[Bug tree-optimization/53550] [4.6/4.7/4.8 regression] ICE with -O{1,2,3} -fprefetch-loop-arrays in build2_stat, at tree.c:3803

[jakub at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=53550

--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> 2012-06-01 
12:29:11 UTC ---
I'd think
--- tree-ssa-loop-niter.c.jj    2012-05-30 14:38:55.000000000 +0200
+++ tree-ssa-loop-niter.c    2012-06-01 14:27:43.202456745 +0200
@@ -1276,13 +1276,14 @@ number_of_iterations_cond (struct loop *
      practice, but it is simple enough to manage.  */
   if (!integer_zerop (iv0->step) && !integer_zerop (iv1->step))
     {
+      tree step_type = POINTER_TYPE_P (type) ? sizetype : type;
       if (code != NE_EXPR)
     return false;

-      iv0->step = fold_binary_to_constant (MINUS_EXPR, type,
+      iv0->step = fold_binary_to_constant (MINUS_EXPR, step_type,
                        iv0->step, iv1->step);
       iv0->no_overflow = false;
-      iv1->step = build_int_cst (type, 0);
+      iv1->step = build_int_cst (step_type, 0);
       iv1->no_overflow = true;
     }
would match what tree-chrec.c is doing.