Manage execution of a standalone Celeritas optical stepping loop. More...

#include <Runner.hh>

Public Types
Type aliases
using	Input = inp::OpticalStandaloneInput

using	SPConstParams = std::shared_ptr< CoreParams const >

using	SpanConstTrackInit = DirectGeneratorAction::SpanConstData

using	SpanConstGenDist = GeneratorAction::SpanConstData

Public Member Functions
	Runner (Input &&)
	Construct with optical problem input definition.

void	insert ()
	Set the number of pending tracks for a primary generator.

void	insert (SpanConstTrackInit)
	Insert track initializers.

void	insert (SpanConstGenDist)
	Insert distributions for generating through scintillation or Cherenkov.

Result	operator() ()
	Generate optical photons and transport to completion.

CounterAccumStats	exchange_counters ()
	Get and reset accumulated track counters.

ActionTimes::MapStrDbl	exchange_action_times ()
	Get accumulated wall times for each action.

StepTimes::VecDbl	exchange_step_times ()
	Get and reset the wall time for each step iteration.

SPConstParams const &	params () const
	Access the shared params.

setup::OpticalProblemLoaded const &	problem () const
	Access the problem setup.

Detailed Description

Manage execution of a standalone Celeritas optical stepping loop.

Used by the celer-optical app and the LarStandaloneRunner .

Note: When parallelizing on the CPU using OpenMP, this class expects track-level parallelism be enabled (transporting all tracks with a single stream and state) rather than event-level parallelism (transporting events on separate streams with one state per stream). Similarly on the GPU a single stream and state will be used.

Member Function Documentation

CounterAccumStats celeritas::optical::Runner::exchange_counters ( )

Get and reset accumulated track counters.

void celeritas::optical::Runner::insert ( SpanConstGenDist data )

Insert distributions for generating through scintillation or Cherenkov.

Todo:: We could accumulate and update the number of pending tracks directly in GeneratorAction::insert, though this would be less efficient for some run modes, e.g. offloading distributions through accel where we already know the number of pending tracks.

The documentation for this class was generated from the following files: