MPI-AMRVAC  3.1
The MPI - Adaptive Mesh Refinement - Versatile Advection Code (development version)
mod_comm_lib.t
Go to the documentation of this file.
1 module mod_comm_lib
2 
3  implicit none
4  private
5 
6 
7  public :: comm_start
8  public :: comm_finalize
9  public :: init_comm_types
10  public :: mpistop
11 
12 contains
13 
14 
15  !> Initialize the MPI environment
16  subroutine comm_start
17  use mod_global_parameters
18 
19  integer(kind=MPI_ADDRESS_KIND) :: lb
20  integer(kind=MPI_ADDRESS_KIND) :: sizes
21 
22  ! Initialize MPI
23  call mpi_init(ierrmpi)
24 
25  ! Each process stores its rank, which ranges from 0 to N-1, where N is the
26  ! number of processes.
27  call mpi_comm_rank(mpi_comm_world,mype,ierrmpi)
28 
29  ! Store the number of processes
30  call mpi_comm_size(mpi_comm_world,npe,ierrmpi)
31 
32  ! Use the default communicator, which contains all the processes
33  icomm = mpi_comm_world
34 
35  ! Get size of double/integer
36  call mpi_type_get_extent(mpi_real,lb,sizes,ierrmpi)
37  if (sizes /= size_real) call mpistop("Incompatible real size")
38  call mpi_type_get_extent(mpi_double_precision,lb,sizes,ierrmpi)
39  if (sizes /= size_double) call mpistop("Incompatible double size")
40  call mpi_type_get_extent(mpi_integer,lb,sizes,ierrmpi)
41  if (sizes /= size_int) call mpistop("Incompatible integer size")
42  call mpi_type_get_extent(mpi_logical,lb,sizes,ierrmpi)
43  if (sizes /= size_logical) call mpistop("Incompatible logical size")
44 
45  end subroutine comm_start
46 
47  !> Finalize (or shutdown) the MPI environment
48  subroutine comm_finalize
49  use mod_global_parameters
50 
51  call mpi_barrier(mpi_comm_world,ierrmpi)
52  call mpi_finalize(ierrmpi)
53 
54  end subroutine comm_finalize
55 
56  !> Create and store the MPI types that will be used for parallel communication
57  subroutine init_comm_types
58  use mod_global_parameters
59 
60  integer, dimension(ndim+1) :: sizes, subsizes, start
61  integer :: i^d, ic^d, nx^d, nxco^d, nxg^d, idir
62 
63  nx^d=ixmhi^d-ixmlo^d+1;
64  nxg^d=ixghi^d-ixglo^d+1;
65  nxco^d=nx^d/2;
66 
67  ^d&sizes(^d)=ixghi^d;
68  sizes(ndim+1)=nw
69  ^d&subsizes(^d)=nxg^d;
70  subsizes(ndim+1)=nw
71  ^d&start(^d)=ixglo^d-1;
72  start(ndim+1)=0
73  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
74  mpi_order_fortran,mpi_double_precision, &
75  type_block,ierrmpi)
76  call mpi_type_commit(type_block,ierrmpi)
77  size_block={nxg^d*}*nw*size_double
78 
79  ^d&sizes(^d)=ixghi^d/2+nghostcells;
80  sizes(ndim+1)=nw
81  ^d&subsizes(^d)=nxco^d;
82  subsizes(ndim+1)=nw
83  ^d&start(^d)=ixmlo^d-1;
84  start(ndim+1)=0
85  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
86  mpi_order_fortran,mpi_double_precision, &
87  type_coarse_block,ierrmpi)
88  call mpi_type_commit(type_coarse_block,ierrmpi)
89 
90  if(stagger_grid) then
91  ^d&sizes(^d)=ixghi^d+1;
92  sizes(ndim+1)=nws
93  ^d&subsizes(^d)=nx^d+1;
94  subsizes(ndim+1)=nws
95  ^d&start(^d)=ixmlo^d-1;
96  start(ndim+1)=0
97  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
98  mpi_order_fortran,mpi_double_precision, &
99  type_block_io_stg,ierrmpi)
100  call mpi_type_commit(type_block_io_stg,ierrmpi)
101  size_block_io_stg={(nx^d+1)*}*nws*size_double
102 
103  ^d&sizes(^d)=ixghi^d/2+nghostcells+1;
104  sizes(ndim+1)=nws
105  {do ic^db=1,2\}
106  do idir=1,ndim
107  ^d&subsizes(^d)=nxco^d+kr(ic^d,1)*kr(idir,^d);
108  subsizes(ndim+1)=1
109  ^d&start(^d)=ixmlo^d-kr(ic^d,1)*kr(idir,^d);
110  start(ndim+1)=idir-1
111 
112  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
