int ompi_mtl_mxm_module_init(void)
{
#if MXM_API < MXM_VERSION(2,0)
ompi_mtl_mxm_ep_conn_info_t ep_info;
#endif
void *ep_address;
size_t ep_address_len;
mxm_error_t err;
uint32_t jobid;
uint64_t mxlr;
ompi_proc_t **procs;
unsigned ptl_bitmap;
size_t totps, proc;
int lr, nlps;
int rc;
mxlr = 0;
lr = -1;
jobid = ompi_mtl_mxm_get_job_id();
if (0 == jobid) {
MXM_ERROR("Failed to generate jobid");
return OMPI_ERROR;
}
if (NULL == (procs = ompi_proc_world(&totps))) {
MXM_ERROR("Unable to obtain process list");
return OMPI_ERROR;
}
if (totps < (size_t)ompi_mtl_mxm.mxm_np) {
MXM_VERBOSE(1, "MXM support will be disabled because of total number "
"of processes (%lu) is less than the minimum set by the "
"mtl_mxm_np MCA parameter (%u)", totps, ompi_mtl_mxm.mxm_np);
return OMPI_ERR_NOT_SUPPORTED;
}
MXM_VERBOSE(1, "MXM support enabled");
if (ORTE_NODE_RANK_INVALID == (lr = ompi_process_info.my_node_rank)) {
MXM_ERROR("Unable to obtain local node rank");
return OMPI_ERROR;
}
nlps = ompi_process_info.num_local_peers + 1;
for (proc = 0; proc < totps; proc++) {
if (OPAL_PROC_ON_LOCAL_NODE(procs[proc]->proc_flags)) {
mxlr = max(mxlr, procs[proc]->proc_name.vpid);
}
}
/* Setup the endpoint options and local addresses to bind to. */
#if MXM_API < MXM_VERSION(2,0)
ptl_bitmap = ompi_mtl_mxm.mxm_ctx_opts->ptl_bitmap;
#else
ptl_bitmap = 0;
#endif
/* Open MXM endpoint */
err = ompi_mtl_mxm_create_ep(ompi_mtl_mxm.mxm_context, &ompi_mtl_mxm.ep,
ptl_bitmap, lr, jobid, mxlr, nlps);
if (MXM_OK != err) {
opal_show_help("help-mtl-mxm.txt", "unable to create endpoint", true,
mxm_error_string(err));
return OMPI_ERROR;
}
/*
* Get address for each PTL on this endpoint, and share it with other ranks.
*/
#if MXM_API < MXM_VERSION(2,0)
if ((ptl_bitmap & MXM_BIT(MXM_PTL_SELF)) &&
OMPI_SUCCESS != ompi_mtl_mxm_get_ep_address(&ep_info, MXM_PTL_SELF)) {
return OMPI_ERROR;
}
if ((ptl_bitmap & MXM_BIT(MXM_PTL_RDMA)) &&
OMPI_SUCCESS != ompi_mtl_mxm_get_ep_address(&ep_info, MXM_PTL_RDMA)) {
return OMPI_ERROR;
}
if ((ptl_bitmap & MXM_BIT(MXM_PTL_SHM)) &&
OMPI_SUCCESS != ompi_mtl_mxm_get_ep_address(&ep_info, MXM_PTL_SHM)) {
return OMPI_ERROR;
}
ep_address = &ep_info;
ep_address_len = sizeof(ep_info);
#else
rc = ompi_mtl_mxm_get_ep_address(&ep_address, &ep_address_len);
if (OMPI_SUCCESS != rc) {
return rc;
}
#endif
rc = ompi_mtl_mxm_send_ep_address(ep_address, ep_address_len);
if (OMPI_SUCCESS != rc) {
MXM_ERROR("Modex session failed.");
return rc;
}
#if MXM_API >= MXM_VERSION(2,0)
free(ep_address);
#endif
/* Register the MXM progress function */
opal_progress_register(ompi_mtl_mxm_progress);
#if MXM_API >= MXM_VERSION(2,0)
if (ompi_mtl_mxm.using_mem_hooks) {
opal_mem_hooks_register_release(ompi_mtl_mxm_mem_release_cb, NULL);
}
#endif
return OMPI_SUCCESS;
}
static mxm_error_t
ompi_mtl_mxm_create_ep(mxm_h ctx, mxm_ep_h *ep, unsigned ptl_bitmap, int lr,
uint32_t jobid, uint64_t mxlr, int nlps)
{
mxm_error_t err;
#if MXM_API < MXM_VERSION(1,5)
mxm_ep_opts_t ep_opt;
struct sockaddr_mxm_local_proc sa_bind_self;
struct sockaddr_mxm_ib_local sa_bind_rdma;
struct sockaddr_mxm_shm_proc sa_bind_shm;
mxm_fill_ep_opts(&ep_opt);
sa_bind_self.sa_family = AF_MXM_LOCAL_PROC;
sa_bind_self.context_id = lr;
sa_bind_rdma.sa_family = AF_MXM_IB_LOCAL;
sa_bind_rdma.lid = 0;
sa_bind_rdma.pkey = 0;
sa_bind_rdma.qp_num = 0;
sa_bind_rdma.sl = 0;
sa_bind_shm.sa_family = AF_MXM_SHM_PROC;
sa_bind_shm.jobid = jobid;
sa_bind_shm.process_id = lr;
sa_bind_shm.context_id = mxlr;
sa_bind_shm.num_procs = nlps;
ep_opt.ptl_bind_addr[MXM_PTL_SELF] =
(ptl_bitmap & MXM_BIT(MXM_PTL_SELF)) ?
(struct sockaddr*) &sa_bind_self : NULL;
ep_opt.ptl_bind_addr[MXM_PTL_RDMA] =
(ptl_bitmap & MXM_BIT(MXM_PTL_RDMA)) ?
(struct sockaddr*) &sa_bind_rdma : NULL;
ep_opt.ptl_bind_addr[MXM_PTL_SHM] =
(ptl_bitmap & MXM_BIT(MXM_PTL_SHM)) ?
(struct sockaddr*) &sa_bind_shm : NULL;
MXM_VERBOSE(1, "MXM version is old, consider to upgrade");
err = mxm_ep_create(ctx, &ep_opt, ep);
#elif MXM_API < MXM_VERSION(2,0)
mxm_ep_opts_t *ep_opts;
err = mxm_config_read_ep_opts(&ep_opts);
if (err != MXM_OK) {
MXM_ERROR("Failed to parse MXM configuration");
return err;
}
ep_opts->job_id = jobid;
ep_opts->local_rank = lr;
ep_opts->num_local_procs = nlps;
err = mxm_ep_create(ctx, ep_opts, ep);
mxm_config_free(ep_opts);
#else
mxm_ep_opts_t *ep_opts;
err = mxm_config_read_ep_opts(&ep_opts);
if (err != MXM_OK) {
MXM_ERROR("Failed to parse MXM configuration");
return err;
}
err = mxm_ep_create(ctx, ep_opts, ep);
mxm_config_free_ep_opts(ep_opts);
#endif
return err;
}
