Construct CoreParams from new problem input in celer-sim #1601
Conversation
This means it's more difficult to get the processed input if there's an error, but it makes the structure easier.
Test summary 4 497 files 6 923 suites 13m 11s ⏱️ Results for commit 82ee9c6. ♻️ This comment has been updated with latest results. |
|
Currently I'm hitting an error trying to run celer-sim on the GPU: /home/alund/celeritas_project/celeritas/app/celer-sim/celer-sim.cc:273: critical: While running input at simple-cms-gpu.inp.json: celeritas: runtime error: celeritas::activate_device may be called only once per application
/home/alund/celeritas_project/celeritas/src/corecel/sys/Device.cc:308: '!d' failed |
|
Glad someone here is paying attention 😅 I'll check it out on wildstyle and fix. |
|
@amandalund all tests pass on a debug+openmp+cuda build on wildstyle! |
|
Awesome, thanks! I'll get back to this today. |
There was a problem hiding this comment.
This looks good @sethrj! One kind of odd thing is the results on this branch were different from develop (but still statistically the same) when I tried running the cms2018 problem below. Is there any reason the results might have changed?
cms2018 json input
{ "geometry_file": "/home/alund/celeritas_project/regression/input/cms2018.gdml", "physics_options": { "coulomb_scattering": false, "eloss_fluctuation": true, "em_bins_per_decade": 56, "lpm": true, "msc": "urban", "physics": "em_basic", "rayleigh_scattering": true }, "brem_combined": false, "primary_options": { "seed": 12345, "direction": {"distribution": "isotropic"}, "energy": 10000, "num_events": 8, "pdg": 11, "position": [0, 0, 0], "primaries_per_event": 130 }, "initializer_capacity": 8388608, "num_track_slots": 65536, "field": [0.0, 0.0, 1.0], "step_diagnostic": true, "step_diagnostic_bins": 1000, "max_events": 1024, "secondary_stack_factor": 3.0, "seed": 20220905, "merge_events": false, "action_times": false, "use_device": false }
| // celeritas::device is reset) | ||
| if (celeritas::device()) | ||
| { | ||
| celeritas::device().create_streams(0); |
There was a problem hiding this comment.
We also do this in SharedParams, maybe we should have a Device::destroy_streams()?
| @@ -207,6 +265,9 @@ inp::StandaloneInput to_input(RunnerInput const& ri) | |||
| si.problem = load_problem(ri); | |||
| if (!ri.physics_file.empty()) | |||
| { | |||
| CELER_VALIDATE( | |||
| ends_with(ri.physics_file, ".root"), | |||
| << R"(physics_file must be a ROOT input: use GDML for geometry_file and if forcing an ORANGE geometry, use the `ORANGE_FORCE_INPUT` environment variable)"); | |||
There was a problem hiding this comment.
Should GDML physics/org.json ORANGE be deprecated input formats? Or are we not worrying about backward compatibility here.
| return EventReader{rfe.event_file, nullptr}.num_events(); | ||
| }, | ||
| }, | ||
| si.events); |
There was a problem hiding this comment.
Since we're getting the number of events here, could we also update the number of streams to not be more than the number of events?
| namespace setup | ||
| { | ||
| //---------------------------------------------------------------------------// | ||
| struct StandaloneLoaded |
| if constexpr (CELERITAS_CORE_GEO == CELERITAS_CORE_GEO_ORANGE) | ||
| { | ||
| static char const fi_hack_envname[] = "ORANGE_FORCE_INPUT"; | ||
| auto const& filename = celeritas::getenv(fi_hack_envname); |
There was a problem hiding this comment.
Why is this hack necessary?
|
|
||
| //---------------------------------------------------------------------------// | ||
| /*! | ||
| * Costruct magnetic field from variant input. |
This follow-on to #1562 moves the core construction from
app/celer-simtoceleritas/setup. I removed a couple of small capabilities related to the deprecated celer-sim json input for now:Changed in
inp:inp::StateCapacityinto core/opticalstd::optionalrather than sentinel values