CPU demo loop

[sethrj]

CPU demo loop

[amandalund]

Do we need space for more than one track state on the CPU? Seems like we would have an outer loop over primaries and initialize the track from each primary/transport it one by one. The track initialization code (extend_from_primaries, initialize_tracks, extend_from_secondaries) is all device-specific, so we'll different logic on the CPU anyway.

[paulromano]

The way I've been structuring things so far is to have a loop over thread IDs when adding an equivalent CPU "kernel", e.g.
https://github.com/paulromano/celeritas/blob/a611b0ba8adc23d56e8d086a6552153ed8678af9/src/physics/em/detail/KleinNishina.cc#L32-L33

This exposes some loop-level parallelism so that if we wanted to actually use CPU threads, it would be pretty straightforward. The alternative would be for the CPU kernels to omit that loop and instead have an outer loop over primaries, as you suggest. Either way, we'll need to have an outer loop for the case where # primaries > # tracks, so the question is really just on the inner loop do we want to allow more than one primary/track?

[paulromano]

One issue I've run into with trying to run the demo loop on CPU: right now, the problem consists of 15 primaries but we're only setting room in the state data for a single primary:

celeritas/app/demo-loop/LDemoRun.cc

Line 190 in b62d827

resize(&state_storage, build_params_refs<MemSpace::host>(params), 1);

It looks like this is because VGNavCollection on the host only permits its size to be one, so increasing the above size just leads to precondition errors on the size of the navigation state. From what I can gather, the options to get around this are:

1. Use NavStatePool on host as we currently do for device to allow multiple CPU "threads"
2. Extend the demo loop logic to account for the case where # primaries > # tracks (not yet implemented for GPU), and continue with a single primary on CPU

Any thoughts on the best path forward would be appreciated

[sethrj]

I think either approach would be fine, but single-threaded should be our initial goal and I will approve most expedients to get us there. :)

[sethrj]

Closed by #273 and #310