int opal_btl_usnic_component_register(void)
{
int tmp, ret = 0;
static int max_modules;
static int stats_relative;
static int want_numa_device_assignment;
static int sd_num;
static int rd_num;
static int prio_sd_num;
static int prio_rd_num;
static int cq_num;
static int av_eq_num;
static int udp_port_base;
static int max_tiny_msg_size;
static int eager_limit;
static int rndv_eager_limit;
static int pack_lazy_threshold;
static int max_short_packets;
#define CHECK(expr) do {\
tmp = (expr); \
if (OPAL_SUCCESS != tmp) ret = tmp; \
} while (0)
CHECK(reg_int("max_btls",
"Maximum number of usNICs to use (default: 0 = as many as are available)",
0, &max_modules,
REGINT_GE_ZERO, OPAL_INFO_LVL_2));
mca_btl_usnic_component.max_modules = (size_t) max_modules;
CHECK(reg_string("if_include",
"Comma-delimited list of usNIC devices/networks to be used (e.g. \"eth3,usnic_0,10.10.0.0/16\"; empty value means to use all available usNICs). Mutually exclusive with btl_usnic_if_exclude.",
NULL, &mca_btl_usnic_component.if_include,
REGSTR_EMPTY_OK, OPAL_INFO_LVL_1));
CHECK(reg_string("if_exclude",
"Comma-delimited list of usNIC devices/networks to be excluded (empty value means to not exclude any usNICs). Mutually exclusive with btl_usnic_if_include.",
NULL, &mca_btl_usnic_component.if_exclude,
REGSTR_EMPTY_OK, OPAL_INFO_LVL_1));
CHECK(reg_int("stats",
"A non-negative integer specifying the frequency at which each usnic BTL will output statistics (default: 0 seconds, meaning that statistics are disabled)",
0, &mca_btl_usnic_component.stats_frequency, 0,
OPAL_INFO_LVL_4));
mca_btl_usnic_component.stats_enabled =
(bool) (mca_btl_usnic_component.stats_frequency > 0);
CHECK(reg_int("stats_relative",
"If stats are enabled, output relative stats between the timestamps (vs. cumulative stats since the beginning of the job) (default: 0 -- i.e., absolute)",
0, &stats_relative, 0, OPAL_INFO_LVL_4));
mca_btl_usnic_component.stats_relative = (bool) stats_relative;
#if RCACHE_VERSION == 30
CHECK(reg_string("mpool_hints", "Hints to use when selecting mpool",
NULL, &mca_btl_usnic_component.usnic_mpool_hints,
REGSTR_EMPTY_OK,
OPAL_INFO_LVL_5));
CHECK(reg_string("rcache", "Name of the registration cache to be used",
"grdma", &mca_btl_usnic_component.usnic_rcache_name, 0,
OPAL_INFO_LVL_5));
#else
CHECK(reg_string("mpool", "Name of the memory pool to be used",
"grdma", &mca_btl_usnic_component.usnic_mpool_name, 0,
OPAL_INFO_LVL_5));
#endif
want_numa_device_assignment = OPAL_HAVE_HWLOC ? 1 : -1;
CHECK(reg_int("want_numa_device_assignment",
"If 1, use only Cisco VIC ports thare are a minimum NUMA distance from the MPI process for short messages. If 0, use all available Cisco VIC ports for short messages. This parameter is meaningless (and ignored) unless MPI proceses are bound to processor cores. Defaults to 1 if NUMA support is included in Open MPI; -1 otherwise.",
want_numa_device_assignment,
&want_numa_device_assignment,
0, OPAL_INFO_LVL_5));
mca_btl_usnic_component.want_numa_device_assignment =
(1 == want_numa_device_assignment) ? true : false;
CHECK(reg_int("sd_num", "Maximum send descriptors to post (-1 = pre-set defaults; depends on number and type of devices available)",
-1, &sd_num, REGINT_NEG_ONE_OK, OPAL_INFO_LVL_5));
mca_btl_usnic_component.sd_num = (int32_t) sd_num;
CHECK(reg_int("rd_num", "Number of pre-posted receive buffers (-1 = pre-set defaults; depends on number and type of devices available)",
-1, &rd_num, REGINT_NEG_ONE_OK, OPAL_INFO_LVL_5));
