On Tue, Nov 24, 2015 at 11:39 AM, Kyrill Tkachov <kyrylo.tkac...@arm.com> wrote: > Hi all, > > The ICE in this PR occurs when trying to dump the CFG to a .dot file. > The i > vc assert at the bottom of the loop in get_loop_body_in_bfs_order > gets triggered when i == 1 and vc == 1. This is not a problem for that loop > because it contains only a header and it's not going to go around the loop > again > to access uninitialized memory in the blocks array. > > Richi suggested restructuring the loop to initialise the blocks array with > the header > early on and moving the assert to the beginning of the loop. > This fix works as far as I can test. > > Bootstrapped and tested on arm, aarch64, x86_64. > > Ok for trunk?
Ok. Thanks, Richard. > Thanks, > Kyrill > > 2015-11-24 Richard Biener <rguent...@suse.de> > Kyrylo Tkachov <kyrylo.tkac...@arm.com> > > PR middle-end/68375 > * cfgloop.c (get_loop_body_in_bfs_order): Restructure loop to avoid > bogus assertion. > > 2015-11-24 Kyrylo Tkachov <kyrylo.tkac...@arm.com> > > PR middle-end/68375 > * gcc.dg/pr68375.c: New test.
