Peano

Searching...

No Matches

#include <algorithm>
#include <cstring>
#include <random>
#include "Benchmarks-main.h"
#include "Constants.h"
#include "tarch/NonCriticalAssertions.h"
#include "tarch/logging/Log.h"
#include "tarch/multicore/multicore.h"
#include "tarch/multicore/Core.h"
#include "peano4/peano.h"
#include "peano4/grid/Spacetree.h"
#include "peano4/parallel/SpacetreeSet.h"
#include "repositories/DataRepository.h"
#include "repositories/StepRepository.h"
#include "repositories/GlobalState.h"
#include "swift2/UserInterface.h"
#include "swift2/kernels/legacy/Density.h"
#include "swift2/kernels/legacy/HydroForce.h"
#include "swift2/kernels/legacy/Leapfrog.h"
#include "swift2/kernels/legacy/SmoothingLength.h"
#include "swift2/timestepping/Leapfrog.h"
#include "swift2/timestepping/TimeStepping.h"
#include "swift2/kernels/ParticleSetIterators.h"
#include "NohProblem.h"
#include "tarch/timing/Measurement.h"
#include "tarch/multicore/otter.h"
#include "globaldata/HydroPart.h"
#include "vertexdata/HydroPartSet.h"

Include dependency graph for BenchmarkDriver.cpp:

Go to the source code of this file.

Functions
void	initParticle (globaldata::HydroPart &particle, int numberOfParticles, int index)
	Initialise the particles with some meaningful information.
void	initParticles (int numberOfParticles, bool initScatteredData)
void	validateOutcomes (const std::string &name, int numberOfParticles)
	Validate the scattered data against the continuous data.
void	prepareForceBenchmark (int numberOfParticles, bool initScatteredData)
	Prepare the force benchmark.
std::vector< int >	CoalescedMemoryIndices (TwoPowerD)
std::vector< int >	ScatteredMemoryIndices (1)
void	createCoalescedMemoryIndices ()
	Create CoalescedMemory particle indicator.
void	allocateParticles (int numberOfParticles)
void	deleteParticles (int numberOfParticles)
void	assessKernel (std::function< void(int) > kernelCallInLoop, const std::string &name, int numberOfParticles, bool firstAssessCallForThisProgramPhase)
void	runBenchmark (int numberOfParticles)
void	initGridTraversalEvent ()
	Initialise the global variable gridTraversalEvent.
int	main (int argc, char **argv)
	Main routine of the SPH code.

Variables
tarch::logging::Log	_log ("::")
peano4::grid::GridTraversalEvent	gridTraversalEvent
	See initGridTraversalEvent()
vertexdata::HydroPartSet *	workItemParticleSet
globaldata::HydroPart *	scatteredWorkItems
vertexdata::HydroPartSet *	scatteredWorkItemParticleSet

Function Documentation

◆ allocateParticles()

void allocateParticles ( int numberOfParticles )

Definition at line 299 of file BenchmarkDriver.cpp.

References tarch::allocateMemory(), assertionEquals, tarch::Heap, scatteredWorkItemParticleSet, scatteredWorkItems, and workItemParticleSet.

Referenced by runBenchmark().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ assessKernel()

void assessKernel	(	std::function< void(int) >	kernelCallInLoop,
		const std::string &	name,
		int	numberOfParticles,
		bool	firstAssessCallForThisProgramPhase )

Parameters

firstAssessCallForThisProgramPhase If it is true, we do not check the continuous data structure against the scattered one.

Definition at line 357 of file BenchmarkDriver.cpp.

References tarch::timing::Watch::getCalendarTime(), tarch::timing::Measurement::getValue(), logInfo, benchmarks::name, tarch::timing::Measurement::setValue(), tarch::timing::Watch::stop(), tarch::timing::Measurement::toString(), and validateOutcomes().

Referenced by runBenchmark(), runBenchmarks(), and runBenchmarks().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ CoalescedMemoryIndices()

std::vector< int > CoalescedMemoryIndices ( TwoPowerD )

References TwoPowerD.

Referenced by createCoalescedMemoryIndices(), and runBenchmark().

Here is the caller graph for this function:

◆ createCoalescedMemoryIndices()

void createCoalescedMemoryIndices ( )

Create CoalescedMemory particle indicator.

Whenever we enter a cell, we are given all the particles associated to the \( 2^d \) vertices of the cell as one big container. If we work with CoalescedMemory storage schemes, these particles will technically be stored as \( 2^d \) continuous sequences. In the benchmark, all particles of one work item are held in one continuous sequence, which is a special case of this. Nevertheless, we create a vector of integeres, which pretends that they were split up further.

So if chunk has 18 entries, it might, in 2d, holds the entries 5,5,5,3.

Definition at line 284 of file BenchmarkDriver.cpp.

References CoalescedMemoryIndices(), ScatteredMemoryIndices(), and TwoPowerD.

Referenced by main().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ deleteParticles()

void deleteParticles ( int numberOfParticles )

Definition at line 337 of file BenchmarkDriver.cpp.

References assertionEquals, tarch::freeMemory(), tarch::Heap, scatteredWorkItemParticleSet, scatteredWorkItems, and workItemParticleSet.

Referenced by toolbox::particles::SieveParticles< T >::deleteParticles(), toolbox::particles::SieveParticles< T >::exchangeSieveListsGlobally(), and runBenchmark().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ initGridTraversalEvent()

void initGridTraversalEvent ( )

Initialise the global variable gridTraversalEvent.

We work with one big traversal event which represents the unit square/cube. It is not refined, and none of the adjacent vertices are hanging. The traversal event is not used directly, but the markers are derived from it later throughout the benchmark run.

Definition at line 1353 of file BenchmarkDriver.cpp.

References gridTraversalEvent.

Referenced by main().

Here is the caller graph for this function:

◆ initParticle()

void initParticle	(	globaldata::HydroPart &	particle,
		int	numberOfParticles,
		int	index )

Initialise the particles with some meaningful information.

We basically spread them out evenly (so there's no division by zero if we compute distances) and then use the Noh2d setup to give each particle a meaningful value.

Definition at line 70 of file BenchmarkDriver.cpp.

References benchmarks::swift2::hydro::kernel_throughput::initNohProblemIC().

Referenced by initParticles().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ initParticles()

void initParticles	(	int	numberOfParticles,
		bool	initScatteredData )

Definition at line 103 of file BenchmarkDriver.cpp.

References initParticle(), scatteredWorkItemParticleSet, and workItemParticleSet.

Referenced by prepareForceBenchmark(), and runBenchmark().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ main()

int main	(	int	argc,
		char **	argv )

Main routine of the SPH code.

Definition at line 1382 of file BenchmarkDriver.cpp.

References createCoalescedMemoryIndices(), peano4::fillLookupTables(), tarch::multicore::Core::getInstance(), tarch::multicore::Core::getNumberOfThreads(), initGridTraversalEvent(), tarch::initNonCriticalAssertionEnvironment(), peano4::initParallelEnvironment(), peano4::initSingletons(), tarch::multicore::initSmartMPI(), logError, logInfo, swift2::parseCommandLineArguments(), runBenchmark(), peano4::shutdownParallelEnvironment(), and peano4::shutdownSingletons().

Here is the call graph for this function:

◆ prepareForceBenchmark()

void prepareForceBenchmark	(	int	numberOfParticles,
		bool	initScatteredData )

Prepare the force benchmark.

To prepare for the force benchmark, we run through the following steps:

Initialise the particles through an initParticles() call. This initialises the workItemParticleSet sequence.
invoke resetSmoothingLengthIterationCounter() and resetAcceleration() on the particles. For this, we run over all the workItemParticleSets and, within each set, use the forAllParticles function to initialise the particles.

Definition at line 232 of file BenchmarkDriver.cpp.

References swift2::kernels::alwaysUpdateInVertexKernel(), swift2::kernels::forAllParticles(), swift2::kernels::kernelrealisation::Generic, gridTraversalEvent, initParticles(), swift2::kernels::legacy::resetAcceleration(), swift2::kernels::legacy::resetSmoothingLengthIterationCounter(), scatteredWorkItemParticleSet, and workItemParticleSet.

Referenced by runBenchmark().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ runBenchmark()

void runBenchmark ( int numberOfParticles )

Definition at line 386 of file BenchmarkDriver.cpp.

Referenced by main().

Here is the call graph for this function:

Here is the caller graph for this function:

◆ ScatteredMemoryIndices()

std::vector< int > ScatteredMemoryIndices ( 1 )

Referenced by createCoalescedMemoryIndices().

Here is the caller graph for this function:

◆ validateOutcomes()

void validateOutcomes	(	const std::string &	name,
		int	numberOfParticles )

Validate the scattered data against the continuous data.

We assume that the benchmark always assesses the generic kernel operating on scattered data first and that there's exactly one run with this data set. That means that we can assume that the scattered data is our validation data and the kernels over the continuous dataset have to yield exactly the same outcome. We validate this here.

Definition at line 136 of file BenchmarkDriver.cpp.

References tarch::la::equals(), benchmarks::name, scatteredWorkItemParticleSet, and workItemParticleSet.

Referenced by assessKernel().

Here is the call graph for this function: