gcc has an internal floating point implementation located in libgcc.a
(if its enabled). If you look at the gcc source in gcc/config/arm (the
various t-* files) you'll see that all of the other arm targets have the
"in-gcc" floating point enabled. Linux is the only target that sets
hard-float, and uses an external floating point library if soft-float is
set.
I have made the changes on my copy of gcc to enable the in-gcc floating
point, however I have not yet run the tests to determine if it is
working correctly. (It seems to be on a float at least, havn't tried a
double yet.)
I do not want to use the in kernel floating point emulator. (Reasons
are varied, but it is not what I want to do..) Once I get the gcc
working the way I want, I'm going to recompile the kernel and turn off
NWFPE, then I'll see if it's really working the way I think it should!
--Mark
Scott Bambrough wrote:
>
> libfloat is only used when the -msoft-float flag is passed. Otherwise the
> compiler issues instructions for the FPA11 coprocessor for floating point. This
> requires the actual hardware, or an emulator such as the one Russell provides or
> the NWFPE.
>
> To use -msoft-float with gcc requires a library implementing the functions that
> the gcc calls instead of using machine based floating point operations.
> libfloat implements this for ARM systems. AFAIK, gcc has no implementation of
> software floating point, and even if it did, I doubt it supports ARM doubles
> correctly.
>
> Scott
>
> Mark Hatle wrote:
> >
> > I hope this is the correct place to ask...
> >
> > I am working on compiling the toolchain for StrongArm. I'm starting
> > with the Netwinder stuff, however I noticed that they were using an
> > external library (libfloat) instead of the built in gcc functions.
> >
> > Is there a reason for this? As far as I can tell all of the
> > architectures I am familiar with use the built in gcc soft floating
> > point.
> >
> > Thanx!
> > --Mark
> >
> > unsubscribe: body of `unsubscribe linux-arm' to [EMAIL PROTECTED]
> > ++ Please use [EMAIL PROTECTED] for ++
> > ++ kernel-related discussions. ++
>
> --
> Scott Bambrough - Software Engineer
> REBEL.COM http://www.rebel.com
> NetWinder http://www.netwinder.org
unsubscribe: body of `unsubscribe linux-arm' to [EMAIL PROTECTED]
++ Please use [EMAIL PROTECTED] for ++
++ kernel-related discussions. ++
