specrec.c

mpeg4 video codec mpeg4 video codec

*/
static void quant_to_spec(ic_stream *ics, real_t *spec_data, uint16_t frame_len)
{
    uint8_t g, sfb, win;
    uint16_t width, bin, k, gindex;

    ALIGN real_t tmp_spec[1024] = {0};

    k = 0;
    gindex = 0;

    for (g = 0; g < ics->num_window_groups; g++)
    {
        uint16_t j = 0;
        uint16_t gincrease = 0;
        uint16_t win_inc = ics->swb_offset[ics->num_swb];

        for (sfb = 0; sfb < ics->num_swb; sfb++)
        {
            width = ics->swb_offset[sfb+1] - ics->swb_offset[sfb];

            for (win = 0; win < ics->window_group_length[g]; win++)
            {
                for (bin = 0; bin < width; bin += 4)
                {
                    tmp_spec[gindex+(win*win_inc)+j+bin+0] = spec_data[k+0];
                    tmp_spec[gindex+(win*win_inc)+j+bin+1] = spec_data[k+1];
                    tmp_spec[gindex+(win*win_inc)+j+bin+2] = spec_data[k+2];
                    tmp_spec[gindex+(win*win_inc)+j+bin+3] = spec_data[k+3];
                    gincrease += 4;
                    k += 4;
                }
            }
            j += width;
        }
        gindex += gincrease;
    }

    memcpy(spec_data, tmp_spec, frame_len*sizeof(real_t));
}

static INLINE real_t iquant(int16_t q, const real_t *tab, uint8_t *error)
{
#ifdef FIXED_POINT
    static const real_t errcorr[] = {
        REAL_CONST(0), REAL_CONST(1.0/8.0), REAL_CONST(2.0/8.0), REAL_CONST(3.0/8.0),
        REAL_CONST(4.0/8.0),  REAL_CONST(5.0/8.0), REAL_CONST(6.0/8.0), REAL_CONST(7.0/8.0),
        REAL_CONST(0)
    };
    real_t x1, x2;
    int16_t sgn = 1;

    if (q < 0)
    {
        q = -q;
        sgn = -1;
    }

    if (q < IQ_TABLE_SIZE)
        return sgn * tab[q];

    /* linear interpolation */
    x1 = tab[q>>3];
    x2 = tab[(q>>3) + 1];
    return sgn * 16 * (MUL_R(errcorr[q&7],(x2-x1)) + x1);
#else
    if (q < 0)
    {
        /* tab contains a value for all possible q [0,8192] */
        if (-q < IQ_TABLE_SIZE)
            return -tab[-q];

        *error = 17;
        return 0;
    } else {
        /* tab contains a value for all possible q [0,8192] */
        if (q < IQ_TABLE_SIZE)
            return tab[q];

        *error = 17;
        return 0;
    }
#endif
}

static uint8_t inverse_quantization(real_t *x_invquant, const int16_t *x_quant, const uint16_t frame_len)
{
    int16_t i;
    uint8_t error = 0; /* Init error flag */
    const real_t *tab = iq_table;

    for (i = 0; i < frame_len; i+=4)
    {
        x_invquant[i] = iquant(x_quant[i], tab, &error);
        x_invquant[i+1] = iquant(x_quant[i+1], tab, &error);
        x_invquant[i+2] = iquant(x_quant[i+2], tab, &error);
        x_invquant[i+3] = iquant(x_quant[i+3], tab, &error);
    }

    return error;
}

#ifndef FIXED_POINT
ALIGN static const real_t pow2sf_tab[] = {
    2.9802322387695313E-008, 5.9604644775390625E-008, 1.1920928955078125E-007,
    2.384185791015625E-007, 4.76837158203125E-007, 9.5367431640625E-007,
    1.9073486328125E-006, 3.814697265625E-006, 7.62939453125E-006,
    1.52587890625E-005, 3.0517578125E-005, 6.103515625E-005,
    0.0001220703125, 0.000244140625, 0.00048828125,
    0.0009765625, 0.001953125, 0.00390625,
    0.0078125, 0.015625, 0.03125,
    0.0625, 0.125, 0.25,
    0.5, 1.0, 2.0,
    4.0, 8.0, 16.0, 32.0,
    64.0, 128.0, 256.0,
    512.0, 1024.0, 2048.0,
    4096.0, 8192.0, 16384.0,
    32768.0, 65536.0, 131072.0,
    262144.0, 524288.0, 1048576.0,
    2097152.0, 4194304.0, 8388608.0,
    16777216.0, 33554432.0, 67108864.0,
    134217728.0, 268435456.0, 536870912.0,
    1073741824.0, 2147483648.0, 4294967296.0,
    8589934592.0, 17179869184.0, 34359738368.0,
    68719476736.0, 137438953472.0, 274877906944.0
};
#endif

ALIGN static real_t pow2_table[] =
{
#if 0
    COEF_CONST(0.59460355750136053335874998528024), /* 2^-0.75 */
    COEF_CONST(0.70710678118654752440084436210485), /* 2^-0.5 */
    COEF_CONST(0.84089641525371454303112547623321), /* 2^-0.25 */
#endif
    COEF_CONST(1.0),
    COEF_CONST(1.1892071150027210667174999705605), /* 2^0.25 */
    COEF_CONST(1.4142135623730950488016887242097), /* 2^0.5 */
    COEF_CONST(1.6817928305074290860622509524664) /* 2^0.75 */
};

void apply_scalefactors(faacDecHandle hDecoder, ic_stream *ics,
                        real_t *x_invquant, uint16_t frame_len)
{
    uint8_t g, sfb;
    uint16_t top;
    int32_t exp, frac;
    uint8_t groups = 0;
    uint16_t nshort = frame_len/8;

    for (g = 0; g < ics->num_window_groups; g++)
    {
        uint16_t k = 0;

        /* using this nshort*groups doesn't hurt long blocks, because
           long blocks only have 1 group, so that means 'groups' is
           always 0 for long blocks
        */
        for (sfb = 0; sfb < ics->max_sfb; sfb++)
        {
            top = ics->sect_sfb_offset[g][sfb+1];

            /* this could be scalefactor for IS or PNS, those can be negative or bigger then 255 */
            /* just ignore them */
            if (ics->scale_factors[g][sfb] < 0 || ics->scale_factors[g][sfb] > 255)
            {
                exp = 0;
                frac = 0;
            } else {
                /* ics->scale_factors[g][sfb] must be between 0 and 255 */
                exp = (ics->scale_factors[g][sfb] /* - 100 */) >> 2;
                frac = (ics->scale_factors[g][sfb] /* - 100 */) & 3;
            }

#ifdef FIXED_POINT
            exp -= 25;
            /* IMDCT pre-scaling */
            if (hDecoder->object_type == LD)
            {
                exp -= 6 /*9*/;
            } else {
                if (ics->window_sequence == EIGHT_SHORT_SEQUENCE)
                    exp -= 4 /*7*/;
                else
                    exp -= 7 /*10*/;
            }
#endif

            /* minimum size of a sf band is 4 and always a multiple of 4 */
            for ( ; k < top; k += 4)
            {
#ifdef FIXED_POINT
                if (exp < 0)
                {
                    x_invquant[k+(groups*nshort)] >>= -exp;
                    x_invquant[k+(groups*nshort)+1] >>= -exp;
                    x_invquant[k+(groups*nshort)+2] >>= -exp;
                    x_invquant[k+(groups*nshort)+3] >>= -exp;
                } else {
                    x_invquant[k+(groups*nshort)] <<= exp;
                    x_invquant[k+(groups*nshort)+1] <<= exp;
                    x_invquant[k+(groups*nshort)+2] <<= exp;
                    x_invquant[k+(groups*nshort)+3] <<= exp;
                }
#else
                x_invquant[k+(groups*nshort)]   = x_invquant[k+(groups*nshort)]   * pow2sf_tab[exp/*+25*/];
                x_invquant[k+(groups*nshort)+1] = x_invquant[k+(groups*nshort)+1] * pow2sf_tab[exp/*+25*/];
                x_invquant[k+(groups*nshort)+2] = x_invquant[k+(groups*nshort)+2] * pow2sf_tab[exp/*+25*/];
                x_invquant[k+(groups*nshort)+3] = x_invquant[k+(groups*nshort)+3] * pow2sf_tab[exp/*+25*/];
#endif

                x_invquant[k+(groups*nshort)]   = MUL_C(x_invquant[k+(groups*nshort)],pow2_table[frac /* + 3*/]);
                x_invquant[k+(groups*nshort)+1] = MUL_C(x_invquant[k+(groups*nshort)+1],pow2_table[frac /* + 3*/]);
                x_invquant[k+(groups*nshort)+2] = MUL_C(x_invquant[k+(groups*nshort)+2],pow2_table[frac /* + 3*/]);
                x_invquant[k+(groups*nshort)+3] = MUL_C(x_invquant[k+(groups*nshort)+3],pow2_table[frac /* + 3*/]);
            }
        }
        groups += ics->window_group_length[g];
    }
}

