2014-03-27 17:56:00 +00:00
|
|
|
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
2013-07-11 20:47:08 +00:00
|
|
|
/*
|
|
|
|
* Copyright (c) 2004-2006 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) 2010-2012 Sandia National Laboratories. All rights reserved.
|
2014-03-27 17:56:00 +00:00
|
|
|
* Copyright (c) 2014 Los Alamos National Security, LLC. All rights
|
|
|
|
* reserved.
|
Per the PMIx RFC:
WHAT: Merge the PMIx branch into the devel repo, creating a new
OPAL “lmix” framework to abstract PMI support for all RTEs.
Replace the ORTE daemon-level collectives with a new PMIx
server and update the ORTE grpcomm framework to support
server-to-server collectives
WHY: We’ve had problems dealing with variations in PMI implementations,
and need to extend the existing PMI definitions to meet exascale
requirements.
WHEN: Mon, Aug 25
WHERE: https://github.com/rhc54/ompi-svn-mirror.git
Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding.
All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level.
Accordingly, we have:
* created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations.
* Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported.
* Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint
* removed the prior OMPI/OPAL modex code
* added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform.
* retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand
This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
|
|
|
* Copyright (c) 2014 Intel, Inc. All rights reserved.
|
2013-07-11 20:47:08 +00:00
|
|
|
* $COPYRIGHT$
|
|
|
|
*
|
|
|
|
* Additional copyrights may follow
|
|
|
|
*
|
|
|
|
* $HEADER$
|
|
|
|
*/
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
#include "opal_config.h"
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
#include <sys/types.h>
|
|
|
|
#include <unistd.h>
|
|
|
|
#include <stdio.h>
|
2014-03-27 17:56:00 +00:00
|
|
|
#include "opal_stdint.h"
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
#include "opal/class/opal_bitmap.h"
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
#include "opal/constants.h"
|
|
|
|
#include "opal/mca/btl/btl.h"
|
2013-07-11 20:47:08 +00:00
|
|
|
#include "opal/datatype/opal_convertor.h"
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
#include "opal/util/proc.h"
|
Per the PMIx RFC:
WHAT: Merge the PMIx branch into the devel repo, creating a new
OPAL “lmix” framework to abstract PMI support for all RTEs.
Replace the ORTE daemon-level collectives with a new PMIx
server and update the ORTE grpcomm framework to support
server-to-server collectives
WHY: We’ve had problems dealing with variations in PMI implementations,
and need to extend the existing PMI definitions to meet exascale
requirements.
WHEN: Mon, Aug 25
WHERE: https://github.com/rhc54/ompi-svn-mirror.git
Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding.
All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level.
Accordingly, we have:
* created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations.
* Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported.
* Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint
* removed the prior OMPI/OPAL modex code
* added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform.
* retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand
This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
|
|
|
#include "opal/mca/pmix/pmix.h"
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
#include "btl_portals4.h"
|
|
|
|
#include "btl_portals4_recv.h"
|
|
|
|
|
|
|
|
mca_btl_portals4_module_t mca_btl_portals4_module = {
|
2014-04-14 19:29:26 +00:00
|
|
|
.super = {
|
|
|
|
.btl_component = &mca_btl_portals4_component.super,
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-04-14 19:29:26 +00:00
|
|
|
/* NOTE: All the default values are set in
|
2013-07-11 20:47:08 +00:00
|
|
|
component_open() */
|
|
|
|
|
2014-04-14 19:29:26 +00:00
|
|
|
.btl_add_procs = mca_btl_portals4_add_procs,
|
|
|
|
.btl_del_procs = mca_btl_portals4_del_procs,
|
|
|
|
.btl_finalize = mca_btl_portals4_finalize,
|
|
|
|
.btl_alloc = mca_btl_portals4_alloc,
|
|
|
|
.btl_free = mca_btl_portals4_free,
|
|
|
|
.btl_prepare_src = mca_btl_portals4_prepare_src,
|
|
|
|
.btl_prepare_dst = mca_btl_portals4_prepare_dst,
|
|
|
|
.btl_send = mca_btl_portals4_send,
|
|
|
|
.btl_get = mca_btl_portals4_get,
|
|
|
|
.btl_dump = mca_btl_base_dump,
|
2013-07-11 20:47:08 +00:00
|
|
|
},
|
|
|
|
};
|
|
|
|
|
|
|
|
int
|
|
|
|
mca_btl_portals4_add_procs(struct mca_btl_base_module_t* btl_base,
|
|
|
|
size_t nprocs,
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
struct opal_proc_t **procs,
|
2013-07-11 20:47:08 +00:00
|
|
|
struct mca_btl_base_endpoint_t** btl_peer_data,
|
|
|
|
opal_bitmap_t* reachable)
|
|
|
|
{
|
2014-01-09 20:01:42 +00:00
|
|
|
struct mca_btl_portals4_module_t* portals4_btl = (struct mca_btl_portals4_module_t*) btl_base;
|
2013-07-11 20:47:08 +00:00
|
|
|
int ret;
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
struct opal_proc_t *curr_proc = NULL;
|
2013-07-11 20:47:08 +00:00
|
|
|
ptl_process_t *id;
|
|
|
|
size_t i, size;
|
|
|
|
bool need_activate = false;
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
opal_output_verbose(50, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"mca_btl_portals4_add_procs: Adding %d procs (%d) for NI %d", (int) nprocs,
|
|
|
|
(int) portals4_btl->portals_num_procs, portals4_btl->interface_num);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
if (0 == portals4_btl->portals_num_procs) {
|
2013-07-11 20:47:08 +00:00
|
|
|
need_activate = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
for (i = 0 ; i < nprocs ; ++i) {
|
|
|
|
curr_proc = procs[i];
|
|
|
|
|
|
|
|
/* portals doesn't support heterogeneous yet... */
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
if (opal_proc_local_get()->proc_arch != curr_proc->proc_arch) {
|
2013-07-11 20:47:08 +00:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
Per the PMIx RFC:
WHAT: Merge the PMIx branch into the devel repo, creating a new
OPAL “lmix” framework to abstract PMI support for all RTEs.
Replace the ORTE daemon-level collectives with a new PMIx
server and update the ORTE grpcomm framework to support
server-to-server collectives
WHY: We’ve had problems dealing with variations in PMI implementations,
and need to extend the existing PMI definitions to meet exascale
requirements.
WHEN: Mon, Aug 25
WHERE: https://github.com/rhc54/ompi-svn-mirror.git
Several community members have been working on a refactoring of the current PMI support within OMPI. Although the APIs are common, Slurm and Cray implement a different range of capabilities, and package them differently. For example, Cray provides an integrated PMI-1/2 library, while Slurm separates the two and requires the user to specify the one to be used at runtime. In addition, several bugs in the Slurm implementations have caused problems requiring extra coding.
All this has led to a slew of #if’s in the PMI code and bugs when the corner-case logic for one implementation accidentally traps the other. Extending this support to other implementations would have increased this complexity to an unacceptable level.
Accordingly, we have:
* created a new OPAL “pmix” framework to abstract the PMI support, with separate components for Cray, Slurm PMI-1, and Slurm PMI-2 implementations.
* Replaced the current ORTE grpcomm daemon-based collective operation with an integrated PMIx server, and updated the grpcomm APIs to provide more flexible, multi-algorithm support for collective operations. At this time, only the xcast and allgather operations are supported.
* Replaced the current global collective id with a signature based on the names of the participating procs. The allows an unlimited number of collectives to be executed by any group of processes, subject to the requirement that only one collective can be active at a time for a unique combination of procs. Note that a proc can be involved in any number of simultaneous collectives - it is the specific combination of procs that is subject to the constraint
* removed the prior OMPI/OPAL modex code
* added new macros for executing modex send/recv to simplify use of the new APIs. The send macros allow the caller to specify whether or not the BTL supports async modex operations - if so, then the non-blocking “fence” operation is used, if the active PMIx component supports it. Otherwise, the default is a full blocking modex exchange as we currently perform.
* retained the current flag that directs us to use a blocking fence operation, but only to retrieve data upon demand
This commit was SVN r32570.
2014-08-21 18:56:47 +00:00
|
|
|
OPAL_MODEX_RECV(ret, &mca_btl_portals4_component.super.btl_version,
|
|
|
|
curr_proc, (void**) &id, &size);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
if (OPAL_SUCCESS != ret) {
|
|
|
|
opal_output_verbose(0, opal_btl_base_framework.framework_output,
|
|
|
|
"btl/portals4: opal_modex_recv failed: %d", ret);
|
2013-07-11 20:47:08 +00:00
|
|
|
return ret;
|
2014-01-09 20:01:42 +00:00
|
|
|
}
|
|
|
|
if (size < sizeof(ptl_process_t)) { /* no available connection */
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
return OPAL_ERROR;
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
2014-01-09 20:01:42 +00:00
|
|
|
if ((size % sizeof(ptl_process_t)) != 0) {
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
opal_output_verbose(0, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"btl/portals4: invalid format in modex");
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
return OPAL_ERROR;
|
2013-08-30 16:54:55 +00:00
|
|
|
}
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"btl/portals4: %d NI(s) declared in the modex", (int) (size/sizeof(ptl_process_t))));
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
btl_peer_data[i] = malloc(sizeof(mca_btl_base_endpoint_t));
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
if (NULL == btl_peer_data[i]) return OPAL_ERROR;
|
2014-01-09 20:01:42 +00:00
|
|
|
|
|
|
|
/* If the modex received one id per interface (this is the
|
|
|
|
normal case), store the id of the corresponding
|
|
|
|
interface */
|
|
|
|
if (size / sizeof(ptl_process_t) >= portals4_btl->interface_num)
|
|
|
|
btl_peer_data[i]->ptl_proc = id[portals4_btl->interface_num];
|
|
|
|
else btl_peer_data[i]->ptl_proc = *id;
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"add_procs: nid=%x pid=%x for NI %d\n",
|
|
|
|
btl_peer_data[i]->ptl_proc.phys.nid,
|
|
|
|
btl_peer_data[i]->ptl_proc.phys.pid,
|
|
|
|
portals4_btl->interface_num));
|
|
|
|
|
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_num_procs, 1);
|
2013-07-11 20:47:08 +00:00
|
|
|
/* and here we can reach */
|
|
|
|
opal_bitmap_set_bit(reachable, i);
|
|
|
|
}
|
2014-01-09 20:01:42 +00:00
|
|
|
if (need_activate && portals4_btl->portals_num_procs > 0) {
|
|
|
|
ret = mca_btl_portals4_recv_enable(portals4_btl);
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
return OPAL_SUCCESS;
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
int
|
|
|
|
mca_btl_portals4_del_procs(struct mca_btl_base_module_t *btl,
|
|
|
|
size_t nprocs,
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
struct opal_proc_t **procs,
|
2013-07-11 20:47:08 +00:00
|
|
|
struct mca_btl_base_endpoint_t **btl_peer_data)
|
|
|
|
{
|
2014-01-09 20:01:42 +00:00
|
|
|
struct mca_btl_portals4_module_t* portals4_btl = (struct mca_btl_portals4_module_t*) btl;
|
2013-07-11 20:47:08 +00:00
|
|
|
size_t i;
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
opal_output_verbose(50, opal_btl_base_framework.framework_output,
|
2013-07-11 20:47:08 +00:00
|
|
|
"mca_btl_portals4_del_procs: Removing %d procs (%d)", (int) nprocs,
|
2014-01-09 20:01:42 +00:00
|
|
|
(int) portals4_btl->portals_num_procs);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2013-08-30 16:54:55 +00:00
|
|
|
/* See comment in btl_portals4_endpoint.h about why we look at the
|
|
|
|
portals4 entry in proc_endpoints instead of the peer_data */
|
2013-07-11 20:47:08 +00:00
|
|
|
for (i = 0 ; i < nprocs ; ++i) {
|
2014-01-09 20:01:42 +00:00
|
|
|
free(btl_peer_data[i]);
|
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_num_procs, -1);
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
if (0 == portals4_btl->portals_num_procs)
|
|
|
|
mca_btl_portals4_free_module(portals4_btl);
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
return OPAL_SUCCESS;
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
mca_btl_base_descriptor_t*
|
|
|
|
mca_btl_portals4_alloc(struct mca_btl_base_module_t* btl_base,
|
|
|
|
struct mca_btl_base_endpoint_t* endpoint,
|
|
|
|
uint8_t order,
|
|
|
|
size_t size,
|
|
|
|
uint32_t flags)
|
|
|
|
{
|
2014-01-09 20:01:42 +00:00
|
|
|
struct mca_btl_portals4_module_t* portals4_btl = (struct mca_btl_portals4_module_t*) btl_base;
|
2013-07-11 20:47:08 +00:00
|
|
|
mca_btl_portals4_frag_t* frag;
|
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
if (size <= portals4_btl->super.btl_eager_limit) {
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_ALLOC_EAGER(portals4_btl, frag);
|
2013-07-11 20:47:08 +00:00
|
|
|
if (NULL == frag) return NULL;
|
|
|
|
frag->segments[0].base.seg_len = size;
|
|
|
|
} else {
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_ALLOC_MAX(portals4_btl, frag);
|
2013-07-11 20:47:08 +00:00
|
|
|
if (NULL == frag) return NULL;
|
|
|
|
frag->segments[0].base.seg_len =
|
2014-01-09 20:01:42 +00:00
|
|
|
size <= portals4_btl->super.btl_max_send_size ?
|
|
|
|
size : portals4_btl->super.btl_max_send_size ;
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
frag->md_h = PTL_INVALID_HANDLE;
|
2014-07-10 16:31:15 +00:00
|
|
|
frag->base.des_local_count = 1;
|
2013-07-11 20:47:08 +00:00
|
|
|
frag->base.des_flags = flags | MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
|
|
|
|
frag->base.order = MCA_BTL_NO_ORDER;
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2013-07-11 20:47:08 +00:00
|
|
|
"mca_btl_portals4_alloc: %p\n", (void *) &frag->base));
|
|
|
|
return &frag->base;
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
mca_btl_portals4_free(struct mca_btl_base_module_t* btl_base,
|
|
|
|
mca_btl_base_descriptor_t* des)
|
|
|
|
{
|
2014-01-09 20:01:42 +00:00
|
|
|
struct mca_btl_portals4_module_t* portals4_btl = (struct mca_btl_portals4_module_t*) btl_base;
|
2013-07-11 20:47:08 +00:00
|
|
|
mca_btl_portals4_frag_t* frag = (mca_btl_portals4_frag_t*) des;
|
|
|
|
|
|
|
|
if (BTL_PORTALS4_FRAG_TYPE_EAGER == frag->type) {
|
|
|
|
/* don't ever unlink eager frags */
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_RETURN_EAGER(portals4_btl, frag);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
} else if (BTL_PORTALS4_FRAG_TYPE_MAX == frag->type) {
|
|
|
|
if (frag->me_h != PTL_INVALID_HANDLE) {
|
|
|
|
frag->me_h = PTL_INVALID_HANDLE;
|
|
|
|
}
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_RETURN_MAX(portals4_btl, frag);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
} else if (BTL_PORTALS4_FRAG_TYPE_USER == frag->type) {
|
|
|
|
if (frag->me_h != PTL_INVALID_HANDLE) {
|
|
|
|
frag->me_h = PTL_INVALID_HANDLE;
|
|
|
|
}
|
2014-01-09 20:01:42 +00:00
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, -1);
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"mca_btl_portals4_free: Decrementing portals_outstanding_ops=%d\n", portals4_btl->portals_outstanding_ops));
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_RETURN_USER(portals4_btl, frag);
|
2013-07-11 20:47:08 +00:00
|
|
|
} else {
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
return OPAL_ERR_BAD_PARAM;
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
return OPAL_SUCCESS;
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
/**
|
|
|
|
* Pack data and return a descriptor that can be
|
|
|
|
* used for send/put.
|
|
|
|
*
|
|
|
|
* @param btl (IN) BTL module
|
|
|
|
* @param peer (IN) BTL peer addressing
|
|
|
|
*/
|
|
|
|
|
|
|
|
mca_btl_base_descriptor_t*
|
|
|
|
mca_btl_portals4_prepare_src(struct mca_btl_base_module_t* btl_base,
|
|
|
|
struct mca_btl_base_endpoint_t* peer,
|
|
|
|
mca_mpool_base_registration_t* registration,
|
|
|
|
struct opal_convertor_t* convertor,
|
|
|
|
uint8_t order,
|
|
|
|
size_t reserve,
|
|
|
|
size_t* size,
|
|
|
|
uint32_t flags)
|
|
|
|
{
|
2014-01-09 20:01:42 +00:00
|
|
|
struct mca_btl_portals4_module_t* portals4_btl = (struct mca_btl_portals4_module_t*) btl_base;
|
2013-07-11 20:47:08 +00:00
|
|
|
mca_btl_portals4_frag_t* frag;
|
|
|
|
size_t max_data = *size;
|
|
|
|
struct iovec iov;
|
|
|
|
uint32_t iov_count = 1;
|
|
|
|
int ret;
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2013-07-11 20:47:08 +00:00
|
|
|
"mca_btl_portals4_prepare_src reserve=%ld size=%ld max_data=%ld\n", reserve, *size, max_data));
|
|
|
|
|
|
|
|
if (0 != reserve || 0 != opal_convertor_need_buffers(convertor)) {
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "mca_btl_portals4_prepare_src NEED BUFFERS or RESERVE\n"));
|
2013-07-11 20:47:08 +00:00
|
|
|
frag = (mca_btl_portals4_frag_t*) mca_btl_portals4_alloc(btl_base, peer, MCA_BTL_NO_ORDER, max_data + reserve, flags);
|
|
|
|
if (NULL == frag) {
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (max_data + reserve > frag->size) {
|
|
|
|
max_data = frag->size - reserve;
|
|
|
|
}
|
|
|
|
iov.iov_len = max_data;
|
|
|
|
iov.iov_base = (unsigned char*) frag->segments[0].base.seg_addr.pval + reserve;
|
|
|
|
ret = opal_convertor_pack(convertor, &iov, &iov_count, &max_data );
|
|
|
|
|
|
|
|
*size = max_data;
|
|
|
|
if ( ret < 0 ) {
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
frag->segments[0].base.seg_len = max_data + reserve;
|
2014-07-10 16:31:15 +00:00
|
|
|
frag->base.des_local_count = 1;
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
} else {
|
|
|
|
/* no need to pack - rdma operation out of user's buffer */
|
|
|
|
ptl_me_t me;
|
|
|
|
|
|
|
|
/* reserve space in the event queue for rdma operations immediately */
|
2014-01-09 20:01:42 +00:00
|
|
|
while (OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, 1) >
|
|
|
|
portals4_btl->portals_max_outstanding_ops) {
|
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, -1);
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "Call to mca_btl_portals4_component_progress (1)\n"));
|
2013-07-11 20:47:08 +00:00
|
|
|
mca_btl_portals4_component_progress();
|
|
|
|
}
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_ALLOC_USER(portals4_btl, frag);
|
2013-07-11 20:47:08 +00:00
|
|
|
if (NULL == frag){
|
2014-01-09 20:01:42 +00:00
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, -1);
|
2013-07-11 20:47:08 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"mca_btl_portals4_prepare_src: Incrementing portals_outstanding_ops=%d\n", portals4_btl->portals_outstanding_ops));
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
iov.iov_len = max_data;
|
|
|
|
iov.iov_base = NULL;
|
|
|
|
|
|
|
|
opal_convertor_pack(convertor, &iov, &iov_count, &max_data );
|
|
|
|
|
|
|
|
frag->segments[0].base.seg_len = max_data;
|
|
|
|
frag->segments[0].base.seg_addr.pval = iov.iov_base;
|
2014-01-09 20:01:42 +00:00
|
|
|
frag->segments[0].key = OPAL_THREAD_ADD64(&(portals4_btl->portals_rdma_key), 1);
|
2014-07-10 16:31:15 +00:00
|
|
|
frag->base.des_local_count = 1;
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
/* either a put or get. figure out which later */
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2013-07-11 20:47:08 +00:00
|
|
|
"rdma src posted for frag 0x%lx, callback 0x%lx, bits %"PRIu64", flags say %d" ,
|
|
|
|
(unsigned long) frag,
|
|
|
|
(unsigned long) frag->base.des_cbfunc,
|
|
|
|
frag->segments[0].key, flags));
|
|
|
|
|
|
|
|
/* create a match entry */
|
|
|
|
me.start = frag->segments[0].base.seg_addr.pval;
|
|
|
|
me.length = frag->segments[0].base.seg_len;
|
|
|
|
me.ct_handle = PTL_CT_NONE;
|
|
|
|
me.min_free = 0;
|
|
|
|
me.uid = PTL_UID_ANY;
|
|
|
|
me.options = PTL_ME_OP_GET | PTL_ME_USE_ONCE |
|
|
|
|
PTL_ME_EVENT_LINK_DISABLE |
|
|
|
|
PTL_ME_EVENT_COMM_DISABLE |
|
|
|
|
PTL_ME_EVENT_UNLINK_DISABLE;
|
|
|
|
me.match_id.phys.nid = peer->ptl_proc.phys.nid;
|
|
|
|
me.match_id.phys.pid = peer->ptl_proc.phys.pid;
|
|
|
|
me.match_bits = frag->segments[0].key;
|
|
|
|
me.ignore_bits = BTL_PORTALS4_PROTOCOL_MASK |
|
|
|
|
BTL_PORTALS4_CONTEXT_MASK |
|
|
|
|
BTL_PORTALS4_SOURCE_MASK;
|
|
|
|
me.ignore_bits = 0;
|
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
ret = PtlMEAppend(portals4_btl->portals_ni_h,
|
|
|
|
portals4_btl->recv_idx,
|
2013-07-11 20:47:08 +00:00
|
|
|
&me,
|
|
|
|
PTL_PRIORITY_LIST,
|
|
|
|
frag,
|
|
|
|
&(frag->me_h));
|
|
|
|
if (PTL_OK != ret) {
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
opal_output_verbose(1, opal_btl_base_framework.framework_output,
|
2013-07-11 20:47:08 +00:00
|
|
|
"%s:%d: PtlMEAppend failed: %d\n",
|
|
|
|
__FILE__, __LINE__, ret);
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_RETURN_USER(portals4_btl, frag);
|
2014-01-09 20:01:42 +00:00
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, -1);
|
2013-07-11 20:47:08 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2013-07-11 20:47:08 +00:00
|
|
|
"PtlMEAppend (prepare_src) frag=%p, me_h=%d start=%p length=%ld nid=%x pid=%x match_bits=%lx\n",
|
|
|
|
(void *)frag, frag->me_h, me.start, me.length,
|
|
|
|
me.match_id.phys.nid, me.match_id.phys.pid, me.match_bits));
|
|
|
|
}
|
2014-07-10 16:31:15 +00:00
|
|
|
frag->base.des_local = &frag->segments[0].base;
|
|
|
|
frag->base.des_remote = NULL;
|
|
|
|
frag->base.des_remote_count = 0;
|
2013-07-11 20:47:08 +00:00
|
|
|
frag->base.des_flags = flags | MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
|
|
|
|
frag->base.order = MCA_BTL_NO_ORDER;
|
|
|
|
return &frag->base;
|
|
|
|
}
|
|
|
|
|
|
|
|
mca_btl_base_descriptor_t*
|
|
|
|
mca_btl_portals4_prepare_dst(struct mca_btl_base_module_t* btl_base,
|
|
|
|
struct mca_btl_base_endpoint_t* peer,
|
|
|
|
mca_mpool_base_registration_t* registration,
|
|
|
|
struct opal_convertor_t* convertor,
|
|
|
|
uint8_t order,
|
|
|
|
size_t reserve,
|
|
|
|
size_t* size,
|
|
|
|
uint32_t flags)
|
|
|
|
{
|
2014-01-09 20:01:42 +00:00
|
|
|
struct mca_btl_portals4_module_t* portals4_btl = (struct mca_btl_portals4_module_t*) btl_base;
|
2013-07-11 20:47:08 +00:00
|
|
|
mca_btl_portals4_frag_t* frag;
|
|
|
|
|
|
|
|
/* reserve space in the event queue for rdma operations immediately */
|
2014-01-09 20:01:42 +00:00
|
|
|
while (OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, 1) >
|
|
|
|
portals4_btl->portals_max_outstanding_ops) {
|
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, -1);
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "Call to mca_btl_portals4_component_progress (2)\n"));
|
2013-07-11 20:47:08 +00:00
|
|
|
mca_btl_portals4_component_progress();
|
|
|
|
}
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_BTL_PORTALS4_FRAG_ALLOC_USER(portals4_btl, frag);
|
2013-07-11 20:47:08 +00:00
|
|
|
if (NULL == frag) {
|
2014-01-09 20:01:42 +00:00
|
|
|
OPAL_THREAD_ADD32(&portals4_btl->portals_outstanding_ops, -1);
|
2013-07-11 20:47:08 +00:00
|
|
|
return NULL;
|
|
|
|
}
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"mca_btl_portals4_prepare_dst: Incrementing portals_outstanding_ops=%d\n", portals4_btl->portals_outstanding_ops));
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
frag->segments[0].base.seg_len = *size;
|
|
|
|
opal_convertor_get_current_pointer( convertor, (void**)&(frag->segments[0].base.seg_addr.pval) );
|
2014-01-09 20:01:42 +00:00
|
|
|
frag->segments[0].key = OPAL_THREAD_ADD64(&(portals4_btl->portals_rdma_key), 1);
|
2014-07-10 16:31:15 +00:00
|
|
|
frag->base.des_remote = NULL;
|
|
|
|
frag->base.des_remote_count = 0;
|
|
|
|
frag->base.des_local = &frag->segments[0].base;
|
|
|
|
frag->base.des_local_count = 1;
|
2013-07-11 20:47:08 +00:00
|
|
|
frag->base.des_flags = flags | MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
|
|
|
|
frag->base.order = MCA_BTL_NO_ORDER;
|
|
|
|
frag->md_h = PTL_INVALID_HANDLE;
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2013-07-11 20:47:08 +00:00
|
|
|
"mca_btl_portals4_prepare_dst &base=%p reserve=%ld size=%ld pid=%x key=%ld\n",
|
|
|
|
(void *)&frag->base, reserve, *size, peer->ptl_proc.phys.pid, frag->segments[0].key));
|
|
|
|
return &frag->base;
|
|
|
|
}
|
|
|
|
|
|
|
|
int
|
|
|
|
mca_btl_portals4_finalize(struct mca_btl_base_module_t *btl)
|
|
|
|
{
|
2014-01-09 20:01:42 +00:00
|
|
|
struct mca_btl_portals4_module_t* portals4_btl = (struct mca_btl_portals4_module_t*) btl;
|
|
|
|
|
|
|
|
mca_btl_portals4_free_module(portals4_btl);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
OBJ_DESTRUCT(&portals4_btl->portals_frag_eager);
|
|
|
|
OBJ_DESTRUCT(&portals4_btl->portals_frag_max);
|
|
|
|
OBJ_DESTRUCT(&portals4_btl->portals_frag_user);
|
|
|
|
OBJ_DESTRUCT(&portals4_btl->portals_recv_blocks);
|
|
|
|
|
|
|
|
free(portals4_btl);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"mca_btl_portals4_finalize NI %d: OK\n", portals4_btl->interface_num));
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
return OPAL_SUCCESS;
|
2014-01-09 20:01:42 +00:00
|
|
|
}
|
|
|
|
|
|
|
|
void mca_btl_portals4_free_module(mca_btl_portals4_module_t *portals4_btl)
|
|
|
|
{
|
|
|
|
int ret;
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"mca_btl_portals4_free_module portals_outstanding_ops=%d\n", portals4_btl->portals_outstanding_ops));
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
/* sanity check */
|
2014-01-09 20:01:42 +00:00
|
|
|
assert(portals4_btl->portals_outstanding_ops >= 0);
|
2013-07-11 20:47:08 +00:00
|
|
|
|
|
|
|
/* finalize all communication */
|
2014-01-09 20:01:42 +00:00
|
|
|
while (portals4_btl->portals_outstanding_ops > 0) {
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output,
|
2014-01-09 20:01:42 +00:00
|
|
|
"mca_btl_portals4_free_module portals_outstanding_ops: %d",
|
|
|
|
portals4_btl->portals_outstanding_ops));
|
2013-07-11 20:47:08 +00:00
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "Call to mca_btl_portals4_component_progress (3)\n"));
|
2013-07-11 20:47:08 +00:00
|
|
|
mca_btl_portals4_component_progress();
|
|
|
|
}
|
|
|
|
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
#if OPAL_PORTALS4_MAX_MD_SIZE < OPAL_PORTALS4_MAX_VA_SIZE
|
2014-01-09 20:01:42 +00:00
|
|
|
if (NULL != portals4_btl->send_md_hs) {
|
2013-07-11 21:00:11 +00:00
|
|
|
int i;
|
2013-08-05 18:49:48 +00:00
|
|
|
int num_mds = mca_btl_portals4_get_num_mds();
|
2013-07-11 21:00:11 +00:00
|
|
|
|
|
|
|
for (i = 0 ; i < num_mds ; ++i) {
|
2014-01-09 20:01:42 +00:00
|
|
|
if (!PtlHandleIsEqual(portals4_btl->send_md_hs[i], PTL_INVALID_HANDLE)) {
|
|
|
|
PtlMDRelease(portals4_btl->send_md_hs[i]);
|
|
|
|
portals4_btl->send_md_hs[i] = PTL_INVALID_HANDLE;
|
2013-07-11 21:00:11 +00:00
|
|
|
}
|
|
|
|
}
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
free(portals4_btl->send_md_hs);
|
|
|
|
portals4_btl->send_md_hs = NULL;
|
2013-07-11 21:00:11 +00:00
|
|
|
}
|
|
|
|
#else
|
2014-01-09 20:01:42 +00:00
|
|
|
if (!PtlHandleIsEqual(portals4_btl->send_md_h, PTL_INVALID_HANDLE)) {
|
|
|
|
PtlMDRelease(portals4_btl->send_md_h);
|
|
|
|
portals4_btl->send_md_h = PTL_INVALID_HANDLE;
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|
2013-07-11 21:00:11 +00:00
|
|
|
#endif
|
2014-01-09 20:01:42 +00:00
|
|
|
if (!PtlHandleIsEqual(portals4_btl->zero_md_h, PTL_INVALID_HANDLE)) {
|
|
|
|
PtlMDRelease(portals4_btl->zero_md_h);
|
|
|
|
portals4_btl->zero_md_h = PTL_INVALID_HANDLE;
|
|
|
|
}
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
if (!PtlHandleIsEqual(portals4_btl->long_overflow_me_h, PTL_INVALID_HANDLE)) {
|
|
|
|
PtlMEUnlink(portals4_btl->long_overflow_me_h);
|
|
|
|
portals4_btl->long_overflow_me_h = PTL_INVALID_HANDLE;
|
|
|
|
}
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
if ((ptl_pt_index_t) ~0UL != mca_btl_portals4_module.recv_idx) {
|
|
|
|
PtlPTFree(portals4_btl->portals_ni_h, portals4_btl->recv_idx);
|
|
|
|
portals4_btl->recv_idx= (ptl_pt_index_t) ~0UL;
|
|
|
|
}
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
if (PTL_EQ_NONE != portals4_btl->recv_eq_h) {
|
|
|
|
ret = PtlEQFree(portals4_btl->recv_eq_h);
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
if (PTL_OK != ret) OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "Error freeing EQ recv: %d", ret));
|
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "PtlEQFree: recv_eq_h=%d portals4_btl=%p",
|
2014-01-09 20:01:42 +00:00
|
|
|
portals4_btl->recv_eq_h, (void*)portals4_btl));
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
portals4_btl->recv_eq_h = PTL_EQ_NONE;
|
|
|
|
}
|
|
|
|
if (!PtlHandleIsEqual(portals4_btl->portals_ni_h, PTL_INVALID_HANDLE)) {
|
|
|
|
ret = PtlNIFini(portals4_btl->portals_ni_h);
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
if (PTL_OK != ret) OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "Error returned by PtlNIFini: %d\n", ret));
|
|
|
|
OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "PtlNIFini: portals_ni_h=%d portals4_btl=%p",
|
2014-01-09 20:01:42 +00:00
|
|
|
portals4_btl->portals_ni_h, (void*)portals4_btl));
|
2013-07-11 20:47:08 +00:00
|
|
|
|
2014-01-09 20:01:42 +00:00
|
|
|
portals4_btl->portals_ni_h = PTL_INVALID_HANDLE;
|
|
|
|
}
|
|
|
|
ret = mca_btl_portals4_recv_disable(portals4_btl);
|
George did the work and deserves all the credit for it. Ralph did the merge, and deserves whatever blame results from errors in it :-)
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
2014-07-26 00:47:28 +00:00
|
|
|
if (PTL_OK != ret) OPAL_OUTPUT_VERBOSE((90, opal_btl_base_framework.framework_output, "Error freeing recv list: %d", ret));
|
2013-07-11 20:47:08 +00:00
|
|
|
}
|