Rockbox mail archiveSubject: Re: Fwd: iRiver Codec - RT encoding and Fixed point library
Re: Fwd: iRiver Codec - RT encoding and Fixed point library
Date: Wed, 13 Jul 2005 21:32:29 +0100 (BST)
There may well be some sort of fixed point instructions in certain
versions of the coldfire. However, not on the CF5249 that H1xx/H3xx use,
(other than the more simple fractional mode of the EMAC unit, which we are
already using in all of our codecs).
If there is a fixed-point version of Shine MP3 encoder already optimised
for ARM then that is probably a good starting point. I'm not sure of the
merits of 'emulating' floating point using a fixed-point CPU, however (it
will either require major rewrites of the code or would just run
suboptimally - in either case, rewriting to use fixed-point data would
probably be a necessity)
> I follow the rockbox porting effort to iriver from the very beginning
> and I'm quite impressed by the work done so far.
> My freetime is limited and maybe I won't be able to write any line of
> C (but who knows), but I was curious about the "Real Time encoding"
> There are comments in the wiki about ShineMP3 encoder and fixed point.
> Is there anybody here interested by the subject?
> Quote from the wiki:
> What about a fixed point library ? Can't we use a floating point
> algorithm with a fixed point library ? Furthermore, it seems that
> there is some sort of fixed point instructions in certain versions of
> the coldfire.
> Does someone here know if we could use floating point code with a
> fixed point (mainly inlined) library ? Is lame too complicated for
> this use ? (maybe it is).
> Furthermore, such a library could be used by other modules of rockbox...
> What do you think about this ?
Received on 2005-07-13