- 
                Notifications
    You must be signed in to change notification settings 
- Fork 121
Open
Description
Performance optimization of the moving immersed boundary (MIBM) code has been explicitly delayed, as stated in the initial PRs. Optimizations for MIBMs must be supported, as it it currently the rate-limiting step in all MIBM cases. Potential sources of improvement include:
- Finish moving all MIBM calculations to the GPU, as some of the calculations are currently performed serially on the CPU and then copied over. Most of these calculations should be quite easy to parallelize
- Possibly add time-step movement to the ghost points. We have a method of tracking how the centroid of the MIB changes between time steps. Because the ghost points are measured to be particular points on the lagrangian system, it should be straight-forward to replace the re-calculation (which is performed each time step) of the ghost points with some form of time evolution. The ability to implement this optimization will depend exactly on the nuance of how ghost point locations evolve over time.
More to be added later.
Metadata
Metadata
Assignees
Labels
No labels