mpi_gbl Module Reference

MPI library interfaces. More...

Data Types
interface	mpi_mod_allgather
interface	mpi_mod_allgatherv_in_place
interface	mpi_mod_bcast
interface	mpi_mod_file_read_cfp
interface	mpi_mod_file_set_view
interface	mpi_mod_file_write
interface	mpi_mod_gatherv
interface	mpi_mod_isend
interface	mpi_mod_recv
interface	mpi_mod_recv_dynamic
interface	mpi_mod_rotate_arrays_around_ring
interface	mpi_mod_rotate_cfp_arrays_around_ring
interface	mpi_mod_scan_sum
interface	mpi_mod_send
interface	mpi_mod_send_dynamic
interface	mpi_reduce_inplace_sum_cfp
interface	mpi_reduce_sum
interface	mpi_reduceall_inplace_sum_cfp
interface	mpi_reduceall_max
interface	mpi_reduceall_min
interface	mpi_reduceall_sum_cfp
type	MPIBlockArray
	Convenience MPI type wrapper with garbage collection. More...
interface	naive_mpi_reduce_inplace_sum

Functions/Subroutines
integer(mpiint) function	mpi_mod_comm (comm_opt)
	Comm or default comm. More...
subroutine	mpi_mod_rank (rank, comm)
	Wrapper around MPI_Comm_rank. More...
subroutine	mpi_mod_barrier (error, comm)
	Interface to the routine MPI_BARRIER. More...
subroutine	mpi_xermsg (mod_name, routine_name, err_msg, err, level)
	Analogue of the xermsg routine. This routine uses xermsg to first print out the error message and then either aborts the program or exits the routine depending on the level of the message (warning/error) which is defined in the same way as in xermsg. More...
subroutine	check_mpi_running
	This is a lightweight routine which aborts the program if MPI is not found running. More...
subroutine	mpi_mod_print_info (u)
	Display information about current MPI setup. More...
subroutine	mpi_mod_start (do_stdout, allow_shared_memory)
	Initializes MPI, assigns the rank for each process and finds out how many processors we're using. It also maps the current floating point precision (kind=cfp) to the corresponding MPI numeric type. This routine also sets the unit for standard output that each rank will use (the value of stdout). In case of serial run the value stdout is not set here and is left to the default value input_unit as specified in the module const. More...
subroutine	setup_intranode_local_mem_communicator
	It is not used anywhere in the code yet but it might become useful later on. It creates intra-node communicators without the memory sharing capability. More...
logical function	strings_are_same (str1, str2, length)
subroutine	mpi_mod_finalize
	Terminates the MPI session and stops the program. It is a blocking routine. More...
subroutine	mpi_mod_create_block_array (this, n, elemtype)
	Set up a new MPI type describing a long array composed of blocks. More...
subroutine	mpi_mod_free_block_array (this)
	Release MPI handle of the block array type. More...
subroutine	mpi_mod_bcast_logical (val, from, comm)
subroutine	mpi_mod_bcast_int32 (val, from, comm)
subroutine	mpi_mod_bcast_int64 (val, from, comm)
subroutine	mpi_mod_bcast_int32_array (val, from, comm)
subroutine	mpi_mod_bcast_int64_array (val, from, comm)
subroutine	mpi_mod_bcast_int32_3d_array (val, from, comm)
subroutine	mpi_mod_bcast_int64_3d_array (val, from, comm)
subroutine	mpi_mod_bcast_wp (val, from, comm)
subroutine	mpi_mod_bcast_ep (val, from, comm)
subroutine	mpi_mod_bcast_wp_array (val, from, comm)
subroutine	mpi_mod_bcast_ep_array (val, from, comm)
subroutine	mpi_mod_bcast_wp_3d_array (val, from, comm)
subroutine	mpi_mod_bcast_ep_3d_array (val, from, comm)
subroutine	mpi_mod_bcast_character (val, from, comm)
subroutine	mpi_mod_bcast_character_array (val, from, comm)
subroutine	mpi_mod_send_wp_array (to, buffer, tag, n, comm)
subroutine	mpi_mod_send_ep_array (to, buffer, tag, n, comm)
subroutine	mpi_mod_send_int32_array (to, buffer, tag, n, comm)
subroutine	mpi_mod_send_int64_array (to, buffer, tag, n, comm)
subroutine	mpi_mod_isend_int32_array (to, buffer, tag, n, comm)
subroutine	mpi_mod_isend_int64_array (to, buffer, tag, n, comm)
subroutine	mpi_mod_recv_wp_array (from, tag, buffer, n, comm)
subroutine	mpi_mod_recv_ep_array (from, tag, buffer, n, comm)
subroutine	mpi_mod_recv_int32_array (from, tag, buffer, n, comm)
subroutine	mpi_mod_recv_int64_array (from, tag, buffer, n, comm)
subroutine	mpi_mod_send_wp_array_dynamic (to, buffer, tag, comm)
subroutine	mpi_mod_send_ep_array_dynamic (to, buffer, tag, comm)
subroutine	mpi_mod_recv_wp_array_dynamic (from, tag, buffer, n, comm)
subroutine	mpi_mod_recv_ep_array_dynamic (from, tag, buffer, n, comm)
subroutine	mpi_mod_file_open_read (filename, fh, ierr, comm)
subroutine	mpi_mod_file_set_view_wp (fh, disp, ierr, wp_dummy)
	Sets view for a file containing floating point numbers with kind=wp. More...
subroutine	mpi_mod_file_set_view_ep (fh, disp, ierr, ep_dummy)
	Sets view for a file containing floating point numbers with kind=ep. More...
subroutine	mpi_mod_file_read_wp (fh, buffer, buflen, ierr)
subroutine	mpi_mod_file_read_ep (fh, buffer, buflen, ierr)
subroutine	mpi_mod_file_close (fh, ierr)
subroutine	mpi_reduce_inplace_sum_wp (buffer, nelem)
subroutine	mpi_reduce_inplace_sum_ep (buffer, nelem)
subroutine	mpi_reduce_sum_wp (src, dest, nelem)
subroutine	mpi_reduce_sum_ep (src, dest, nelem)
subroutine	mpi_reduceall_sum_wp (src, dest, nelem, comm)
subroutine	mpi_reduceall_sum_ep (src, dest, nelem, comm)
subroutine	mpi_reduceall_max_int32 (src, dest, comm_opt)
	Choose largest elements among processes from 32-bit integer arrays. More...
subroutine	mpi_reduceall_max_int64 (src, dest, comm_opt)
	Choose largest elements among processes from 64-bit integer arrays. More...
subroutine	mpi_reduceall_min_int32 (src, dest, comm_opt)
	Choose smallest elements among processes from 32-bit integer arrays. More...
subroutine	mpi_reduceall_min_int64 (src, dest, comm_opt)
	Choose smallest elements among processes from 64-bit integer arrays. More...
subroutine	naive_mpi_reduce_inplace_sum_wp (buffer, nelem, split)
subroutine	naive_mpi_reduce_inplace_sum_ep (buffer, nelem, split)
subroutine	mpi_reduceall_inplace_sum_wp (buffer, nelem, comm)
subroutine	mpi_reduceall_inplace_sum_ep (buffer, nelem, comm)
subroutine	mpi_mod_rotate_arrays_around_ring_wp (elem_count, int_array, wp_array, max_num_elements, comm_opt)
	This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation. More...
subroutine	mpi_mod_rotate_arrays_around_ring_ep (elem_count, int_array, ep_array, max_num_elements, comm_opt)
	This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation. More...
subroutine	mpi_mod_rotate_wp_arrays_around_ring (elem_count, wp_array, max_num_elements, communicator)
	This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation. More...
subroutine	mpi_mod_rotate_ep_arrays_around_ring (elem_count, ep_array, max_num_elements, communicator)
	This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation. More...
subroutine	mpi_mod_rotate_int_arrays_around_ring (elem_count, int_array, max_num_elements, communicator)
	This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation. More...
subroutine	mpi_mod_allgather_int32 (send, receive, comm)
	All gather for one integer send from every process. More...
subroutine	mpi_mod_allgather_int64 (send, receive, comm)
subroutine	mpi_mod_allgatherv_in_place_int32_array (sendcount, receive, comm)
subroutine	mpi_mod_allgatherv_in_place_int64_array (sendcount, receive, comm)
subroutine	mpi_mod_allgatherv_in_place_real64_array (sendcount, receive, comm)
subroutine	mpi_mod_allgatherv_in_place_real128_array (sendcount, receive, comm)
subroutine	mpi_mod_allgather_character (send, receive, comm)
	All gather for one string of length MPI_MAX_PROCESSOR_NAME send from every process. More...
subroutine	mpi_mod_gatherv_int32 (sendbuf, recvbuf, recvcounts, displs, to, comm)
	Gather for 32bit integer array send from every process and received on a given rank. More...
subroutine	mpi_mod_gatherv_int64 (sendbuf, recvbuf, recvcounts, displs, to, comm)
	Gather for 64bit integer array send from every process and received on a given rank. More...
subroutine	mpi_mod_gatherv_real64 (sendbuf, recvbuf, recvcounts, displs, to, comm)
	Gather for 64bit float array send from every process and received on a given rank. More...
subroutine	mpi_mod_gatherv_real128 (sendbuf, recvbuf, recvcounts, displs, to, comm)
	Gather for 128bit float array send from every process and received on a given rank. More...
real(wp) function	mpi_mod_wtime ()
	Interface to the routine MPI_BARRIER. More...
subroutine	mpi_mod_file_open_write (filename, fh, ierr, comm)
subroutine	mpi_mod_file_write_int32 (fh, n)
	Let master process write one 4-byte integer to stream file. More...
subroutine	mpi_mod_file_write_array1d_int32 (fh, array, length)
	Let master process write array of 4-byte integers to stream file. More...
subroutine	mpi_mod_file_write_array2d_int32 (fh, array, length1, length2)
	Let master process write 2d array of 4-byte integers to stream file. More...
subroutine	mpi_mod_file_write_real64 (fh, x)
	Let master process write an 8-byte real to stream file. More...
subroutine	mpi_mod_file_write_array1d_real64 (fh, array, length)
	Let master process write array of 8-byte reals to stream file. More...
subroutine	mpi_mod_file_write_array2d_real64 (fh, array, length1, length2)
	Let master process write 2d array of 8-byte reals to stream file. More...
subroutine	mpi_mod_file_write_array3d_real64 (fh, array, length1, length2, length3)
	Let master process write 3d array of 8-byte reals to stream file. More...
subroutine	mpi_mod_file_set_size (fh, sz)
	Collectively resize file. More...
subroutine	mpi_mod_file_write_darray2d_real64 (fh, m, n, nprow, npcol, rb, cb, locA, myrows, mycols, comm)
	Write block-cyclically distributed matrix to a stream file. More...
subroutine	mpi_mod_scan_sum_int32 (src, dst, comm_opt)
	Wrapper aroung MPI_Scan. More...
subroutine	mpi_mod_scan_sum_int64 (src, dst, comm_opt)
	Wrapper aroung MPI_Scan. More...

