ts-7000
[Top] [All Lists]

[ts-7000] Re: What is needed to have floating point capabilities on 7250

To:
Subject: [ts-7000] Re: What is needed to have floating point capabilities on 7250?
From: "Ray" <>
Date: Sun, 29 Jul 2007 06:17:49 -0000
I am new to this list, so maybe I am missing something, but
what is wrong with just compiling gnuplot for your target.

I am running debian on TS7400 just installed gnuplot it seems 
to work fine although you have to output to a file, since I 
have no X. (actually no video display). 

Should be ok for a web server if that's what you are doing.
Also since you are just visualizing some data, speed is not an
issue. Not like doing some compute intensive graphics..

On the subject of fixed vs float wars, I am firmly on the side of
fixed point, although I have written my own floating point 
libraries in assembler (many years ago). The only time you
**REALLY** need floating point for realtime is when a variable
has a very large dynamic range, and those instances are fairly
rare. Nowadays it seems everything has hardware floating point
and as a result it becomes the easy way out. 

Meantime, try compiling something like 

         float x,y,z;
         x=123.456;
         y=567.890;
         z = x*y;
         printf("z=%f\n",z);

I would be surprised if it didn't work..... :-)


Hope this helps.

Regards Ray


--- In  "j.chitte" <> wrote:
>
> --- In  Christopher Friedt <cfriedt@> 
> wrote:
> >
> > j.chitte wrote:
> > > I have managed to find some information on the subject which 
> seemed 
> > > to indicate fp emulation could be done but I'd need uclibc which 
> in 
> > > itself required a 2.6.x kernel (which is not currently supported 
> by 
> > > TS).
> > 
> > This is not the case. uclibc and glibc both have soft-floats, it's 
> just 
> > a question of how they're handled. You can do software floating 
> point at 
> > the moment, but it will be quite slow - I think that's the 
> intention of 
> > the warning.
> > 
> 
> Hmm, in view of the complexity of this I may have to see if I can 
> live with that solution.
> 
> > > I need to asses how much customisation effort this is going to 
> imply 
> > > and whether I need to find another way to implement what I need 
> > > without fp.
> > 
> > You could also used fixed point integer arithmetic, and go without 
> any 
> > system calls for softfloat - maybe there's even a flag you could 
> pass to 
> > gcc to optimize for fixed-point integer math? It won't be terribly 
> fast 
> > though.
> > 
> > > 
> > > Could someone give me details on what is needed to achieve softfp 
> on 
> > > 7250?
> > 
> > If you're interested in using something new(er) to speed up FP, 
> then you 
> > probably would want to use the EABI. It speeds up soft-floats by a 
> large 
> > factor.
> > 
> > http://www.linuxdevices.com/articles/AT5920399313.html
> > http://wiki.debian.org/ArmEabiPort
> > 
> > I don't know if anyone has actually gotten an eabi kernel / 
> userland 
> > running on the 7[23]xx boards yet, but I'm currently working on it. 
> I've 
> > built a toolchain as follows:
> > 
> > USE="-*" crossdev -v \
> > --kernel   2.6.21    \
> > --libc     2.5-r3    \
> > --binutils 2.17      \
> > --gcc      4.2.0     \
> > --target arm-softfloat-linux-gnueabi
> >
> 
> OK, I tried to follow what you posted earlier but glibc failed 
> telling my headers where too old (2.6.20). Is 2.6.21 an minimum here 
> or is that possibly a bug? I thought EABI work started around 2.6.11 .
> 
> I've been having some issues building qemu as well. Odd since I had 
> it working beatifully a few months back.
> 
> Could you cast an eye over this topic if you have a chance?
> http://forums.gentoo.org/viewtopic-p-4162111.html#4162111
> 
> I'm quite interested in your idea of using this as a means of doing 
> "native" arm compiling form gentoo as you suggested.
> 
> Thanks for all your explainations. It's taken some digging but I'm 
> starting to get an overview of the start of the art on all this. 
> 
> js.
>  
> > Both the userland and kernel should either both be compiled w/ the 
> EABI 
> > or both not, ... it's possible to have a mixed environment but it 
> would 
> > likely be very complicated.
> > 
> > ... I have yet to get a functional kernel out of EABI though :P I 
> seem 
> > to be hitting the 'invalid machine id' problem in the list's 
> archives, 
> > which leads me to believe that the compiler isn't generating the 
> right 
> > endianness or some other register issue. I probably just need to 
> change 
> > my cflags when compiling linux to point specifically to the ep93xx 
> > devices, but haven't touched it for a few days.
> > 
> > 
> > Let us know if you make any interesting progress?
> > 
> > ~/Chris
> >
>




 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
    http://groups.yahoo.com/group/ts-7000/

<*> Your email settings:
    Individual Email | Traditional

<*> To change settings online go to:
    http://groups.yahoo.com/group/ts-7000/join
    (Yahoo! ID required)

<*> To change settings via email:
     
    

<*> To unsubscribe from this group, send an email to:
    

<*> Your use of Yahoo! Groups is subject to:
    http://docs.yahoo.com/info/terms/
 

<Prev in Thread] Current Thread [Next in Thread>
Admin

Disclaimer: Neither Andrew Taylor nor the University of NSW School of Computer and Engineering take any responsibility for the contents of this archive. It is purely a compilation of material sent by many people to the birding-aus mailing list. It has not been checked for accuracy nor its content verified in any way. If you wish to get material removed from the archive or have other queries about the archive e-mail Andrew Taylor at this address: andrewt@cse.unsw.EDU.AU