Closed RudolfWeeber closed 6 years ago
@RudolfWeeber , was this fixed by #1587?
On Thu, Nov 02, 2017 at 04:58:12PM +0000, Florian Weik wrote:
@RudolfWeeber , was this fixed by #1587? Yes, except the one for IMMERSED_BOUNDARIES which I left alone (@sgekle) The runtime error collection there can probably also be removed, but it only becomes relevant when virtual sites are switchable at runtime and IMMERSED_BOUNDARIES can be included in the maxset configuration.
Collection includes an
all_reduce
which, in effect acts as barrier. Several algorithms call this during the integration loop without checking whether they are on:There is one non-ifdef-ed
check_runtime_errors()
somewhere in the middle of the integration loop. Maybe, that's the only one which should stay. It might however be moved to the point at which theannounce_resort_particles
happens, which also is an effective barrier.Also an unconditional ghost comm there.