------- Comment #7 from ubizjak at gmail dot com 2009-09-03 08:53 ------- (In reply to comment #3) > Hmm, dropping the stmt looks like it would be a "hack". Alex - if I just set > flag_var_tracking_assignments to 1 if I encounter a GIMPLE_DEBUG during > read-in > I get > > t.c: In function 'foo': > t.c:1:6: error: unrecognizable insn: > (debug_insn 6 5 14 t.c:3 (var_location:SI hex (mem/c/i:SI (plus:SI (reg/f:SI > 20 > frame) > (const_int -4 [0xfffffffc])) [0 hex+0 S4 A32])) -1 (nil)) > t.c:1:6: internal compiler error: in insn_default_length, at > config/i386/i386.md:544 > Please submit a full bug report, > with preprocessed source if appropriate. > See <http://gcc.gnu.org/bugs.html> for instructions. > > is there sth else that I need to do to "enable" VTA?
You should use NONDEBUG_INSN_P instead of INSN_P somewhere. Backtrace will tell you where. I have similar problem on alpha, see perhaps PR 41238. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=41230
