1 !***********************************************************************
2 !* GNU Lesser General Public License
4 !* This file is part of the GFDL Flexible Modeling System (FMS).
6 !* FMS is free software: you can redistribute it and/or modify it under
7 !* the terms of the GNU Lesser General Public License as published by
8 !* the Free Software Foundation, either version 3 of the License, or (at
9 !* your option) any later version.
11 !* FMS is distributed in the hope that it will be useful, but WITHOUT
12 !* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
13 !* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
16 !* You should have received a copy of the GNU Lesser General Public
17 !* License along with FMS. If not, see <http://www.gnu.org/licenses/>.
18 !***********************************************************************
19 !-----------------------------------------------------------------------
20 ! Communication for message-passing codes
22 ! AUTHOR: V. Balaji (V.Balaji@noaa.gov)
23 ! SGI/GFDL Princeton University
25 !-----------------------------------------------------------------------
27 !> @defgroup mpp_mod mpp_mod
29 !> @brief This module defines interfaces for common operations using message-passing libraries.
30 !! Any type-less arguments in the documentation are MPP_TYPE_ which is defined by the pre-processor
31 !! to create multiple subroutines out of one implementation for use in an interface. See the note
32 !! below for more information
34 !> @author V. Balaji <"V.Balaji@noaa.gov">
36 !! A set of simple calls to provide a uniform interface
37 !! to different message-passing libraries. It currently can be
38 !! implemented either in the SGI/Cray native SHMEM library or in the MPI
39 !! standard. Other libraries (e.g MPI-2, Co-Array Fortran) can be
40 !! incorporated as the need arises.
42 !! The data transfer between a processor and its own memory is based
43 !! on <TT>load</TT> and <TT>store</TT> operations upon
44 !! memory. Shared-memory systems (including distributed shared memory
45 !! systems) have a single address space and any processor can acquire any
46 !! data within the memory by <TT>load</TT> and
47 !! <TT>store</TT>. The situation is different for distributed
48 !! parallel systems. Specialized MPP systems such as the T3E can simulate
49 !! shared-memory by direct data acquisition from remote memory. But if
50 !! the parallel code is distributed across a cluster, or across the Net,
51 !! messages must be sent and received using the protocols for
52 !! long-distance communication, such as TCP/IP. This requires a
53 !! ``handshaking'' between nodes of the distributed system. One can think
54 !! of the two different methods as involving <TT>put</TT>s or
55 !! <TT>get</TT>s (e.g the SHMEM library), or in the case of
56 !! negotiated communication (e.g MPI), <TT>send</TT>s and
59 !! The difference between SHMEM and MPI is that SHMEM uses one-sided
60 !! communication, which can have very low-latency high-bandwidth
61 !! implementations on tightly coupled systems. MPI is a standard
62 !! developed for distributed computing across loosely-coupled systems,
63 !! and therefore incurs a software penalty for negotiating the
64 !! communication. It is however an open industry standard whereas SHMEM
65 !! is a proprietary interface. Besides, the <TT>put</TT>s or
66 !! <TT>get</TT>s on which it is based cannot currently be implemented in
67 !! a cluster environment (there are recent announcements from Compaq that
70 !! The message-passing requirements of climate and weather codes can be
71 !! reduced to a fairly simple minimal set, which is easily implemented in
72 !! any message-passing API. <TT>mpp_mod</TT> provides this API.
74 !! Features of <TT>mpp_mod</TT> include:
76 !! <li> Simple, minimal API, with free access to underlying API for </li>
77 !! more complicated stuff.<BR/>
78 !! <li> Design toward typical use in climate/weather CFD codes. </li>
79 !! <li> Performance to be not significantly lower than any native API. </li>
82 !! This module is used to develop higher-level calls for
83 !! domain decomposition (@ref mpp_domains) and parallel I/O (@ref fms2_io)
85 !! Parallel computing is initially daunting, but it soon becomes
86 !! second nature, much the way many of us can now write vector code
87 !! without much effort. The key insight required while reading and
88 !! writing parallel code is in arriving at a mental grasp of several
89 !! independent parallel execution streams through the same code (the SPMD
90 !! model). Each variable you examine may have different values for each
91 !! stream, the processor ID being an obvious example. Subroutines and
92 !! function calls are particularly subtle, since it is not always obvious
93 !! from looking at a call what synchronization between execution streams
94 !! it implies. An example of erroneous code would be a global barrier
95 !! call (see @ref mpp_sync below) placed
96 !! within a code block that not all PEs will execute, e.g:
99 !! if( pe.EQ.0 )call mpp_sync()
102 !! Here only PE 0 reaches the barrier, where it will wait
103 !! indefinitely. While this is a particularly egregious example to
104 !! illustrate the coding flaw, more subtle versions of the same are
105 !! among the most common errors in parallel code.
107 !! It is therefore important to be conscious of the context of a
108 !! subroutine or function call, and the implied synchronization. There
109 !! are certain calls here (e.g <TT>mpp_declare_pelist, mpp_init,
110 !! mpp_set_stack_size</TT>) which must be called by all
111 !! PEs. There are others which must be called by a subset of PEs (here
112 !! called a <TT>pelist</TT>) which must be called by all the PEs in the
113 !! <TT>pelist</TT> (e.g <TT>mpp_max, mpp_sum, mpp_sync</TT>). Still
114 !! others imply no synchronization at all. I will make every effort to
115 !! highlight the context of each call in the MPP modules, so that the
116 !! implicit synchronization is spelt out.
118 !! For performance it is necessary to keep synchronization as limited
119 !! as the algorithm being implemented will allow. For instance, a single
120 !! message between two PEs should only imply synchronization across the
121 !! PEs in question. A <I>global</I> synchronization (or <I>barrier</I>)
122 !! is likely to be slow, and is best avoided. But codes first
123 !! parallelized on a Cray T3E tend to have many global syncs, as very
124 !! fast barriers were implemented there in hardware.
126 !! Another reason to use pelists is to run a single program in MPMD
127 !! mode, where different PE subsets work on different portions of the
128 !! code. A typical example is to assign an ocean model and atmosphere
129 !! model to different PE subsets, and couple them concurrently instead of
130 !! running them serially. The MPP module provides the notion of a
131 !! <I>current pelist</I>, which is set when a group of PEs branch off
132 !! into a subset. Subsequent calls that omit the <TT>pelist</TT> optional
133 !! argument (seen below in many of the individual calls) assume that the
134 !! implied synchronization is across the current pelist. The calls
135 !! <TT>mpp_root_pe</TT> and <TT>mpp_npes</TT> also return the values
136 !! appropriate to the current pelist. The <TT>mpp_set_current_pelist</TT>
137 !! call is provided to set the current pelist.
141 !! @note F90 is a strictly-typed language, and the syntax pass of the
142 !! compiler requires matching of type, kind and rank (TKR). Most calls
143 !! listed here use a generic type, shown here as <TT>MPP_TYPE_</TT>. This
144 !! is resolved in the pre-processor stage to any of a variety of
145 !! types. In general the MPP operations work on 4-byte and 8-byte
146 !! variants of <TT>integer, real, complex, logical</TT> variables, of
147 !! rank 0 to 5, leading to 48 specific module procedures under the same
148 !! generic interface. Any of the variables below shown as
149 !! <TT>MPP_TYPE_</TT> is treated in this way.
153 ! Define rank(X) for PGI compiler
154 #if defined( __PGI) || defined (__FLANG)
155 #define rank(X) size(shape(X))
159 #if defined(use_libMPI)
163 use iso_fortran_env, only : INPUT_UNIT, OUTPUT_UNIT, ERROR_UNIT
164 use mpp_parameter_mod, only : MPP_VERBOSE, MPP_DEBUG, ALL_PES, ANY_PE, NULL_PE
165 use mpp_parameter_mod, only : NOTE, WARNING, FATAL, MPP_CLOCK_DETAILED,MPP_CLOCK_SYNC
166 use mpp_parameter_mod, only : CLOCK_COMPONENT, CLOCK_SUBCOMPONENT, CLOCK_MODULE_DRIVER
167 use mpp_parameter_mod, only : CLOCK_MODULE, CLOCK_ROUTINE, CLOCK_LOOP, CLOCK_INFRA
168 use mpp_parameter_mod, only : MAX_EVENTS, MAX_BINS, MAX_EVENT_TYPES, MAX_CLOCKS
169 use mpp_parameter_mod, only : MAXPES, EVENT_WAIT, EVENT_ALLREDUCE, EVENT_BROADCAST
170 use mpp_parameter_mod, only : EVENT_ALLTOALL
171 use mpp_parameter_mod, only : EVENT_TYPE_CREATE, EVENT_TYPE_FREE
172 use mpp_parameter_mod, only : EVENT_RECV, EVENT_SEND, MPP_READY, MPP_WAIT
173 use mpp_parameter_mod, only : mpp_parameter_version=>version
174 use mpp_parameter_mod, only : DEFAULT_TAG
175 use mpp_parameter_mod, only : COMM_TAG_1, COMM_TAG_2, COMM_TAG_3, COMM_TAG_4
176 use mpp_parameter_mod, only : COMM_TAG_5, COMM_TAG_6, COMM_TAG_7, COMM_TAG_8
177 use mpp_parameter_mod, only : COMM_TAG_9, COMM_TAG_10, COMM_TAG_11, COMM_TAG_12
178 use mpp_parameter_mod, only : COMM_TAG_13, COMM_TAG_14, COMM_TAG_15, COMM_TAG_16
179 use mpp_parameter_mod, only : COMM_TAG_17, COMM_TAG_18, COMM_TAG_19, COMM_TAG_20
180 use mpp_parameter_mod, only : MPP_FILL_INT,MPP_FILL_DOUBLE
181 use mpp_data_mod, only : stat, mpp_stack, ptr_stack, status, ptr_status, sync, ptr_sync
182 use mpp_data_mod, only : mpp_from_pe, ptr_from, remote_data_loc, ptr_remote
183 use mpp_data_mod, only : mpp_data_version=>version
189 !--- public parameters -----------------------------------------------
190 public :: MPP_VERBOSE, MPP_DEBUG, ALL_PES, ANY_PE, NULL_PE, NOTE, WARNING, FATAL
191 public :: MPP_CLOCK_SYNC, MPP_CLOCK_DETAILED, CLOCK_COMPONENT, CLOCK_SUBCOMPONENT
192 public :: CLOCK_MODULE_DRIVER, CLOCK_MODULE, CLOCK_ROUTINE, CLOCK_LOOP, CLOCK_INFRA
193 public :: MAXPES, EVENT_RECV, EVENT_SEND
194 public :: COMM_TAG_1, COMM_TAG_2, COMM_TAG_3, COMM_TAG_4
195 public :: COMM_TAG_5, COMM_TAG_6, COMM_TAG_7, COMM_TAG_8
196 public :: COMM_TAG_9, COMM_TAG_10, COMM_TAG_11, COMM_TAG_12
197 public :: COMM_TAG_13, COMM_TAG_14, COMM_TAG_15, COMM_TAG_16
198 public :: COMM_TAG_17, COMM_TAG_18, COMM_TAG_19, COMM_TAG_20
199 public :: MPP_FILL_INT,MPP_FILL_DOUBLE,MPP_INFO_NULL,MPP_COMM_NULL
200 public :: mpp_init_test_full_init, mpp_init_test_init_true_only, mpp_init_test_peset_allocated
201 public :: mpp_init_test_clocks_init, mpp_init_test_datatype_list_init, mpp_init_test_logfile_init
202 public :: mpp_init_test_read_namelist, mpp_init_test_etc_unit, mpp_init_test_requests_allocated
204 !--- public interface from mpp_util.h ------------------------------
205 public :: stdin, stdout, stderr, stdlog, warnlog, lowercase, uppercase, mpp_error, mpp_error_state
206 public :: mpp_set_warn_level, mpp_sync, mpp_sync_self, mpp_pe
207 public :: mpp_npes, mpp_root_pe, mpp_set_root_pe, mpp_declare_pelist
208 public :: mpp_get_current_pelist, mpp_set_current_pelist, mpp_get_current_pelist_name
209 public :: mpp_clock_id, mpp_clock_set_grain, mpp_record_timing_data, get_unit
210 public :: read_ascii_file, read_input_nml, mpp_clock_begin, mpp_clock_end
211 public :: get_ascii_file_num_lines, get_ascii_file_num_lines_and_length
212 public :: mpp_record_time_start, mpp_record_time_end
214 !--- public interface from mpp_comm.h ------------------------------
215 public :: mpp_chksum, mpp_max, mpp_min, mpp_sum, mpp_transmit, mpp_send, mpp_recv
217 public :: mpp_broadcast, mpp_init, mpp_exit
218 public :: mpp_gather, mpp_scatter, mpp_alltoall
219 public :: mpp_type, mpp_byte, mpp_type_create, mpp_type_free
221 !*********************************************************************
225 !*********************************************************************
226 !> Communication information for message passing libraries
228 !> peset hold communicators as SHMEM-compatible triads (start, log2(stride), num)
232 character(len=32) :: name
233 integer, pointer :: list(:) =>NULL()
235 integer :: start, log2stride !< dummy variables when libMPI is defined.
236 integer :: id, group !< MPI communicator and group id for this PE set.
237 end type communicator
239 !> Communication event profile
243 character(len=16) :: name
244 integer(i8_kind), dimension(MAX_EVENTS) :: ticks, bytes
248 !> a clock contains an array of event profiles for a region
252 character(len=32) :: name
253 integer(i8_kind) :: hits
254 integer(i8_kind) :: tick
255 integer(i8_kind) :: total_ticks
257 logical :: sync_on_begin, detailed
259 type(event), pointer :: events(:) =>NULL() !> if needed, allocate to MAX_EVENT_TYPES
260 logical :: is_on !> initialize to false. set true when calling mpp_clock_begin
261 !! set false when calling mpp_clock_end
264 !> Summary of information from a clock run
266 type :: Clock_Data_Summary
268 character(len=16) :: name
269 real(r8_kind) :: msg_size_sums(MAX_BINS)
270 real(r8_kind) :: msg_time_sums(MAX_BINS)
271 real(r8_kind) :: total_data
272 real(r8_kind) :: total_time
273 integer(i8_kind) :: msg_size_cnts(MAX_BINS)
274 integer(i8_kind) :: total_cnts
275 end type Clock_Data_Summary
277 !> holds name and clock data for use in @ref mpp_util.h
279 type :: Summary_Struct
281 character(len=16) :: name
282 type (Clock_Data_Summary) :: event(MAX_EVENT_TYPES)
283 end type Summary_Struct
285 !> Data types for generalized data transfer (e.g. MPI_Type)
289 integer :: counter !> Number of instances of this type
291 integer, allocatable :: sizes(:)
292 integer, allocatable :: subsizes(:)
293 integer, allocatable :: starts(:)
294 integer :: etype !> Elementary data type (e.g. MPI_BYTE)
295 integer :: id !> Identifier within message passing library (e.g. MPI)
297 type(mpp_type), pointer :: prev => null()
298 type(mpp_type), pointer :: next => null()
301 !> Persisent elements for linked list interaction
303 type :: mpp_type_list
305 type(mpp_type), pointer :: head => null()
306 type(mpp_type), pointer :: tail => null()
308 end type mpp_type_list
310 !***********************************************************************
312 ! public interface from mpp_util.h
314 !***********************************************************************
315 !> @brief Error handler.
317 !> It is strongly recommended that all error exits pass through
318 !! <TT>mpp_error</TT> to assure the program fails cleanly. An individual
319 !! PE encountering a <TT>STOP</TT> statement, for instance, can cause the
320 !! program to hang. The use of the <TT>STOP</TT> statement is strongly
323 !! Calling mpp_error with no arguments produces an immediate error
331 !! The argument order
333 !! call mpp_error( routine, errormsg, errortype )
335 !! is also provided to support legacy code. In this version of the
336 !! call, none of the arguments may be omitted.
338 !! The behaviour of <TT>mpp_error</TT> for a <TT>WARNING</TT> can be
339 !! controlled with an additional call <TT>mpp_set_warn_level</TT>.
341 !! call mpp_set_warn_level(ERROR)
343 !! causes <TT>mpp_error</TT> to treat <TT>WARNING</TT>
344 !! exactly like <TT>FATAL</TT>.
346 !! call mpp_set_warn_level(WARNING)
348 !! resets to the default behaviour described above.
350 !! <TT>mpp_error</TT> also has an internal error state which
351 !! maintains knowledge of whether a warning has been issued. This can be
352 !! used at startup in a subroutine that checks if the model has been
353 !! properly configured. You can generate a series of warnings using
354 !! <TT>mpp_error</TT>, and then check at the end if any warnings has been
355 !! issued using the function <TT>mpp_error_state()</TT>. If the value of
356 !! this is <TT>WARNING</TT>, at least one warning has been issued, and
357 !! the user can take appropriate action:
360 !! if( ... )call mpp_error( WARNING, '...' )
361 !! if( ... )call mpp_error( WARNING, '...' )
362 !! if( ... )call mpp_error( WARNING, '...' )
364 !! if( mpp_error_state().EQ.WARNING )call mpp_error( FATAL, '...' )
367 !! <br> Example usage:
369 !! call mpp_error( errortype, routine, errormsg )
372 !! One of <TT>NOTE</TT>, <TT>WARNING</TT> or <TT>FATAL</TT>
373 !! (these definitions are acquired by use association).
374 !! <TT>NOTE</TT> writes <TT>errormsg</TT> to <TT>STDOUT</TT>.
375 !! <TT>WARNING</TT> writes <TT>errormsg</TT> to <TT>STDERR</TT>.
376 !! <TT>FATAL</TT> writes <TT>errormsg</TT> to <TT>STDERR</TT>,
377 !! and induces a clean error exit with a call stack traceback.
378 !! @param routine Calling routine name
379 !! @param errmsg Message to output
383 module procedure mpp_error_basic
384 module procedure mpp_error_mesg
385 module procedure mpp_error_noargs
386 module procedure mpp_error_is
387 module procedure mpp_error_rs
388 module procedure mpp_error_ia
389 module procedure mpp_error_ra
390 module procedure mpp_error_ia_ia
391 module procedure mpp_error_ia_ra
392 module procedure mpp_error_ra_ia
393 module procedure mpp_error_ra_ra
394 module procedure mpp_error_ia_is
395 module procedure mpp_error_ia_rs
396 module procedure mpp_error_ra_is
397 module procedure mpp_error_ra_rs
398 module procedure mpp_error_is_ia
399 module procedure mpp_error_is_ra
400 module procedure mpp_error_rs_ia
401 module procedure mpp_error_rs_ra
402 module procedure mpp_error_is_is
403 module procedure mpp_error_is_rs
404 module procedure mpp_error_rs_is
405 module procedure mpp_error_rs_rs
407 !> Takes a given integer or real array and returns it as a string
408 !> @param[in] array An array of integers or reals
409 !> @returns string equivalent of given array
411 interface array_to_char
412 module procedure iarray_to_char
413 module procedure rarray_to_char
416 !***********************************************************************
418 ! public interface from mpp_comm.h
420 !***********************************************************************
422 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
424 ! ROUTINES TO INITIALIZE/FINALIZE MPP MODULE: mpp_init, mpp_exit !
426 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
428 !> @fn mpp_mod::mpp_init::mpp_init( flags, localcomm, test_level)
430 !> @brief Initialize @ref mpp_mod
432 !> Called to initialize the <TT>mpp_mod</TT> package. It is recommended
433 !! that this call be the first executed line in your program. It sets the
434 !! number of PEs assigned to this run (acquired from the command line, or
435 !! through the environment variable <TT>NPES</TT>), and associates an ID
436 !! number to each PE. These can be accessed by calling @ref mpp_npes and
438 !! <br> Example usage:
440 !! call mpp_init( flags )
443 !! <TT>flags</TT> can be set to <TT>MPP_VERBOSE</TT> to
444 !! have <TT>mpp_mod</TT> keep you informed of what it's up to.
446 !! Debugging flag to set amount of initialization tasks performed
448 !> @fn mpp_mod::mpp_exit()
449 !> @brief Exit <TT>@ref mpp_mod</TT>.
451 !> Called at the end of the run, or to re-initialize <TT>mpp_mod</TT>,
452 !! should you require that for some odd reason.
454 !! This call implies synchronization across all PEs.
456 !! <br>Example usage:
461 !#####################################################################
463 !> @fn subroutine mpp_set_stack_size(n)
464 !> @brief Allocate module internal workspace.
465 !> @param Integer to set stack size to(in words)
466 !> <TT>mpp_mod</TT> maintains a private internal array called
467 !! <TT>mpp_stack</TT> for private workspace. This call sets the length,
468 !! in words, of this array.
470 !! The <TT>mpp_init</TT> call sets this
471 !! workspace length to a default of 32768, and this call may be used if a
472 !! longer workspace is needed.
474 !! This call implies synchronization across all PEs.
476 !! This workspace is symmetrically allocated, as required for
477 !! efficient communication on SGI and Cray MPP systems. Since symmetric
478 !! allocation must be performed by <I>all</I> PEs in a job, this call
479 !! must also be called by all PEs, using the same value of
480 !! <TT>n</TT>. Calling <TT>mpp_set_stack_size</TT> from a subset of PEs,
481 !! or with unequal argument <TT>n</TT>, may cause the program to hang.
483 !! If any MPP call using <TT>mpp_stack</TT> overflows the declared
484 !! stack array, the program will abort with a message specifying the
485 !! stack length that is required. Many users wonder why, if the required
486 !! stack length can be computed, it cannot also be specified at that
487 !! point. This cannot be automated because there is no way for the
488 !! program to know if all PEs are present at that call, and with equal
489 !! values of <TT>n</TT>. The program must be rerun by the user with the
490 !! correct argument to <TT>mpp_set_stack_size</TT>, called at an
491 !! appropriate point in the code where all PEs are known to be present.
492 !! @verbose call mpp_set_stack_size(n)
495 public :: mpp_set_stack_size
498 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
500 ! DATA TRANSFER TYPES: mpp_type_create !
