|
Message
From: Matjaz Breskvar <phoenix@o...>
Date: Wed, 12 Nov 2003 13:39:40 +0100
Subject: Re: [openrisc] gdb stub instead of JTAG?
* Scott Furman (sfurman@r...) wrote:
> Matjaz Breskvar wrote:
>
> >* Robert Cragie (rcc@j...) wrote:
> >
> >
> >>I had problems with diag_printf() with optimisations turned on but works
> >>OK
> >>with optimisation turned off. I am not using multi-tasking facilities of
> >>eCos yet.
> >>
> >>Has anyone else experienced problems using optimisation (-02) with the
> >>compiler?
> >>
> >>
> >
> >Yes, there were reports of problems compiling eCos with -O2 optimizations
> >on opencores gcc-3.2.3 compiler. i think going to -O solved it. (it in the
> >archives a few months ago)
> >
> I encountered bugs with both long long and (soft) double floating point
> support using gcc 3.1 and have made some fixes to the compiler to solve
> these problems. I haven't submitted my patches because my changes
> modify the ABI, rendering object files incompatible with code compiled
> by the unmodified compiler. (The calling convention I use is more
> efficient than the existing 3.1 version: 64-bit values (long longs and
> doubles) are passed by value in registers rather than as references.)
>
> Is anyone actively maintaining the gcc versions ? I could share my
> changes if we were willing to tolerate an ABI change.
i'm the opencores maintener for gcc. please do send the
patches. i'll look into patch and then we can have
discussion about the ABI change. personaly i'm not against a change
if there is a clear benefit in doing it.
> > - linker may miscalculate section lenght or corrupt the binary
> >when linking with -z. If you avoid giving section directives
> >to gcc it should be ok.
> >
> I have a bunch of linker fixes which cure the problems with linking with
> the -r flag. I don't think they would fix problems of the sort you
> describe, but it's worth a try. I'll submit my patches tomorrow.
great. if you have write access to opencores cvs you may commit
it yourself.
p.
|
Browse
