sonic.c

ffmpeg的完整源代码和作者自己写的文档。不但有在Linux的工程哦

/*
 * Simple free lossless/lossy audio codec
 * Copyright (c) 2004 Alex Beregszaszi
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg 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.
 *
 * FFmpeg 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 FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */
#include "avcodec.h"
#include "bitstream.h"
#include "golomb.h"

/**
 * @file sonic.c
 * Simple free lossless/lossy audio codec
 * Based on Paul Francis Harrison's Bonk (http://www.logarithmic.net/pfh/bonk)
 * Written and designed by Alex Beregszaszi
 *
 * TODO:
 *  - CABAC put/get_symbol
 *  - independent quantizer for channels
 *  - >2 channels support
 *  - more decorrelation types
 *  - more tap_quant tests
 *  - selectable intlist writers/readers (bonk-style, golomb, cabac)
 */

#define MAX_CHANNELS 2

#define MID_SIDE 0
#define LEFT_SIDE 1
#define RIGHT_SIDE 2

typedef struct SonicContext {
    int lossless, decorrelation;

    int num_taps, downsampling;
    double quantization;

    int channels, samplerate, block_align, frame_size;

    int *tap_quant;
    int *int_samples;
    int *coded_samples[MAX_CHANNELS];

    // for encoding
    int *tail;
    int tail_size;
    int *window;
    int window_size;

    // for decoding
    int *predictor_k;
    int *predictor_state[MAX_CHANNELS];
} SonicContext;

#define LATTICE_SHIFT   10
#define SAMPLE_SHIFT    4
#define LATTICE_FACTOR  (1 << LATTICE_SHIFT)
#define SAMPLE_FACTOR   (1 << SAMPLE_SHIFT)

#define BASE_QUANT      0.6
#define RATE_VARIATION  3.0

static inline int divide(int a, int b)
{
    if (a < 0)
        return -( (-a + b/2)/b );
    else
        return (a + b/2)/b;
}

static inline int shift(int a,int b)
{
    return (a+(1<<(b-1))) >> b;
}

static inline int shift_down(int a,int b)
{
    return (a>>b)+((a<0)?1:0);
}

#if 1
static inline int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
{
    int i;

    for (i = 0; i < entries; i++)
        set_se_golomb(pb, buf[i]);

    return 1;
}

static inline int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
{
    int i;

    for (i = 0; i < entries; i++)
        buf[i] = get_se_golomb(gb);

    return 1;
}

#else

#define ADAPT_LEVEL 8

static int bits_to_store(uint64_t x)
{
    int res = 0;

    while(x)
    {
        res++;
        x >>= 1;
    }
    return res;
}

static void write_uint_max(PutBitContext *pb, unsigned int value, unsigned int max)
{
    int i, bits;

    if (!max)
        return;

    bits = bits_to_store(max);

    for (i = 0; i < bits-1; i++)
        put_bits(pb, 1, value & (1 << i));

    if ( (value | (1 << (bits-1))) <= max)
        put_bits(pb, 1, value & (1 << (bits-1)));
}

static unsigned int read_uint_max(GetBitContext *gb, int max)
{
    int i, bits, value = 0;

    if (!max)
        return 0;

    bits = bits_to_store(max);

    for (i = 0; i < bits-1; i++)
        if (get_bits1(gb))
            value += 1 << i;

    if ( (value | (1<<(bits-1))) <= max)
        if (get_bits1(gb))
            value += 1 << (bits-1);

    return value;
}

static int intlist_write(PutBitContext *pb, int *buf, int entries, int base_2_part)
{
    int i, j, x = 0, low_bits = 0, max = 0;
    int step = 256, pos = 0, dominant = 0, any = 0;
    int *copy, *bits;

    copy = av_mallocz(4* entries);
    if (!copy)
        return -1;

    if (base_2_part)
    {
        int energy = 0;

        for (i = 0; i < entries; i++)
            energy += abs(buf[i]);

        low_bits = bits_to_store(energy / (entries * 2));
        if (low_bits > 15)
            low_bits = 15;

        put_bits(pb, 4, low_bits);
    }

    for (i = 0; i < entries; i++)
    {
        put_bits(pb, low_bits, abs(buf[i]));
        copy[i] = abs(buf[i]) >> low_bits;
        if (copy[i] > max)
            max = abs(copy[i]);
    }

    bits = av_mallocz(4* entries*max);
    if (!bits)
    {
//        av_free(copy);
        return -1;
    }

    for (i = 0; i <= max; i++)
    {
        for (j = 0; j < entries; j++)
            if (copy[j] >= i)
                bits[x++] = copy[j] > i;
    }

