1
1
openmpi/opal/mca/btl/ugni/btl_ugni_rdma.h

150 строки
7.0 KiB
C
Исходник Обычный вид История

/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
/*
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
* Copyright (c) 2011-2018 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2011 UT-Battelle, LLC. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#if !defined(MCA_BTL_UGNI_RDMA_H)
#define MCA_BTL_UGNI_RDMA_H
#include "btl_ugni.h"
#include "btl_ugni_frag.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 "btl_ugni_device.h"
int mca_btl_ugni_start_eager_get (mca_btl_base_endpoint_t *ep,
mca_btl_ugni_eager_ex_frag_hdr_t hdr,
mca_btl_ugni_base_frag_t *frag);
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
static inline void init_post_desc (mca_btl_ugni_post_descriptor_t *post_desc,
mca_btl_base_endpoint_t *endpoint, int order,
gni_post_type_t op_type, uint64_t lcl_addr,
gni_mem_handle_t lcl_mdh, uint64_t rem_addr,
gni_mem_handle_t rem_mdh, uint64_t bufsize,
gni_cq_handle_t cq_hndl, mca_btl_base_rdma_completion_fn_t cbfunc,
void *cbcontext, void *cbdata,
mca_btl_base_registration_handle_t *local_handle) {
post_desc->endpoint = endpoint;
post_desc->cbfunc = cbfunc;
post_desc->ctx = cbcontext;
post_desc->cbdata = cbdata;
post_desc->local_handle = local_handle;
post_desc->gni_desc.type = op_type;
post_desc->gni_desc.cq_mode = GNI_CQMODE_GLOBAL_EVENT;
if (MCA_BTL_NO_ORDER == order) {
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
post_desc->gni_desc.dlvr_mode = GNI_DLVMODE_PERFORMANCE;
} else {
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
post_desc->gni_desc.dlvr_mode = GNI_DLVMODE_NO_ADAPT;
}
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
post_desc->gni_desc.local_addr = (uint64_t) lcl_addr;
post_desc->gni_desc.local_mem_hndl = lcl_mdh;
post_desc->gni_desc.remote_addr = (uint64_t) rem_addr;
post_desc->gni_desc.remote_mem_hndl = rem_mdh;
post_desc->gni_desc.length = bufsize;
post_desc->gni_desc.rdma_mode = 0;
post_desc->gni_desc.src_cq_hndl = cq_hndl;
}
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
__opal_attribute_always_inline__
static inline int mca_btl_ugni_post_fma (struct mca_btl_base_endpoint_t *endpoint, gni_post_type_t op_type,
size_t size, void *local_address, uint64_t remote_address,
mca_btl_base_registration_handle_t *local_handle,
mca_btl_base_registration_handle_t *remote_handle,
int order, mca_btl_base_rdma_completion_fn_t cbfunc,
void *cbcontext, void *cbdata)
{
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
mca_btl_ugni_post_descriptor_t post_desc;
gni_mem_handle_t local_gni_handle = {0, 0};
if (local_handle) {
local_gni_handle = local_handle->gni_handle;
}
/* Post descriptor (CQ is ignored for FMA transactions) -- The CQ associated with the endpoint
* is used. */
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
init_post_desc (&post_desc, endpoint, order, op_type, (intptr_t) local_address, local_gni_handle,
remote_address, remote_handle->gni_handle, size, 0, cbfunc, cbcontext, cbdata,
local_handle);
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
return mca_btl_ugni_endpoint_post_fma (endpoint, &post_desc);
}
static inline int mca_btl_ugni_post_bte (mca_btl_base_endpoint_t *endpoint, gni_post_type_t op_type,
size_t size, void *local_address, uint64_t remote_address,
mca_btl_base_registration_handle_t *local_handle,
mca_btl_base_registration_handle_t *remote_handle,
int order, mca_btl_base_rdma_completion_fn_t cbfunc,
void *cbcontext, void *cbdata)
{
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
mca_btl_ugni_module_t *module = mca_btl_ugni_ep_btl (endpoint);
mca_btl_ugni_post_descriptor_t post_desc;
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 rc;
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
/* There is a performance benefit to throttling the total number of active BTE tranactions. Not sure
* what the optimium is but the limit is inforced as a soft limit. */
if (module->active_rdma_count >= mca_btl_ugni_component.active_rdma_threshold) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
(void) OPAL_THREAD_FETCH_ADD32 (&module->active_rdma_count, 1);
/* Post descriptor */
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
init_post_desc (&post_desc, endpoint, order, op_type, (intptr_t) local_address, local_handle->gni_handle,
remote_address, remote_handle->gni_handle, size, 0, cbfunc, cbcontext, cbdata,
local_handle);
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
rc = mca_btl_ugni_endpoint_post_rdma (endpoint, &post_desc);
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 (OPAL_UNLIKELY(OPAL_SUCCESS != rc)) {
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
(void) OPAL_THREAD_FETCH_ADD32 (&module->active_rdma_count, -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 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
static inline int mca_btl_ugni_post_cqwrite (mca_btl_base_endpoint_t *endpoint, mca_btl_ugni_cq_t *cq,
gni_mem_handle_t irq_mhndl, uint64_t value,
mca_btl_base_rdma_completion_fn_t cbfunc,
void *cbcontext, void *cbdata)
{
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
mca_btl_ugni_post_descriptor_t post_desc;
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
post_desc.gni_desc.type = GNI_POST_CQWRITE;
post_desc.gni_desc.cqwrite_value = value; /* up to 48 bytes here, not used for now */
post_desc.gni_desc.cq_mode = GNI_CQMODE_GLOBAL_EVENT;
post_desc.gni_desc.dlvr_mode = GNI_DLVMODE_IN_ORDER;
post_desc.gni_desc.src_cq_hndl = cq->gni_handle;
post_desc.gni_desc.remote_mem_hndl = irq_mhndl;
post_desc.cq = cq;
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
return mca_btl_ugni_endpoint_post_cqwrite (endpoint, &post_desc);
}
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
__opal_attribute_always_inline__
static inline int mca_btl_ugni_post (mca_btl_base_endpoint_t *endpoint, int get, size_t size,
void *local_address, uint64_t remote_address,
mca_btl_base_registration_handle_t *local_handle,
mca_btl_base_registration_handle_t *remote_handle,
int order, mca_btl_base_rdma_completion_fn_t cbfunc,
void *cbcontext, void *cbdata)
{
const gni_post_type_t fma_ops[2] = {GNI_POST_FMA_PUT, GNI_POST_FMA_GET};
const gni_post_type_t rdma_ops[2] = {GNI_POST_RDMA_PUT, GNI_POST_RDMA_GET};
const size_t fma_limit = (size_t) (get ? mca_btl_ugni_component.ugni_fma_get_limit :
mca_btl_ugni_component.ugni_fma_put_limit);
btl/ugni: improve multi-threaded RDMA performance This commit improves the injection rate and latency for RDMA operations. This is done by the following improvements: - If C11's _Thread_local keyword is available then always use the same virtual device index for the same thread when using RDMA. If the keyword is not available then attempt to use any device that isn't already in use. The binding support is enabled by default but can be disabled via the btl_ugni_bind_devices MCA variable. - When posting FMA and RDMA operations always attempt to reap completions after posting the operation. This allows us to better balance the work of reaping completions across all application threads. - Limit the total number of outstanding BTE transactions. This fixes a performance bug when using many threads. - Split out RDMA and local SMSG completion queue sizes. The RDMA queue size is better tuned for performance with RMA-MT. - Split out put and get FMA limits. The old btl_ugni_fma_limit MCA variable is deprecated. The new variable names are: btl_ugni_fma_put_limit and btl_ugni_fma_get_limit. - Change how post descriptors are handled. They are no longer allocated seperately from the RDMA endpoints. - Some cleanup to move error code out of the critical path. - Disable the FMA sharing flag on the CDM when we detect that there should be enough FMA descriptors for the number of virtual devices we plan will create. If the user sets this flag we will not unset it. This change should improve the small-message RMA performance by ~ 10%. Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
2018-02-15 12:57:21 -07:00
if (size <= fma_limit) {
return mca_btl_ugni_post_fma (endpoint, fma_ops[get], size, local_address, remote_address,
local_handle, remote_handle, order, cbfunc, cbcontext, cbdata);
}
return mca_btl_ugni_post_bte (endpoint, rdma_ops[get], size, local_address, remote_address,
local_handle, remote_handle, order, cbfunc, cbcontext, cbdata);
}
#endif /* MCA_BTL_UGNI_RDMA_H */