502 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
504 !> @brief Create a mpp_type variable
505 !> @param[in] field A field of any numerical or logical type
506 !> @param[in] array_of_subsizes Integer array of subsizes
507 !> @param[in] array_of_starts Integer array of starts
508 !> @param[out] dtype_out Output variable for created @ref mpp_type
510 interface mpp_type_create
511 module procedure mpp_type_create_int4
512 module procedure mpp_type_create_int8
513 module procedure mpp_type_create_real4
514 module procedure mpp_type_create_real8
515 module procedure mpp_type_create_cmplx4
516 module procedure mpp_type_create_cmplx8
517 module procedure mpp_type_create_logical4
518 module procedure mpp_type_create_logical8
519 end interface mpp_type_create
521 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
523 ! GLOBAL REDUCTION ROUTINES: mpp_max, mpp_sum, mpp_min !
525 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
527 !> @brief Reduction operations.
528 !> Find the max of scalar a from the PEs in pelist
529 !! result is also automatically broadcast to all PEs
531 !! call mpp_max( a, pelist )
533 !> @param a <TT>real</TT> or <TT>integer</TT>, of 4-byte of 8-byte kind.
534 !> @param pelist If <TT>pelist</TT> is omitted, the context is assumed to be the
535 !! current pelist. This call implies synchronization across the PEs in
536 !! <TT>pelist</TT>, or the current pelist if <TT>pelist</TT> is absent.
539 module procedure mpp_max_real8_0d
540 module procedure mpp_max_real8_1d
541 module procedure mpp_max_int8_0d
542 module procedure mpp_max_int8_1d
543 module procedure mpp_max_real4_0d
544 module procedure mpp_max_real4_1d
545 module procedure mpp_max_int4_0d
546 module procedure mpp_max_int4_1d
549 !> @brief Reduction operations.
550 !> Find the min of scalar a from the PEs in pelist
551 !! result is also automatically broadcast to all PEs
553 !! call mpp_min( a, pelist )
555 !> @param a <TT>real</TT> or <TT>integer</TT>, of 4-byte of 8-byte kind.
556 !> @param pelist If <TT>pelist</TT> is omitted, the context is assumed to be the
557 !! current pelist. This call implies synchronization across the PEs in
558 !! <TT>pelist</TT>, or the current pelist if <TT>pelist</TT> is absent.
561 module procedure mpp_min_real8_0d
562 module procedure mpp_min_real8_1d
563 module procedure mpp_min_int8_0d
564 module procedure mpp_min_int8_1d
565 module procedure mpp_min_real4_0d
566 module procedure mpp_min_real4_1d
567 module procedure mpp_min_int4_0d
568 module procedure mpp_min_int4_1d
572 !> @brief Reduction operation.
574 !> <TT>MPP_TYPE_</TT> corresponds to any 4-byte and 8-byte variant of
575 !! <TT>integer, real, complex</TT> variables, of rank 0 or 1. A
576 !! contiguous block from a multi-dimensional array may be passed by its
577 !! starting address and its length, as in <TT>f77</TT>.
579 !! Library reduction operators are not required or guaranteed to be
580 !! bit-reproducible. In any case, changing the processor count changes
581 !! the data layout, and thus very likely the order of operations. For
582 !! bit-reproducible sums of distributed arrays, consider using the
583 !! <TT>mpp_global_sum</TT> routine provided by the
584 !! @ref mpp_domains module.
586 !! The <TT>bit_reproducible</TT> flag provided in earlier versions of
587 !! this routine has been removed.
590 !! If <TT>pelist</TT> is omitted, the context is assumed to be the
591 !! current pelist. This call implies synchronization across the PEs in
592 !! <TT>pelist</TT>, or the current pelist if <TT>pelist</TT> is absent.
594 !! call mpp_sum( a, length, pelist )
598 module procedure mpp_sum_int8
599 module procedure mpp_sum_int8_scalar
600 module procedure mpp_sum_int8_2d
601 module procedure mpp_sum_int8_3d
602 module procedure mpp_sum_int8_4d
603 module procedure mpp_sum_int8_5d
604 module procedure mpp_sum_real8
605 module procedure mpp_sum_real8_scalar
606 module procedure mpp_sum_real8_2d
607 module procedure mpp_sum_real8_3d
608 module procedure mpp_sum_real8_4d
609 module procedure mpp_sum_real8_5d
611 module procedure mpp_sum_cmplx8
612 module procedure mpp_sum_cmplx8_scalar
613 module procedure mpp_sum_cmplx8_2d
614 module procedure mpp_sum_cmplx8_3d
615 module procedure mpp_sum_cmplx8_4d
616 module procedure mpp_sum_cmplx8_5d
618 module procedure mpp_sum_int4
619 module procedure mpp_sum_int4_scalar
620 module procedure mpp_sum_int4_2d
621 module procedure mpp_sum_int4_3d
622 module procedure mpp_sum_int4_4d
623 module procedure mpp_sum_int4_5d
624 module procedure mpp_sum_real4
625 module procedure mpp_sum_real4_scalar
626 module procedure mpp_sum_real4_2d
627 module procedure mpp_sum_real4_3d
628 module procedure mpp_sum_real4_4d
629 module procedure mpp_sum_real4_5d
631 module procedure mpp_sum_cmplx4
632 module procedure mpp_sum_cmplx4_scalar
633 module procedure mpp_sum_cmplx4_2d
634 module procedure mpp_sum_cmplx4_3d
635 module procedure mpp_sum_cmplx4_4d
636 module procedure mpp_sum_cmplx4_5d
640 !> Calculates sum of a given numerical array across pe's for adjoint domains
643 module procedure mpp_sum_int8_ad
644 module procedure mpp_sum_int8_scalar_ad
645 module procedure mpp_sum_int8_2d_ad
646 module procedure mpp_sum_int8_3d_ad
647 module procedure mpp_sum_int8_4d_ad
648 module procedure mpp_sum_int8_5d_ad
649 module procedure mpp_sum_real8_ad
650 module procedure mpp_sum_real8_scalar_ad
651 module procedure mpp_sum_real8_2d_ad
652 module procedure mpp_sum_real8_3d_ad
653 module procedure mpp_sum_real8_4d_ad
654 module procedure mpp_sum_real8_5d_ad
656 module procedure mpp_sum_cmplx8_ad
657 module procedure mpp_sum_cmplx8_scalar_ad
658 module procedure mpp_sum_cmplx8_2d_ad
659 module procedure mpp_sum_cmplx8_3d_ad
660 module procedure mpp_sum_cmplx8_4d_ad
661 module procedure mpp_sum_cmplx8_5d_ad
663 module procedure mpp_sum_int4_ad
664 module procedure mpp_sum_int4_scalar_ad
665 module procedure mpp_sum_int4_2d_ad
666 module procedure mpp_sum_int4_3d_ad
667 module procedure mpp_sum_int4_4d_ad
668 module procedure mpp_sum_int4_5d_ad
669 module procedure mpp_sum_real4_ad
670 module procedure mpp_sum_real4_scalar_ad
671 module procedure mpp_sum_real4_2d_ad
672 module procedure mpp_sum_real4_3d_ad
673 module procedure mpp_sum_real4_4d_ad
674 module procedure mpp_sum_real4_5d_ad
676 module procedure mpp_sum_cmplx4_ad
677 module procedure mpp_sum_cmplx4_scalar_ad
678 module procedure mpp_sum_cmplx4_2d_ad
679 module procedure mpp_sum_cmplx4_3d_ad
680 module procedure mpp_sum_cmplx4_4d_ad
681 module procedure mpp_sum_cmplx4_5d_ad
685 !> @brief Gather data sent from pelist onto the root pe
686 !! Wrapper for MPI_gather, can be used with and without indices
689 !> @param sbuf MPP_TYPE_ data buffer to send
690 !> @param rbuf MPP_TYPE_ data buffer to receive
691 !> @param pelist integer(:) optional pelist to gather from, defaults to current
693 !> <BR> Example usage:
695 !! call mpp_gather(send_buffer,recv_buffer, pelist)
696 !! call mpp_gather(is, ie, js, je, pelist, array_seg, data, is_root_pe)
699 module procedure mpp_gather_logical_1d
700 module procedure mpp_gather_int4_1d
701 module procedure mpp_gather_int8_1d
702 module procedure mpp_gather_real4_1d
703 module procedure mpp_gather_real8_1d
704 module procedure mpp_gather_logical_1dv
705 module procedure mpp_gather_int4_1dv
706 module procedure mpp_gather_int8_1dv
707 module procedure mpp_gather_real4_1dv
708 module procedure mpp_gather_real8_1dv
709 module procedure mpp_gather_pelist_logical_2d
710 module procedure mpp_gather_pelist_logical_3d
711 module procedure mpp_gather_pelist_int4_2d
712 module procedure mpp_gather_pelist_int4_3d
713 module procedure mpp_gather_pelist_int8_2d
714 module procedure mpp_gather_pelist_int8_3d
715 module procedure mpp_gather_pelist_real4_2d
716 module procedure mpp_gather_pelist_real4_3d
717 module procedure mpp_gather_pelist_real8_2d
718 module procedure mpp_gather_pelist_real8_3d
721 !> @brief Scatter (ie - is) * (je - js) contiguous elements of array data from the designated root pe
722 !! into contigous members of array segment in each pe that is included in the pelist argument.
725 !> @param is, ie integer start and end index of the first dimension of the segment array
726 !> @param je, js integer start and end index of the second dimension of the segment array
727 !> @param pelist integer(:) the PE list of target pes, needs to be monotonically increasing
728 !> @param array_seg MPP_TYPE_ 2D array that the data is to be copied into
729 !> @param data MPP_TYPE_ the source array
730 !> @param is_root_pe logical true if calling from root pe
731 !> @param ishift integer offsets specifying the first elelement in the data array
732 !> @param nk integer size of third dimension for 3D calls
734 !> <BR> Example usage:
736 !! call mpp_scatter(is, ie, js, je, pelist, segment, data, .true.)
738 interface mpp_scatter
739 module procedure mpp_scatter_pelist_int4_2d
740 module procedure mpp_scatter_pelist_int4_3d
741 module procedure mpp_scatter_pelist_int8_2d
742 module procedure mpp_scatter_pelist_int8_3d
743 module procedure mpp_scatter_pelist_real4_2d
744 module procedure mpp_scatter_pelist_real4_3d
745 module procedure mpp_scatter_pelist_real8_2d
746 module procedure mpp_scatter_pelist_real8_3d
749 !#####################################################################
750 !> @brief Scatter a vector across all PEs
752 !> Transpose the vector and PE index
753 !! Wrapper for the MPI_alltoall function, includes more generic _V and _W
754 !! versions if given displacements/data types
756 !! Generic MPP_TYPE_ implentations:
757 !! <li> @ref mpp_alltoall_ </li>
758 !! <li> @ref mpp_alltoallv_ </li>
759 !! <li> @ref mpp_alltoallw_ </li>
762 interface mpp_alltoall
763 module procedure mpp_alltoall_int4
764 module procedure mpp_alltoall_int8
765 module procedure mpp_alltoall_real4
766 module procedure mpp_alltoall_real8
768 module procedure mpp_alltoall_cmplx4
771 module procedure mpp_alltoall_cmplx8
773 module procedure mpp_alltoall_logical4
774 module procedure mpp_alltoall_logical8
775 module procedure mpp_alltoall_int4_v
776 module procedure mpp_alltoall_int8_v
777 module procedure mpp_alltoall_real4_v
778 module procedure mpp_alltoall_real8_v
780 module procedure mpp_alltoall_cmplx4_v
783 module procedure mpp_alltoall_cmplx8_v
785 module procedure mpp_alltoall_logical4_v
786 module procedure mpp_alltoall_logical8_v
787 module procedure mpp_alltoall_int4_w
788 module procedure mpp_alltoall_int8_w
789 module procedure mpp_alltoall_real4_w
790 module procedure mpp_alltoall_real8_w
792 module procedure mpp_alltoall_cmplx4_w
795 module procedure mpp_alltoall_cmplx8_w
797 module procedure mpp_alltoall_logical4_w
798 module procedure mpp_alltoall_logical8_w
802 !#####################################################################
803 !> @brief Basic message-passing call.
805 !> <TT>MPP_TYPE_</TT> corresponds to any 4-byte and 8-byte variant of
806 !! <TT>integer, real, complex, logical</TT> variables, of rank 0 or 1. A
807 !! contiguous block from a multi-dimensional array may be passed by its
808 !! starting address and its length, as in <TT>f77</TT>.
810 !! <TT>mpp_transmit</TT> is currently implemented as asynchronous
811 !! outward transmission and synchronous inward transmission. This follows
812 !! the behaviour of <TT>shmem_put</TT> and <TT>shmem_get</TT>. In MPI, it
813 !! is implemented as <TT>mpi_isend</TT> and <TT>mpi_recv</TT>. For most
814 !! applications, transmissions occur in pairs, and are here accomplished
817 !! The special PE designations <TT>NULL_PE</TT>,
818 !! <TT>ANY_PE</TT> and <TT>ALL_PES</TT> are provided by use
821 !! <TT>NULL_PE</TT>: is used to disable one of the pair of
822 !! transmissions.<BR/>
823 !! <TT>ANY_PE</TT>: is used for unspecific remote
824 !! destination. (Please note that <TT>put_pe=ANY_PE</TT> has no meaning
825 !! in the MPI context, though it is available in the SHMEM invocation. If
826 !! portability is a concern, it is best avoided).<BR/>
827 !! <TT>ALL_PES</TT>: is used for broadcast operations.
829 !! It is recommended that
830 !! @ref mpp_broadcast be used for
833 !! The following example illustrates the use of
834 !! <TT>NULL_PE</TT> and <TT>ALL_PES</TT>:
837 !! real, dimension(n) :: a
840 !! call mpp_transmit( a, n, p, a, n, NULL_PE )
843 !! call mpp_transmit( a, n, NULL_PE, a, n, 0 )
846 !! call mpp_transmit( a, n, ALL_PES, a, n, 0 )
849 !! The do loop and the broadcast operation above are equivalent.
