From: Karl Kurbjun <kkurbjun_at_gmail.com>
Date: Mon, 12 Nov 2007 21:11:27 -0700

Sorry, to clarify, after building the new crosscompiler you should not
need the -mfpu=fpa option (at least I did not).

-Karl

On Nov 12, 2007 9:06 PM, Karl Kurbjun <kkurbjun_at_gmail.com> wrote:
> Cat,
>
> Thanks to Dave's work on building ARM multilib you can use the
> arm926ej-s option. If you recompile your arm gcc crosscompiler with
> rockboxdev.sh it should give you full armv5 (arm926ej-s) support.
> I'll commit a change to the configure script so that it adds the mcpu
> option by default. I've been working with it for a day or so and it
> seems to work well.
>
> Regards,
> Karl
>
>
> On Nov 12, 2007 8:52 PM, Catalin Patulea <cat_at_vv.carleton.ca> wrote:
> > Hey guys,
> >
> > I'm working on some code for the m:robe and I noticed that the current
> > assembler for the m:robe target (arm9tdmi) doesn't support the BKPT
> > instruction. I forget exactly what error message I get, but using
> > "-mcpu=arm926ej-s" solved the problem (which makes sense, because
> > that's a closer match to the CPU in the m:robe). I ended up getting a
> > different error at that point, regarding heterogenous use of
> > floating-point instructions. I found that compiling all of Rockbox
> > with "-mfpu=fpa" solved that problem as well. The resulting code seems
> > to run, though I haven't really tested much. I think the code size
> > also decreased marginally.
> >
> > My question is: why do I need -mfpu=fpa and is it safe to check in a
> > change like that? Does Rockbox use floating-point? I can dig up the
> > actual error messages if you think they're valuable.
> >
> > -Cat
> >
>
Received on 2007-11-13