ac3enc.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

/*
 * The simplest AC3 encoder
 * Copyright (c) 2000 Fabrice Bellard.
 *
 * 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 ac3enc.c
 * The simplest AC3 encoder.
 */
//#define DEBUG
//#define DEBUG_BITALLOC
#include "avcodec.h"
#include "bitstream.h"
#include "crc.h"
#include "ac3.h"

typedef struct AC3EncodeContext {
    PutBitContext pb;
    int nb_channels;
    int nb_all_channels;
    int lfe_channel;
    int bit_rate;
    unsigned int sample_rate;
    unsigned int bsid;
    unsigned int frame_size_min; /* minimum frame size in case rounding is necessary */
    unsigned int frame_size; /* current frame size in words */
    unsigned int bits_written;
    unsigned int samples_written;
    int halfratecod;
    unsigned int frmsizecod;
    unsigned int fscod; /* frequency */
    unsigned int acmod;
    int lfe;
    unsigned int bsmod;
    short last_samples[AC3_MAX_CHANNELS][256];
    unsigned int chbwcod[AC3_MAX_CHANNELS];
    int nb_coefs[AC3_MAX_CHANNELS];

    /* bitrate allocation control */
    int sgaincod, sdecaycod, fdecaycod, dbkneecod, floorcod;
    AC3BitAllocParameters bit_alloc;
    int csnroffst;
    int fgaincod[AC3_MAX_CHANNELS];
    int fsnroffst[AC3_MAX_CHANNELS];
    /* mantissa encoding */
    int mant1_cnt, mant2_cnt, mant4_cnt;
} AC3EncodeContext;

static int16_t costab[64];
static int16_t sintab[64];
static int16_t fft_rev[512];
static int16_t xcos1[128];
static int16_t xsin1[128];

#define MDCT_NBITS 9
#define N         (1 << MDCT_NBITS)

/* new exponents are sent if their Norm 1 exceed this number */
#define EXP_DIFF_THRESHOLD 1000

static void fft_init(int ln);

static inline int16_t fix15(float a)
{
    int v;
    v = (int)(a * (float)(1 << 15));
    if (v < -32767)
        v = -32767;
    else if (v > 32767)
        v = 32767;
    return v;
}

typedef struct IComplex {
    short re,im;
} IComplex;

static void fft_init(int ln)
{
    int i, j, m, n;
    float alpha;

    n = 1 << ln;

    for(i=0;i<(n/2);i++) {
        alpha = 2 * M_PI * (float)i / (float)n;
        costab[i] = fix15(cos(alpha));
        sintab[i] = fix15(sin(alpha));
    }

    for(i=0;i<n;i++) {
        m=0;
        for(j=0;j<ln;j++) {
            m |= ((i >> j) & 1) << (ln-j-1);
        }
        fft_rev[i]=m;
    }
}

/* butter fly op */
#define BF(pre, pim, qre, qim, pre1, pim1, qre1, qim1) \
{\
  int ax, ay, bx, by;\
  bx=pre1;\
  by=pim1;\
  ax=qre1;\
  ay=qim1;\
  pre = (bx + ax) >> 1;\
  pim = (by + ay) >> 1;\
  qre = (bx - ax) >> 1;\
  qim = (by - ay) >> 1;\
}

#define MUL16(a,b) ((a) * (b))

#define CMUL(pre, pim, are, aim, bre, bim) \
{\
   pre = (MUL16(are, bre) - MUL16(aim, bim)) >> 15;\
   pim = (MUL16(are, bim) + MUL16(bre, aim)) >> 15;\
}


/* do a 2^n point complex fft on 2^ln points. */
static void fft(IComplex *z, int ln)
{
    int        j, l, np, np2;
    int        nblocks, nloops;
    register IComplex *p,*q;
    int tmp_re, tmp_im;

    np = 1 << ln;

    /* reverse */
    for(j=0;j<np;j++) {
        int k;
        IComplex tmp;
        k = fft_rev[j];
        if (k < j) {
            tmp = z[k];
            z[k] = z[j];
            z[j] = tmp;
        }
    }

    /* pass 0 */

    p=&z[0];
    j=(np >> 1);
    do {
        BF(p[0].re, p[0].im, p[1].re, p[1].im,
           p[0].re, p[0].im, p[1].re, p[1].im);
        p+=2;
    } while (--j != 0);

    /* pass 1 */

    p=&z[0];
    j=np >> 2;
    do {
        BF(p[0].re, p[0].im, p[2].re, p[2].im,
           p[0].re, p[0].im, p[2].re, p[2].im);
        BF(p[1].re, p[1].im, p[3].re, p[3].im,
           p[1].re, p[1].im, p[3].im, -p[3].re);
        p+=4;
    } while (--j != 0);

    /* pass 2 .. ln-1 */

    nblocks = np >> 3;
    nloops = 1 << 2;
    np2 = np >> 1;
    do {
        p = z;
        q = z + nloops;
        for (j = 0; j < nblocks; ++j) {

            BF(p->re, p->im, q->re, q->im,
               p->re, p->im, q->re, q->im);

            p++;
            q++;
            for(l = nblocks; l < np2; l += nblocks) {
                CMUL(tmp_re, tmp_im, costab[l], -sintab[l], q->re, q->im);
                BF(p->re, p->im, q->re, q->im,
                   p->re, p->im, tmp_re, tmp_im);
                p++;
                q++;
            }
            p += nloops;
            q += nloops;
        }
        nblocks = nblocks >> 1;
        nloops = nloops << 1;
    } while (nblocks != 0);
}

/* do a 512 point mdct */
static void mdct512(int32_t *out, int16_t *in)
{
    int i, re, im, re1, im1;
    int16_t rot[N];
    IComplex x[N/4];

    /* shift to simplify computations */
    for(i=0;i<N/4;i++)
        rot[i] = -in[i + 3*N/4];
    for(i=N/4;i<N;i++)
        rot[i] = in[i - N/4];

    /* pre rotation */
    for(i=0;i<N/4;i++) {
        re = ((int)rot[2*i] - (int)rot[N-1-2*i]) >> 1;
        im = -((int)rot[N/2+2*i] - (int)rot[N/2-1-2*i]) >> 1;
        CMUL(x[i].re, x[i].im, re, im, -xcos1[i], xsin1[i]);
    }

    fft(x, MDCT_NBITS - 2);

    /* post rotation */
    for(i=0;i<N/4;i++) {
        re = x[i].re;
        im = x[i].im;
        CMUL(re1, im1, re, im, xsin1[i], xcos1[i]);
        out[2*i] = im1;
        out[N/2-1-2*i] = re1;
    }
}

/* XXX: use another norm ? */
static int calc_exp_diff(uint8_t *exp1, uint8_t *exp2, int n)
{
    int sum, i;
    sum = 0;
    for(i=0;i<n;i++) {
        sum += abs(exp1[i] - exp2[i]);
    }
    return sum;
}

static void compute_exp_strategy(uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
                                 uint8_t exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
                                 int ch, int is_lfe)
{
    int i, j;
    int exp_diff;

    /* estimate if the exponent variation & decide if they should be
       reused in the next frame */
    exp_strategy[0][ch] = EXP_NEW;
    for(i=1;i<NB_BLOCKS;i++) {
        exp_diff = calc_exp_diff(exp[i][ch], exp[i-1][ch], N/2);
#ifdef DEBUG
        av_log(NULL, AV_LOG_DEBUG, "exp_diff=%d\n", exp_diff);
#endif
        if (exp_diff > EXP_DIFF_THRESHOLD)
            exp_strategy[i][ch] = EXP_NEW;
        else
            exp_strategy[i][ch] = EXP_REUSE;
    }
    if (is_lfe)
        return;

    /* now select the encoding strategy type : if exponents are often
       recoded, we use a coarse encoding */
    i = 0;
    while (i < NB_BLOCKS) {
        j = i + 1;
        while (j < NB_BLOCKS && exp_strategy[j][ch] == EXP_REUSE)
            j++;
        switch(j - i) {
        case 1:
            exp_strategy[i][ch] = EXP_D45;
            break;
        case 2:
        case 3:
            exp_strategy[i][ch] = EXP_D25;
            break;
        default:
            exp_strategy[i][ch] = EXP_D15;
            break;
        }
        i = j;
    }
}

/* set exp[i] to min(exp[i], exp1[i]) */
static void exponent_min(uint8_t exp[N/2], uint8_t exp1[N/2], int n)
{
    int i;

    for(i=0;i<n;i++) {
        if (exp1[i] < exp[i])
            exp[i] = exp1[i];
    }
}

/* update the exponents so that they are the ones the decoder will
   decode. Return the number of bits used to code the exponents */
static int encode_exp(uint8_t encoded_exp[N/2],
                      uint8_t exp[N/2],
                      int nb_exps,
                      int exp_strategy)
{
    int group_size, nb_groups, i, j, k, exp_min;
    uint8_t exp1[N/2];

    switch(exp_strategy) {
    case EXP_D15:
        group_size = 1;
        break;
    case EXP_D25:
        group_size = 2;
        break;
    default:
    case EXP_D45:
        group_size = 4;
        break;
    }
    nb_groups = ((nb_exps + (group_size * 3) - 4) / (3 * group_size)) * 3;

    /* for each group, compute the minimum exponent */
    exp1[0] = exp[0]; /* DC exponent is handled separately */
    k = 1;
    for(i=1;i<=nb_groups;i++) {
        exp_min = exp[k];
        assert(exp_min >= 0 && exp_min <= 24);
        for(j=1;j<group_size;j++) {
            if (exp[k+j] < exp_min)
                exp_min = exp[k+j];
        }
        exp1[i] = exp_min;
        k += group_size;
    }

    /* constraint for DC exponent */
    if (exp1[0] > 15)
        exp1[0] = 15;

    /* Decrease the delta between each groups to within 2
     * so that they can be differentially encoded */
    for (i=1;i<=nb_groups;i++)
        exp1[i] = FFMIN(exp1[i], exp1[i-1] + 2);
    for (i=nb_groups-1;i>=0;i--)
        exp1[i] = FFMIN(exp1[i], exp1[i+1] + 2);

    /* now we have the exponent values the decoder will see */
    encoded_exp[0] = exp1[0];
    k = 1;
    for(i=1;i<=nb_groups;i++) {
        for(j=0;j<group_size;j++) {
            encoded_exp[k+j] = exp1[i];
        }
        k += group_size;
    }

#if defined(DEBUG)
    av_log(NULL, AV_LOG_DEBUG, "exponents: strategy=%d\n", exp_strategy);
    for(i=0;i<=nb_groups * group_size;i++) {
        av_log(NULL, AV_LOG_DEBUG, "%d ", encoded_exp[i]);
    }
    av_log(NULL, AV_LOG_DEBUG, "\n");
#endif

    return 4 + (nb_groups / 3) * 7;
}

/* return the size in bits taken by the mantissa */
static int compute_mantissa_size(AC3EncodeContext *s, uint8_t *m, int nb_coefs)
{
    int bits, mant, i;

    bits = 0;
    for(i=0;i<nb_coefs;i++) {
        mant = m[i];
        switch(mant) {
        case 0:
            /* nothing */
            break;
        case 1:
            /* 3 mantissa in 5 bits */
            if (s->mant1_cnt == 0)
                bits += 5;
            if (++s->mant1_cnt == 3)
                s->mant1_cnt = 0;
            break;
        case 2:
            /* 3 mantissa in 7 bits */
            if (s->mant2_cnt == 0)
                bits += 7;
            if (++s->mant2_cnt == 3)
                s->mant2_cnt = 0;
            break;
        case 3:
            bits += 3;
            break;
        case 4:
            /* 2 mantissa in 7 bits */
            if (s->mant4_cnt == 0)
                bits += 7;
            if (++s->mant4_cnt == 2)
                s->mant4_cnt = 0;
            break;
        case 14:
            bits += 14;
            break;
        case 15:
            bits += 16;
            break;
        default:
            bits += mant - 1;
            break;
        }
    }
    return bits;
}


static void bit_alloc_masking(AC3EncodeContext *s,
                              uint8_t encoded_exp[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
                              uint8_t exp_strategy[NB_BLOCKS][AC3_MAX_CHANNELS],
                              int16_t psd[NB_BLOCKS][AC3_MAX_CHANNELS][N/2],
                              int16_t mask[NB_BLOCKS][AC3_MAX_CHANNELS][50])
{
    int blk, ch;
    int16_t bndpsd[NB_BLOCKS][AC3_MAX_CHANNELS][50];

    for(blk=0; blk<NB_BLOCKS; blk++) {
        for(ch=0;ch<s->nb_all_channels;ch++) {
            if(exp_strategy[blk][ch] == EXP_REUSE) {
                memcpy(psd[blk][ch], psd[blk-1][ch], (N/2)*sizeof(int16_t));
                memcpy(mask[blk][ch], mask[blk-1][ch], 50*sizeof(int16_t));
            } else {
                ff_ac3_bit_alloc_calc_psd(encoded_exp[blk][ch], 0,
                                          s->nb_coefs[ch],
                                          psd[blk][ch], bndpsd[blk][ch]);
                ff_ac3_bit_alloc_calc_mask(&s->bit_alloc, bndpsd[blk][ch],
                                           0, s->nb_coefs[ch],
                                           ff_fgaintab[s->fgaincod[ch]],
                                           ch == s->lfe_channel,
                                           DBA_NONE, 0, NULL, NULL, NULL,
                                           mask[blk][ch]);