mpegaudiodec.c

mediastreamer2是开源的网络传输媒体流的库

    register const MPA_INT *w, *w2, *p;
    int j, offset, v;
    OUT_INT *samples2;
#if FRAC_BITS <= 15
    int sum, sum2;
#else
    int64_t sum, sum2;
#endif

    dct32(tmp, sb_samples);

    offset = *synth_buf_offset;
    synth_buf = synth_buf_ptr + offset;

    for(j=0;j<32;j++) {
        v = tmp[j];
#if FRAC_BITS <= 15
        /* NOTE: can cause a loss in precision if very high amplitude
           sound */
        v = av_clip_int16(v);
#endif
        synth_buf[j] = v;
    }
    /* copy to avoid wrap */
    memcpy(synth_buf + 512, synth_buf, 32 * sizeof(MPA_INT));

    samples2 = samples + 31 * incr;
    w = window;
    w2 = window + 31;

    sum = *dither_state;
    p = synth_buf + 16;
    SUM8(sum, +=, w, p);
    p = synth_buf + 48;
    SUM8(sum, -=, w + 32, p);
    *samples = round_sample(&sum);
    samples += incr;
    w++;

    /* we calculate two samples at the same time to avoid one memory
       access per two sample */
    for(j=1;j<16;j++) {
        sum2 = 0;
        p = synth_buf + 16 + j;
        SUM8P2(sum, +=, sum2, -=, w, w2, p);
        p = synth_buf + 48 - j;
        SUM8P2(sum, -=, sum2, -=, w + 32, w2 + 32, p);

        *samples = round_sample(&sum);
        samples += incr;
        sum += sum2;
        *samples2 = round_sample(&sum);
        samples2 -= incr;
        w++;
        w2--;
    }

    p = synth_buf + 32;
    SUM8(sum, -=, w + 32, p);
    *samples = round_sample(&sum);
    *dither_state= sum;

    offset = (offset - 32) & 511;
    *synth_buf_offset = offset;
}

#define C3 FIXHR(0.86602540378443864676/2)

/* 0.5 / cos(pi*(2*i+1)/36) */
static const int icos36[9] = {
    FIXR(0.50190991877167369479),
    FIXR(0.51763809020504152469), //0
    FIXR(0.55168895948124587824),
    FIXR(0.61038729438072803416),
    FIXR(0.70710678118654752439), //1
    FIXR(0.87172339781054900991),
    FIXR(1.18310079157624925896),
    FIXR(1.93185165257813657349), //2
    FIXR(5.73685662283492756461),
};

/* 0.5 / cos(pi*(2*i+1)/36) */
static const int icos36h[9] = {
    FIXHR(0.50190991877167369479/2),
    FIXHR(0.51763809020504152469/2), //0
    FIXHR(0.55168895948124587824/2),
    FIXHR(0.61038729438072803416/2),
    FIXHR(0.70710678118654752439/2), //1
    FIXHR(0.87172339781054900991/2),
    FIXHR(1.18310079157624925896/4),
    FIXHR(1.93185165257813657349/4), //2
//    FIXHR(5.73685662283492756461),
};

/* 12 points IMDCT. We compute it "by hand" by factorizing obvious
   cases. */
static void imdct12(int *out, int *in)
{
    int in0, in1, in2, in3, in4, in5, t1, t2;

    in0= in[0*3];
    in1= in[1*3] + in[0*3];
    in2= in[2*3] + in[1*3];
    in3= in[3*3] + in[2*3];
    in4= in[4*3] + in[3*3];
    in5= in[5*3] + in[4*3];
    in5 += in3;
    in3 += in1;

    in2= MULH(2*in2, C3);
    in3= MULH(4*in3, C3);

    t1 = in0 - in4;
    t2 = MULH(2*(in1 - in5), icos36h[4]);

    out[ 7]=
    out[10]= t1 + t2;
    out[ 1]=
    out[ 4]= t1 - t2;

    in0 += in4>>1;
    in4 = in0 + in2;
    in5 += 2*in1;
    in1 = MULH(in5 + in3, icos36h[1]);
    out[ 8]=
    out[ 9]= in4 + in1;
    out[ 2]=
    out[ 3]= in4 - in1;

    in0 -= in2;
    in5 = MULH(2*(in5 - in3), icos36h[7]);
    out[ 0]=
    out[ 5]= in0 - in5;
    out[ 6]=
    out[11]= in0 + in5;
}

