ctx->decoded[0], NULL, count);

    /* Now apply the predictor decoding */
    predictor_decode_mono(ctx, count);

    /* Pseudo-stereo - just copy left channel to right channel */
    if (ctx->channels == 2) {
        memcpy(ctx->decoded[1], ctx->decoded[0], count * sizeof(*ctx->decoded[1]));
    }
}

static void ape_unpack_stereo(APEContext *ctx, int count)
{
    int32_t left, right;
    int32_t *decoded0 = ctx->decoded[0];
    int32_t *decoded1 = ctx->decoded[1];

    if (ctx->frameflags & APE_FRAMECODE_STEREO_SILENCE) {
        /* We are pure silence, so we're done. */
        av_log(ctx->avctx, AV_LOG_DEBUG, "pure silence stereo\n");
        return;
    }

    entropy_decode(ctx, count, 1);
    ape_apply_filters(ctx, decoded0, decoded1, count);

    /* Now apply the predictor decoding */
    predictor_decode_stereo(ctx, count);

    /* Decorrelate and scale to output depth */
    while (count--) {
        left = *decoded1 - (*decoded0 / 2);
        right = left + *decoded0;

        *(decoded0++) = left;
        *(decoded1++) = right;
    }
}

static int ape_decode_frame(AVCodecContext *avctx, void *data,
                            int *got_frame_ptr, AVPacket *avpkt)
{
    const uint8_t *buf = avpkt->data;
    APEContext *s = avctx->priv_data;
    uint8_t *sample8;
    int16_t *sample16;
    int32_t *sample24;
    int i, ch, ret;
    int blockstodecode;
    int bytes_used = 0;

    /* this should never be negative, but bad things will happen if it is, so
       check it just to make sure. */
    av_assert0(s->samples >= 0);

    if(!s->samples){
        uint32_t nblocks, offset;
        int buf_size;

        if (!avpkt->size) {
            *got_frame_ptr = 0;
            return 0;
        }
        if (avpkt->size < 8) {
            av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
            return AVERROR_INVALIDDATA;
        }
        buf_size = avpkt->size & ~3;
        if (buf_size != avpkt->size) {
            av_log(avctx, AV_LOG_WARNING, "packet size is not a multiple of 4. "
                   "extra bytes at the end will be skipped.\n");
        }

        av_fast_malloc(&s->data, &s->data_size, buf_size);
        if (!s->data)
            return AVERROR(ENOMEM);
        s->dsp.bswap_buf((uint32_t*)s->data, (const uint32_t*)buf, buf_size >> 2);
        s->ptr = s->data;
        s->data_end = s->data + buf_size;

        nblocks = bytestream_get_be32(&s->ptr);
        offset  = bytestream_get_be32(&s->ptr);
        if (offset > 3) {
            av_log(avctx, AV_LOG_ERROR, "Incorrect offset passed\n");
            s->data = NULL;
            return AVERROR_INVALIDDATA;
        }
        if (s->data_end - s->ptr < offset) {
            av_log(avctx, AV_LOG_ERROR, "Packet is too small\n");
            return AVERROR_INVALIDDATA;
        }
        s->ptr += offset;

        if (!nblocks || nblocks > INT_MAX) {
            av_log(avctx, AV_LOG_ERROR, "Invalid sample count: %u.\n", nblocks);
            return AVERROR_INVALIDDATA;
        }
        s->samples = nblocks;

        /* Initialize the frame decoder */
        if (init_frame_decoder(s) < 0) {
            av_log(avctx, AV_LOG_ERROR, "Error reading frame header\n");
            return AVERROR_INVALIDDATA;
        }

        bytes_used = avpkt->size;
    }

    if (!s->data) {
        *got_frame_ptr = 0;
        return avpkt->size;
    }

    blockstodecode = FFMIN(s->blocks_per_loop, s->samples);

    /* reallocate decoded sample buffer if needed */
    av_fast_malloc(&s->decoded_buffer, &s->decoded_size,
                   2 * FFALIGN(blockstodecode, 8) * sizeof(*s->decoded_buffer));
    if (!s->decoded_buffer)
        return AVERROR(ENOMEM);
    memset(s->decoded_buffer, 0, s->decoded_size);
    s->decoded[0] = s->decoded_buffer;
    s->decoded[1] = s->decoded_buffer + FFALIGN(blockstodecode, 8);

    /* get output buffer */
    s->frame.nb_samples = blockstodecode;
    if ((ret = ff_get_buffer(avctx, &s->frame)) < 0) {
        av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
        return ret;
    }

    s->error=0;

    if ((s->channels == 1) || (s->frameflags & APE_FRAMECODE_PSEUDO_STEREO))
        ape_unpack_mono(s, blockstodecode);
    else
        ape_unpack_stereo(s, blockstodecode);
    emms_c();

    if (s->error) {
        s->samples=0;
        av_log(avctx, AV_LOG_ERROR, "Error decoding frame\n");
        return AVERROR_INVALIDDATA;
    }

    switch (s->bps) {
    case 8:
        for (ch = 0; ch < s->channels; ch++) {
            sample8 = (uint8_t *)s->frame.data[ch];
            for (i = 0; i < blockstodecode; i++)
                *sample8++ = (s->decoded[ch][i] + 0x80) & 0xff;
        }
        break;
    case 16:
        for (ch = 0; ch < s->channels; ch++) {
            sample16 = (int16_t *)s->frame.data[ch];
            for (i = 0; i < blockstodecode; i++)
                *sample16++ = s->decoded[ch][i];
        }
        break;
    case 24:
        for (ch = 0; ch < s->channels; ch++) {
            sample24 = (int32_t *)s->frame.data[ch];
            for (i = 0; i < blockstodecode; i++)
                *sample24++ = s->decoded[ch][i] << 8;
        }
        break;
    }

    s->samples -= blockstodecode;

    *got_frame_ptr   = 1;
    *(AVFrame *)data = s->frame;

    return bytes_used;
}

static void ape_flush(AVCodecContext *avctx)
{
    APEContext *s = avctx->priv_data;
    s->samples= 0;
}

#define OFFSET(x) offsetof(APEContext, x)
#define PAR (AV_OPT_FLAG_DECODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM)
static const AVOption options[] = {
    { "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" },
    { "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" },
    { NULL},
};

static const AVClass ape_decoder_class = {
    .class_name = "APE decoder",
    .item_name  = av_default_item_name,
    .option     = options,
    .version    = LIBAVUTIL_VERSION_INT,
};

AVCodec ff_ape_decoder = {
    .name           = "ape",
    .type           = AVMEDIA_TYPE_AUDIO,
    .id             = AV_CODEC_ID_APE,
    .priv_data_size = sizeof(APEContext),
    .init           = ape_decode_init,
    .close          = ape_decode_close,
    .decode         = ape_decode_frame,
    .capabilities   = CODEC_CAP_SUBFRAMES | CODEC_CAP_DELAY | CODEC_CAP_DR1,
    .flush          = ape_flush,
    .long_name      = NULL_IF_CONFIG_SMALL("Monkey's Audio"),
    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_U8P,
                                                      AV_SAMPLE_FMT_S16P,
                                                      AV_SAMPLE_FMT_S32P,
                                                      AV_SAMPLE_FMT_NONE },
    .priv_class     = &ape_decoder_class,
};