Re: SPE configuration in GCC

[Mark Mitchell <mark at codesourcery dot com>]

David Edelsohn wrote:
> > > > > > Mark Mitchell writes:
>
>
> Mark> I understand that reason.  I think that it would make more sense for the 
> Mark> default to match the usual GCC convention (all variants supported) and 
> Mark> have the AIX configuration redefine TARGET_SPE to zero (to avoid the 
> Mark> bloat) than the current situation.  David, comments?
>
> 	powerpc-linux (mentioned in the first message) is not the default.

True.

> SPE is not supported by *BSD or BeOS or most other PowerPC targets.  It
> seemed easier to disable it as the default and enable it for the specific
> targets that support SPE than vice versa.
>
> 	I would consider changing the behavior for powerpc-elf (not
> powerpc-linux), but all other powerpc targets, including linux, need to be
> updated to zero that macro.  Compiler performance on Linux, Darwin, AIX,
> *BSD, etc. is important.

So, it sounds like you would consider a patch which modified rs6000.h, 
so long as all the OS header files for which particular variants were 
inappropriate then zero'd the macros.  This certainly isn't a 4.0 
change, so I'll put in my possible-changes-to-implement-for-4.1 queue, 
if you're agreed with the idea in principle.

I'm working with a customer which is distributing cross compilers to 
GNU/Linux.  They are supporting both SPE/E500 and "traditional" PowerPC. 
 They use different kernels and different GLIBC builds, for these 
environments.  However, they would like to use a single compiler.  I 
think this is a valid scenario where one wants a single compiler that 
can do both SPE/E500 GNU/Linux and traditional GNU/Linux.  So, if you 
don't think the change I'm suggesting is appropriate for GNU/Linux, then 
I'll just make a separate configuration header for this customer.

Thanks,

--
Mark Mitchell
CodeSourcery, LLC
mark@codesourcery.com
(916) 791-8304