Re: variable pool for unit-at-a-time C/objC

[Jan Hubicka <jh at suse dot cz>]

> On Thursday, Jun 19, 2003, at 17:13 US/Eastern, Jan Hubicka wrote:
> 
> >
> >Hi,
> >this patch implements new variable pool - datastructure with interface
> >symmetric to cgraph that is used to hold global variables.  At the 
> >moment it
> >just implements the interface and reachability analysis (via checking 
> >what
> >assembly names has been output and what variables are needed by 
> >functions
> >to be expanded in cgraphunit), so it can deal with cyclic references 
> >dead
> >from outside like:
> >
> >static char *a=(char *)&b;
> >static char *b=(char *)&a;
> >
> >That is probably not big deal.  It is mostly usefull to make C++ 
> >transition
> >easier, but I would like to implement some simple optimizations, like
> >constification of variables never writen to and unification of the 
> >same objects
> >whose address is never taken.
> >
> >Regtested/bootstrapped i386/mainline together with the previous patch 
> >to
> >cleanup unit-at-a-time.  OK?
> >
> 
> Does this fix also PR optimization/9702 by a chance?

This looks to me more like shared constant pool issue.  This is old
problem I always wanted to find time to take a look at, but didn't
suceed to find time yet :(

Honza
> 
> Thanks,
> Andrew Pinski
>