[FFmpeg-devel] Information about FFT implementation in ffmpeg

[Vitor Sessak]

On Mon, Jul 25, 2011 at 10:30 AM, jogging song <joggingsong at gmail.com> wrote:
> Hi, all
>
> I hope to learn more about FFT implementation in ffmpeg.
> FFT in ffmpeg performs on float type data.
> When twiddle factors are generated, only one fourth of
> the length is calculated. Others are copied from generated
> twiddle factors and only cosine is generated.
> When the precision is considered, more twiddle factors
> are required.

This has nothing to do with precision.

> Such as in the function ff_init_ff_cos_tabs
> for(i=0; i<=m/4; i++)
>
>   tab[i] = cos(i*freq);
>
> When i > m /4, the cosine value calculated from the
> above equation is not equal to the value copied
> from the generated value. One or two significant
> digits are different.

I expect the sign to be different, not just a few significant digits.
In the C implementation only the values for i < m/4 are used. Higher
values are used only in ASM implementations to avoid shuffling.

> Now I try to generate twiddle factors of half of the length
> for both cosine value and sine value, and use these
> twiddle factors to calculate FFT. I need to modify the
> source code.

Why?

> Is there any materials about the FFT implementation in ffmpeg?

It is a pretty standard split-radix FFT implementation. Google for
"split radix FFT".

-Vitor