Re: [tree-ssa] ccp/aliasing problem

[Andrew Pinski <pinskia at physics dot uc dot edu>]

On Feb 11, 2004, at 05:00, Diego Novillo wrote:

> On Wed, 2004-02-11 at 06:39, Zdenek Dvorak wrote:
>
> > which is wrong, since __first._M_current is the same place as
> > *(int * const *)&__first and I have lost the dependency.  Where's the
> > problem?
> Aliasing.  It's hard to tell without the actual test case, but it seems
> to me that __first ought to be in the alias set of D38552's memory tag.
> See if the aliasing changes I committed yesterday make a difference.  
> If
> not, please open a PR and assign it to me.
>
> Also, are you sure that the input program is not pulling a fast one?


The problem is that the C/C++ front-ends are lowering things like this:
(&temp->a) to (int**)(temp) which is what is causing this aliasing 
problem.
There is an already a bug filed for this, PR 14029.

Thanks,
Andrew Pinski