knowing Message Passing Interface (MPI)
Table of Contents
- Abstract
- What is MPI?
- Getting Started
- Environment Management Routines
- Point to Point Communication Routines
- Collective Communication Routines
- Derived Data Types
- Group and Communicator Management Routines
- Virtual Topologies
- A Brief Word on MPI-2
- LLNL Specific Information and Recommendations
- References and More Information
- Appendix A: MPI-1 Routine Index
- Exercise
| Abstract |
The Message Passing Interface Standard (MPI) is a message passing library
standard based on the consensus of the MPI Forum, which has over 40
participating organizations, including vendors, researchers, software library
developers, and users. The goal of the Message Passing Interface is to
establish a portable, efficient, and flexible standard for message passing
that will be widely used for writing message passing programs. As such, MPI
is the first standardized, vendor independent, message passing library. The
advantages of developing message passing software using MPI closely match the
design goals of portability, efficiency, and flexibility. MPI is not an
IEEE or ISO standard, but has in fact, become the "industry standard" for
writing message passing programs on HPC platforms.
The goal of this tutorial is to teach those unfamiliar with MPI how to develop and run parallel programs according to the MPI standard. The primary topics that are presented focus on those which are the most useful for new MPI programmers. The tutorial begins with an introduction, background, and basic information for getting started with MPI. This is followed by a detailed look at the MPI routines that are most useful for new MPI programmers, including MPI Environment Management, Point-to-Point Communications, and Collective Communications routines. Numerous examples in both C and Fortran are provided, as well as a lab exercise.
The tutorial materials also include more advanced topics such as Derived Data Types, Group and Communicator Management Routines, and Virtual Topologies. However, these are not actually presented during the lecture, but are meant to serve as "further reading" for those who are interested.
Level/Prerequisites: Ideal for those who are new to parallel
programming with MPI. A basic understanding of parallel programming
in C or Fortran is assumed. For those who are unfamiliar with Parallel
Programming in general, the material covered in
EC3500: Introduction To Parallel
Computing would be helpful.
| What is MPI? |
An Interface Specification:
- M P I =
Message
Passing
Interface
- MPI is a specification
for the developers and users of message passing
libraries. By itself, it is NOT a library - but rather the specification
of what such a library should be.
- Simply stated, the goal of the Message Passing Interface is to provide a
widely used standard for writing message passing programs. The interface
attempts to be
- practical
- portable
- efficient
- flexible
- Interface specifications have been defined for C/C++ and Fortran programs.
History and Evolution:
- MPI resulted from the efforts of numerous individuals and groups
over the course of a 2 year period between 1992 and 1994. Some history:
- 1980s - early 1990s: Distributed memory, parallel computing
develops, as do a number of incompatible software tools for
writing such programs - usually with tradeoffs between portability,
performance, functionality and price. Recognition of the need
for a standard arose.
- April, 1992: Workshop on Standards for Message Passing in a
Distributed Memory Environment, sponsored by the Center for
Research on Parallel Computing, Williamsburg, Virginia.
The basic features essential to a standard message passing
interface were discussed, and a working group established to
continue the standardization process. Preliminary draft proposal
developed subsequently.
- November 1992: - Working group meets in Minneapolis. MPI draft
proposal (MPI1) from ORNL presented. Group adopts procedures
and organization to form the
MPI Forum.
MPIF eventually comprised of about 175 individuals from
40 organizations including parallel computer vendors, software
writers, academia and application scientists.
- November 1993: Supercomputing 93 conference - draft MPI standard
presented.
- Final version of draft released in May, 1994 - available on the
at: www.mcs.anl.gov/Projects/mpi/standard.html
- MPI-2 picked up where the first MPI specification left off, and addressed
topics which go beyond the first MPI specification. The original MPI
then became known as MPI-1.
MPI-2 is briefly covered later. Was finalized in 1996.
- Today, MPI implementations are a combination of MPI-1 and MPI-2. A few implementations include the full functionality of both.
Reasons for Using MPI:
- Standardization
- MPI is the only message passing library
which can be considered a standard. It is supported on
virtually all HPC platforms. Practically, it has replaced
all previous message passing libraries.
- Portability
- There is no need to modify your source code
when you port your application to a different platform that supports
(and is compliant with) the MPI standard.
- Performance Opportunities
- Vendor implementations should be able to
exploit native hardware features to optimize performance.
For more information about MPI performance see the
MPI Performance Topics
(http://www.llnl.gov/computing/tutorials/mpi_performance) tutorial.
- Functionality
- Over 115 routines are defined.
- Availability - A variety of implementations are available, both vendor and public domain.
Programming Model:
- MPI lends itself to most (if not all) distributed memory parallel
programming models. In addition, MPI is commonly used to implement
(behind the scenes)
some shared memory models, such as Data Parallel, on distributed
memory architectures.
- Hardware platforms:
- Distributed Memory: Originally, MPI was targeted for distributed memory systems.
- Shared Memory: As shared memory systems became more popular, particularly SMP / NUMA architectures, MPI implementations for these platforms appeared.
- Hybrid: MPI is now used on just about any common parallel architecture including massively parallel machines, SMP clusters, workstation clusters and heterogeneous networks.
- All parallelism is explicit: the programmer is responsible for
correctly identifying parallelism and implementing parallel
algorithms using MPI constructs.
- The number of tasks dedicated to run a parallel program is static. New tasks can not be dynamically spawned during run time. (MPI-2 addresses this issue).
| Getting Started |
Header File:
- Required for all programs/routines which make MPI library calls.
C include file Fortran include file #include "mpi.h" include 'mpif.h'
Format of MPI Calls:
- C names are case sensitive; Fortran names are not.
| C Binding | |||||
|---|---|---|---|---|---|
| Format: | | Example:
| Error code:
| Returned as "rc". MPI_SUCCESS if successful
| | |
| Fortran Binding | |||||
|---|---|---|---|---|---|
| Format: | call mpi_xxxxx(parameter,..., ierr) Example:
| Error code:
| Returned as "ierr" parameter. MPI_SUCCESS if successful
| | |
General MPI Program Structure:
 |
Communicators and Groups:
- MPI uses objects called communicators and groups to define which
collection of processes may communicate with each other. Most MPI
routines require you to specify a communicator as an argument.
- Communicators and groups will be covered in more detail later. For now,
simply use MPI_COMM_WORLD whenever a communicator is required -
it is the predefined communicator that includes all of your MPI processes.
Rank:
- Within a communicator, every process has its own unique, integer
identifier assigned by the system when the process initializes. A rank
is sometimes also called a "process ID". Ranks are contiguous and begin
at zero.
- Used by the programmer to specify the source and destination of messages. Often used conditionally by the application to control program execution (if rank=0 do this / if rank=1 do that).
| Environment Management Routines |
MPI environment management routines are used for an assortment of purposes, such as initializing and terminating the MPI environment, querying the environment and identity, etc. Most of the commonly used ones are described below.
- MPI_Init
- Initializes the MPI execution environment. This function must be called
in every MPI program, must be called before any other MPI functions
and must be called only once in an MPI program. For C programs, MPI_Init
may be used to pass the command line arguments to all processes,
although this is not required by the standard and is implementation
dependent.
MPI_Init (&argc,&argv)
MPI_INIT (ierr) - MPI_Comm_size
- Determines the number of processes in the group associated with a
communicator. Generally used within the communicator MPI_COMM_WORLD
to determine the number of processes being used by your application.
MPI_Comm_size (comm,&size)
MPI_COMM_SIZE (comm,size,ierr) - MPI_Comm_rank
- Determines the rank of the calling process within the communicator.
Initially, each process will be assigned a unique integer rank
between 0 and number of processors - 1 within the communicator
MPI_COMM_WORLD. This rank is often referred to as a task ID.
If a process becomes associated with other communicators, it will have
a unique rank within each of these as well.
MPI_Comm_rank (comm,&rank)
MPI_COMM_RANK (comm,rank,ierr) - MPI_Abort
- Terminates all MPI processes associated with the communicator. In
most MPI implementations it terminates ALL processes regardless of
the communicator specified.
MPI_Abort (comm,errorcode)
MPI_ABORT (comm,errorcode,ierr) -
MPI_Get_processor_name
- Returns the processor name. Also
returns the length of the name. The buffer for "name" must be at
least MPI_MAX_PROCESSOR_NAME characters in size. What is returned into
"name" is implementation dependent - may not be the same as the output
of the "hostname" or "host" shell commands.
MPI_Get_processor_name (&name,&resultlength)
MPI_GET_PROCESSOR_NAME (name,resultlength,ierr) - MPI_Initialized
- Indicates whether MPI_Init has been called - returns flag as either
logical true (1) or false(0). MPI requires that MPI_Init
be called once and only once by each process. This may pose a problem
for modules that want to use MPI and are prepared to call MPI_Init
if necessary. MPI_Initialized solves this problem.
