Michael Cheng wrote:
> On Mon, 26 Oct 1998, Rafael W. Luebbert wrote:
>
> > > First question: NDA with whom? FhG?
> > It is not with FhG, but I can't say more than that, sorry to be vague.
> So you wouldn't be able to directly share with us coding tips or source
> code? Technical advice only?
> I assume you didn't get a 'cease and desist' letter from FhG then?
I fully intend to license a working, compliant coder through FhG when I have
reached that level of stability, but, no, I have not received a "cease and desist"
letter from FhG.
I do not mind helping, and will help in coding and technical advice as I can.
>
>
> Where did you get the technical knowledge to do this? are you a
> compsci/dsp person? or are you just very good at reading the cryptic ISO
> documents.
I am not in the computer field at all. This is kind of a long story, but basically
I discovered MPEG on the net, then went and bought the standard, then
went and read everything else I could get my hands on, and started hacking.
David Renelt should have a nice interview article on his web site soon, which
describes all the ?boring details... ;)
>
> ugh.
> Admittedly, I've never looked at the acuracy of the routines in comparison
> to the standard. I'd sort of assumed they were right, and spent time just
> speeding them up.
> The code is pretty fast now and the focus should shift towards improving
> the quality.
>
> > Adding the joint stereo routines would probably yield the best return
> > on your "coding" investment, especially MS stereo. I have "played"
> cool. Get out your ISO documents folks, it's time to start reading :)
>
> later
> mike
I don't know about the Intel side, but on the PowerPC side, there was not a horribly
high amount of computational time spent in the FFT subroutines. What percentages
of overall CPU time are you seeing spent in FFT()?
In my early testing, I saw the window_subband() and the filter_subband() sucking up
much more than the FFT (or the MDCT subroutines).
More later,
Rafael
