Skip to content

osc/rdma, btl/tcp: fix various issues with osc/rdma #9400

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
wants to merge 9 commits into from
Closed

osc/rdma, btl/tcp: fix various issues with osc/rdma #9400

Changes from 1 commit
Commits
Show all changes
9 commits
Select commit Hold shift + click to select a range
09f8f75
btl/base,ofi: adjust the MCA_BTL_ATOMIC_SUPPORTS_GLOB flag
wzamazon Sep 17, 2021
c8495bd
osc/rdma: fix bugs in evaluating using_cpu_atomics was
wzamazon Sep 20, 2021
131bb67
osc/rdma: fix the fail condition in osc_rdma_new_peer()
wzamazon Sep 21, 2021
4a845a9
osc/rdma: do not use local leader optimization for active message RDMA
wzamazon Sep 21, 2021
3c0ae03
osc/rdma: add pid to shm file name
wzamazon Sep 25, 2021
6c4cb75
osc/rdma: use pointer to btl to replace btl_index in peer
wzamazon Sep 26, 2021
b0197d9
osc/rdma: use btl/self for self communication as last resort
wzamazon Sep 26, 2021
ef501d8
btl/self: claim support of MCA_BTL_ATOMIC_SUPPORTS_GLOB
wzamazon Sep 27, 2021
5d9e5a2
osc/rdma: adjust the way use_cpu_atomics is evaluated
wzamazon Sep 27, 2021
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
18 changes: 14 additions & 4 deletions ompi/mca/osc/rdma/osc_rdma_peer.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -73,7 +73,19 @@ static int ompi_osc_rdma_peer_btl_endpoint (struct ompi_osc_rdma_module_t *modul
}
}

/* unlikely but can happen when creating a peer for self */
if (peer_id == ompi_comm_rank (module->comm)) {
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm really concerned about this approach. I suppose it works, but it feels like the right place for this decision is during the initial btl selection logic.

for (int btl_index = 0 ; btl_index < num_btls ; ++btl_index) {
struct mca_btl_base_module_t *btl;

btl = bml_endpoint->btl_rdma.bml_btls[btl_index].btl;
if (strcmp(btl->btl_component->btl_version.mca_component_name, "self")==0) {
*btl_out = btl;
*endpoint = bml_endpoint->btl_eager.bml_btls[btl_index].btl_endpoint;
return OMPI_SUCCESS;
}
}
}

return OMPI_ERR_UNREACH;
}

Expand All @@ -86,9 +98,7 @@ int ompi_osc_rdma_new_peer (struct ompi_osc_rdma_module_t *module, int peer_id,

/* find a btl/endpoint to use for this peer */
int ret = ompi_osc_rdma_peer_btl_endpoint (module, peer_id, &btl, &endpoint);
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret &&
!(module->selected_btls[0]->btl_atomic_flags & MCA_BTL_ATOMIC_SUPPORTS_GLOB) &&
(peer_id != ompi_comm_rank (module->comm)))) {
if (OPAL_UNLIKELY(OMPI_SUCCESS != ret)) {
return ret;
}

Expand Down