int MPID_nem_scif_get_business_card(int my_rank, char **bc_val_p, int *val_max_sz_p)
{
int mpi_errno = MPI_SUCCESS;
int str_errno = MPIU_STR_SUCCESS;
int ret;
char hostname[512];
uint16_t self;
MPIDI_STATE_DECL(MPID_STATE_MPID_NEM_SCIF_GET_BUSINESS_CARD);
MPIDI_FUNC_ENTER(MPID_STATE_MPID_NEM_SCIF_GET_BUSINESS_CARD);
hostname[sizeof(hostname) - 1] = 0;
gethostname(hostname, sizeof(hostname) - 1);
str_errno =
MPIU_Str_add_string_arg(bc_val_p, val_max_sz_p,
MPIDI_CH3I_HOST_DESCRIPTION_KEY, hostname);
if (str_errno) {
MPIU_ERR_CHKANDJUMP(str_errno == MPIU_STR_NOMEM, mpi_errno,
MPI_ERR_OTHER, "**buscard_len");
MPIU_ERR_SETANDJUMP(mpi_errno, MPI_ERR_OTHER, "**buscard");
}
str_errno = MPIU_Str_add_int_arg(bc_val_p, val_max_sz_p, MPIDI_CH3I_PORT_KEY, listen_port);
if (str_errno) {
MPIU_ERR_CHKANDJUMP(str_errno == MPIU_STR_NOMEM, mpi_errno,
MPI_ERR_OTHER, "**buscard_len");
MPIU_ERR_SETANDJUMP(mpi_errno, MPI_ERR_OTHER, "**buscard");
}
ret = scif_get_nodeIDs(NULL, 0, &self);
MPIU_ERR_CHKANDJUMP2(ret == -1, mpi_errno, MPI_ERR_OTHER,
"**scif_get_nodeIDs", "**scif_get_nodeIDs %s %d",
MPIU_Strerror(errno), errno);
str_errno = MPIU_Str_add_int_arg(bc_val_p, val_max_sz_p, MPIDI_CH3I_NODE_KEY, self);
if (str_errno) {
MPIU_ERR_CHKANDJUMP(str_errno == MPIU_STR_NOMEM, mpi_errno,
MPI_ERR_OTHER, "**buscard_len");
MPIU_ERR_SETANDJUMP(mpi_errno, MPI_ERR_OTHER, "**buscard");
}
fn_exit:
MPIDI_FUNC_EXIT(MPID_STATE_MPID_NEM_SCIF_GET_BUSINESS_CARD);
return mpi_errno;
fn_fail:
goto fn_exit;
}
int mca_btl_scif_module_init (void)
{
int rc;
/* create an endpoint to listen for connections */
mca_btl_scif_module.scif_fd = scif_open ();
if (-1 == mca_btl_scif_module.scif_fd) {
BTL_VERBOSE(("scif_open failed. errno = %d", errno));
return OPAL_ERROR;
}
/* bind the endpoint to a port */
mca_btl_scif_module.port_id.port = scif_bind (mca_btl_scif_module.scif_fd, 0);
if (-1 == mca_btl_scif_module.port_id.port) {
BTL_VERBOSE(("scif_bind failed. errno = %d", errno));
scif_close (mca_btl_scif_module.scif_fd);
mca_btl_scif_module.scif_fd = -1;
return OPAL_ERROR;
}
/* determine this processes node id */
rc = scif_get_nodeIDs (NULL, 0, &mca_btl_scif_module.port_id.node);
if (-1 == rc) {
BTL_VERBOSE(("btl/scif error getting node id of this node"));
return OPAL_ERROR;
}
/* Listen for connections */
/* TODO - base the maximum backlog off something */
rc = scif_listen (mca_btl_scif_module.scif_fd, 64);
if (-1 == rc) {
BTL_VERBOSE(("scif_listen failed. errno = %d", errno));
scif_close (mca_btl_scif_module.scif_fd);
mca_btl_scif_module.scif_fd = -1;
return OPAL_ERROR;
}
BTL_VERBOSE(("btl/scif: listening @ port %u on node %u\n",
mca_btl_scif_module.port_id.port, mca_btl_scif_module.port_id.node));
OBJ_CONSTRUCT(&mca_btl_scif_module.dma_frags, opal_free_list_t);
OBJ_CONSTRUCT(&mca_btl_scif_module.eager_frags, opal_free_list_t);
return OPAL_SUCCESS;
}
};
typedef request& request_reference_type;
typedef const request& request_const_reference_type;
communicator(int argc, char* argv[])
{
instance_ = this;
HAM_DEBUG( HAM_LOG << "FLAG_FALSE = " << FLAG_FALSE << ", NO_BUFFER_INDEX = " << NO_BUFFER_INDEX << std::endl; )
// basic SCIF initialisation
uint16_t nodes[MAX_SCIF_NODES];
uint16_t self;
int err = 0;
num_nodes_ = scif_get_nodeIDs(nodes, MAX_SCIF_NODES, &self);
errno_handler(num_nodes_, "scif_get_nodeIDs");
// command line configuration
ham_process_count = num_nodes_; // number of participating processes
ham_address = self; // this processes' address
ham_host_address = 0; // the address of the host process
ham_host_node = 0; // SCIF network node of the host process, default is 0 (which is the host machines)
// command line options
boost::program_options::options_description desc("HAM Options");
desc.add_options()
("ham-help", "Shows this message")
("ham-process-count", boost::program_options::value(&ham_process_count)->default_value(ham_process_count), "Number of processes the job consists of.")
("ham-address", boost::program_options::value(&ham_address)->default_value(ham_address), "This processes address, between 0 and host-process-count minus 1.")
("ham-host-node", boost::program_options::value(&ham_host_node)->default_value(ham_host_node), "The SCIF network node on which the host process is started (0, the host machine by default).")
