Re: [PATCH] Fix PR35607, another revenge of invariant addresses (ivopts)

[Richard Guenther <rguenther at suse dot de>]

On Mon, 17 Mar 2008, Zdenek Dvorak wrote:

> Hi,
> 
> > On Sun, 16 Mar 2008, Richard Guenther wrote:
> > 
> > > On Sun, 16 Mar 2008, Richard Guenther wrote:
> > > 
> > > > On Sun, 16 Mar 2008, Zdenek Dvorak wrote:
> > > > > I am more inclined to sacrifice the simplicity of that check (see the
> > > > > patch below).  Is that known to cause problems?
> > > > 
> > > > Previously that was rejected as being too expensive (if I remember
> > > > correctly Eric proposed sth similar last year).  (it also looks
> > > > somewhat incomplete as it doesn't handle all handled_component_refs,
> > > > also indirect_refs are never invariant, also IMHO for ARRAY_REF
> > > > indices we should require a literal constant (another nested
> > > > &foo[0] should not be allowed)).
> > > 
> > > http://gcc.gnu.org/ml/gcc-patches/2008-01/msg00095.html

-->

> > On 7/4/07 6:24 PM, Eric Botcazou wrote:
> > > OK, my understanding is that is_gimple_min_invariant as currently 
> > written is 
> > > correct as "non-recursive" predicate (for the updated grammar), i.e. 
> > during 
> > > gimplification."
> > 
> > this seems to be incorrect - re-gimplifying of &foo[(int)a_2] does
> > not result in any gimplification work as is_gimple_min_invariant is
> > used to check if recursing is required.  So this creates a
> > chicken-and-egg problem here (and maybe is the point that needs
> > fixing?)

<---

> maybe you wanted to send this to the list, or cc Eric?  I certainly
> agree with this statement,

Indeed.  Copied the list now.  And Diego who had much to say about
gimple grammar in the threads referenced above (read between the
two arrows ;).

Richard.