I converged to an implementation based on flagged cells: I assume that the flagged cells at the deepest level are the ultimate cells one is interested with. From there, I track theses cells up and down the hierarchy, including through all layers of virtual grids.
The trade-off between speed and memory is done by storing only the flagged cells vector rather than the full grid at each level, but looping through it regularly. I will run a quick profiling test to check that it does not suddenly breaks done on bigger cases.
Main additions is in the multilevelcontext/mask file. Let me know if anything seems absurd.
Generate Ramses ic_refmap and ic_pvar files.
I converged to an implementation based on flagged cells: I assume that the flagged cells at the deepest level are the ultimate cells one is interested with. From there, I track theses cells up and down the hierarchy, including through all layers of virtual grids.
The trade-off between speed and memory is done by storing only the flagged cells vector rather than the full grid at each level, but looping through it regularly. I will run a quick profiling test to check that it does not suddenly breaks done on bigger cases.
Main additions is in the multilevelcontext/mask file. Let me know if anything seems absurd.