Parallel related data structures and functions. More...

#include <assert.h>
#include <mpi.h>
#include <vector>
#include "utils.hpp"

Classes
class	pargemslr::ParallelLogClass
	The data structure for parallel computing, including data structures for MPI and CUDA. More...

Macros
#define	PARGEMSLR_TIMES_NUM 60

#define	PARGEMSLR_BUILDTIME_PARTITION 0

#define	PARGEMSLR_BUILDTIME_IE 1

#define	PARGEMSLR_BUILDTIME_METIS 2

#define	PARGEMSLR_BUILDTIME_STRUCTURE 3

#define	PARGEMSLR_BUILDTIME_RCM 4

#define	PARGEMSLR_BUILDTIME_ILUT 5

#define	PARGEMSLR_BUILDTIME_LRC 6

#define	PARGEMSLR_BUILDTIME_ARNOLDI 7

#define	PARGEMSLR_BUILDTIME_BUILD_RES 8

#define	PARGEMSLR_BUILDTIME_SOLVELU 9

#define	PARGEMSLR_BUILDTIME_SOLVELU_L 10

#define	PARGEMSLR_BUILDTIME_SOLVELR 11

#define	PARGEMSLR_BUILDTIME_SOLVEEBFC 12

#define	PARGEMSLR_BUILDTIME_EXTRACTMAT 13

#define	PARGEMSLR_BUILDTIME_MOVEDATA 14

#define	PARGEMSLR_BUILDTIME_LOCALPERM 15

#define	PARGEMSLR_BUILDTIME_EMV 16

#define	PARGEMSLR_BUILDTIME_FMV 17

#define	PARGEMSLR_BUILDTIME_GEN_MAT 18

#define	PARGEMSLR_BUILDTIME_DECOMP 19

#define	PARGEMSLR_BUILDTIME_MGS 20

#define	PARGEMSLR_BUILDTIME_EBFC 21

#define	PARGEMSLR_PRECTIME_PRECOND 30

#define	PARGEMSLR_PRECTIME_ILUT 31

#define	PARGEMSLR_PRECTIME_ILUT_L 32

#define	PARGEMSLR_PRECTIME_LRC 33

#define	PARGEMSLR_PRECTIME_EMV 34

#define	PARGEMSLR_PRECTIME_FMV 35

#define	PARGEMSLR_PRECTIME_INNER 36

#define	PARGEMSLR_PRECTIME_MOVEDATA 37

#define	PARGEMSLR_PRECTIME_LOCALPERM 38

#define	PARGEMSLR_ITERTIME_AMV 40

#define	PARGEMSLR_ITERTIME_MGS 41

#define	PARGEMSLR_TOTAL_GEN_MAT_TIME 50

#define	PARGEMSLR_TOTAL_SETUP_TIME 51

#define	PARGEMSLR_TOTAL_SOLVE_TIME 52

#define	PARGEMSLR_BUILDTIME_BMV 53

#define	PARGEMSLR_BUILDTIME_CMV 54

#define	PARGEMSLR_BUILDTIME_SMV 55

#define	PARGEMSLR_PRECTIME_BMV 56

#define	PARGEMSLR_PRECTIME_CMV 57

#define	PARGEMSLR_PRECTIME_SMV 58

#define	PARGEMSLR_GLOBAL_FIRM_TIME_CALL(num, ...)

#define	PARGEMSLR_LOCAL_FIRM_TIME_CALL(num, ...)

#define	PARGEMSLR_FIRM_TIME_CALL(comm, num, ...)

#define	PARGEMSLR_GLOBAL_TIME_CALL(num, ...)

#define	PARGEMSLR_LOCAL_TIME_CALL(num, ...)

#define	PARGEMSLR_TIME_CALL(comm, num, ...)

#define	PARGEMSLR_PRINT_TIMING_RESULT(print_level, ...)

#define	PARGEMSLR_RESET_TIME

#define	PARGEMSLR_MPI_CALL(...)

#define	PARGEMSLR_CUDA_SYNCHRONIZE

#define	PARGEMSLR_GLOBAL_SEQUENTIAL_RUN(...)

