Re: [autovect] [patch] and aliasing question

[Daniel Berlin <dberlin at dberlin dot org>]

On Thu, 2005-05-26 at 08:35 -0400, Diego Novillo wrote:
> On Thu, May 26, 2005 at 03:31:45PM +0300, Ira Rosen wrote:
> 
> > BTW, the original pointer also has no flow-sensitive information.
> > 
> It wouldn't.  It's a default name for a PARM_DECL.  I *think*
> Dan's new points-to analysis will give it some more meaningful
> information.

When I started, the answer would have been no, because i couldn't get a
correct list of global variables easily (saying it only points to global
variables referenced in this procedure isn't actually correct).

However, I think cgraph provides us a way to walk the global variables,
and that info is now *correct*, so this should be possible now.

I'll do it as a followup after i replace the old analyzer.


>   In principle, this pointer would point to all the
> locals whose address escapes and all globals with their address
> taken.  IPA information would give you a more precise set.
> 
> 
> Diego.