Variables
integer, parameter, public	mpiint = kind(0)
integer, parameter, public	mpiaddr = longint
integer, parameter, public	mpicnt = longint
integer, parameter, public	mpiofs = longint
logical, public, protected	mpi_running = .false.
	This can be set only once at the beginning of the program run by a call to MPI_MOD_START. The value of this variable is accessed by MPI_MOD_CHECK which aborts the program if MPI is not running. This variable stays .false. if the library was compiled without MPI support. More...
logical, public, protected	mpi_started = .false.
	This is similar to mpi_running, but will be always set to true once MPI_MOD_START is called, whether or not the library was built with MPI support. So, while mpi_running can be used to determine whether the program is running under MPI, the variable mpi_started indicates whether the call to MPI_MOD_START had been already done (even in serial build). More...
integer(kind=mpiint), public, protected	nprocs = -1
	Total number of processes. Set by mpi_mod_start. More...
integer(kind=mpiint), public, protected	myrank = -1
	The local process number (rank). Set by mpi_mod_start. More...
integer(kind=mpiint), public, protected	local_nprocs = 1
	Total number of processes on the node this task is bound to. Set by mpi_mod_start. More...
integer(kind=mpiint), public, protected	local_rank = 0
	The local process number (rank) on the node this task is bound to. Set by mpi_mod_start. More...
logical, public, protected	shared_enabled = .false.
	The routine mpi_start creates the shared_communicator which groups all tasks sitting on the same node. If we're using MPI 3.0 standard the shared_communicator enables creation of shared memory areas. In this case shared_enabled is set to .true. If we're not using MPI 3.0 then shared_enabled is set to .false. More...
integer(kind=mpiint), parameter, public	master = 0
	ID of the master process. More...
integer, public, protected	max_data_count = huge(dummy_32bit_integer)
	Largest integer (e.g. number of elements) supported by the standard MPI 32-bit integer API. More...
character(len=mpi_max_processor_name), parameter, public, protected	procname = "N/A"
	Name of the processor on which the current process is running. More...
integer(kind=mpiint), public, protected	shared_communicator = -1
	Intra-node communicator created by mpi_mod_start. More...

Detailed Description

MPI library interfaces.

Authors

Z Masin, J Benda

Date

2014 - 2019

Convenience generic interfaces for MPI calls. Specific implementations available for 4-byte and 8-byte integers and 8-byte and 16-byte real data types.

Warning

This module is also used by SCATCI but always in DOUBLE PRECISION. However, SCATCI must always be linked to the integral library of precision used to calculate the integrals file. Therefore, in order to ensure correct functionality of SCATCI we need to ensure this module supports calls to both DOUBLE and QUAD precision MPI routines within the same program run. This is achieved by interfacing all routines manipulating floating point variables for both wp and ep precisions.

Function/Subroutine Documentation

check_mpi_running()

subroutine mpi_gbl::check_mpi_running

This is a lightweight routine which aborts the program if MPI is not found running.

mpi_mod_allgather_character()

subroutine mpi_gbl::mpi_mod_allgather_character	(	character(len=*), intent(in)	send,
		character(len=*), dimension(:), intent(out)	receive,
		integer(mpiint), intent(in), optional	comm
	)

All gather for one string of length MPI_MAX_PROCESSOR_NAME send from every process.

Here is the call graph for this function:

mpi_mod_allgather_int32()

subroutine mpi_gbl::mpi_mod_allgather_int32	(	integer(int32), intent(in)	send,
		integer(int32), dimension(:), intent(out)	receive,
		integer(mpiint), intent(in), optional	comm
	)

All gather for one integer send from every process.

Here is the call graph for this function:

mpi_mod_allgather_int64()

subroutine mpi_gbl::mpi_mod_allgather_int64	(	integer(int64), intent(in)	send,
		integer(int64), dimension(:), intent(out)	receive,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_allgatherv_in_place_int32_array()

subroutine mpi_gbl::mpi_mod_allgatherv_in_place_int32_array	(	integer(mpiint), intent(in)	sendcount,
		integer(int32), dimension(:), intent(inout)	receive,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_allgatherv_in_place_int64_array()

subroutine mpi_gbl::mpi_mod_allgatherv_in_place_int64_array	(	integer(mpiint), intent(in)	sendcount,
		integer(int64), dimension(:), intent(inout)	receive,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_allgatherv_in_place_real128_array()

subroutine mpi_gbl::mpi_mod_allgatherv_in_place_real128_array	(	integer(mpiint), intent(in)	sendcount,
		real(real128), dimension(:), intent(inout)	receive,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_allgatherv_in_place_real64_array()

subroutine mpi_gbl::mpi_mod_allgatherv_in_place_real64_array	(	integer(mpiint), intent(in)	sendcount,
		real(real64), dimension(:), intent(inout)	receive,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_barrier()

subroutine mpi_gbl::mpi_mod_barrier	(	integer, intent(out)	error,
		integer(mpiint), intent(in), optional	comm
	)

Interface to the routine MPI_BARRIER.

Here is the call graph for this function:

mpi_mod_bcast_character()

subroutine mpi_gbl::mpi_mod_bcast_character	(	character(len=*), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_character_array()

subroutine mpi_gbl::mpi_mod_bcast_character_array	(	character(len=*), dimension(:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_ep()

subroutine mpi_gbl::mpi_mod_bcast_ep	(	real(ep), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_ep_3d_array()

subroutine mpi_gbl::mpi_mod_bcast_ep_3d_array	(	real(ep), dimension(:,:,:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_ep_array()

subroutine mpi_gbl::mpi_mod_bcast_ep_array	(	real(ep), dimension(:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_int32()

subroutine mpi_gbl::mpi_mod_bcast_int32	(	integer(int32), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_int32_3d_array()

subroutine mpi_gbl::mpi_mod_bcast_int32_3d_array	(	integer(int32), dimension(:,:,:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_int32_array()

subroutine mpi_gbl::mpi_mod_bcast_int32_array	(	integer(int32), dimension(:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_int64()

subroutine mpi_gbl::mpi_mod_bcast_int64	(	integer(int64), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_int64_3d_array()

subroutine mpi_gbl::mpi_mod_bcast_int64_3d_array	(	integer(int64), dimension(:,:,:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_int64_array()

subroutine mpi_gbl::mpi_mod_bcast_int64_array	(	integer(int64), dimension(:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_logical()

subroutine mpi_gbl::mpi_mod_bcast_logical	(	logical, intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_wp()

subroutine mpi_gbl::mpi_mod_bcast_wp	(	real(wp), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_wp_3d_array()

subroutine mpi_gbl::mpi_mod_bcast_wp_3d_array	(	real(wp), dimension(:,:,:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_bcast_wp_array()

subroutine mpi_gbl::mpi_mod_bcast_wp_array	(	real(wp), dimension(:), intent(inout)	val,
		integer(mpiint), intent(in)	from,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_comm()

integer(mpiint) function mpi_gbl::mpi_mod_comm

(

integer(mpiint), intent(in), optional

comm_opt

)

Comm or default comm.

Authors

J Benda

Date

2019

Auxiliary function used in other subroutines for brevity. If an argument is given, it will return its value. If no argument is given, the function returns MPI_COMM_WORLD.

mpi_mod_create_block_array()

subroutine mpi_gbl::mpi_mod_create_block_array	(	class(mpiblockarray), intent(inout)	this,
		integer, intent(in)	n,
		integer(mpiint), intent(in)	elemtype
	)

Set up a new MPI type describing a long array composed of blocks.

Authors

J Benda

Date

2019

Creates a MPI type handle describing a structured data type that can be used to transfer very long arrays of data. The array is represented as a set of blocks, whose element count does not exceed the maximal value of the MPI integer type. To avoid resource leaks, the returned handle should be released by a call to MPI_Type_free.

Parameters

[in]	this	Block array object to initialize.
[in]	n	Number of elements in the array.
[in]	elemtype	MPI datatype handle of elements of the array.

Here is the call graph for this function:

mpi_mod_file_close()

subroutine mpi_gbl::mpi_mod_file_close	(	integer(mpiint), intent(inout)	fh,
		integer, intent(out)	ierr
	)

mpi_mod_file_open_read()

subroutine mpi_gbl::mpi_mod_file_open_read	(	character(len=*), intent(in)	filename,
		integer(mpiint), intent(out)	fh,
		integer, intent(out)	ierr,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_file_open_write()

subroutine mpi_gbl::mpi_mod_file_open_write	(	character(len=*), intent(in)	filename,
		integer(mpiint), intent(out)	fh,
		integer, intent(out)	ierr,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_file_read_ep()

subroutine mpi_gbl::mpi_mod_file_read_ep	(	integer, intent(in)	fh,
		real(kind=ep1), dimension(:), intent(out)	buffer,
		integer, intent(in)	buflen,
		integer, intent(out)	ierr
	)

Here is the call graph for this function:

mpi_mod_file_read_wp()

subroutine mpi_gbl::mpi_mod_file_read_wp	(	integer, intent(in)	fh,
		real(kind=wp), dimension(:), intent(out)	buffer,
		integer, intent(in)	buflen,
		integer, intent(out)	ierr
	)

Here is the call graph for this function:

mpi_mod_file_set_size()

subroutine mpi_gbl::mpi_mod_file_set_size	(	integer(mpiint), intent(in)	fh,
		integer, intent(in)	sz
	)

Collectively resize file.

Authors

J Benda

Date

2019

Change size of a file; mostly used to truncate the file to zero size. Only available when compiled with MPI.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	sz	New size of the file.

Here is the call graph for this function:

mpi_mod_file_set_view_ep()

subroutine mpi_gbl::mpi_mod_file_set_view_ep	(	integer, intent(in)	fh,
		integer, intent(in)	disp,
		integer, intent(out)	ierr,
		real(kind=ep), intent(in)	ep_dummy
	)

Sets view for a file containing floating point numbers with kind=ep.

Here is the call graph for this function:

mpi_mod_file_set_view_wp()

subroutine mpi_gbl::mpi_mod_file_set_view_wp	(	integer, intent(in)	fh,
		integer, intent(in)	disp,
		integer, intent(out)	ierr,
		real(kind=wp), intent(in)	wp_dummy
	)

Sets view for a file containing floating point numbers with kind=wp.

Here is the call graph for this function:

mpi_mod_file_write_array1d_int32()

subroutine mpi_gbl::mpi_mod_file_write_array1d_int32	(	integer(mpiint), intent(in)	fh,
		integer(int32), dimension(length), intent(in)	array,
		integer, intent(in)	length
	)

Let master process write array of 4-byte integers to stream file.

Authors

J Benda

Date

2019

Uses blocking collective MPI output routine to write array of integers. Only master process will do the writing, other processes behave as if the provided length parameter was zero. The integers are written at the master's file position, which gets updated. Positions of other processes in the file remain as they were.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	array	Integer array to write.
[in]	length	Number of elements in the array.

Here is the call graph for this function:

mpi_mod_file_write_array1d_real64()

subroutine mpi_gbl::mpi_mod_file_write_array1d_real64	(	integer(mpiint), intent(in)	fh,
		real(real64), dimension(length), intent(in)	array,
		integer, intent(in)	length
	)

Let master process write array of 8-byte reals to stream file.

Authors

J Benda

Date

2019

Uses blocking collective MPI output routine to write array of reals. Only master process will do the writing, other processes behave as if the provided length parameter was zero. The reals are written at the master's file position, which gets updated. Positions of other processes in the file remain as they were.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	array	Real array to write.
[in]	length	Number of elements in the array.

Here is the call graph for this function:

mpi_mod_file_write_array2d_int32()

subroutine mpi_gbl::mpi_mod_file_write_array2d_int32	(	integer(mpiint), intent(in)	fh,
		integer(int32), dimension(length1, length2), intent(in)	array,
		integer, intent(in)	length1,
		integer, intent(in)	length2
	)

Let master process write 2d array of 4-byte integers to stream file.

Authors

J Benda

Date

2019

Convenience wrapper around mpi_mod_file_write_array1d_int32, which does the real work.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	array	Two-dimensional integer array to write.
[in]	length1	Number of rows (= leading dimension).
[in]	length2	Number of columns.

Here is the call graph for this function:

mpi_mod_file_write_array2d_real64()

subroutine mpi_gbl::mpi_mod_file_write_array2d_real64	(	integer(mpiint), intent(in)	fh,
		real(real64), dimension(length1, length2), intent(in)	array,
		integer, intent(in)	length1,
		integer, intent(in)	length2
	)

Let master process write 2d array of 8-byte reals to stream file.

Authors

J Benda

Date

2019

Convenience wrapper around mpi_mod_file_write_array1d_real64, which does the real work.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	array	Two-dimensional real array to write.
[in]	length1	Number of rows (= leading dimension).
[in]	length2	Number of columns.

Here is the call graph for this function:

mpi_mod_file_write_array3d_real64()

subroutine mpi_gbl::mpi_mod_file_write_array3d_real64	(	integer(mpiint), intent(in)	fh,
		real(real64), dimension(length1, length2, length3), intent(in)	array,
		integer, intent(in)	length1,
		integer, intent(in)	length2,
		integer, intent(in)	length3
	)

Let master process write 3d array of 8-byte reals to stream file.

Authors

J Benda

Date

2019

Convenience wrapper around mpi_mod_file_write_array1d_real64, which does the real work.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	array	Three-dimensional real array to write.
[in]	length1	Leading dimension.
[in]	length2	Next dimension.
[in]	length3	Next dimension.

Here is the call graph for this function:

mpi_mod_file_write_darray2d_real64()

subroutine mpi_gbl::mpi_mod_file_write_darray2d_real64	(	integer(mpiint), intent(in)	fh,
		integer, intent(in)	m,
		integer, intent(in)	n,
		integer, intent(in)	nprow,
		integer, intent(in)	npcol,
		integer, intent(in)	rb,
		integer, intent(in)	cb,
		real(real64), dimension(myrows, mycols), intent(in)	locA,
		integer, intent(in)	myrows,
		integer, intent(in)	mycols,
		integer(mpiint), intent(in), optional	comm
	)

Write block-cyclically distributed matrix to a stream file.

Authors

J Benda

Date

2019

Assumes ScaLAPACK-compatible row-major (!) block-cyclic distribution. Writes the data at the master process file position, which it first broadcasts to other processes.

Warning

At the moment, there is a limitation on the element count of the local portion of the distributed matrix, which mustn't exceed 2**31 elements (i.e. 16 GiB).

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	m	Number of rows in the distributed matrix.
[in]	n	Number of columns in the distributed matrix.
[in]	nprow	Number of rows in the process grid.
[in]	npcol	Number of columns in the process grid.
[in]	rb	Block row count.
[in]	cb	Block column column.
[in]	locA	Local portion of the distributed matrix.
[in]	myrows	Number of rows in locA.
[in]	mycols	Number of columns in locA.
[in]	comm_opt	Optional communicator on which the file is shared (MPI world used if not provided).

Here is the call graph for this function:

mpi_mod_file_write_int32()

subroutine mpi_gbl::mpi_mod_file_write_int32	(	integer(mpiint), intent(in)	fh,
		integer(int32), intent(in)	n
	)

Let master process write one 4-byte integer to stream file.

Authors

J Benda

Date

2019

Convenience wrapper around mpi_mod_file_write_array1d_int32, which does the real work.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	n	Integer to write.

Here is the call graph for this function:

mpi_mod_file_write_real64()

subroutine mpi_gbl::mpi_mod_file_write_real64	(	integer(mpiint), intent(in)	fh,
		real(real64), intent(in)	x
	)

Let master process write an 8-byte real to stream file.

Authors

J Benda

Date

2019

Convenience wrapper around mpi_mod_file_write_array1d_real64, which does the real work.

Parameters

[in]	fh	File handle as returned by mpi_mod_file_open_*.
[in]	x	Real number to write.

Here is the call graph for this function:

mpi_mod_finalize()

subroutine mpi_gbl::mpi_mod_finalize

Terminates the MPI session and stops the program. It is a blocking routine.

Here is the call graph for this function:

mpi_mod_free_block_array()

subroutine mpi_gbl::mpi_mod_free_block_array

(

type(mpiblockarray), intent(inout)

this

)

Release MPI handle of the block array type.

Authors

J Benda

Date

2019

This is a destructor for MPIBlockArray. It frees the handle for the MPI derived type (if any).

mpi_mod_gatherv_int32()

subroutine mpi_gbl::mpi_mod_gatherv_int32	(	integer(int32), dimension(:), intent(in)	sendbuf,
		integer(int32), dimension(:), intent(out)	recvbuf,
		integer(mpiint), dimension(:), intent(in)	recvcounts,
		integer(mpiint), dimension(:), intent(in)	displs,
		integer(mpiint), intent(in)	to,
		integer(mpiint), intent(in), optional	comm
	)

Gather for 32bit integer array send from every process and received on a given rank.

Here is the call graph for this function:

mpi_mod_gatherv_int64()

subroutine mpi_gbl::mpi_mod_gatherv_int64	(	integer(int64), dimension(:), intent(in)	sendbuf,
		integer(int64), dimension(:), intent(out)	recvbuf,
		integer(mpiint), dimension(:), intent(in)	recvcounts,
		integer(mpiint), dimension(:), intent(in)	displs,
		integer(mpiint), intent(in)	to,
		integer(mpiint), intent(in), optional	comm
	)

Gather for 64bit integer array send from every process and received on a given rank.

Here is the call graph for this function:

mpi_mod_gatherv_real128()

subroutine mpi_gbl::mpi_mod_gatherv_real128	(	real(real128), dimension(:), intent(in)	sendbuf,
		real(real128), dimension(:), intent(out)	recvbuf,
		integer(mpiint), dimension(:), intent(in)	recvcounts,
		integer(mpiint), dimension(:), intent(in)	displs,
		integer(mpiint), intent(in)	to,
		integer(mpiint), intent(in), optional	comm
	)

Gather for 128bit float array send from every process and received on a given rank.

Here is the call graph for this function:

mpi_mod_gatherv_real64()

subroutine mpi_gbl::mpi_mod_gatherv_real64	(	real(real64), dimension(:), intent(in)	sendbuf,
		real(real64), dimension(:), intent(out)	recvbuf,
		integer(mpiint), dimension(:), intent(in)	recvcounts,
		integer(mpiint), dimension(:), intent(in)	displs,
		integer(mpiint), intent(in)	to,
		integer(mpiint), intent(in), optional	comm
	)

Gather for 64bit float array send from every process and received on a given rank.

Here is the call graph for this function:

mpi_mod_isend_int32_array()

subroutine mpi_gbl::mpi_mod_isend_int32_array	(	integer(mpiint), intent(in)	to,
		integer(int32), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_isend_int64_array()

subroutine mpi_gbl::mpi_mod_isend_int64_array	(	integer(mpiint), intent(in)	to,
		integer(int64), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_print_info()

subroutine mpi_gbl::mpi_mod_print_info

(

integer, intent(in)

u

)

Display information about current MPI setup.

Authors

Z Masin, J Benda

Date

2014 - 2019

Originally part of mpi_mod_start, but separated for better application-level control of output from the library.

Parameters

[in]

u

Unit for the text output (mostly stdout).

Here is the call graph for this function:

mpi_mod_rank()

subroutine mpi_gbl::mpi_mod_rank	(	integer(mpiint), intent(out)	rank,
		integer(mpiint), intent(in), optional	comm
	)

Wrapper around MPI_Comm_rank.

Authors

J Benda

Date

2019

Wrapper around a standard MPI rank query which returns zero when compiled without MPI.

Here is the call graph for this function:

mpi_mod_recv_ep_array()

subroutine mpi_gbl::mpi_mod_recv_ep_array	(	integer(mpiint), intent(in)	from,
		integer, intent(in)	tag,
		real(ep), dimension(:), intent(out)	buffer,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_recv_ep_array_dynamic()

subroutine mpi_gbl::mpi_mod_recv_ep_array_dynamic	(	integer(mpiint), intent(in)	from,
		integer, intent(in)	tag,
		real(ep), dimension(:), intent(out), allocatable	buffer,
		integer, intent(out)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_recv_int32_array()

subroutine mpi_gbl::mpi_mod_recv_int32_array	(	integer(mpiint), intent(in)	from,
		integer, intent(in)	tag,
		integer(int32), dimension(:), intent(out)	buffer,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_recv_int64_array()

subroutine mpi_gbl::mpi_mod_recv_int64_array	(	integer(mpiint), intent(in)	from,
		integer, intent(in)	tag,
		integer(int64), dimension(:), intent(out)	buffer,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_recv_wp_array()

subroutine mpi_gbl::mpi_mod_recv_wp_array	(	integer(mpiint), intent(in)	from,
		integer, intent(in)	tag,
		real(wp), dimension(:), intent(out)	buffer,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_recv_wp_array_dynamic()

subroutine mpi_gbl::mpi_mod_recv_wp_array_dynamic	(	integer(mpiint), intent(in)	from,
		integer, intent(in)	tag,
		real(wp), dimension(:), intent(out), allocatable	buffer,
		integer, intent(out)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_rotate_arrays_around_ring_ep()

subroutine mpi_gbl::mpi_mod_rotate_arrays_around_ring_ep	(	integer(longint), intent(inout)	elem_count,
		integer(longint), dimension(:), intent(inout)	int_array,
		real(ep), dimension(:), intent(inout)	ep_array,
		integer(longint), intent(in)	max_num_elements,
		integer(mpiint), intent(in), optional	comm_opt
	)

This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation.

mpi_mod_rotate_arrays_around_ring_wp()

subroutine mpi_gbl::mpi_mod_rotate_arrays_around_ring_wp	(	integer(longint), intent(inout)	elem_count,
		integer(longint), dimension(:), intent(inout)	int_array,
		real(wp), dimension(:), intent(inout)	wp_array,
		integer(longint), intent(in)	max_num_elements,
		integer(mpiint), intent(in), optional	comm_opt
	)

This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation.

mpi_mod_rotate_ep_arrays_around_ring()

subroutine mpi_gbl::mpi_mod_rotate_ep_arrays_around_ring	(	integer(longint), intent(inout)	elem_count,
		real(ep), dimension(:), intent(inout)	ep_array,
		integer(longint), intent(in)	max_num_elements,
		integer(mpiint), intent(in), optional	communicator
	)

This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation.

