[vkoppejan]

Hi everyone,

The problem in short:

point cloud with Np particles -> take a single particles center coordinates

-> add 28 sub-volume points per particle (i.e. r_test = r_center + r_subvolpoint),

where r is a (x,y,z) coordinate.

sub-volume point -> findCell(subvolpoint) (using one of the many options) ->cellI,

where cellI is the outcome of the findCell algorithm

* outcome 1, cellI >= 0 (awesome, no problem, do stuff on partition owned by proc)

* outcome 2, cellI < 0 (oops, see below)

Let's say I gather all outcomes 2 into a array/vector/list or array for each proc, how can I reduce these among all procs (make a single vector/list/array out of the coordinates), do another findCell loop in parallel (preferably) where each procs only acts if cellI > 0.

Thanks in advance,

Victor

p.s.

The basic idea behind this thing is to minimize any reduce operations since they can be slow on multiple nodes on a cluster.