[tree-ssa] More aliasing fixes

[Diego Novillo]

On Thu, 2003-12-18 at 08:45, Jan Hubicka wrote:

> But DOM shall not invalidate A's value just because P's MT is writen
> into.  A's tag is not.
>
It will.  Value numbering uses all the virtual operands.

> But I think in your counting of VUSES and VDEFS you didn't considered
> that for store operation to *p for instance, *p MT is in VDEFS, but a
> and b MT is in both VDEFS and VUSES as it is not clear that it modifies
> it.
>
VDEF implies VUSE.  We won't add a VUSE<a> if the statement already has
a VDEF<a>.

> Still have to find time to think it over.  Jeff also mentioned some SGI
> paper about the issue, do you have pointer by any chance?
> 
SGI paper?  I don't recall an SGI paper talking about this.  Sorry.


Diego.