NNPartMgr< Gnl, GainMgr, ConstrMgr > Class Template Reference

No-Nonsense Partitioning Algorithm Manager. More...

#include <NNPartMgr.hpp>

Inheritance diagram for NNPartMgr< Gnl, GainMgr, ConstrMgr >:

Collaboration diagram for NNPartMgr< Gnl, GainMgr, ConstrMgr >:

Public Member Functions
	NNPartMgr (const Gnl &hyprgraph, GainMgr &gain_mgr, ConstrMgr &constr_mgr, size_t num_parts)
	Constructs a new NNPartMgr object.
Public Member Functions inherited from PartMgrBase< Gnl, GainMgr, ConstrMgr >
	PartMgrBase (const Gnl &hyprgraph, GainMgr &gain_mgr, ConstrMgr &constr_mgr, size_t num_parts)
	Construct a new Part Mgr Base object.
void	init (std::span< std::uint8_t > part)
	Initializes the partition manager with the given partition.
auto	legalize (std::span< std::uint8_t > part) -> LegalCheck
	Legalizes the partition to satisfy balance constraints.
void	optimize (std::span< std::uint8_t > part)
	Optimizes the partition using the FM algorithm.

Protected Member Functions
void	_optimize_1pass (std::span< std::uint8_t > part) override
	Performs a single pass of the No-Nonsense optimization algorithm.

Additional Inherited Members
Public Types inherited from PartMgrBase< Gnl, GainMgr, ConstrMgr >
using	GainCalc_ = typename GainMgr::GainCalc_
using	GainMgr_ = GainMgr
using	ConstrMgr_ = ConstrMgr
Public Attributes inherited from PartMgrBase< Gnl, GainMgr, ConstrMgr >
int	total_cost {}
Protected Attributes inherited from PartMgrBase< Gnl, GainMgr, ConstrMgr >
const Gnl &	hyprgraph
	Reference to the hypergraph being partitioned.
GainMgr &	gain_mgr
	Gain manager for computing and managing gains.
ConstrMgr &	validator
	Constraint manager for validating partition constraints.
size_t	num_parts
	Number of partitions.

Detailed Description

template<typename Gnl, typename GainMgr, typename ConstrMgr>
class NNPartMgr< Gnl, GainMgr, ConstrMgr >

No-Nonsense Partitioning Algorithm Manager.

NNPartMgr reuses the shared FM algorithm skeleton from PartMgrBase (Template Method pattern: init/legalize/optimize) and overrides only the single-pass behaviour: unlike FM it is a pure greedy local search that stops at the first non-positive-gain move (no look-ahead / snapshot / rollback) and never locks a moved vertex. The gain-computation and constraint-validation strategies remain injected as template parameters.

Template Parameters

Gnl
GainMgr
ConstrMgr

Constructor & Destructor Documentation

NNPartMgr()

template<typename Gnl , typename GainMgr , typename ConstrMgr >

NNPartMgr< Gnl, GainMgr, ConstrMgr >::NNPartMgr ( const Gnl &  hyprgraph, GainMgr &  gain_mgr, ConstrMgr &  constr_mgr, size_t  num_parts  )

inline

Constructs a new NNPartMgr object.

Parameters

[in]	hyprgraph	The hypergraph to be partitioned
[in,out]	gain_mgr	The gain manager used for the partitioning
[in,out]	constr_mgr	The constraint manager used for the partitioning
[in]	num_parts	The number of partitions to create

Member Function Documentation

_optimize_1pass()

template<typename Gnl , typename GainMgr , typename ConstrMgr >

void NNPartMgr< Gnl, GainMgr, ConstrMgr >::_optimize_1pass ( std::span< std::uint8_t >  part )

overrideprotectedvirtual

Performs a single pass of the No-Nonsense optimization algorithm.

Pure greedy local search: repeatedly takes the highest-gain move from the gain buckets while its gain is strictly positive, and stops at the first non-positive-gain move. It performs no snapshotting/rollback and does not lock moved vertices.

Parameters

[in,out]

part

The partition vector to optimize

Reimplemented from PartMgrBase< Gnl, GainMgr, ConstrMgr >.

---

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

• ckpttn/NNPartMgr.hpp