Re: [PATCH] Fix IVOPTs problem

[Zdenek Dvorak <rakdver at atrey dot karlin dot mff dot cuni dot cz>]

Hello,

> On Fri, 2005-08-05 at 11:48 +0200, Richard Guenther wrote:
> > Compiling gcc.c-torture/execute/20010910-1.c with IVOPTs enabled
> > leads to ssa verification failures because 
> > tree-ssa-loop-ivopts.c:rewrite_use does not update ssa form
> > (it calls update_stmt but misses mark_new_vars_to_rename).
> But are the transformations performed by ivopts supposed to
> be changing the aliasing information?

no, ivopts should not change virtual operands of any statement.  So
there is indeed a bug in ivopts, but this is not the right fix.

Zdenek

> ie, the appearance of a new V_MAY_DEF indicates that the
> aliasing information on that statement has changed.  So at
> the heart of this patch is whether or not IVOPTS is allowed
> to change a statement's aliasing properties.
> 
> If the answer is yes, then clearly this patch is OK.  If the
> answer is no, then clearly this patch is not OK.  I tend to
> think the former, mostly because we ran into a similar situation
> with the vectorizer, but I'd like to be a little more sure.
> 
> I was going to try and do a little investigative work on my
> own, but couldn't get 20010910-1.c to fail.  Do I need to
> install some uncommitted patch (if so, a pointer would be
> greatly appreciated).  Target triplet and compile-time
> flags would be helpful too.
> 
> jeff
> 
>