REVIEW: fix review comments

openucx · Jan 3, 2024 · c2aa99f · c2aa99f
1 parent 3c0dad3
commit c2aa99f
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 8 deletions.
diff --git a/src/components/tl/nccl/tl_nccl.c b/src/components/tl/nccl/tl_nccl.c
@@ -40,8 +40,8 @@ static ucs_config_field_t ucc_tl_nccl_context_config_table[] = {
     },
 
     {"BLOCKING", "yes",
-     "If set to yes will use non-blocking mode communicator behavior, "
-     "if set to no will use blocking mode",
+     "If set to no will use non-blocking mode communicator behavior, "
+     "if set to yes will use blocking mode",
      ucs_offsetof(ucc_tl_nccl_context_config_t, nccl_cfg_blocking),
      UCS_CONFIG_TYPE_BOOL},
 

diff --git a/src/components/tl/nccl/tl_nccl_coll.c b/src/components/tl/nccl/tl_nccl_coll.c
@@ -144,9 +144,11 @@ ucc_status_t ucc_tl_nccl_init_task(ucc_base_coll_args_t *coll_args,
         return UCC_ERR_NOT_SUPPORTED;
     }
 
-    status = ucc_tl_nccl_comm_init(nccl_team);
-    if (ucc_unlikely(status != UCC_OK)) {
-        return status;
+    if (ucc_unlikely(nccl_team->comm_state != TL_NCCL_COMM_STATE_READY)) {
+        status = ucc_tl_nccl_comm_init(nccl_team);
+        if (ucc_unlikely(status != UCC_OK)) {
+            return status;
+        }
     }
 
     task = ucc_mpool_get(&nccl_ctx->req_mp);

diff --git a/src/components/tl/nccl/tl_nccl_team.c b/src/components/tl/nccl/tl_nccl_team.c
@@ -121,6 +121,10 @@ ucc_status_t ucc_tl_nccl_comm_init(ucc_tl_nccl_team_t *team)
     ucc_rank_t   trank    = UCC_TL_TEAM_RANK(team);
     ucc_status_t status;
     ncclResult_t nccl_status;
+#if NCCL_USE_NON_BLOCKING
+    ncclConfig_t nccl_cfg = NCCL_CONFIG_INITIALIZER;
+    ncclResult_t async_status;
+#endif
 
     if (team->comm_state == TL_NCCL_COMM_STATE_READY) {
         return UCC_OK;
@@ -138,9 +142,6 @@ ucc_status_t ucc_tl_nccl_comm_init(ucc_tl_nccl_team_t *team)
                                               cudaStreamNonBlocking),
                     exit_err, status);
 #if NCCL_USE_NON_BLOCKING
-    ncclConfig_t nccl_cfg = NCCL_CONFIG_INITIALIZER;
-    ncclResult_t async_status;
-
     /*
     * if NCCL comm initialized during first call to collective init a.k.a lazy init
     * we need to use blocking init to correctly fallback to other TL in case of error