is there trivial way to change floats to fixed points?

Evgeniy Ginzburg nad.oby at gmail.com
Tue May 12 16:46:19 CEST 2009


On Tue, May 12, 2009 at 5:27 PM, Visti Andresen <talpa at talpa.dk> wrote:
> On Tue, 12 May 2009 13:28:06 +0300
> Aapo Rantalainen <aapo.rantalainen at gmail.com> wrote:
>
>> Is there trivial (=automatic) way to change floating point arithmetic
>> to fixed point arithmetic?
>
> Not that I'm aware of...
> Floating and fixed point numbers have different limitations and
> therefor not all software would work with fixed point.
>
> But it "should be possible" (not trivial) to extend gcc to allow
> faking floats using fixed point?
>
>>
>> I have some compilable and runnable code, but it is too slow on
>> Freerunner and I think it is because of used floats. It runs on Nokia
>> Internet Tablet very well and I think it has almost same specks than
>> Freerunner expect it has FPU.
>>
>> So is there some other way than reading and learning whole source code
>> and make changes? (Maybe some compile options to use fixed points
>> instead of floats?)
>>
AFAIR one of developers (GSoC) used double precision fixed point
arithmetic for voice recognition
code might be public so it will be nice hint for you.
Here it is
http://lists.openmoko.org/pipermail/openmoko-devel/2008-June/003153.html
-- 
So long, and thanks for all the fish.



More information about the devel mailing list