TL/UCP: refresh active radix after phase

Author: jeffnvidia

src/components/tl/ucp/allgather/allgather_knomial.c

@@ -63,6 +63,7 @@ void ucc_tl_ucp_allgather_knomial_progress(ucc_coll_task_t *coll_task)
     ucc_coll_args_t       *args      = &TASK_ARGS(task);
     ucc_tl_ucp_team_t     *team      = TASK_TEAM(task);
     ucc_tl_ucp_context_t  *ctx       = UCC_TL_UCP_TEAM_CTX(team);
+    ucc_kn_radix_t         radix     = task->allgather_kn.p.radix;
     uint8_t                node_type = task->allgather_kn.p.node_type;
     ucc_knomial_pattern_t *p         = &task->allgather_kn.p;
     void                  *rbuf      = GET_DST(args);
@@ -152,7 +153,7 @@ void ucc_tl_ucp_allgather_knomial_progress(ucc_coll_task_t *coll_task)
                                    &local_seg_offset);
         sbuf = PTR_OFFSET(rbuf, local_seg_offset * dt_size);
 
-        for (loop_step = p->radix - 1; loop_step > 0; loop_step--) {
+        for (loop_step = radix - 1; loop_step > 0; loop_step--) {
             peer = ucc_knomial_pattern_get_loop_peer(p, rank, loop_step);
             if (peer == UCC_KN_PEER_NULL)
                 continue;
@@ -170,7 +171,7 @@ void ucc_tl_ucp_allgather_knomial_progress(ucc_coll_task_t *coll_task)
                           task, out);
         }
 
-        for (loop_step = 1; loop_step < p->radix; loop_step++) {
+        for (loop_step = 1; loop_step < radix; loop_step++) {
             peer = ucc_knomial_pattern_get_loop_peer(p, rank, loop_step);
             if (peer == UCC_KN_PEER_NULL)
                 continue;
@@ -198,6 +199,7 @@ void ucc_tl_ucp_allgather_knomial_progress(ucc_coll_task_t *coll_task)
             return;
         }
         ucc_kn_ag_pattern_next_iter(p);
+        radix = p->radix;
     }
 
     if (KN_NODE_PROXY == node_type) {