autopas::LiveInfo Class Reference

This class is able to gather and store important information for a tuning phase from a container and functor. More...

#include <LiveInfo.h>

Public Types
using	InfoType = std::variant< bool, double, size_t, ContainerOption, TraversalOption, LoadEstimatorOption, DataLayoutOption, Newton3Option >
	The type of an info.

Public Member Functions
template<class Particle_T >
void	gather (ContainerIterator< Particle_T, true, false > particleIter, size_t rebuildFrequency, size_t numOwnedParticles, std::array< double, 3 > boxMin, std::array< double, 3 > boxMax, double cutoff, double skin)
	Gathers key statistics that define the computational profile of the simulation, in order to provide lower dimensional and human understandable inputs for the tuning strategies.
const auto &	get () const
	Returns a map of all infos.
template<typename T >
T	get (const std::string &key) const
	Gets a single value from the infos that corresponds to the given string.
std::string	toString () const
	Creates a string containing all live info gathered.
std::pair< std::string, std::string >	getCSVLine () const
	Generate a csv representation containing all values from the toString() method.

Friends
std::ostream &	operator<< (std::ostream &out, const LiveInfo &info)
	Stream operator to write the LiveInfo to a stream.
std::istream &	operator>> (std::istream &in, LiveInfo &info)
	Stream operator to read the LiveInfo in from a stream.

Detailed Description

This class is able to gather and store important information for a tuning phase from a container and functor.

Infos are identified by a string. Their type is InfoType.

Member Function Documentation

gather()

template<class Particle_T >

void autopas::LiveInfo::gather ( ContainerIterator< Particle_T, true, false >  particleIter, size_t  rebuildFrequency, size_t  numOwnedParticles, std::array< double, 3 >  boxMin, std::array< double, 3 >  boxMax, double  cutoff, double  skin  )

inline

Gathers key statistics that define the computational profile of the simulation, in order to provide lower dimensional and human understandable inputs for the tuning strategies.

A lot of the information is based on a couple of different spatial bin resolutions:

• cells: Bin dimensions are the same as cells in the Linked Cells container with CSF 1.
• particleDependentBins: Bin dimensions are designed such that the are approximately 10 particles per bin.
• blurredBins: The domain is divided equally into 3x3x3 "blurred" bins

Note

It is not clear how useful the statistics derived from the particleDependentBins are, as the "resolution" varies depending on the number of particles. E.g. Consider a sparse simulation with a "macroscopic" heterogeneity and a dense simulation with a "microscopic" heterogeneity but at a "macroscopic" level is rather homogeneous. These could have the same particleDependentBin homogeneity (relative std. dev. of bin counts) but would respond to traversals very differently. This could, however, provide a useful metric for homogeneity which is somewhat independent of particle density (e.g. could be useful for determining the best traversal with one statistic independently of what cell-size factor is chosen)

Currently, it provides: -— Bin Independent Statistics -—

• numOwnedParticles: The number of particles in the container that are marked as owned.
• numHaloParticles: The number of particles in the container that are marked as halo.
• cutoff: The configured cutoff radius.
• skin: The configured skin radius.
• domainSizeX: The size of the domain on the x-axis.
• domainSizeY: The size of the domain on the y-axis.
• domainSizeZ: The size of the domain on the z-axis.
• particleSize: The number of bytes one particle in AoS layout needs.
• threadCount: The number of OMP threads that can be used.
• rebuildFrequency: The current verlet-rebuild-frequency of the simulation. -— Cell Statistics -—
• numCells: The number of cell-bins in the domain.
• numEmptyCells: The number of cell-bins that are empty.
• minParticlesPerCell: The minimum number of particles in a cell-bin.
• maxParticlesPerCell: The maximum number of particles in a cell-bin.
• medianParticlesPerCell: The median number of particles per cell-bin.
• lowerQuartileParticlesPerCell: The lower quartile number of particles per cell-bin.
• upperQuartileParticlesPerCell: The upper quartile number of particles per cell-bin.
• meanParticlesPerCell: The average number of particles per cell-bin.
• particlesPerCellStdDev: The standard deviation in the number of particles per cell-bin.
• relativeParticlesPerCellStdDev: The, relative to the mean, standard deviation in the number of particles per cell-bin.
• estimatedNumNeighborInteractions: Rough estimation of number of neighbor interactions. Assumes that neighboring cell-bins contain roughly the same number of particles. Estimation does not work well if this is not the case. -— Particle Dependent Bin Statistics -—
• particleDependentBinMaxDensity: The maximum density of particles of a particle-dependent bin.
• particleDependentBinDensityStdDev: The standard deviation in the density of particles per particle-dependent bin. In earlier versions of AutoPas, this was referred to as "Homogeneity". -— Blurred Bin Statistics -—
• maxParticlesPerBlurredBin: The maximum number of particles in a blurred bin.
• minParticlesPerBlurredBin: The minimum number of particles in a blurred bin.
• medianParticlesPerBlurredBin: The median number of particles per blurred bin.
• lowerQuartileParticlesPerBlurredBin: The lower quartile number of particles per blurred bin.
• upperQuartileParticlesPerBlurredBin: The upper quartile number of particles per blurred bin.
• meanParticlesPerBlurredBin: The mean number of particles per blurred bin.
• particlesPerBlurredBinStdDev: The standard deviation in the number of particles per blurred bin.
• relativeParticlesPerBlurredBinStdDev: The, relative to the mean, standard deviation in the number of particles per blurred bin.

Template Parameters

Particle_T

The type of particle the container stores.

Parameters

particleIter	Iterator to the first particle in the AutoPas container.
rebuildFrequency	The current verlet rebuild frequency that is used in the simulation.
numOwnedParticles	The number of owned particles in the container.
boxMin	Lower left corner of (sub)domain.
boxMax	Upper right corner of (sub)domain.
cutoff	The cutoff.
skin	The (Verlet) skin.

get() [1/2]

const auto & autopas::LiveInfo::get ( ) const

inline

Returns a map of all infos.

Returns

A map of all infos.

get() [2/2]

template<typename T >

T autopas::LiveInfo::get ( const std::string &  key ) const

inline

Gets a single value from the infos that corresponds to the given string.

Template Parameters

T	return type

Parameters

key	string key for the infos map

Returns

getCSVLine()

std::pair< std::string, std::string > autopas::LiveInfo::getCSVLine ( ) const

inline

Generate a csv representation containing all values from the toString() method.

Since the keys are not necessarily known in advance, this method generates a CSV header and a CSV line.

Returns

A pair of strings in the form of (header, line).

toString()

std::string autopas::LiveInfo::toString ( ) const

inline

Creates a string containing all live info gathered.

Returns

A string containing all live info gathered.

Friends And Related Symbol Documentation

operator<<

std::ostream & operator<< ( std::ostream &  out, const LiveInfo &  info  )

friend

Stream operator to write the LiveInfo to a stream.

Parameters

out
info

Returns

operator>>

std::istream & operator>> ( std::istream &  in, LiveInfo &  info  )

friend

Stream operator to read the LiveInfo in from a stream.

Parameters

in
info

Returns

---

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

• autopas/tuning/tuningStrategy/LiveInfo.h