851 !! Two overloaded calls <TT>mpp_send</TT> and
852 !! <TT>mpp_recv</TT> have also been
853 !! provided. <TT>mpp_send</TT> calls <TT>mpp_transmit</TT>
854 !! with <TT>get_pe=NULL_PE</TT>. <TT>mpp_recv</TT> calls
855 !! <TT>mpp_transmit</TT> with <TT>put_pe=NULL_PE</TT>. Thus
856 !! the do loop above could be written more succinctly:
861 !! call mpp_send( a, n, p )
864 !! call mpp_recv( a, n, 0 )
869 !! call mpp_transmit( put_data, put_len, put_pe, get_data, get_len, get_pe )
872 interface mpp_transmit
873 module procedure mpp_transmit_real8
874 module procedure mpp_transmit_real8_scalar
875 module procedure mpp_transmit_real8_2d
876 module procedure mpp_transmit_real8_3d
877 module procedure mpp_transmit_real8_4d
878 module procedure mpp_transmit_real8_5d
880 module procedure mpp_transmit_cmplx8
881 module procedure mpp_transmit_cmplx8_scalar
882 module procedure mpp_transmit_cmplx8_2d
883 module procedure mpp_transmit_cmplx8_3d
884 module procedure mpp_transmit_cmplx8_4d
885 module procedure mpp_transmit_cmplx8_5d
887 module procedure mpp_transmit_int8
888 module procedure mpp_transmit_int8_scalar
889 module procedure mpp_transmit_int8_2d
890 module procedure mpp_transmit_int8_3d
891 module procedure mpp_transmit_int8_4d
892 module procedure mpp_transmit_int8_5d
893 module procedure mpp_transmit_logical8
894 module procedure mpp_transmit_logical8_scalar
895 module procedure mpp_transmit_logical8_2d
896 module procedure mpp_transmit_logical8_3d
897 module procedure mpp_transmit_logical8_4d
898 module procedure mpp_transmit_logical8_5d
900 module procedure mpp_transmit_real4
901 module procedure mpp_transmit_real4_scalar
902 module procedure mpp_transmit_real4_2d
903 module procedure mpp_transmit_real4_3d
904 module procedure mpp_transmit_real4_4d
905 module procedure mpp_transmit_real4_5d
908 module procedure mpp_transmit_cmplx4
909 module procedure mpp_transmit_cmplx4_scalar
910 module procedure mpp_transmit_cmplx4_2d
911 module procedure mpp_transmit_cmplx4_3d
912 module procedure mpp_transmit_cmplx4_4d
913 module procedure mpp_transmit_cmplx4_5d
915 module procedure mpp_transmit_int4
916 module procedure mpp_transmit_int4_scalar
917 module procedure mpp_transmit_int4_2d
918 module procedure mpp_transmit_int4_3d
919 module procedure mpp_transmit_int4_4d
920 module procedure mpp_transmit_int4_5d
921 module procedure mpp_transmit_logical4
922 module procedure mpp_transmit_logical4_scalar
923 module procedure mpp_transmit_logical4_2d
924 module procedure mpp_transmit_logical4_3d
925 module procedure mpp_transmit_logical4_4d
926 module procedure mpp_transmit_logical4_5d
928 !> @brief Recieve data from another PE
930 !> @param[out] get_data scalar or array to get written with received data
931 !> @param get_len size of array to recv from get_data
932 !> @param from_pe PE number to receive from
933 !> @param block true for blocking, false for non-blocking. Defaults to true
934 !> @param tag communication tag
935 !> @param[out] request MPI request handle
938 module procedure mpp_recv_real8
939 module procedure mpp_recv_real8_scalar
940 module procedure mpp_recv_real8_2d
941 module procedure mpp_recv_real8_3d
942 module procedure mpp_recv_real8_4d
943 module procedure mpp_recv_real8_5d
945 module procedure mpp_recv_cmplx8
946 module procedure mpp_recv_cmplx8_scalar
947 module procedure mpp_recv_cmplx8_2d
948 module procedure mpp_recv_cmplx8_3d
949 module procedure mpp_recv_cmplx8_4d
950 module procedure mpp_recv_cmplx8_5d
952 module procedure mpp_recv_int8
953 module procedure mpp_recv_int8_scalar
954 module procedure mpp_recv_int8_2d
955 module procedure mpp_recv_int8_3d
956 module procedure mpp_recv_int8_4d
957 module procedure mpp_recv_int8_5d
958 module procedure mpp_recv_logical8
959 module procedure mpp_recv_logical8_scalar
960 module procedure mpp_recv_logical8_2d
961 module procedure mpp_recv_logical8_3d
962 module procedure mpp_recv_logical8_4d
963 module procedure mpp_recv_logical8_5d
965 module procedure mpp_recv_real4
966 module procedure mpp_recv_real4_scalar
967 module procedure mpp_recv_real4_2d
968 module procedure mpp_recv_real4_3d
969 module procedure mpp_recv_real4_4d
970 module procedure mpp_recv_real4_5d
973 module procedure mpp_recv_cmplx4
974 module procedure mpp_recv_cmplx4_scalar
975 module procedure mpp_recv_cmplx4_2d
976 module procedure mpp_recv_cmplx4_3d
977 module procedure mpp_recv_cmplx4_4d
978 module procedure mpp_recv_cmplx4_5d
980 module procedure mpp_recv_int4
981 module procedure mpp_recv_int4_scalar
982 module procedure mpp_recv_int4_2d
983 module procedure mpp_recv_int4_3d
984 module procedure mpp_recv_int4_4d
985 module procedure mpp_recv_int4_5d
986 module procedure mpp_recv_logical4
987 module procedure mpp_recv_logical4_scalar
988 module procedure mpp_recv_logical4_2d
989 module procedure mpp_recv_logical4_3d
990 module procedure mpp_recv_logical4_4d
991 module procedure mpp_recv_logical4_5d
993 !> Send data to a receiving PE.
995 !> @param put_data scalar or array to get sent to a receiving PE
996 !> @param put_len size of data to send from put_data
997 !> @param to_pe PE number to send to
998 !> @param block true for blocking, false for non-blocking. Defaults to true
999 !> @param tag communication tag
1000 !> @param[out] request MPI request handle
1001 !! <br> Example usage:
1002 !! @code{.F90} call mpp_send(data, ie, pe) @endcode
1005 module procedure mpp_send_real8
1006 module procedure mpp_send_real8_scalar
1007 module procedure mpp_send_real8_2d
1008 module procedure mpp_send_real8_3d
1009 module procedure mpp_send_real8_4d
1010 module procedure mpp_send_real8_5d
1012 module procedure mpp_send_cmplx8
1013 module procedure mpp_send_cmplx8_scalar
1014 module procedure mpp_send_cmplx8_2d
1015 module procedure mpp_send_cmplx8_3d
1016 module procedure mpp_send_cmplx8_4d
1017 module procedure mpp_send_cmplx8_5d
1019 module procedure mpp_send_int8
1020 module procedure mpp_send_int8_scalar
1021 module procedure mpp_send_int8_2d
1022 module procedure mpp_send_int8_3d
1023 module procedure mpp_send_int8_4d
1024 module procedure mpp_send_int8_5d
1025 module procedure mpp_send_logical8
1026 module procedure mpp_send_logical8_scalar
1027 module procedure mpp_send_logical8_2d
1028 module procedure mpp_send_logical8_3d
1029 module procedure mpp_send_logical8_4d
1030 module procedure mpp_send_logical8_5d
1032 module procedure mpp_send_real4
1033 module procedure mpp_send_real4_scalar
1034 module procedure mpp_send_real4_2d
1035 module procedure mpp_send_real4_3d
1036 module procedure mpp_send_real4_4d
1037 module procedure mpp_send_real4_5d
1040 module procedure mpp_send_cmplx4
1041 module procedure mpp_send_cmplx4_scalar
1042 module procedure mpp_send_cmplx4_2d
1043 module procedure mpp_send_cmplx4_3d
1044 module procedure mpp_send_cmplx4_4d
1045 module procedure mpp_send_cmplx4_5d
1047 module procedure mpp_send_int4
1048 module procedure mpp_send_int4_scalar
1049 module procedure mpp_send_int4_2d
1050 module procedure mpp_send_int4_3d
1051 module procedure mpp_send_int4_4d
1052 module procedure mpp_send_int4_5d
1053 module procedure mpp_send_logical4
1054 module procedure mpp_send_logical4_scalar
1055 module procedure mpp_send_logical4_2d
1056 module procedure mpp_send_logical4_3d
1057 module procedure mpp_send_logical4_4d
1058 module procedure mpp_send_logical4_5d
1062 !> @brief Perform parallel broadcasts
1064 !> The <TT>mpp_broadcast</TT> call has been added because the original
1065 !! syntax (using <TT>ALL_PES</TT> in <TT>mpp_transmit</TT>) did not
1066 !! support a broadcast across a pelist.
