MPI for Python

Author

Lisandro Dalcin

Contact

dalcinl@gmail.com

Date

Nov 25, 2021

Abstract

This document describes the MPI for Python package. MPI for Python provides Python bindings for the Message Passing Interface (MPI) standard, allowing Python applications to exploit multiple processors on workstations, clusters and supercomputers.

This package builds on the MPI specification and provides an object oriented interface resembling the MPI-2 C++ bindings. It supports point-to-point (sends, receives) and collective (broadcasts, scatters, gathers) communication of any picklable Python object, as well as efficient communication of Python objects exposing the Python buffer interface (e.g. NumPy arrays and builtin bytes/array/memoryview objects).

Introduction

Over the last years, high performance computing has become an affordable resource to many more researchers in the scientific community than ever before. The conjunction of quality open source software and commodity hardware strongly influenced the now widespread popularity of Beowulf class clusters and cluster of workstations.

Among many parallel computational models, message-passing has proven to be an effective one. This paradigm is specially suited for (but not limited to) distributed memory architectures and is used in today’s most demanding scientific and engineering application related to modeling, simulation, design, and signal processing. However, portable message-passing parallel programming used to be a nightmare in the past because of the many incompatible options developers were faced to. Fortunately, this situation definitely changed after the MPI Forum released its standard specification.

High performance computing is traditionally associated with software development using compiled languages. However, in typical applications programs, only a small part of the code is time-critical enough to require the efficiency of compiled languages. The rest of the code is generally related to memory management, error handling, input/output, and user interaction, and those are usually the most error prone and time-consuming lines of code to write and debug in the whole development process. Interpreted high-level languages can be really advantageous for this kind of tasks.

For implementing general-purpose numerical computations, MATLAB 1 is the dominant interpreted programming language. In the open source side, Octave and Scilab are well known, freely distributed software packages providing compatibility with the MATLAB language. In this work, we present MPI for Python, a new package enabling applications to exploit multiple processors using standard MPI “look and feel” in Python scripts.

1

MATLAB is a registered trademark of The MathWorks, Inc.

What is MPI?

MPI, [mpi-using] [mpi-ref] the Message Passing Interface, is a standardized and portable message-passing system designed to function on a wide variety of parallel computers. The standard defines the syntax and semantics of library routines and allows users to write portable programs in the main scientific programming languages (Fortran, C, or C++).

Since its release, the MPI specification [mpi-std1] [mpi-std2] has become the leading standard for message-passing libraries for parallel computers. Implementations are available from vendors of high-performance computers and from well known open source projects like MPICH [mpi-mpich] and Open MPI [mpi-openmpi].

What is Python?

Python is a modern, easy to learn, powerful programming language. It has efficient high-level data structures and a simple but effective approach to object-oriented programming with dynamic typing and dynamic binding. It supports modules and packages, which encourages program modularity and code reuse. Python’s elegant syntax, together with its interpreted nature, make it an ideal language for scripting and rapid application development in many areas on most platforms.

The Python interpreter and the extensive standard library are available in source or binary form without charge for all major platforms, and can be freely distributed. It is easily extended with new functions and data types implemented in C or C++. Python is also suitable as an extension language for customizable applications.

Python is an ideal candidate for writing the higher-level parts of large-scale scientific applications [Hinsen97] and driving simulations in parallel architectures [Beazley97] like clusters of PC’s or SMP’s. Python codes are quickly developed, easily maintained, and can achieve a high degree of integration with other libraries written in compiled languages.

Overview

MPI for Python provides an object oriented approach to message passing which grounds on the standard MPI-2 C++ bindings. The interface was designed with focus in translating MPI syntax and semantics of standard MPI-2 bindings for C++ to Python. Any user of the standard C/C++ MPI bindings should be able to use this module without need of learning a new interface.

Communicating Python Objects and Array Data

The Python standard library supports different mechanisms for data persistence. Many of them rely on disk storage, but pickling and marshaling can also work with memory buffers.

The pickle modules provide user-extensible facilities to serialize general Python objects using ASCII or binary formats. The marshal module provides facilities to serialize built-in Python objects using a binary format specific to Python, but independent of machine architecture issues.

MPI for Python can communicate any built-in or user-defined Python object taking advantage of the features provided by the pickle module. These facilities will be routinely used to build binary representations of objects to communicate (at sending processes), and restoring them back (at receiving processes).

Although simple and general, the serialization approach (i.e., pickling and unpickling) previously discussed imposes important overheads in memory as well as processor usage, especially in the scenario of objects with large memory footprints being communicated. Pickling general Python objects, ranging from primitive or container built-in types to user-defined classes, necessarily requires computer resources. Processing is also needed for dispatching the appropriate serialization method (that depends on the type of the object) and doing the actual packing. Additional memory is always needed, and if its total amount is not known a priori, many reallocations can occur. Indeed, in the case of large numeric arrays, this is certainly unacceptable and precludes communication of objects occupying half or more of the available memory resources.

MPI for Python supports direct communication of any object exporting the single-segment buffer interface. This interface is a standard Python mechanism provided by some types (e.g., strings and numeric arrays), allowing access in the C side to a contiguous memory buffer (i.e., address and length) containing the relevant data. This feature, in conjunction with the capability of constructing user-defined MPI datatypes describing complicated memory layouts, enables the implementation of many algorithms involving multidimensional numeric arrays (e.g., image processing, fast Fourier transforms, finite difference schemes on structured Cartesian grids) directly in Python, with negligible overhead, and almost as fast as compiled Fortran, C, or C++ codes.

Communicators

In MPI for Python, Comm is the base class of communicators. The Intracomm and Intercomm classes are sublcasses of the Comm class. The Comm.Is_inter method (and Comm.Is_intra, provided for convenience but not part of the MPI specification) is defined for communicator objects and can be used to determine the particular communicator class.

The two predefined intracommunicator instances are available: COMM_SELF and COMM_WORLD. From them, new communicators can be created as needed.

The number of processes in a communicator and the calling process rank can be respectively obtained with methods Comm.Get_size and Comm.Get_rank. The associated process group can be retrieved from a communicator by calling the Comm.Get_group method, which returns an instance of the Group class. Set operations with Group objects like like Group.Union, Group.Intersection and Group.Difference are fully supported, as well as the creation of new communicators from these groups using Comm.Create and Comm.Create_group.

New communicator instances can be obtained with the Comm.Clone, Comm.Dup and Comm.Split methods, as well methods Intracomm.Create_intercomm and Intercomm.Merge.

Virtual topologies (Cartcomm, Graphcomm and Distgraphcomm classes, which are specializations of the Intracomm class) are fully supported. New instances can be obtained from intracommunicator instances with factory methods Intracomm.Create_cart and Intracomm.Create_graph.

Point-to-Point Communications

Point to point communication is a fundamental capability of message passing systems. This mechanism enables the transmission of data between a pair of processes, one side sending, the other receiving.

MPI provides a set of send and receive functions allowing the communication of typed data with an associated tag. The type information enables the conversion of data representation from one architecture to another in the case of heterogeneous computing environments; additionally, it allows the representation of non-contiguous data layouts and user-defined datatypes, thus avoiding the overhead of (otherwise unavoidable) packing/unpacking operations. The tag information allows selectivity of messages at the receiving end.

Blocking Communications

MPI provides basic send and receive functions that are blocking. These functions block the caller until the data buffers involved in the communication can be safely reused by the application program.

In MPI for Python, the Comm.Send, Comm.Recv and Comm.Sendrecv methods of communicator objects provide support for blocking point-to-point communications within Intracomm and Intercomm instances. These methods can communicate memory buffers. The variants Comm.send, Comm.recv and Comm.sendrecv can communicate general Python objects.

Nonblocking Communications

On many systems, performance can be significantly increased by overlapping communication and computation. This is particularly true on systems where communication can be executed autonomously by an intelligent, dedicated communication controller.

MPI provides nonblocking send and receive functions. They allow the possible overlap of communication and computation. Non-blocking communication always come in two parts: posting functions, which begin the requested operation; and test-for-completion functions, which allow to discover whether the requested operation has completed.

In MPI for Python, the Comm.Isend and Comm.Irecv methods initiate send and receive operations, respectively. These methods return a Request instance, uniquely identifying the started operation. Its completion can be managed using the Request.Test, Request.Wait and Request.Cancel methods. The management of Request objects and associated memory buffers involved in communication requires a careful, rather low-level coordination. Users must ensure that objects exposing their memory buffers are not accessed at the Python level while they are involved in nonblocking message-passing operations.

Persistent Communications

Often a communication with the same argument list is repeatedly executed within an inner loop. In such cases, communication can be further optimized by using persistent communication, a particular case of nonblocking communication allowing the reduction of the overhead between processes and communication controllers. Furthermore , this kind of optimization can also alleviate the extra call overheads associated to interpreted, dynamic languages like Python.

In MPI for Python, the Comm.Send_init and Comm.Recv_init methods create persistent requests for a send and receive operation, respectively. These methods return an instance of the Prequest class, a subclass of the Request class. The actual communication can be effectively started using the Prequest.Start method, and its completion can be managed as previously described.

Collective Communications

Collective communications allow the transmittal of data between multiple processes of a group simultaneously. The syntax and semantics of collective functions is consistent with point-to-point communication. Collective functions communicate typed data, but messages are not paired with an associated tag; selectivity of messages is implied in the calling order. Additionally, collective functions come in blocking versions only.

The more commonly used collective communication operations are the following.

  • Barrier synchronization across all group members.

  • Global communication functions

    • Broadcast data from one member to all members of a group.

    • Gather data from all members to one member of a group.

    • Scatter data from one member to all members of a group.

  • Global reduction operations such as sum, maximum, minimum, etc.

In MPI for Python, the Comm.Bcast, Comm.Scatter, Comm.Gather, Comm.Allgather, Comm.Alltoall methods provide support for collective communications of memory buffers. The lower-case variants Comm.bcast, Comm.scatter, Comm.gather, Comm.allgather and Comm.alltoall can communicate general Python objects. The vector variants (which can communicate different amounts of data to each process) Comm.Scatterv, Comm.Gatherv, Comm.Allgatherv, Comm.Alltoallv and Comm.Alltoallw are also supported, they can only communicate objects exposing memory buffers.

Global reducion operations on memory buffers are accessible through the Comm.Reduce, Comm.Reduce_scatter, Comm.Allreduce, Intracomm.Scan and Intracomm.Exscan methods. The lower-case variants Comm.reduce, Comm.allreduce, Intracomm.scan and Intracomm.exscan can communicate general Python objects; however, the actual required reduction computations are performed sequentially at some process. All the predefined (i.e., SUM, PROD, MAX, etc.) reduction operations can be applied.

Support for GPU-aware MPI

Several MPI implementations, including Open MPI and MVAPICH, support passing GPU pointers to MPI calls to avoid explict data movement between the host and the device. On the Python side, GPU arrays have been implemented by many libraries that need GPU computation, such as CuPy, Numba, PyTorch, and PyArrow. In order to increase library interoperability, two kinds of zero-copy data exchange protocols are defined and agreed upon: DLPack and CUDA Array Interface. For example, a CuPy array can be passed to a Numba CUDA-jit kernel.

MPI for Python provides an experimental support for GPU-aware MPI. This feature requires:

  1. mpi4py is built against a GPU-aware MPI library.

  2. The Python GPU arrays are compliant with either of the protocols.

See the Tutorial section for further information. We note that

  • Whether or not a MPI call can work for GPU arrays depends on the underlying MPI implementation, not on mpi4py.

  • This support is currently experimental and subject to change in the future.

Dynamic Process Management

In the context of the MPI-1 specification, a parallel application is static; that is, no processes can be added to or deleted from a running application after it has been started. Fortunately, this limitation was addressed in MPI-2. The new specification added a process management model providing a basic interface between an application and external resources and process managers.

This MPI-2 extension can be really useful, especially for sequential applications built on top of parallel modules, or parallel applications with a client/server model. The MPI-2 process model provides a mechanism to create new processes and establish communication between them and the existing MPI application. It also provides mechanisms to establish communication between two existing MPI applications, even when one did not start the other.

In MPI for Python, new independent process groups can be created by calling the Intracomm.Spawn method within an intracommunicator. This call returns a new intercommunicator (i.e., an Intercomm instance) at the parent process group. The child process group can retrieve the matching intercommunicator by calling the Comm.Get_parent class method. At each side, the new intercommunicator can be used to perform point to point and collective communications between the parent and child groups of processes.

Alternatively, disjoint groups of processes can establish communication using a client/server approach. Any server application must first call the Open_port function to open a port and the Publish_name function to publish a provided service, and next call the Intracomm.Accept method. Any client applications can first find a published service by calling the Lookup_name function, which returns the port where a server can be contacted; and next call the Intracomm.Connect method. Both Intracomm.Accept and Intracomm.Connect methods return an Intercomm instance. When connection between client/server processes is no longer needed, all of them must cooperatively call the Comm.Disconnect method. Additionally, server applications should release resources by calling the Unpublish_name and Close_port functions.

One-Sided Communications

One-sided communications (also called Remote Memory Access, RMA) supplements the traditional two-sided, send/receive based MPI communication model with a one-sided, put/get based interface. One-sided communication that can take advantage of the capabilities of highly specialized network hardware. Additionally, this extension lowers latency and software overhead in applications written using a shared-memory-like paradigm.

The MPI specification revolves around the use of objects called windows; they intuitively specify regions of a process’s memory that have been made available for remote read and write operations. The published memory blocks can be accessed through three functions for put (remote send), get (remote write), and accumulate (remote update or reduction) data items. A much larger number of functions support different synchronization styles; the semantics of these synchronization operations are fairly complex.

In MPI for Python, one-sided operations are available by using instances of the Win class. New window objects are created by calling the Win.Create method at all processes within a communicator and specifying a memory buffer . When a window instance is no longer needed, the Win.Free method should be called.

The three one-sided MPI operations for remote write, read and reduction are available through calling the methods Win.Put, Win.Get, and Win.Accumulate respectively within a Win instance. These methods need an integer rank identifying the target process and an integer offset relative the base address of the remote memory block being accessed.

The one-sided operations read, write, and reduction are implicitly nonblocking, and must be synchronized by using two primary modes. Active target synchronization requires the origin process to call the Win.Start and Win.Complete methods at the origin process, and target process cooperates by calling the Win.Post and Win.Wait methods. There is also a collective variant provided by the Win.Fence method. Passive target synchronization is more lenient, only the origin process calls the Win.Lock and Win.Unlock methods. Locks are used to protect remote accesses to the locked remote window and to protect local load/store accesses to a locked local window.

Parallel Input/Output

The POSIX standard provides a model of a widely portable file system. However, the optimization needed for parallel input/output cannot be achieved with this generic interface. In order to ensure efficiency and scalability, the underlying parallel input/output system must provide a high-level interface supporting partitioning of file data among processes and a collective interface supporting complete transfers of global data structures between process memories and files. Additionally, further efficiencies can be gained via support for asynchronous input/output, strided accesses to data, and control over physical file layout on storage devices. This scenario motivated the inclusion in the MPI-2 standard of a custom interface in order to support more elaborated parallel input/output operations.

The MPI specification for parallel input/output revolves around the use objects called files. As defined by MPI, files are not just contiguous byte streams. Instead, they are regarded as ordered collections of typed data items. MPI supports sequential or random access to any integral set of these items. Furthermore, files are opened collectively by a group of processes.

The common patterns for accessing a shared file (broadcast, scatter, gather, reduction) is expressed by using user-defined datatypes. Compared to the communication patterns of point-to-point and collective communications, this approach has the advantage of added flexibility and expressiveness. Data access operations (read and write) are defined for different kinds of positioning (using explicit offsets, individual file pointers, and shared file pointers), coordination (non-collective and collective), and synchronism (blocking, nonblocking, and split collective with begin/end phases).

In MPI for Python, all MPI input/output operations are performed through instances of the File class. File handles are obtained by calling the File.Open method at all processes within a communicator and providing a file name and the intended access mode. After use, they must be closed by calling the File.Close method. Files even can be deleted by calling method File.Delete.

After creation, files are typically associated with a per-process view. The view defines the current set of data visible and accessible from an open file as an ordered set of elementary datatypes. This data layout can be set and queried with the File.Set_view and File.Get_view methods respectively.

Actual input/output operations are achieved by many methods combining read and write calls with different behavior regarding positioning, coordination, and synchronism. Summing up, MPI for Python provides the thirty (30) methods defined in MPI-2 for reading from or writing to files using explicit offsets or file pointers (individual or shared), in blocking or nonblocking and collective or noncollective versions.

Environmental Management

Initialization and Exit

Module functions Init or Init_thread and Finalize provide MPI initialization and finalization respectively. Module functions Is_initialized and Is_finalized provide the respective tests for initialization and finalization.

Note

MPI_Init() or MPI_Init_thread() is actually called when you import the MPI module from the mpi4py package, but only if MPI is not already initialized. In such case, calling Init or Init_thread from Python is expected to generate an MPI error, and in turn an exception will be raised.

Note

MPI_Finalize() is registered (by using Python C/API function Py_AtExit()) for being automatically called when Python processes exit, but only if mpi4py actually initialized MPI. Therefore, there is no need to call Finalize from Python to ensure MPI finalization.

Implementation Information

  • The MPI version number can be retrieved from module function Get_version. It returns a two-integer tuple (version, subversion).

  • The Get_processor_name function can be used to access the processor name.

  • The values of predefined attributes attached to the world communicator can be obtained by calling the Comm.Get_attr method within the COMM_WORLD instance.

Timers

MPI timer functionalities are available through the Wtime and Wtick functions.

Error Handling

In order facilitate handle sharing with other Python modules interfacing MPI-based parallel libraries, the predefined MPI error handlers ERRORS_RETURN and ERRORS_ARE_FATAL can be assigned to and retrieved from communicators using methods Comm.Set_errhandler and Comm.Get_errhandler, and similarly for windows and files.

When the predefined error handler ERRORS_RETURN is set, errors returned from MPI calls within Python code will raise an instance of the exception class Exception, which is a subclass of the standard Python exception RuntimeError.

Note

After import, mpi4py overrides the default MPI rules governing inheritance of error handlers. The ERRORS_RETURN error handler is set in the predefined COMM_SELF and COMM_WORLD communicators, as well as any new Comm, Win, or File instance created through mpi4py. If you ever pass such handles to C/C++/Fortran library code, it is recommended to set the ERRORS_ARE_FATAL error handler on them to ensure MPI errors do not pass silently.

Warning

Importing with from mpi4py.MPI import * will cause a name clashing with the standard Python Exception base class.

Tutorial

Warning

Under construction. Contributions very welcome!

Tip

Rolf Rabenseifner at HLRS developed a comprehensive MPI-3.1/4.0 course with slides and a large set of exercises including solutions. This material is available online for self-study. The slides and exercises show the C, Fortran, and Python (mpi4py) interfaces. For performance reasons, most Python exercises use NumPy arrays and communication routines involving buffer-like objects.

Tip

Victor Eijkhout at TACC authored the book Parallel Programming for Science and Engineering. This book is available online in PDF and HTML formats. The book covers parallel programming with MPI and OpenMP in C/C++ and Fortran, and MPI in Python using mpi4py.

MPI for Python supports convenient, pickle-based communication of generic Python object as well as fast, near C-speed, direct array data communication of buffer-provider objects (e.g., NumPy arrays).

  • Communication of generic Python objects

    You have to use methods with all-lowercase names, like Comm.send, Comm.recv, Comm.bcast, Comm.scatter, Comm.gather . An object to be sent is passed as a parameter to the communication call, and the received object is simply the return value.

    The Comm.isend and Comm.irecv methods return Request instances; completion of these methods can be managed using the Request.test and Request.wait methods.

    The Comm.recv and Comm.irecv methods may be passed a buffer object that can be repeatedly used to receive messages avoiding internal memory allocation. This buffer must be sufficiently large to accommodate the transmitted messages; hence, any buffer passed to Comm.recv or Comm.irecv must be at least as long as the pickled data transmitted to the receiver.

    Collective calls like Comm.scatter, Comm.gather, Comm.allgather, Comm.alltoall expect a single value or a sequence of Comm.size elements at the root or all process. They return a single value, a list of Comm.size elements, or None.

    Note

    MPI for Python uses the highest protocol version available in the Python runtime (see the HIGHEST_PROTOCOL constant in the pickle module). The default protocol can be changed at import time by setting the MPI4PY_PICKLE_PROTOCOL environment variable, or at runtime by assigning a different value to the PROTOCOL attribute of the pickle object within the MPI module.

  • Communication of buffer-like objects

    You have to use method names starting with an upper-case letter, like Comm.Send, Comm.Recv, Comm.Bcast, Comm.Scatter, Comm.Gather.

    In general, buffer arguments to these calls must be explicitly specified by using a 2/3-list/tuple like [data, MPI.DOUBLE], or [data, count, MPI.DOUBLE] (the former one uses the byte-size of data and the extent of the MPI datatype to define count).

    For vector collectives communication operations like Comm.Scatterv and Comm.Gatherv, buffer arguments are specified as [data, count, displ, datatype], where count and displ are sequences of integral values.

    Automatic MPI datatype discovery for NumPy/GPU arrays and PEP-3118 buffers is supported, but limited to basic C types (all C/C99-native signed/unsigned integral types and single/double precision real/complex floating types) and availability of matching datatypes in the underlying MPI implementation. In this case, the buffer-provider object can be passed directly as a buffer argument, the count and MPI datatype will be inferred.

    If mpi4py is built against a GPU-aware MPI implementation, GPU arrays can be passed to upper-case methods as long as they have either the __dlpack__ and __dlpack_device__ methods or the __cuda_array_interface__ attribute that are compliant with the respective standard specifications. Moreover, only C-contiguous or Fortran-contiguous GPU arrays are supported. It is important to note that GPU buffers must be fully ready before any MPI routines operate on them to avoid race conditions. This can be ensured by using the synchronization API of your array library. mpi4py does not have access to any GPU-specific functionality and thus cannot perform this operation automatically for users.

Running Python scripts with MPI

Most MPI programs can be run with the command mpiexec. In practice, running Python programs looks like:

$ mpiexec -n 4 python script.py

to run the program with 4 processors.

Point-to-Point Communication

  • Python objects (pickle under the hood):

    from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

if rank == 0:
    data = {'a': 7, 'b': 3.14}
    comm.send(data, dest=1, tag=11)
elif rank == 1:
    data = comm.recv(source=0, tag=11)

  • Python objects with non-blocking communication:

    from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

if rank == 0:
    data = {'a': 7, 'b': 3.14}
    req = comm.isend(data, dest=1, tag=11)
    req.wait()
elif rank == 1:
    req = comm.irecv(source=0, tag=11)
    data = req.wait()

  • NumPy arrays (the fast way!):

    from mpi4py import MPI
import numpy

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

# passing MPI datatypes explicitly
if rank == 0:
    data = numpy.arange(1000, dtype='i')
    comm.Send([data, MPI.INT], dest=1, tag=77)
elif rank == 1:
    data = numpy.empty(1000, dtype='i')
    comm.Recv([data, MPI.INT], source=0, tag=77)

# automatic MPI datatype discovery
if rank == 0:
    data = numpy.arange(100, dtype=numpy.float64)
    comm.Send(data, dest=1, tag=13)
elif rank == 1:
    data = numpy.empty(100, dtype=numpy.float64)
    comm.Recv(data, source=0, tag=13)

Collective Communication

  • Broadcasting a Python dictionary:

    from mpi4py import MPI

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

if rank == 0:
    data = {'key1' : [7, 2.72, 2+3j],
            'key2' : ( 'abc', 'xyz')}
else:
    data = None
data = comm.bcast(data, root=0)

  • Scattering Python objects:

    from mpi4py import MPI

comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()

if rank == 0:
    data = [(i+1)**2 for i in range(size)]
else:
    data = None
data = comm.scatter(data, root=0)
assert data == (rank+1)**2

  • Gathering Python objects:

    from mpi4py import MPI

comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()

data = (rank+1)**2
data = comm.gather(data, root=0)
if rank == 0:
    for i in range(size):
        assert data[i] == (i+1)**2
else:
    assert data is None

  • Broadcasting a NumPy array:

    from mpi4py import MPI
import numpy as np

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

if rank == 0:
    data = np.arange(100, dtype='i')
else:
    data = np.empty(100, dtype='i')
comm.Bcast(data, root=0)
for i in range(100):
    assert data[i] == i

  • Scattering NumPy arrays:

    from mpi4py import MPI
import numpy as np

comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()

sendbuf = None
if rank == 0:
    sendbuf = np.empty([size, 100], dtype='i')
    sendbuf.T[:,:] = range(size)
recvbuf = np.empty(100, dtype='i')
comm.Scatter(sendbuf, recvbuf, root=0)
assert np.allclose(recvbuf, rank)

  • Gathering NumPy arrays:

    from mpi4py import MPI
import numpy as np

comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()

sendbuf = np.zeros(100, dtype='i') + rank
recvbuf = None
if rank == 0:
    recvbuf = np.empty([size, 100], dtype='i')
comm.Gather(sendbuf, recvbuf, root=0)
if rank == 0:
    for i in range(size):
        assert np.allclose(recvbuf[i,:], i)

  • Parallel matrix-vector product:

    from mpi4py import MPI
import numpy

def matvec(comm, A, x):
    m = A.shape[0] # local rows
    p = comm.Get_size()
    xg = numpy.zeros(m*p, dtype='d')
    comm.Allgather([x,  MPI.DOUBLE],
                   [xg, MPI.DOUBLE])
    y = numpy.dot(A, xg)
    return y

MPI-IO

  • Collective I/O with NumPy arrays:

    from mpi4py import MPI
import numpy as np

amode = MPI.MODE_WRONLY|MPI.MODE_CREATE
comm = MPI.COMM_WORLD
fh = MPI.File.Open(comm, "./datafile.contig", amode)

buffer = np.empty(10, dtype=np.int)
buffer[:] = comm.Get_rank()

offset = comm.Get_rank()*buffer.nbytes
fh.Write_at_all(offset, buffer)

fh.Close()

  • Non-contiguous Collective I/O with NumPy arrays and datatypes:

    from mpi4py import MPI
import numpy as np

comm = MPI.COMM_WORLD
rank = comm.Get_rank()
size = comm.Get_size()

amode = MPI.MODE_WRONLY|MPI.MODE_CREATE
fh = MPI.File.Open(comm, "./datafile.noncontig", amode)

item_count = 10

buffer = np.empty(item_count, dtype='i')
buffer[:] = rank

filetype = MPI.INT.Create_vector(item_count, 1, size)
filetype.Commit()

displacement = MPI.INT.Get_size()*rank
fh.Set_view(displacement, filetype=filetype)

fh.Write_all(buffer)
filetype.Free()
fh.Close()

Dynamic Process Management

  • Compute Pi - Master (or parent, or client) side:

    #!/usr/bin/env python
from mpi4py import MPI
import numpy
import sys

comm = MPI.COMM_SELF.Spawn(sys.executable,
                           args=['cpi.py'],
                           maxprocs=5)

N = numpy.array(100, 'i')
comm.Bcast([N, MPI.INT], root=MPI.ROOT)
PI = numpy.array(0.0, 'd')
comm.Reduce(None, [PI, MPI.DOUBLE],
            op=MPI.SUM, root=MPI.ROOT)
print(PI)

comm.Disconnect()

  • Compute Pi - Worker (or child, or server) side:

    #!/usr/bin/env python
from mpi4py import MPI
import numpy

comm = MPI.Comm.Get_parent()
size = comm.Get_size()
rank = comm.Get_rank()

N = numpy.array(0, dtype='i')
comm.Bcast([N, MPI.INT], root=0)
h = 1.0 / N; s = 0.0
for i in range(rank, N, size):
    x = h * (i + 0.5)
    s += 4.0 / (1.0 + x**2)
PI = numpy.array(s * h, dtype='d')
comm.Reduce([PI, MPI.DOUBLE], None,
            op=MPI.SUM, root=0)

comm.Disconnect()

CUDA-aware MPI + Python GPU arrays

  • Reduce-to-all CuPy arrays:

    from mpi4py import MPI
import cupy as cp

comm = MPI.COMM_WORLD
size = comm.Get_size()
rank = comm.Get_rank()

sendbuf = cp.arange(10, dtype='i')
recvbuf = cp.empty_like(sendbuf)
assert hasattr(sendbuf, '__cuda_array_interface__')
assert hasattr(recvbuf, '__cuda_array_interface__')
cp.cuda.get_current_stream().synchronize()
comm.Allreduce(sendbuf, recvbuf)

assert cp.allclose(recvbuf, sendbuf*size)

One-Sided Communications

  • Read from (write to) the entire RMA window:

    import numpy as np
from mpi4py import MPI
from mpi4py.util import dtlib

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

datatype = MPI.FLOAT
np_dtype = dtlib.to_numpy_dtype(datatype)
itemsize = datatype.Get_size()

N = 10
win_size = N * itemsize if rank == 0 else 0
win = MPI.Win.Allocate(win_size, comm=comm)

buf = np.empty(N, dtype=np_dtype)
if rank == 0:
    buf.fill(42)
    win.Lock(rank=0)
    win.Put(buf, target_rank=0)
    win.Unlock(rank=0)
    comm.Barrier()
else:
    comm.Barrier()
    win.Lock(rank=0)
    win.Get(buf, target_rank=0)
    win.Unlock(rank=0)
    assert np.all(buf == 42)

  • Accessing a part of the RMA window using the target argument, which is defined as (offset, count, datatype):

    import numpy as np
from mpi4py import MPI
from mpi4py.util import dtlib

comm = MPI.COMM_WORLD
rank = comm.Get_rank()

datatype = MPI.FLOAT
np_dtype = dtlib.to_numpy_dtype(datatype)
itemsize = datatype.Get_size()

N = comm.Get_size() + 1
win_size = N * itemsize if rank == 0 else 0
win = MPI.Win.Allocate(
    size=win_size,
    disp_unit=itemsize,
    comm=comm,
)
if rank == 0:
    mem = np.frombuffer(win, dtype=np_dtype)
    mem[:] = np.arange(len(mem), dtype=np_dtype)
comm.Barrier()

buf = np.zeros(3, dtype=np_dtype)
target = (rank, 2, datatype)
win.Lock(rank=0)
win.Get(buf, target_rank=0, target=target)
win.Unlock(rank=0)
assert np.all(buf == [rank, rank+1, 0])

Wrapping with SWIG

  • C source:

    /* file: helloworld.c */
void sayhello(MPI_Comm comm)
{
  int size, rank;
  MPI_Comm_size(comm, &size);
  MPI_Comm_rank(comm, &rank);
  printf("Hello, World! "
         "I am process %d of %d.\n",
         rank, size);
}

  • SWIG interface file:

    // file: helloworld.i
%module helloworld
%{
#include <mpi.h>
#include "helloworld.c"
}%

%include mpi4py/mpi4py.i
%mpi4py_typemap(Comm, MPI_Comm);
void sayhello(MPI_Comm comm);

  • Try it in the Python prompt:

    >>> from mpi4py import MPI
>>> import helloworld
>>> helloworld.sayhello(MPI.COMM_WORLD)
Hello, World! I am process 0 of 1.

Wrapping with F2Py

  • Fortran 90 source:

    ! file: helloworld.f90
subroutine sayhello(comm)
  use mpi
  implicit none
  integer :: comm, rank, size, ierr
  call MPI_Comm_size(comm, size, ierr)
  call MPI_Comm_rank(comm, rank, ierr)
  print *, 'Hello, World! I am process ',rank,' of ',size,'.'
end subroutine sayhello

  • Compiling example using f2py

    $ f2py -c --f90exec=mpif90 helloworld.f90 -m helloworld

  • Try it in the Python prompt:

    >>> from mpi4py import MPI
>>> import helloworld
>>> fcomm = MPI.COMM_WORLD.py2f()
>>> helloworld.sayhello(fcomm)
Hello, World! I am process 0 of 1.

mpi4py

This is the MPI for Python package.

The Message Passing Interface (MPI) is a standardized and portable message-passing system designed to function on a wide variety of parallel computers. The MPI standard defines the syntax and semantics of library routines and allows users to write portable programs in the main scientific programming languages (Fortran, C, or C++). Since its release, the MPI specification has become the leading standard for message-passing libraries for parallel computers.

MPI for Python provides MPI bindings for the Python programming language, allowing any Python program to exploit multiple processors. This package build on the MPI specification and provides an object oriented interface which closely follows MPI-2 C++ bindings.

Runtime configuration options

mpi4py.rc

This object has attributes exposing runtime configuration options that become effective at import time of the MPI module.

Attributes Summary

initialize

Automatic MPI initialization at import

threads

Request initialization with thread support

thread_level

Level of thread support to request

finalize

Automatic MPI finalization at exit

fast_reduce

Use tree-based reductions for objects

recv_mprobe

Use matched probes to receive objects

errors

Error handling policy

Attributes Documentation

mpi4py.rc.initialize

Automatic MPI initialization at import.

Type

bool

Default

True

See also

MPI4PY_RC_INITIALIZE

mpi4py.rc.threads

Request initialization with thread support.

Type

bool

Default

True

See also

MPI4PY_RC_THREADS

mpi4py.rc.thread_level

Level of thread support to request.

Type

str

Default

"multiple"

Choices

"multiple", "serialized", "funneled", "single"

See also

MPI4PY_RC_THREAD_LEVEL

mpi4py.rc.finalize

Automatic MPI finalization at exit.

Type

None or bool

Default

None

See also

MPI4PY_RC_FINALIZE

mpi4py.rc.fast_reduce

Use tree-based reductions for objects.

Type

bool

Default

True

See also

MPI4PY_RC_FAST_REDUCE

mpi4py.rc.recv_mprobe

Use matched probes to receive objects.

Type

bool

Default

True

See also

MPI4PY_RC_RECV_MPROBE

mpi4py.rc.errors

Error handling policy.

Type

str

Default

"exception"

Choices

"exception", "default", "fatal"

See also

MPI4PY_RC_ERRORS

Example

MPI for Python features automatic initialization and finalization of the MPI execution environment. By using the mpi4py.rc object, MPI initialization and finalization can be handled programatically:

import mpi4py
mpi4py.rc.initialize = False  # do not initialize MPI automatically
mpi4py.rc.finalize = False    # do not finalize MPI automatically

from mpi4py import MPI # import the 'MPI' module

MPI.Init()      # manual initialization of the MPI environment
...             # your finest code here ...
MPI.Finalize()  # manual finalization of the MPI environment

Environment variables

The following environment variables override the corresponding attributes of the mpi4py.rc and MPI.pickle objects at import time of the MPI module.

Note

For variables of boolean type, accepted values are 0 and 1 (interpreted as False and True, respectively), and strings specifying a YAML boolean value (case-insensitive).

MPI4PY_RC_INITIALIZE
Type

bool

Default

True

Whether to automatically initialize MPI at import time of the mpi4py.MPI module.

See also

mpi4py.rc.initialize

New in version 3.1.0.

MPI4PY_RC_FINALIZE
Type

None | bool

Default

None

Choices

None, True, False

Whether to automatically finalize MPI at exit time of the Python process.

See also

mpi4py.rc.finalize

New in version 3.1.0.

MPI4PY_RC_THREADS
Type

bool

Default

True

Whether to initialize MPI with thread support.

See also

mpi4py.rc.threads

New in version 3.1.0.

MPI4PY_RC_THREAD_LEVEL
Default

"multiple"

Choices

"single", "funneled", "serialized", "multiple"

The level of required thread support.

See also

mpi4py.rc.thread_level

New in version 3.1.0.

MPI4PY_RC_FAST_REDUCE
Type

bool

Default

True

Whether to use tree-based reductions for objects.

See also

mpi4py.rc.fast_reduce

New in version 3.1.0.

MPI4PY_RC_RECV_MPROBE
Type

bool

Default

True

Whether to use matched probes to receive objects.

See also

mpi4py.rc.recv_mprobe

MPI4PY_RC_ERRORS
Default

"exception"

Choices

"exception", "default", "fatal"

Controls default MPI error handling policy.

See also

mpi4py.rc.errors

New in version 3.1.0.

MPI4PY_PICKLE_PROTOCOL
Type

int

Default

pickle.HIGHEST_PROTOCOL

Controls the default pickle protocol to use when communicating Python objects.

See also

PROTOCOL attribute of the MPI.pickle object within the MPI module.

New in version 3.1.0.

MPI4PY_PICKLE_THRESHOLD
Type

int

Default

262144

Controls the default buffer size threshold for switching from in-band to out-of-band buffer handling when using pickle protocol version 5 or higher.

See also

Module mpi4py.util.pkl5.

New in version 3.1.2.

Miscellaneous functions

mpi4py.profile(name, *, path=None, logfile=None)

Support for the MPI profiling interface.

Parameters

  • name (str) – Name of the profiler library to load.

  • path (sequence of str, optional) – Additional paths to search for the profiler.

  • logfile (str, optional) – Filename prefix for dumping profiler output.

Return type

None

mpi4py.get_config()

Return a dictionary with information about MPI.

Return type

Dict[str, str]

mpi4py.get_include()

Return the directory in the package that contains header files.

Extension modules that need to compile against mpi4py should use this function to locate the appropriate include directory. Using Python distutils (or perhaps NumPy distutils):

import mpi4py
Extension('extension_name', ...
          include_dirs=[..., mpi4py.get_include()])
Return type

str

mpi4py.MPI

Classes

Ancillary

Datatype([datatype])

Datatype object

Status([status])

Status object

Request([request])

Request handle

Prequest([request])

Persistent request handle

Grequest([request])

Generalized request handle

Op([op])

Operation object

Group([group])

Group of processes

Info([info])

Info object

Communication

Comm([comm])

Communicator

Intracomm([comm])

Intracommunicator

Topocomm([comm])

Topology intracommunicator

Cartcomm([comm])

Cartesian topology intracommunicator

Graphcomm([comm])

General graph topology intracommunicator

Distgraphcomm([comm])

Distributed graph topology intracommunicator

Intercomm([comm])

Intercommunicator

Message([message])

Matched message handle

One-sided operations

Win([win])

Window handle

Input/Output

File([file])

File handle

Error handling

Errhandler([errhandler])

Error handler

Exception([ierr])

Exception class

Auxiliary

Pickle([dumps, loads, protocol])

Pickle/unpickle Python objects

memory(buf)

Memory buffer

Functions

Version inquiry

Get_version()

Obtain the version number of the MPI standard supported by the implementation as a tuple (version, subversion)

Get_library_version()

Obtain the version string of the MPI library

Initialization and finalization

Init()

Initialize the MPI execution environment

Init_thread([required])

Initialize the MPI execution environment

Finalize()

Terminate the MPI execution environment

Is_initialized()

Indicates whether Init has been called

Is_finalized()

Indicates whether Finalize has completed

Query_thread()

Return the level of thread support provided by the MPI library

Is_thread_main()

Indicate whether this thread called Init or Init_thread

Memory allocation

Alloc_mem(size[, info])

Allocate memory for message passing and RMA

Free_mem(mem)

Free memory allocated with Alloc_mem()

Address manipulation

Get_address(location)

Get the address of a location in memory

Aint_add(base, disp)

Return the sum of base address and displacement

Aint_diff(addr1, addr2)

Return the difference between absolute addresses

Timer

Wtick()

Return the resolution of Wtime

Wtime()

Return an elapsed time on the calling processor

Error handling

Get_error_class(errorcode)

Convert an error code into an error class

Get_error_string(errorcode)

Return the error string for a given error class or error code

Add_error_class()

Add an error class to the known error classes

Add_error_code(errorclass)

Add an error code to an error class

Add_error_string(errorcode, string)

Associate an error string with an error class or errorcode

Dynamic process management

Open_port([info])

Return an address that can be used to establish connections between groups of MPI processes

Close_port(port_name)

Close a port

Publish_name(service_name, port_name[, info])

Publish a service name

Unpublish_name(service_name, port_name[, info])

Unpublish a service name

Lookup_name(service_name[, info])

Lookup a port name given a service name

Miscellanea

Attach_buffer(buf)

Attach a user-provided buffer for sending in buffered mode

Detach_buffer()

Remove an existing attached buffer

Compute_dims(nnodes, dims)

Return a balanced distribution of processes per coordinate direction

Get_processor_name()

Obtain the name of the calling processor

Register_datarep(datarep, read_fn, write_fn, ...)

Register user-defined data representations

Pcontrol(level)

Control profiling

Utilities

get_vendor()

Infomation about the underlying MPI implementation

Attributes

UNDEFINED

int UNDEFINED

ANY_SOURCE

int ANY_SOURCE

ANY_TAG

int ANY_TAG

PROC_NULL

int PROC_NULL

ROOT

int ROOT

BOTTOM

Bottom BOTTOM

IN_PLACE

InPlace IN_PLACE

KEYVAL_INVALID

int KEYVAL_INVALID

TAG_UB

int TAG_UB

HOST

int HOST

IO

int IO

WTIME_IS_GLOBAL

int WTIME_IS_GLOBAL

UNIVERSE_SIZE

int UNIVERSE_SIZE

APPNUM

int APPNUM

LASTUSEDCODE

int LASTUSEDCODE

WIN_BASE

int WIN_BASE

WIN_SIZE

int WIN_SIZE

WIN_DISP_UNIT

int WIN_DISP_UNIT

WIN_CREATE_FLAVOR

int WIN_CREATE_FLAVOR

WIN_FLAVOR

int WIN_FLAVOR

WIN_MODEL

int WIN_MODEL

SUCCESS

int SUCCESS

ERR_LASTCODE

int ERR_LASTCODE

ERR_COMM

int ERR_COMM

ERR_GROUP

int ERR_GROUP

ERR_TYPE

int ERR_TYPE

ERR_REQUEST

int ERR_REQUEST

ERR_OP

int ERR_OP

ERR_BUFFER

int ERR_BUFFER

ERR_COUNT

int ERR_COUNT

ERR_TAG

int ERR_TAG

ERR_RANK

int ERR_RANK

ERR_ROOT

int ERR_ROOT

ERR_TRUNCATE

int ERR_TRUNCATE

ERR_IN_STATUS

int ERR_IN_STATUS

ERR_PENDING

int ERR_PENDING

ERR_TOPOLOGY

int ERR_TOPOLOGY

ERR_DIMS

int ERR_DIMS

ERR_ARG

int ERR_ARG

ERR_OTHER

int ERR_OTHER

ERR_UNKNOWN

int ERR_UNKNOWN

ERR_INTERN

int ERR_INTERN

ERR_INFO

int ERR_INFO

ERR_FILE

int ERR_FILE

ERR_WIN

int ERR_WIN

ERR_KEYVAL

int ERR_KEYVAL

ERR_INFO_KEY

int ERR_INFO_KEY

ERR_INFO_VALUE

int ERR_INFO_VALUE

ERR_INFO_NOKEY

int ERR_INFO_NOKEY

ERR_ACCESS

int ERR_ACCESS

ERR_AMODE

int ERR_AMODE

ERR_BAD_FILE

int ERR_BAD_FILE

ERR_FILE_EXISTS

int ERR_FILE_EXISTS

ERR_FILE_IN_USE

int ERR_FILE_IN_USE

ERR_NO_SPACE

int ERR_NO_SPACE

ERR_NO_SUCH_FILE

int ERR_NO_SUCH_FILE

ERR_IO

int ERR_IO

ERR_READ_ONLY

int ERR_READ_ONLY

ERR_CONVERSION

int ERR_CONVERSION

ERR_DUP_DATAREP

int ERR_DUP_DATAREP

ERR_UNSUPPORTED_DATAREP

int ERR_UNSUPPORTED_DATAREP

ERR_UNSUPPORTED_OPERATION

int ERR_UNSUPPORTED_OPERATION

ERR_NAME

int ERR_NAME

ERR_NO_MEM

int ERR_NO_MEM

ERR_NOT_SAME

int ERR_NOT_SAME

ERR_PORT

int ERR_PORT

ERR_QUOTA

int ERR_QUOTA

ERR_SERVICE

int ERR_SERVICE

ERR_SPAWN

int ERR_SPAWN

ERR_BASE

int ERR_BASE

ERR_SIZE

int ERR_SIZE

ERR_DISP

int ERR_DISP

ERR_ASSERT

int ERR_ASSERT

ERR_LOCKTYPE

int ERR_LOCKTYPE

ERR_RMA_CONFLICT

int ERR_RMA_CONFLICT

ERR_RMA_SYNC

int ERR_RMA_SYNC

ERR_RMA_RANGE

int ERR_RMA_RANGE

ERR_RMA_ATTACH

int ERR_RMA_ATTACH

ERR_RMA_SHARED

int ERR_RMA_SHARED

ERR_RMA_FLAVOR

int ERR_RMA_FLAVOR

ORDER_C

int ORDER_C

ORDER_F

int ORDER_F

ORDER_FORTRAN

int ORDER_FORTRAN

TYPECLASS_INTEGER

int TYPECLASS_INTEGER

TYPECLASS_REAL

int TYPECLASS_REAL

TYPECLASS_COMPLEX

int TYPECLASS_COMPLEX

DISTRIBUTE_NONE

int DISTRIBUTE_NONE

DISTRIBUTE_BLOCK

int DISTRIBUTE_BLOCK

DISTRIBUTE_CYCLIC

int DISTRIBUTE_CYCLIC

DISTRIBUTE_DFLT_DARG

int DISTRIBUTE_DFLT_DARG

COMBINER_NAMED

int COMBINER_NAMED

COMBINER_DUP

int COMBINER_DUP

COMBINER_CONTIGUOUS

int COMBINER_CONTIGUOUS

COMBINER_VECTOR

int COMBINER_VECTOR

COMBINER_HVECTOR

int COMBINER_HVECTOR

COMBINER_INDEXED

int COMBINER_INDEXED

COMBINER_HINDEXED

int COMBINER_HINDEXED

COMBINER_INDEXED_BLOCK

int COMBINER_INDEXED_BLOCK

COMBINER_HINDEXED_BLOCK

int COMBINER_HINDEXED_BLOCK

COMBINER_STRUCT

int COMBINER_STRUCT

COMBINER_SUBARRAY

int COMBINER_SUBARRAY

COMBINER_DARRAY

int COMBINER_DARRAY

COMBINER_RESIZED

int COMBINER_RESIZED

COMBINER_F90_REAL

int COMBINER_F90_REAL

COMBINER_F90_COMPLEX

int COMBINER_F90_COMPLEX

COMBINER_F90_INTEGER

int COMBINER_F90_INTEGER

IDENT

int IDENT

CONGRUENT

int CONGRUENT

SIMILAR

int SIMILAR

UNEQUAL

int UNEQUAL

CART

int CART

GRAPH

int GRAPH

DIST_GRAPH

int DIST_GRAPH

UNWEIGHTED

int UNWEIGHTED

WEIGHTS_EMPTY

int WEIGHTS_EMPTY

COMM_TYPE_SHARED

int COMM_TYPE_SHARED

BSEND_OVERHEAD

int BSEND_OVERHEAD

WIN_FLAVOR_CREATE

int WIN_FLAVOR_CREATE

WIN_FLAVOR_ALLOCATE

int WIN_FLAVOR_ALLOCATE

WIN_FLAVOR_DYNAMIC

int WIN_FLAVOR_DYNAMIC

WIN_FLAVOR_SHARED

int WIN_FLAVOR_SHARED

WIN_SEPARATE

int WIN_SEPARATE

WIN_UNIFIED

int WIN_UNIFIED

MODE_NOCHECK

int MODE_NOCHECK

MODE_NOSTORE

int MODE_NOSTORE

MODE_NOPUT

int MODE_NOPUT

MODE_NOPRECEDE

int MODE_NOPRECEDE

MODE_NOSUCCEED

int MODE_NOSUCCEED

LOCK_EXCLUSIVE

int LOCK_EXCLUSIVE

LOCK_SHARED

int LOCK_SHARED

MODE_RDONLY

int MODE_RDONLY

MODE_WRONLY

int MODE_WRONLY

MODE_RDWR

int MODE_RDWR

MODE_CREATE

int MODE_CREATE

MODE_EXCL

int MODE_EXCL

MODE_DELETE_ON_CLOSE

int MODE_DELETE_ON_CLOSE

MODE_UNIQUE_OPEN

int MODE_UNIQUE_OPEN

MODE_SEQUENTIAL

int MODE_SEQUENTIAL

MODE_APPEND

int MODE_APPEND

SEEK_SET

int SEEK_SET

SEEK_CUR

int SEEK_CUR

SEEK_END

int SEEK_END

DISPLACEMENT_CURRENT

int DISPLACEMENT_CURRENT

DISP_CUR

int DISP_CUR

THREAD_SINGLE

int THREAD_SINGLE

THREAD_FUNNELED

int THREAD_FUNNELED

THREAD_SERIALIZED

int THREAD_SERIALIZED

THREAD_MULTIPLE

int THREAD_MULTIPLE

VERSION

int VERSION

SUBVERSION

int SUBVERSION

MAX_PROCESSOR_NAME

int MAX_PROCESSOR_NAME

MAX_ERROR_STRING

int MAX_ERROR_STRING

MAX_PORT_NAME

int MAX_PORT_NAME

MAX_INFO_KEY

int MAX_INFO_KEY

MAX_INFO_VAL

int MAX_INFO_VAL

MAX_OBJECT_NAME

int MAX_OBJECT_NAME

MAX_DATAREP_STRING

int MAX_DATAREP_STRING

MAX_LIBRARY_VERSION_STRING

int MAX_LIBRARY_VERSION_STRING

DATATYPE_NULL

Datatype DATATYPE_NULL

UB

Datatype UB

LB

Datatype LB

PACKED

Datatype PACKED

BYTE

Datatype BYTE

AINT

Datatype AINT

OFFSET

Datatype OFFSET

COUNT

Datatype COUNT

CHAR

Datatype CHAR

WCHAR

Datatype WCHAR

SIGNED_CHAR

Datatype SIGNED_CHAR

SHORT

Datatype SHORT

INT

Datatype INT

LONG

Datatype LONG

LONG_LONG

Datatype LONG_LONG

UNSIGNED_CHAR

Datatype UNSIGNED_CHAR

UNSIGNED_SHORT

Datatype UNSIGNED_SHORT

UNSIGNED

Datatype UNSIGNED

UNSIGNED_LONG

Datatype UNSIGNED_LONG

UNSIGNED_LONG_LONG

Datatype UNSIGNED_LONG_LONG

FLOAT

Datatype FLOAT

DOUBLE

Datatype DOUBLE

LONG_DOUBLE

Datatype LONG_DOUBLE

C_BOOL

Datatype C_BOOL

INT8_T

Datatype INT8_T

INT16_T

Datatype INT16_T

INT32_T

Datatype INT32_T

INT64_T

Datatype INT64_T

UINT8_T

Datatype UINT8_T

UINT16_T

Datatype UINT16_T

UINT32_T

Datatype UINT32_T

UINT64_T

Datatype UINT64_T

C_COMPLEX

Datatype C_COMPLEX

C_FLOAT_COMPLEX

Datatype C_FLOAT_COMPLEX

C_DOUBLE_COMPLEX

Datatype C_DOUBLE_COMPLEX

C_LONG_DOUBLE_COMPLEX

Datatype C_LONG_DOUBLE_COMPLEX

CXX_BOOL

Datatype CXX_BOOL

CXX_FLOAT_COMPLEX

Datatype CXX_FLOAT_COMPLEX

CXX_DOUBLE_COMPLEX

Datatype CXX_DOUBLE_COMPLEX

CXX_LONG_DOUBLE_COMPLEX

Datatype CXX_LONG_DOUBLE_COMPLEX

SHORT_INT

Datatype SHORT_INT

INT_INT

Datatype INT_INT

TWOINT

Datatype TWOINT

LONG_INT

Datatype LONG_INT

FLOAT_INT

Datatype FLOAT_INT

DOUBLE_INT

Datatype DOUBLE_INT

LONG_DOUBLE_INT

Datatype LONG_DOUBLE_INT

CHARACTER

Datatype CHARACTER

LOGICAL

Datatype LOGICAL

INTEGER

Datatype INTEGER

REAL

Datatype REAL

DOUBLE_PRECISION

Datatype DOUBLE_PRECISION

COMPLEX

Datatype COMPLEX

DOUBLE_COMPLEX

Datatype DOUBLE_COMPLEX

LOGICAL1

Datatype LOGICAL1

LOGICAL2

Datatype LOGICAL2

LOGICAL4

Datatype LOGICAL4

LOGICAL8

Datatype LOGICAL8

INTEGER1

Datatype INTEGER1

INTEGER2

Datatype INTEGER2

INTEGER4

Datatype INTEGER4

INTEGER8

Datatype INTEGER8

INTEGER16

Datatype INTEGER16

REAL2

Datatype REAL2

REAL4

Datatype REAL4

REAL8

Datatype REAL8

REAL16

Datatype REAL16

COMPLEX4

Datatype COMPLEX4

COMPLEX8

Datatype COMPLEX8

COMPLEX16

Datatype COMPLEX16

COMPLEX32

Datatype COMPLEX32

UNSIGNED_INT

Datatype UNSIGNED_INT

SIGNED_SHORT

Datatype SIGNED_SHORT

SIGNED_INT

Datatype SIGNED_INT

SIGNED_LONG

Datatype SIGNED_LONG

SIGNED_LONG_LONG

Datatype SIGNED_LONG_LONG

BOOL

Datatype BOOL

SINT8_T

Datatype SINT8_T

SINT16_T

Datatype SINT16_T

SINT32_T

Datatype SINT32_T

SINT64_T

Datatype SINT64_T

F_BOOL

Datatype F_BOOL

F_INT

Datatype F_INT

F_FLOAT

Datatype F_FLOAT

F_DOUBLE

Datatype F_DOUBLE

F_COMPLEX

Datatype F_COMPLEX

F_FLOAT_COMPLEX

Datatype F_FLOAT_COMPLEX

F_DOUBLE_COMPLEX

Datatype F_DOUBLE_COMPLEX

REQUEST_NULL

Request REQUEST_NULL

MESSAGE_NULL

Message MESSAGE_NULL

MESSAGE_NO_PROC

Message MESSAGE_NO_PROC

OP_NULL

Op OP_NULL

MAX

Op MAX

MIN

Op MIN

SUM

Op SUM

PROD

Op PROD

LAND

Op LAND

BAND

Op BAND

LOR

Op LOR

BOR

Op BOR

LXOR

Op LXOR

BXOR

Op BXOR

MAXLOC

Op MAXLOC

MINLOC

Op MINLOC

REPLACE

Op REPLACE

NO_OP

Op NO_OP

GROUP_NULL

Group GROUP_NULL

GROUP_EMPTY

Group GROUP_EMPTY

INFO_NULL

Info INFO_NULL

INFO_ENV

Info INFO_ENV

ERRHANDLER_NULL

Errhandler ERRHANDLER_NULL

ERRORS_RETURN

Errhandler ERRORS_RETURN

ERRORS_ARE_FATAL

Errhandler ERRORS_ARE_FATAL

COMM_NULL

Comm COMM_NULL

COMM_SELF

Intracomm COMM_SELF

COMM_WORLD

Intracomm COMM_WORLD

WIN_NULL

Win WIN_NULL

FILE_NULL

File FILE_NULL

pickle

Pickle pickle

mpi4py.futures

New in version 3.0.0.

This package provides a high-level interface for asynchronously executing callables on a pool of worker processes using MPI for inter-process communication.

concurrent.futures

The mpi4py.futures package is based on concurrent.futures from the Python standard library. More precisely, mpi4py.futures provides the MPIPoolExecutor class as a concrete implementation of the abstract class Executor. The submit() interface schedules a callable to be executed asynchronously and returns a Future object representing the execution of the callable. Future instances can be queried for the call result or exception. Sets of Future instances can be passed to the wait() and as_completed() functions.

Note

The concurrent.futures package was introduced in Python 3.2. A backport targeting Python 2.7 is available on PyPI. The mpi4py.futures package uses concurrent.futures if available, either from the Python 3 standard library or the Python 2.7 backport if installed. Otherwise, mpi4py.futures uses a bundled copy of core functionality backported from Python 3.5 to work with Python 2.7.

See also

Module concurrent.futures

Documentation of the concurrent.futures standard module.

MPIPoolExecutor

The MPIPoolExecutor class uses a pool of MPI processes to execute calls asynchronously. By performing computations in separate processes, it allows to side-step the global interpreter lock but also means that only picklable objects can be executed and returned. The __main__ module must be importable by worker processes, thus MPIPoolExecutor instances may not work in the interactive interpreter.

MPIPoolExecutor takes advantage of the dynamic process management features introduced in the MPI-2 standard. In particular, the MPI.Intracomm.Spawn method of MPI.COMM_SELF is used in the master (or parent) process to spawn new worker (or child) processes running a Python interpreter. The master process uses a separate thread (one for each MPIPoolExecutor instance) to communicate back and forth with the workers. The worker processes serve the execution of tasks in the main (and only) thread until they are signaled for completion.

Note

The worker processes must import the main script in order to unpickle any callable defined in the __main__ module and submitted from the master process. Furthermore, the callables may need access to other global variables. At the worker processes, mpi4py.futures executes the main script code (using the runpy module) under the __worker__ namespace to define the __main__ module. The __main__ and __worker__ modules are added to sys.modules (both at the master and worker processes) to ensure proper pickling and unpickling.

Warning

During the initial import phase at the workers, the main script cannot create and use new MPIPoolExecutor instances. Otherwise, each worker would attempt to spawn a new pool of workers, leading to infinite recursion. mpi4py.futures detects such recursive attempts to spawn new workers and aborts the MPI execution environment. As the main script code is run under the __worker__ namespace, the easiest way to avoid spawn recursion is using the idiom if __name__ == '__main__': ... in the main script.

class mpi4py.futures.MPIPoolExecutor(max_workers=None, initializer=None, initargs=(), **kwargs)

An Executor subclass that executes calls asynchronously using a pool of at most max_workers processes. If max_workers is None or not given, its value is determined from the MPI4PY_FUTURES_MAX_WORKERS environment variable if set, or the MPI universe size if set, otherwise a single worker process is spawned. If max_workers is lower than or equal to 0, then a ValueError will be raised.

initializer is an optional callable that is called at the start of each worker process before executing any tasks; initargs is a tuple of arguments passed to the initializer. If initializer raises an exception, all pending tasks and any attempt to submit new tasks to the pool will raise a BrokenExecutor exception.

Other parameters:

  • python_exe: Path to the Python interpreter executable used to spawn worker processes, otherwise sys.executable is used.

  • python_args: list or iterable with additional command line flags to pass to the Python executable. Command line flags determined from inspection of sys.flags, sys.warnoptions and sys._xoptions in are passed unconditionally.

  • mpi_info: dict or iterable yielding (key, value) pairs. These (key, value) pairs are passed (through an MPI.Info object) to the MPI.Intracomm.Spawn call used to spawn worker processes. This mechanism allows telling the MPI runtime system where and how to start the processes. Check the documentation of the backend MPI implementation about the set of keys it interprets and the corresponding format for values.

  • globals: dict or iterable yielding (name, value) pairs to initialize the main module namespace in worker processes.

  • main: If set to False, do not import the __main__ module in worker processes. Setting main to False prevents worker processes from accessing definitions in the parent __main__ namespace.

  • path: list or iterable with paths to append to sys.path in worker processes to extend the module search path.

  • wdir: Path to set the current working directory in worker processes using os.chdir(). The initial working directory is set by the MPI implementation. Quality MPI implementations should honor a wdir info key passed through mpi_info, although such feature is not mandatory.

  • env: dict or iterable yielding (name, value) pairs with environment variables to update os.environ in worker processes. The initial environment is set by the MPI implementation. MPI implementations may allow setting the initial environment through mpi_info, however such feature is not required nor recommended by the MPI standard.

submit(func, *args, **kwargs)

Schedule the callable, func, to be executed as func(*args, **kwargs) and returns a Future object representing the execution of the callable.

executor = MPIPoolExecutor(max_workers=1)
future = executor.submit(pow, 321, 1234)
print(future.result())
map(func, *iterables, timeout=None, chunksize=1, **kwargs)

Equivalent to map(func, *iterables) except func is executed asynchronously and several calls to func may be made concurrently, out-of-order, in separate processes. The returned iterator raises a TimeoutError if __next__() is called and the result isn’t available after timeout seconds from the original call to map(). timeout can be an int or a float. If timeout is not specified or None, there is no limit to the wait time. If a call raises an exception, then that exception will be raised when its value is retrieved from the iterator. This method chops iterables into a number of chunks which it submits to the pool as separate tasks. The (approximate) size of these chunks can be specified by setting chunksize to a positive integer. For very long iterables, using a large value for chunksize can significantly improve performance compared to the default size of one. By default, the returned iterator yields results in-order, waiting for successive tasks to complete . This behavior can be changed by passing the keyword argument unordered as True, then the result iterator will yield a result as soon as any of the tasks complete.

executor = MPIPoolExecutor(max_workers=3)
for result in executor.map(pow, [2]*32, range(32)):
    print(result)
starmap(func, iterable, timeout=None, chunksize=1, **kwargs)

Equivalent to itertools.starmap(func, iterable). Used instead of map() when argument parameters are already grouped in tuples from a single iterable (the data has been “pre-zipped”). map(func, *iterable) is equivalent to starmap(func, zip(*iterable)).

executor = MPIPoolExecutor(max_workers=3)
iterable = ((2, n) for n in range(32))
for result in executor.starmap(pow, iterable):
    print(result)
shutdown(wait=True, cancel_futures=False)

Signal the executor that it should free any resources that it is using when the currently pending futures are done executing. Calls to submit() and map() made after shutdown() will raise RuntimeError.

If wait is True then this method will not return until all the pending futures are done executing and the resources associated with the executor have been freed. If wait is False then this method will return immediately and the resources associated with the executor will be freed when all pending futures are done executing. Regardless of the value of wait, the entire Python program will not exit until all pending futures are done executing.

If cancel_futures is True, this method will cancel all pending futures that the executor has not started running. Any futures that are completed or running won’t be cancelled, regardless of the value of cancel_futures.

You can avoid having to call this method explicitly if you use the with statement, which will shutdown the executor instance (waiting as if shutdown() were called with wait set to True).

import time
with MPIPoolExecutor(max_workers=1) as executor:
    future = executor.submit(time.sleep, 2)
assert future.done()
bootup(wait=True)

Signal the executor that it should allocate eagerly any required resources (in particular, MPI worker processes). If wait is True, then bootup() will not return until the executor resources are ready to process submissions. Resources are automatically allocated in the first call to submit(), thus calling bootup() explicitly is seldom needed.

MPI4PY_FUTURES_MAX_WORKERS

If the max_workers parameter to MPIPoolExecutor is None or not given, the MPI4PY_FUTURES_MAX_WORKERS environment variable provides fallback value for the maximum number of MPI worker processes to spawn.

Note

As the master process uses a separate thread to perform MPI communication with the workers, the backend MPI implementation should provide support for MPI.THREAD_MULTIPLE. However, some popular MPI implementations do not support yet concurrent MPI calls from multiple threads. Additionally, users may decide to initialize MPI with a lower level of thread support. If the level of thread support in the backend MPI is less than MPI.THREAD_MULTIPLE, mpi4py.futures will use a global lock to serialize MPI calls. If the level of thread support is less than MPI.THREAD_SERIALIZED, mpi4py.futures will emit a RuntimeWarning.

Warning

If the level of thread support in the backend MPI is less than MPI.THREAD_SERIALIZED (i.e, it is either MPI.THREAD_SINGLE or MPI.THREAD_FUNNELED), in theory mpi4py.futures cannot be used. Rather than raising an exception, mpi4py.futures emits a warning and takes a “cross-fingers” attitude to continue execution in the hope that serializing MPI calls with a global lock will actually work.

MPICommExecutor

Legacy MPI-1 implementations (as well as some vendor MPI-2 implementations) do not support the dynamic process management features introduced in the MPI-2 standard. Additionally, job schedulers and batch systems in supercomputing facilities may pose additional complications to applications using the MPI_Comm_spawn() routine.

With these issues in mind, mpi4py.futures supports an additonal, more traditional, SPMD-like usage pattern requiring MPI-1 calls only. Python applications are started the usual way, e.g., using the mpiexec command. Python code should make a collective call to the MPICommExecutor context manager to partition the set of MPI processes within a MPI communicator in one master processes and many workers processes. The master process gets access to an MPIPoolExecutor instance to submit tasks. Meanwhile, the worker process follow a different execution path and team-up to execute the tasks submitted from the master.

Besides alleviating the lack of dynamic process managment features in legacy MPI-1 or partial MPI-2 implementations, the MPICommExecutor context manager may be useful in classic MPI-based Python applications willing to take advantage of the simple, task-based, master/worker approach available in the mpi4py.futures package.

class mpi4py.futures.MPICommExecutor(comm=None, root=0)

Context manager for MPIPoolExecutor. This context manager splits a MPI (intra)communicator comm (defaults to MPI.COMM_WORLD if not provided or None) in two disjoint sets: a single master process (with rank root in comm) and the remaining worker processes. These sets are then connected through an intercommunicator. The target of the with statement is assigned either an MPIPoolExecutor instance (at the master) or None (at the workers).

from mpi4py import MPI
from mpi4py.futures import MPICommExecutor

with MPICommExecutor(MPI.COMM_WORLD, root=0) as executor:
    if executor is not None:
       future = executor.submit(abs, -42)
       assert future.result() == 42
       answer = set(executor.map(abs, [-42, 42]))
       assert answer == {42}

Warning

If MPICommExecutor is passed a communicator of size one (e.g., MPI.COMM_SELF), then the executor instace assigned to the target of the with statement will execute all submitted tasks in a single worker thread, thus ensuring that task execution still progress asynchronously. However, the GIL will prevent the main and worker threads from running concurrently in multicore processors. Moreover, the thread context switching may harm noticeably the performance of CPU-bound tasks. In case of I/O-bound tasks, the GIL is not usually an issue, however, as a single worker thread is used, it progress one task at a time. We advice against using MPICommExecutor with communicators of size one and suggest refactoring your code to use instead a ThreadPoolExecutor.

Command line

Recalling the issues related to the lack of support for dynamic process managment features in MPI implementations, mpi4py.futures supports an alternative usage pattern where Python code (either from scripts, modules, or zip files) is run under command line control of the mpi4py.futures package by passing -m mpi4py.futures to the python executable. The mpi4py.futures invocation should be passed a pyfile path to a script (or a zipfile/directory containing a __main__.py file). Additionally, mpi4py.futures accepts -m mod to execute a module named mod, -c cmd to execute a command string cmd, or even - to read commands from standard input (sys.stdin). Summarizing, mpi4py.futures can be invoked in the following ways:

  • $ mpiexec -n numprocs python -m mpi4py.futures pyfile [arg] ...

  • $ mpiexec -n numprocs python -m mpi4py.futures -m mod [arg] ...

  • $ mpiexec -n numprocs python -m mpi4py.futures -c cmd [arg] ...

  • $ mpiexec -n numprocs python -m mpi4py.futures - [arg] ...

Before starting the main script execution, mpi4py.futures splits MPI.COMM_WORLD in one master (the process with rank 0 in MPI.COMM_WORLD) and numprocs - 1 workers and connects them through an MPI intercommunicator. Afterwards, the master process proceeds with the execution of the user script code, which eventually creates MPIPoolExecutor instances to submit tasks. Meanwhile, the worker processes follow a different execution path to serve the master. Upon successful termination of the main script at the master, the entire MPI execution environment exists gracefully. In case of any unhandled exception in the main script, the master process calls MPI.COMM_WORLD.Abort(1) to prevent deadlocks and force termination of entire MPI execution environment.

Warning

Running scripts under command line control of mpi4py.futures is quite similar to executing a single-process application that spawn additional workers as required. However, there is a very important difference users should be aware of. All MPIPoolExecutor instances created at the master will share the pool of workers. Tasks submitted at the master from many different executors will be scheduled for execution in random order as soon as a worker is idle. Any executor can easily starve all the workers (e.g., by calling MPIPoolExecutor.map() with long iterables). If that ever happens, submissions from other executors will not be serviced until free workers are available.

See also

Command line

Documentation on Python command line interface.

Examples

The following julia.py script computes the Julia set and dumps an image to disk in binary PGM format. The code starts by importing MPIPoolExecutor from the mpi4py.futures package. Next, some global constants and functions implement the computation of the Julia set. The computations are protected with the standard if __name__ == '__main__': ... idiom. The image is computed by whole scanlines submitting all these tasks at once using the map method. The result iterator yields scanlines in-order as the tasks complete. Finally, each scanline is dumped to disk.

julia.py
 1from mpi4py.futures import MPIPoolExecutor
 2
 3x0, x1, w = -2.0, +2.0, 640*2
 4y0, y1, h = -1.5, +1.5, 480*2
 5dx = (x1 - x0) / w
 6dy = (y1 - y0) / h
 7
 8c = complex(0, 0.65)
 9
10def julia(x, y):
11    z = complex(x, y)
12    n = 255
13    while abs(z) < 3 and n > 1:
14        z = z**2 + c
15        n -= 1
16    return n
17
18def julia_line(k):
19    line = bytearray(w)
20    y = y1 - k * dy
21    for j in range(w):
22        x = x0 + j * dx
23        line[j] = julia(x, y)
24    return line
25
26if __name__ == '__main__':
27
28    with MPIPoolExecutor() as executor:
29        image = executor.map(julia_line, range(h))
30        with open('julia.pgm', 'wb') as f:
31            f.write(b'P5 %d %d %d\n' % (w, h, 255))
32            for line in image:
33                f.write(line)

The recommended way to execute the script is by using the mpiexec command specifying one MPI process (master) and (optional but recommended) the desired MPI universe size, which determines the number of additional dynamically spawned processes (workers). The MPI universe size is provided either by a batch system or set by the user via command-line arguments to mpiexec or environment variables. Below we provide examples for MPICH and Open MPI implementations 1. In all of these examples, the mpiexec command launches a single master process running the Python interpreter and executing the main script. When required, mpi4py.futures spawns the pool of 16 worker processes. The master submits tasks to the workers and waits for the results. The workers receive incoming tasks, execute them, and send back the results to the master.

When using MPICH implementation or its derivatives based on the Hydra process manager, users can set the MPI universe size via the -usize argument to mpiexec:

$ mpiexec -n 1 -usize 17 python julia.py

or, alternatively, by setting the MPIEXEC_UNIVERSE_SIZE environment variable:

$ MPIEXEC_UNIVERSE_SIZE=17 mpiexec -n 1 python julia.py

In the Open MPI implementation, the MPI universe size can be set via the -host argument to mpiexec:

$ mpiexec -n 1 -host <hostname>:17 python julia.py

Another way to specify the number of workers is to use the mpi4py.futures-specific environment variable MPI4PY_FUTURES_MAX_WORKERS:

$ MPI4PY_FUTURES_MAX_WORKERS=16 mpiexec -n 1 python julia.py

Note that in this case, the MPI universe size is ignored.

Alternatively, users may decide to execute the script in a more traditional way, that is, all the MPI processes are started at once. The user script is run under command-line control of mpi4py.futures passing the -m flag to the python executable:

$ mpiexec -n 17 python -m mpi4py.futures julia.py

As explained previously, the 17 processes are partitioned in one master and 16 workers. The master process executes the main script while the workers execute the tasks submitted by the master.

1

When using an MPI implementation other than MPICH or Open MPI, please check the documentation of the implementation and/or batch system for the ways to specify the desired MPI universe size.

GIL

See global interpreter lock.

mpi4py.util

New in version 3.1.0.

The mpi4py.util package collects miscellaneous utilities within the intersection of Python and MPI.

mpi4py.util.pkl5

New in version 3.1.0.

pickle protocol 5 (see PEP 574) introduced support for out-of-band buffers, allowing for more efficient handling of certain object types with large memory footprints.

MPI for Python uses the traditional in-band handling of buffers. This approach is appropriate for communicating non-buffer Python objects, or buffer-like objects with small memory footprints. For point-to-point communication, in-band buffer handling allows for the communication of a pickled stream with a single MPI message, at the expense of additional CPU and memory overhead in the pickling and unpickling steps.

The mpi4py.util.pkl5 module provides communicator wrapper classes reimplementing pickle-based point-to-point communication methods using pickle protocol 5. Handling out-of-band buffers necessarily involve multiple MPI messages, thus increasing latency and hurting performance in case of small size data. However, in case of large size data, the zero-copy savings of out-of-band buffer handling more than offset the extra latency costs. Additionally, these wrapper methods overcome the infamous 2 GiB message count limit (MPI-1 to MPI-3).

Note

Support for pickle protocol 5 is available in the pickle module within the Python standard library since Python 3.8. Previous Python 3 releases can use the pickle5 backport, which is available on PyPI and can be installed with:

python -m pip install pickle5
class mpi4py.util.pkl5.Request(request=None)

Request.

Custom request class for nonblocking communications.

Note

Request is not a subclass of mpi4py.MPI.Request

Parameters

request (Iterable[MPI.Request]) –

Return type

Request

Free()

Free a communication request.

Return type

None

cancel()

Cancel a communication request.

Return type

None

get_status(status=None)

Non-destructive test for the completion of a request.

Parameters

status (Optional[Status]) –

Return type

bool

test(status=None)

Test for the completion of a request.

Parameters

status (Optional[Status]) –

Return type

Tuple[bool, Optional[Any]]

wait(status=None)

Wait for a request to complete.

Parameters

status (Optional[Status]) –

Return type

Any

classmethod testall(requests, statuses=None)

Test for the completion of all requests.

Classmethod

classmethod waitall(requests, statuses=None)

Wait for all requests to complete.

Classmethod

class mpi4py.util.pkl5.Message(message=None)

Message.

Custom message class for matching probes.

Note

Message is not a subclass of mpi4py.MPI.Message

Parameters

message (Iterable[MPI.Message]) –

Return type

Message

recv(status=None)

Blocking receive of matched message.

Parameters

status (Optional[Status]) –

Return type

Any

irecv()

Nonblocking receive of matched message.

Return type

Request

classmethod probe(comm, source=ANY_SOURCE, tag=ANY_TAG, status=None)

Blocking test for a matched message.

Classmethod

classmethod iprobe(comm, source=ANY_SOURCE, tag=ANY_TAG, status=None)

Nonblocking test for a matched message.

Classmethod

class mpi4py.util.pkl5.Comm

Communicator.

Base communicator wrapper class.

send(obj, dest, tag=0)

Blocking send in standard mode.

Parameters

  • obj (Any) –

  • dest (int) –

  • tag (int) –

Return type

None

bsend(obj, dest, tag=0)

Blocking send in buffered mode.

Parameters

  • obj (Any) –

  • dest (int) –

  • tag (int) –

Return type

None

ssend(obj, dest, tag=0)

Blocking send in synchronous mode.

Parameters

  • obj (Any) –

  • dest (int) –

  • tag (int) –

Return type

None

isend(obj, dest, tag=0)

Nonblocking send in standard mode.

Parameters

  • obj (Any) –

  • dest (int) –

  • tag (int) –

Return type

Request

ibsend(obj, dest, tag=0)

Nonblocking send in buffered mode.

Parameters

  • obj (Any) –

  • dest (int) –

  • tag (int) –

Return type

Request

issend(obj, dest, tag=0)

Nonblocking send in synchronous mode.

Parameters

  • obj (Any) –

  • dest (int) –

  • tag (int) –

Return type

Request

recv(buf=None, source=ANY_SOURCE, tag=ANY_TAG, status=None)

Blocking receive.

Parameters

  • buf (Optional[Buffer]) –

  • source (int) –

  • tag (int) –

  • status (Optional[Status]) –

Return type

Any

irecv(buf=None, source=ANY_SOURCE, tag=ANY_TAG)

Nonblocking receive.

Warning

This method cannot be supported reliably and raises RuntimeError.

Parameters

  • buf (Optional[Buffer]) –

  • source (int) –

  • tag (int) –

Return type

Request

sendrecv(sendobj, dest, sendtag=0, recvbuf=None, source=ANY_SOURCE, recvtag=ANY_TAG, status=None)

Send and receive.

Parameters

  • sendobj (Any) –

  • dest (int) –

  • sendtag (int) –

  • recvbuf (Optional[Buffer]) –

  • source (int) –

  • recvtag (int) –

  • status (Optional[Status]) –

Return type

Any

mprobe(source=ANY_SOURCE, tag=ANY_TAG, status=None)

Blocking test for a matched message.

Parameters

  • source (int) –

  • tag (int) –

  • status (Optional[Status]) –

Return type

Message

improbe(source=ANY_SOURCE, tag=ANY_TAG, status=None)

Nonblocking test for a matched message.

Parameters

  • source (int) –

  • tag (int) –

  • status (Optional[Status]) –

Return type

Optional[Message]

bcast(obj, root=0)

Broadcast.

Parameters

  • obj (Any) –

  • root (int) –

Return type

Any

class mpi4py.util.pkl5.Intracomm

Intracommunicator.

Intracommunicator wrapper class.

class mpi4py.util.pkl5.Intercomm

Intercommunicator.

Intercommunicator wrapper class.

Examples

test-pkl5-1.py
 1import numpy as np
 2from mpi4py import MPI
 3from mpi4py.util import pkl5
 4
 5comm = pkl5.Intracomm(MPI.COMM_WORLD)  # comm wrapper
 6size = comm.Get_size()
 7rank = comm.Get_rank()
 8dst = (rank + 1) % size
 9src = (rank - 1) % size
10
11sobj = np.full(1024**3, rank, dtype='i4')  # > 4 GiB
12sreq = comm.isend(sobj, dst, tag=42)
13robj = comm.recv (None, src, tag=42)
14sreq.Free()
15
16assert np.min(robj) == src
17assert np.max(robj) == src
test-pkl5-2.py
 1import numpy as np
 2from mpi4py import MPI
 3from mpi4py.util import pkl5
 4
 5comm = pkl5.Intracomm(MPI.COMM_WORLD)  # comm wrapper
 6size = comm.Get_size()
 7rank = comm.Get_rank()
 8dst = (rank + 1) % size
 9src = (rank - 1) % size
10
11sobj = np.full(1024**3, rank, dtype='i4')  # > 4 GiB
12sreq = comm.isend(sobj, dst, tag=42)
13
14status = MPI.Status()
15rmsg = comm.mprobe(status=status)
16assert status.Get_source() == src
17assert status.Get_tag() == 42
18rreq = rmsg.irecv()
19robj = rreq.wait()
20
21sreq.Free()
22assert np.max(robj) == src
23assert np.min(robj) == src

mpi4py.util.dtlib

New in version 3.1.0.

The mpi4py.util.dtlib module provides converter routines between NumPy and MPI datatypes.

mpi4py.util.dtlib.from_numpy_dtype(dtype)

Convert NumPy datatype to MPI datatype.

Parameters

dtype (numpy.typing.DTypeLike) – NumPy dtype-like object.

Return type

Datatype

mpi4py.util.dtlib.to_numpy_dtype(datatype)

Convert MPI datatype to NumPy datatype.

Parameters

datatype (Datatype) – MPI datatype.

Return type

numpy.dtype

mpi4py.run

New in version 3.0.0.

At import time, mpi4py initializes the MPI execution environment calling MPI_Init_thread() and installs an exit hook to automatically call MPI_Finalize() just before the Python process terminates. Additionally, mpi4py overrides the default ERRORS_ARE_FATAL error handler in favor of ERRORS_RETURN, which allows translating MPI errors in Python exceptions. These departures from standard MPI behavior may be controversial, but are quite convenient within the highly dynamic Python programming environment. Third-party code using mpi4py can just from mpi4py import MPI and perform MPI calls without the tedious initialization/finalization handling. MPI errors, once translated automatically to Python exceptions, can be dealt with the common try…except…finally clauses; unhandled MPI exceptions will print a traceback which helps in locating problems in source code.

