imc.c

mediastreamer2是开源的网络传输媒体流的库

/*
 * IMC compatible decoder
 * Copyright (c) 2002-2004 Maxim Poliakovski
 * Copyright (c) 2006 Benjamin Larsson
 * Copyright (c) 2006 Konstantin Shishkov
 *
 * 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
 */

/**
 *  @file imc.c IMC - Intel Music Coder
 *  A mdct based codec using a 256 points large transform
 *  divied into 32 bands with some mix of scale factors.
 *  Only mono is supported.
 *
 */


#include <math.h>
#include <stddef.h>
#include <stdio.h>

#define ALT_BITSTREAM_READER
#include "avcodec.h"
#include "bitstream.h"
#include "dsputil.h"

#include "imcdata.h"

#define IMC_BLOCK_SIZE 64
#define IMC_FRAME_ID 0x21
#define BANDS 32
#define COEFFS 256

typedef struct {
    float old_floor[BANDS];
    float flcoeffs1[BANDS];
    float flcoeffs2[BANDS];
    float flcoeffs3[BANDS];
    float flcoeffs4[BANDS];
    float flcoeffs5[BANDS];
    float flcoeffs6[BANDS];
    float CWdecoded[COEFFS];

    /** MDCT tables */
    //@{
    float mdct_sine_window[COEFFS];
    float post_cos[COEFFS];
    float post_sin[COEFFS];
    float pre_coef1[COEFFS];
    float pre_coef2[COEFFS];
    float last_fft_im[COEFFS];
    //@}

    int bandWidthT[BANDS];     ///< codewords per band
    int bitsBandT[BANDS];      ///< how many bits per codeword in band
    int CWlengthT[COEFFS];     ///< how many bits in each codeword
    int levlCoeffBuf[BANDS];
    int bandFlagsBuf[BANDS];   ///< flags for each band
    int sumLenArr[BANDS];      ///< bits for all coeffs in band
    int skipFlagRaw[BANDS];    ///< skip flags are stored in raw form or not
    int skipFlagBits[BANDS];   ///< bits used to code skip flags
    int skipFlagCount[BANDS];  ///< skipped coeffients per band
    int skipFlags[COEFFS];     ///< skip coefficient decoding or not
    int codewords[COEFFS];     ///< raw codewords read from bitstream
    float sqrt_tab[30];
    GetBitContext gb;
    VLC huffman_vlc[4][4];
    int decoder_reset;
    float one_div_log2;

    DSPContext dsp;
    FFTContext fft;
    DECLARE_ALIGNED_16(FFTComplex, samples[COEFFS/2]);
    DECLARE_ALIGNED_16(float, out_samples[COEFFS]);
} IMCContext;


static int imc_decode_init(AVCodecContext * avctx)
{
    int i, j;
    IMCContext *q = avctx->priv_data;
    double r1, r2;

    q->decoder_reset = 1;

    for(i = 0; i < BANDS; i++)
        q->old_floor[i] = 1.0;

    /* Build mdct window, a simple sine window normalized with sqrt(2) */
    for(i = 0; i < COEFFS; i++)
        q->mdct_sine_window[i] = sin((i + 0.5) / 512.0 * M_PI) * sqrt(2.0);
    for(i = 0; i < COEFFS/2; i++){
        q->post_cos[i] = cos(i / 256.0 * M_PI);
        q->post_sin[i] = sin(i / 256.0 * M_PI);

        r1 = sin((i * 4.0 + 1.0) / 1024.0 * M_PI);
        r2 = cos((i * 4.0 + 1.0) / 1024.0 * M_PI);

        if (i & 0x1)
        {
            q->pre_coef1[i] =  (r1 + r2) * sqrt(2.0);
            q->pre_coef2[i] = -(r1 - r2) * sqrt(2.0);
        }
        else
        {
            q->pre_coef1[i] = -(r1 + r2) * sqrt(2.0);
            q->pre_coef2[i] =  (r1 - r2) * sqrt(2.0);
        }

        q->last_fft_im[i] = 0;
    }

    /* Generate a square root table */

    for(i = 0; i < 30; i++) {
        q->sqrt_tab[i] = sqrt(i);
    }

    /* initialize the VLC tables */
    for(i = 0; i < 4 ; i++) {
        for(j = 0; j < 4; j++) {
            init_vlc (&q->huffman_vlc[i][j], 9, imc_huffman_sizes[i],
                     imc_huffman_lens[i][j], 1, 1,
                     imc_huffman_bits[i][j], 2, 2, 1);
        }
    }
    q->one_div_log2 = 1/log(2);

    ff_fft_init(&q->fft, 7, 1);
    dsputil_init(&q->dsp, avctx);
    return 0;
}

static void imc_calculate_coeffs(IMCContext* q, float* flcoeffs1, float* flcoeffs2, int* bandWidthT,
                                float* flcoeffs3, float* flcoeffs5)
{
    float   workT1[BANDS];
    float   workT2[BANDS];
    float   workT3[BANDS];
    float   snr_limit = 1.e-30;
    float   accum = 0.0;
    int i, cnt2;

    for(i = 0; i < BANDS; i++) {
        flcoeffs5[i] = workT2[i] = 0.0;
        if (bandWidthT[i]){
            workT1[i] = flcoeffs1[i] * flcoeffs1[i];
            flcoeffs3[i] = 2.0 * flcoeffs2[i];
        } else {
            workT1[i] = 0.0;
            flcoeffs3[i] = -30000.0;
        }
        workT3[i] = bandWidthT[i] * workT1[i] * 0.01;
        if (workT3[i] <= snr_limit)
            workT3[i] = 0.0;
    }

    for(i = 0; i < BANDS; i++) {
        for(cnt2 = i; cnt2 < cyclTab[i]; cnt2++)
            flcoeffs5[cnt2] = flcoeffs5[cnt2] + workT3[i];
        workT2[cnt2-1] = workT2[cnt2-1] + workT3[i];
    }

    for(i = 1; i < BANDS; i++) {
        accum = (workT2[i-1] + accum) * imc_weights1[i-1];
        flcoeffs5[i] += accum;
    }

    for(i = 0; i < BANDS; i++)
        workT2[i] = 0.0;

    for(i = 0; i < BANDS; i++) {
        for(cnt2 = i-1; cnt2 > cyclTab2[i]; cnt2--)
            flcoeffs5[cnt2] += workT3[i];
        workT2[cnt2+1] += workT3[i];
    }

    accum = 0.0;

    for(i = BANDS-2; i >= 0; i--) {
        accum = (workT2[i+1] + accum) * imc_weights2[i];
        flcoeffs5[i] += accum;
        //there is missing code here, but it seems to never be triggered
    }
}


static void imc_read_level_coeffs(IMCContext* q, int stream_format_code, int* levlCoeffs)
{
    int i;
    VLC *hufftab[4];
    int start = 0;
    const uint8_t *cb_sel;
    int s;

    s = stream_format_code >> 1;
    hufftab[0] = &q->huffman_vlc[s][0];
    hufftab[1] = &q->huffman_vlc[s][1];
    hufftab[2] = &q->huffman_vlc[s][2];
    hufftab[3] = &q->huffman_vlc[s][3];
    cb_sel = imc_cb_select[s];

    if(stream_format_code & 4)
        start = 1;
    if(start)
        levlCoeffs[0] = get_bits(&q->gb, 7);
    for(i = start; i < BANDS; i++){
        levlCoeffs[i] = get_vlc2(&q->gb, hufftab[cb_sel[i]]->table, hufftab[cb_sel[i]]->bits, 2);
        if(levlCoeffs[i] == 17)
            levlCoeffs[i] += get_bits(&q->gb, 4);
    }
}

static void imc_decode_level_coefficients(IMCContext* q, int* levlCoeffBuf, float* flcoeffs1,
                                         float* flcoeffs2)
{
    int i, level;
    float tmp, tmp2;
    //maybe some frequency division thingy

    flcoeffs1[0] = 20000.0 / pow (2, levlCoeffBuf[0] * 0.18945); // 0.18945 = log2(10) * 0.05703125
    flcoeffs2[0] = log(flcoeffs1[0])/log(2);
    tmp = flcoeffs1[0];
    tmp2 = flcoeffs2[0];

    for(i = 1; i < BANDS; i++) {
        level = levlCoeffBuf[i];
        if (level == 16) {
            flcoeffs1[i] = 1.0;
            flcoeffs2[i] = 0.0;
        } else {
            if (level < 17)
                level -=7;
            else if (level <= 24)
                level -=32;
            else
                level -=16;

            tmp  *= imc_exp_tab[15 + level];
            tmp2 += 0.83048 * level;  // 0.83048 = log2(10) * 0.25
            flcoeffs1[i] = tmp;
            flcoeffs2[i] = tmp2;
        }
    }
}


static void imc_decode_level_coefficients2(IMCContext* q, int* levlCoeffBuf, float* old_floor, float* flcoeffs1,
                                          float* flcoeffs2) {
    int i;
        //FIXME maybe flag_buf = noise coding and flcoeffs1 = new scale factors
        //      and flcoeffs2 old scale factors
        //      might be incomplete due to a missing table that is in the binary code
    for(i = 0; i < BANDS; i++) {
        flcoeffs1[i] = 0;
        if(levlCoeffBuf[i] < 16) {
            flcoeffs1[i] = imc_exp_tab2[levlCoeffBuf[i]] * old_floor[i];
            flcoeffs2[i] = (levlCoeffBuf[i]-7) * 0.83048 + flcoeffs2[i]; // 0.83048 = log2(10) * 0.25
        } else {
            flcoeffs1[i] = old_floor[i];
        }
    }
}

/**
 * Perform bit allocation depending on bits available
 */
static int bit_allocation (IMCContext* q, int stream_format_code, int freebits, int flag) {
    int i, j;
    const float limit = -1.e20;
    float highest = 0.0;
    int indx;
    int t1 = 0;
    int t2 = 1;
    float summa = 0.0;
    int iacc = 0;
    int summer = 0;
    int rres, cwlen;
    float lowest = 1.e10;
    int low_indx = 0;
    float workT[32];
    int flg;
    int found_indx = 0;

    for(i = 0; i < BANDS; i++)
        highest = FFMAX(highest, q->flcoeffs1[i]);

    for(i = 0; i < BANDS-1; i++) {
        q->flcoeffs4[i] = q->flcoeffs3[i] - log(q->flcoeffs5[i])/log(2);
    }
    q->flcoeffs4[BANDS - 1] = limit;

    highest = highest * 0.25;

    for(i = 0; i < BANDS; i++) {
        indx = -1;
        if ((band_tab[i+1] - band_tab[i]) == q->bandWidthT[i])
            indx = 0;

        if ((band_tab[i+1] - band_tab[i]) > q->bandWidthT[i])
            indx = 1;

        if (((band_tab[i+1] - band_tab[i])/2) >= q->bandWidthT[i])
            indx = 2;

        if (indx == -1)
            return -1;

        q->flcoeffs4[i] = q->flcoeffs4[i] + xTab[(indx*2 + (q->flcoeffs1[i] < highest)) * 2 + flag];
    }

    if (stream_format_code & 0x2) {
        q->flcoeffs4[0] = limit;
        q->flcoeffs4[1] = limit;
        q->flcoeffs4[2] = limit;
        q->flcoeffs4[3] = limit;
    }

    for(i = (stream_format_code & 0x2)?4:0; i < BANDS-1; i++) {
        iacc += q->bandWidthT[i];
        summa += q->bandWidthT[i] * q->flcoeffs4[i];
    }
    q->bandWidthT[BANDS-1] = 0;
    summa = (summa * 0.5 - freebits) / iacc;


    for(i = 0; i < BANDS/2; i++) {
        rres = summer - freebits;
        if((rres >= -8) && (rres <= 8)) break;

        summer = 0;
        iacc = 0;

        for(j = (stream_format_code & 0x2)?4:0; j < BANDS; j++) {
            cwlen = av_clip((int)((q->flcoeffs4[j] * 0.5) - summa + 0.5), 0, 6);

            q->bitsBandT[j] = cwlen;
            summer += q->bandWidthT[j] * cwlen;

            if (cwlen > 0)
                iacc += q->bandWidthT[j];
        }

        flg = t2;
        t2 = 1;
        if (freebits < summer)
            t2 = -1;
        if (i == 0)
            flg = t2;
        if(flg != t2)
            t1++;

        summa = (float)(summer - freebits) / ((t1 + 1) * iacc) + summa;
    }

    for(i = (stream_format_code & 0x2)?4:0; i < BANDS; i++) {
        for(j = band_tab[i]; j < band_tab[i+1]; j++)
            q->CWlengthT[j] = q->bitsBandT[i];
    }

    if (freebits > summer) {
        for(i = 0; i < BANDS; i++) {
            workT[i] = (q->bitsBandT[i] == 6) ? -1.e20 : (q->bitsBandT[i] * -2 + q->flcoeffs4[i] - 0.415);
        }

        highest = 0.0;

        do{
            if (highest <= -1.e20)
                break;

            found_indx = 0;
            highest = -1.e20;

            for(i = 0; i < BANDS; i++) {
                if (workT[i] > highest) {
                    highest = workT[i];
                    found_indx = i;
                }
            }

            if (highest > -1.e20) {
                workT[found_indx] -= 2.0;
                if (++(q->bitsBandT[found_indx]) == 6)
                    workT[found_indx] = -1.e20;

                for(j = band_tab[found_indx]; j < band_tab[found_indx+1] && (freebits > summer); j++){
                    q->CWlengthT[j]++;
                    summer++;
                }
            }
        }while (freebits > summer);
    }
    if (freebits < summer) {