/*
* Register an MPI_T performance variable of type CLASS_COUNTER.
*/
static void register_pvar_counter(char *name, char *desc, size_t offset)
{
int rc __opal_attribute_unused__;
rc = mca_base_component_pvar_register(&mca_btl_usnic_component.super.btl_version,
name, desc,
OPAL_INFO_LVL_5,
MCA_BASE_PVAR_CLASS_COUNTER,
pvar_type,
NULL, /* enumeration */
MCA_BASE_VAR_BIND_NO_OBJECT,
(MCA_BASE_PVAR_FLAG_READONLY |
MCA_BASE_PVAR_FLAG_CONTINUOUS),
usnic_pvar_read,
NULL, /* write function */
usnic_pvar_notify,
(void *) offset);
assert(rc >= 0);
}
static int btl_scif_component_register(void)
{
(void) mca_base_var_group_component_register(&mca_btl_scif_component.super.btl_version,
"SCIF byte transport layer");
mca_btl_scif_component.scif_free_list_num = 8;
(void) mca_base_component_var_register(&mca_btl_scif_component.super.btl_version,
"free_list_num", "Initial fragment free list size",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_scif_component.scif_free_list_num);
mca_btl_scif_component.scif_free_list_max = 16384;
(void) mca_base_component_var_register(&mca_btl_scif_component.super.btl_version,
"free_list_max", "Maximum fragment free list size",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_scif_component.scif_free_list_max);
mca_btl_scif_component.scif_free_list_inc = 64;
(void) mca_base_component_var_register(&mca_btl_scif_component.super.btl_version,
"free_list_inc", "Fragment free list size increment",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_scif_component.scif_free_list_inc);
mca_btl_scif_component.segment_size = 8 * 1024;
(void) mca_base_component_var_register(&mca_btl_scif_component.super.btl_version,
"segment_size", "Size of memory segment to "
"allocate for each remote process (default: "
"8k)", MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_scif_component.segment_size);
mca_btl_scif_component.rma_use_cpu = false;
(void) mca_base_component_var_register(&mca_btl_scif_component.super.btl_version,
"rma_use_cpu", "Use CPU instead of DMA "
"for RMA copies (default: false)", MCA_BASE_VAR_TYPE_BOOL,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_scif_component.rma_use_cpu);
mca_btl_scif_component.rma_sync = true;
(void) mca_base_component_var_register(&mca_btl_scif_component.super.btl_version,
"rma_sync", "Use synchronous RMA instead of "
"an RMA fence (default: true)", MCA_BASE_VAR_TYPE_BOOL,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_scif_component.rma_sync);
#if defined(SCIF_TIMING)
mca_btl_scif_component.aquire_buffer_time = 0.0;
(void) mca_base_component_pvar_register(&mca_btl_scif_component.super.btl_version,
"aquire_buffer_time", "Aggregate time spent "
"aquiring send buffers", OPAL_INFO_LVL_9,
MCA_BASE_PVAR_CLASS_AGGREGATE, MCA_BASE_VAR_TYPE_DOUBLE,
NULL, MCA_BASE_VAR_BIND_NO_OBJECT, MCA_BASE_PVAR_FLAG_READONLY |
MCA_BASE_PVAR_FLAG_CONTINUOUS, NULL, NULL, NULL,
&mca_btl_scif_component.aquire_buffer_time);
mca_btl_scif_component.send_time = 0.0;
(void) mca_base_component_pvar_register(&mca_btl_scif_component.super.btl_version,
"send_time", "Aggregate time spent writing to "
"send buffers", OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_AGGREGATE,
MCA_BASE_VAR_TYPE_DOUBLE, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
NULL, NULL, NULL, &mca_btl_scif_component.send_time);
mca_btl_scif_component.sendi_time = 0.0;
(void) mca_base_component_pvar_register(&mca_btl_scif_component.super.btl_version,
"sendi_time", "Aggregate time spent writing to "
"send buffers in sendi", OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_AGGREGATE,
MCA_BASE_VAR_TYPE_DOUBLE, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
NULL, NULL, NULL, &mca_btl_scif_component.sendi_time);
mca_btl_scif_component.get_time = 0.0;
(void) mca_base_component_pvar_register(&mca_btl_scif_component.super.btl_version,
"get_time", "Aggregate time spent in DMA read (scif_readfrom)",
OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_AGGREGATE,
MCA_BASE_VAR_TYPE_DOUBLE, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
NULL, NULL, NULL, &mca_btl_scif_component.get_time);
mca_btl_scif_component.get_count = 0;
(void) mca_base_component_pvar_register(&mca_btl_scif_component.super.btl_version,
"get_count", "Number of times btl_scif_get was called",
OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_COUNTER,
MCA_BASE_VAR_TYPE_UNSIGNED_LONG, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
NULL, NULL, NULL, &mca_btl_scif_component.get_count);
mca_btl_scif_component.put_time = 0.0;
(void) mca_base_component_pvar_register(&mca_btl_scif_component.super.btl_version,
"put_time", "Aggregate time spent in DMA write (scif_writeto)",
OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_AGGREGATE,
MCA_BASE_VAR_TYPE_DOUBLE, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
NULL, NULL, NULL, &mca_btl_scif_component.put_time);
mca_btl_scif_component.put_count = 0;
(void) mca_base_component_pvar_register(&mca_btl_scif_component.super.btl_version,
"put_count", "Number of times btl_scif_put was called",
OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_COUNTER,
MCA_BASE_VAR_TYPE_UNSIGNED_LONG, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
NULL, NULL, NULL, &mca_btl_scif_component.put_count);
#endif
mca_btl_scif_module.super.btl_exclusivity = MCA_BTL_EXCLUSIVITY_HIGH;
mca_btl_scif_module.super.btl_eager_limit = 1 * 1024;
mca_btl_scif_module.super.btl_rndv_eager_limit = 1 * 1024;
mca_btl_scif_module.super.btl_rdma_pipeline_frag_size = 4 * 1024 * 1024;
mca_btl_scif_module.super.btl_max_send_size = 1 * 1024;
mca_btl_scif_module.super.btl_rdma_pipeline_send_length = 1 * 1024;
/* threshold for put */
mca_btl_scif_module.super.btl_min_rdma_pipeline_size = 1 * 1024;
mca_btl_scif_module.super.btl_flags = MCA_BTL_FLAGS_SEND |
MCA_BTL_FLAGS_RDMA | MCA_BTL_FLAGS_SEND_INPLACE;
mca_btl_scif_module.super.btl_registration_handle_size = sizeof (mca_btl_base_registration_handle_t);
mca_btl_scif_module.super.btl_bandwidth = 50000; /* Mbs */
mca_btl_scif_module.super.btl_latency = 2; /* Microsecs */
/* Call the BTL based to register its MCA params */
mca_btl_base_param_register(&mca_btl_scif_component.super.btl_version,
&mca_btl_scif_module.super);
return OPAL_SUCCESS;
}
/*
* Setup the usnic_X device enumeration pvar
*/
static void setup_mpit_pvars_enum(void)
{
int i;
int rc __opal_attribute_unused__;
mca_base_var_enum_value_t *devices;
static mca_base_var_enum_t *devices_enum;
opal_btl_usnic_module_t *m;
unsigned char *c;
struct sockaddr_in *sin;
devices = calloc(mca_btl_usnic_component.num_modules + 1,
sizeof(*devices));
assert(devices != NULL);
for (i = 0; i < mca_btl_usnic_component.num_modules; ++i) {
char *str;
m = mca_btl_usnic_component.usnic_active_modules[i];
sin = m->fabric_info->src_addr;
c = (unsigned char*) &sin->sin_addr.s_addr;
devices[i].value = i;
rc = asprintf(&str, "%s,%hhu.%hhu.%hhu.%hhu/%" PRIu32,
m->fabric_info->fabric_attr->name,
c[0], c[1], c[2], c[3],
usnic_netmask_to_cidrlen(sin->sin_addr.s_addr));
assert(rc > 0);
devices[i].string = str;
}
devices[i].string = NULL;
rc = mca_base_var_enum_create("btl_usnic", devices, &devices_enum);
assert(OPAL_SUCCESS == rc);
rc = mca_base_component_pvar_register(&mca_btl_usnic_component.super.btl_version,
"devices",
"Enumeration representing which slot in btl_usnic_* MPI_T pvar value arrays correspond to which usnic_X Linux device",
OPAL_INFO_LVL_5,
MCA_BASE_PVAR_CLASS_STATE,
MCA_BASE_VAR_TYPE_INT,
devices_enum,
MCA_BASE_VAR_BIND_NO_OBJECT,
(MCA_BASE_PVAR_FLAG_READONLY |
MCA_BASE_PVAR_FLAG_CONTINUOUS),
usnic_pvar_enum_read,
NULL, /* write function */
usnic_pvar_notify,
NULL /* context */);
assert(rc >= 0);
/* Free the strings (mca_base_var_enum_create() strdup()'ed them
into private storage, so we don't need them any more) */
for (int i = 0; i < mca_btl_usnic_component.num_modules; ++i) {
free((char*) devices[i].string);
}
free(devices);
/* The devices_enum has been RETAIN'ed by the pvar, so we can
RELEASE it here, and the enum will be destroyed when the pvar
is destroyed. */
OBJ_RELEASE(devices_enum);
}
static int
btl_ugni_component_register(void)
{
mca_base_var_enum_t *new_enum;
gni_nic_device_t device_type;
int rc;
(void) mca_base_var_group_component_register(&mca_btl_ugni_component.super.btl_version,
"uGNI byte transport layer");
mca_btl_ugni_component.ugni_free_list_num = 8;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"free_list_num", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_free_list_num);
mca_btl_ugni_component.ugni_free_list_max = 4096;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"free_list_max", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_free_list_max);
mca_btl_ugni_component.ugni_free_list_inc = 64;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"free_list_inc", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_free_list_inc);
mca_btl_ugni_component.ugni_eager_num = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"eager_num", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_eager_num);
mca_btl_ugni_component.ugni_eager_max = 128;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"eager_max", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_eager_max);
mca_btl_ugni_component.ugni_eager_inc = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"eager_inc", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_eager_inc);
mca_btl_ugni_component.remote_cq_size = 40000;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"remote_cq_size", "Remote SMSG completion queue "
"size (default 40000)", MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.remote_cq_size);
mca_btl_ugni_component.local_cq_size = 8192;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"local_cq_size", "Local completion queue size "
"(default 8192)", MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.local_cq_size);
mca_btl_ugni_component.ugni_smsg_limit = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_limit", "Maximum size message that "
"will be sent using the SMSG/MSGQ protocol "
"(0 - autoselect(default), 16k max)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_smsg_limit);
mca_btl_ugni_component.smsg_max_credits = 32;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_max_credits", "Maximum number of "
"outstanding SMSG/MSGQ message (default 32)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.smsg_max_credits);
mca_btl_ugni_component.ugni_fma_limit = 1024;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"fma_limit", "Maximum size message that "
"will be sent using the FMA (Fast Memory "
"Access) protocol (default 1024, 64k max)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_fma_limit);
mca_btl_ugni_component.rdma_max_retries = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"rdma_max_retries", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.rdma_max_retries);
mca_btl_ugni_component.smsg_max_retries = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_max_retries", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.smsg_max_retries);
