https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78742
--- Comment #6 from rguenther at suse dot de <rguenther at suse dot de> --- On Fri, 9 Dec 2016, mpolacek at gcc dot gnu.org wrote: > https://gcc.gnu.org/bugzilla/show_bug.cgi?id=78742 > > --- Comment #5 from Marek Polacek <mpolacek at gcc dot gnu.org> --- > FWIW, my untested fix was > > --- a/gcc/tree-data-ref.c > +++ b/gcc/tree-data-ref.c > @@ -2781,6 +2781,10 @@ can_use_analyze_subscript_affine_affine (tree *chrec_a, > tree *chrec_b) > /* FIXME: For the moment not handled. Might be refined later. */ > return false; > > + if (!cst_and_fits_in_hwi (CHREC_RIGHT (*chrec_a)) > + || !cst_and_fits_in_hwi (CHREC_RIGHT (*chrec_b))) > + return false; > + > type = chrec_type (*chrec_a); > left_a = CHREC_LEFT (*chrec_a); > left_b = chrec_convert (type, CHREC_LEFT (*chrec_b), NULL); In fact initialize_matrix_A relies on the value fitting an int ... I'll see to a better fix.