Re: [patch, vectorizer] Fix PR tree-optimization/37174: ICE in tree-vectorizer.h

[Ira Rosen <IRAR at il dot ibm dot com>]

Dorit Nuzman/Haifa/IBM wrote on 21/08/2008 11:28:23:

> it's ok, except we usually do that by explicitly checking if the
> stmt is outside the loop, like so:
>   bb = gimple_bb (*def_stmt);
>   if (!flow_bb_inside_loop_p (loop, bb))
>  ....
>
> By relying on the def-type ('dt') being vect_loop_def:
>
> > -      if (def_stmt && vinfo_for_stmt (def_stmt)
> > +      if (def_stmt && dt[i] == vect_loop_def && vinfo_for_stmt
(def_stmt)
> >            && STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (def_stmt)))
>
> ...you may miss patterns that involve other def-types, like
> reduction, induction, etc. It's probably ok for now, cause SLP
> doesn't supports reduction/induction anyhow, but we would probably
> want to change that in the future.

OK, I am now testing:

--- tree-vect-analyze.c 2008-08-20 15:18:54.000000000 +0300
+++ tree-vect-analyze.c.p       2008-08-21 12:23:31.000000000 +0300
@@ -2508,6 +2508,7 @@ vect_get_and_check_slp_defs (loop_vec_in
   stmt_vec_info stmt_info =
     vinfo_for_stmt (VEC_index (gimple, SLP_TREE_SCALAR_STMTS (slp_node),
0));
   enum gimple_rhs_class rhs_class;
+  struct loop *loop = LOOP_VINFO_LOOP (loop_vinfo);

   rhs_class = get_gimple_rhs_class (gimple_assign_rhs_code (stmt));
   number_of_oprnds = gimple_num_ops (stmt) - 1;        /* RHS only */
@@ -2531,7 +2532,9 @@ vect_get_and_check_slp_defs (loop_vec_in
       /* Check if DEF_STMT is a part of a pattern and get the def stmt
from);
          the pattern. Check that all the stmts of the node are in the
          pattern.  */
-      if (def_stmt && vinfo_for_stmt (def_stmt)
+      if (def_stmt
+          && flow_bb_inside_loop_p (loop, gimple_bb (def_stmt))
+          && vinfo_for_stmt (def_stmt)
           && STMT_VINFO_IN_PATTERN_P (vinfo_for_stmt (def_stmt)))
         {
           if (!*first_stmt_dt0)

Thanks,
Ira