    // store bitstream
    while (pos < x)
    {
        int steplet = step >> 8;

        if (pos + steplet > x)
            steplet = x - pos;

        for (i = 0; i < steplet; i++)
            if (bits[i+pos] != dominant)
                any = 1;

        put_bits(pb, 1, any);

        if (!any)
        {
            pos += steplet;
            step += step / ADAPT_LEVEL;
        }
        else
        {
            int interloper = 0;

            while (((pos + interloper) < x) && (bits[pos + interloper] == dominant))
                interloper++;

            // note change
            write_uint_max(pb, interloper, (step >> 8) - 1);

            pos += interloper + 1;
            step -= step / ADAPT_LEVEL;
        }

        if (step < 256)
        {
            step = 65536 / step;
            dominant = !dominant;
        }
    }

    // store signs
    for (i = 0; i < entries; i++)
        if (buf[i])
            put_bits(pb, 1, buf[i] < 0);

//    av_free(bits);
//    av_free(copy);

    return 0;
}

static int intlist_read(GetBitContext *gb, int *buf, int entries, int base_2_part)
{
    int i, low_bits = 0, x = 0;
    int n_zeros = 0, step = 256, dominant = 0;
    int pos = 0, level = 0;
    int *bits = av_mallocz(4* entries);

    if (!bits)
        return -1;

    if (base_2_part)
    {
        low_bits = get_bits(gb, 4);

        if (low_bits)
            for (i = 0; i < entries; i++)
                buf[i] = get_bits(gb, low_bits);
    }

//    av_log(NULL, AV_LOG_INFO, "entries: %d, low bits: %d\n", entries, low_bits);

    while (n_zeros < entries)
    {
        int steplet = step >> 8;

        if (!get_bits1(gb))
        {
            for (i = 0; i < steplet; i++)
                bits[x++] = dominant;

            if (!dominant)
                n_zeros += steplet;

            step += step / ADAPT_LEVEL;
        }
        else
        {
            int actual_run = read_uint_max(gb, steplet-1);

//            av_log(NULL, AV_LOG_INFO, "actual run: %d\n", actual_run);

            for (i = 0; i < actual_run; i++)
                bits[x++] = dominant;

            bits[x++] = !dominant;

            if (!dominant)
                n_zeros += actual_run;
            else
                n_zeros++;

            step -= step / ADAPT_LEVEL;
        }

        if (step < 256)
        {
            step = 65536 / step;
            dominant = !dominant;
        }
    }

    // reconstruct unsigned values
    n_zeros = 0;
    for (i = 0; n_zeros < entries; i++)
    {
        while(1)
        {
            if (pos >= entries)
            {
                pos = 0;
                level += 1 << low_bits;
            }

            if (buf[pos] >= level)
                break;

            pos++;
        }

        if (bits[i])
            buf[pos] += 1 << low_bits;
        else
            n_zeros++;

        pos++;
    }
//    av_free(bits);

    // read signs
    for (i = 0; i < entries; i++)
        if (buf[i] && get_bits1(gb))
            buf[i] = -buf[i];

//    av_log(NULL, AV_LOG_INFO, "zeros: %d pos: %d\n", n_zeros, pos);

    return 0;
}
#endif

static void predictor_init_state(int *k, int *state, int order)
{
    int i;

    for (i = order-2; i >= 0; i--)
    {
        int j, p, x = state[i];

        for (j = 0, p = i+1; p < order; j++,p++)
            {
            int tmp = x + shift_down(k[j] * state[p], LATTICE_SHIFT);
            state[p] += shift_down(k[j]*x, LATTICE_SHIFT);
            x = tmp;
        }
    }
}

static int predictor_calc_error(int *k, int *state, int order, int error)
{
    int i, x = error - shift_down(k[order-1] * state[order-1], LATTICE_SHIFT);