Typedefs
typedef class pargemslr::ParallelLogClass	pargemslr::parallel_log
	The data structure for parallel computing, including data structures for MPI and CUDA. More...

typedef class pargemslr::ParallelLogClass *	pargemslr::parallel_logp

Functions
int	pargemslr::PargemslrSetOpenmpNumThreads (int nthreads)
	Set the OpenMP thread number for each MPI process. More...

int	pargemslr::PargemslrNLocalToNGlobal (int n_local, long int &n_start, long int &n_global, MPI_Comm &comm)
	Each MPI rank holds n_local, get the n_start and n_global. More...

int	pargemslr::PargemslrNLocalToNGlobal (int nrow_local, int ncol_local, long int &nrow_start, long int &ncol_start, long int &nrow_global, long int &ncol_global, MPI_Comm &comm)
	Each MPI rank holds two n_locals, get the n_starts and n_globals. More...

int	pargemslr::PargemslrInit (int argc, char **argv)
	Initilize MPI, OpenMP, and CUDA. Note that if you have already called MPI_Init, call other init functions instead. More...

int	pargemslr::PargemslrInitMpi (MPI_Comm comm)
	Initilize MPI data struct with MPI_Comm. More...

int	pargemslr::PargemslrInitOpenMP (int nthreads)
	Initilize OpenMP and MKL. More...

int	pargemslr::PargemslrInitCUDA ()
	Initilize CUDA. More...

int	pargemslr::PargemslrPrintParallelInfo ()
	Print the parallel information to output. More...

int	pargemslr::PargemslrFinalize ()
	Finalize MPI, OpenMP, and CUDA. Note that if you don't want to call MPI_Finalize here, call other finalize functions. More...

int	pargemslr::PargemslrFinalizeMpi ()
	Finalize MPI data structure. Note that MPI_Finalize won't be called here. More...

int	pargemslr::PargemslrFinalizeOpenMP ()
	Finalize OpenMP data structure. More...

int	pargemslr::PargemslrFinalizeCUDA ()
	Finalize CUDA data structure. More...

int	pargemslr::PargemslrMpiTime (MPI_Comm comm, double &t)
	Get current time using MPI_Wtime. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiIsend (T buf, int count, int dest, int tag, MPI_Comm comm, MPI_Request request)
	MPI_Isend. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiIsend (T buf, int count, int dest, int tag, MPI_Comm comm, MPI_Request request)
	MPI_Isend. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiIrecv (T buf, int count, int source, int tag, MPI_Comm comm, MPI_Request request)
	MPI_Irecv. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiIrecv (T buf, int count, int source, int tag, MPI_Comm comm, MPI_Request request)
	MPI_Irecv. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiSend (T *buf, int count, int dest, int tag, MPI_Comm comm)
	MPI_Send. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiSend (T *buf, int count, int dest, int tag, MPI_Comm comm)
	MPI_Send. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiRecv (T buf, int count, int source, int tag, MPI_Comm comm, MPI_Status status)
	MPI_Irecv. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiRecv (T buf, int count, int source, int tag, MPI_Comm comm, MPI_Status status)
	MPI_Irecv. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiBcast (T *buf, int count, int root, MPI_Comm comm)
	MPI_Bcast. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiBcast (T *buf, int count, int root, MPI_Comm comm)
	MPI_Bcast. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiScan (T sendbuf, T recvbuf, int count, MPI_Op op, MPI_Comm comm)
	MPI_Scan. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiScan (T sendbuf, T recvbuf, int count, MPI_Op op, MPI_Comm comm)
	MPI_Scan. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiReduce (T sendbuf, T recvbuf, int count, MPI_Op op, int root, MPI_Comm comm)
	MPI_Reduce. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiReduce (T sendbuf, T recvbuf, int count, MPI_Op op, int root, MPI_Comm comm)
	MPI_Reduce. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllreduce (T sendbuf, T recvbuf, int count, MPI_Op op, MPI_Comm comm)
	MPI_Allreduce. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllreduce (T sendbuf, T recvbuf, int count, MPI_Op op, MPI_Comm comm)
	MPI_Allreduce. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllreduceInplace (T *buf, int count, MPI_Op op, MPI_Comm comm)
	In place MPI_Allreduce. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllreduceInplace (T *buf, int count, MPI_Op op, MPI_Comm comm)
	In place MPI_Allreduce. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiGather (T sendbuf, int count, T recvbuf, int root, MPI_Comm comm)
	MPI_Gather. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiGather (T sendbuf, int count, T recvbuf, int root, MPI_Comm comm)
	MPI_Gather. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllgather (T sendbuf, int count, T recvbuf, MPI_Comm comm)
	MPI_Allgather. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllgather (T sendbuf, int count, T recvbuf, MPI_Comm comm)
	MPI_Allgather. More...