MPI_Initialized (&flag)
MPI_INITIALIZED (flag,ierr) - MPI_Wtime
- Returns an elapsed wall clock time in seconds (double precision) on the
calling processor.
MPI_Wtime ()
MPI_WTIME () - MPI_Wtick
- Returns the resolution in seconds (double precision) of MPI_Wtime.
MPI_Wtick ()
MPI_WTICK () - MPI_Finalize
- Terminates the MPI execution environment. This function should be
the last MPI routine called in every MPI program - no other MPI
routines may be called after it.
MPI_Finalize ()
MPI_FINALIZE (ierr)
Examples: Environment Management Routines
C Language - Environment Management Routines Example
#include "mpi.h"
#include <stdio.h>
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, rc;
rc = MPI_Init(&argc,&argv);
if (rc != MPI_SUCCESS) {
printf ("Error starting MPI program. Terminating.\n");
MPI_Abort(MPI_COMM_WORLD, rc);
}
MPI_Comm_size(MPI_COMM_WORLD,&numtasks);
MPI_Comm_rank(MPI_COMM_WORLD,&rank);
printf ("Number of tasks= %d My rank= %d\n", numtasks,rank);
/******* do some work *******/
MPI_Finalize();
}
|
Fortran - Environment Management Routines Example
program simple
include 'mpif.h'
integer numtasks, rank, ierr, rc
call MPI_INIT(ierr)
if (ierr .ne. MPI_SUCCESS) then
print *,'Error starting MPI program. Terminating.'
call MPI_ABORT(MPI_COMM_WORLD, rc, ierr)
end if
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
print *, 'Number of tasks=',numtasks,' My rank=',rank
C ****** do some work ******
call MPI_FINALIZE(ierr)
end
|
| Point to Point Communication Routines |
General Concepts
Types of Point-to-Point Operations:
- MPI point-to-point operations typically involve message passing between
two, and only two, different MPI tasks. One task is performing a send
operation and the other task is performing a matching receive operation.
- There are different types of send and receive routines used for
different purposes. For example:
- Synchronous send
- Blocking send / blocking receive
- Non-blocking send / non-blocking receive
- Buffered send
- Combined send/receive
- "Ready" send
- Any type of send routine can be paired with any type of receive routine.
- MPI also provides several routines associated with send - receive operations, such as those used to wait for a message's arrival or probe to find out if a message has arrived.
Buffering:
- In a perfect world, every send operation would be perfectly
synchronized with its matching receive. This is rarely the case.
Somehow or other, the MPI implementation must be able to deal with
storing data when the two tasks are out of sync.
- Consider the following two cases:
- A send operation occurs 5 seconds before the receive is ready - where is the message while the receive is pending?
- Multiple sends arrive at the same receiving task which can only accept one send at a time - what happens to the messages that are "backing up"?
- The MPI implementation (not the MPI standard) decides what happens to
data in these types of cases. Typically, a system buffer area
is reserved to hold data in transit. For example:
- System buffer space is:
- Opaque to the programmer and managed entirely by the MPI library
- A finite resource that can be easy to exhaust
- Often mysterious and not well documented
- Able to exist on the sending side, the receiving side, or both
- Something that may improve program performance because it allows send - receive operations to be asynchronous.
- User managed address space (i.e. your program variables) is called the application buffer. MPI also provides for a user managed send buffer.
Blocking vs. Non-blocking:
- Most of the MPI point-to-point routines can be used in either blocking
or non-blocking mode.
- Blocking:
- A blocking send routine will only "return" after it is safe to modify the application buffer (your send data) for reuse. Safe means that modifications will not affect the data intended for the receive task. Safe does not imply that the data was actually received - it may very well be sitting in a system buffer.
- A blocking send can be synchronous which means there is handshaking occurring with the receive task to confirm a safe send.
- A blocking send can be asynchronous if a system buffer is used to hold the data for eventual delivery to the receive.
- A blocking receive only "returns" after the data has arrived and is ready for use by the program.
- Non-blocking:
- Non-blocking send and receive routines behave similarly - they will return almost immediately. They do not wait for any communication events to complete, such as message copying from user memory to system buffer space or the actual arrival of message.
- Non-blocking operations simply "request" the MPI library to perform the operation when it is able. The user can not predict when that will happen.
- It is unsafe to modify the application buffer (your variable space) until you know for a fact the requested non-blocking operation was actually performed by the library. There are "wait" routines used to do this.
- Non-blocking communications are primarily used to overlap computation with communication and exploit possible performance gains.
Order and Fairness:
- Order:
- MPI guarantees that messages will not overtake each other.
- If a sender sends two messages (Message 1 and Message 2) in succession to the same destination, and both match the same receive, the receive operation will receive Message 1 before Message 2.
- If a receiver posts two receives (Receive 1 and Receive 2), in succession, and both are looking for the same message, Receive 1 will receive the message before Receive 2.
- Order rules do not apply if there are multiple threads participating in the communication operations.
- Fairness:
- MPI does not guarantee fairness - it's up to the programmer to prevent "operation starvation".
- Example: task 0 sends a message to task 2. However, task 1 sends a competing message that matches task 2's receive. Only one of the sends will complete.
| Point to Point Communication Routines |
MPI Message Passing Routine Arguments
MPI point-to-point communication routines generally have an argument list that takes one of the following formats:
| Blocking sends | | Non-blocking sends
| Blocking receive
| Non-blocking receive
| |
- Buffer
- Program (application) address space that references the data that is
to be sent or received. In most cases, this is simply the variable
name that is be sent/received. For C programs, this argument is
passed by reference and usually must be prepended with an ampersand:
&var1
- Data Count
- Indicates the number of data elements of a particular type to be sent.
- Data Type
- For reasons of portability, MPI predefines its elementary data types.
The table below lists those required by the standard.
C Data Types Fortran Data Types MPI_CHAR signed char MPI_CHARACTER character(1) MPI_SHORT signed short int MPI_INT signed int MPI_INTEGER integer MPI_LONG signed long int MPI_UNSIGNED_CHAR unsigned char MPI_UNSIGNED_SHORT unsigned short int MPI_UNSIGNED unsigned int MPI_UNSIGNED_LONG unsigned long int MPI_FLOAT float MPI_REAL real MPI_DOUBLE double MPI_DOUBLE_PRECISION double precision MPI_LONG_DOUBLE long double MPI_COMPLEX complex MPI_DOUBLE_COMPLEX double complex MPI_LOGICAL logical MPI_BYTE 8 binary digits MPI_BYTE 8 binary digits MPI_PACKED data packed or unpacked with MPI_Pack()/ MPI_Unpack MPI_PACKED data packed or unpacked with MPI_Pack()/ MPI_Unpack Notes:
- Programmers may also create their own data types (see Derived Data Types).
- MPI_BYTE and MPI_PACKED do not correspond to standard C or Fortran types.
- The MPI standard includes the following optional data types:
- C: MPI_LONG_LONG_INT
- Fortran: MPI_INTEGER1, MPI_INTEGER2, MPI_INTEGER4, MPI_REAL2, MPI_REAL4, MPI_REAL8
- Some implementations may include additional elementary data types (MPI_LOGICAL2, MPI_COMPLEX32, etc.). Check the MPI header file.
- Destination
- An argument to send routines that indicates the process where a
message should be delivered. Specified as the rank of the receiving
process.
- Source
- An argument to receive routines that indicates the originating process
of the message. Specified as the rank of the sending process. This may
be set to the wild card MPI_ANY_SOURCE to receive a message from any task.
- Tag
- Arbitrary non-negative integer assigned by the programmer to uniquely
identify a message. Send and receive operations should match message
tags. For a receive operation, the wild card MPI_ANY_TAG can be used
to receive any message regardless of its tag. The MPI standard guarantees
that integers 0-32767 can be used as tags, but most implementations allow
a much larger range than this.
- Communicator
- Indicates the communication context, or set of processes for which the
source or destination fields are valid. Unless the programmer is
explicitly creating new communicators, the predefined communicator
MPI_COMM_WORLD is usually used.
- Status
- For a receive operation, indicates the source of the message and the
tag of the message. In C, this argument is a pointer to a predefined
structure MPI_Status (ex. stat.MPI_SOURCE stat.MPI_TAG).
In Fortran, it is an integer array of size MPI_STATUS_SIZE (ex.
stat(MPI_SOURCE) stat(MPI_TAG)). Additionally, the actual number of
bytes received are obtainable from Status via the MPI_Get_count
routine.
- Request
- Used by non-blocking send and receive operations. Since non-blocking operations may return before the requested system buffer space is obtained, the system issues a unique "request number". The programmer uses this system assigned "handle" later (in a WAIT type routine) to determine completion of the non-blocking operation. In C, this argument is a pointer to a predefined structure MPI_Request. In Fortran, it is an integer.
| Point to Point Communication Routines |
Blocking Message Passing Routines
The more commonly used MPI blocking message passing routines are described below.
- MPI_Send
- Basic blocking send operation. Routine returns only after the
application buffer in the sending task is free for reuse. Note that
this routine may be implemented differently on different systems. The
MPI standard permits the use of a system buffer but does not require it.
Some implementations may actually use a synchronous send (discussed
below) to implement the basic blocking send.
MPI_Send (&buf,count,datatype,dest,tag,comm)
MPI_SEND (buf,count,datatype,dest,tag,comm,ierr) - MPI_Recv
- Receive a message and block until the requested data is available in
the application buffer in the receiving task.
MPI_Recv (&buf,count,datatype,source,tag,comm,&status)
MPI_RECV (buf,count,datatype,source,tag,comm,status,ierr) - MPI_Ssend
- Synchronous blocking send: Send a message and block until the
application buffer in the sending task is free for reuse and the
destination process has started to receive the message.
MPI_Ssend (&buf,count,datatype,dest,tag,comm)
MPI_SSEND (buf,count,datatype,dest,tag,comm,ierr) - MPI_Bsend
- Buffered blocking send: permits the programmer to allocate the required
amount of buffer space into which data can be copied until it is
delivered. Insulates against the problems associated with insufficient
system buffer space. Routine returns after the data has been copied
from application buffer space to the allocated send buffer. Must be
used with the MPI_Buffer_attach routine.
MPI_Bsend (&buf,count,datatype,dest,tag,comm)
MPI_BSEND (buf,count,datatype,dest,tag,comm,ierr) - MPI_Buffer_attach
MPI_Buffer_detach - Used by programmer to allocate/deallocate message buffer space to be
used by the MPI_Bsend routine. The size argument is specified in
actual data bytes - not a count of data elements.
Only one buffer can be attached to a process at a time. Note that the
IBM implementation uses MPI_BSEND_OVERHEAD bytes of the allocated buffer
for overhead.
MPI_Buffer_attach (&buffer,size)
MPI_Buffer_detach (&buffer,size)
MPI_BUFFER_ATTACH (buffer,size,ierr)
MPI_BUFFER_DETACH (buffer,size,ierr) - MPI_Rsend
- Blocking ready send. Should only be used if the programmer is certain
that the matching receive has already been posted.
MPI_Rsend (&buf,count,datatype,dest,tag,comm)
MPI_RSEND (buf,count,datatype,dest,tag,comm,ierr) - MPI_Sendrecv
- Send a message and post a receive before blocking. Will block until
the sending application buffer is free for reuse and until the receiving
application buffer contains the received message.
MPI_Sendrecv (&sendbuf,sendcount,sendtype,dest,sendtag,
...... &recvbuf,recvcount,recvtype,source,recvtag,
...... comm,&status)
MPI_SENDRECV (sendbuf,sendcount,sendtype,dest,sendtag,
...... recvbuf,recvcount,recvtype,source,recvtag,
...... comm,status,ierr) - MPI_Wait
MPI_Waitany
MPI_Waitall
MPI_Waitsome - MPI_Wait blocks until a specified non-blocking send or receive
operation has completed. For multiple non-blocking operations, the
programmer can specify any, all or some completions.
MPI_Wait (&request,&status)
MPI_Waitany (count,&array_of_requests,&index,&status)
MPI_Waitall (count,&array_of_requests,&array_of_statuses)
MPI_Waitsome (incount,&array_of_requests,&outcount,
...... &array_of_offsets, &array_of_statuses)
MPI_WAIT (request,status,ierr)
MPI_WAITANY (count,array_of_requests,index,status,ierr)
MPI_WAITALL (count,array_of_requests,array_of_statuses,
...... ierr)
MPI_WAITSOME (incount,array_of_requests,outcount,
...... array_of_offsets, array_of_statuses,ierr) - MPI_Probe
- Performs a blocking test for a message. The "wildcards" MPI_ANY_SOURCE
and MPI_ANY_TAG may be used to test for a message from any source or
with any tag. For the C routine, the actual source and tag will be
returned in the status structure as status.MPI_SOURCE and
status.MPI_TAG. For the Fortran routine, they will be returned in
the integer array status(MPI_SOURCE) and status(MPI_TAG).
MPI_Probe (source,tag,comm,&status)
MPI_PROBE (source,tag,comm,status,ierr)
Examples: Blocking Message Passing Routines
Task 0 pings task 1 and awaits return ping
C Language - Blocking Message Passing
Routines Example
#include "mpi.h"
#include <stdio.h>
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, dest, source, rc, count, tag=1;
char inmsg, outmsg='x';
MPI_Status Stat;
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if (rank == 0) {
dest = 1;
source = 1;
rc = MPI_Send(&outmsg, 1, MPI_CHAR, dest, tag, MPI_COMM_WORLD);
rc = MPI_Recv(&inmsg, 1, MPI_CHAR, source, tag, MPI_COMM_WORLD, &Stat);
}
else if (rank == 1) {
dest = 0;
source = 0;
rc = MPI_Recv(&inmsg, 1, MPI_CHAR, source, tag, MPI_COMM_WORLD, &Stat);
rc = MPI_Send(&outmsg, 1, MPI_CHAR, dest, tag, MPI_COMM_WORLD);
}
rc = MPI_Get_count(&Stat, MPI_CHAR, &count);
printf("Task %d: Received %d char(s) from task %d with tag %d \n",
rank, count, Stat.MPI_SOURCE, Stat.MPI_TAG);
MPI_Finalize();
}
|
Fortran - Blocking Message Passing
Routines Example
program ping
include 'mpif.h'
integer numtasks, rank, dest, source, count, tag, ierr
integer stat(MPI_STATUS_SIZE)
character inmsg, outmsg
outmsg = 'x'
tag = 1
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
if (rank .eq. 0) then
dest = 1
source = 1
call MPI_SEND(outmsg, 1, MPI_CHARACTER, dest, tag,
& MPI_COMM_WORLD, ierr)
call MPI_RECV(inmsg, 1, MPI_CHARACTER, source, tag,
& MPI_COMM_WORLD, stat, ierr)
else if (rank .eq. 1) then
dest = 0
source = 0
call MPI_RECV(inmsg, 1, MPI_CHARACTER, source, tag,
& MPI_COMM_WORLD, stat, err)
call MPI_SEND(outmsg, 1, MPI_CHARACTER, dest, tag,
& MPI_COMM_WORLD, err)
endif
call MPI_GET_COUNT(stat, MPI_CHARACTER, count, ierr)
print *, 'Task ',rank,': Received', count, 'char(s) from task',
& stat(MPI_SOURCE), 'with tag',stat(MPI_TAG)
call MPI_FINALIZE(ierr)
end
|
| Point to Point Communication Routines |
Non-Blocking Message Passing Routines
The more commonly used MPI non-blocking message passing routines are described below.
- MPI_Isend
- Identifies an area in memory to serve as a send buffer. Processing
continues immediately without waiting for the message to be copied out
from the application buffer. A communication request handle is
returned for handling the pending message status. The program should
not modify the application buffer until subsequent calls to MPI_Wait
or MPI_Test indicate that the non-blocking send has completed.
MPI_Isend (&buf,count,datatype,dest,tag,comm,&request)
MPI_ISEND (buf,count,datatype,dest,tag,comm,request,ierr) - MPI_Irecv
- Identifies an area in memory to serve as a receive buffer. Processing
continues immediately without actually waiting for the message to be
received and copied into the the application buffer. A communication
request handle is returned for handling the pending message status.
The program must use calls to MPI_Wait or MPI_Test to determine when the
non-blocking receive operation completes and the requested message is
available in the application buffer.
MPI_Irecv (&buf,count,datatype,source,tag,comm,&request)
MPI_IRECV (buf,count,datatype,source,tag,comm,request,ierr) - MPI_Issend
- Non-blocking synchronous send. Similar to MPI_Isend(), except
MPI_Wait() or MPI_Test() indicates when the destination process has
received the message.
MPI_Issend (&buf,count,datatype,dest,tag,comm,&request)
MPI_ISSEND (buf,count,datatype,dest,tag,comm,request,ierr) - MPI_Ibsend
- Non-blocking buffered send. Similar to MPI_Bsend() except MPI_Wait()
or MPI_Test() indicates when the destination process has received the
message. Must be used with the MPI_Buffer_attach routine.
MPI_Ibsend (&buf,count,datatype,dest,tag,comm,&request)
MPI_IBSEND (buf,count,datatype,dest,tag,comm,request,ierr) - MPI_Irsend
- Non-blocking ready send. Similar to MPI_Rsend() except MPI_Wait()
or MPI_Test() indicates when the destination process has received the
message. Should only be used if the programmer is certain that the
matching receive has already been posted.
MPI_Irsend (&buf,count,datatype,dest,tag,comm,&request)
MPI_IRSEND (buf,count,datatype,dest,tag,comm,request,ierr) - MPI_Test
MPI_Testany
MPI_Testall
MPI_Testsome - MPI_Test checks the status of a specified non-blocking send or receive
operation. The "flag" parameter is returned logical true (1) if the
operation has completed, and logical false (0) if not. For multiple
non-blocking operations, the programmer can specify any, all or some
completions.
MPI_Test (&request,&flag,&status)
MPI_Testany (count,&array_of_requests,&index,&flag,&status)
MPI_Testall (count,&array_of_requests,&flag,&array_of_statuses)
MPI_Testsome (incount,&array_of_requests,&outcount,
...... &array_of_offsets, &array_of_statuses)
MPI_TEST (request,flag,status,ierr)
MPI_TESTANY (count,array_of_requests,index,flag,status,ierr)
MPI_TESTALL (count,array_of_requests,flag,array_of_statuses,ierr)
MPI_TESTSOME (incount,array_of_requests,outcount,
...... array_of_offsets, array_of_statuses,ierr) - MPI_Iprobe
- Performs a non-blocking test for a message. The "wildcards"
MPI_ANY_SOURCE and MPI_ANY_TAG may be used to test for a message
from any source or with any tag. The integer "flag" parameter is returned
logical true (1) if a message has arrived, and logical false (0) if not.
For the C routine, the actual source and tag will be
returned in the status structure as status.MPI_SOURCE and
status.MPI_TAG. For the Fortran routine, they will be returned in
the integer array status(MPI_SOURCE) and status(MPI_TAG).
MPI_Iprobe (source,tag,comm,&flag,&status)
MPI_IPROBE (source,tag,comm,flag,status,ierr)
Examples: Non-Blocking Message Passing Routines
Nearest neighbor exchange in ring topology
C Language - Non-Blocking Message
Passing Routines Example
#include "mpi.h"
#include <stdio.h>
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, next, prev, buf[2], tag1=1, tag2=2;
MPI_Request reqs[4];
MPI_Status stats[4];
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
prev = rank-1;
next = rank+1;
if (rank == 0) prev = numtasks - 1;
if (rank == (numtasks - 1)) next = 0;
MPI_Irecv(&buf[0], 1, MPI_INT, prev, tag1, MPI_COMM_WORLD, &reqs[0]);
MPI_Irecv(&buf[1], 1, MPI_INT, next, tag2, MPI_COMM_WORLD, &reqs[1]);
MPI_Isend(&rank, 1, MPI_INT, prev, tag2, MPI_COMM_WORLD, &reqs[2]);
MPI_Isend(&rank, 1, MPI_INT, next, tag1, MPI_COMM_WORLD, &reqs[3]);
{ do some work }
MPI_Waitall(4, reqs, stats);
MPI_Finalize();
}
|
Fortran - Non-Blocking Message
Passing Routines Example
program ringtopo
include 'mpif.h'
integer numtasks, rank, next, prev, buf(2), tag1, tag2, ierr
integer stats(MPI_STATUS_SIZE,4), reqs(4)
tag1 = 1
tag2 = 2
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
prev = rank - 1
next = rank + 1
if (rank .eq. 0) then
prev = numtasks - 1
endif
if (rank .eq. numtasks - 1) then
next = 0
endif
call MPI_IRECV(buf(1), 1, MPI_INTEGER, prev, tag1,
& MPI_COMM_WORLD, reqs(1), ierr)
call MPI_IRECV(buf(2), 1, MPI_INTEGER, next, tag2,
& MPI_COMM_WORLD, reqs(2), ierr)
call MPI_ISEND(rank, 1, MPI_INTEGER, prev, tag2,
& MPI_COMM_WORLD, reqs(3), ierr)
call MPI_ISEND(rank, 1, MPI_INTEGER, next, tag1,
& MPI_COMM_WORLD, reqs(4), ierr)
C do some work
call MPI_WAITALL(4, reqs, stats, ierr);
call MPI_FINALIZE(ierr)
end
|
| Collective Communication Routines |
All or None:
- Collective communication must involve all processes in the
scope of a communicator. All processes are by default, members in the
communicator MPI_COMM_WORLD.
- It is the programmer's responsibility to insure that all processes within a communicator participate in any collective operations.
Types of Collective Operations:
- Synchronization - processes wait until all members of the group
have reached the synchronization point.
- Data Movement - broadcast, scatter/gather, all to all.
- Collective Computation (reductions) - one member of the group collects data from the other members and performs an operation (min, max, add, multiply, etc.) on that data.
Programming Considerations and Restrictions:
- Collective operations are blocking.
- Collective communication routines do not take message tag arguments.
- Collective operations within subsets of processes are accomplished
by first partitioning the subsets into new groups and then
attaching the new groups to new communicators (discussed in the
Group and Communicator
Management Routines section).
- Can only be used with MPI predefined datatypes - not with MPI Derived Data Types.
Collective Communication Routines
- MPI_Barrier
- Creates a barrier synchronization in a group. Each task, when reaching
the MPI_Barrier call, blocks until all tasks in the group reach the same
MPI_Barrier call.
MPI_Barrier (comm)
MPI_BARRIER (comm,ierr)
Examples: Collective Communications
Perform a scatter operation on the rows of an array
C Language - Collective Communications
Example
#include "mpi.h"
#include <stdio.h>
#define SIZE 4
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, sendcount, recvcount, source;
float sendbuf[SIZE][SIZE] = {
{1.0, 2.0, 3.0, 4.0},
{5.0, 6.0, 7.0, 8.0},
{9.0, 10.0, 11.0, 12.0},
{13.0, 14.0, 15.0, 16.0} };
float recvbuf[SIZE];
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
if (numtasks == SIZE) {
source = 1;
sendcount = SIZE;
recvcount = SIZE;
MPI_Scatter(sendbuf,sendcount,MPI_FLOAT,recvbuf,recvcount,
MPI_FLOAT,source,MPI_COMM_WORLD);
printf("rank= %d Results: %f %f %f %f\n",rank,recvbuf[0],
recvbuf[1],recvbuf[2],recvbuf[3]);
}
else
printf("Must specify %d processors. Terminating.\n",SIZE);
MPI_Finalize();
}
|
Fortran - Collective Communications
Example
program scatter
include 'mpif.h'
integer SIZE
parameter(SIZE=4)
integer numtasks, rank, sendcount, recvcount, source, ierr
real*4 sendbuf(SIZE,SIZE), recvbuf(SIZE)
C Fortran stores this array in column major order, so the
C scatter will actually scatter columns, not rows.
data sendbuf /1.0, 2.0, 3.0, 4.0,
& 5.0, 6.0, 7.0, 8.0,
& 9.0, 10.0, 11.0, 12.0,
& 13.0, 14.0, 15.0, 16.0 /
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
if (numtasks .eq. SIZE) then
source = 1
sendcount = SIZE
recvcount = SIZE
call MPI_SCATTER(sendbuf, sendcount, MPI_REAL, recvbuf,
& recvcount, MPI_REAL, source, MPI_COMM_WORLD, ierr)
print *, 'rank= ',rank,' Results: ',recvbuf
else
print *, 'Must specify',SIZE,' processors. Terminating.'
endif
call MPI_FINALIZE(ierr)
end
|
Sample program output:
rank= 0 Results: 1.000000 2.000000 3.000000 4.000000 rank= 1 Results: 5.000000 6.000000 7.000000 8.000000 rank= 2 Results: 9.000000 10.000000 11.000000 12.000000 rank= 3 Results: 13.000000 14.000000 15.000000 16.000000
| Derived Data Types |
- As previously mentioned, MPI
predefines its primitive data types:
C Data Types Fortran Data Types MPI_CHAR
MPI_SHORT
MPI_INT
MPI_LONG
MPI_UNSIGNED_CHAR
MPI_UNSIGNED_SHORT
MPI_UNSIGNED_LONG
MPI_UNSIGNED
MPI_FLOAT
MPI_DOUBLE
MPI_LONG_DOUBLE
MPI_BYTE
MPI_PACKEDMPI_CHARACTER
MPI_INTEGER
MPI_REAL
MPI_DOUBLE_PRECISION
MPI_COMPLEX
MPI_DOUBLE_COMPLEX
MPI_LOGICAL
MPI_BYTE
MPI_PACKED - MPI also provides facilities for you to define your own data structures
based upon sequences of the MPI primitive data types. Such user defined
structures are called derived data types.
- Primitive data types are contiguous. Derived data types allow you to
specify non-contiguous data in a convenient manner and to treat it as
though it was contiguous.
- MPI provides several methods for constructing derived data types:
- Contiguous
- Vector
- Indexed
- Struct
Derived Data Type Routines
- MPI_Type_contiguous
- The simplest constructor. Produces a new data type by making count
copies of an existing data type.