mpi_mod_rotate_int_arrays_around_ring()

subroutine mpi_gbl::mpi_mod_rotate_int_arrays_around_ring	(	integer(longint), intent(inout)	elem_count,
		integer(longint), dimension(:), intent(inout)	int_array,
		integer(longint), intent(in)	max_num_elements,
		integer(mpiint), intent(in), optional	communicator
	)

This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation.

mpi_mod_rotate_wp_arrays_around_ring()

subroutine mpi_gbl::mpi_mod_rotate_wp_arrays_around_ring	(	integer(longint), intent(inout)	elem_count,
		real(wp), dimension(:), intent(inout)	wp_array,
		integer(longint), intent(in)	max_num_elements,
		integer(mpiint), intent(in), optional	communicator
	)

This will rotate a combination of the number of elements, an integer array and float array once around in a ring formation.

mpi_mod_scan_sum_int32()

subroutine mpi_gbl::mpi_mod_scan_sum_int32	(	integer(int32), intent(in)	src,
		integer(int32), intent(inout)	dst,
		integer(mpiint), intent(in), optional	comm_opt
	)

Wrapper aroung MPI_Scan.

Authors

J Benda

Date

2019

Calls MPI_Scan for 32-bit scalar integer with operation MPI_SUM.

Here is the call graph for this function:

mpi_mod_scan_sum_int64()

subroutine mpi_gbl::mpi_mod_scan_sum_int64	(	integer(int64), intent(in)	src,
		integer(int64), intent(inout)	dst,
		integer(mpiint), intent(in), optional	comm_opt
	)

Wrapper aroung MPI_Scan.

Authors

J Benda

Date

2019

Calls MPI_Scan for 64-bit scalar initeger with operation MPI_SUM.

Here is the call graph for this function:

mpi_mod_send_ep_array()

subroutine mpi_gbl::mpi_mod_send_ep_array	(	integer(mpiint), intent(in)	to,
		real(ep), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_send_ep_array_dynamic()

subroutine mpi_gbl::mpi_mod_send_ep_array_dynamic	(	integer(mpiint), intent(in)	to,
		real(ep), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer(mpiint), intent(in), optional	comm
	)

mpi_mod_send_int32_array()

subroutine mpi_gbl::mpi_mod_send_int32_array	(	integer(mpiint), intent(in)	to,
		integer(int32), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_send_int64_array()

subroutine mpi_gbl::mpi_mod_send_int64_array	(	integer(mpiint), intent(in)	to,
		integer(int64), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_send_wp_array()

subroutine mpi_gbl::mpi_mod_send_wp_array	(	integer(mpiint), intent(in)	to,
		real(wp), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer, intent(in)	n,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_mod_send_wp_array_dynamic()

subroutine mpi_gbl::mpi_mod_send_wp_array_dynamic	(	integer(mpiint), intent(in)	to,
		real(wp), dimension(:), intent(in)	buffer,
		integer, intent(in)	tag,
		integer(mpiint), intent(in), optional	comm
	)

mpi_mod_start()

subroutine mpi_gbl::mpi_mod_start	(	logical, intent(in), optional	do_stdout,
		logical, intent(in), optional	allow_shared_memory
	)

Initializes MPI, assigns the rank for each process and finds out how many processors we're using. It also maps the current floating point precision (kind=cfp) to the corresponding MPI numeric type. This routine also sets the unit for standard output that each rank will use (the value of stdout). In case of serial run the value stdout is not set here and is left to the default value input_unit as specified in the module const.

Warning

This must the first statement in every level3 program.

Here is the call graph for this function:

mpi_mod_wtime()

real(wp) function mpi_gbl::mpi_mod_wtime

Interface to the routine MPI_BARRIER.

Here is the call graph for this function:

mpi_reduce_inplace_sum_ep()

subroutine mpi_gbl::mpi_reduce_inplace_sum_ep	(	real(kind=ep), dimension(:), intent(inout)	buffer,
		integer, intent(in)	nelem
	)

Here is the call graph for this function:

mpi_reduce_inplace_sum_wp()

subroutine mpi_gbl::mpi_reduce_inplace_sum_wp	(	real(kind=wp), dimension(:), intent(inout)	buffer,
		integer, intent(in)	nelem
	)

Here is the call graph for this function:

mpi_reduce_sum_ep()

subroutine mpi_gbl::mpi_reduce_sum_ep	(	real(kind=ep), dimension(:), intent(in)	src,
		real(kind=ep), dimension(:), intent(in)	dest,
		integer, intent(in)	nelem
	)

Here is the call graph for this function:

mpi_reduce_sum_wp()

subroutine mpi_gbl::mpi_reduce_sum_wp	(	real(kind=wp), dimension(:), intent(in)	src,
		real(kind=wp), dimension(:), intent(in)	dest,
		integer, intent(in)	nelem
	)

Here is the call graph for this function:

mpi_reduceall_inplace_sum_ep()

subroutine mpi_gbl::mpi_reduceall_inplace_sum_ep	(	real(kind=ep), dimension(:), intent(inout)	buffer,
		integer, intent(in)	nelem,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_reduceall_inplace_sum_wp()

subroutine mpi_gbl::mpi_reduceall_inplace_sum_wp	(	real(kind=wp), dimension(:), intent(inout)	buffer,
		integer, intent(in)	nelem,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_reduceall_max_int32()

subroutine mpi_gbl::mpi_reduceall_max_int32	(	integer(int32), intent(in)	src,
		integer(int32), intent(inout)	dest,
		integer(mpiint), intent(in), optional	comm_opt
	)

Choose largest elements among processes from 32-bit integer arrays.

Author

Z Masin, J Benda

Date

2017 - 2020

mpi_reduceall_max_int64()

subroutine mpi_gbl::mpi_reduceall_max_int64	(	integer(int64), intent(in)	src,
		integer(int64), intent(inout)	dest,
		integer(mpiint), intent(in), optional	comm_opt
	)

Choose largest elements among processes from 64-bit integer arrays.

