Powerpc IRQ handling breaks strict EABI compliance

[Sergei Organov]

Valette Eric <eric.valette at free.fr> writes:
> > 64K for read/write data + 64K for read-only data.
> 
> How do you intent to choose which part of the data will fit in the 64K?

I don't. Compiler does it for me. By default it puts data variables/constants
of the length <= 8 bytes to these small segments and still I wasn't able to
overflow them. The length margin could be changed using -G switch to gcc.

> 
> > Unfortunately not. The only thing I know for sure is that my application
> > doesn't fit into a few of my boards without EABI and does fit with it. Code
> > speed improvement is a guess I've looking at the assembly code generated by
> > gcc in both cases.
> 
> Which is not really a prof for speed improvement...

I didn't say it is. But to proof it is not would be even harder...

> 
> > Just tried to re-build one of my applications without EABI (RTEMS is still
> > compiled with EABI). The difference in code size is:
> > -meabi -msdata=eabi: 362032
> 
> > -mno-eabi -msdata=none: 378720
> > If newlib/glib is also compiled without EABI, the code size is:
> 
> >                         380320
> > So for my application it's about 5% difference.
> 
> Did you try the new gcc 3.x -Os option to see the result?

No, I didn't. But I expect the decrease of code size due to -Os would be
additive w.r.t. the decrease due to the use of EABI.

-- 
Sergei.