MPI_Type_contiguous (count,oldtype,&newtype)
MPI_TYPE_CONTIGUOUS (count,oldtype,newtype,ierr) - MPI_Type_vector
MPI_Type_hvector - Similar to contiguous, but allows for regular gaps (stride) in the
displacements. MPI_Type_hvector is identical to
MPI_Type_vector except that stride is specified in bytes.
MPI_Type_vector (count,blocklength,stride,oldtype,&newtype)
MPI_TYPE_VECTOR (count,blocklength,stride,oldtype,newtype,ierr) - MPI_Type_indexed
MPI_Type_hindexed - An array of displacements of the input data type is provided as the map
for the new data type. MPI_Type_hindexed is identical to
MPI_Type_indexed except that offsets are specified in bytes.
MPI_Type_indexed (count,blocklens[],offsets[],old_type,&newtype)
MPI_TYPE_INDEXED (count,blocklens(),offsets(),old_type,newtype,ierr) - MPI_Type_struct
- The new data type is formed according to completely defined map of the
component data types.
MPI_Type_struct (count,blocklens[],offsets[],old_types,&newtype)
MPI_TYPE_STRUCT (count,blocklens(),offsets(),old_types,newtype,ierr) - MPI_Type_extent
- Returns the size in bytes of the specified data type. Useful for
the MPI subroutines that require specification of offsets in bytes.
MPI_Type_extent (datatype,&extent)
MPI_TYPE_EXTENT (datatype,extent,ierr) - MPI_Type_commit
- Commits new datatype to the system. Required for all user constructed
(derived) datatypes.
MPI_Type_commit (&datatype)
MPI_TYPE_COMMIT (datatype,ierr) - MPI_Type_free
- Deallocates the specified datatype object. Use of this routine is
especially important to prevent memory exhaustion if many datatype
objects are created, as in a loop.
MPI_Type_free (&datatype)
MPI_TYPE_FREE (datatype,ierr)
Examples: Contiguous Derived Data Type
Create a data type representing a row of an array and distribute a
different row to all processes.
C Language - Contiguous Derived Data Type
Example
#include "mpi.h"
#include <stdio.h>
#define SIZE 4
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, source=0, dest, tag=1, i;
float a[SIZE][SIZE] =
{1.0, 2.0, 3.0, 4.0,
5.0, 6.0, 7.0, 8.0,
9.0, 10.0, 11.0, 12.0,
13.0, 14.0, 15.0, 16.0};
float b[SIZE];
MPI_Status stat;
MPI_Datatype rowtype;
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
MPI_Type_contiguous(SIZE, MPI_FLOAT, &rowtype);
MPI_Type_commit(&rowtype);
if (numtasks == SIZE) {
if (rank == 0) {
for (i=0; i<numtasks; i++)
MPI_Send(&a[i][0], 1, rowtype, i, tag, MPI_COMM_WORLD);
}
MPI_Recv(b, SIZE, MPI_FLOAT, source, tag, MPI_COMM_WORLD, &stat);
printf("rank= %d b= %3.1f %3.1f %3.1f %3.1f\n",
rank,b[0],b[1],b[2],b[3]);
}
else
printf("Must specify %d processors. Terminating.\n",SIZE);
MPI_Type_free(&rowtype);
MPI_Finalize();
}
|
Fortran - Contiguous Derived Data Type
Example
program contiguous
include 'mpif.h'
integer SIZE
parameter(SIZE=4)
integer numtasks, rank, source, dest, tag, i, ierr
real*4 a(0:SIZE-1,0:SIZE-1), b(0:SIZE-1)
integer stat(MPI_STATUS_SIZE), columntype
C Fortran stores this array in column major order
data a /1.0, 2.0, 3.0, 4.0,
& 5.0, 6.0, 7.0, 8.0,
& 9.0, 10.0, 11.0, 12.0,
& 13.0, 14.0, 15.0, 16.0 /
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
call MPI_TYPE_CONTIGUOUS(SIZE, MPI_REAL, columntype, ierr)
call MPI_TYPE_COMMIT(columntype, ierr)
tag = 1
if (numtasks .eq. SIZE) then
if (rank .eq. 0) then
do 10 i=0, numtasks-1
call MPI_SEND(a(0,i), 1, columntype, i, tag,
& MPI_COMM_WORLD,ierr)
10 continue
endif
source = 0
call MPI_RECV(b, SIZE, MPI_REAL, source, tag,
& MPI_COMM_WORLD, stat, ierr)
print *, 'rank= ',rank,' b= ',b
else
print *, 'Must specify',SIZE,' processors. Terminating.'
endif
call MPI_TYPE_FREE(columntype, ierr)
call MPI_FINALIZE(ierr)
end
|
Sample program output:
rank= 0 b= 1.0 2.0 3.0 4.0 rank= 1 b= 5.0 6.0 7.0 8.0 rank= 2 b= 9.0 10.0 11.0 12.0 rank= 3 b= 13.0 14.0 15.0 16.0
Examples: Vector Derived Data Type
Create a data type representing a column of an array and distribute
different columns to all processes.
C Language - Vector Derived Data Type
Example
#include "mpi.h"
#include <stdio.h>
#define SIZE 4
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, source=0, dest, tag=1, i;
float a[SIZE][SIZE] =
{1.0, 2.0, 3.0, 4.0,
5.0, 6.0, 7.0, 8.0,
9.0, 10.0, 11.0, 12.0,
13.0, 14.0, 15.0, 16.0};
float b[SIZE];
MPI_Status stat;
MPI_Datatype columntype;
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
MPI_Type_vector(SIZE, 1, SIZE, MPI_FLOAT, &columntype);
MPI_Type_commit(&columntype);
if (numtasks == SIZE) {
if (rank == 0) {
for (i=0; i<numtasks; i++)
MPI_Send(&a[0][i], 1, columntype, i, tag, MPI_COMM_WORLD);
}
MPI_Recv(b, SIZE, MPI_FLOAT, source, tag, MPI_COMM_WORLD, &stat);
printf("rank= %d b= %3.1f %3.1f %3.1f %3.1f\n",
rank,b[0],b[1],b[2],b[3]);
}
else
printf("Must specify %d processors. Terminating.\n",SIZE);
MPI_Type_free(&columntype);
MPI_Finalize();
}
|
Fortran - Vector Derived Data Type
Example
program vector
include 'mpif.h'
integer SIZE
parameter(SIZE=4)
integer numtasks, rank, source, dest, tag, i, ierr
real*4 a(0:SIZE-1,0:SIZE-1), b(0:SIZE-1)
integer stat(MPI_STATUS_SIZE), rowtype
C Fortran stores this array in column major order
data a /1.0, 2.0, 3.0, 4.0,
& 5.0, 6.0, 7.0, 8.0,
& 9.0, 10.0, 11.0, 12.0,
& 13.0, 14.0, 15.0, 16.0 /
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
call MPI_TYPE_VECTOR(SIZE, 1, SIZE, MPI_REAL, rowtype, ierr)
call MPI_TYPE_COMMIT(rowtype, ierr)
tag = 1
if (numtasks .eq. SIZE) then
if (rank .eq. 0) then
do 10 i=0, numtasks-1
call MPI_SEND(a(i,0), 1, rowtype, i, tag,
& MPI_COMM_WORLD, ierr)
10 continue
endif
source = 0
call MPI_RECV(b, SIZE, MPI_REAL, source, tag,
& MPI_COMM_WORLD, stat, ierr)
print *, 'rank= ',rank,' b= ',b
else
print *, 'Must specify',SIZE,' processors. Terminating.'
endif
call MPI_TYPE_FREE(rowtype, ierr)
call MPI_FINALIZE(ierr)
end
|
Sample program output:
rank= 0 b= 1.0 5.0 9.0 13.0 rank= 1 b= 2.0 6.0 10.0 14.0 rank= 2 b= 3.0 7.0 11.0 15.0 rank= 3 b= 4.0 8.0 12.0 16.0
Examples: Indexed Derived Data Type
Create a datatype by extracting variable portions of an array and distribute
to all tasks.
