Full supported and unchanged. Initialization happens on the CPU, before data is moved down to the GPU.
Fully support, as an out of place sort
Supported, but requires a full data copy
Supported, but requires a full data copy
Supported, but requires a full data copy
Fully supported
Supported in two ways:
- GPU: If a user writes GPU-aware code in their deck, it will work and be fast
- CPU: A user can use existing diagnostics by requesting the code to ship the data back to the CPU, at the expense of speed (see "New Features")
Basic TA support added, but not heavily tested.
Needs manual porting
Only the inbuilt TA collisions are supported at this time
Not used by any major uses. Will be ported on request
Users can request that particle or field data be copied back to the host via intervals set in the input deck. This copy happens before user diagnostics to allow the use of existing diagnostics code at the expense of performance