113  mpi_order_fortran,mpi_double_precision, &
114  type_coarse_block_stg(idir,ic^d),ierrmpi)
115  call mpi_type_commit(type_coarse_block_stg(idir,ic^d),ierrmpi)
116  end do
117  {end do\}
118 
119  ^d&sizes(^d)=ixghi^d+1;
120  sizes(ndim+1)=nws
121  {do ic^db=1,2\}
122  do idir=1,^nd
123  ^d&subsizes(^d)=nxco^d+kr(ic^d,1)*kr(idir,^d);
124  subsizes(ndim+1)=1
125  ^d&start(^d)=ixmlo^d-kr(ic^d,1)*kr(idir,^d)+(ic^d-1)*nxco^d;
126  start(ndim+1)=idir-1
127  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
128  mpi_order_fortran,mpi_double_precision, &
129  type_sub_block_stg(idir,ic^d),ierrmpi)
130  call mpi_type_commit(type_sub_block_stg(idir,ic^d),ierrmpi)
131  end do
132  {end do\}
133  end if
134 
135  ^d&sizes(^d)=ixghi^d;
136  sizes(ndim+1)=nw
137  {do ic^db=1,2\}
138  ^d&subsizes(^d)=nxco^d;
139  subsizes(ndim+1)=nw
140  ^d&start(^d)=ixmlo^d-1+(ic^d-1)*nxco^d;
141  start(ndim+1)=0
142  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
143  mpi_order_fortran,mpi_double_precision, &
144  type_sub_block(ic^d),ierrmpi)
145  call mpi_type_commit(type_sub_block(ic^d),ierrmpi)
146  {end do\}
147 
148  ^d&sizes(^d)=ixghi^d;
149  sizes(ndim+1)=nw
150  ^d&subsizes(^d)=nx^d;
151  subsizes(ndim+1)=nw
152  ^d&start(^d)=ixmlo^d-1;
153  start(ndim+1)=0
154  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
155  mpi_order_fortran,mpi_double_precision, &
156  type_block_io,ierrmpi)
157  call mpi_type_commit(type_block_io,ierrmpi)
158  size_block_io={nx^d*}*nw*size_double
159 
160  ^d&sizes(^d)=ixmhi^d-ixmlo^d+1;
161  sizes(ndim+1)=^nd
162  ^d&subsizes(^d)=sizes(^d);
163  subsizes(ndim+1)=^nd
164  ^d&start(^d)=0;
165  start(ndim+1)=0
166  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
167  mpi_order_fortran,mpi_double_precision, &
168  type_block_xcc_io,ierrmpi)
169  call mpi_type_commit(type_block_xcc_io,ierrmpi)
170 
171  ^d&sizes(^d)=ixmhi^d-ixmlo^d+2;
172  sizes(ndim+1)=^nd
173  ^d&subsizes(^d)=sizes(^d);
174  subsizes(ndim+1)=^nd
175  ^d&start(^d)=0;
176  start(ndim+1)=0
177  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
178  mpi_order_fortran,mpi_double_precision, &
179  type_block_xc_io,ierrmpi)
180  call mpi_type_commit(type_block_xc_io,ierrmpi)
181 
182  ^d&sizes(^d)=ixmhi^d-ixmlo^d+1;
183  sizes(ndim+1)=nw+nwauxio
184  ^d&subsizes(^d)=sizes(^d);
185  subsizes(ndim+1)=nw+nwauxio
186  ^d&start(^d)=0;
187  start(ndim+1)=0
188  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
189  mpi_order_fortran,mpi_double_precision, &
190  type_block_wcc_io,ierrmpi)
191  call mpi_type_commit(type_block_wcc_io,ierrmpi)
192 
193  ^d&sizes(^d)=ixmhi^d-ixmlo^d+2;
194  sizes(ndim+1)=nw+nwauxio
195  ^d&subsizes(^d)=sizes(^d);
196  subsizes(ndim+1)=nw+nwauxio
197  ^d&start(^d)=0;
198  start(ndim+1)=0
199  call mpi_type_create_subarray(ndim+1,sizes,subsizes,start, &
200  mpi_order_fortran,mpi_double_precision, &
201  type_block_wc_io,ierrmpi)
202  call mpi_type_commit(type_block_wc_io,ierrmpi)
203 
204  end subroutine init_comm_types
205 
206  !> Exit MPI-AMRVAC with an error message
207  subroutine mpistop(message)
208  use mod_global_parameters
209 
210  character(len=*), intent(in) :: message !< The error message
211  integer :: ierrcode
212 
213  write(*, *) "ERROR for processor", mype, ":"
214  write(*, *) trim(message)
215 
216  call mpi_abort(icomm, ierrcode, ierrmpi)
217 
218  end subroutine mpistop
219 
220 end module mod_comm_lib
mod_comm_lib
Definition: mod_comm_lib.t:1
mod_comm_lib::init_comm_types
subroutine, public init_comm_types
Create and store the MPI types that will be used for parallel communication.
