Re: [tree-ssa]: Overlap of variable with same variable?

[Daniel Berlin <dberlin at dberlin dot org>]

Works for insert before, i think.

I think i'm left now with one last insert problem somewhere, because DCE
is still busting things (removing way too many phi nodes).  If I turn off
DCE or PRE, it bootstraps okay, which tells me it's probably an insert
problem.

I'll hunt down what's going on.



On Thu, 29 May 2003, Andrew MacLeod wrote:

> On Thu, 2003-05-29 at 19:56, Daniel Berlin wrote:
> > When I hit the abort, it says there is an overlap between t and var
> > Except that:
> >
> > (gdb) p debug_generic_expr (t)
> > pretmp.389_37
> > $1 = void
> > (gdb) p debug_generic_expr (var)
> > pretmp.389
> >
> > This is not a possible overlap, they are the same variable.
>
> Actually it can be an overlap. pretmp.389_XXX has been coalesced with
> pretmp.389 already, and it interfered with pretmp.389_37. That forces
> _37 to get a new variable. Look in the list of coalesces And I bet its
> there.  I should providea better mechanism for debugging all the bits
> that have gone on... perhaps a dump of the conflict graph, but the
> mechanism it provides wasnt very helpful.
>
> >
> > If it helps, it could be we aren't updating the end of the bb tree
> > properly either , since in this case, pretmp.389_37 is saved only to a
> > statement at the end of a bb, and thus, if end_bb wasn't updated
> > properly, the out-of-ssa dropper would miss it.
> >
> > Any ideas?
> >
>
> First, look back, Im sure you'll find pretmp.389 coalesced with
> something which interferes with _37
>
> I'll think about what better debugging could be available.
> Andrew
>
>