1
1
openmpi/ompi/mpi/cxx/comm.cc

160 строки
5.8 KiB
C++
Исходник Обычный вид История

// -*- c++ -*-
//
// Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
// University Research and Technology
// Corporation. All rights reserved.
// Copyright (c) 2004-2005 The University of Tennessee and The University
// of Tennessee Research Foundation. All rights
// reserved.
// Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
// University of Stuttgart. All rights reserved.
// Copyright (c) 2004-2005 The Regents of the University of California.
// All rights reserved.
// Copyright (c) 2007-2008 Cisco Systems, Inc. All rights reserved.
// $COPYRIGHT$
//
// Additional copyrights may follow
//
// $HEADER$
//
// do not include ompi_config.h because it kills the free/malloc defines
#include "mpi.h"
#include "ompi/mpi/cxx/mpicxx.h"
#ifdef HAVE_SCHED_H
#include <sched.h>
#endif
#include "opal/threads/mutex.h"
#include "opal/class/opal_list.h"
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00
#include "ompi/communicator/communicator.h"
#include "ompi/attribute/attribute.h"
#include "ompi/errhandler/errhandler.h"
//
// These functions are all not inlined because they need to use locks to
// protect the handle maps and it would be bad to have those in headers
// because that would require that we always install the lock headers.
// Instead we take the function call hit (we're locking - who cares about
// a function call. And these aren't exactly the performance critical
// functions) and make everyone's life easier.
//
// construction
MPI::Comm::Comm()
{
}
// copy
MPI::Comm::Comm(const Comm_Null& data) : Comm_Null(data)
{
}
// This function needs some internal OMPI types, so it's not inlined
MPI::Errhandler
MPI::Comm::Create_errhandler(MPI::Comm::_MPI2CPP_ERRHANDLERFN_* function)
{
MPI_Errhandler c_errhandler =
ompi_errhandler_create(OMPI_ERRHANDLER_TYPE_COMM,
(ompi_errhandler_generic_handler_fn_t*) function,
OMPI_ERRHANDLER_LANG_CXX);
c_errhandler->eh_cxx_dispatch_fn =
(ompi_errhandler_cxx_dispatch_fn_t*)
ompi_mpi_cxx_comm_errhandler_invoke;
return c_errhandler;
}
//JGS I took the const out because it causes problems when trying to
//call this function with the predefined NULL_COPY_FN etc.
int
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00
MPI::Comm::do_create_keyval(MPI_Comm_copy_attr_function* c_copy_fn,
MPI_Comm_delete_attr_function* c_delete_fn,
Copy_attr_function* cxx_copy_fn,
Delete_attr_function* cxx_delete_fn,
void* extra_state, int &keyval)
{
int ret, count = 0;
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00
ompi_attribute_fn_ptr_union_t copy_fn;
ompi_attribute_fn_ptr_union_t delete_fn;
keyval_intercept_data_t *cxx_extra_state;
// If both the callbacks are C, then do the simple thing -- no
// need for all the C++ machinery.
if (NULL != c_copy_fn && NULL != c_delete_fn) {
copy_fn.attr_communicator_copy_fn =
(MPI_Comm_internal_copy_attr_function*) c_copy_fn;
delete_fn.attr_communicator_delete_fn = c_delete_fn;
ret = ompi_attr_create_keyval(COMM_ATTR, copy_fn, delete_fn,
&keyval, extra_state, 0, NULL);
if (MPI_SUCCESS != ret) {
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, ret,
"MPI::Comm::Create_keyval");
}
}
// If either callback is C++, then we have to use the C++
// callbacks for both, because we have to generate a new
// extra_state. And since we only get one extra_state (i.e., we
// don't get one extra_state for the copy callback and another
// extra_state for the delete callback), we have to use the C++
// callbacks for both (and therefore translate the C++-special
// extra_state into the user's original extra_state). Ensure to
// malloc() the struct here (vs new) so that it can be free()'ed
// by the C attribute base.
