TL/MLX5: simplify CI test to debug

openucx · Mar 11, 2024 · 2dbadc1 · 2dbadc1
1 parent 3594368
commit 2dbadc1
Show file tree

Hide file tree

Showing 2 changed files with 3 additions and 3 deletions.
diff --git a/.ci/scripts/run_tests_ucc_mpi.sh b/.ci/scripts/run_tests_ucc_mpi.sh
@@ -119,7 +119,7 @@ for MT in "" "-T"; do
         tlmlx5_args+=" -x UCC_LOG_LEVEL=debug -x UCC_COLL_TRACE=info "
         echo $CX7_DEV
         tlmlx5_colls="alltoall"
-        mpirun $(mpi_params $PPN) $tlmlx5_args $EXE $MT $TG --mtypes host,cuda -c $tlmlx5_colls -t world -d uint8 -O 0 -m 1:128
+        mpirun $(mpi_params $PPN) $tlmlx5_args /opt/nvidia/src/ucc/build/test/mpi/ucc_test_mpi --mtypes host -c $tlmlx5_colls -t world -d uint8 -O 0 -m 128
     fi
     echo "INFO: UCC MPI unit tests (TL/MLX5) ... DONE"
 

diff --git a/src/components/tl/mlx5/alltoall/alltoall.c b/src/components/tl/mlx5/alltoall/alltoall.c
@@ -83,7 +83,7 @@ ucc_status_t ucc_tl_mlx5_team_init_alltoall(ucc_tl_mlx5_team_t *team)
     node_size = node->group_size;
     nnodes    = ucc_topo_nnodes(topo);
     team_size = UCC_TL_TEAM_SIZE(team);
-
+    // while(1) {;};
     if (!ucc_topo_isoppn(topo)) {
         tl_debug(ctx->super.super.lib,
                  "disabling mlx5 a2a for team with non-uniform ppn, "
@@ -93,7 +93,7 @@ ucc_status_t ucc_tl_mlx5_team_init_alltoall(ucc_tl_mlx5_team_t *team)
     }
     ppn = ucc_topo_max_ppn(topo);
 
-    if (net->status == UCC_SBGP_NOT_EXISTS) {
+    if (nnodes == 1) {
         tl_debug(ctx->super.super.lib,
                  "disabling mlx5 a2a for single node team");
         goto non_fatal_error;