#ifdef USE_SSE
void apply_scalefactors_sse(faacDecHandle hDecoder, ic_stream *ics,
                            real_t *x_invquant, uint16_t frame_len)
{
    uint8_t g, sfb;
    uint16_t top;
    int32_t exp, frac;
    uint8_t groups = 0;
    uint16_t nshort = frame_len/8;

    for (g = 0; g < ics->num_window_groups; g++)
    {
        uint16_t k = 0;

        /* using this nshort*groups doesn't hurt long blocks, because
           long blocks only have 1 group, so that means 'groups' is
           always 0 for long blocks
        */
        for (sfb = 0; sfb < ics->max_sfb; sfb++)
        {
            top = ics->sect_sfb_offset[g][sfb+1];

            exp = (ics->scale_factors[g][sfb] /* - 100 */) >> 2;
            frac = (ics->scale_factors[g][sfb] /* - 100 */) & 3;

            /* minimum size of a sf band is 4 and always a multiple of 4 */
            for ( ; k < top; k += 4)
            {
                __m128 m1 = _mm_load_ps(&x_invquant[k+(groups*nshort)]);
                __m128 m2 = _mm_load_ps1(&pow2sf_tab[exp /*+25*/]);
                __m128 m3 = _mm_load_ps1(&pow2_table[frac /* + 3*/]);
                __m128 m4 = _mm_mul_ps(m1, m2);
                __m128 m5 = _mm_mul_ps(m3, m4);
                _mm_store_ps(&x_invquant[k+(groups*nshort)], m5);
            }
        }
        groups += ics->window_group_length[g];
    }
}
#endif

static uint8_t allocate_single_channel(faacDecHandle hDecoder, uint8_t channel,
                                       uint8_t output_channels)
{
    uint8_t mul = 1;

#ifdef MAIN_DEC
    /* MAIN object type prediction */
    if (hDecoder->object_type == MAIN)
    {
        /* allocate the state only when needed */
        if (hDecoder->pred_stat[channel] == NULL)
        {
            hDecoder->pred_stat[channel] = (pred_state*)faad_malloc(hDecoder->frameLength * sizeof(pred_state));
            reset_all_predictors(hDecoder->pred_stat[channel], hDecoder->frameLength);
        }
    }
#endif

#ifdef LTP_DEC
    if (is_ltp_ot(hDecoder->object_type))
    {
        /* allocate the state only when needed */
        if (hDecoder->lt_pred_stat[channel] == NULL)
        {
            hDecoder->lt_pred_stat[channel] = (int16_t*)faad_malloc(hDecoder->frameLength*4 * sizeof(int16_t));
            memset(hDecoder->lt_pred_stat[channel], 0, hDecoder->frameLength*4 * sizeof(int16_t));
        }
    }
#endif

    if (hDecoder->time_out[channel] == NULL)
    {
        mul = 1;
#ifdef SBR_DEC
        hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 0;
        if ((hDecoder->sbr_present_flag == 1) || (hDecoder->forceUpSampling == 1))
        {
            /* SBR requires 2 times as much output data */
            mul = 2;
            hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 1;
        }
#endif
        hDecoder->time_out[channel] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
        memset(hDecoder->time_out[channel], 0, mul*hDecoder->frameLength*sizeof(real_t));
    }
#if (defined(PS_DEC) || defined(DRM_PS))
    if (output_channels == 2)
    {
        if (hDecoder->time_out[channel+1] == NULL)
        {
            hDecoder->time_out[channel+1] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
            memset(hDecoder->time_out[channel+1], 0, mul*hDecoder->frameLength*sizeof(real_t));
        }
    }
#endif

    if (hDecoder->fb_intermed[channel] == NULL)
    {
        hDecoder->fb_intermed[channel] = (real_t*)faad_malloc(hDecoder->frameLength*sizeof(real_t));
        memset(hDecoder->fb_intermed[channel], 0, hDecoder->frameLength*sizeof(real_t));
    }

