Re: Is store_expr(e,t,v!=m) safe?

[Jan Hubicka <jh at suse dot cz>]

> > On Mon, Apr 07, 2003 at 11:56:52PM +0200, Jan Hubicka wrote:
> > > I think it does not mask it.  The problem is that a[0]=x is converted
> > > into load+logicals+store of the whole "a" that fits in DImode that
> > > doesn't hurt as long as we don't decide to spit "a" into constant memory
> > > area.  Avoid purge_addressof from doing the (pessimizing) transfromation
> > > and later changing the mind is the right way to go IMO.
> > 
> > I disagree.  Indeed, even my patch does not totally fix
> > the problem.  There are additional examples in 
> > 
> >   http://gcc.gnu.org/ml/gcc-patches/2002-12/msg00533.html
> > 
> > that continue to fail.

Also do you have testcase that fails with Zdenek's patch and does not
fail with yours?

Honza
> 
> How are these related to the addressof pass?  It looks like problem in
> the expansion that should either decide to clear the flag or not use
> memcpy+initialize nonzero elements sequence.
> 
> I see that the unchanging flag goes away when I increase the size of the
> structure, so probably someone already takes care for this but forgets
> about the case when we go trought addressof.  I can check this tomorrow.
> 
> Honza
> > 
> > 
> > r~