comparison fft.c @ 10153:7a63015e4627 libavcodec

ARM: NEON optimised FFT and MDCT Vorbis and AC3 ~3x faster. Parts by Naotoshi Nojiri, naonoj gmail

10152:ed85bbd5dccb

{
    int i, j, m, n;
    float alpha, c1, s1, s2;
    int split_radix = 1;
    int av_unused has_vectors;

    if (nbits < 2 || nbits > 16)
        goto fail;
    s->nbits = nbits;
    n = 1 << nbits;

    has_vectors = mm_support();
    if (has_vectors & FF_MM_ALTIVEC) {
        s->fft_calc = ff_fft_calc_altivec;
        split_radix = 0;
    }
#endif

    if (split_radix) {
        for(j=4; j<=nbits; j++) {
            int m = 1<<j;

                tab[i] = cos(i*freq);
            for(i=1; i<m/4; i++)
                tab[m/2-i] = tab[i];
        }
        for(i=0; i<n; i++)
            s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] = i;
        s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
    } else {
        int np, nblocks, np2, l;
        FFTComplex *q;

10153:7a63015e4627

{
    int i, j, m, n;
    float alpha, c1, s1, s2;
    int split_radix = 1;
    int av_unused has_vectors;
    int revtab_shift = 0;

    if (nbits < 2 || nbits > 16)
        goto fail;
    s->nbits = nbits;
    n = 1 << nbits;

    has_vectors = mm_support();
    if (has_vectors & FF_MM_ALTIVEC) {
        s->fft_calc = ff_fft_calc_altivec;
        split_radix = 0;
    }
#elif HAVE_NEON
    s->fft_permute = ff_fft_permute_neon;
    s->fft_calc    = ff_fft_calc_neon;
    s->imdct_calc  = ff_imdct_calc_neon;
    s->imdct_half  = ff_imdct_half_neon;
    revtab_shift = 3;
#endif

    if (split_radix) {
        for(j=4; j<=nbits; j++) {
            int m = 1<<j;

                tab[i] = cos(i*freq);
            for(i=1; i<m/4; i++)
                tab[m/2-i] = tab[i];
        }
        for(i=0; i<n; i++)
            s->revtab[-split_radix_permutation(i, n, s->inverse) & (n-1)] =
                i << revtab_shift;
        s->tmp_buf = av_malloc(n * sizeof(FFTComplex));
    } else {
        int np, nblocks, np2, l;
        FFTComplex *q;