[visit-users] host cells or ghost nodes?

Favre Jean jfavre at cscs.ch
Mon Mar 29 15:45:42 EDT 2010


Dear List

I have a parallel reader, producing a rectilinear mesh. My data is node-centric. Motivated by the latest CG&A paper on streaming of ghost data, co-authored by Hank, I decided to add proper ghost-nodes to my already working reader. (I use version 2.0RC)

However,  InverseGhostZone tells me it is all empty. I am wondering if that operator can only show ghost cells, and if it is not capable of finding cells touched by nodes marked as ghost-nodes?

I have checked that the array I passed to mark ghost-nodes is correctly pseudo-colored, showing me the two different values I put in (0, or DUPLICATED_NODE).

When calculating divergence(vector field), I get discontinuities all along my process boundaries. I interpret that as a sign that either my ghost-node marking does not work at all, or that the div() calculation does not take it into account properly. I use a vtkExtentTranslator to compute the correct dimensions of each process's domain+ghost.

Question: should InverseGhostZone show me the cells where at least one node is marked as ghost, or must I also marked the cells as ghost inside GetMesh()? I have checked the manual, but it only briefly covers ghost cells, and does not cover ghost nodes.

Thanks for any tip.

-----------------
Jean M. Favre
Swiss National Supercomputing Center



More information about the visit-users mailing list