This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC] tree-ssa-ccp.c: Propagate even more ADDR_EXPRs.

On Tue, 2005-05-31 at 11:03 -0400, Diego Novillo wrote:
> On Tue, May 31, 2005 at 08:43:58AM -0600, Jeffrey A Law wrote:
> > I'm not sure if we want to bother with this in CCP.  The forward 
> > propagation code already handles this case rather effectively.  What's
> > the real benefit from doing the same optimization in CCP that we're
> > already doing in forwprop.c?
> > 
> The propagation done by CCP is inherently more powerful, but this
> case may be so rare in practice that I wouldn't be surprised if
> it didn't any difference either way.
I'd be surprised if we saw a difference as well.  About the only thing
that's done by forwprop.c that isn't done by CCP is some trivial
cost analysis -- ie, don't propagate an ADDR_EXPR into a loop nest,
don't propagate to multiple use sites (which undoes CSE).

> > [ Now you could argue that all the ADDR_EXPR stuff in forwprop.c could
> >   be moved into CCP.  They're not really constants or invariants, but
> >   with some work we could probably convince CCP to handle them.  ]
> > 
> Either that, or re-implement forwprop using the propagation
> engine.
I don't think the bulk of forwprop really fits well into the
propagation engine, except for the ADDR_EXPR bits.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]