template<typename T >
std::enable_if<!PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllgatherv (T sendbuf, int count, T recvbuf, int recvcounts, int recvdisps, MPI_Comm comm)
	MPI_Allgatherv. More...

template<typename T >
std::enable_if< PargemslrIsComplex< T >::value, int >::type	pargemslr::PargemslrMpiAllgatherv (T sendbuf, int count, T recvbuf, int recvcounts, int recvdisps, MPI_Comm comm)
	MPI_Allgatherv. More...

template<typename T >
MPI_Datatype	pargemslr::PargemslrMpiDataType ()
	Get the MPI_Datatype. More...

template<>
MPI_Datatype	pargemslr::PargemslrMpiDataType< int > ()
	Get the MPI_Datatype. More...

template<>
MPI_Datatype	pargemslr::PargemslrMpiDataType< long int > ()
	Get the MPI_Datatype. More...

template<>
MPI_Datatype	pargemslr::PargemslrMpiDataType< float > ()
	Get the MPI_Datatype. More...

template<>
MPI_Datatype	pargemslr::PargemslrMpiDataType< double > ()
	Get the MPI_Datatype. More...

template<>
MPI_Datatype	pargemslr::PargemslrMpiDataType< complexs > ()
	Get the MPI_Datatype. More...

template<>
MPI_Datatype	pargemslr::PargemslrMpiDataType< complexd > ()
	Get the MPI_Datatype. More...

Macro Definition Documentation

◆ PARGEMSLR_FIRM_TIME_CALL

#define PARGEMSLR_FIRM_TIME_CALL	(	comm,
		num,
		...
	)

Value:

   {\
   pargemslr::PargemslrMpiTime( (comm), pargemslr::ParallelLogClass::_times_buffer_start[num]);\
   (__VA_ARGS__);\
   pargemslr::PargemslrMpiTime( (comm), pargemslr::ParallelLogClass::_times_buffer_end[num]);\
   pargemslr::ParallelLogClass::_times[num] += pargemslr::ParallelLogClass::_times_buffer_end[num] - pargemslr::ParallelLogClass::_times_buffer_start[num];\
}

◆ PARGEMSLR_GLOBAL_FIRM_TIME_CALL

#define PARGEMSLR_GLOBAL_FIRM_TIME_CALL	(	num,
		...
	)

Value:

   {\
   pargemslr::PargemslrMpiTime( (*(pargemslr::ParallelLogClass::_gcomm)), pargemslr::ParallelLogClass::_times_buffer_start[num]);\
   (__VA_ARGS__);\
   pargemslr::PargemslrMpiTime( (*(pargemslr::ParallelLogClass::_gcomm)), pargemslr::ParallelLogClass::_times_buffer_end[num]);\
   pargemslr::ParallelLogClass::_times[num] += pargemslr::ParallelLogClass::_times_buffer_end[num] - pargemslr::ParallelLogClass::_times_buffer_start[num];\
}

◆ PARGEMSLR_GLOBAL_SEQUENTIAL_RUN

#define PARGEMSLR_GLOBAL_SEQUENTIAL_RUN ( ... )

Value:

   {\
   for(int pgsri = 0 ; pgsri < pargemslr::parallel_log::_gsize ; pgsri++)\
   {\
      if( pargemslr::parallel_log::_grank == pgsri)\
      {\
         (__VA_ARGS__);\
      }\
      MPI_Barrier(*(pargemslr::parallel_log::_gcomm));\
   }\
}

◆ PARGEMSLR_GLOBAL_TIME_CALL

#define PARGEMSLR_GLOBAL_TIME_CALL	(	num,
		...
	)

Value:

   {\
   (__VA_ARGS__);\
}

◆ PARGEMSLR_LOCAL_FIRM_TIME_CALL

#define PARGEMSLR_LOCAL_FIRM_TIME_CALL	(	num,
		...
	)

Value:

   {\
   PARGEMSLR_CUDA_SYNCHRONIZE;\
   pargemslr::ParallelLogClass::_times_buffer_start[num] = MPI_Wtime();\
   (__VA_ARGS__);\
   PARGEMSLR_CUDA_SYNCHRONIZE;\
   pargemslr::ParallelLogClass::_times_buffer_end[num] = MPI_Wtime();\
   pargemslr::ParallelLogClass::_times[num] += pargemslr::ParallelLogClass::_times_buffer_end[num] - pargemslr::ParallelLogClass::_times_buffer_start[num];\
}

◆ PARGEMSLR_LOCAL_TIME_CALL

#define PARGEMSLR_LOCAL_TIME_CALL	(	num,
		...
	)

Value:

   {\
   (__VA_ARGS__);\
}

◆ PARGEMSLR_MPI_CALL

#define PARGEMSLR_MPI_CALL ( ... )

Value:

   {\
   (__VA_ARGS__);\
}

◆ PARGEMSLR_PRINT_TIMING_RESULT

#define PARGEMSLR_PRINT_TIMING_RESULT	(	print_level,
		...
	)

Value:

   {\
   if(__VA_ARGS__)\
   {\
      PARGEMSLR_PRINT("\n");\
      PargemslrPrintDashLine(pargemslr::pargemslr_global::_dash_line_width);\
      PARGEMSLR_PRINT("Time info:\n");\
      PARGEMSLR_PRINT("\tLoad matrix time:   %fs\n",pargemslr::ParallelLogClass::_times[PARGEMSLR_TOTAL_GEN_MAT_TIME]);\
      PARGEMSLR_PRINT("\tPartition time:     %fs\n",pargemslr::ParallelLogClass::_times[PARGEMSLR_BUILDTIME_PARTITION]);\
      PARGEMSLR_PRINT("\tSetup time:         %fs\n",pargemslr::ParallelLogClass::_times[PARGEMSLR_TOTAL_SETUP_TIME]-pargemslr::ParallelLogClass::_times[PARGEMSLR_BUILDTIME_PARTITION]);\
      PARGEMSLR_PRINT("\tSolve time:         %fs\n",pargemslr::ParallelLogClass::_times[PARGEMSLR_TOTAL_SOLVE_TIME]);\
      PARGEMSLR_PRINT("\tTotal time:         %fs\n",pargemslr::ParallelLogClass::_times[PARGEMSLR_TOTAL_SETUP_TIME]+pargemslr::ParallelLogClass::_times[PARGEMSLR_TOTAL_SOLVE_TIME]);\
      PargemslrPrintDashLine(pargemslr::pargemslr_global::_dash_line_width);\
      PARGEMSLR_PRINT("\n");\
   }\
}

◆ PARGEMSLR_TIME_CALL

#define PARGEMSLR_TIME_CALL	(	comm,
		num,
		...
	)

Value:

   {\
   (__VA_ARGS__);\
}

Typedef Documentation

◆ parallel_log

typedef class pargemslr::ParallelLogClass pargemslr::parallel_log

The data structure for parallel computing, including data structures for MPI and CUDA.
All CUDA information are shared, local MPI information can be different.

Function Documentation

◆ PargemslrFinalize()

int pargemslr::PargemslrFinalize ( )

Finalize MPI, OpenMP, and CUDA. Note that if you don't want to call MPI_Finalize here, call other finalize functions.