#if 1
    int *k_ptr = &(k[order-2]),
        *state_ptr = &(state[order-2]);
    for (i = order-2; i >= 0; i--, k_ptr--, state_ptr--)
    {
        int k_value = *k_ptr, state_value = *state_ptr;
        x -= shift_down(k_value * state_value, LATTICE_SHIFT);
        state_ptr[1] = state_value + shift_down(k_value * x, LATTICE_SHIFT);
    }
#else
    for (i = order-2; i >= 0; i--)
    {
        x -= shift_down(k[i] * state[i], LATTICE_SHIFT);
        state[i+1] = state[i] + shift_down(k[i] * x, LATTICE_SHIFT);
    }
#endif

    // don't drift too far, to avoid overflows
    if (x >  (SAMPLE_FACTOR<<16)) x =  (SAMPLE_FACTOR<<16);
    if (x < -(SAMPLE_FACTOR<<16)) x = -(SAMPLE_FACTOR<<16);

    state[0] = x;

    return x;
}

#ifdef CONFIG_ENCODERS
// Heavily modified Levinson-Durbin algorithm which
// copes better with quantization, and calculates the
// actual whitened result as it goes.

static void modified_levinson_durbin(int *window, int window_entries,
        int *out, int out_entries, int channels, int *tap_quant)
{
    int i;
    int *state = av_mallocz(4* window_entries);

    memcpy(state, window, 4* window_entries);

    for (i = 0; i < out_entries; i++)
    {
        int step = (i+1)*channels, k, j;
        double xx = 0.0, xy = 0.0;
#if 1
        int *x_ptr = &(window[step]), *state_ptr = &(state[0]);
        j = window_entries - step;
        for (;j>=0;j--,x_ptr++,state_ptr++)
        {
            double x_value = *x_ptr, state_value = *state_ptr;
            xx += state_value*state_value;
            xy += x_value*state_value;
        }
#else
        for (j = 0; j <= (window_entries - step); j++);
        {
            double stepval = window[step+j], stateval = window[j];
//            xx += (double)window[j]*(double)window[j];
//            xy += (double)window[step+j]*(double)window[j];
            xx += stateval*stateval;
            xy += stepval*stateval;
        }
#endif
        if (xx == 0.0)
            k = 0;
        else
            k = (int)(floor(-xy/xx * (double)LATTICE_FACTOR / (double)(tap_quant[i]) + 0.5));

        if (k > (LATTICE_FACTOR/tap_quant[i]))
            k = LATTICE_FACTOR/tap_quant[i];
        if (-k > (LATTICE_FACTOR/tap_quant[i]))
            k = -(LATTICE_FACTOR/tap_quant[i]);

        out[i] = k;
        k *= tap_quant[i];

#if 1
        x_ptr = &(window[step]);
        state_ptr = &(state[0]);
        j = window_entries - step;
        for (;j>=0;j--,x_ptr++,state_ptr++)
        {
            int x_value = *x_ptr, state_value = *state_ptr;
            *x_ptr = x_value + shift_down(k*state_value,LATTICE_SHIFT);
            *state_ptr = state_value + shift_down(k*x_value, LATTICE_SHIFT);
        }
#else
        for (j=0; j <= (window_entries - step); j++)
        {
            int stepval = window[step+j], stateval=state[j];
            window[step+j] += shift_down(k * stateval, LATTICE_SHIFT);
            state[j] += shift_down(k * stepval, LATTICE_SHIFT);
        }
#endif
    }

    av_free(state);
}
#endif /* CONFIG_ENCODERS */

static int samplerate_table[] =
    { 44100, 22050, 11025, 96000, 48000, 32000, 24000, 16000, 8000 };

#ifdef CONFIG_ENCODERS