-
Notifications
You must be signed in to change notification settings - Fork 11
Home
Roland Haas edited this page Jan 11, 2024
·
50 revisions
As a first step we will use Reprimand to reduce the complexity of code we implement. Instructions for this are found here
- ideas: call scheduled functions with full MFab: done
- ideas: use OpenMP tasking or similar to have a more task based
scheduling - work item ideas
- virtual function on GPUs
- vertex centered grid functions by default: done
- test multi-node GPU support: done
- checkpoint / restart: done
- test a qc0 run or similar to compare results and speed to Carpet w/o subcycling in time
- subcycling in time, likely using runge-kutta prolongation
- allow flexibiliy of subcycling vs no subcycling on a per level case, currently partially implemented
- lead by Liwei Ji @lwJi, code at: https://github.com/lwJi/Infino.jl/tree/main
- ET wiki with information: https://docs.einsteintoolkit.org/et-docs/Prolongation
Science runs would be say a MHD driven supernova with tracer particles and a neutrino evolution scheme. This would require these abilities:
- cell centered (volume averaged) grid with AMR for fluid: done
- vertex centered (sampled) grid with AMR for metric: done
- staggered grid for magnetic field
- subcycling in time
- Runge-Kutta prolongation into mesh refinement boundaries to avoid "buffer" zones (which AMReX may not support)
- some sort refluxing for fluid due to subcycling in time
- some sort of monopole avoiding prolongation scheme for B
- some sort of scheme that conserves magnetic flux at mesh refinement boundaries
- support for 2D (prevent refinement in the Z-direction)
- interpolators to get values of fields at particle locations: done
- reduction operators for min, max, sum etc.: done
- checkpoint and recovery: done
- 3d output for visualization: done
- GPU support
- for RHS calculation: done
- for AMR operators: done
- for everything else