Returns: Return error message.

◆ PargemslrFinalizeCUDA()

int pargemslr::PargemslrFinalizeCUDA ( )

Finalize CUDA data structure.

Returns: Return error message.

◆ PargemslrFinalizeMpi()

int pargemslr::PargemslrFinalizeMpi ( )

Finalize MPI data structure. Note that MPI_Finalize won't be called here.

Returns: Return error message.

◆ PargemslrFinalizeOpenMP()

int pargemslr::PargemslrFinalizeOpenMP ( )

Finalize OpenMP data structure.

Returns: Return error message.

◆ PargemslrInit()

int pargemslr::PargemslrInit	(	int *	argc,
		char ***	argv
	)

Initilize MPI, OpenMP, and CUDA. Note that if you have already called MPI_Init, call other init functions instead.

Parameters

[in,out]	argc	Input of the main function.
[in,out]	argv	Input of the main function.

Returns: Return error message.

◆ PargemslrInitCUDA()

int pargemslr::PargemslrInitCUDA ( )

Initilize CUDA.

Returns: Return error message.

◆ PargemslrInitMpi()

int pargemslr::PargemslrInitMpi ( MPI_Comm comm )

Initilize MPI data struct with MPI_Comm. The ParGEMSLR package will duplicate this MPI_Comm.

Parameters

[in] comm The comm for ParGEMSLR, typically should be MPI_COMM_WORLD.

Returns: Return error message.

◆ PargemslrInitOpenMP()

int pargemslr::PargemslrInitOpenMP ( int nthreads )

Initilize OpenMP and MKL.

Parameters

[in] nthreads The max number of OpenMP threads.

Returns: Return error message.

◆ PargemslrMpiAllgather() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllgather	(	T *	sendbuf,
		int	count,
		T *	recvbuf,
		MPI_Comm	comm
	)

MPI_Allgather.

Parameters

[in]	sendbuf	Pointer to the send data.
[in]	count	Size of each single send/recv.
[out]	recvbuf	Pointer to the recv data.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiAllgather() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllgather	(	T *	sendbuf,
		int	count,
		T *	recvbuf,
		MPI_Comm	comm
	)

MPI_Allgather.

Parameters

[in]	sendbuf	Pointer to the send data.
[in]	count	Size of each single send/recv.
[out]	recvbuf	Pointer to the recv data.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiAllgatherv() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllgatherv	(	T *	sendbuf,
		int	count,
		T *	recvbuf,
		int *	recvcounts,
		int *	recvdisps,
		MPI_Comm	comm
	)

MPI_Allgatherv.

Parameters

[in]	sendbuf	Pointer to the send data.
[in]	count	Size of each single send/recv.
[out]	recvbuf	Pointer to the recv data.
[in]	recvcounts	Number of elements on each processor.
[in]	recvdisps	Displacement of elements on each processor.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiAllgatherv() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllgatherv	(	T *	sendbuf,
		int	count,
		T *	recvbuf,
		int *	recvcounts,
		int *	recvdisps,
		MPI_Comm	comm
	)

MPI_Allgatherv.

Parameters

[in]	sendbuf	Pointer to the send data.
[in]	count	Size of each single send/recv.
[out]	recvbuf	Pointer to the recv data.
[in]	recvcounts	Number of elements on each processor.
[in]	recvdisps	Displacement of elements on each processor.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiAllreduce() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllreduce	(	T *	sendbuf,
		T *	recvbuf,
		int	count,
		MPI_Op	op,
		MPI_Comm	comm
	)

MPI_Allreduce.

Parameters

[in]	sendbuf	Pointer to the send data.
[out]	recvbuf	Pointer to the recv data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiAllreduce() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllreduce	(	T *	sendbuf,
		T *	recvbuf,
		int	count,
		MPI_Op	op,
		MPI_Comm	comm
	)

MPI_Allreduce.

