Re: [RFA] PowerPC e500mc core support

[Edmar Wienskoski-RA8797 <edmar at freescale dot com>]

Joseph S. Myers wrote:
> On Mon, 5 May 2008, Edmar Wienskoski-RA8797 wrote:
>
>   
> > Freescale would like to contribute this patch to gcc. It creates a scheduler
> > and proper default flags for the e500mc core.
> >     
>
> Could you describe how this has been tested?
>   
I built the compiler with "--with-cpu=e500mc", used the result to build 
glibc, and run all testsuite
with no regressions in respect to the 603 target.
> You're adding a processor with both ISEL and FPRs, a new combination.  If
Yes. It is a new chip that has both.
>  
> you're not using a configuration that uses the e500.h header, TARGET_ISEL 
> is just defined to be 0, so the enabling of ISEL is ineffective.  If you 
>   
Right, my error. I did not verify if the resulting compiler actually 
generated an isel instruction.
> are using this header in your testing, you get the error from 
> CHECK_E500_OPTIONS that this combination isn't supported.  The only reason 
> not to get that error would be that after you set rs6000_isel there is 
> code that does
>
>   if (TARGET_E500)
>     {
> ...
>     }
>   else if (rs6000_select[1].string != NULL)
>     {
> ...
>       if (!rs6000_explicit_options.isel)
>         rs6000_isel = 0;
>     }
>
> and so rs6000_isel would be reset again - and this comes before the 
> CHECK_E500_OPTIONS.  (The new processor doesn't fall within the existing 
> TARGET_E500 definition; whether it should depends on how it relates to the 
> few uses of this macro for the presence of such things as string 
> instructions and lwsync.)
>
> So you need to make sure you test for a configuration where ISEL is 
> supported at all (one using e500.h); remove the error for the ISEL, FPRs 
> combination; remove the disabling of ISEL for explicit use of this 
> processor; and fix any ICEs arising from the combination.  (The error was 
> added because of very simple code producing an ICE with the combination in 
> the past, when the combination arose through user error.)
>   
I will work on this..

Thanks,
Edmar