adam_mpih_nvf_object

ADAM, MPI handler class definition, NVF backend.

Extend common mpih class adding NVF features.

Source: src/lib/nvf/adam_mpih_nvf_object.F90

Dependencies

Contents

• mpih_nvf_object
• initialize
• check_cuda_error
• load_from_file
• print_device_properties
• set_cuda_dimensions
• description

Variables

Name	Type	Attributes	Description
CUDA_SILENT_ERR	integer(kind=I4P)	parameter	CUDA silent error code.
INI_SECTION_NAME	character(len=4)	parameter	INI (config) file section name containing configs.

Derived Types

mpih_nvf_object

MPI handler class, NVF backend.

Inheritance

Extends: mpih_object

Components

Name	Type	Attributes	Description
myrank	integer(kind=I4P)		MPI rank process.
myrankstr	character(len=:)	allocatable	MPI rank process stringified.
procs_number	integer(kind=I4P)		Number of MPI processes.
memory_avail	real(kind=R8P)		CPU memory available (GB) for each process.
error	integer(kind=I4P)		Error traping flag.
timing	real(kind=R8P)		Tic toc timing.
tictoc	integer(kind=I4P)		Next is tic or toc?
req_send_recv	integer(kind=I4P)	allocatable	MPI request receive flags.
mydev	integer(kind=I4P)		My GPU rank.
local_comm	integer(kind=I4P)		Local communicator.
iercuda	integer(kind=I4P)		Error trapping flag for CUDAFortran.
cblk	integer(kind=I4P)		Default CUDA block dimensions.
grid	type(dim3)		CUDA grid and block.
tBlock	type(dim3)		CUDA grid and block.

Type-Bound Procedures

Name	Attributes	Description
abort	pass(self)	Handy MPI abort wrapper.
barrier	pass(self)	Handy MPI barrier wrapper.
error_stop	pass(self)	Stop run with error output.
finalize	pass(self)	Handy MPI finalize wrapper.
print_message	pass(self)	Print a message on stdout with rank prefix.
tictoc_timing	pass(self)	Return the last tic toc timing.
tic	pass(self)	Start a tic toc timing.
toc	pass(self)	Stop a tic toc timing.
initialize	pass(self)	Initialize MPI handler data.
check_cuda_error	pass(self)	Check if CUDA error occurs and abort in case.
description	pass(self)	Return pretty-printed object description.
load_from_file	pass(self)	Load config from file.
print_device_properties	pass(self)	Pretty print device properties.
set_cuda_dimensions	pass(self)	Compute CUDA grid dimensions for GPU parallel computations.

Subroutines

initialize

Initialize MPI handler data.

subroutine initialize(self, do_mpi_init, do_device_init, myrankstr_char_length, verbose)

Arguments

Name	Type	Intent	Attributes	Description
self	class(mpih_nvf_object)	inout		MPI handler.
do_mpi_init	logical	in	optional	Flag to activate MPI init call.
do_device_init	logical	in	optional	Flag to activate device init call.
myrankstr_char_length	integer(kind=I4P)	in	optional	MPI ID string length.
verbose	logical	in	optional	Trigger verbose output.

Call graph

check_cuda_error

Check if CUDA error occurs and abort in case.

subroutine check_cuda_error(self, error_code, msg)

Arguments

Name	Type	Intent	Attributes	Description
self	class(mpih_nvf_object)	inout		MPI handler.
error_code	integer(kind=I4P)	in	optional	Abort error code.
msg	character(len=*)	in	optional	Error message.

Call graph

load_from_file

Load config from file.

subroutine load_from_file(self, file_parameters, go_on_fail)

Arguments

Name	Type	Intent	Attributes	Description
self	class(mpih_nvf_object)	inout		MPI handler.
file_parameters	type(file_ini)	in		Simulation parameters ini file handler.
go_on_fail	logical	in	optional	Go on if load fails.

Call graph

print_device_properties

Pretty print device properties.

subroutine print_device_properties(self, device_properties)

Arguments

Name	Type	Intent	Attributes	Description
self	class(mpih_nvf_object)	in		MPI handler.
device_properties	type(cudadeviceprop)	in		Device properties.

Call graph

set_cuda_dimensions

Compute CUDA grid dimensions for GPU parallel computations.

subroutine set_cuda_dimensions(self, cgrd, cblk)

Arguments

Name	Type	Intent	Attributes	Description
self	class(mpih_nvf_object)	inout		MPI handler.
cgrd	integer(kind=I4P)	in		CUDA grid dimensions.
cblk	integer(kind=I4P)	in	optional	CUDA block dimensions.

Call graph

Functions

description

Return a pretty-formatted object description.

Attributes: pure

Returns: character(len=:)

function description(self) result(desc)

Arguments

Name	Type	Intent	Attributes	Description
self	class(mpih_nvf_object)	in		MPI handler.

Call graph