Parameters

[in]	sendbuf	Pointer to the send data.
[out]	recvbuf	Pointer to the recv data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiAllreduceInplace() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllreduceInplace	(	T *	buf,
		int	count,
		MPI_Op	op,
		MPI_Comm	comm
	)

In place MPI_Allreduce.

Parameters

[in,out]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiAllreduceInplace() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiAllreduceInplace	(	T *	buf,
		int	count,
		MPI_Op	op,
		MPI_Comm	comm
	)

In place MPI_Allreduce.

Parameters

[in,out]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiBcast() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiBcast	(	T *	buf,
		int	count,
		int	root,
		MPI_Comm	comm
	)

MPI_Bcast.

Parameters

[in,out]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	root	Root's MPI rank.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiBcast() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiBcast	(	T *	buf,
		int	count,
		int	root,
		MPI_Comm	comm
	)

MPI_Bcast.

Parameters

[in,out]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	root	Root's MPI rank.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiDataType()

template<typename T >

MPI_Datatype pargemslr::PargemslrMpiDataType ( )

Get the MPI_Datatype.

Returns: Return the MPI_Datatype.

◆ PargemslrMpiDataType< complexd >()

template<>

MPI_Datatype pargemslr::PargemslrMpiDataType< complexd > ( )

Get the MPI_Datatype.

Returns: Return the MPI_Datatype.

◆ PargemslrMpiDataType< complexs >()

template<>

MPI_Datatype pargemslr::PargemslrMpiDataType< complexs > ( )

Get the MPI_Datatype.

Returns: Return the MPI_Datatype.

◆ PargemslrMpiDataType< double >()

template<>

MPI_Datatype pargemslr::PargemslrMpiDataType< double > ( )

Get the MPI_Datatype.

Returns: Return the MPI_Datatype.

◆ PargemslrMpiDataType< float >()

template<>

MPI_Datatype pargemslr::PargemslrMpiDataType< float > ( )

Get the MPI_Datatype.

Returns: Return the MPI_Datatype.

◆ PargemslrMpiDataType< int >()

template<>

MPI_Datatype pargemslr::PargemslrMpiDataType< int > ( )

Get the MPI_Datatype.

Returns: Return the MPI_Datatype.

◆ PargemslrMpiDataType< long int >()

template<>

MPI_Datatype pargemslr::PargemslrMpiDataType< long int > ( )

Get the MPI_Datatype.

Returns: Return the MPI_Datatype.

◆ PargemslrMpiGather() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiGather	(	T *	sendbuf,
		int	count,
		T *	recvbuf,
		int	root,
		MPI_Comm	comm
	)

MPI_Gather.

Parameters

[in]	sendbuf	Pointer to the send data.
[in]	count	Size of each single send/recv.
[out]	recvbuf	Pointer to the recv data.
[in]	root	The MPI rank of the reciver.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiGather() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiGather	(	T *	sendbuf,
		int	count,
		T *	recvbuf,
		int	root,
		MPI_Comm	comm
	)

MPI_Gather.

Parameters

[in]	sendbuf	Pointer to the send data.
[in]	count	Size of each single send/recv.
[out]	recvbuf	Pointer to the recv data.
[in]	root	The MPI rank of the reciver.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiIrecv() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiIrecv	(	T *	buf,
		int	count,
		int	source,
		int	tag,
		MPI_Comm	comm,
		MPI_Request *	request
	)

MPI_Irecv.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	dest	Rank of the source MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.
[out]	request	MPI_Request.

Returns: Return error message.

◆ PargemslrMpiIrecv() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiIrecv	(	T *	buf,
		int	count,
		int	source,
		int	tag,
		MPI_Comm	comm,
		MPI_Request *	request
	)

MPI_Irecv.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	source	Rank of the source MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.
[out]	request	MPI_Request.

Returns: Return error message.

◆ PargemslrMpiIsend() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiIsend	(	T *	buf,
		int	count,
		int	dest,
		int	tag,
		MPI_Comm	comm,
		MPI_Request *	request
	)

MPI_Isend.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	dest	Rank of the target MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.
[out]	request	MPI_Request.

Returns: Return error message.

