------- Comment #2 from rguenth at gcc dot gnu dot org 2006-04-27 09:44 -------
Note that loop information looks corrupt:
(gdb) print *loop
$2 = {num = 144085752, header = 0xb7fbb178, latch = 0xb7e4c3c0,
lpt_decision = {decision = LPT_NONE, times = 0}, ninsns = 0, av_ninsns = 0,
num_nodes = 0, depth = -1, pred = 0x0, level = 2, outer = 0x0, inner = 0x0,
next = 0x0, copy = 0x0, aux = 0x0, nb_iterations = 0x0,
estimated_nb_iterations = 0x69, bounds = 0xb7fbb178,
single_exit = 0x8948888, parallel_p = 1 '\001'}
(gdb) print *loop->header
$3 = {stmt_list = 0x0, preds = 0x0, succs = 0x896ad70, aux = 0x8948888,
loop_father = 0x896d608, dom = {0x896d608, 0xb7fbb188},
prev_bb = 0xb7fbb188, next_bb = 0xb7fbb190, il = {rtl = 0xb7fbb190},
phi_nodes = 0x896d750, predictions = 0x8947938,
count = -5189358841774755424, index = -1208241752, loop_depth = -1208241752,
frequency = -1208241744, flags = -1208241744}
The loop walking in peel_loops_completely looks bogus, as we're modifying
the loop tree.
--
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=27335
