This is the mail archive of the
gcc-bugs@gcc.gnu.org
mailing list for the GCC project.
[Bug tree-optimization/50208] [4.7 Regression] ICE: in vinfo_for_stmt at tree-vectorizer.h:598 with -ftree-vectorize -fno-tree-fre -fno-tree-pre
- From: "jakub at gcc dot gnu.org" <gcc-bugzilla at gcc dot gnu dot org>
- To: gcc-bugs at gcc dot gnu dot org
- Date: Mon, 29 Aug 2011 10:44:42 +0000
- Subject: [Bug tree-optimization/50208] [4.7 Regression] ICE: in vinfo_for_stmt at tree-vectorizer.h:598 with -ftree-vectorize -fno-tree-fre -fno-tree-pre
- Auto-submitted: auto-generated
- References: <bug-50208-4@http.gcc.gnu.org/bugzilla/>
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=50208
--- Comment #2 from Jakub Jelinek <jakub at gcc dot gnu.org> 2011-08-29 10:44:42 UTC ---
The problem is that def_stmt is a gimple stmt from before the loop and
new_loop_vec_info only initializes gimple_uid of stmts in the loop being
processed.
So perhaps before testing for vinfo_for_stmt it should
|| !gimple_bb (def_stmt)
|| !flow_bb_inside_loop_p (loop, gimple_bb (def_stmt))
and perhaps give up in those cases too? Loop would need to be initialized,
perhaps similarly to vect_recog_dot_prod_pattern?
loop_vec_info loop_info = STMT_VINFO_LOOP_VINFO (vinfo_for_stmt (stmt));
struct loop *loop = LOOP_VINFO_LOOP (loop_info);
? I have no idea whether this can be invoked from slp vectorizer and what to
do in that case.