Reply To: Manipulating populations with postprocessors on GPU

[jmctighe]

After thinking on this a little bit, would it be possible to store the cell data adjacent to the boundaries in such a way that they are globally available? The cell data could be stored before executing the boundary condition post processor. Something like this would also more easily support domain decomposition if each block could access all of the necessary cell data.

An alternative approach, if possible, might be to execute the boundary condition post processor on the host and then perform the collision and streaming on the device. Obviously, this would not be as efficient and does not support domain decomposition, but could give a computational boost over the single-CPU serial process.

Interested to know if either of these approaches might be possible within the current framework.