Skip to content

Commit 63f215a

Browse files
committed
oshmem: Add actual symmetric remote keys deduplication calls
Invoke the lookup call, but only for ranks from different node where, scale is actually problematic. Signed-off-by: Thomas Vegas <[email protected]>
1 parent b2d3fde commit 63f215a

File tree

1 file changed

+14
-2
lines changed

1 file changed

+14
-2
lines changed

oshmem/mca/spml/ucx/spml_ucx.c

Lines changed: 14 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -240,6 +240,7 @@ int mca_spml_ucx_ctx_mkey_add(mca_spml_ucx_ctx_t *ucx_ctx, int pe, uint32_t segn
240240
{
241241
int rc;
242242
ucs_status_t err;
243+
ucp_rkey_h rkey;
243244

244245
rc = mca_spml_ucx_ctx_mkey_new(ucx_ctx, pe, segno, ucx_mkey);
245246
if (OSHMEM_SUCCESS != rc) {
@@ -248,11 +249,18 @@ int mca_spml_ucx_ctx_mkey_add(mca_spml_ucx_ctx_t *ucx_ctx, int pe, uint32_t segn
248249
}
249250

250251
if (mkey->u.data) {
251-
err = ucp_ep_rkey_unpack(ucx_ctx->ucp_peers[pe].ucp_conn, mkey->u.data, &((*ucx_mkey)->rkey));
252+
err = ucp_ep_rkey_unpack(ucx_ctx->ucp_peers[pe].ucp_conn, mkey->u.data, &rkey);
252253
if (UCS_OK != err) {
253254
SPML_UCX_ERROR("failed to unpack rkey: %s", ucs_status_string(err));
254255
return OSHMEM_ERROR;
255256
}
257+
258+
if (!oshmem_proc_on_local_node(pe)) {
259+
rkey = mca_spml_ucx_rkey_store_get(&mca_spml_ucx.rkey_store, rkey);
260+
}
261+
262+
(*ucx_mkey)->rkey = rkey;
263+
256264
rc = mca_spml_ucx_ctx_mkey_cache(ucx_ctx, mkey, segno, pe);
257265
if (OSHMEM_SUCCESS != rc) {
258266
SPML_UCX_ERROR("mca_spml_ucx_ctx_mkey_cache failed");
@@ -267,7 +275,11 @@ int mca_spml_ucx_ctx_mkey_del(mca_spml_ucx_ctx_t *ucx_ctx, int pe, uint32_t segn
267275
ucp_peer_t *ucp_peer;
268276
int rc;
269277
ucp_peer = &(ucx_ctx->ucp_peers[pe]);
270-
ucp_rkey_destroy(ucx_mkey->rkey);
278+
if (!oshmem_proc_on_local_node(pe)) {
279+
mca_spml_ucx_rkey_store_put(&mca_spml_ucx.rkey_store, ucx_mkey->rkey);
280+
} else {
281+
ucp_rkey_destroy(ucx_mkey->rkey);
282+
}
271283
ucx_mkey->rkey = NULL;
272284
rc = mca_spml_ucx_peer_mkey_cache_del(ucp_peer, segno);
273285
if(OSHMEM_SUCCESS != rc){

0 commit comments

Comments
 (0)