Collaboration diagram for MPIContext:

Public Member Functions
	MPIContext ()

int	GetRank () const

int	GetSize () const

	MPIContext ()

int	GetRank () const

int	GetSize () const

	MPIContext ()

int	GetRank () const

int	GetSize () const

	MPIContext ()

int	GetRank () const

int	GetSize () const

	MPIContext ()

int	GetRank () const

int	GetSize () const

Public Attributes
int	drank

int	dsize

int	dcolor

int	grank

int	gsize

int	kpar

int	nproc_in_pool

int	my_pool

int	rank_in_pool

int	nstogroup

int	MY_BNDGROUP

int	rank_in_stogroup

int	nproc_in_stogroup

int	KPAR

int	NPROC_IN_POOL

int	MY_POOL

int	RANK_IN_POOL

Private Attributes
int	_rank

int	_size

Detailed Description

The tested functions are wrappers of MPI_Bcast in ABACUS, as defined in source_base/parallel_common.h. The source is process 0 in all MPI_Bcast wrappers.

The tested functions are: i. Parallel_Global::split_diag_world(), which is used in David diagonalization in pw basis calculation. ii. Parallel_Global::split_grid_world() iii. Parallel_Global::MyProd(std::complex<double> *in,std::complex<double> *inout,int *len,MPI_Datatype *dptr); iv. Parallel_Global::init_pools(); v. Parallel_Global::divide_pools(void);

The tested functions are mainly wrappers of MPI_Allreduce and MPI_Allgather in ABACUS, as defined in source_base/ parallel_reduce.h.

The logic to test MPI_Allreduce wrapper functions is to calculate the sum of the total array in two ways, one by using MPI_Allreduce with only 1 number, another one by using MPI_Allreduce with n numbers. The total array is deemed as the sum of local arrays with the same length.

ReduceIntAll: Tests two variations of reduce_all()
ReduceDoubleAll: Tests two variations of reduce_all()
ReduceComplexAll: Tests two variations of reduce_complex_all()
GatherIntAll: Tests gather_int_all() and gather_min_int_all()
GatherDoubleAll: Tests gather_min_double_all() and gather_max_double_all()
ReduceIntDiag: Tests reduce_int_diag()
ReduceDoubleDiag: Tests reduce_double_diag()
ReduceIntGrid: Tests reduce_int_grid()
ReduceDoubleGrid: Tests reduce_double_grid() 10.ReduceDoublePool: Tests two variations of reduce_pool() and two variations of reduce_double_allpool() 11.ReduceComplexPool: Tests two variations of reduce_pool() 12.GatherDoublePool: Tests gather_min_double_pool() and gather_max_double_pool()

The tested functions: i. Parallel_Global::init_pools() is the public interface to call the private function Parallel_Global::divide_pools(), which divide all processes into KPAR groups. ii.Parallel_Kpoints::kinf() is the public interface to call another three functions: get_nks_pool(), get_startk_pool(), get_whichpool(), which divide all kpoints into KPAR groups. iii.Parallel_Kpoints::gatherkvec() is an interface to gather kpoints vectors from all processors. The default number of processes is set to 4 in parallel_kpoints_test.sh. One may modify it to do more tests, or adapt this unittest to local environment.

Test fixture for class Parallel_K2D

Constructor & Destructor Documentation

◆ MPIContext() [1/5]

MPIContext::MPIContext ( )

inline

◆ MPIContext() [2/5]

MPIContext::MPIContext ( )

inline

◆ MPIContext() [3/5]

MPIContext::MPIContext ( )

inline

◆ MPIContext() [4/5]

MPIContext::MPIContext ( )

inline

◆ MPIContext() [5/5]

MPIContext::MPIContext ( )

inline

Member Function Documentation

◆ GetRank() [1/5]

int MPIContext::GetRank ( ) const

inline

Here is the caller graph for this function:

◆ GetRank() [2/5]

int MPIContext::GetRank ( ) const

inline

◆ GetRank() [3/5]

int MPIContext::GetRank ( ) const

inline

◆ GetRank() [4/5]

int MPIContext::GetRank ( ) const

inline