/* cos(pi*i/18) */
#define C1 FIXHR(0.98480775301220805936/2)
#define C2 FIXHR(0.93969262078590838405/2)
#define C3 FIXHR(0.86602540378443864676/2)
#define C4 FIXHR(0.76604444311897803520/2)
#define C5 FIXHR(0.64278760968653932632/2)
#define C6 FIXHR(0.5/2)
#define C7 FIXHR(0.34202014332566873304/2)
#define C8 FIXHR(0.17364817766693034885/2)


/* using Lee like decomposition followed by hand coded 9 points DCT */
static void imdct36(int *out, int *buf, int *in, int *win)
{
    int i, j, t0, t1, t2, t3, s0, s1, s2, s3;
    int tmp[18], *tmp1, *in1;

    for(i=17;i>=1;i--)
        in[i] += in[i-1];
    for(i=17;i>=3;i-=2)
        in[i] += in[i-2];

    for(j=0;j<2;j++) {
        tmp1 = tmp + j;
        in1 = in + j;
#if 0
//more accurate but slower
        int64_t t0, t1, t2, t3;
        t2 = in1[2*4] + in1[2*8] - in1[2*2];

        t3 = (in1[2*0] + (int64_t)(in1[2*6]>>1))<<32;
        t1 = in1[2*0] - in1[2*6];
        tmp1[ 6] = t1 - (t2>>1);
        tmp1[16] = t1 + t2;

        t0 = MUL64(2*(in1[2*2] + in1[2*4]),    C2);
        t1 = MUL64(   in1[2*4] - in1[2*8] , -2*C8);
        t2 = MUL64(2*(in1[2*2] + in1[2*8]),   -C4);

        tmp1[10] = (t3 - t0 - t2) >> 32;
        tmp1[ 2] = (t3 + t0 + t1) >> 32;
        tmp1[14] = (t3 + t2 - t1) >> 32;

        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
        t2 = MUL64(2*(in1[2*1] + in1[2*5]),    C1);
        t3 = MUL64(   in1[2*5] - in1[2*7] , -2*C7);
        t0 = MUL64(2*in1[2*3], C3);

        t1 = MUL64(2*(in1[2*1] + in1[2*7]),   -C5);

        tmp1[ 0] = (t2 + t3 + t0) >> 32;
        tmp1[12] = (t2 + t1 - t0) >> 32;
        tmp1[ 8] = (t3 - t1 - t0) >> 32;
#else
        t2 = in1[2*4] + in1[2*8] - in1[2*2];

        t3 = in1[2*0] + (in1[2*6]>>1);
        t1 = in1[2*0] - in1[2*6];
        tmp1[ 6] = t1 - (t2>>1);
        tmp1[16] = t1 + t2;

        t0 = MULH(2*(in1[2*2] + in1[2*4]),    C2);
        t1 = MULH(   in1[2*4] - in1[2*8] , -2*C8);
        t2 = MULH(2*(in1[2*2] + in1[2*8]),   -C4);

        tmp1[10] = t3 - t0 - t2;
        tmp1[ 2] = t3 + t0 + t1;
        tmp1[14] = t3 + t2 - t1;

        tmp1[ 4] = MULH(2*(in1[2*5] + in1[2*7] - in1[2*1]), -C3);
        t2 = MULH(2*(in1[2*1] + in1[2*5]),    C1);
        t3 = MULH(   in1[2*5] - in1[2*7] , -2*C7);
        t0 = MULH(2*in1[2*3], C3);

        t1 = MULH(2*(in1[2*1] + in1[2*7]),   -C5);

        tmp1[ 0] = t2 + t3 + t0;
        tmp1[12] = t2 + t1 - t0;
        tmp1[ 8] = t3 - t1 - t0;
#endif
    }

    i = 0;
    for(j=0;j<4;j++) {
        t0 = tmp[i];
        t1 = tmp[i + 2];
        s0 = t1 + t0;
        s2 = t1 - t0;

        t2 = tmp[i + 1];
        t3 = tmp[i + 3];
        s1 = MULH(2*(t3 + t2), icos36h[j]);
        s3 = MULL(t3 - t2, icos36[8 - j]);

        t0 = s0 + s1;
        t1 = s0 - s1;
        out[(9 + j)*SBLIMIT] =  MULH(t1, win[9 + j]) + buf[9 + j];
        out[(8 - j)*SBLIMIT] =  MULH(t1, win[8 - j]) + buf[8 - j];
        buf[9 + j] = MULH(t0, win[18 + 9 + j]);
        buf[8 - j] = MULH(t0, win[18 + 8 - j]);

        t0 = s2 + s3;
        t1 = s2 - s3;
        out[(9 + 8 - j)*SBLIMIT] =  MULH(t1, win[9 + 8 - j]) + buf[9 + 8 - j];
        out[(        j)*SBLIMIT] =  MULH(t1, win[        j]) + buf[        j];
        buf[9 + 8 - j] = MULH(t0, win[18 + 9 + 8 - j]);
        buf[      + j] = MULH(t0, win[18         + j]);
        i += 4;
    }

    s0 = tmp[16];
    s1 = MULH(2*tmp[17], icos36h[4]);
    t0 = s0 + s1;
    t1 = s0 - s1;
    out[(9 + 4)*SBLIMIT] =  MULH(t1, win[9 + 4]) + buf[9 + 4];
    out[(8 - 4)*SBLIMIT] =  MULH(t1, win[8 - 4]) + buf[8 - 4];
    buf[9 + 4] = MULH(t0, win[18 + 9 + 4]);
    buf[8 - 4] = MULH(t0, win[18 + 8 - 4]);
}

/* return the number of decoded frames */
static int mp_decode_layer1(MPADecodeContext *s)
{
    int bound, i, v, n, ch, j, mant;
    uint8_t allocation[MPA_MAX_CHANNELS][SBLIMIT];
    uint8_t scale_factors[MPA_MAX_CHANNELS][SBLIMIT];

    if (s->mode == MPA_JSTEREO)
        bound = (s->mode_ext + 1) * 4;
    else
        bound = SBLIMIT;

    /* allocation bits */
    for(i=0;i<bound;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            allocation[ch][i] = get_bits(&s->gb, 4);
        }
    }
    for(i=bound;i<SBLIMIT;i++) {
        allocation[0][i] = get_bits(&s->gb, 4);
    }

    /* scale factors */
    for(i=0;i<bound;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (allocation[ch][i])
                scale_factors[ch][i] = get_bits(&s->gb, 6);
        }
    }
    for(i=bound;i<SBLIMIT;i++) {
        if (allocation[0][i]) {
            scale_factors[0][i] = get_bits(&s->gb, 6);
            scale_factors[1][i] = get_bits(&s->gb, 6);
        }
    }