C Language - Indexed Derived Data Type
Example
#include "mpi.h"
#include <stdio.h>
#define NELEMENTS 6
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, source=0, dest, tag=1, i;
int blocklengths[2], displacements[2];
float a[16] =
{1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0,
9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0};
float b[NELEMENTS];
MPI_Status stat;
MPI_Datatype indextype;
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
blocklengths[0] = 4;
blocklengths[1] = 2;
displacements[0] = 5;
displacements[1] = 12;
MPI_Type_indexed(2, blocklengths, displacements, MPI_FLOAT, &indextype);
MPI_Type_commit(&indextype);
if (rank == 0) {
for (i=0; i<numtasks; i++)
MPI_Send(a, 1, indextype, i, tag, MPI_COMM_WORLD);
}
MPI_Recv(b, NELEMENTS, MPI_FLOAT, source, tag, MPI_COMM_WORLD, &stat);
printf("rank= %d b= %3.1f %3.1f %3.1f %3.1f %3.1f %3.1f\n",
rank,b[0],b[1],b[2],b[3],b[4],b[5]);
MPI_Type_free(&indextype);
MPI_Finalize();
}
|
Fortran - Indexed Derived Data Type
Example
program indexed
include 'mpif.h'
integer NELEMENTS
parameter(NELEMENTS=6)
integer numtasks, rank, source, dest, tag, i, ierr
integer blocklengths(0:1), displacements(0:1)
real*4 a(0:15), b(0:NELEMENTS-1)
integer stat(MPI_STATUS_SIZE), indextype
data a /1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0,
& 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0 /
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
blocklengths(0) = 4
blocklengths(1) = 2
displacements(0) = 5
displacements(1) = 12
call MPI_TYPE_INDEXED(2, blocklengths, displacements, MPI_REAL,
& indextype, ierr)
call MPI_TYPE_COMMIT(indextype, ierr)
tag = 1
if (rank .eq. 0) then
do 10 i=0, numtasks-1
call MPI_SEND(a, 1, indextype, i, tag, MPI_COMM_WORLD, ierr)
10 continue
endif
source = 0
call MPI_RECV(b, NELEMENTS, MPI_REAL, source, tag, MPI_COMM_WORLD,
& stat, ierr)
print *, 'rank= ',rank,' b= ',b
call MPI_TYPE_FREE(indextype, ierr)
call MPI_FINALIZE(ierr)
end
|
Sample program output:
rank= 0 b= 6.0 7.0 8.0 9.0 13.0 14.0 rank= 1 b= 6.0 7.0 8.0 9.0 13.0 14.0 rank= 2 b= 6.0 7.0 8.0 9.0 13.0 14.0 rank= 3 b= 6.0 7.0 8.0 9.0 13.0 14.0
Examples: Struct Derived Data Type
Create a data type that represents a particle and distribute an array
of such particles to all processes.
C Language - Struct Derived Data Type
Example
#include "mpi.h"
#include <stdio.h>
#define NELEM 25
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, source=0, dest, tag=1, i;
typedef struct {
float x, y, z;
float velocity;
int n, type;
} Particle;
Particle p[NELEM], particles[NELEM];
MPI_Datatype particletype, oldtypes[2];
int blockcounts[2];
/* MPI_Aint type used to be consistent with syntax of */
/* MPI_Type_extent routine */
MPI_Aint offsets[2], extent;
MPI_Status stat;
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
/* Setup description of the 4 MPI_FLOAT fields x, y, z, velocity */
offsets[0] = 0;
oldtypes[0] = MPI_FLOAT;
blockcounts[0] = 4;
/* Setup description of the 2 MPI_INT fields n, type */
/* Need to first figure offset by getting size of MPI_FLOAT */
MPI_Type_extent(MPI_FLOAT, &extent);
offsets[1] = 4 * extent;
oldtypes[1] = MPI_INT;
blockcounts[1] = 2;
/* Now define structured type and commit it */
MPI_Type_struct(2, blockcounts, offsets, oldtypes, &particletype);
MPI_Type_commit(&particletype);
/* Initialize the particle array and then send it to each task */
if (rank == 0) {
for (i=0; i<NELEM; i++) {
particles[i].x = i * 1.0;
particles[i].y = i * -1.0;
particles[i].z = i * 1.0;
particles[i].velocity = 0.25;
particles[i].n = i;
particles[i].type = i % 2;
}
for (i=0; i<numtasks; i++)
MPI_Send(particles, NELEM, particletype, i, tag, MPI_COMM_WORLD);
}
MPI_Recv(p, NELEM, particletype, source, tag, MPI_COMM_WORLD, &stat);
/* Print a sample of what was received */
printf("rank= %d %3.2f %3.2f %3.2f %3.2f %d %d\n", rank,p[3].x,
p[3].y,p[3].z,p[3].velocity,p[3].n,p[3].type);
MPI_Type_free(&particletype);
MPI_Finalize();
}
|
Fortran - Struct Derived Data Type
Example
program struct
include 'mpif.h'
integer NELEM
parameter(NELEM=25)
integer numtasks, rank, source, dest, tag, i, ierr
integer stat(MPI_STATUS_SIZE)
type Particle
sequence
real*4 x, y, z, velocity
integer n, type
end type Particle
type (Particle) p(NELEM), particles(NELEM)
integer particletype, oldtypes(0:1), blockcounts(0:1),
& offsets(0:1), extent
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
C Setup description of the 4 MPI_REAL fields x, y, z, velocity
offsets(0) = 0
oldtypes(0) = MPI_REAL
blockcounts(0) = 4
C Setup description of the 2 MPI_INTEGER fields n, type
C Need to first figure offset by getting size of MPI_REAL
call MPI_TYPE_EXTENT(MPI_REAL, extent, ierr)
offsets(1) = 4 * extent
oldtypes(1) = MPI_INTEGER
blockcounts(1) = 2
C Now define structured type and commit it
call MPI_TYPE_STRUCT(2, blockcounts, offsets, oldtypes,
& particletype, ierr)
call MPI_TYPE_COMMIT(particletype, ierr)
C Initialize the particle array and then send it to each task
tag = 1
if (rank .eq. 0) then
do 10 i=0, NELEM-1
particles(i) = Particle ( 1.0*i, -1.0*i, 1.0*i,
& 0.25, i, mod(i,2) )
10 continue
do 20 i=0, numtasks-1
call MPI_SEND(particles, NELEM, particletype, i, tag,
& MPI_COMM_WORLD, ierr)
20 continue
endif
source = 0
call MPI_RECV(p, NELEM, particletype, source, tag,
& MPI_COMM_WORLD, stat, ierr)
print *, 'rank= ',rank,' p(3)= ',p(3)
call MPI_TYPE_FREE(particletype, ierr)
call MPI_FINALIZE(ierr)
end
|
Sample program output:
rank= 0 3.00 -3.00 3.00 0.25 3 1 rank= 2 3.00 -3.00 3.00 0.25 3 1 rank= 1 3.00 -3.00 3.00 0.25 3 1 rank= 3 3.00 -3.00 3.00 0.25 3 1
| Group and Communicator Management Routines |
Groups vs. Communicators:
- A group is an ordered set of processes. Each process in a group is
associated with a unique integer rank. Rank values start at zero and go
to N-1, where N is the number of processes in the group.
In MPI, a group is represented within system memory as an object.
It is accessible to the programmer only by a "handle". A group is always
associated with a communicator object.
- A communicator encompasses a group of processes that may communicate with
each other. All MPI messages must specify a communicator. In the
simplest sense, the communicator is an extra "tag" that must be included
with MPI calls.
Like groups, communicators are represented within system memory as
objects and are accessible to the programmer only by "handles".
For example, the handle for the communicator that comprises all tasks
is MPI_COMM_WORLD.
- From the programmer's perspective, a group and a communicator are one. The group routines are primarily used to specify which processes should be used to construct a communicator.
Primary Purposes of Group and Communicator Objects:
- Allow you to organize tasks, based upon function, into task groups.
- Enable Collective Communications operations across a subset of
related tasks.
- Provide basis for implementing user defined virtual topologies
- Provide for safe communications
Programming Considerations and Restrictions:
- Groups/communicators are dynamic - they can be created and destroyed
during program execution.
- Processes may be in more than one group/communicator. They will have
a unique rank within each group/communicator.
- MPI provides over 40 routines related to groups, communicators,
and virtual topologies.
- Typical usage:
- Extract handle of global group from MPI_COMM_WORLD using MPI_Comm_group
- Form new group as a subset of global group using MPI_Group_incl
- Create new communicator for new group using MPI_Comm_create
- Determine new rank in new communicator using MPI_Comm_rank
- Conduct communications using any MPI message passing routine
- When finished, free up new communicator and group (optional) using MPI_Comm_free and MPI_Group_free
Group and Communicator Management Routines
Create two different process groups for separate collective communications exchange. Requires creating new communicators also.
C Language - Group and Communicator
Routines Example
#include "mpi.h"
#include <stdio.h>
#define NPROCS 8
int main(argc,argv)
int argc;
char *argv[]; {
int rank, new_rank, sendbuf, recvbuf, numtasks,
ranks1[4]={0,1,2,3}, ranks2[4]={4,5,6,7};
MPI_Group orig_group, new_group;
MPI_Comm new_comm;
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
if (numtasks != NPROCS) {
printf("Must specify MP_PROCS= %d. Terminating.\n",NPROCS);
MPI_Finalize();
exit(0);
}
sendbuf = rank;
/* Extract the original group handle */
MPI_Comm_group(MPI_COMM_WORLD, &orig_group);
/* Divide tasks into two distinct groups based upon rank */
if (rank < NPROCS/2) {
MPI_Group_incl(orig_group, NPROCS/2, ranks1, &new_group);
}
else {
MPI_Group_incl(orig_group, NPROCS/2, ranks2, &new_group);
}
/* Create new new communicator and then perform collective communications */
MPI_Comm_create(MPI_COMM_WORLD, new_group, &new_comm);
MPI_Allreduce(&sendbuf, &recvbuf, 1, MPI_INT, MPI_SUM, new_comm);
MPI_Group_rank (new_group, &new_rank);
printf("rank= %d newrank= %d recvbuf= %d\n",rank,new_rank,recvbuf);
MPI_Finalize();
}
|
Fortran - Group and Communicator
Routines Example
program group
include 'mpif.h'
integer NPROCS
parameter(NPROCS=8)
integer rank, new_rank, sendbuf, recvbuf, numtasks
integer ranks1(4), ranks2(4), ierr
integer orig_group, new_group, new_comm
data ranks1 /0, 1, 2, 3/, ranks2 /4, 5, 6, 7/
call MPI_INIT(ierr)
call MPI_COMM_RANK(MPI_COMM_WORLD, rank, ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
if (numtasks .ne. NPROCS) then
print *, 'Must specify MPROCS= ',NPROCS,' Terminating.'
call MPI_FINALIZE(ierr)
stop
endif
sendbuf = rank
C Extract the original group handle
call MPI_COMM_GROUP(MPI_COMM_WORLD, orig_group, ierr)
C Divide tasks into two distinct groups based upon rank
if (rank .lt. NPROCS/2) then
call MPI_GROUP_INCL(orig_group, NPROCS/2, ranks1,
& new_group, ierr)
else
call MPI_GROUP_INCL(orig_group, NPROCS/2, ranks2,
& new_group, ierr)
endif
call MPI_COMM_CREATE(MPI_COMM_WORLD, new_group,
& new_comm, ierr)
call MPI_ALLREDUCE(sendbuf, recvbuf, 1, MPI_INTEGER,
& MPI_SUM, new_comm, ierr)
call MPI_GROUP_RANK(new_group, new_rank, ierr)
print *, 'rank= ',rank,' newrank= ',new_rank,' recvbuf= ',
& recvbuf
call MPI_FINALIZE(ierr)
end
|
Sample program output:
rank= 7 newrank= 3 recvbuf= 22 rank= 0 newrank= 0 recvbuf= 6 rank= 1 newrank= 1 recvbuf= 6 rank= 2 newrank= 2 recvbuf= 6 rank= 6 newrank= 2 recvbuf= 22 rank= 3 newrank= 3 recvbuf= 6 rank= 4 newrank= 0 recvbuf= 22 rank= 5 newrank= 1 recvbuf= 22
| Virtual Topologies |
What Are They?
- In terms of MPI, a virtual topology describes a mapping/ordering of
MPI processes into a geometric "shape".
- The two main types of topologies supported by MPI are Cartesian (grid)
and Graph.
- MPI topologies are virtual - there may be no relation between the
physical structure of the parallel machine and the process topology.
- Virtual topologies are built upon MPI communicators and groups.
- Must be "programmed" by the application developer.
Why Use Them?
- Convenience
- Virtual topologies may be useful for applications with specific communication patterns - patterns that match an MPI topology structure.
- For example, a Cartesian topology might prove convenient for an application that requires 4-way nearest neighbor communications for grid based data.
- Communication Efficiency
- Some hardware architectures may impose penalties for communications between successively distant "nodes".
- A particular implementation may optimize process mapping based upon the physical characteristics of a given parallel machine.
- The mapping of processes into an MPI virtual topology is dependent upon the MPI implementation, and may be totally ignored.
Example:
A simplified mapping of processes into a Cartesian virtual topology appears below:
Virtual Topology Routines
Create a 4 x 4 Cartesian topology from 16 processors and have each process exchange its rank with four neighbors.
C Language - Cartesian Virtual Topology
Example
#include "mpi.h"
#include <stdio.h>
#define SIZE 16
#define UP 0
#define DOWN 1
#define LEFT 2
#define RIGHT 3
int main(argc,argv)
int argc;
char *argv[]; {
int numtasks, rank, source, dest, outbuf, i, tag=1,
inbuf[4]={MPI_PROC_NULL,MPI_PROC_NULL,MPI_PROC_NULL,MPI_PROC_NULL,},
nbrs[4], dims[2]={4,4},
periods[2]={0,0}, reorder=0, coords[2];
MPI_Request reqs[8];
MPI_Status stats[8];
MPI_Comm cartcomm;
MPI_Init(&argc,&argv);
MPI_Comm_size(MPI_COMM_WORLD, &numtasks);
if (numtasks == SIZE) {
MPI_Cart_create(MPI_COMM_WORLD, 2, dims, periods, reorder, &cartcomm);
MPI_Comm_rank(cartcomm, &rank);
MPI_Cart_coords(cartcomm, rank, 2, coords);
MPI_Cart_shift(cartcomm, 0, 1, &nbrs[UP], &nbrs[DOWN]);
MPI_Cart_shift(cartcomm, 1, 1, &nbrs[LEFT], &nbrs[RIGHT]);
outbuf = rank;
for (i=0; i<4; i++) {
dest = nbrs[i];
source = nbrs[i];
MPI_Isend(&outbuf, 1, MPI_INT, dest, tag,
MPI_COMM_WORLD, &reqs[i]);
MPI_Irecv(&inbuf[i], 1, MPI_INT, source, tag,
MPI_COMM_WORLD, &reqs[i+4]);
}
MPI_Waitall(8, reqs, stats);
printf("rank= %d coords= %d %d neighbors(u,d,l,r)= %d %d %d %d\n",
rank,coords[0],coords[1],nbrs[UP],nbrs[DOWN],nbrs[LEFT],
nbrs[RIGHT]);
printf("rank= %d inbuf(u,d,l,r)= %d %d %d %d\n",
rank,inbuf[UP],inbuf[DOWN],inbuf[LEFT],inbuf[RIGHT]);
}
else
printf("Must specify %d processors. Terminating.\n",SIZE);
MPI_Finalize();
}
|
Fortran - Cartesian Virtual Topology
Example
program cartesian
include 'mpif.h'
integer SIZE, UP, DOWN, LEFT, RIGHT
parameter(SIZE=16)
parameter(UP=1)
parameter(DOWN=2)
parameter(LEFT=3)
parameter(RIGHT=4)
integer numtasks, rank, source, dest, outbuf, i, tag, ierr,
& inbuf(4), nbrs(4), dims(2), coords(2),
& stats(MPI_STATUS_SIZE, 8), reqs(8), cartcomm,
& periods(2), reorder
data inbuf /MPI_PROC_NULL,MPI_PROC_NULL,MPI_PROC_NULL,
& MPI_PROC_NULL/, dims /4,4/, tag /1/,
& periods /0,0/, reorder /0/
call MPI_INIT(ierr)
call MPI_COMM_SIZE(MPI_COMM_WORLD, numtasks, ierr)
if (numtasks .eq. SIZE) then
call MPI_CART_CREATE(MPI_COMM_WORLD, 2, dims, periods, reorder,
& cartcomm, ierr)
call MPI_COMM_RANK(cartcomm, rank, ierr)
call MPI_CART_COORDS(cartcomm, rank, 2, coords, ierr)
print *,'rank= ',rank,'coords= ',coords
call MPI_CART_SHIFT(cartcomm, 0, 1, nbrs(UP), nbrs(DOWN), ierr)
call MPI_CART_SHIFT(cartcomm, 1, 1, nbrs(LEFT), nbrs(RIGHT),
& ierr)
outbuf = rank
do i=1,4
dest = nbrs(i)
source = nbrs(i)
call MPI_ISEND(outbuf, 1, MPI_INTEGER, dest, tag,
& MPI_COMM_WORLD, reqs(i), ierr)
call MPI_IRECV(inbuf(i), 1, MPI_INTEGER, source, tag,
& MPI_COMM_WORLD, reqs(i+4), ierr)
enddo
call MPI_WAITALL(8, reqs, stats, ierr)
print *,'rank= ',rank,' coords= ',coords,
& ' neighbors(u,d,l,r)= ',nbrs
print *,'rank= ',rank,' ',
& ' inbuf(u,d,l,r)= ',inbuf
else
print *, 'Must specify',SIZE,' processors. Terminating.'
endif
call MPI_FINALIZE(ierr)
end
|
Sample program output: (partial)
rank= 0 coords= 0 0 neighbors(u,d,l,r)= -3 4 -3 1
rank= 0 inbuf(u,d,l,r)= -3 4 -3 1
rank= 1 coords= 0 1 neighbors(u,d,l,r)= -3 5 0 2
rank= 1 inbuf(u,d,l,r)= -3 5 0 2
rank= 2 coords= 0 2 neighbors(u,d,l,r)= -3 6 1 3
rank= 2 inbuf(u,d,l,r)= -3 6 1 3
. . . . .
rank= 14 coords= 3 2 neighbors(u,d,l,r)= 10 -3 13 15
rank= 14 inbuf(u,d,l,r)= 10 -3 13 15
rank= 15 coords= 3 3 neighbors(u,d,l,r)= 11 -3 14 -3
rank= 15 inbuf(u,d,l,r)= 11 -3 14 -3
| A Brief Word on MPI-2 |
History:
- Intentionally, the MPI specification did not address several
"difficult" issues. For reasons of expediency, these issues
were deferred to a second specification, called MPI-2.
