552c9ca5a0
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.
343 строки
13 KiB
C
343 строки
13 KiB
C
/* -*- Mode: C; c-basic-offset:4 ; indent-tabs-mode:nil -*- */
|
|
/*
|
|
* Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
|
|
* University Research and Technology
|
|
* Corporation. All rights reserved.
|
|
* Copyright (c) 2004-2009 The University of Tennessee and The University
|
|
* of Tennessee Research Foundation. All rights
|
|
* reserved.
|
|
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
|
|
* University of Stuttgart. All rights reserved.
|
|
* Copyright (c) 2004-2005 The Regents of the University of California.
|
|
* All rights reserved.
|
|
* Copyright (c) 2007-2010 Cisco Systems, Inc. All rights reserved.
|
|
* Copyright (c) 2010 Oracle and/or its affiliates. All rights reserved
|
|
* Copyright (c) 2013 Los Alamos National Security, LLC. All rights
|
|
* reserved.
|
|
* $COPYRIGHT$
|
|
*
|
|
* Additional copyrights may follow
|
|
*
|
|
* $HEADER$
|
|
*/
|
|
|
|
#include "ompi_config.h"
|
|
#include "opal/mca/event/event.h"
|
|
#include "mpi.h"
|
|
#include "ompi/runtime/params.h"
|
|
#include "ompi/mca/pml/pml.h"
|
|
#include "ompi/mca/pml/base/pml_base_bsend.h"
|
|
#include "pml_ob1.h"
|
|
#include "pml_ob1_hdr.h"
|
|
#include "pml_ob1_sendreq.h"
|
|
#include "pml_ob1_recvreq.h"
|
|
#include "pml_ob1_rdmafrag.h"
|
|
#include "pml_ob1_recvfrag.h"
|
|
#include "ompi/mca/bml/base/base.h"
|
|
#include "pml_ob1_component.h"
|
|
#include "opal/mca/allocator/base/base.h"
|
|
#include "opal/mca/base/mca_base_pvar.h"
|
|
#include "opal/runtime/opal_params.h"
|
|
|
|
OBJ_CLASS_INSTANCE( mca_pml_ob1_pckt_pending_t,
|
|
ompi_free_list_item_t,
|
|
NULL,
|
|
NULL );
|
|
|
|
static int mca_pml_ob1_component_register(void);
|
|
static int mca_pml_ob1_component_open(void);
|
|
static int mca_pml_ob1_component_close(void);
|
|
static mca_pml_base_module_t*
|
|
mca_pml_ob1_component_init( int* priority, bool enable_progress_threads,
|
|
bool enable_mpi_threads );
|
|
static int mca_pml_ob1_component_fini(void);
|
|
int mca_pml_ob1_output = 0;
|
|
static int mca_pml_ob1_verbose = 0;
|
|
|
|
mca_pml_base_component_2_0_0_t mca_pml_ob1_component = {
|
|
|
|
/* First, the mca_base_component_t struct containing meta
|
|
information about the component itself */
|
|
|
|
{
|
|
MCA_PML_BASE_VERSION_2_0_0,
|
|
|
|
"ob1", /* MCA component name */
|
|
OMPI_MAJOR_VERSION, /* MCA component major version */
|
|
OMPI_MINOR_VERSION, /* MCA component minor version */
|
|
OMPI_RELEASE_VERSION, /* MCA component release version */
|
|
mca_pml_ob1_component_open, /* component open */
|
|
mca_pml_ob1_component_close, /* component close */
|
|
NULL,
|
|
mca_pml_ob1_component_register
|
|
},
|
|
{
|
|
/* The component is checkpoint ready */
|
|
MCA_BASE_METADATA_PARAM_CHECKPOINT
|
|
},
|
|
|
|
mca_pml_ob1_component_init, /* component init */
|
|
mca_pml_ob1_component_fini /* component finalize */
|
|
|
|
};
|
|
|
|
void *mca_pml_ob1_seg_alloc( struct mca_mpool_base_module_t* mpool,
|
|
size_t* size,
|
|
mca_mpool_base_registration_t** registration);
|
|
|
|
void mca_pml_ob1_seg_free( struct mca_mpool_base_module_t* mpool,
|
|
void* segment );
|
|
|
|
static inline int mca_pml_ob1_param_register_int(
|
|
const char* param_name,
|
|
int default_value,
|
|
int *storage)
|
|
{
|
|
*storage = default_value;
|
|
(void) mca_base_component_var_register(&mca_pml_ob1_component.pmlm_version, param_name,
|
|
NULL, MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
|
|
OPAL_INFO_LVL_9,
|
|
MCA_BASE_VAR_SCOPE_READONLY, storage);
|
|
return *storage;
|
|
}
|
|
|
|
static inline unsigned int mca_pml_ob1_param_register_uint(
|
|
const char* param_name,
|
|
unsigned int default_value,
|
|
unsigned int *storage)
|
|
{
|
|
*storage = default_value;
|
|
(void) mca_base_component_var_register(&mca_pml_ob1_component.pmlm_version, param_name,
|
|
NULL, MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, 0, 0,
|
|
OPAL_INFO_LVL_9,
|
|
MCA_BASE_VAR_SCOPE_READONLY, storage);
|
|
return *storage;
|
|
}
|
|
|
|
static inline size_t mca_pml_ob1_param_register_sizet(
|
|
const char* param_name,
|
|
size_t default_value,
|
|
size_t *storage)
|
|
{
|
|
*storage = default_value;
|
|
(void) mca_base_component_var_register(&mca_pml_ob1_component.pmlm_version, param_name,
|
|
NULL, MCA_BASE_VAR_TYPE_SIZE_T, NULL, 0, 0,
|
|
OPAL_INFO_LVL_9,
|
|
MCA_BASE_VAR_SCOPE_READONLY, storage);
|
|
return *storage;
|
|
}
|
|
|
|
static int mca_pml_ob1_comm_size_notify (mca_base_pvar_t *pvar, mca_base_pvar_event_t event, void *obj_handle, int *count)
|
|
{
|
|
if (MCA_BASE_PVAR_HANDLE_BIND == event) {
|
|
/* Return the size of the communicator as the number of values */
|
|
*count = ompi_comm_size ((ompi_communicator_t *) obj_handle);
|
|
}
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static int mca_pml_ob1_get_unex_msgq_size (const struct mca_base_pvar_t *pvar, void *value, void *obj_handle)
|
|
{
|
|
ompi_communicator_t *comm = (ompi_communicator_t *) obj_handle;
|
|
mca_pml_ob1_comm_t *pml_comm = comm->c_pml_comm;
|
|
int comm_size = ompi_comm_size (comm);
|
|
unsigned *values = (unsigned *) value;
|
|
mca_pml_ob1_comm_proc_t *pml_proc;
|
|
int i;
|
|
|
|
for (i = 0 ; i < comm_size ; ++i) {
|
|
pml_proc = pml_comm->procs + i;
|
|
|
|
values[i] = opal_list_get_size (&pml_proc->unexpected_frags);
|
|
}
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static int mca_pml_ob1_get_posted_recvq_size (const struct mca_base_pvar_t *pvar, void *value, void *obj_handle)
|
|
{
|
|
ompi_communicator_t *comm = (ompi_communicator_t *) obj_handle;
|
|
mca_pml_ob1_comm_t *pml_comm = comm->c_pml_comm;
|
|
int comm_size = ompi_comm_size (comm);
|
|
unsigned *values = (unsigned *) value;
|
|
mca_pml_ob1_comm_proc_t *pml_proc;
|
|
int i;
|
|
|
|
for (i = 0 ; i < comm_size ; ++i) {
|
|
pml_proc = pml_comm->procs + i;
|
|
|
|
values[i] = opal_list_get_size (&pml_proc->specific_receives);
|
|
}
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static int mca_pml_ob1_component_register(void)
|
|
{
|
|
mca_pml_ob1_param_register_int("verbose", 0, &mca_pml_ob1_verbose);
|
|
|
|
mca_pml_ob1_param_register_int("free_list_num", 4, &mca_pml_ob1.free_list_num);
|
|
mca_pml_ob1_param_register_int("free_list_max", -1, &mca_pml_ob1.free_list_max);
|
|
mca_pml_ob1_param_register_int("free_list_inc", 64, &mca_pml_ob1.free_list_inc);
|
|
mca_pml_ob1_param_register_int("priority", 20, &mca_pml_ob1.priority);
|
|
mca_pml_ob1_param_register_sizet("send_pipeline_depth", 3, &mca_pml_ob1.send_pipeline_depth);
|
|
mca_pml_ob1_param_register_sizet("recv_pipeline_depth", 4, &mca_pml_ob1.recv_pipeline_depth);
|
|
|
|
/* NTH: we can get into a live-lock situation in the RDMA failure path so disable
|
|
RDMA retries for now. Falling back to send may suck but it is better than
|
|
hanging */
|
|
mca_pml_ob1.rdma_retries_limit = 0;
|
|
/* mca_pml_ob1_param_register_sizet("rdma_retries_limit", 5, &mca_pml_ob1.rdma_retries_limit); */
|
|
|
|
mca_pml_ob1_param_register_int("max_rdma_per_request", 4, &mca_pml_ob1.max_rdma_per_request);
|
|
mca_pml_ob1_param_register_int("max_send_per_range", 4, &mca_pml_ob1.max_send_per_range);
|
|
|
|
mca_pml_ob1_param_register_uint("unexpected_limit", 128, &mca_pml_ob1.unexpected_limit);
|
|
|
|
mca_pml_ob1.allocator_name = "bucket";
|
|
(void) mca_base_component_var_register(&mca_pml_ob1_component.pmlm_version, "allocator",
|
|
"Name of allocator component for unexpected messages",
|
|
MCA_BASE_VAR_TYPE_STRING, NULL, 0, 0, OPAL_INFO_LVL_9,
|
|
MCA_BASE_VAR_SCOPE_READONLY, &mca_pml_ob1.allocator_name);
|
|
|
|
(void) mca_base_pvar_register ("ompi", "pml", "ob1", "unexpected_msgq_length", "Number of unexpected messages "
|
|
"received by each peer in a communicator", OPAL_INFO_LVL_4, MPI_T_PVAR_CLASS_SIZE,
|
|
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, MPI_T_BIND_MPI_COMM,
|
|
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
|
|
mca_pml_ob1_get_unex_msgq_size, NULL, mca_pml_ob1_comm_size_notify, NULL);
|
|
|
|
(void) mca_base_pvar_register ("ompi", "pml", "ob1", "posted_recvq_length", "Number of unmatched receives "
|
|
"posted for each peer in a communicator", OPAL_INFO_LVL_4, MPI_T_PVAR_CLASS_SIZE,
|
|
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, MPI_T_BIND_MPI_COMM,
|
|
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
|
|
mca_pml_ob1_get_posted_recvq_size, NULL, mca_pml_ob1_comm_size_notify, NULL);
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
static int mca_pml_ob1_component_open(void)
|
|
{
|
|
mca_pml_ob1_output = opal_output_open(NULL);
|
|
opal_output_set_verbosity(mca_pml_ob1_output, mca_pml_ob1_verbose);
|
|
|
|
mca_pml_ob1.enabled = false;
|
|
return mca_base_framework_open(&ompi_bml_base_framework, 0);
|
|
}
|
|
|
|
|
|
static int mca_pml_ob1_component_close(void)
|
|
{
|
|
int rc;
|
|
|
|
if (OMPI_SUCCESS != (rc = mca_base_framework_close(&ompi_bml_base_framework))) {
|
|
return rc;
|
|
}
|
|
opal_output_close(mca_pml_ob1_output);
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
|
|
static mca_pml_base_module_t*
|
|
mca_pml_ob1_component_init( int* priority,
|
|
bool enable_progress_threads,
|
|
bool enable_mpi_threads )
|
|
{
|
|
mca_allocator_base_component_t* allocator_component;
|
|
|
|
opal_output_verbose( 10, mca_pml_ob1_output,
|
|
"in ob1, my priority is %d\n", mca_pml_ob1.priority);
|
|
|
|
if((*priority) > mca_pml_ob1.priority) {
|
|
*priority = mca_pml_ob1.priority;
|
|
return NULL;
|
|
}
|
|
*priority = mca_pml_ob1.priority;
|
|
|
|
allocator_component = mca_allocator_component_lookup( mca_pml_ob1.allocator_name );
|
|
if(NULL == allocator_component) {
|
|
opal_output(0, "mca_pml_ob1_component_init: can't find allocator: %s\n", mca_pml_ob1.allocator_name);
|
|
return NULL;
|
|
}
|
|
|
|
mca_pml_ob1.allocator = allocator_component->allocator_init(true,
|
|
mca_pml_ob1_seg_alloc,
|
|
mca_pml_ob1_seg_free, NULL);
|
|
if(NULL == mca_pml_ob1.allocator) {
|
|
opal_output(0, "mca_pml_ob1_component_init: unable to initialize allocator\n");
|
|
return NULL;
|
|
}
|
|
|
|
if(OMPI_SUCCESS != mca_bml_base_init( enable_progress_threads,
|
|
enable_mpi_threads)) {
|
|
return NULL;
|
|
}
|
|
|
|
/* Set this here (vs in component_open()) because
|
|
opal_leave_pinned* may have been set after MCA params were
|
|
read (e.g., by the openib btl) */
|
|
mca_pml_ob1.leave_pinned = (1 == opal_leave_pinned);
|
|
mca_pml_ob1.leave_pinned_pipeline = (int) opal_leave_pinned_pipeline;
|
|
|
|
return &mca_pml_ob1.super;
|
|
}
|
|
|
|
int mca_pml_ob1_component_fini(void)
|
|
{
|
|
int rc;
|
|
|
|
/* Shutdown BML */
|
|
if(OMPI_SUCCESS != (rc = mca_bml.bml_finalize()))
|
|
return rc;
|
|
|
|
if(!mca_pml_ob1.enabled)
|
|
return OMPI_SUCCESS; /* never selected.. return success.. */
|
|
mca_pml_ob1.enabled = false; /* not anymore */
|
|
|
|
OBJ_DESTRUCT(&mca_pml_ob1.rdma_pending);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.pckt_pending);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.recv_pending);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.send_pending);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.non_existing_communicator_pending);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.buffers);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.pending_pckts);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.recv_frags);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.rdma_frags);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.lock);
|
|
OBJ_DESTRUCT(&mca_pml_ob1.send_ranges);
|
|
|
|
if( NULL != mca_pml_ob1.allocator ) {
|
|
(void)mca_pml_ob1.allocator->alc_finalize(mca_pml_ob1.allocator);
|
|
mca_pml_ob1.allocator = NULL;
|
|
}
|
|
|
|
#if 0
|
|
if (mca_pml_base_send_requests.fl_num_allocated !=
|
|
mca_pml_base_send_requests.super.opal_list_length) {
|
|
opal_output(0, "ob1 send requests: %d allocated %d returned\n",
|
|
mca_pml_base_send_requests.fl_num_allocated,
|
|
mca_pml_base_send_requests.super.opal_list_length);
|
|
}
|
|
if (mca_pml_base_recv_requests.fl_num_allocated !=
|
|
mca_pml_base_recv_requests.super.opal_list_length) {
|
|
opal_output(0, "ob1 recv requests: %d allocated %d returned\n",
|
|
mca_pml_base_recv_requests.fl_num_allocated,
|
|
mca_pml_base_recv_requests.super.opal_list_length);
|
|
}
|
|
#endif
|
|
|
|
return OMPI_SUCCESS;
|
|
}
|
|
|
|
void *mca_pml_ob1_seg_alloc( struct mca_mpool_base_module_t* mpool,
|
|
size_t* size,
|
|
mca_mpool_base_registration_t** registration) {
|
|
return malloc(*size);
|
|
}
|
|
|
|
void mca_pml_ob1_seg_free( struct mca_mpool_base_module_t* mpool,
|
|
void* segment ) {
|
|
free(segment);
|
|
}
|