Re: Antworten: Re: [avr-gcc-list] Problem with the calling convention

[E. Weddington]

On 14 Jul 2004 at 18:06, Joerg Wunsch wrote:

> "Martin Krüger " <address@hidden> wrote:
>
> > avr-gcc -c -Os  test.c (without optimization the code seems to be ok)
> >
> > avr-objdump -D test.o
>
> You could have called avr-gcc with -S directly instead. ;-)
>
> > The we will see the call to testfunc32
> >
> > mov     r25, r19
> > mov     r24, r18
> > mov     r23, r17
> > mov     r22, r16
> > rcall   .+0
>
> That's quite OK.
>
> > This code is executed after all the other function-calls.
> > Any of these could have modified the values of  r19,r18
>
> Sure, but since they are not used again afterwards, this is not a
> problem.
>
> However, I can see a problem.  If I add something like
>
> unsigned long foo;
>
> ....
>
> testfunc(...)
> {
>  ...
>  foo = d;
> }
>
> it still doesn't save r18/r19.  In that case, I think it /is/ a bug.
>
> Please submit a GCC bug report for this (but use my addition, since it
> makes the bug obvious).
>

Joerg,

Do we know if this is a regression? If we can discover that this is a
regression then we're more likely to get a fix in say 3.4.2 instead of having
to wait until 3.5.

Eric