annotate rdft.c @ 11032:01bd040f8607 libavcodec

Unroll main loop so the edge==0 case is seperate. This allows many things to be simplified away. h264 decoder is overall 1% faster with a mbaff sample and 0.1% slower with the cathedral sample, probably because the slow loop filter code must be loaded into the code cache for each first MB of each row but isnt used for the following MBs.

/*
 * (I)RDFT transforms
 * Copyright (c) 2009 Alex Converse <alex dot converse at gmail dot com>
 *
 * 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
 */
#include <math.h>
#include "dsputil.h"

/**
 * @file libavcodec/rdft.c
 * (Inverse) Real Discrete Fourier Transforms.
 */

/* sin(2*pi*x/n) for 0<=x<n/4, followed by n/2<=x<3n/4 */
#if !CONFIG_HARDCODED_TABLES
SINTABLE(16);
SINTABLE(32);
SINTABLE(64);
SINTABLE(128);
SINTABLE(256);
SINTABLE(512);
SINTABLE(1024);
SINTABLE(2048);
SINTABLE(4096);
SINTABLE(8192);
SINTABLE(16384);
SINTABLE(32768);
SINTABLE(65536);
#endif
SINTABLE_CONST FFTSample * const ff_sin_tabs[] = {
    NULL, NULL, NULL, NULL,
    ff_sin_16, ff_sin_32, ff_sin_64, ff_sin_128, ff_sin_256, ff_sin_512, ff_sin_1024,
    ff_sin_2048, ff_sin_4096, ff_sin_8192, ff_sin_16384, ff_sin_32768, ff_sin_65536,
};

av_cold int ff_rdft_init(RDFTContext *s, int nbits, enum RDFTransformType trans)
{
    int n = 1 << nbits;
    int i;
    const double theta = (trans == RDFT || trans == IRIDFT ? -1 : 1)*2*M_PI/n;

    s->nbits           = nbits;
    s->inverse         = trans == IRDFT || trans == IRIDFT;
    s->sign_convention = trans == RIDFT || trans == IRIDFT ? 1 : -1;

    if (nbits < 4 || nbits > 16)
        return -1;

    if (ff_fft_init(&s->fft, nbits-1, trans == IRDFT || trans == RIDFT) < 0)
        return -1;

    ff_init_ff_cos_tabs(nbits);
    s->tcos = ff_cos_tabs[nbits];
    s->tsin = ff_sin_tabs[nbits]+(trans == RDFT || trans == IRIDFT)*(n>>2);
#if !CONFIG_HARDCODED_TABLES
    for (i = 0; i < (n>>2); i++) {
        s->tsin[i] = sin(i*theta);
    }
#endif
    return 0;
}

/** Map one real FFT into two parallel real even and odd FFTs. Then interleave
 * the two real FFTs into one complex FFT. Unmangle the results.
 * ref: http://www.engineeringproductivitytools.com/stuff/T0001/PT10.HTM
 */
void ff_rdft_calc_c(RDFTContext* s, FFTSample* data)
{
    int i, i1, i2;
    FFTComplex ev, od;
    const int n = 1 << s->nbits;
    const float k1 = 0.5;
    const float k2 = 0.5 - s->inverse;
    const FFTSample *tcos = s->tcos;
    const FFTSample *tsin = s->tsin;

    if (!s->inverse) {
        ff_fft_permute(&s->fft, (FFTComplex*)data);
        ff_fft_calc(&s->fft, (FFTComplex*)data);
    }
    /* i=0 is a special case because of packing, the DC term is real, so we
       are going to throw the N/2 term (also real) in with it. */
    ev.re = data[0];
    data[0] = ev.re+data[1];
    data[1] = ev.re-data[1];
    for (i = 1; i < (n>>2); i++) {
        i1 = 2*i;
        i2 = n-i1;
        /* Separate even and odd FFTs */
        ev.re =  k1*(data[i1  ]+data[i2  ]);
        od.im = -k2*(data[i1  ]-data[i2  ]);
        ev.im =  k1*(data[i1+1]-data[i2+1]);
        od.re =  k2*(data[i1+1]+data[i2+1]);
        /* Apply twiddle factors to the odd FFT and add to the even FFT */
        data[i1  ] =  ev.re + od.re*tcos[i] - od.im*tsin[i];
        data[i1+1] =  ev.im + od.im*tcos[i] + od.re*tsin[i];
        data[i2  ] =  ev.re - od.re*tcos[i] + od.im*tsin[i];
        data[i2+1] = -ev.im + od.im*tcos[i] + od.re*tsin[i];
    }
    data[2*i+1]=s->sign_convention*data[2*i+1];
    if (s->inverse) {
        data[0] *= k1;
        data[1] *= k1;
        ff_fft_permute(&s->fft, (FFTComplex*)data);
        ff_fft_calc(&s->fft, (FFTComplex*)data);
    }
}

void ff_rdft_calc(RDFTContext *s, FFTSample *data)
{
    ff_rdft_calc_c(s, data);
}

av_cold void ff_rdft_end(RDFTContext *s)
{
    ff_fft_end(&s->fft);
}