Re: [tree-ssa] Make PHI nodes reachable by GGC

[Jan Hubicka <hubicka at ucw dot cz>]

> On Fri, Nov 28, 2003 at 11:59:24PM +0100, Jan Hubicka wrote:
> > This still seems to be better than moving whole CFG into GGC memory, but
> > I can implement that too if this is preferred.
> 
> I'd prefer the whole CFG be GCed.

The problem with this idea is that GGC gets into incredibly large
recursion depths while walking linked lists.
We have no way to expose the fact that there are multiple linked lists
per node, so we probably need some extension to deal with this more
nicely.
> 
> >     ann = bb_ann (bb);
> >     TREE_CHAIN (phi) = ann->phi_nodes;
> >     ann->phi_nodes = phi;
> > +   VARRAY_TREE (tree_phi_root, bb->index) = phi;
> 
> This just wastes space in the bb annotation.  We might as well just
> move the PHI out to the array entirely and be done with it.

OK, it seemed to be consistent to have the GGC hack on side and keep CFG
looking like real datastructure, but I see that we have ancesor function
for phi_nodes, so it is not big deal to return in back later.  Will
update the patch.

Honza
> 
> 
> r~