Re: Fix PR39580

[Matthias Klose <doko at ubuntu dot com>]

On 22.04.2009 15:22, Richard Guenther wrote:
> 2009/4/22 Andrey Belevantsev<abel@ispras.ru>:
> > Hello,
> >
> > PR 39580 is a test that ices with selective scheduling on ia64.  As
> > explained in the audit trail, the assert that was hit with the testcase did
> > not describe all situations happening when the history of changes of two
> > expressions is being merged.  The expression history vector contains one
> > entry for each insn uid, on which the expression was changed, and for each
> > entry we save the form of expression before and after the transformation.
> >   As long as the assert condition was true, it meant that  whatever
> > unifications we perform, this single entry is enough to find all unified
> > expressions.  The PR testcase shows the counterexample for that.  In the
> > corner situations, only one unified expression will be found below the merge
> > point, which does not influence correctness but only effectiveness of
> > unification.
> >
> > Extending assert to handle the PR case would make it useless, as it used to
> > assert that only certain situations could happen, and now we know that all
> > cases can happen.  The best solution IMHO is to remove the assert.  This
> > fixes the testcase and bootstraps/regtests fine on ia64 with all default
> > languages enabled.
> >
> > OK for trunk?
>
> Ok.

this was reported to the debian bug tracker, seen for 4.4.4. the patch applies 
on the 4.4 branch as well and fixes the ICE. currently running the testsuites on 
ia64-linux-gnu and i486-linux-gnu. Is this ok for the 4.4 branch if the 
testsuites run without regressions?

  Matthias

> Thanks,
> Richard.
>
> > Andrey
> >
> > 2009-04-22  Andrey Belevantsev<abel@ispras.ru>
> >
> >         PR rtl-optimization/39580
> >         * sel-sched-ir.c (insert_in_history_vect): Remove incorrect
> > gcc_assert.
> >
> >
> > Index: gcc/sel-sched-ir.c
> > ===================================================================
> > *** gcc/sel-sched-ir.c  (revision 146520)
> > --- gcc/sel-sched-ir.c  (working copy)
> > *************** insert_in_history_vect (VEC (expr_histor
> > *** 1512,1525 ****
> >       {
> >         expr_history_def *phist = VEC_index (expr_history_def, vect, ind);
> >
> > -       /* When merging, either old vinsns are the *same* or, if not, both
> > -          old and new vinsns are different pointers.  In the latter case,
> > -          though, new vinsns should be equal.  */
> > -       gcc_assert (phist->old_expr_vinsn == old_expr_vinsn
> > -                   || (phist->new_expr_vinsn != new_expr_vinsn
> > -&&  (vinsn_equal_p
> > -                           (phist->old_expr_vinsn, old_expr_vinsn))));
> > -
> >         /* It is possible that speculation types of expressions that were
> >            propagated through different paths will be different here. In this
> >            case, merge the status to get the correct check later.  */
> > --- 1512,1517 ----