1068 !! <TT>MPP_TYPE_</TT> corresponds to any 4-byte and 8-byte variant of
1069 !! <TT>integer, real, complex, logical</TT> variables, of rank 0 or 1. A
1070 !! contiguous block from a multi-dimensional array may be passed by its
1071 !! starting address and its length, as in <TT>f77</TT>.
1073 !! Global broadcasts through the <TT>ALL_PES</TT> argument to
1074 !! @ref mpp_transmit are still provided for
1075 !! backward-compatibility.
1077 !! If <TT>pelist</TT> is omitted, the context is assumed to be the
1078 !! current pelist. <TT>from_pe</TT> must belong to the current
1079 !! pelist. This call implies synchronization across the PEs in
1080 !! <TT>pelist</TT>, or the current pelist if <TT>pelist</TT> is absent.
1082 !! <br>Example usage:
1084 !! call mpp_broadcast( data, length, from_pe, pelist )
1086 !> @param[inout] data Data to broadcast
1087 !> @param length Length of data to broadcast
1088 !> @param from_pe PE to send the data from
1089 !> @param pelist List of PE's to broadcast across, if not provided uses current list
1091 interface mpp_broadcast
1092 module procedure mpp_broadcast_char
1093 module procedure mpp_broadcast_real8
1094 module procedure mpp_broadcast_real8_scalar
1095 module procedure mpp_broadcast_real8_2d
1096 module procedure mpp_broadcast_real8_3d
1097 module procedure mpp_broadcast_real8_4d
1098 module procedure mpp_broadcast_real8_5d
1100 module procedure mpp_broadcast_cmplx8
1101 module procedure mpp_broadcast_cmplx8_scalar
1102 module procedure mpp_broadcast_cmplx8_2d
1103 module procedure mpp_broadcast_cmplx8_3d
1104 module procedure mpp_broadcast_cmplx8_4d
1105 module procedure mpp_broadcast_cmplx8_5d
1107 module procedure mpp_broadcast_int8
1108 module procedure mpp_broadcast_int8_scalar
1109 module procedure mpp_broadcast_int8_2d
1110 module procedure mpp_broadcast_int8_3d
1111 module procedure mpp_broadcast_int8_4d
1112 module procedure mpp_broadcast_int8_5d
1113 module procedure mpp_broadcast_logical8
1114 module procedure mpp_broadcast_logical8_scalar
1115 module procedure mpp_broadcast_logical8_2d
1116 module procedure mpp_broadcast_logical8_3d
1117 module procedure mpp_broadcast_logical8_4d
1118 module procedure mpp_broadcast_logical8_5d
1120 module procedure mpp_broadcast_real4
1121 module procedure mpp_broadcast_real4_scalar
1122 module procedure mpp_broadcast_real4_2d
1123 module procedure mpp_broadcast_real4_3d
1124 module procedure mpp_broadcast_real4_4d
1125 module procedure mpp_broadcast_real4_5d
1128 module procedure mpp_broadcast_cmplx4
1129 module procedure mpp_broadcast_cmplx4_scalar
1130 module procedure mpp_broadcast_cmplx4_2d
1131 module procedure mpp_broadcast_cmplx4_3d
1132 module procedure mpp_broadcast_cmplx4_4d
1133 module procedure mpp_broadcast_cmplx4_5d
1135 module procedure mpp_broadcast_int4
1136 module procedure mpp_broadcast_int4_scalar
1137 module procedure mpp_broadcast_int4_2d
1138 module procedure mpp_broadcast_int4_3d
1139 module procedure mpp_broadcast_int4_4d
1140 module procedure mpp_broadcast_int4_5d
1141 module procedure mpp_broadcast_logical4
1142 module procedure mpp_broadcast_logical4_scalar
1143 module procedure mpp_broadcast_logical4_2d
1144 module procedure mpp_broadcast_logical4_3d
1145 module procedure mpp_broadcast_logical4_4d
1146 module procedure mpp_broadcast_logical4_5d
1149 !#####################################################################
1151 !> @brief Calculate parallel checksums
1153 !> \e mpp_chksum is a parallel checksum routine that returns an
1154 !! identical answer for the same array irrespective of how it has been
1155 !! partitioned across processors. \e int_kind is the KIND
1156 !! parameter corresponding to long integers (see discussion on
1157 !! OS-dependent preprocessor directives) defined in
1158 !! the file platform.F90. \e MPP_TYPE_ corresponds to any
1159 !! 4-byte and 8-byte variant of \e integer, \e real, \e complex, \e logical
1160 !! variables, of rank 0 to 5.
1162 !! Integer checksums on FP data use the F90 <TT>TRANSFER()</TT>
1165 !! This provides identical results on a single-processor job, and to perform
1166 !! serial checksums on a single processor of a parallel job, you only
1167 !! need to use the optional <TT>pelist</TT> argument.
1170 !! integer :: pe, chksum
1173 !! chksum = mpp_chksum( a, (/pe/) )
1176 !! The additional functionality of <TT>mpp_chksum</TT> over
1177 !! serial checksums is to compute the checksum across the PEs in
1178 !! <TT>pelist</TT>. The answer is guaranteed to be the same for
1179 !! the same distributed array irrespective of how it has been
1182 !! If <TT>pelist</TT> is omitted, the context is assumed to be the
1183 !! current pelist. This call implies synchronization across the PEs in
1184 !! <TT>pelist</TT>, or the current pelist if <TT>pelist</TT> is absent.
