I don't have a time to look the whole source, but I mention one point...
>>>>> "M" == Mathew Hendry <[EMAIL PROTECTED]> writes:
>> - pentium optimized fft - djbfft
M> http://pobox.com/~djb/djbfft.html
>> (gives out-of-order output,that i'm not too sure how to grok)
M> A header is supplied to aid in grokking. We could always
M> reorder it (more lookup tables :). One other thing: djbfft
M> overwrites the input data. Does this matter for LAME?
LAME's l3psy routine uses FFT the way like this
(1) buffering the sampling data to sync MDCT and PSYCHO model.
(2) store (buffered data) * (window function) into work(wsamp_r)
(3) FFT on the work memory.
LAME's FFT also overwrites the input.
I thought (1) could be removed and the iteration loop of (2) could be
integrate into 1st loop of FFT. I did it :)
Could someone examine this ?
http://www.isoternet.org/~tominaga/lame-beta/lame-1017-4.tar.bz2
---
Takehiro TOMINAGA // May the source be with you!
--
MP3 ENCODER mailing list ( http://geek.rcc.se/mp3encoder/ )