mca_btl_usnic_component.rd_num = (int32_t) rd_num;
CHECK(reg_int("prio_sd_num", "Maximum priority send descriptors to post (-1 = pre-set defaults; depends on number and type of devices available)",
-1, &prio_sd_num, REGINT_NEG_ONE_OK, OPAL_INFO_LVL_5));
mca_btl_usnic_component.prio_sd_num = (int32_t) prio_sd_num;
CHECK(reg_int("prio_rd_num", "Number of pre-posted priority receive buffers (-1 = pre-set defaults; depends on number and type of devices available)",
-1, &prio_rd_num, REGINT_NEG_ONE_OK, OPAL_INFO_LVL_5));
mca_btl_usnic_component.prio_rd_num = (int32_t) prio_rd_num;
CHECK(reg_int("cq_num", "Number of completion queue entries (-1 = pre-set defaults; depends on number and type of devices available; will error if (sd_num+rd_num)>cq_num)",
-1, &cq_num, REGINT_NEG_ONE_OK, OPAL_INFO_LVL_5));
mca_btl_usnic_component.cq_num = (int32_t) cq_num;
CHECK(reg_int("av_eq_num", "Number of event queue entries for peer address resolution",
1024, &av_eq_num, REGINT_GE_ONE, OPAL_INFO_LVL_5));
mca_btl_usnic_component.av_eq_num = (int32_t) av_eq_num;
CHECK(reg_int("base_udp_port", "Base UDP port to use for usNIC communications. If 0, system will pick the port number. If non-zero, it will be added to each process' local rank to obtain the final port number (default: 0)",
0, &udp_port_base, REGINT_GE_ZERO, OPAL_INFO_LVL_5));
mca_btl_usnic_component.udp_port_base = (int) udp_port_base;
CHECK(reg_int("retrans_timeout", "Number of microseconds before retransmitting a frame",
5000, &mca_btl_usnic_component.retrans_timeout,
REGINT_GE_ONE, OPAL_INFO_LVL_5));
CHECK(reg_int("priority_limit", "Max size of \"priority\" messages (0 = use pre-set defaults; depends on number and type of devices available)",
0, &max_tiny_msg_size,
REGINT_GE_ZERO, OPAL_INFO_LVL_5));
opal_btl_usnic_module_template.max_tiny_msg_size =
(size_t) max_tiny_msg_size;
CHECK(reg_int("eager_limit", "Eager send limit (0 = use pre-set defaults; depends on number and type of devices available)",
0, &eager_limit, REGINT_GE_ZERO, OPAL_INFO_LVL_5));
opal_btl_usnic_module_template.super.btl_eager_limit = eager_limit;
CHECK(reg_int("rndv_eager_limit", "Eager rendezvous limit (0 = use pre-set defaults; depends on number and type of devices available)",
0, &rndv_eager_limit, REGINT_GE_ZERO, OPAL_INFO_LVL_5));
opal_btl_usnic_module_template.super.btl_rndv_eager_limit =
rndv_eager_limit;
CHECK(reg_int("pack_lazy_threshold", "Convertor packing on-the-fly threshold (-1 = always pack eagerly, 0 = always pack lazily, otherwise will pack on the fly if fragment size is > limit)",
USNIC_DFLT_PACK_LAZY_THRESHOLD, &pack_lazy_threshold, REGINT_NEG_ONE_OK, OPAL_INFO_LVL_5));
mca_btl_usnic_component.pack_lazy_threshold = pack_lazy_threshold;
CHECK(reg_int("max_short_packets", "Number of abnormally-short packets received before outputting a warning (0 = never show the warning)",
25, &max_short_packets,
REGINT_GE_ZERO, OPAL_INFO_LVL_5));
mca_btl_usnic_component.max_short_packets = max_short_packets;
/* Default to bandwidth auto-detection */
opal_btl_usnic_module_template.super.btl_bandwidth = 0;
opal_btl_usnic_module_template.super.btl_latency = 2;
/* Show "cannot find route" warnings? */
mca_btl_usnic_component.show_route_failures = true;
CHECK(reg_bool("show_route_failures",
"Whether to show a warning when route failures between MPI process peers are detected (default = 1, enabled; 0 = disabled)",
mca_btl_usnic_component.show_route_failures,
&mca_btl_usnic_component.show_route_failures,
OPAL_INFO_LVL_3));
/* Connectivity verification */
mca_btl_usnic_component.connectivity_enabled = true;
CHECK(reg_bool("connectivity_check",
"Whether to enable the usNIC connectivity check upon first send (default = 1, enabled; 0 = disabled)",
mca_btl_usnic_component.connectivity_enabled,
&mca_btl_usnic_component.connectivity_enabled,
OPAL_INFO_LVL_3));
mca_btl_usnic_component.connectivity_ack_timeout = 250;
CHECK(reg_int("connectivity_ack_timeout",
"Timeout, in milliseconds, while waiting for an ACK while verification connectivity between usNIC interfaces. If 0, the connectivity check is disabled (must be >=0).",
mca_btl_usnic_component.connectivity_ack_timeout,
&mca_btl_usnic_component.connectivity_ack_timeout,
REGINT_GE_ZERO, OPAL_INFO_LVL_3));
mca_btl_usnic_component.connectivity_num_retries = 40;
CHECK(reg_int("connectivity_error_num_retries",
"Number of times to retry usNIC connectivity verification before aborting the MPI job (must be >0).",
mca_btl_usnic_component.connectivity_num_retries,
&mca_btl_usnic_component.connectivity_num_retries,
REGINT_GE_ONE, OPAL_INFO_LVL_3));
mca_btl_usnic_component.connectivity_map_prefix = NULL;
CHECK(reg_string("connectivity_map",
"Write a per-process file containing the usNIC connectivity map. If this parameter is specified, it is the filename prefix emitted by each MPI process. The full filename emitted by each process is of the form: <prefix>-<hostname>.<pid>.<jobid>.<MCW rank>.txt.",
mca_btl_usnic_component.connectivity_map_prefix,
&mca_btl_usnic_component.connectivity_map_prefix,
REGSTR_EMPTY_OK, OPAL_INFO_LVL_3));
return ret;
}
int mca_bcol_iboffload_register_params(void)
{
mca_base_var_enum_t *new_enum;
char *msg;
int ret = OMPI_SUCCESS, tmp;
#define CHECK(expr) do { \
tmp = (expr); \
if (OMPI_SUCCESS != tmp) ret = tmp; \
} while (0)
/* register openib component parameters */
CHECK(reg_int("k_nomial_radix", NULL,
"The radix of the K-nomial tree for scatther-gather type algorithms"
"(starts from 2)", 2, &mca_bcol_iboffload_component.k_nomial_radix,
REGINT_GE_ONE));
CHECK(reg_int("priority", NULL,
"IB offload component priority"
"(from 0(low) to 90 (high))", 90,
&mca_bcol_iboffload_component.super.priority, 0));
CHECK(reg_int("verbose", NULL,
"Output some verbose IB offload BTL information "
"(0 = no output, nonzero = output)", 0,
&mca_bcol_iboffload_component.verbose, 0));
CHECK(reg_bool("warn_default_gid_prefix", NULL,
"Warn when there is more than one active ports and at least one of them connected to the network with only default GID prefix configured (0 = do not warn; any other value = warn)",
true, &mca_bcol_iboffload_component.warn_default_gid_prefix));
CHECK(reg_bool("warn_nonexistent_if", NULL,
"Warn if non-existent devices and/or ports are specified in the bcol_iboffla_if_[in|ex]clude MCA parameters (0 = do not warn; any other value = warn)",
true, &mca_bcol_iboffload_component.warn_nonexistent_if));
CHECK(reg_int("max_pipeline_depth", NULL,
"The maximal number of fragments of the same collective request that can be transferred in parallel", 3,
(int *) &mca_bcol_iboffload_component.max_pipeline_depth, 0));
CHECK(reg_int("max_mqe_tasks", NULL,
"Maximum number of MQEs for each iboffload module",
1024, &mca_bcol_iboffload_component.max_mqe_tasks, 0));
CHECK(reg_int("max_mq_size", NULL,
"Maximum size of each MQ for each iboffload module",
1024, &mca_bcol_iboffload_component.max_mq_size, 0));
CHECK(reg_int("free_list_num", NULL,
"Intial size of free lists (must be >= 1)",
256, &mca_bcol_iboffload_component.free_list_num,
REGINT_GE_ONE));
CHECK(reg_int("free_list_max", NULL,
"Maximum size of free lists "
"(-1 = infinite, otherwise must be >= 0)",
-1, &mca_bcol_iboffload_component.free_list_max,
REGINT_NEG_ONE_OK | REGINT_GE_ONE));
CHECK(reg_int("free_list_inc", NULL,
"Increment size of free lists (must be >= 1)",
32, &mca_bcol_iboffload_component.free_list_inc,
REGINT_GE_ONE));
CHECK(reg_string("mpool", NULL,
"Name of the memory pool to be used (it is unlikely that you will ever want to change this",
"rdma", &mca_bcol_iboffload_component.mpool_name,
0));
CHECK(reg_int("cq_size", "cq_size",
"Size of the OpenFabrics completion "
"queue (will automatically be set to a minimum of "
"(2 * number_of_peers * bcol_iboffload_rd_num))",
1024, &mca_bcol_iboffload_component.cq_size, REGINT_GE_ONE));
CHECK(reg_int("exchange_tree_order", NULL,
"The order of the exchange tree. "
"Must be power of two.",
2, &mca_bcol_iboffload_component.exchange_tree_order, REGINT_GE_ONE));
CHECK(reg_int("knomial_tree_order", NULL,
"The order of the knomial exchange tree. ",
3, &mca_bcol_iboffload_component.knomial_tree_order, REGINT_GE_ONE));
CHECK(reg_int("max_inline_data", "max_inline_data",
"Maximum size of inline data segment "
"(-1 = run-time probe to discover max value, "
"otherwise must be >= 0). "
"If not explicitly set, use max_inline_data from "
"the INI file containing device-specific parameters",
128, (int *) &mca_bcol_iboffload_component.max_inline_data,
REGINT_NEG_ONE_OK | REGINT_GE_ZERO));
#if 0
CHECK(reg_string("pkey", "ib_pkey_val",
"OpenFabrics partition key (pkey) value. "
"Unsigned integer decimal or hex values are allowed (e.g., \"3\" or \"0x3f\") and will be masked against the maximum allowable IB paritition key value (0x7fff)",
"0", &pkey, 0));
/* Pasha
mca_bcol_iboffload_component.pkey_val =
ompi_btl_openib_ini_intify(pkey) & MCA_BTL_IB_PKEY_MASK;
free(pkey);
*/
#endif
CHECK(reg_string("receive_queues", NULL,
"Colon-delimited, comma delimited list of receive queues: P,4096,8,6,4:P,32768,8,6,4",
"P,512,256,192,128", &mca_bcol_iboffload_component.receive_queues,
0));
CHECK(reg_int("qp_ous_rd_atom", NULL,
"InfiniBand outstanding atomic reads (must be >= 0)", 4,
(int *) &mca_bcol_iboffload_component.qp_ous_rd_atom, REGINT_GE_ZERO));
asprintf(&msg, "OpenFabrics MTU, in bytes (if not specified in INI files). Valid values are: %d=256 bytes, %d=512 bytes, %d=1024 bytes, %d=2048 bytes, %d=4096 bytes",
IBV_MTU_256,
IBV_MTU_512,
IBV_MTU_1024,
IBV_MTU_2048,
IBV_MTU_4096);
if (NULL == msg) {
/* Don't try to recover from this */
return OMPI_ERR_OUT_OF_RESOURCE;
}
CHECK(mca_base_var_enum_create("infiniband mtu", mtu_values, &new_enum));
mca_bcol_iboffload_component.mtu = IBV_MTU_1024;
tmp = mca_base_component_var_register(&mca_bcol_iboffload_component.super.bcol_version,
"mtu", msg, MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_READONLY,
&mca_bcol_iboffload_component.mtu);
OBJ_RELEASE(new_enum);
free(msg);
if (0 > tmp) ret = tmp;
tmp = mca_base_var_register_synonym(tmp, "ompi", "bcol", "iboffload", "ib_mtu",
MCA_BASE_VAR_SYN_FLAG_DEPRECATED);
if (0 > tmp) ret = tmp;
CHECK(reg_int("ib_min_rnr_timer", NULL, "InfiniBand minimum "
"\"receiver not ready\" timer, in seconds "
"(must be >= 0 and <= 31)",
1 , &mca_bcol_iboffload_component.min_rnr_timer, 0));
CHECK(reg_int("ib_timeout", NULL, "InfiniBand transmit timeout, plugged into formula: 4.096 microseconds * "
"(2^bcol_iboffload_ib_timeout) (must be >= 0 and <= 31)",
20, &mca_bcol_iboffload_component.timeout, 0));
CHECK(reg_int("ib_retry_count", NULL, "InfiniBand transmit retry count "
"(must be >= 0 and <= 7)",
7, &mca_bcol_iboffload_component.retry_count, 0));
CHECK(reg_int("ib_rnr_retry", NULL, "InfiniBand \"receiver not ready\" "
"retry count; applies *only* to SRQ/XRC queues. PP queues "
"use RNR retry values of 0 because Open MPI performs "
"software flow control to guarantee that RNRs never occur "
"(must be >= 0 and <= 7; 7 = \"infinite\")",
7, &mca_bcol_iboffload_component.rnr_retry, 0));
CHECK(reg_int("ib_max_rdma_dst_ops", NULL, "InfiniBand maximum pending RDMA "
"destination operations "
"(must be >= 0)",
4, &mca_bcol_iboffload_component.max_rdma_dst_ops, REGINT_GE_ZERO));
CHECK(reg_int("ib_service_level", NULL, "InfiniBand service level "
"(must be >= 0 and <= 15)",
0, &mca_bcol_iboffload_component.service_level, 0));
CHECK(reg_int("buffer_alignment", NULL,
"Prefered communication buffer alignment, in bytes "
"(must be > 0 and power of two)",
64, &mca_bcol_iboffload_component.buffer_alignment, REGINT_GE_ZERO));
/* register parmeters controlling message fragementation */
CHECK(reg_int("min_frag_size", NULL,
"Minimum fragment size",
getpagesize(), &mca_bcol_iboffload_component.super.min_frag_size,
REGINT_GE_ONE));
CHECK(reg_int("max_frag_size", NULL,
"Maximum fragment size",
FRAG_SIZE_NO_LIMIT, &mca_bcol_iboffload_component.super.max_frag_size,
REGINT_NONZERO));
CHECK(reg_bool("can_use_user_buffers", NULL,
"User memory can be used by the collective algorithms",
true, &mca_bcol_iboffload_component.super.can_use_user_buffers));
CHECK(reg_int("barrier_mode", NULL,
"Barrier mode: 0 - Recursive doubling; 1 - Recursive K-ing",
0, &mca_bcol_iboffload_component.barrier_mode, REGINT_GE_ZERO));
CHECK(reg_int("max_progress_pull", NULL,
"Max number of progress pull checks",
8, &mca_bcol_iboffload_component.max_progress_pull, REGINT_GE_ZERO));
CHECK(reg_int("use_brucks_smsg_alltoall_rdma", NULL,
"Use brucks algorithm for smsg alltoall and RDMA semantics 1 = No Temp buffer recycling"
"1 = Alg with no Temp Buffer Recycling (faster), 2 = Alg with temp Buffer Recycling (slower)",
0, &mca_bcol_iboffload_component.use_brucks_smsg_alltoall_rdma, 0));
CHECK(reg_int("use_brucks_smsg_alltoall_sr", NULL,
"Use brucks algorithm for smsg alltoall and Send/Recv semantics "
"1 = Alg with RTR (faster), 2 = Alg with RNR (slower)",
0, &mca_bcol_iboffload_component.use_brucks_smsg_alltoall_sr, 0));
CHECK(reg_int("alltoall_bruck_radix", NULL,
"Radix for Bruck algorithm for smsg alltoall",
3, &mca_bcol_iboffload_component.k_alltoall_bruck_radix, 0));
CHECK(reg_int("k_alltoall_bruck_radix", NULL,
"Temp Buffer alignment for Bruck algorithm for smsg alltoall",
64, &mca_bcol_iboffload_component.tmp_buf_alignment, 0));
/*
CHECK(reg_string("if_include", NULL,
"Comma-delimited list of devices/ports to be used (e.g. \"mthca0,mthca1:2\"; empty value means to use all ports found). Mutually exclusive with bcol_iboffload_if_exclude.",
NULL, &mca_bcol_iboffload_component.if_include,
0));
CHECK(reg_string("if_exclude", NULL,
"Comma-delimited list of device/ports to be excluded (empty value means to not exclude any ports). Mutually exclusive with bcol_iboffload_if_include.",
NULL, &mca_bcol_iboffload_component.if_exclude,
0));
*/
CHECK(mca_bcol_iboffload_verify_params());
/* Register any MCA params for the connect pseudo-components */
if (OMPI_SUCCESS == ret) {
ret = ompi_common_ofacm_base_register(&mca_bcol_iboffload_component.super.bcol_version);
}
return ret;
}
