Re: SPE configuration in GCC

[Mark Mitchell <mark at codesourcery dot com>]

David Edelsohn wrote:

> 	As Aldy mentioned, there already is a powerpc-linuxspe
> configuration that seems to do what your customer wants.
> gcc/config/rs6000/linuxspe.h makes the options active and defaults to a
> 8540 scheduling model, but does not enable the SPE features by default.

I don't think that's quite true.  It seems to default to enabling SPE 
via SUBTARGET_OVERRIDE_OIPTIONS.  It also passes "-mspe -me500" to the 
assembler unconditionally.  Our customer wants SPE to be an option, not 
the default.

> 	I am wliling to consider a patch to rs6000.h and the other
> headers, but I am not sure what benefit it provides over the
> configurations that Aldy already designed and implemented.

I think that the best solution, then, is probably a new configuration 
header for this customer.  That's easy enough to arrange.  We've 
certainly established that the current situation is a conscious choice, 
which was my primary objective.

As a secondary issue, it would be nice to have a GCC-wide policy about 
whether "target-elf" should support all variants of a chip, or just 
some.  Based on my knowledge of users, I'd suggest that it should 
support all variants.

As another secondary issue, I consider it a bug in the PowerPC back end 
that command-line options (like -misel) are silently ignored in the 
default configuration.  If you agree, I can file a PR about that.

Thanks,

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