◆ PargemslrMpiIsend() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiIsend	(	T *	buf,
		int	count,
		int	dest,
		int	tag,
		MPI_Comm	comm,
		MPI_Request *	request
	)

MPI_Isend.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	dest	Rank of the target MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.
[out]	request	MPI_Request.

Returns: Return error message.

◆ PargemslrMpiRecv() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiRecv	(	T *	buf,
		int	count,
		int	source,
		int	tag,
		MPI_Comm	comm,
		MPI_Status *	status
	)

MPI_Irecv.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	source	Rank of the source MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.
[out]	status	MPI_Status.

Returns: Return error message.

◆ PargemslrMpiRecv() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiRecv	(	T *	buf,
		int	count,
		int	source,
		int	tag,
		MPI_Comm	comm,
		MPI_Status *	status
	)

MPI_Irecv.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	source	Rank of the source MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.
[out]	status	MPI_Status.

Returns: Return error message.

◆ PargemslrMpiReduce() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiReduce	(	T *	sendbuf,
		T *	recvbuf,
		int	count,
		MPI_Op	op,
		int	root,
		MPI_Comm	comm
	)

MPI_Reduce.

Parameters

[in]	sendbuf	Pointer to the send data.
[out]	recvbuf	Pointer to the recv data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	root	Root's MPI rank.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiReduce() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiReduce	(	T *	sendbuf,
		T *	recvbuf,
		int	count,
		MPI_Op	op,
		int	root,
		MPI_Comm	comm
	)

MPI_Reduce.

Parameters

[in]	sendbuf	Pointer to the send data.
[out]	recvbuf	Pointer to the recv data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	root	Root's MPI rank.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiScan() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiScan	(	T *	sendbuf,
		T *	recvbuf,
		int	count,
		MPI_Op	op,
		MPI_Comm	comm
	)

MPI_Scan.

Parameters

[in]	sendbuf	Pointer to the send data.
[out]	recvbuf	Pointer to the recv data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiScan() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiScan	(	T *	sendbuf,
		T *	recvbuf,
		int	count,
		MPI_Op	op,
		MPI_Comm	comm
	)

MPI_Scan.

Parameters

[in]	sendbuf	Pointer to the send data.
[out]	recvbuf	Pointer to the recv data.
[in]	count	Number of elements.
[in]	op	MPI_Op.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiSend() [1/2]

template<typename T >

std::enable_if<!PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiSend	(	T *	buf,
		int	count,
		int	dest,
		int	tag,
		MPI_Comm	comm
	)

MPI_Send.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	dest	Rank of the target MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.

Returns: Return error message.

◆ PargemslrMpiSend() [2/2]

template<typename T >

std::enable_if< PargemslrIsComplex< T >::value, int >::type pargemslr::PargemslrMpiSend	(	T *	buf,
		int	count,
		int	dest,
		int	tag,
		MPI_Comm	comm
	)

MPI_Send.

Parameters

[in]	buf	Pointer to the data.
[in]	count	Number of elements.
[in]	dest	Rank of the target MPI rank.
[in]	tag	Tag of the message.
[in]	comm	MPI_Comm.
[out]	request	MPI_Request.

Returns: Return error message.

◆ PargemslrMpiTime()

int pargemslr::PargemslrMpiTime	(	MPI_Comm	comm,
		double &	t
	)

Get current time using MPI_Wtime.

Parameters

[in]	comm	The MPI_Comm.
[out]	t	The time.

Returns: Return error message.

◆ PargemslrNLocalToNGlobal() [1/2]

int pargemslr::PargemslrNLocalToNGlobal	(	int	n_local,
		long int &	n_start,
		long int &	n_global,
		MPI_Comm &	comm
	)

Each MPI rank holds n_local, get the n_start and n_global.

Parameters

[in]	n_local	The local size.
[out]	n_start	The start index.
[out]	n_global	The global size.
[out]	comm	The MPI_comm.

Returns: Return error message.

◆ PargemslrNLocalToNGlobal() [2/2]

int pargemslr::PargemslrNLocalToNGlobal	(	int	nrow_local,
		int	ncol_local,
		long int &	nrow_start,
		long int &	ncol_start,
		long int &	nrow_global,
		long int &	ncol_global,
		MPI_Comm &	comm
	)

Each MPI rank holds two n_locals, get the n_starts and n_globals.

Parameters

[in]	nrow_local	The first local size.
[in]	ncol_local	The second local size.
[out]	nrow_start	The first start index.
[out]	ncol_start	The second start index.
[out]	nrow_global	The first global size.
[out]	ncol_global	The second global size.
[out]	comm	The MPI_comm.

Returns: Return error message.

◆ PargemslrPrintParallelInfo()

int pargemslr::PargemslrPrintParallelInfo ( )

Print the parallel information to output.

Returns: Return error message.

◆ PargemslrSetOpenmpNumThreads()

int pargemslr::PargemslrSetOpenmpNumThreads ( int nthreads )

Set the OpenMP thread number for each MPI process.

Parameters

[in] nthreads The number of threads per MPI process.

Returns: Return error message.

Classes

Macros

Typedefs

Functions

Macro Definition Documentation

◆ PARGEMSLR_FIRM_TIME_CALL

◆ PARGEMSLR_GLOBAL_FIRM_TIME_CALL

◆ PARGEMSLR_GLOBAL_SEQUENTIAL_RUN

◆ PARGEMSLR_GLOBAL_TIME_CALL

◆ PARGEMSLR_LOCAL_FIRM_TIME_CALL

◆ PARGEMSLR_LOCAL_TIME_CALL

◆ PARGEMSLR_MPI_CALL

◆ PARGEMSLR_PRINT_TIMING_RESULT

◆ PARGEMSLR_TIME_CALL

Typedef Documentation

◆ parallel_log

Function Documentation

◆ PargemslrFinalize()

◆ PargemslrFinalizeCUDA()

◆ PargemslrFinalizeMpi()

◆ PargemslrFinalizeOpenMP()

◆ PargemslrInit()

◆ PargemslrInitCUDA()

◆ PargemslrInitMpi()

◆ PargemslrInitOpenMP()

◆ PargemslrMpiAllgather() [1/2]

◆ PargemslrMpiAllgather() [2/2]

◆ PargemslrMpiAllgatherv() [1/2]

◆ PargemslrMpiAllgatherv() [2/2]

◆ PargemslrMpiAllreduce() [1/2]

◆ PargemslrMpiAllreduce() [2/2]

◆ PargemslrMpiAllreduceInplace() [1/2]

◆ PargemslrMpiAllreduceInplace() [2/2]

◆ PargemslrMpiBcast() [1/2]

◆ PargemslrMpiBcast() [2/2]

◆ PargemslrMpiDataType()

◆ PargemslrMpiDataType< complexd >()

◆ PargemslrMpiDataType< complexs >()

◆ PargemslrMpiDataType< double >()

◆ PargemslrMpiDataType< float >()

◆ PargemslrMpiDataType< int >()

◆ PargemslrMpiDataType< long int >()

◆ PargemslrMpiGather() [1/2]

◆ PargemslrMpiGather() [2/2]

◆ PargemslrMpiIrecv() [1/2]

◆ PargemslrMpiIrecv() [2/2]

◆ PargemslrMpiIsend() [1/2]

◆ PargemslrMpiIsend() [2/2]

◆ PargemslrMpiRecv() [1/2]

◆ PargemslrMpiRecv() [2/2]

◆ PargemslrMpiReduce() [1/2]

◆ PargemslrMpiReduce() [2/2]

◆ PargemslrMpiScan() [1/2]

◆ PargemslrMpiScan() [2/2]

◆ PargemslrMpiSend() [1/2]

◆ PargemslrMpiSend() [2/2]

◆ PargemslrMpiTime()

◆ PargemslrNLocalToNGlobal() [1/2]

◆ PargemslrNLocalToNGlobal() [2/2]

◆ PargemslrPrintParallelInfo()

◆ PargemslrSetOpenmpNumThreads()