From 2d0f76c1a28570652c4ae4069425407800f021ca Mon Sep 17 00:00:00 2001 From: Thomas Vegas Date: Thu, 23 May 2024 16:22:59 +0300 Subject: [PATCH] Allow get zcopy for smallest sizes --- .ci/run_nccl_tests.sh | 3 +-- src/ucx_uct_lib.c | 6 ++++++ 2 files changed, 7 insertions(+), 2 deletions(-) diff --git a/.ci/run_nccl_tests.sh b/.ci/run_nccl_tests.sh index 6d094d70..c2724a4f 100755 --- a/.ci/run_nccl_tests.sh +++ b/.ci/run_nccl_tests.sh @@ -37,7 +37,6 @@ MPIRUN_OPTIONS_COMMON="\ -x UCX_MEMTYPE_CACHE=n \ -x HCOLL_ENABLE_SHARP=0 \ -x HCOLL_ENABLE_MCAST_ALL=0 \ --x UCX_RC_MLX5_TX_INLINE_RESP=0 \ -mca pml ucx \ -mca coll_hcoll_enable 1 \ --map-by node \ @@ -110,7 +109,7 @@ for TEST_EXE in ${NCCL_TEST_EXE[@]}; do #=================== # Enable ucx_rma tests once this is resolved: https://redmine.mellanox.com/issues/3037941 # for P2P_LAYER in ucx ucx_rma ib - for P2P_LAYER in ucx_uct_read ib ucx_uct ucx; do + for P2P_LAYER in ib ucx ucx_uct ucx_uct_read; do MPIRUN_OPTIONS_PLUGIN_P2P_LAYER="-x NCCL_PLUGIN_P2P=${P2P_LAYER}" #=================== diff --git a/src/ucx_uct_lib.c b/src/ucx_uct_lib.c index cd09d904..3a6441f0 100644 --- a/src/ucx_uct_lib.c +++ b/src/ucx_uct_lib.c @@ -46,6 +46,12 @@ static uct_iface_h nccl_uct_resource_iface_open(uct_worker_h worker, UCXCHECK(uct_md_iface_config_read(md, tl->tl_name, NULL, NULL, &config), return NULL, "read MD iface config for TL '%s'", tl->tl_name); + if (uct_config_modify(config, "RC_MLX5_TX_INLINE_RESP", "0") != UCS_OK) { + WARN("Failed to modify MD configuration for '%s'", tl->tl_name); + uct_config_release(config); + return NULL; + } + params.field_mask = UCT_IFACE_PARAM_FIELD_OPEN_MODE | UCT_IFACE_PARAM_FIELD_DEVICE | UCT_IFACE_PARAM_FIELD_STATS_ROOT | UCT_IFACE_PARAM_FIELD_RX_HEADROOM;