Author

Z Masin, J Benda

Date

2017 - 2020

mpi_reduceall_min_int32()

subroutine mpi_gbl::mpi_reduceall_min_int32	(	integer(int32), intent(in)	src,
		integer(int32), intent(inout)	dest,
		integer(mpiint), intent(in), optional	comm_opt
	)

Choose smallest elements among processes from 32-bit integer arrays.

Author

Z Masin, J Benda

Date

2017 - 2020

mpi_reduceall_min_int64()

subroutine mpi_gbl::mpi_reduceall_min_int64	(	integer(int64), intent(in)	src,
		integer(int64), intent(inout)	dest,
		integer(mpiint), intent(in), optional	comm_opt
	)

Choose smallest elements among processes from 64-bit integer arrays.

Author

Z Masin, J Benda

Date

2017 - 2020

mpi_reduceall_sum_ep()

subroutine mpi_gbl::mpi_reduceall_sum_ep	(	real(kind=ep), dimension(:), intent(in)	src,
		real(kind=ep), dimension(:), intent(in)	dest,
		integer, intent(in)	nelem,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_reduceall_sum_wp()

subroutine mpi_gbl::mpi_reduceall_sum_wp	(	real(kind=wp), dimension(:), intent(in)	src,
		real(kind=wp), dimension(:), intent(in)	dest,
		integer, intent(in)	nelem,
		integer(mpiint), intent(in), optional	comm
	)

Here is the call graph for this function:

mpi_xermsg()

subroutine mpi_gbl::mpi_xermsg	(	character(len=*), intent(in)	mod_name,
		character(len=*), intent(in)	routine_name,
		character(len=*), intent(in)	err_msg,
		integer, intent(in)	err,
		integer, intent(in)	level
	)

Analogue of the xermsg routine. This routine uses xermsg to first print out the error message and then either aborts the program or exits the routine depending on the level of the message (warning/error) which is defined in the same way as in xermsg.

naive_mpi_reduce_inplace_sum_ep()

subroutine mpi_gbl::naive_mpi_reduce_inplace_sum_ep	(	real(kind=ep), dimension(:), intent(inout)	buffer,
		integer, intent(in)	nelem,
		logical, intent(in)	split
	)

Here is the call graph for this function:

naive_mpi_reduce_inplace_sum_wp()

subroutine mpi_gbl::naive_mpi_reduce_inplace_sum_wp	(	real(kind=wp), dimension(:), intent(inout)	buffer,
		integer, intent(in)	nelem,
		logical, intent(in)	split
	)

Here is the call graph for this function:

setup_intranode_local_mem_communicator()

subroutine mpi_gbl::setup_intranode_local_mem_communicator

It is not used anywhere in the code yet but it might become useful later on. It creates intra-node communicators without the memory sharing capability.

Here is the call graph for this function:

strings_are_same()

logical function mpi_gbl::strings_are_same	(	character(len=length), intent(in)	str1,
		character(len=length), intent(in)	str2,
		integer, intent(in)	length
	)

Variable Documentation

local_nprocs

integer(kind=mpiint), public, protected mpi_gbl::local_nprocs = 1

Total number of processes on the node this task is bound to. Set by mpi_mod_start.

local_rank

integer(kind=mpiint), public, protected mpi_gbl::local_rank = 0

The local process number (rank) on the node this task is bound to. Set by mpi_mod_start.

master

integer(kind=mpiint), parameter, public mpi_gbl::master = 0

ID of the master process.

Warning

This must be set to 0 so don't change it.

max_data_count

integer, public, protected mpi_gbl::max_data_count = huge(dummy_32bit_integer)

Largest integer (e.g. number of elements) supported by the standard MPI 32-bit integer API.

mpi_running

logical, public, protected mpi_gbl::mpi_running = .false.

This can be set only once at the beginning of the program run by a call to MPI_MOD_START. The value of this variable is accessed by MPI_MOD_CHECK which aborts the program if MPI is not running. This variable stays .false. if the library was compiled without MPI support.

mpi_started

logical, public, protected mpi_gbl::mpi_started = .false.

This is similar to mpi_running, but will be always set to true once MPI_MOD_START is called, whether or not the library was built with MPI support. So, while mpi_running can be used to determine whether the program is running under MPI, the variable mpi_started indicates whether the call to MPI_MOD_START had been already done (even in serial build).

mpiaddr

integer, parameter, public mpi_gbl::mpiaddr = longint

mpicnt

integer, parameter, public mpi_gbl::mpicnt = longint

mpiint

integer, parameter, public mpi_gbl::mpiint = kind(0)

mpiofs

integer, parameter, public mpi_gbl::mpiofs = longint

myrank

integer(kind=mpiint), public, protected mpi_gbl::myrank = -1

The local process number (rank). Set by mpi_mod_start.

nprocs

integer(kind=mpiint), public, protected mpi_gbl::nprocs = -1

Total number of processes. Set by mpi_mod_start.

procname

character(len=*), parameter, public, protected mpi_gbl::procname = "N/A"

Name of the processor on which the current process is running.

shared_communicator

integer(kind=mpiint), public, protected mpi_gbl::shared_communicator = -1

Intra-node communicator created by mpi_mod_start.

shared_enabled

logical, public, protected mpi_gbl::shared_enabled = .false.

The routine mpi_start creates the shared_communicator which groups all tasks sitting on the same node. If we're using MPI 3.0 standard the shared_communicator enables creation of shared memory areas. In this case shared_enabled is set to .true. If we're not using MPI 3.0 then shared_enabled is set to .false.