Unfortunately, the interplay of automatic MPI finalization and unhandled exceptions may lead to deadlocks. In unattended runs, these deadlocks will drain the battery of your laptop, or burn precious allocation hours in your supercomputing facility.

Consider the following snippet of Python code. Assume this code is stored in a standard Python script file and run with mpiexec in two or more processes.

from mpi4py import MPI
assert MPI.COMM_WORLD.Get_size() > 1
rank = MPI.COMM_WORLD.Get_rank()
if rank == 0:
    1/0
    MPI.COMM_WORLD.send(None, dest=1, tag=42)
elif rank == 1:
    MPI.COMM_WORLD.recv(source=0, tag=42)

Process 0 raises ZeroDivisionError exception before performing a send call to process 1. As the exception is not handled, the Python interpreter running in process 0 will proceed to exit with non-zero status. However, as mpi4py installed a finalizer hook to call MPI_Finalize() before exit, process 0 will block waiting for other processes to also enter the MPI_Finalize() call. Meanwhile, process 1 will block waiting for a message to arrive from process 0, thus never reaching to MPI_Finalize(). The whole MPI execution environment is irremediably in a deadlock state.

To alleviate this issue, mpi4py offers a simple, alternative command line execution mechanism based on using the -m flag and implemented with the runpy module. To use this features, Python code should be run passing -m mpi4py in the command line invoking the Python interpreter. In case of unhandled exceptions, the finalizer hook will call MPI_Abort() on the MPI_COMM_WORLD communicator, thus effectively aborting the MPI execution environment.

Warning

When a process is forced to abort, resources (e.g. open files) are not cleaned-up and any registered finalizers (either with the atexit module, the Python C/API function Py_AtExit(), or even the C standard library function atexit()) will not be executed. Thus, aborting execution is an extremely impolite way of ensuring process termination. However, MPI provides no other mechanism to recover from a deadlock state.

Interface options

The use of -m mpi4py to execute Python code on the command line resembles that of the Python interpreter.

  • mpiexec -n numprocs python -m mpi4py pyfile [arg] ...

  • mpiexec -n numprocs python -m mpi4py -m mod [arg] ...

  • mpiexec -n numprocs python -m mpi4py -c cmd [arg] ...

  • mpiexec -n numprocs python -m mpi4py - [arg] ...

<pyfile>

Execute the Python code contained in pyfile, which must be a filesystem path referring to either a Python file, a directory containing a __main__.py file, or a zipfile containing a __main__.py file.

-m <mod>

Search sys.path for the named module mod and execute its contents.

-c <cmd>

Execute the Python code in the cmd string command.

-

Read commands from standard input (sys.stdin).

See also

Command line

Documentation on Python command line interface.

Reference

mpi4py.MPI

Message Passing Interface.

Citation

If MPI for Python been significant to a project that leads to an academic publication, please acknowledge that fact by citing the project.

Installation

Requirements

You need to have the following software properly installed in order to build MPI for Python:

  • A working MPI implementation, preferably supporting MPI-3 and built with shared/dynamic libraries.

    Note

    If you want to build some MPI implementation from sources, check the instructions at Building MPI from sources in the appendix.

  • Python 2.7, 3.5 or above.

    Note

    Some MPI-1 implementations do require the actual command line arguments to be passed in MPI_Init(). In this case, you will need to use a rebuilt, MPI-enabled, Python interpreter executable. MPI for Python has some support for alleviating you from this task. Check the instructions at MPI-enabled Python interpreter in the appendix.

Using pip

If you already have a working MPI (either if you installed it from sources or by using a pre-built package from your favourite GNU/Linux distribution) and the mpicc compiler wrapper is on your search path, you can use pip:

$ python -m pip install mpi4py

Note

If the mpicc compiler wrapper is not on your search path (or if it has a different name) you can use env to pass the environment variable MPICC providing the full path to the MPI compiler wrapper executable:

$ env MPICC=/path/to/mpicc python -m pip install mpi4py

Warning

pip keeps previouly built wheel files on its cache for future reuse. If you want to reinstall the mpi4py package using a different or updated MPI implementation, you have to either first remove the cached wheel file with:

$ python -m pip cache remove mpi4py

or ask pip to disable the cache:

$ python -m pip install --no-cache-dir mpi4py

Using distutils

The MPI for Python package is available for download at the project website generously hosted by GitHub. You can use curl or wget to get a release tarball.

  • Using curl:

    $ curl -O https://github.com/mpi4py/mpi4py/releases/download/X.Y.Z/mpi4py-X.Y.Z.tar.gz

  • Using wget:

    $ wget https://github.com/mpi4py/mpi4py/releases/download/X.Y.Z/mpi4py-X.Y.Z.tar.gz

After unpacking the release tarball:

$ tar -zxf mpi4py-X.Y.Z.tar.gz
$ cd mpi4py-X.Y.Z

the package is ready for building.

MPI for Python uses a standard distutils-based build system. However, some distutils commands (like build) have additional options:

--mpicc=

Lets you specify a special location or name for the mpicc compiler wrapper.

--mpi=

Lets you pass a section with MPI configuration within a special configuration file.

--configure

Runs exhaustive tests for checking about missing MPI types, constants, and functions. This option should be passed in order to build MPI for Python against old MPI-1 or MPI-2 implementations, possibly providing a subset of MPI-3.

If you use a MPI implementation providing a mpicc compiler wrapper (e.g., MPICH, Open MPI), it will be used for compilation and linking. This is the preferred and easiest way of building MPI for Python.

If mpicc is located somewhere in your search path, simply run the build command:

$ python setup.py build

If mpicc is not in your search path or the compiler wrapper has a different name, you can run the build command specifying its location:

$ python setup.py build --mpicc=/where/you/have/mpicc

Alternatively, you can provide all the relevant information about your MPI implementation by editing the file called mpi.cfg. You can use the default section [mpi] or add a new, custom section, for example [other_mpi] (see the examples provided in the mpi.cfg file as a starting point to write your own section):

[mpi]

include_dirs         = /usr/local/mpi/include
libraries            = mpi
library_dirs         = /usr/local/mpi/lib
runtime_library_dirs = /usr/local/mpi/lib

[other_mpi]

include_dirs         = /opt/mpi/include ...
libraries            = mpi ...
library_dirs         = /opt/mpi/lib ...
runtime_library_dirs = /op/mpi/lib ...

...

and then run the build command, perhaps specifying you custom configuration section:

$ python setup.py build --mpi=other_mpi

After building, the package is ready for install.

If you have root privileges (either by log-in as the root user of by using sudo) and you want to install MPI for Python in your system for all users, just do:

$ python setup.py install

The previous steps will install the mpi4py package at standard location prefix/lib/pythonX.X/site-packages.

If you do not have root privileges or you want to install MPI for Python for your private use, just do:

$ python setup.py install --user

Testing

To quickly test the installation:

$ mpiexec -n 5 python -m mpi4py.bench helloworld
Hello, World! I am process 0 of 5 on localhost.
Hello, World! I am process 1 of 5 on localhost.
Hello, World! I am process 2 of 5 on localhost.
Hello, World! I am process 3 of 5 on localhost.
Hello, World! I am process 4 of 5 on localhost.

If you installed from source, issuing at the command line:

$ mpiexec -n 5 python demo/helloworld.py

or (in the case of ancient MPI-1 implementations):

$ mpirun -np 5 python `pwd`/demo/helloworld.py

will launch a five-process run of the Python interpreter and run the test script demo/helloworld.py from the source distribution.

You can also run all the unittest scripts:

$ mpiexec -n 5 python test/runtests.py

or, if you have nose unit testing framework installed:

$ mpiexec -n 5 nosetests -w test

or, if you have py.test unit testing framework installed:

$ mpiexec -n 5 py.test test/

Appendix

MPI-enabled Python interpreter

Warning

These days it is no longer required to use the MPI-enabled Python interpreter in most cases, and, therefore, it is not built by default anymore because it is too difficult to reliably build a Python interpreter across different distributions. If you know that you still really need it, see below on how to use the build_exe and install_exe commands.

Some MPI-1 implementations (notably, MPICH 1) do require the actual command line arguments to be passed at the time MPI_Init() is called. In this case, you will need to use a re-built, MPI-enabled, Python interpreter binary executable. A basic implementation (targeting Python 2.X) of what is required is shown below:

#include <Python.h>
#include <mpi.h>

int main(int argc, char *argv[])
{
   int status, flag;
   MPI_Init(&argc, &argv);
   status = Py_Main(argc, argv);
   MPI_Finalized(&flag);
   if (!flag) MPI_Finalize();
   return status;
}

The source code above is straightforward; compiling it should also be. However, the linking step is more tricky: special flags have to be passed to the linker depending on your platform. In order to alleviate you for such low-level details, MPI for Python provides some pure-distutils based support to build and install an MPI-enabled Python interpreter executable:

$ cd mpi4py-X.X.X
$ python setup.py build_exe [--mpi=<name>|--mpicc=/path/to/mpicc]
$ [sudo] python setup.py install_exe [--install-dir=$HOME/bin]

After the above steps you should have the MPI-enabled interpreter installed as prefix/bin/pythonX.X-mpi (or $HOME/bin/pythonX.X-mpi). Assuming that prefix/bin (or $HOME/bin) is listed on your PATH, you should be able to enter your MPI-enabled Python interactively, for example:

$ python2.7-mpi
Python 2.7.8 (default, Nov 10 2014, 08:19:18)
[GCC 4.9.2 20141101 (Red Hat 4.9.2-1)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
>>> import sys
>>> sys.executable
'/usr/bin/python2.7-mpi'
>>>

Building MPI from sources

In the list below you have some executive instructions for building some of the open-source MPI implementations out there with support for shared/dynamic libraries on POSIX environments.

  • MPICH

    $ tar -zxf mpich-X.X.X.tar.gz
$ cd mpich-X.X.X
$ ./configure --enable-shared --prefix=/usr/local/mpich
$ make
$ make install

  • Open MPI

    $ tar -zxf openmpi-X.X.X tar.gz
$ cd openmpi-X.X.X
$ ./configure --prefix=/usr/local/openmpi
$ make all
$ make install

  • MPICH 1

    $ tar -zxf mpich-X.X.X.tar.gz
$ cd mpich-X.X.X
$ ./configure --enable-sharedlib --prefix=/usr/local/mpich1
$ make
$ make install

Perhaps you will need to set the LD_LIBRARY_PATH environment variable (using export, setenv or what applies to your system) pointing to the directory containing the MPI libraries . In case of getting runtime linking errors when running MPI programs, the following lines can be added to the user login shell script (.profile, .bashrc, etc.).

  • MPICH

    MPI_DIR=/usr/local/mpich
export LD_LIBRARY_PATH=$MPI_DIR/lib:$LD_LIBRARY_PATH

  • Open MPI

    MPI_DIR=/usr/local/openmpi
export LD_LIBRARY_PATH=$MPI_DIR/lib:$LD_LIBRARY_PATH

  • MPICH 1

    MPI_DIR=/usr/local/mpich1
export LD_LIBRARY_PATH=$MPI_DIR/lib/shared:$LD_LIBRARY_PATH:
export MPICH_USE_SHLIB=yes

    Warning

    MPICH 1 support for dynamic libraries is not completely transparent. Users should set the environment variable MPICH_USE_SHLIB to yes in order to avoid link problems when using the mpicc compiler wrapper.