Barriers should not be re-initialized. You initialize them once at the start of the program and then just use them. They are designed to reset automagically. Any re-init call is exposing yourself to random failures.
On Fri, Jan 9, 2015 at 11:58 AM, Stuart Haslam <[email protected]> wrote: > On Fri, Jan 09, 2015 at 11:09:33AM -0500, Mike Holmes wrote: > > I think this may have to come out again - can anyone confirm this > behaviour > > Regression started running this multiple times and it is hanging with > high > > load maybe 5% of the time. > > > > while true; do ./test/validation/odp_synchronizers; done > > Hangs pretty quickly for me, one thread is in custom_barrier_wait() on > custom_barrier2 and the rest are in custom_barrier_wait() on > custom_barrier1. > > The custom_barrier2.wait_cnt seems to always be 8, so I wonder if the > slow thread is reinitialising the barriers before all the other threads > have made it out of the barrier2 custom_barrier_wait(). Maybe the > barrier test is missing a barrier? > > -- > Stuart > > _______________________________________________ > lng-odp mailing list > [email protected] > http://lists.linaro.org/mailman/listinfo/lng-odp >
_______________________________________________ lng-odp mailing list [email protected] http://lists.linaro.org/mailman/listinfo/lng-odp