1185 !! <br> Example usage:
1187 !! mpp_chksum( var, pelist )
1189 !! @param var Data to calculate checksum of
1190 !! @param pelist Optional list of PE's to include in checksum calculation if not using
1192 !! @return Parallel checksum of var across given or implicit pelist
1194 !! Generic MPP_TYPE_ implentations:
1195 !! <li> @ref mpp_chksum_</li>
1196 !! <li> @ref mpp_chksum_int_</li>
1197 !! <li> @ref mpp_chksum_int_rmask_</li>
1200 interface mpp_chksum
1201 module procedure mpp_chksum_i8_1d
1202 module procedure mpp_chksum_i8_2d
1203 module procedure mpp_chksum_i8_3d
1204 module procedure mpp_chksum_i8_4d
1205 module procedure mpp_chksum_i8_5d
1206 module procedure mpp_chksum_i8_1d_rmask
1207 module procedure mpp_chksum_i8_2d_rmask
1208 module procedure mpp_chksum_i8_3d_rmask
1209 module procedure mpp_chksum_i8_4d_rmask
1210 module procedure mpp_chksum_i8_5d_rmask
1212 module procedure mpp_chksum_i4_1d
1213 module procedure mpp_chksum_i4_2d
1214 module procedure mpp_chksum_i4_3d
1215 module procedure mpp_chksum_i4_4d
1216 module procedure mpp_chksum_i4_5d
1217 module procedure mpp_chksum_i4_1d_rmask
1218 module procedure mpp_chksum_i4_2d_rmask
1219 module procedure mpp_chksum_i4_3d_rmask
1220 module procedure mpp_chksum_i4_4d_rmask
1221 module procedure mpp_chksum_i4_5d_rmask
1223 module procedure mpp_chksum_r8_0d
1224 module procedure mpp_chksum_r8_1d
1225 module procedure mpp_chksum_r8_2d
1226 module procedure mpp_chksum_r8_3d
1227 module procedure mpp_chksum_r8_4d
1228 module procedure mpp_chksum_r8_5d
1230 module procedure mpp_chksum_r4_0d
1231 module procedure mpp_chksum_r4_1d
1232 module procedure mpp_chksum_r4_2d
1233 module procedure mpp_chksum_r4_3d
1234 module procedure mpp_chksum_r4_4d
1235 module procedure mpp_chksum_r4_5d
1237 module procedure mpp_chksum_c8_0d
1238 module procedure mpp_chksum_c8_1d
1239 module procedure mpp_chksum_c8_2d
1240 module procedure mpp_chksum_c8_3d
1241 module procedure mpp_chksum_c8_4d
1242 module procedure mpp_chksum_c8_5d
1245 module procedure mpp_chksum_c4_0d
1246 module procedure mpp_chksum_c4_1d
1247 module procedure mpp_chksum_c4_2d
1248 module procedure mpp_chksum_c4_3d
1249 module procedure mpp_chksum_c4_4d
1250 module procedure mpp_chksum_c4_5d
1254 !> @addtogroup mpp_mod
1256 !***********************************************************************
1260 !***********************************************************************
1261 integer, parameter :: PESET_MAX = 10000
1262 integer :: current_peset_max = 32
1263 type(communicator), allocatable :: peset(:) !< Will be allocated starting from 0, 0 is a dummy used
1264 !! to hold single-PE "self" communicator
1265 logical :: module_is_initialized = .false.
1266 logical :: debug = .false.
1267 integer :: npes=1, root_pe=0, pe=0
1268 integer(i8_kind) :: tick, ticks_per_sec, max_ticks, start_tick, end_tick, tick0=0
1269 integer :: mpp_comm_private
1270 logical :: first_call_system_clock_mpi=.TRUE.
1271 real(r8_kind) :: mpi_count0=0 !< use to prevent integer overflow
1272 real(r8_kind) :: mpi_tick_rate=0.d0 !< clock rate for mpi_wtick()
1273 logical :: mpp_record_timing_data=.TRUE.
1274 type(clock),save :: clocks(MAX_CLOCKS)
1275 integer :: log_unit, etc_unit
1276 integer :: warn_unit !< unit number of the warning log
1277 character(len=32), parameter :: configfile='logfile'
1278 character(len=32), parameter :: warnfile='warnfile' !< base name for warninglog (appends ".<PE>.out")
1279 integer :: peset_num=0, current_peset_num=0
1280 integer :: world_peset_num !<the world communicator
1282 integer :: clock_num=0, num_clock_ids=0,current_clock=0, previous_clock(MAX_CLOCKS)=0
1285 type(mpp_type_list) :: datatypes
1286 type(mpp_type), target :: mpp_byte
1288 integer :: cur_send_request = 0
1289 integer :: cur_recv_request = 0
1290 integer, allocatable :: request_send(:)
1291 integer, allocatable :: request_recv(:)
1292 integer, allocatable :: size_recv(:)
1293 integer, allocatable :: type_recv(:)
1294 ! if you want to save the non-root PE information uncomment out the following line
1295 ! and comment out the assigment of etcfile to '/dev/null'
1297 character(len=32) :: etcfile='._mpp.nonrootpe.msgs'
1299 character(len=32) :: etcfile='/dev/null'
1302 !> Use the intrinsics in iso_fortran_env
1303 integer :: in_unit=INPUT_UNIT, out_unit=OUTPUT_UNIT, err_unit=ERROR_UNIT
1304 integer :: stdout_unit
1306 !--- variables used in mpp_util.h
1307 type(Summary_Struct) :: clock_summary(MAX_CLOCKS)
1308 logical :: warnings_are_fatal = .FALSE.
1309 integer :: error_state=0
1310 integer :: clock_grain=CLOCK_LOOP-1
1312 !--- variables used in mpp_comm.h
1313 integer :: clock0 !<measures total runtime from mpp_init to mpp_exit
1314 integer :: mpp_stack_size=0, mpp_stack_hwm=0
1315 logical :: verbose=.FALSE.
1317 integer :: get_len_nocomm = 0 !< needed for mpp_transmit_nocomm.h
1319 !--- variables used in mpp_comm_mpi.inc
1320 integer, parameter :: mpp_init_test_full_init = -1
1321 integer, parameter :: mpp_init_test_init_true_only = 0
1322 integer, parameter :: mpp_init_test_peset_allocated = 1
1323 integer, parameter :: mpp_init_test_clocks_init = 2
1324 integer, parameter :: mpp_init_test_datatype_list_init = 3
1325 integer, parameter :: mpp_init_test_logfile_init = 4
1326 integer, parameter :: mpp_init_test_read_namelist = 5
1327 integer, parameter :: mpp_init_test_etc_unit = 6
1328 integer, parameter :: mpp_init_test_requests_allocated = 7
1330 !> MPP_INFO_NULL acts as an analagous mpp-macro for MPI_INFO_NULL to share with fms2_io NetCDF4
1331 !! mpi-io. The default value for the no-mpi case comes from Intel MPI and MPICH. OpenMPI sets
1332 !! a default value of '0'
1333 #if defined(use_libMPI)
1334 integer, parameter :: MPP_INFO_NULL = MPI_INFO_NULL
1336 integer, parameter :: MPP_INFO_NULL = 469762048
1339 !> MPP_COMM_NULL acts as an analagous mpp-macro for MPI_COMM_NULL to share with fms2_io NetCDF4
1340 !! mpi-io. The default value for the no-mpi case comes from Intel MPI and MPICH. OpenMPI sets
1341 !! a default value of '2'
1342 #if defined(use_libMPI)
1343 integer, parameter :: MPP_COMM_NULL = MPI_COMM_NULL
1345 integer, parameter :: MPP_COMM_NULL = 67108864
1348 !***********************************************************************
1349 ! variables needed for subroutine read_input_nml (include/mpp_util.inc)
1351 ! public variable needed for reading input nml file from an internal file
1352 character(len=:), dimension(:), allocatable, target, public :: input_nml_file
1353 logical :: read_ascii_file_on = .FALSE.
1354 !***********************************************************************
1356 ! Include variable "version" to be written to log file.
1357 #include<file_version.h>
1360 integer, parameter :: MAX_REQUEST_MIN = 10000
1361 integer :: request_multiply = 20
1363 logical :: etc_unit_is_stderr = .false.
1364 integer :: max_request = 0
1365 logical :: sync_all_clocks = .false.
1366 namelist /mpp_nml/ etc_unit_is_stderr, request_multiply, mpp_record_timing_data, sync_all_clocks
1369 #include <system_clock.fh>
1370 #include <mpp_util.inc>
1371 #include <mpp_comm.inc>
1375 ! close documentation grouping