    /* compute samples */
    for(j=0;j<12;j++) {
        for(i=0;i<bound;i++) {
            for(ch=0;ch<s->nb_channels;ch++) {
                n = allocation[ch][i];
                if (n) {
                    mant = get_bits(&s->gb, n + 1);
                    v = l1_unscale(n, mant, scale_factors[ch][i]);
                } else {
                    v = 0;
                }
                s->sb_samples[ch][j][i] = v;
            }
        }
        for(i=bound;i<SBLIMIT;i++) {
            n = allocation[0][i];
            if (n) {
                mant = get_bits(&s->gb, n + 1);
                v = l1_unscale(n, mant, scale_factors[0][i]);
                s->sb_samples[0][j][i] = v;
                v = l1_unscale(n, mant, scale_factors[1][i]);
                s->sb_samples[1][j][i] = v;
            } else {
                s->sb_samples[0][j][i] = 0;
                s->sb_samples[1][j][i] = 0;
            }
        }
    }
    return 12;
}

static int mp_decode_layer2(MPADecodeContext *s)
{
    int sblimit; /* number of used subbands */
    const unsigned char *alloc_table;
    int table, bit_alloc_bits, i, j, ch, bound, v;
    unsigned char bit_alloc[MPA_MAX_CHANNELS][SBLIMIT];
    unsigned char scale_code[MPA_MAX_CHANNELS][SBLIMIT];
    unsigned char scale_factors[MPA_MAX_CHANNELS][SBLIMIT][3], *sf;
    int scale, qindex, bits, steps, k, l, m, b;

    /* select decoding table */
    table = ff_mpa_l2_select_table(s->bit_rate / 1000, s->nb_channels,
                            s->sample_rate, s->lsf);
    sblimit = ff_mpa_sblimit_table[table];
    alloc_table = ff_mpa_alloc_tables[table];

    if (s->mode == MPA_JSTEREO)
        bound = (s->mode_ext + 1) * 4;
    else
        bound = sblimit;

    dprintf(s->avctx, "bound=%d sblimit=%d\n", bound, sblimit);

    /* sanity check */
    if( bound > sblimit ) bound = sblimit;

    /* parse bit allocation */
    j = 0;
    for(i=0;i<bound;i++) {
        bit_alloc_bits = alloc_table[j];
        for(ch=0;ch<s->nb_channels;ch++) {
            bit_alloc[ch][i] = get_bits(&s->gb, bit_alloc_bits);
        }
        j += 1 << bit_alloc_bits;
    }
    for(i=bound;i<sblimit;i++) {
        bit_alloc_bits = alloc_table[j];
        v = get_bits(&s->gb, bit_alloc_bits);
        bit_alloc[0][i] = v;
        bit_alloc[1][i] = v;
        j += 1 << bit_alloc_bits;
    }

#ifdef DEBUG
    {
        for(ch=0;ch<s->nb_channels;ch++) {
            for(i=0;i<sblimit;i++)
                dprintf(s->avctx, " %d", bit_alloc[ch][i]);
            dprintf(s->avctx, "\n");
        }
    }
#endif

    /* scale codes */
    for(i=0;i<sblimit;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (bit_alloc[ch][i])
                scale_code[ch][i] = get_bits(&s->gb, 2);
        }
    }

    /* scale factors */
    for(i=0;i<sblimit;i++) {
        for(ch=0;ch<s->nb_channels;ch++) {
            if (bit_alloc[ch][i]) {
                sf = scale_factors[ch][i];
                switch(scale_code[ch][i]) {
                default:
                case 0:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[1] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    break;
                case 2:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[1] = sf[0];
                    sf[2] = sf[0];
                    break;
                case 1:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    sf[1] = sf[0];
                    break;
                case 3:
                    sf[0] = get_bits(&s->gb, 6);
                    sf[2] = get_bits(&s->gb, 6);
                    sf[1] = sf[2];
                    break;
                }
            }
        }
    }

#ifdef DEBUG
    for(ch=0;ch<s->nb_channels;ch++) {
        for(i=0;i<sblimit;i++) {
            if (bit_alloc[ch][i]) {
                sf = scale_factors[ch][i];
                dprintf(s->avctx, " %d %d %d", sf[0], sf[1], sf[2]);
            } else {
                dprintf(s->avctx, " -");
            }
        }
        dprintf(s->avctx, "\n");
    }
#endif

    /* samples */