8.7.3. Initialization


Up: MPI and Threads Next: New Attribute Caching Functions Previous: Clarifications

The following function may be used to initialize MPI, and initialize the MPI thread environment, instead of MPI_INIT.

MPI_INIT_THREAD(required, provided)
IN requireddesired level of thread support (integer)
OUT providedprovided level of thread support (integer)

int MPI_Init_thread(int *argc, char *((*argv)[]), int required, int *provided)

MPI_INIT_THREAD(REQUIRED, PROVIDED, IERROR)
INTEGER REQUIRED, PROVIDED, IERROR

int MPI::Init_thread(int& argc, char**& argv, int required)
int MPI::Init_thread(int required)


Advice to users.

In C and C++, the passing of argc and argv is optional. In C, this is accomplished by passing the appropriate null pointer. In C++, this is accomplished with two separate bindings to cover these two cases. This is as with MPI_INIT as discussed in Section Passing NULL to MPI_Init . ( End of advice to users.)

This call initializes MPI in the same way that a call to MPI_INIT would. In addition, it initializes the thread environment. The argument required is used to specify the desired level of thread support. The possible values are listed in increasing order of thread support.

{ MPI_THREAD_SINGLE}
Only one thread will execute.
{ MPI_THREAD_FUNNELED}
The process may be multi-threaded, but only the main thread will make MPI calls (all MPI calls are ``funneled'' to the main thread).
{ MPI_THREAD_SERIALIZED}
The process may be multi-threaded, and multiple threads may make MPI calls, but only one at a time: MPI calls are not made concurrently from two distinct threads (all MPI calls are ``serialized'').
{ MPI_THREAD_MULTIPLE}
Multiple threads may call MPI, with no restrictions.

These values are monotonic; i.e., MPI_THREAD_SINGLE < MPI_THREAD_FUNNELED < MPI_THREAD_SERIALIZED < MPI_THREAD_MULTIPLE.

Different processes in MPI_COMM_WORLD may require different levels of thread support.

The call returns in provided information about the actual level of thread support that will be provided by MPI. It can be one of the four values listed above.

The level(s) of thread support that can be provided by MPI_INIT_THREAD will depend on the implementation, and may depend on information provided by the user before the program started to execute (e.g., with arguments to mpiexec). If possible, the call will return provided = required. Failing this, the call will return the least supported level such that provided > required (thus providing a stronger level of support than required by the user). Finally, if the user requirement cannot be satisfied, then the call will return in provided the highest supported level.

A thread compliant MPI implementation will be able to return provided
= MPI_THREAD_MULTIPLE. Such an implementation may always return provided
= MPI_THREAD_MULTIPLE, irrespective of the value of required. At the other extreme, an MPI library that is not thread compliant may always return provided = MPI_THREAD_SINGLE, irrespective of the value of required.

A call to MPI_INIT has the same effect as a call to MPI_INIT_THREAD with a required = MPI_THREAD_SINGLE.

Vendors may provide (implementation dependent) means to specify the level(s) of thread support available when the MPI program is started, e.g., with arguments to mpiexec. This will affect the outcome of calls to MPI_INIT and MPI_INIT_THREAD. Suppose, for example, that an MPI program has been started so that only MPI_THREAD_MULTIPLE is available. Then MPI_INIT_THREAD will return provided = MPI_THREAD_MULTIPLE, irrespective of the value of required; a call to MPI_INIT will also initialize the MPI thread support level to MPI_THREAD_MULTIPLE. Suppose, on the other hand, that an MPI program has been started so that all four levels of thread support are available. Then, a call to MPI_INIT_THREAD will return provided = required; on the other hand, a call to MPI_INIT will initialize the MPI thread support level to MPI_THREAD_SINGLE.


Rationale.

Various optimizations are possible when MPI code is executed single-threaded, or is executed on multiple threads, but not concurrently: mutual exclusion code may be omitted. Furthermore, if only one thread executes, then the MPI library can use library functions that are not thread safe, without risking conflicts with user threads. Also, the model of one communication thread, multiple computation threads fits well many applications. E.g., if the process code is a sequential Fortran/C/C++ program with MPI calls that has been parallelized by a compiler for execution on an SMP node, in a cluster of SMPs, then the process computation is multi-threaded, but MPI calls will likely execute on a single thread.

The design accommodates a static specification of the thread support level, for environments that require static binding of libraries, and for compatibility for current multi-threaded MPI codes. ( End of rationale.)

Advice to implementors.

If provided is not MPI_THREAD_SINGLE then the MPI library should not invoke C/ C++/Fortran library calls that are not thread safe, e.g., in an environment where malloc is not thread safe, then malloc should not be used by the MPI library.

Some implementors may want to use different MPI libraries for different levels of thread support. They can do so using dynamic linking and selecting which library will be linked when MPI_INIT_THREAD is invoked. If this is not possible, then optimizations for lower levels of thread support will occur only when the level of thread support required is specified at link time. ( End of advice to implementors.)
The following function can be used to query the current level of thread support.

MPI_QUERY_THREAD(provided)
OUT providedprovided level of thread support (integer)

int MPI_Query_thread(int *provided)

MPI_QUERY_THREAD(PROVIDED, IERROR)
INTEGER PROVIDED, IERROR

int MPI::Query_thread()

The call returns in provided the current level of thread support. This will be the value returned in provided by MPI_INIT_THREAD, if MPI was initialized by a call to MPI_INIT_THREAD().

MPI_IS_THREAD_MAIN(flag)
OUT flagtrue if calling thread is main thread, false otherwise (logical)

int MPI_Is_thread_main(int *flag)

MPI_IS_THREAD_MAIN(FLAG, IERROR)
LOGICAL FLAG
INTEGER IERROR

bool MPI::Is_thread_main()

This function can be called by a thread to find out whether it is the main thread (the thread that called MPI_INIT or MPI_INIT_THREAD).

All routines listed in this section must be supported by all MPI implementations.


Rationale.

MPI libraries are required to provide these calls even if they do not support threads, so that portable code that contains invocations to these functions be able to link correctly. MPI_INIT continues to be supported so as to provide compatibility with current MPI codes. ( End of rationale.)

Advice to users.

It is possible to spawn threads before MPI is initialized, but no MPI call other than MPI_INITIALIZED should be executed by these threads, until MPI_INIT_THREAD is invoked by one thread (which, thereby, becomes the main thread). In particular, it is possible to enter the MPI execution with a multi-threaded process.

The level of thread support provided is a global property of the MPI process that can be specified only once, when MPI is initialized on that process (or before). Portable third party libraries have to be written so as to accommodate any provided level of thread support. Otherwise, their usage will be restricted to specific level(s) of thread support. If such a library can run only with specific level(s) of thread support, e.g., only with MPI_THREAD_MULTIPLE, then MPI_QUERY_THREAD can be used to check whether the user initialized MPI to the correct level of thread support and, if not, raise an exception. ( End of advice to users.)



Up: MPI and Threads Next: New Attribute Caching Functions Previous: Clarifications


Return to MPI-2 Standard Index
Return to MPI 1.1 Standard Index
Return to MPI Forum Home Page

MPI-2.0 of July 18, 1997
HTML Generated on September 10, 2001