Re: cross-gcc : FPU type error

[Dimitry Andric <dimitry at andric dot com>]

On 2004-05-26 at 17:05:32 Marius Groeger wrote:

>> Yup, running glibc + all apps with softfloat, VFP format.  However,

> What is "all apps" in your project? If you go all the way up to X11,
> this can be quite a lot...

This is an embedded target running ARM Linux with only some dedicated
applications, running busybox for most grunt work.  So it's not much
work to simply recompile everything from scratch.

But indeed, I haven't tried building X11, not sure I want to. :)


> Actually, -msoft-float in kernel or firmware Makefiles usually means
> "no float operations, please".

Of course you can do float operations, but the compiler just won't
emit any hardware FPU instructions. :)  My gcc's spec file is set up
so that software VFP is the default, if you don't specify any option,
i.e.:

no option     -> softfloat, VFP format
-msoft-float  -> softfloat, FPA format
-mhard-float  -> hardfloat, FPA format


> I agree that a -Werror-on-fp option would be nice. But don't forget
> that to be really sure, you'd need that on the linker level to catch
> malicious asm sources or inline asm as well!

Normally the linker refuses to link if .o files disagree on the float
format.  So hopefully such an error will be caught before you ship
the product. ;)

Attachment: pgp00000.pgp
Description: PGP signature