mca_btl_ugni_component.max_mem_reg = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"max_mem_reg", "Maximum number of "
"memory registrations a process can "
"hold (0 - autoselect, -1 - unlimited)"
" (default 0)", MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_3, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.max_mem_reg);
mca_btl_ugni_component.mbox_increment = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"mbox_inc", "Number of SMSG mailboxes to "
"allocate in each block (0 - autoselect(default))",
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.mbox_increment);
mca_btl_ugni_component.smsg_page_size = 2 << 20;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_page_size", "Page size to use for SMSG "
"mailbox allocation (default 2M)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.smsg_page_size);
mca_btl_ugni_component.progress_thread_requested = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"request_progress_thread",
"Enable to request ugni btl progress thread - requires MPI_THREAD_MULTIPLE support",
MCA_BASE_VAR_TYPE_BOOL, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.progress_thread_requested);
/* performance variables */
mca_btl_ugni_progress_thread_wakeups = 0;
(void) mca_base_component_pvar_register(&mca_btl_ugni_component.super.btl_version,
"progress_thread_wakeups", "Number of times the progress thread "
"has been woken", OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_COUNTER,
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS, NULL,
NULL, NULL, &mca_btl_ugni_progress_thread_wakeups);
/* btl/ugni can only support only a fixed set of mpools (these mpools have compatible resource
* structures) */
rc = mca_base_var_enum_create ("btl_ugni_mpool", mpool_values, &new_enum);
if (OPAL_SUCCESS != rc) {
return rc;
}
mca_btl_ugni_component.mpool_type = MCA_BTL_UGNI_MPOOL_UDREG;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"mpool", "mpool to use", MCA_BASE_VAR_TYPE_INT, new_enum,
0, MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.mpool_type);
OBJ_RELEASE(new_enum);
/* ensure we loose send exclusivity to sm and vader if they are enabled */
mca_btl_ugni_module.super.btl_exclusivity = MCA_BTL_EXCLUSIVITY_HIGH - 2;
/* smsg threshold */
mca_btl_ugni_module.super.btl_eager_limit = 8 * 1024;
mca_btl_ugni_module.super.btl_rndv_eager_limit = 8 * 1024;
mca_btl_ugni_module.super.btl_rdma_pipeline_frag_size = 4 * 1024 * 1024;
mca_btl_ugni_module.super.btl_max_send_size = 8 * 1024;
mca_btl_ugni_module.super.btl_rdma_pipeline_send_length = 8 * 1024;
mca_btl_ugni_module.super.btl_get_limit = 1 * 1024 * 1024;
/* determine if there are get alignment restrictions */
GNI_GetDeviceType (&device_type);
/*
* see def. of ALIGNMENT_MASK to figure this one out
*/
/* both gemini and aries have a 4-byte alignment requirement on remote addresses */
mca_btl_ugni_module.super.btl_get_alignment = 4;
/* threshold for put */
mca_btl_ugni_module.super.btl_min_rdma_pipeline_size = 8 * 1024;
mca_btl_ugni_module.super.btl_flags = MCA_BTL_FLAGS_SEND |
MCA_BTL_FLAGS_RDMA | MCA_BTL_FLAGS_SEND_INPLACE | MCA_BTL_FLAGS_ATOMIC_OPS |
MCA_BTL_FLAGS_ATOMIC_FOPS;
mca_btl_ugni_module.super.btl_atomic_flags = MCA_BTL_ATOMIC_SUPPORTS_ADD |
MCA_BTL_ATOMIC_SUPPORTS_AND | MCA_BTL_ATOMIC_SUPPORTS_OR | MCA_BTL_ATOMIC_SUPPORTS_XOR |
MCA_BTL_ATOMIC_SUPPORTS_CSWAP;
mca_btl_ugni_module.super.btl_registration_handle_size = sizeof (mca_btl_base_registration_handle_t);
mca_btl_ugni_module.super.btl_bandwidth = 40000; /* Mbs */
mca_btl_ugni_module.super.btl_latency = 2; /* Microsecs */
mca_btl_ugni_module.super.btl_get_local_registration_threshold = 0;
mca_btl_ugni_module.super.btl_put_local_registration_threshold = mca_btl_ugni_component.ugni_fma_limit;
/* Call the BTL based to register its MCA params */
mca_btl_base_param_register(&mca_btl_ugni_component.super.btl_version,
&mca_btl_ugni_module.super);
return OPAL_SUCCESS;
}
static int btl_ugni_component_register(void)
{
mca_base_var_enum_t *new_enum;
gni_nic_device_t device_type;
char *mpool_hints_tmp = NULL;
int rc;
(void) mca_base_var_group_component_register(&mca_btl_ugni_component.super.btl_version,
"uGNI byte transport layer");
mca_btl_ugni_component.ugni_free_list_num = 8;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"free_list_num", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_free_list_num);
mca_btl_ugni_component.ugni_free_list_max = 4096;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"free_list_max", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_free_list_max);
mca_btl_ugni_component.ugni_free_list_inc = 64;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"free_list_inc", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_free_list_inc);
mca_btl_ugni_component.ugni_eager_num = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"eager_num", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_eager_num);
mca_btl_ugni_component.ugni_eager_max = 128;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"eager_max", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_eager_max);
mca_btl_ugni_component.ugni_eager_inc = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"eager_inc", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_eager_inc);
mca_btl_ugni_component.remote_cq_size = 40000;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"remote_cq_size", "Remote SMSG completion queue "
"size (default 40000)", MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.remote_cq_size);
mca_btl_ugni_component.local_cq_size = 8192;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"local_cq_size", "Local completion queue size "
"(default 8192)", MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.local_cq_size);
mca_btl_ugni_component.ugni_smsg_limit = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_limit", "Maximum size message that "
"will be sent using the SMSG/MSGQ protocol "
"(0 - autoselect(default), 16k max)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_smsg_limit);
mca_btl_ugni_component.smsg_max_credits = 32;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_max_credits", "Maximum number of "
"outstanding SMSG/MSGQ message (default 32)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.smsg_max_credits);
mca_btl_ugni_component.ugni_fma_limit = 1024;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"fma_limit", "Maximum size message that "
"will be sent using the FMA (Fast Memory "
"Access) protocol (default 1024, 64k max)",
MCA_BASE_VAR_TYPE_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.ugni_fma_limit);
mca_btl_ugni_component.rdma_max_retries = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"rdma_max_retries", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.rdma_max_retries);
mca_btl_ugni_component.smsg_max_retries = 16;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_max_retries", NULL, MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_5, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.smsg_max_retries);
mca_btl_ugni_component.max_mem_reg = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"max_mem_reg", "Maximum number of "
"memory registrations a process can "
"hold (0 - autoselect, -1 - unlimited)"
" (default 0)", MCA_BASE_VAR_TYPE_INT,
NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_3, MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.max_mem_reg);
mca_btl_ugni_component.mbox_increment = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"mbox_inc", "Number of SMSG mailboxes to "
"allocate in each block (0 - autoselect(default))",
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.mbox_increment);
/* communication domain flags */
rc = mca_base_var_enum_create_flag ("btl_ugni_cdm_flags", cdm_flags, (mca_base_var_enum_flag_t **) &new_enum);
if (OPAL_SUCCESS != rc) {
return rc;
}
mca_btl_ugni_component.cdm_flags = GNI_CDM_MODE_FORK_PARTCOPY | GNI_CDM_MODE_ERR_NO_KILL | GNI_CDM_MODE_FAST_DATAGRAM_POLL |
GNI_CDM_MODE_MDD_SHARED | GNI_CDM_MODE_FMA_SHARED | GNI_CDM_MODE_FMA_SMALL_WINDOW;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"cdm_flags", "Flags to set when creating a communication domain "
" (default: fork-fullcopy,cached-amo-enabled,err-no-kill,fast-datagram-poll,"
"fma-shared,fma-small-window)",
MCA_BASE_VAR_TYPE_UNSIGNED_INT, new_enum, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.cdm_flags);
OBJ_RELEASE(new_enum);
mca_btl_ugni_component.virtual_device_count = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"virtual_device_count", "Number of virtual devices to create. Higher numbers may "
"result in better performance when using threads. (default: auto, max: 8)",
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.virtual_device_count);
/* determine if there are get alignment restrictions */
GNI_GetDeviceType (&device_type);
mca_btl_ugni_component.smsg_page_size = 2 << 20;
if (GNI_DEVICE_GEMINI == device_type) {
if (access ("/sys/class/gemini/ghal0/mrt", R_OK)) {
int fd = open ("/sys/class/gemini/ghal0/mrt", O_RDONLY);
char buffer[10];
if (0 <= fd) {
memset (buffer, 0, sizeof (buffer));
read (fd, buffer, sizeof (buffer) - 1);
close (fd);
mca_btl_ugni_ugni_page_size = strtol (buffer, NULL, 10) * 1024;
mca_btl_ugni_component.smsg_page_size = mca_btl_ugni_ugni_page_size;
}
}
}
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"smsg_page_size", "Page size to use for SMSG mailbox allocation (default: detect)",
MCA_BASE_VAR_TYPE_INT, NULL, 0, MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.smsg_page_size);
mca_btl_ugni_component.progress_thread_requested = 0;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"request_progress_thread",
"Enable to request ugni btl progress thread - requires MPI_THREAD_MULTIPLE support",
MCA_BASE_VAR_TYPE_BOOL, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE,
OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL,
&mca_btl_ugni_component.progress_thread_requested);
/* performance variables */
mca_btl_ugni_progress_thread_wakeups = 0;
(void) mca_base_component_pvar_register(&mca_btl_ugni_component.super.btl_version,
"progress_thread_wakeups", "Number of times the progress thread "
"has been woken", OPAL_INFO_LVL_9, MCA_BASE_PVAR_CLASS_COUNTER,
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS, NULL,
NULL, NULL, &mca_btl_ugni_progress_thread_wakeups);
/* register network statistics as performance variables */
for (int i = 0 ; i < GNI_NUM_STATS ; ++i) {
char name[128], desc[128];
size_t str_len = strlen (gni_statistic_str[i]);
assert (str_len < sizeof (name));
/* we can get an all-caps string for the variable from gni_statistic_str. need to make it lowercase
* to match ompi standards */
for (size_t j = 0 ; j < str_len ; ++j) {
name[j] = tolower (gni_statistic_str[i][j]);
desc[j] = ('_' == name[j]) ? ' ' : name[j];
}
name[str_len] = '\0';
desc[str_len] = '\0';
(void) mca_base_component_pvar_register (&mca_btl_ugni_component.super.btl_version, name, desc,
OPAL_INFO_LVL_4, MCA_BASE_PVAR_CLASS_COUNTER,
MCA_BASE_VAR_TYPE_UNSIGNED_INT, NULL, MCA_BASE_VAR_BIND_NO_OBJECT,
MCA_BASE_PVAR_FLAG_READONLY | MCA_BASE_PVAR_FLAG_CONTINUOUS,
mca_btl_ugni_get_stat, NULL, mca_btl_ugni_notify_stat,
(void *) (intptr_t) i);
}
/* btl/ugni can only support only a fixed set of rcache components (these rcache components have compatible resource
* structures) */
rc = mca_base_var_enum_create ("btl_ugni_rcache", rcache_values, &new_enum);
if (OPAL_SUCCESS != rc) {
return rc;
}
/* NTH: there are known *serious* performance issues with udreg. if they are ever resolved it is the preferred rcache */
mca_btl_ugni_component.rcache_type = MCA_BTL_UGNI_RCACHE_GRDMA;
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"rcache", "registration cache to use (default: grdma)", MCA_BASE_VAR_TYPE_INT, new_enum,
0, MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.rcache_type);
OBJ_RELEASE(new_enum);
if (mca_btl_ugni_ugni_page_size) {
rc = asprintf (&mpool_hints_tmp, "page_size=%lu", mca_btl_ugni_ugni_page_size);
if (rc < 0) {
return OPAL_ERR_OUT_OF_RESOURCE;
}
mca_btl_ugni_component.mpool_hints = mpool_hints_tmp;
} else {
mca_btl_ugni_component.mpool_hints = "page_size=2M";
}
(void) mca_base_component_var_register(&mca_btl_ugni_component.super.btl_version,
"mpool_hints", "hints to use when selecting a memory pool (default: "
"\"page_size=2M\")", MCA_BASE_VAR_TYPE_STRING, NULL, 0,
MCA_BASE_VAR_FLAG_SETTABLE, OPAL_INFO_LVL_3,
MCA_BASE_VAR_SCOPE_LOCAL, &mca_btl_ugni_component.mpool_hints);
free (mpool_hints_tmp);
/* ensure we loose send exclusivity to sm and vader if they are enabled */
mca_btl_ugni_module.super.btl_exclusivity = MCA_BTL_EXCLUSIVITY_HIGH - 2;
/* smsg threshold */
mca_btl_ugni_module.super.btl_eager_limit = 8 * 1024;
mca_btl_ugni_module.super.btl_rndv_eager_limit = 8 * 1024;
mca_btl_ugni_module.super.btl_rdma_pipeline_frag_size = 4 * 1024 * 1024;
mca_btl_ugni_module.super.btl_max_send_size = 8 * 1024;
mca_btl_ugni_module.super.btl_rdma_pipeline_send_length = 8 * 1024;
mca_btl_ugni_module.super.btl_get_limit = 1 * 1024 * 1024;
/*
* see def. of ALIGNMENT_MASK to figure this one out
*/
/* both gemini and aries have a 4-byte alignment requirement on remote addresses */
mca_btl_ugni_module.super.btl_get_alignment = 4;
/* threshold for put */
mca_btl_ugni_module.super.btl_min_rdma_pipeline_size = 8 * 1024;
mca_btl_ugni_module.super.btl_flags = MCA_BTL_FLAGS_SEND |
MCA_BTL_FLAGS_RDMA | MCA_BTL_FLAGS_SEND_INPLACE | MCA_BTL_FLAGS_ATOMIC_OPS |
MCA_BTL_FLAGS_ATOMIC_FOPS;
mca_btl_ugni_module.super.btl_atomic_flags = MCA_BTL_ATOMIC_SUPPORTS_ADD |
MCA_BTL_ATOMIC_SUPPORTS_AND | MCA_BTL_ATOMIC_SUPPORTS_OR | MCA_BTL_ATOMIC_SUPPORTS_XOR |
MCA_BTL_ATOMIC_SUPPORTS_CSWAP;
if (GNI_DEVICE_ARIES == device_type) {
/* aries supports additional atomic operations */
mca_btl_ugni_module.super.btl_atomic_flags |= MCA_BTL_ATOMIC_SUPPORTS_MIN | MCA_BTL_ATOMIC_SUPPORTS_MAX |
MCA_BTL_ATOMIC_SUPPORTS_LAND | MCA_BTL_ATOMIC_SUPPORTS_LOR | MCA_BTL_ATOMIC_SUPPORTS_LXOR |
MCA_BTL_ATOMIC_SUPPORTS_32BIT | MCA_BTL_ATOMIC_SUPPORTS_FLOAT;
}
mca_btl_ugni_module.super.btl_registration_handle_size = sizeof (mca_btl_base_registration_handle_t);
mca_btl_ugni_module.super.btl_bandwidth = 40000; /* Mbs */
mca_btl_ugni_module.super.btl_latency = 2; /* Microsecs */
mca_btl_ugni_module.super.btl_get_local_registration_threshold = 0;
mca_btl_ugni_module.super.btl_put_local_registration_threshold = mca_btl_ugni_component.ugni_fma_limit;
/* Call the BTL based to register its MCA params */
mca_btl_base_param_register(&mca_btl_ugni_component.super.btl_version,
&mca_btl_ugni_module.super);
return OPAL_SUCCESS;
}