1
1
openmpi/ompi/mca/mtl/psm2/mtl_psm2_component.c
Matias A Cabral fc8582c560 MTL/PSM2: Do not lower the priority when all processes are local.
The intention of lowering the priority when all processes are local
was to favor Vader BTL. However, in builds including the OFI MTL it
gets selected instead.

Reviewed-by: Spruit, Neil R <neil.r.spruit@intel.com>
Reviewed-by: Gopalakrishnan, Aravind <aravind.gopalakrishnan@intel.com>
Signed-off-by: Matias Cabral <matias.a.cabral@intel.com>
2018-12-04 15:31:09 -08:00

438 строки
16 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) 2004-2005 The University of Tennessee and The University
* of Tennessee Research Foundation. All rights
* reserved.
* Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
* University of Stuttgart. All rights reserved.
* Copyright (c) 2004-2005 The Regents of the University of California.
* All rights reserved.
* Copyright (c) 2006-2010 QLogic Corporation. All rights reserved.
* Copyright (c) 2012-2017 Los Alamos National Security, LLC. All rights
* reserved.
* Copyright (c) 2013-2017 Intel, Inc. All rights reserved
* Copyright (c) 2018 Research Organization for Information Science
* and Technology (RIST). All rights reserved.
* Copyright (c) 2018 Amazon.com, Inc. or its affiliates. All Rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include "opal/mca/event/event.h"
#include "opal/util/output.h"
#include "opal/util/show_help.h"
#include "opal/util/opal_environ.h"
#include "opal/util/printf.h"
#include "ompi/proc/proc.h"
#include "mtl_psm2.h"
#include "mtl_psm2_types.h"
#include "mtl_psm2_request.h"
#include "psm2.h"
#include <sys/types.h>
#include <sys/stat.h>
#include <unistd.h>
#include <glob.h>
static int param_priority;
/* MPI_THREAD_MULTIPLE_SUPPORT */
opal_mutex_t mtl_psm2_mq_mutex = OPAL_MUTEX_STATIC_INIT;
#if OPAL_CUDA_SUPPORT
static bool cuda_envvar_set = false;
#endif
static int ompi_mtl_psm2_component_open(void);
static int ompi_mtl_psm2_component_close(void);
static int ompi_mtl_psm2_component_query(mca_base_module_t **module, int *priority);
static int ompi_mtl_psm2_component_register(void);
static mca_mtl_base_module_t* ompi_mtl_psm2_component_init( bool enable_progress_threads,
bool enable_mpi_threads );
mca_mtl_psm2_component_t mca_mtl_psm2_component = {
{
/* First, the mca_base_component_t struct containing meta
* information about the component itself */
.mtl_version = {
MCA_MTL_BASE_VERSION_2_0_0,
.mca_component_name = "psm2",
MCA_BASE_MAKE_VERSION(component, OMPI_MAJOR_VERSION, OMPI_MINOR_VERSION,
OMPI_RELEASE_VERSION),
.mca_open_component = ompi_mtl_psm2_component_open,
.mca_close_component = ompi_mtl_psm2_component_close,
.mca_query_component = ompi_mtl_psm2_component_query,
.mca_register_component_params = ompi_mtl_psm2_component_register,
},
.mtl_data = {
/* The component is not checkpoint ready */
MCA_BASE_METADATA_PARAM_NONE
},
.mtl_init = ompi_mtl_psm2_component_init,
}
};
struct ompi_mtl_psm2_shadow_variable {
int variable_type;
void *storage;
mca_base_var_storage_t default_value;
const char *env_name;
mca_base_var_info_lvl_t info_level;
const char *mca_name;
const char *description;
mca_base_var_flag_t flags;
};
struct ompi_mtl_psm2_shadow_variable ompi_mtl_psm2_shadow_variables[] = {
{MCA_BASE_VAR_TYPE_STRING, &ompi_mtl_psm2.psm2_devices, {.stringval = "self,shm,hfi"}, "PSM2_DEVICES", OPAL_INFO_LVL_3,
"devices",
"Comma-delimited list of PSM2 devices. Valid values: self, shm, hfi (default: self,shm,hfi. Reduced to self,shm in single node jobs)",0},
{MCA_BASE_VAR_TYPE_STRING, &ompi_mtl_psm2.psm2_memory, {.stringval = "normal"}, "PSM2_MEMORY", OPAL_INFO_LVL_9,
"memory_model", "PSM2 memory usage mode. Valid values: min, normal, large (default: normal)", 0},
{MCA_BASE_VAR_TYPE_UNSIGNED_LONG, &ompi_mtl_psm2.psm2_mq_sendreqs_max, {.ulval = 0}, "PSM2_MQ_SENDREQS_MAX", OPAL_INFO_LVL_3,
"mq_sendreqs_max", "PSM2 maximum number of isend requests in flight (default: unset, let libpsm2 use its default)", MCA_BASE_VAR_FLAG_DEF_UNSET},
{MCA_BASE_VAR_TYPE_UNSIGNED_LONG, &ompi_mtl_psm2.psm2_mq_recvreqs_max, {.ulval = 0}, "PSM2_MQ_RECVREQS_MAX", OPAL_INFO_LVL_3,
"mq_recvreqs_max", "PSM2 maximum number of irecv requests in flight (default: unset, let libpsm2 use its default)", MCA_BASE_VAR_FLAG_DEF_UNSET},
{MCA_BASE_VAR_TYPE_UNSIGNED_LONG, &ompi_mtl_psm2.psm2_mq_rndv_hfi_threshold, {.ulval = 0}, "PSM2_MQ_RNDV_HFI_THRESH", OPAL_INFO_LVL_3,
"hfi_eager_limit", "PSM2 eager to rendezvous threshold (default: unset, let libpsm2 use its defaults)", MCA_BASE_VAR_FLAG_DEF_UNSET},
{MCA_BASE_VAR_TYPE_UNSIGNED_LONG, &ompi_mtl_psm2.psm2_mq_rndv_shm_threshold, {.ulval = 0}, "PSM2_MQ_RNDV_SHM_THRESH", OPAL_INFO_LVL_3,
"shm_eager_limit", "PSM2 shared memory eager to rendezvous threshold (default: unset, let libpsm2 use its default)", MCA_BASE_VAR_FLAG_DEF_UNSET},
{MCA_BASE_VAR_TYPE_BOOL, &ompi_mtl_psm2.psm2_recvthread, {.boolval = true}, "PSM2_RCVTHREAD", OPAL_INFO_LVL_3,
"use_receive_thread", "Use PSM2 progress thread (default: true)"},
{MCA_BASE_VAR_TYPE_BOOL, &ompi_mtl_psm2.psm2_shared_contexts, {.boolval = true}, "PSM2_SHAREDCONTEXTS", OPAL_INFO_LVL_6,
"use_shared_contexts", "Share PSM contexts between MPI processes (default: true)"},
{MCA_BASE_VAR_TYPE_UNSIGNED_LONG, &ompi_mtl_psm2.psm2_max_contexts_per_job, {.ulval = 0}, "PSM2_MAX_CONTEXTS_PER_JOB", OPAL_INFO_LVL_9,
"max_contexts_per_job", "Maximum number of contexts available on a node (default: unset, let libpsm2 use its default)", MCA_BASE_VAR_FLAG_DEF_UNSET},
{MCA_BASE_VAR_TYPE_UNSIGNED_LONG, &ompi_mtl_psm2.psm2_tracemask, {.ulval = 1}, "PSM2_TRACEMASK", OPAL_INFO_LVL_9,
"trace_mask", "PSM2 tracemask value. See PSM2 documentation for accepted values in 0x (default: 1)"},
{MCA_BASE_VAR_TYPE_UNSIGNED_LONG, &ompi_mtl_psm2.psm2_opa_sl, {.ulval = 0}, "HFI_SL", OPAL_INFO_LVL_9,
"opa_service_level", "HFI Service Level (default: unset, let libpsm2 use its defaults)", MCA_BASE_VAR_FLAG_DEF_UNSET},
{-1},
};
static void ompi_mtl_psm2_set_shadow_env (struct ompi_mtl_psm2_shadow_variable *variable)
{
mca_base_var_storage_t *storage = variable->storage;
char *env_value;
int ret = 0;
int var_index = 0;
const mca_base_var_t *mca_base_var;
var_index = mca_base_var_find("ompi", "mtl", "psm2", variable->mca_name);
ret = mca_base_var_get (var_index,&mca_base_var);
/* Something is fundamentally broken if registered variables are
* not found */
if (OPAL_SUCCESS != ret) {
fprintf (stderr, "ERROR setting PSM2 environment variable: %s\n", variable->env_name);
return;
}
/** Skip setting variables for which the default behavior is "unset" */
if ((mca_base_var->mbv_flags & MCA_BASE_VAR_FLAG_DEF_UNSET) &&
(MCA_BASE_VAR_SOURCE_DEFAULT == mca_base_var->mbv_source)){
return ;
}
switch (variable->variable_type) {
case MCA_BASE_VAR_TYPE_BOOL:
ret = opal_asprintf (&env_value, "%s=%d", variable->env_name, storage->boolval ? 1 : 0);
break;
case MCA_BASE_VAR_TYPE_UNSIGNED_LONG:
if (0 == strcmp (variable->env_name, "PSM2_TRACEMASK")) {
/* PSM2 documentation shows the tracemask as a hexidecimal number. to be consitent
* use hexidecimal here. */
ret = opal_asprintf (&env_value, "%s=0x%lx", variable->env_name, storage->ulval);
} else {
ret = opal_asprintf (&env_value, "%s=%lu", variable->env_name, storage->ulval);
}
break;
case MCA_BASE_VAR_TYPE_STRING:
ret = opal_asprintf (&env_value, "%s=%s", variable->env_name, storage->stringval);
break;
}
if (0 > ret) {
fprintf (stderr, "ERROR setting PSM2 environment variable: %s\n", variable->env_name);
} else {
putenv (env_value);
}
}
static void ompi_mtl_psm2_register_shadow_env (struct ompi_mtl_psm2_shadow_variable *variable)
{
mca_base_var_storage_t *storage = variable->storage;
char *env_value;
env_value = getenv (variable->env_name);
switch (variable->variable_type) {
case MCA_BASE_VAR_TYPE_BOOL:
if (env_value) {
int tmp;
(void) mca_base_var_enum_bool.value_from_string (&mca_base_var_enum_bool, env_value, &tmp);
storage->boolval = !!tmp;
} else {
storage->boolval = variable->default_value.boolval;
}
break;
case MCA_BASE_VAR_TYPE_UNSIGNED_LONG:
if (env_value) {
storage->ulval = strtol (env_value, NULL, 0);
} else {
storage->ulval = variable->default_value.ulval;
}
break;
case MCA_BASE_VAR_TYPE_STRING:
if (env_value) {
storage->stringval = env_value;
} else {
storage->stringval = variable->default_value.stringval;
}
break;
}
(void) mca_base_component_var_register (&mca_mtl_psm2_component.super.mtl_version, variable->mca_name, variable->description,
variable->variable_type, NULL, 0, variable->flags, variable->info_level, MCA_BASE_VAR_SCOPE_READONLY,
variable->storage);
}
static int
get_num_total_procs(int *out_ntp)
{
*out_ntp = (int)ompi_process_info.num_procs;
return OMPI_SUCCESS;
}
static int
get_num_local_procs(int *out_nlp)
{
/* num_local_peers does not include us in
* its calculation, so adjust for that */
*out_nlp = (int)(1 + ompi_process_info.num_local_peers);
return OMPI_SUCCESS;
}
static int
ompi_mtl_psm2_component_register(void)
{
int num_local_procs, num_total_procs;
ompi_mtl_psm2.connect_timeout = 180;
(void) mca_base_component_var_register(&mca_mtl_psm2_component.super.mtl_version,
"connect_timeout",
"PSM2 connection timeout value in seconds",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&ompi_mtl_psm2.connect_timeout);
(void) get_num_local_procs(&num_local_procs);
(void) get_num_total_procs(&num_total_procs);
/* set priority high enough to beat ob1's default (also set higher than psm) */
if ((num_local_procs == num_total_procs) && (1 < num_total_procs)) {
/* Disable hfi if all processes are local. However, if running only one
* process assume it is ompi_info or this is most likely going to spawn, for
* which all PSM2 devices are needed */
setenv("PSM2_DEVICES", "self,shm", 0);
}
(void) mca_base_component_var_register (&mca_mtl_psm2_component.super.mtl_version,
"priority", "Priority of the PSM2 MTL component",
MCA_BASE_VAR_TYPE_INT, NULL, 0, 0,
OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY,
&param_priority);
for (int i = 0 ; ompi_mtl_psm2_shadow_variables[i].variable_type >= 0 ; ++i) {
ompi_mtl_psm2_register_shadow_env (ompi_mtl_psm2_shadow_variables + i);
}
ompi_mtl_psm2_register_pvars();
return OMPI_SUCCESS;
}
static int
ompi_mtl_psm2_component_open(void)
{
int res;
glob_t globbuf = {0};
/* Component available only if Omni-Path hardware is present */
res = glob("/dev/hfi1_[0-9]", GLOB_DOOFFS, NULL, &globbuf);
if (globbuf.gl_pathc > 0 || GLOB_NOMATCH==res) {
globfree(&globbuf);
}
if (0 != res) {
res = glob("/dev/hfi1_[0-9][0-9]", GLOB_APPEND, NULL, &globbuf);
if (globbuf.gl_pathc > 0) {
globfree(&globbuf);
}
if (0 != res) {
return OPAL_ERR_NOT_AVAILABLE;
}
}
/* Component available only if at least one hfi1 port is ACTIVE */
bool foundOnlineHfi1Port = false;
size_t i;
char portState[128];
FILE *devFile;
if (glob("/sys/class/infiniband/hfi1_*/ports/*/state",
GLOB_DOOFFS, NULL, &globbuf) != 0) {
return OPAL_ERR_NOT_AVAILABLE;
}
for (i=0;i < globbuf.gl_pathc; i++) {
devFile = fopen(globbuf.gl_pathv[i], "r");
fgets(portState, sizeof(portState), devFile);
fclose(devFile);
if (strstr(portState, "ACTIVE") != NULL) {
/* Found at least one ACTIVE port */
foundOnlineHfi1Port = true;
break;
}
}
globfree(&globbuf);
if (!foundOnlineHfi1Port) {
return OPAL_ERR_NOT_AVAILABLE;
}
return OMPI_SUCCESS;
}
static int
ompi_mtl_psm2_component_query(mca_base_module_t **module, int *priority)
{
/*
* if we get here it means that PSM2 is available so give high priority
*/
*priority = param_priority;
*module = (mca_base_module_t *)&ompi_mtl_psm2.super;
return OMPI_SUCCESS;
}
static int
ompi_mtl_psm2_component_close(void)
{
#if OPAL_CUDA_SUPPORT
if (cuda_envvar_set) {
opal_unsetenv("PSM2_CUDA", &environ);
}
#endif
return OMPI_SUCCESS;
}
static int
get_local_rank(int *out_rank)
{
ompi_node_rank_t my_node_rank;
*out_rank = 0;
if (OMPI_NODE_RANK_INVALID == (my_node_rank =
ompi_process_info.my_node_rank)) {
return OMPI_ERROR;
}
*out_rank = (int)my_node_rank;
return OMPI_SUCCESS;
}
static mca_mtl_base_module_t *
ompi_mtl_psm2_component_init(bool enable_progress_threads,
bool enable_mpi_threads)
{
psm2_error_t err;
int verno_major = PSM2_VERNO_MAJOR;
int verno_minor = PSM2_VERNO_MINOR;
int local_rank = -1, num_local_procs = 0;
#if OPAL_CUDA_SUPPORT
int ret;
char *cuda_env;
glob_t globbuf = {0};
#endif
/* Compute the total number of processes on this host and our local rank
* on that node. We need to provide PSM2 with these values so it can
* allocate hardware contexts appropriately across processes.
*/
if (OMPI_SUCCESS != get_num_local_procs(&num_local_procs)) {
opal_output(0, "Cannot determine number of local processes. "
"Cannot continue.\n");
return NULL;
}
if (OMPI_SUCCESS != get_local_rank(&local_rank)) {
opal_output(0, "Cannot determine local rank. Cannot continue.\n");
return NULL;
}
err = psm2_error_register_handler(NULL /* no ep */,
PSM2_ERRHANDLER_NOP);
if (err) {
opal_output(0, "Error in psm2_error_register_handler (error %s)\n",
psm2_error_get_string(err));
return NULL;
}
for (int i = 0 ; ompi_mtl_psm2_shadow_variables[i].variable_type >= 0 ; ++i) {
ompi_mtl_psm2_set_shadow_env (ompi_mtl_psm2_shadow_variables + i);
}
#if OPAL_CUDA_SUPPORT
/*
* If using CUDA enabled Open MPI, the user likely intends to
* run with CUDA buffers. So, force-set the envvar here if user failed
* to set it.
*/
ret = glob("/sys/module/nvidia", GLOB_DOOFFS, NULL, &globbuf);
if (globbuf.gl_pathc > 0) {
globfree(&globbuf);
}
cuda_env = getenv("PSM2_CUDA");
if (!cuda_env && (0 == ret)) {
opal_show_help("help-mtl-psm2.txt",
"no psm2 cuda env", true,
ompi_process_info.nodename);
opal_setenv("PSM2_CUDA", "1", false, &environ);
cuda_envvar_set = true;
}
#endif
err = psm2_init(&verno_major, &verno_minor);
if (err) {
opal_show_help("help-mtl-psm2.txt",
"psm2 init", true,
psm2_error_get_string(err));
return NULL;
}
/* Complete PSM2 initialization */
ompi_mtl_psm2_module_init(local_rank, num_local_procs);
ompi_mtl_psm2.super.mtl_request_size =
sizeof(mca_mtl_psm2_request_t) -
sizeof(struct mca_mtl_request_t);
return &ompi_mtl_psm2.super;
}