Setup Testcase for cpu - gpu comparison
As discussed in the VF Meetings, we want to have a simple test case comparing the cpu and gpu part of VirtualFluids. As a quality check, we expect to get the same results accordingly to the floating-point accuracy. The test case have the following setup:
- cubic main grid
- sphere solid object
- cubic area of refinement around the sphere.
@m.schoenherr @k.kutscher Please specify the test case here:
-
grid size - MSch: 3x3x3 blocks (CPU) maybe 20x20x20 nodes per block (overall cube 60x60x60 nodes, more nodes are better ;-) )
-
position / size of sphere - MSch: in the center of the refined block
-
position / size of fine grid - MSch: the refined block/grid is located in the center of the testcase
-
Which Collision Kernel? - MSch: actual version of the K17
-
Which Refinement? Probably !162 (merged) ? - MSch: jep, the newest version of the grid refinement
-
Boundary Conditions - MSch: maybe periodic BC in all three directions (and a forcing) - check of the diff. BCs is an extra test
-
MSch: we nee to evaluate flows in all three directions to be sure, that we are independent of the simulated direction -
MSch: calculation in double and single precision?
I probably miss some points. Feel free to add / change things.
- MSch: I've added some suggestions.
Edited by Martin Schönherr