LoadBalanceStrategy.hpp File Reference

implementation of the hilbert curve load balancing More...

#include "shambase/aliases_int.hpp"
#include "shambackends/sycl.hpp"
#include "shambackends/vec.hpp"
#include "shamcomm/logs.hpp"
#include "shamcomm/worldInfo.hpp"
#include <vector>

Include dependency graph for LoadBalanceStrategy.hpp:

This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Classes
struct	shamrock::scheduler::TileWithLoad< Torder, Tweight >
struct	shamrock::scheduler::details::LoadBalancedTile< Torder, Tweight >
struct	shamrock::scheduler::details::LBMetric

Namespaces
namespace	shamrock
	namespace for the main framework

Functions
template<class Torder , class Tweight >
void	shamrock::scheduler::details::apply_ordering (std::vector< LoadBalancedTile< Torder, Tweight > > &lb_vec)
	Sort tiles by their ordering value.
template<class Torder , class Tweight >
std::vector< i32 >	shamrock::scheduler::details::lb_startegy_parallel_sweep (const std::vector< TileWithLoad< Torder, Tweight > > &lb_vector, i32 wsize)
	Load balance using parallel sweep strategy based on accumulated load.
template<class Torder , class Tweight >
std::vector< i32 >	shamrock::scheduler::details::lb_startegy_roundrobin (const std::vector< TileWithLoad< Torder, Tweight > > &lb_vector, i32 wsize)
	Load balance using round-robin strategy ignoring actual load values.
template<class Torder , class Tweight >
LBMetric	shamrock::scheduler::details::compute_LB_metric (const std::vector< TileWithLoad< Torder, Tweight > > &lb_vector, const std::vector< i32 > &new_owners, i32 world_size, f64 strategy_weight)
	Compute load balance quality metrics.
template<class Torder , class Tweight >
std::vector< i32 >	shamrock::scheduler::load_balance (std::vector< TileWithLoad< Torder, Tweight > > &&lb_vector, i32 world_size=shamcomm::world_size())
	load balance the input vector

Detailed Description

implementation of the hilbert curve load balancing

Author

Timothée David–Cléris (tim.s.nosp@m.hamr.nosp@m.ock@p.nosp@m.roto.nosp@m.n.me)

Definition in file LoadBalanceStrategy.hpp.

Function Documentation

apply_ordering()

template<class Torder , class Tweight >

void shamrock::scheduler::details::apply_ordering ( std::vector< LoadBalancedTile< Torder, Tweight > > &  lb_vec )

inline

Sort tiles by their ordering value.

Template Parameters

Torder	Ordering value type
Tweight	Load weight type

Parameters

lb_vec

Vector of load-balanced tiles to sort

Definition at line 58 of file LoadBalanceStrategy.hpp.

Here is the call graph for this function:

compute_LB_metric()

template<class Torder , class Tweight >

LBMetric shamrock::scheduler::details::compute_LB_metric ( const std::vector< TileWithLoad< Torder, Tweight > > &  lb_vector, const std::vector< i32 > &  new_owners, i32  world_size, f64  strategy_weight  )

inline

Compute load balance quality metrics.

Template Parameters

Torder	Ordering value type
Tweight	Load weight type

Parameters

lb_vector	Tiles with load information
new_owners	Owner assignments to evaluate
world_size	Number of workers

Returns

LBMetric Statistics about load distribution (min, max, mean, stddev)

Definition at line 222 of file LoadBalanceStrategy.hpp.

Here is the call graph for this function:

lb_startegy_parallel_sweep()

template<class Torder , class Tweight >

std::vector< i32 > shamrock::scheduler::details::lb_startegy_parallel_sweep ( const std::vector< TileWithLoad< Torder, Tweight > > &  lb_vector, i32  wsize  )

inline

Load balance using parallel sweep strategy based on accumulated load.

Template Parameters

Torder	Ordering value type
Tweight	Load weight type

Parameters

lb_vector	Tiles with load information
wsize	Number of workers

Returns

std::vector<i32> New owner assignments for each tile

Definition at line 75 of file LoadBalanceStrategy.hpp.

Here is the call graph for this function:

lb_startegy_roundrobin()

template<class Torder , class Tweight >

std::vector< i32 > shamrock::scheduler::details::lb_startegy_roundrobin ( const std::vector< TileWithLoad< Torder, Tweight > > &  lb_vector, i32  wsize  )

inline

Load balance using round-robin strategy ignoring actual load values.

Template Parameters

Torder	Ordering value type
Tweight	Load weight type

Parameters

lb_vector	Tiles with load information
wsize	Number of workers

Returns

std::vector<i32> New owner assignments for each tile

Definition at line 144 of file LoadBalanceStrategy.hpp.

Here is the call graph for this function:

load_balance()

template<class Torder , class Tweight >

std::vector< i32 > shamrock::scheduler::load_balance ( std::vector< TileWithLoad< Torder, Tweight > > &&  lb_vector, i32  world_size = shamcomm::world_size()  )

inline

load balance the input vector

Template Parameters

Torder	ordering value (hilbert, morton, ...)
Tweight	weight type

Parameters

lb_vector

Returns

std::vector<i32> The new owner list

Definition at line 274 of file LoadBalanceStrategy.hpp.

Here is the call graph for this function: