1
1
openmpi/opal/mca/btl/ugni/btl_ugni_endpoint.c

448 строки
15 KiB
C
Исходник Обычный вид История

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
* Copyright (c) 2011-2017 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2011-2013 UT-Battelle, LLC. All rights reserved.
* Copyright (c) 2017 Intel, Inc. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "btl_ugni_endpoint.h"
#include "btl_ugni_smsg.h"
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
#include "opal/mca/pmix/pmix.h"
static void mca_btl_ugni_ep_construct (mca_btl_base_endpoint_t *ep)
{
memset ((char *) ep + sizeof(ep->super), 0, sizeof (*ep) - sizeof (ep->super));
OBJ_CONSTRUCT(&ep->frag_wait_list, opal_list_t);
OBJ_CONSTRUCT(&ep->lock, opal_recursive_mutex_t);
}
static void mca_btl_ugni_ep_destruct (mca_btl_base_endpoint_t *ep)
{
OBJ_DESTRUCT(&ep->frag_wait_list);
OBJ_DESTRUCT(&ep->lock);
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
free (ep->remote_attr);
}
OBJ_CLASS_INSTANCE(mca_btl_ugni_endpoint_t, opal_list_item_t,
mca_btl_ugni_ep_construct, mca_btl_ugni_ep_destruct);
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
static int mca_btl_ugni_endpoint_get_modex (mca_btl_base_endpoint_t *ep)
{
mca_btl_ugni_modex_t *modex;
size_t msg_size;
int rc;
assert (NULL != ep && NULL != ep->peer_proc);
/* Receive the modex */
OPAL_MODEX_RECV(rc, &mca_btl_ugni_component.super.btl_version,
&ep->peer_proc->proc_name, (void **)&modex, &msg_size);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
BTL_ERROR(("error receiving modex"));
return rc;
}
ep->ep_rem_addr = modex->addr;
ep->ep_rem_id = modex->id;
BTL_VERBOSE(("received modex for ep %p. addr: %d, id: %d", (void*)ep, ep->ep_rem_addr, ep->ep_rem_id));
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
free (modex);
return OPAL_SUCCESS;
}
int mca_btl_ugni_init_ep (mca_btl_ugni_module_t *ugni_module, mca_btl_ugni_endpoint_t **ep,
mca_btl_ugni_module_t *btl, opal_proc_t *peer_proc)
{
mca_btl_ugni_endpoint_t *endpoint;
int rc;
endpoint = OBJ_NEW(mca_btl_ugni_endpoint_t);
assert (endpoint != NULL);
endpoint->smsg_progressing = 0;
endpoint->state = MCA_BTL_UGNI_EP_STATE_INIT;
endpoint->peer_proc = peer_proc;
/* get the modex info for this endpoint and setup a ugni endpoint. this call may lead
* to re-entry through opal_progress(). */
rc = mca_btl_ugni_endpoint_get_modex (endpoint);
if (OPAL_SUCCESS != rc) {
assert (0);
return rc;
}
/* add this endpoint to the pointer array */
endpoint->index = opal_pointer_array_add (&ugni_module->endpoints, endpoint);
*ep = endpoint;
return OPAL_SUCCESS;
}
void mca_btl_ugni_release_ep (mca_btl_ugni_endpoint_t *ep)
{
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
int rc;
opal_mutex_lock (&ep->lock);
rc = mca_btl_ugni_ep_disconnect (ep, false);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
BTL_VERBOSE(("btl/ugni error disconnecting endpoint"));
}
/* TODO -- Clear space at the end of the endpoint array */
opal_pointer_array_set_item (&ugni_module->endpoints, ep->index, NULL);
opal_mutex_unlock (&ep->lock);
OBJ_RELEASE(ep);
}
static inline int mca_btl_ugni_ep_smsg_get_mbox (mca_btl_base_endpoint_t *ep) {
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
opal_free_list_item_t *mbox;
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
assert (NULL == ep->mailbox);
mbox = opal_free_list_get (&ugni_module->smsg_mboxes);
if (OPAL_UNLIKELY(NULL == mbox)) {
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_OUT_OF_RESOURCE;
}
ep->mailbox = (mca_btl_ugni_smsg_mbox_t *) mbox;
ep->mailbox->attr.index = ep->index;
/* per ugni spec we need to zero mailbox data before connecting */
memset ((char *)ep->mailbox->attr.smsg_attr.msg_buffer + ep->mailbox->attr.smsg_attr.mbox_offset, 0,
ep->mailbox->attr.smsg_attr.buff_size);
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;
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
static int mca_btl_ugni_ep_send_disconnect (mca_btl_base_endpoint_t *ep)
{
int rc;
do {
rc = mca_btl_ugni_endpoint_smsg_send_wtag (ep, NULL, 0, NULL, 0, -1, MCA_BTL_UGNI_TAG_DISCONNECT);
if (OPAL_LIKELY(GNI_RC_NOT_DONE != rc)) {
break;
}
/* most likely got here because we are out of credits. check the remote CQ to get credit return */
(void) mca_btl_ugni_progress_remote_smsg (mca_btl_ugni_ep_btl (ep));
} while (1);
return mca_btl_rc_ugni_to_opal (rc);
}
int mca_btl_ugni_ep_disconnect (mca_btl_base_endpoint_t *ep, bool send_disconnect)
{
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
mca_btl_ugni_device_t *device;
int rc;
if (MCA_BTL_UGNI_EP_STATE_INIT == ep->state) {
/* nothing to do */
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;
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
device = ep->smsg_ep_handle->device;
while (device->dev_smsg_local_cq.active_operations) {
/* ensure all sends are complete before removing and procs */
rc = mca_btl_ugni_progress_local_smsg (ugni_module, device);
if (OPAL_SUCCESS != rc) {
break;
}
}
if (MCA_BTL_UGNI_EP_STATE_CONNECTED == ep->state && send_disconnect) {
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
rc = mca_btl_ugni_ep_send_disconnect (ep);
if (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
BTL_VERBOSE(("could not send disconnect message to peer"));
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
/* wait for the disconnect messagse to go */
do {
/* ensure all sends are complete before removing and procs */
rc = mca_btl_ugni_progress_local_smsg (ugni_module, device);
if (OPAL_SUCCESS != rc) {
break;
}
} while (device->dev_smsg_local_cq.active_operations);
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
(void) opal_atomic_add_32 (&ep->smsg_ep_handle->device->smsg_connections, -1);
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
mca_btl_ugni_device_lock (device);
/* NTH: this call may not need the device lock. seems to work without it but
* the lock is here to be safe. */
(void) mca_btl_ugni_ep_handle_destroy (ep->smsg_ep_handle);
ep->smsg_ep_handle = NULL;
mca_btl_ugni_device_unlock (device);
if (ep->mailbox) {
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
opal_free_list_return (&ugni_module->smsg_mboxes, ((opal_free_list_item_t *) ep->mailbox));
ep->mailbox = NULL;
}
ep->state = MCA_BTL_UGNI_EP_STATE_INIT;
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;
}
static inline int mca_btl_ugni_ep_connect_start (mca_btl_base_endpoint_t *ep) {
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
mca_btl_ugni_device_t *device = ugni_module->devices;
int rc;
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
/* protect against re-entry from opal_progress */
if (OPAL_UNLIKELY(MCA_BTL_UGNI_EP_STATE_CONNECTING == ep->state)) {
return OPAL_ERR_RESOURCE_BUSY;
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
ep->state = MCA_BTL_UGNI_EP_STATE_CONNECTING;
BTL_VERBOSE(("initiating connection to remote peer with address: %u id: %u proc: %p",
ep->ep_rem_addr, ep->ep_rem_id, (void *)ep->peer_proc));
2014-11-13 08:26:25 -07:00
/* bind endpoint to remote address */
/* we bind two endpoints to seperate out local smsg completion and local fma completion */
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
mca_btl_ugni_device_lock (device);
ep->smsg_ep_handle = mca_btl_ugni_ep_handle_create (ep, device->dev_smsg_local_cq.gni_handle, device);
mca_btl_ugni_device_unlock (device);
if (OPAL_UNLIKELY(NULL == ep->smsg_ep_handle)) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
/* build connection data */
rc = mca_btl_ugni_ep_smsg_get_mbox (ep);
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_UNLIKELY(OPAL_SUCCESS != rc)) {
return rc;
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
ep->remote_attr = calloc (1, sizeof (*ep->remote_attr));
if (OPAL_UNLIKELY(NULL == ep->remote_attr)) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
BTL_VERBOSE(("btl/ugni connection to remote peer initiated"));
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;
}
static inline int mca_btl_ugni_ep_connect_finish (mca_btl_base_endpoint_t *ep) {
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
gni_return_t grc;
int rc;
BTL_VERBOSE(("finishing connection. remote attributes: msg_type = %d, msg_buffer = %p, buff_size = %d, "
"mem_hndl = {qword1 = %" PRIu64 ", qword2 = %" PRIu64 "}, mbox = %d, mbox_maxcredit = %d, "
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
"msg_maxsize = %d", ep->remote_attr->smsg_attr.msg_type, ep->remote_attr->smsg_attr.msg_buffer,
ep->remote_attr->smsg_attr.buff_size, ep->remote_attr->smsg_attr.mem_hndl.qword1,
ep->remote_attr->smsg_attr.mem_hndl.qword2, ep->remote_attr->smsg_attr.mbox_offset,
ep->remote_attr->smsg_attr.mbox_maxcredit, ep->remote_attr->smsg_attr.msg_maxsize));
BTL_VERBOSE(("finishing connection. local attributes: msg_type = %d, msg_buffer = %p, buff_size = %d, "
"mem_hndl = {qword1 = %" PRIu64 ", qword2 = %" PRIu64 "}, mbox = %d, mbox_maxcredit = %d, "
"msg_maxsize = %d", ep->mailbox->attr.smsg_attr.msg_type, ep->mailbox->attr.smsg_attr.msg_buffer,
ep->mailbox->attr.smsg_attr.buff_size, ep->mailbox->attr.smsg_attr.mem_hndl.qword1,
ep->mailbox->attr.smsg_attr.mem_hndl.qword2, ep->mailbox->attr.smsg_attr.mbox_offset,
ep->mailbox->attr.smsg_attr.mbox_maxcredit, ep->mailbox->attr.smsg_attr.msg_maxsize));
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
grc = GNI_SmsgInit (ep->smsg_ep_handle->gni_handle, &ep->mailbox->attr.smsg_attr,
&ep->remote_attr->smsg_attr);
if (OPAL_UNLIKELY(GNI_RC_SUCCESS != grc)) {
BTL_ERROR(("error initializing SMSG protocol. rc = %d", grc));
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
return mca_btl_rc_ugni_to_opal (grc);
}
/* set the local event data to the local index and the remote event data to my
* index on the remote peer. This makes lookup of endpoints on completion take
* a single lookup in the endpoints array. we will not be able to change the
* remote peer's index in the endpoint's array after this point. */
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
GNI_EpSetEventData (ep->smsg_ep_handle->gni_handle, ep->index, ep->remote_attr->index);
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
ep->rmt_irq_mem_hndl = ep->remote_attr->rmt_irq_mem_hndl;
ep->state = MCA_BTL_UGNI_EP_STATE_CONNECTED;
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
(void) opal_atomic_add_32 (&ep->smsg_ep_handle->device->smsg_connections, 1);
/* send all pending messages */
BTL_VERBOSE(("endpoint connected. posting %u sends", (unsigned int) opal_list_get_size (&ep->frag_wait_list)));
rc = mca_btl_ugni_progress_send_wait_list (ep);
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_UNLIKELY(OPAL_SUCCESS != rc)) {
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
OPAL_THREAD_LOCK(&ugni_module->ep_wait_list_lock);
if (false == ep->wait_listed) {
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
opal_list_append (&ugni_module->ep_wait_list, &ep->super);
ep->wait_listed = true;
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
OPAL_THREAD_UNLOCK(&ugni_module->ep_wait_list_lock);
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
free (ep->remote_attr);
ep->remote_attr = 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
return OPAL_SUCCESS;
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
static int mca_btl_ugni_directed_ep_post (mca_btl_base_endpoint_t *ep)
{
mca_btl_ugni_module_t *ugni_module = mca_btl_ugni_ep_btl (ep);
gni_return_t rc;
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
BTL_VERBOSE(("posting directed datagram to remote id: %d for endpoint %p", ep->ep_rem_id, (void *)ep));
/* the irq cq is associated with only the first device */
ep->mailbox->attr.rmt_irq_mem_hndl = ugni_module->devices->smsg_irq_mhndl;
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
rc = GNI_EpPostDataWId (ep->smsg_ep_handle->gni_handle, &ep->mailbox->attr, sizeof (ep->mailbox->attr),
ep->remote_attr, sizeof (*ep->remote_attr),
MCA_BTL_UGNI_CONNECT_DIRECTED_ID | ep->index);
if (OPAL_LIKELY(GNI_RC_SUCCESS == rc)) {
(void) opal_atomic_add_32 (&ugni_module->active_datagrams, 1);
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
return mca_btl_rc_ugni_to_opal (rc);
}
int mca_btl_ugni_wildcard_ep_post (mca_btl_ugni_module_t *ugni_module)
{
gni_return_t rc;
BTL_VERBOSE(("posting wildcard datagram"));
memset (&ugni_module->wc_local_attr, 0, sizeof (ugni_module->wc_local_attr));
memset (&ugni_module->wc_remote_attr, 0, sizeof (ugni_module->wc_remote_attr));
rc = GNI_EpPostDataWId (ugni_module->wildcard_ep, &ugni_module->wc_local_attr,
sizeof (ugni_module->wc_local_attr), &ugni_module->wc_remote_attr,
sizeof (ugni_module->wc_remote_attr), MCA_BTL_UGNI_CONNECT_WILDCARD_ID);
return mca_btl_rc_ugni_to_opal (rc);
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
int mca_btl_ugni_ep_connect_progress (mca_btl_base_endpoint_t *ep)
{
int rc;
BTL_VERBOSE(("progressing connection for endpoint %p with state %d", (void *)ep, ep->state));
if (MCA_BTL_UGNI_EP_STATE_CONNECTED == ep->state) {
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;
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
if (MCA_BTL_UGNI_EP_STATE_INIT == ep->state) {
rc = mca_btl_ugni_ep_connect_start (ep);
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 != rc) {
return rc;
}
}
BTL_VERBOSE(("ep->remote_attr->smsg_attr = {.msg_type = %d, .msg_buffer = %p}", ep->remote_attr->smsg_attr.msg_type,
(void*)ep->remote_attr->smsg_attr.msg_buffer));
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
if (GNI_SMSG_TYPE_INVALID == ep->remote_attr->smsg_attr.msg_type) {
/* use datagram to exchange connection information with the remote peer */
if (!ep->dg_posted) {
rc = mca_btl_ugni_directed_ep_post (ep);
if (OPAL_SUCCESS == rc) {
ep->dg_posted = true;
rc = OPAL_ERR_RESOURCE_BUSY;
}
return rc;
}
return OPAL_SUCCESS;
}
return mca_btl_ugni_ep_connect_finish (ep);
}
btl/ugni: improve multi-threaded performance This commit updates the ugni btl to make use of multiple device contexts to improve the multi-threaded RMA performance. This commit contains the following: - Cleanup the endpoint structure by removing unnecessary field. The structure now also contains all the fields originally handled by the common/ugni endpoint. - Clean up the fragment allocation code to remove the need to initialize the my_list member of the fragment structure. This member is not initialized by the free list initializer function. - Remove the (now unused) common/ugni component. btl/ugni no longer need the component. common/ugni was originally split out of btl/ugni to support bcol/ugni. As that component exists there is no reason to keep this component. - Create wrappers for the ugni functionality required by btl/ugni. This was done to ease supporting multiple device contexts. The wrappers are thread safe and currently use a spin lock instead of a mutex. This produces better performance when using multiple threads spread over multiple cores. In the future this lock may be replaced by another serialization mechanism. The wrappers are located in a new file: btl_ugni_device.h. - Remove unnecessary device locking from serial parts of the ugni btl. This includes the first add-procs and module finalize. - Clean up fragment wait list code by moving enqueue into common function. - Expose the communication domain flags as an MCA variable. The defaults have been updated to reflect the recommended setting for knl and haswell. - Avoid allocating fragments for communication with already overloaded peers. - Allocate RDMA endpoints dyncamically. This is needed to support spreading RMA operations accross multiple contexts. - Add support for spreading RMA communication over multiple ugni device contexts. This should greatly improve the threading performance when communicating with multiple peers. By default the number of virtual devices depends on 1) whether opal_using_threads() is set, 2) how many local processes are in the job, and 3) how many bits are available in the pid. The last is used to ensure that each CDM is created with a unique id. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2017-03-12 22:37:35 -06:00
int mca_btl_ugni_endpoint_handle_init_rdma (opal_free_list_item_t *item, void *ctx)
{
mca_btl_ugni_endpoint_handle_t *handle = (mca_btl_ugni_endpoint_handle_t *) item;
mca_btl_ugni_device_t *device = (mca_btl_ugni_device_t *) ctx;
gni_return_t grc;
grc = GNI_EpCreate (device->dev_handle, device->dev_rdma_local_cq.gni_handle, &handle->gni_handle);
handle->device = device;
return mca_btl_rc_ugni_to_opal (grc);
}
static void mca_btl_ugni_endpoint_handle_construct (mca_btl_ugni_endpoint_handle_t *handle)
{
handle->gni_handle = 0;
}
static void mca_btl_ugni_endpoint_handle_destruct (mca_btl_ugni_endpoint_handle_t *handle)
{
if (handle->gni_handle) {
GNI_EpDestroy (handle->gni_handle);
handle->gni_handle = 0;
}
}
OBJ_CLASS_INSTANCE(mca_btl_ugni_endpoint_handle_t, opal_object_t,
mca_btl_ugni_endpoint_handle_construct,
mca_btl_ugni_endpoint_handle_destruct);
mca_btl_ugni_endpoint_handle_t *mca_btl_ugni_ep_handle_create (mca_btl_ugni_endpoint_t *ep, gni_cq_handle_t cq,
mca_btl_ugni_device_t *device)
{
mca_btl_ugni_endpoint_handle_t *ep_handle;
gni_return_t grc;
ep_handle = OBJ_NEW(mca_btl_ugni_endpoint_handle_t);
if (OPAL_UNLIKELY(NULL == ep_handle)) {
return NULL;
}
ep_handle->device = device;
/* create a uGNI endpoint handle and bind it to the remote peer */
grc = GNI_EpCreate (device->dev_handle, cq, &ep_handle->gni_handle);
if (OPAL_LIKELY(GNI_RC_SUCCESS == grc)) {
grc = GNI_EpBind (ep_handle->gni_handle, ep->ep_rem_addr, ep->ep_rem_id);
}
if (GNI_RC_SUCCESS != grc) {
OBJ_RELEASE(ep_handle);
ep_handle = NULL;
}
return ep_handle;
}
int mca_btl_ugni_ep_handle_destroy (mca_btl_ugni_endpoint_handle_t *ep_handle)
{
int rc;
if (NULL == ep_handle || 0 == ep_handle->gni_handle) {
return OPAL_SUCCESS;
}
/* TODO: need to fix, may be outstanding tx's, etc. */
rc = GNI_EpUnbind (ep_handle->gni_handle);
if (OPAL_UNLIKELY(GNI_RC_SUCCESS != rc)) {
/* should warn */
}
OBJ_RELEASE(ep_handle);
return OPAL_SUCCESS;
}