#ifdef SSR_DEC
    if (hDecoder->object_type == SSR)
    {
        if (hDecoder->ssr_overlap[channel] == NULL)
        {
            hDecoder->ssr_overlap[channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
            memset(hDecoder->ssr_overlap[channel], 0, 2*hDecoder->frameLength*sizeof(real_t));
        }
        if (hDecoder->prev_fmd[channel] == NULL)
        {
            uint16_t k;
            hDecoder->prev_fmd[channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
            for (k = 0; k < 2*hDecoder->frameLength; k++)
                hDecoder->prev_fmd[channel][k] = REAL_CONST(-1);
        }
    }
#endif

    return 0;
}

static uint8_t allocate_channel_pair(faacDecHandle hDecoder,
                                     uint8_t channel, uint8_t paired_channel)
{
    uint8_t mul = 1;

#ifdef MAIN_DEC
    /* MAIN object type prediction */
    if (hDecoder->object_type == MAIN)
    {
        /* allocate the state only when needed */
        if (hDecoder->pred_stat[channel] == NULL)
        {
            hDecoder->pred_stat[channel] = (pred_state*)faad_malloc(hDecoder->frameLength * sizeof(pred_state));
            reset_all_predictors(hDecoder->pred_stat[channel], hDecoder->frameLength);
        }
        if (hDecoder->pred_stat[paired_channel] == NULL)
        {
            hDecoder->pred_stat[paired_channel] = (pred_state*)faad_malloc(hDecoder->frameLength * sizeof(pred_state));
            reset_all_predictors(hDecoder->pred_stat[paired_channel], hDecoder->frameLength);
        }
    }
#endif

#ifdef LTP_DEC
    if (is_ltp_ot(hDecoder->object_type))
    {
        /* allocate the state only when needed */
        if (hDecoder->lt_pred_stat[channel] == NULL)
        {
            hDecoder->lt_pred_stat[channel] = (int16_t*)faad_malloc(hDecoder->frameLength*4 * sizeof(int16_t));
            memset(hDecoder->lt_pred_stat[channel], 0, hDecoder->frameLength*4 * sizeof(int16_t));
        }
        if (hDecoder->lt_pred_stat[paired_channel] == NULL)
        {
            hDecoder->lt_pred_stat[paired_channel] = (int16_t*)faad_malloc(hDecoder->frameLength*4 * sizeof(int16_t));
            memset(hDecoder->lt_pred_stat[paired_channel], 0, hDecoder->frameLength*4 * sizeof(int16_t));
        }
    }
#endif

    if (hDecoder->time_out[channel] == NULL)
    {
        mul = 1;
#ifdef SBR_DEC
        hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 0;
        if ((hDecoder->sbr_present_flag == 1) || (hDecoder->forceUpSampling == 1))
        {
            /* SBR requires 2 times as much output data */
            mul = 2;
            hDecoder->sbr_alloced[hDecoder->fr_ch_ele] = 1;
        }
#endif
        hDecoder->time_out[channel] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
        memset(hDecoder->time_out[channel], 0, mul*hDecoder->frameLength*sizeof(real_t));
    }
    if (hDecoder->time_out[paired_channel] == NULL)
    {
        hDecoder->time_out[paired_channel] = (real_t*)faad_malloc(mul*hDecoder->frameLength*sizeof(real_t));
        memset(hDecoder->time_out[paired_channel], 0, mul*hDecoder->frameLength*sizeof(real_t));
    }

    if (hDecoder->fb_intermed[channel] == NULL)
    {
        hDecoder->fb_intermed[channel] = (real_t*)faad_malloc(hDecoder->frameLength*sizeof(real_t));
        memset(hDecoder->fb_intermed[channel], 0, hDecoder->frameLength*sizeof(real_t));
    }
    if (hDecoder->fb_intermed[paired_channel] == NULL)
    {
        hDecoder->fb_intermed[paired_channel] = (real_t*)faad_malloc(hDecoder->frameLength*sizeof(real_t));
        memset(hDecoder->fb_intermed[paired_channel], 0, hDecoder->frameLength*sizeof(real_t));
    }

#ifdef SSR_DEC
    if (hDecoder->object_type == SSR)
    {
        if (hDecoder->ssr_overlap[cpe->channel] == NULL)
        {
            hDecoder->ssr_overlap[cpe->channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));
            memset(hDecoder->ssr_overlap[cpe->channel], 0, 2*hDecoder->frameLength*sizeof(real_t));
        }
        if (hDecoder->ssr_overlap[cpe->paired_channel] == NULL)
        {
            hDecoder->ssr_overlap[cpe->paired_channel] = (real_t*)faad_malloc(2*hDecoder->frameLength*sizeof(real_t));