Definition: mod_comm_lib.t:58
mod_comm_lib::comm_start
subroutine, public comm_start
Initialize the MPI environment.
Definition: mod_comm_lib.t:17
mod_comm_lib::comm_finalize
subroutine, public comm_finalize
Finalize (or shutdown) the MPI environment.
Definition: mod_comm_lib.t:49
mod_comm_lib::mpistop
subroutine, public mpistop(message)
Exit MPI-AMRVAC with an error message.
Definition: mod_comm_lib.t:208
mod_global_parameters
This module contains definitions of global parameters and variables and some generic functions/subrou...
Definition: mod_global_parameters.t:4
mod_global_parameters::type_coarse_block
integer type_coarse_block
MPI type for block coarsened by 2, and for its children blocks.
Definition: mod_global_parameters.t:35
mod_global_parameters::ixghi
integer ixghi
Upper index of grid block arrays.
Definition: mod_global_parameters.t:102
mod_global_parameters::kr
integer, dimension(3, 3) kr
Kronecker delta tensor.
Definition: mod_global_parameters.t:449
mod_global_parameters::ndim
integer, parameter ndim
Number of spatial dimensions for grid variables.
Definition: mod_global_parameters.t:62
mod_global_parameters::stagger_grid
logical stagger_grid
True for using stagger grid.
Definition: mod_global_parameters.t:432
mod_global_parameters::type_coarse_block_stg
integer, dimension(^nd, 2^d &) type_coarse_block_stg
MPI type for staggered block coarsened by 2, and for its children blocks.
Definition: mod_global_parameters.t:37
mod_global_parameters::icomm
integer icomm
The MPI communicator.
Definition: mod_global_parameters.t:25
mod_global_parameters::mype
integer mype
The rank of the current MPI task.
Definition: mod_global_parameters.t:22
mod_global_parameters::d
integer, dimension(:), allocatable, parameter d
Definition: mod_global_parameters.t:88
mod_global_parameters::ierrmpi
integer ierrmpi
A global MPI error return code.
Definition: mod_global_parameters.t:28
mod_global_parameters::size_block
integer size_block
Definition: mod_global_parameters.t:33
mod_global_parameters::npe
integer npe
The number of MPI tasks.
Definition: mod_global_parameters.t:19
mod_global_parameters::type_block
integer type_block
MPI type for block including ghost cells and its size.
Definition: mod_global_parameters.t:33
mod_global_parameters::nghostcells
integer nghostcells
Number of ghost cells surrounding a grid.
Definition: mod_global_parameters.t:111
mod_global_parameters::type_block_io_stg
integer type_block_io_stg
MPI type for IO of staggered variables.
Definition: mod_global_parameters.t:41
mod_global_parameters::size_block_io_stg
integer size_block_io_stg
Definition: mod_global_parameters.t:41
mod_global_parameters::ixglo
integer, parameter ixglo
Lower index of grid block arrays (always 1)
Definition: mod_global_parameters.t:99
Generated by doxygen 1.9.1