- In March 1995, following the release of the initial MPI specification,
the MPI Forum began discussing enhancements to the MPI standard.
Following this:
- December 1995: Supercomputing '95 conference - Birds of a Feather
meeting to discuss proposed extensions to MPI.
- November 1996: Supercomputing '96 conference - MPI-2 draft
made available. Public comments solicited.
Meeting to discuss MPI-2 extensions.
- The draft presented at Supercomputing '96 shortly thereafter became the MPI-2 standard.
- December 1995: Supercomputing '95 conference - Birds of a Feather
meeting to discuss proposed extensions to MPI.
- Currently, most HPC vendors have only partial implementations of MPI-2
Key Areas of New Functionality:
- Dynamic Processes - extensions that remove the static process model
of MPI. Provides routines to create new processes.
- One-Sided Communications - provides routines for one directional
communications. Include shared memory operations (put/get) and
remote accumulate operations.
- Extended Collective Operations - allows for non-blocking collective
operations and application of collective operations to
inter-communicators
- External Interfaces - defines routines that allow developers to
layer on top of MPI, such as for debuggers and profilers.
- Additional Language Bindings - describes C++ bindings and discusses
Fortran-90 issues.
- Parallel I/O - describes MPI support for parallel I/O.
More Information on MPI-2:
- The Argonne National Lab MPI web pages have MPI-2 information. See the References section for links.
| LLNL Specific Information and Recommendations |
MPI Implementations:
- LC supports the following MPI implementations on its production systems:
Platform Implementations Comments IBM AIX IBM MPI threaded library Recommended IBM MPI signal library Not thread safe and not recommended. MPICH Not thread safe Intel Linux Quadrics MPI Recommended. Uses shared memory for on-node communications and message passing over the Quadrics switch for inter-node communications. Not thread-safe MPICH shared memory On-node communications. Not thread safe. Only for machines without a Quadrics switch. COMPAQ Tru64 Compaq MPI shared memory On-node communications. Use -pthread compile flag for thread-safety. MPICH shared memory On-node communications. Not thread safe MPICH P4 Inter-node communications. Not thread safe. Not recommeded due to lack of high speed interconnect between nodes (no switch). - Multiple versions for each implementation are installed on every machine.
- C, C++, Fortran77 and Fortran90 are supported
Compiling, Linking, Running:
- The table below provides a summary of MPI compile/run commands for
all LC production systems. Note that this is a summary only - additional
implementation specific details can be found by consulting the sources
listed in the References and More Information
section.
Platform Implementations Compile/Link
RunIBM AIX IBM MPI threaded library mpcc_r code.c mpxlc_r code.c mpguidec code.c mpCC_r code.C mpxlC_r code.C mpguidec++ code.c mpxlf_r code.f mpguidef77 code.f mpxlf90_r code.F mpguidef90 code.F mpxlf95_r code.f
Set required POE environment variables, then either:
a.out args
poe a.out argsOr, use POE flags instead of environment variables:
a.out args poe_args
poe a.out args poe_args
IBM MPI signal library setenv LLNL_COMPILE_SINGLE_THREADED TRUE mpcc code.c mpxlc code.c mpCC code.C mpxlC code.C mpxlf code.f mpxlf90 code.F mpxlf95 code.f
Set required POE environment variables, then either:
a.out args
poe a.out argsOr, use POE flags instead of environment variables:
a.out args poe_args
poe a.out args poe_args
MPICH setenv LLNL_COMPILE_SINGLE_THREADED TRUE mpicc code.c mpiCC code.C mpif77 code.f mpif90 code.F mpirun -nodes n -np p a.out args - OR - setenv MP_NODES n mpirun -np p a.out args
Intel Linux Quadrics MPI (clusters with a Quadrics switch only)
mpicc code.c mpiicc code.c mpipgcc code.c mpiCC code.C mpipgCC code.C mpif77 code.f mpiifc code.f mpipgf77 code.f mpif90 code.F mpipgf90 code.F srun -n n -p partition a.out args
MPICH shared memory (clusters without a Quadrics switch)
mpicc code.c mpiicc code.c mpipgcc code.c mpiCC code.C mpipgCC code.C mpif77 code.f mpiifc code.f mpipgf77 code.f mpif90 code.F mpipgf90 code.F mpirun -nodes n -np p a.out args
COMPAQ Tru64 Compaq MPI shared memory cc code.c -lmpi -pthread gcc code.c -lmpi -pthread cxx code.C -lmpi -pthread KCC code.C -lmpi -pthread g++ code.C -lmpi -pthread f77 code.f -lfmpi -lmpi -pthread f90 code.F -lfmpi -lmpi -pthread dmpirun -np p a.out args
MPICH shared memory mpicc code.c mpiCC code.C mpicxx code.C mpif77 code.f mpif90 code.F mpirun -np p a.out args
MPICH P4 mpicc_p4 code.c mpiCC_p4 code.C mpicxx_p4 code.C mpif77_p4 code.f mpif90_p4 code.F mpirun_p4 -machinefile user_machine_file -np p a.out args
Environment Variables:
- IBM POE Environment Variables:
- POE environment variables play a critical role in how an MPI application behaves on all ASC IBM systems.
- LLNL explicitly sets several of these variables on each ASC IBM machine. See the /etc/environment file on the ASC system you intend to use for specifics.
- Linux Systems Elan Environment Variables:
- Quadrics MPI also provides a number of Elan environment variables, some of which are automatically set by LC for performance reasons.
32-bit versus 64-bit:
- Since the retirement of the ASC Blue systems, all IBM SPs at LC
are 64-bit architectures.
- All IA32 Linux clusters are of course, 32-bit architectures.
Thunder is a 64-bit Itanium architecture.
- The Compaq clusters are 64-bit architectures.
- 64-bit architectures have 64-bit address space - 18 million
terabytes. 32-bit architectures remain at ~4 GB.
- IBM makes compilation 32-bit by default, but LC overrides this
so that compilations are 64-bit by default. But, it's still
a good idea to use the -q64 compile flag anyway.
- 32-bit objects and 64-bits objects do not mix. An application has
to be completely one or the other.
- There are a number of important issues to consider when porting to or developing 64-bit applications. Consult the IBM and Intel documentation (and the web) for details.
More information:
- See the References and More Information section for the topics that apply to your work. In particular, the LC tutorials and MPI related documents, and the relevant IBM and Intel documentation.
This completes the tutorial.
 |
Please complete the online evaluation form - unless you are doing the exercise, in which case please complete it at the end of the exercise. |
Where would you like to go now?
| References and More Information |
- Author: Blaise Barney, Livermore
Computing.
- MPI web pages at Argonne National Laboratory
http://www-unix.mcs.anl.gov/mpi - "Using MPI", Gropp, Lusk and Skjellum. MIT Press, 1994.
- Livermore Computing specific information:
- MPI at LLNL
http://www.llnl.gov/computing/mpi - IBM SP Systems Overview tutorial
http://www.llnl.gov/computing/tutorials/ibm_sp - Compaq Clusters Overview tutorial
http://www.llnl.gov/computing/tutorials/compaq_clusters - IA 32 Linux Clusters Overview tutorial
http://www.llnl.gov/computing/tutorials/linux_clusters - "MPI and Elan Library Environment Variables" document
www.llnl.gov/computing/mpi/elan.html
- MPI at LLNL
- IBM Parallel Environment Manuals
http://www-1.ibm.com/servers/eserver/pseries/library/sp_books - IBM Compiler Documentation:
Fortran: www-4.ibm.com/software/ad/fortran
C/C++: www-4.ibm.com/software/ad/caix - Intel Compilers
intel.com/software/products/compilers - "RS/6000 SP: Practical MPI Programming", Yukiya Aoyama and Jun Nakano,
RS/6000 Technical Support Center, IBM Japan. Available from IBM's
Redbooks server at
http://www.redbooks.ibm.com.
- "A User's Guide to MPI", Peter S. Pacheco. Department of Mathematics, University of San Francisco.
| Appendix A: MPI-1 Routine Index |
These man pages were derived from an IBM implementation of MPI and may differ from the man pages of other implementations.