--- On Thu, 12/11/08, Kai Ruottu <kai.ruottu@xxxxxxxxxxx> wrote:
> From: Kai Ruottu <kai.ruottu@xxxxxxxxxxx>
> Subject: Re: vfpv2 support in gcc-4?
> To: "Gabi Voiculescu" <boy3dfx2@xxxxxxxxx>
> Date: Thursday, December 11, 2008, 12:40 PM
> Gabi Voiculescu wrote:
>
> > Does the standard gcc=>4 have support for ARM's
> vectored floating point unit version 2 (VFP11 - VFPv2)?
>
> Please see the config template and other config settings
> for your target
> in 'gcc/config.gcc'. Then browse the config headers
> in question in
> 'gcc/config/arm' ! At least in gcc-4.3.2 there
> seemed to be quite a lot
> "support"...
>
Ok.
I did not know about this file.
I also see that gcc/Changelog-*.* provides me with some information on what was modified from one gcc version to the next.
I'll use them to figure out exactly where I start having vfp11 supported.
> > Is it compiled in applications/toolchains when the
> -mfpu=vfp is given in the command line?
>
> What "command line"? When you build a cross GCC,
> the options used in
> the "command lines" are for the host GCC, using a
> '-mfpu=vfp' for a x86
> GCC would cause an error message about "wrong
> option"! If you produce
> a generic GCC for ARM (Linux, I assume) then it requires
> using this as
> an additional option just as a i386-default GCC would
> require a
> '-march=pentium2' or something when compiling for
> 'i686'.... If you
> produce a "custom" GCC which has this option as
> default in code
> production, then you must use one or more suitable
> '--with-cpu=' and
> '--with-fpu=' "GCC configure options",
> not "GCC options". Please see
> the documentation for "Install/Configure" in
> 'gcc.gnu.org'...
Ok. I think I understand now.
The filename in crosstools mislead me: I thought arm_softfloat.dat would refer to the features of the toolchain, not to those of the host.
Thank you for clearing up those issues.
Gabi Voiculescu
