Re: [rtl-optimization] Improve Data Prefetch for IA-64

[Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>]

Hello,

> > Steven Bosscher <stevenb@suse.de>:
> > 
> > >
> > > What happens if you use the memory address unrolling 
> > patch, turn on
> > > -fweb, and set the unrolling parameters properly?
> > >
> > 
> > The memory address unrolling patch can't work on IA-
> > 64,
> 
> Ah, no base+offset.  It could be made to work with a little
> work though.  The IV splitting needs to be taught about GIVs,
> and when that is in place, it should work, right?
> (IMHO the current RTL IV analysis is not very nice to work
> with, we could look into improving that first...)

what problems do you have concretely in mind?

> On the other hand, we may not even need it if the RTL unroller
> will be replaced by the tree one.  I don't know what Zdenek's
> plans are about that, or if this is a wise thing to do at all.

I don't think it is the case; as far as I know, it is recommended
to do non-specific unrolling just before or during scheduling.
(By "non-specific" I mean unrolling the loops without any other
reason other than just getting them unrolled).  

Zdenek