cxx_extra_state = (keyval_intercept_data_t*)
malloc(sizeof(keyval_intercept_data_t));
if (NULL == cxx_extra_state) {
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_NO_MEM,
"MPI::Comm::Create_keyval");
}
cxx_extra_state->c_copy_fn = c_copy_fn;
cxx_extra_state->cxx_copy_fn = cxx_copy_fn;
cxx_extra_state->c_delete_fn = c_delete_fn;
cxx_extra_state->cxx_delete_fn = cxx_delete_fn;
cxx_extra_state->extra_state = extra_state;
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00
// Error check. Must have exactly 2 non-NULL function pointers.
if (NULL != c_copy_fn) {
++count;
}
if (NULL != c_delete_fn) {
++count;
}
if (NULL != cxx_copy_fn) {
++count;
}
if (NULL != cxx_delete_fn) {
++count;
}
if (2 != count) {
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, MPI_ERR_ARG,
"MPI::Comm::Create_keyval");
}
// We do not call MPI_Comm_create_keyval() here because we need to
// pass in the cxx_extra_state to the backend keyval creation so
// that when the keyval is destroyed (i.e., when its refcount goes
// to 0), the cxx_extra_state is free()'ed.
copy_fn.attr_communicator_copy_fn =
(MPI_Comm_internal_copy_attr_function*)
ompi_mpi_cxx_comm_copy_attr_intercept;
delete_fn.attr_communicator_delete_fn =
ompi_mpi_cxx_comm_delete_attr_intercept;
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00
ret = ompi_attr_create_keyval(COMM_ATTR, copy_fn, delete_fn,
&keyval, cxx_extra_state, 0,
cxx_extra_state);
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00
if (OMPI_SUCCESS != ret) {
return OMPI_ERRHANDLER_INVOKE(MPI_COMM_WORLD, ret,
"MPI::Comm::Create_keyval");
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00
}
return MPI_SUCCESS;
}
Fixes trac:817 The C++ bindings were not tracking keyvals properly -- they were freeing some internal meta data when Free_keyval() was called, not when the keyval was actually destroyed (keyvals are refcounted in the C layer, just like all other MPI objects, because they can live for long after their corresponding Free call is invoked). This commit fixes this problem and several other things: * Add infrastructure on the ompi_attribute_keyval_t for an "extra" destructor pointer that will be invoked during the "real" constructor (i.e., when OBJ_RELEASE puts the refcount to 0). This allows calling back into the C++ layer to release meta data associated with the keyval. * Adjust all cases where keyvals are created to pass in relevant destructors (NULL or the C++ destructor). * Do essentially the same for MPI::Comm, MPI::Win, and MPI:Datatype: * Move several functions out of the .cc file into the _inln.h file since they no longer require locks * Make the 4 Create_keyval() functions call a common back-end keyval creation function that does the Right Thing depending on whether C or C++ function pointers were used for the keyval functions. The back-end function does not call the corresponding C MPI_*_create_keyval function, but rather does the work itself so that it can associate a "destructor" callback for the C++ bindings for when the keyval is actually destroyed. * Change a few type names to be more indicative of what they are (mostly dealing with keyvals [not "keys"]). * Add the 3 missing bindings for MPI::Comm::Create_keyval(). * Remove MPI::Comm::comm_map (and associated types) because it's no longer necessary in the intercepts -- it was a by-product of being a portable C++ bindings layer. Now we can just query the C layer directly to figure out what type a communicator is. This solves some logistics / callback issues, too. * Rename several types, variables, and fix many comments in the back-end C attribute implementation to make the names really reflect what they are (keyvals vs. attributes). The previous names heavily overloaded the name "key" and were ''extremely'' confusing. This commit was SVN r13565. The following Trac tickets were found above: Ticket 817 --> https://svn.open-mpi.org/trac/ompi/ticket/817
2007-02-09 02:50:04 +03:00