comparison dca.c @ 11617:bb17732c00ef libavcodec

DCA: break out lfe_interpolation_fir() inner loops to a function This enables SIMD optimisations of this function.

11616:1461e6044153

#include "put_bits.h"
#include "dcadata.h"
#include "dcahuff.h"
#include "dca.h"
#include "synth_filter.h"

//#define TRACE

#define DCA_PRIM_CHANNELS_MAX (5)
#define DCA_SUBBANDS (32)

    int debug_flag;             ///< used for suppressing repeated error messages output
    DSPContext dsp;
    FFTContext imdct;
    SynthFilterContext synth;
} DCAContext;

static const uint16_t dca_vlc_offs[] = {
        0,   512,   640,   768,  1282,  1794,  2436,  3080,  3770,  4454,  5364,
     5372,  5380,  5388,  5392,  5396,  5412,  5420,  5428,  5460,  5492,  5508,

        samples_out+= 32;

    }
}

static void lfe_interpolation_fir(int decimation_select,
                                  int num_deci_sample, float *samples_in,
                                  float *samples_out, float scale,
                                  float bias)
{
    /* samples_in: An array holding decimated samples.

     *   from last subframe as history.
     *
     * samples_out: An array holding interpolated samples
     */

    int decifactor, k, j;
    const float *prCoeff;
    int deciindex;

    /* Select decimation filter */
    if (decimation_select == 1) {

        decifactor = 32;
        prCoeff = lfe_fir_64;
    }
    /* Interpolation */
    for (deciindex = 0; deciindex < num_deci_sample; deciindex++) {
        float *samples_out2 = samples_out + decifactor;
        const float *cf0 = prCoeff;
        const float *cf1 = prCoeff + 256;

        /* One decimated sample generates 2*decifactor interpolated ones */
        for (k = 0; k < decifactor; k++) {
            float v0 = 0.0;
            float v1 = 0.0;
            for (j = 0; j < 256 / decifactor; j++) {
                float s = samples_in[-j];
                v0 += s * *cf0++;
                v1 += s * *--cf1;
            }
            *samples_out++  = (v0 * scale) + bias;
            *samples_out2++ = (v1 * scale) + bias;
        }

        samples_in++;
        samples_out += decifactor;
    }
}

/* downmixing routines */
#define MIX_REAR1(samples, si1, rs, coef) \

    /* Generate LFE samples for this subsubframe FIXME!!! */
    if (s->output & DCA_LFE) {
        int lfe_samples = 2 * s->lfe * s->subsubframes;

        lfe_interpolation_fir(s->lfe, 2 * s->lfe,
                              s->lfe_data + lfe_samples +
                              2 * s->lfe * subsubframe,
                              &s->samples[256 * dca_lfe_index[s->amode]],
                              (1.0/256.0)*s->scale_bias,  s->add_bias);
        /* Outputs 20bits pcm samples */

    dca_init_vlcs();

    dsputil_init(&s->dsp, avctx);
    ff_mdct_init(&s->imdct, 6, 1, 1.0);
    ff_synth_filter_init(&s->synth);

    for(i = 0; i < 6; i++)
        s->samples_chanptr[i] = s->samples + i * 256;
    avctx->sample_fmt = SAMPLE_FMT_S16;

11617:bb17732c00ef

#include "put_bits.h"
#include "dcadata.h"
#include "dcahuff.h"
#include "dca.h"
#include "synth_filter.h"
#include "dcadsp.h"

//#define TRACE

#define DCA_PRIM_CHANNELS_MAX (5)
#define DCA_SUBBANDS (32)

    int debug_flag;             ///< used for suppressing repeated error messages output
    DSPContext dsp;
    FFTContext imdct;
    SynthFilterContext synth;
    DCADSPContext dcadsp;
} DCAContext;

static const uint16_t dca_vlc_offs[] = {
        0,   512,   640,   768,  1282,  1794,  2436,  3080,  3770,  4454,  5364,
     5372,  5380,  5388,  5392,  5396,  5412,  5420,  5428,  5460,  5492,  5508,

        samples_out+= 32;

    }
}

static void lfe_interpolation_fir(DCAContext *s, int decimation_select,
                                  int num_deci_sample, float *samples_in,
                                  float *samples_out, float scale,
                                  float bias)
{
    /* samples_in: An array holding decimated samples.

     *   from last subframe as history.
     *
     * samples_out: An array holding interpolated samples
     */

    int decifactor;
    const float *prCoeff;
    int deciindex;

    /* Select decimation filter */
    if (decimation_select == 1) {

        decifactor = 32;
        prCoeff = lfe_fir_64;
    }
    /* Interpolation */
    for (deciindex = 0; deciindex < num_deci_sample; deciindex++) {
        s->dcadsp.lfe_fir(samples_out, samples_in, prCoeff, decifactor,
                          scale, bias);
        samples_in++;
        samples_out += 2 * decifactor;
    }
}

/* downmixing routines */
#define MIX_REAR1(samples, si1, rs, coef) \

    /* Generate LFE samples for this subsubframe FIXME!!! */
    if (s->output & DCA_LFE) {
        int lfe_samples = 2 * s->lfe * s->subsubframes;

        lfe_interpolation_fir(s, s->lfe, 2 * s->lfe,
                              s->lfe_data + lfe_samples +
                              2 * s->lfe * subsubframe,
                              &s->samples[256 * dca_lfe_index[s->amode]],
                              (1.0/256.0)*s->scale_bias,  s->add_bias);
        /* Outputs 20bits pcm samples */

    dca_init_vlcs();

    dsputil_init(&s->dsp, avctx);
    ff_mdct_init(&s->imdct, 6, 1, 1.0);
    ff_synth_filter_init(&s->synth);
    ff_dcadsp_init(&s->dcadsp);

    for(i = 0; i < 6; i++)
        s->samples_chanptr[i] = s->samples + i * 256;
    avctx->sample_fmt = SAMPLE_FMT_S16;