diff --git a/CMakeLists.txt b/CMakeLists.txt index c8dff7f2f0..6ccaf57f8a 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -29,6 +29,7 @@ option(ENABLE_MSCCL_KERNEL "Enable MSCCL while compiling" option(ENABLE_MSCCLPP "Enable MSCCL++" ON) option(ENABLE_MSCCLPP_CLIP "Enable MSCCL++ CLIP" OFF) option(ENABLE_MSCCLPP_EXECUTOR "Enable MSCCL++ Executor" OFF) +option(ENABLE_NPKIT "Enable NPKit" OFF) option(ENABLE_IFC "Enable indirect function call" OFF) option(INSTALL_DEPENDENCIES "Force install dependencies" OFF) option(ROCTX "Enable ROCTX" ON) @@ -791,9 +792,126 @@ endif() if(HAVE_ROCM_SMI_THREAD_ONLY_MUTEX) target_compile_definitions(rccl PRIVATE USE_ROCM_SMI_THREAD_ONLY_MUTEX) endif() -if(NPKIT_FLAGS) - target_compile_definitions(rccl PRIVATE ${NPKIT_FLAGS}) + +# NPKit flags +## May be better to move these to a separate file +if(ENABLE_NPKIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_TIME_SYNC_GPU) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_TIME_SYNC_CPU) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_RECV_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_RECV_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_RECV_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_RECV_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_RECV_REDUCE_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_RECV_REDUCE_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_SEND_FROM_OUTPUT_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_DIRECT_SEND_FROM_OUTPUT_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_REDUCE_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_RECV_REDUCE_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_FROM_OUTPUT_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_FROM_OUTPUT_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_SIMPLE_WAIT_PEER_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_SIMPLE_WAIT_PEER_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_SIMPLE_REDUCE_OR_COPY_MULTI_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_SIMPLE_REDUCE_OR_COPY_MULTI_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL_WAIT_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL_WAIT_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL_DATA_PROCESS_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL_DATA_PROCESS_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL128_WAIT_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL128_WAIT_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL128_DATA_PROCESS_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_PRIM_LL128_DATA_PROCESS_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_NET_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_NET_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_NET_TEST_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_NET_TEST_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_NET_RECV_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_NET_RECV_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_RECV_REDUCE_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_RECV_REDUCE_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_REDUCE_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_REDUCE_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_REDUCE_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_REDUCE_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_BROADCAST_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_BROADCAST_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_BROADCAST_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_BROADCAST_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_BROADCAST_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_BROADCAST_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_RECV_LOCAL_COPY_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_RECV_LOCAL_COPY_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_RECV_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_RECV_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_RECV_RECV_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_SEND_RECV_RECV_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_RECV_COPY_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_RECV_COPY_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_DIRECT_RECV_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_ALL_GATHER_RING_DIRECT_RECV_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_GENERIC_OP_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_GENERIC_OP_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_REDUCE_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_REDUCE_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_RECV_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_RECV_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_RUN_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_RUN_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_RECV_REDUCE_COPY_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_RECV_REDUCE_COPY_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_INIT_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_MSCCL_INIT_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_BROADCAST_RING_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_BROADCAST_RING_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_SEND_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_SEND_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_COPY_ENTRY) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_COPY_EXIT) + target_compile_definitions(rccl PRIVATE ENABLE_NPKIT_PRIM_COLLECT_DATA_PROCESS_TIME) endif() + if(PROFILE) target_compile_definitions(rccl PRIVATE ENABLE_PROFILING) endif() diff --git a/install.sh b/install.sh index 403f9427fd..cd31682ca1 100755 --- a/install.sh +++ b/install.sh @@ -278,122 +278,8 @@ if [[ "${openmp_test_enabled}" == true ]]; then fi # Enable NPKit -npkit_options="" if [[ "${npkit_enabled}" == true ]]; then - npkit_options="-DENABLE_NPKIT \ - -DENABLE_NPKIT_EVENT_TIME_SYNC_GPU \ - -DENABLE_NPKIT_EVENT_TIME_SYNC_CPU \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_EXIT \ - -DENABLE_NPKIT_EVENT_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_DIRECT_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_DIRECT_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_DIRECT_RECV_ENTRY \ - -DENABLE_NPKIT_EVENT_DIRECT_RECV_EXIT \ - -DENABLE_NPKIT_EVENT_DIRECT_RECV_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_DIRECT_RECV_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_DIRECT_RECV_REDUCE_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_DIRECT_RECV_REDUCE_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_DIRECT_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_DIRECT_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_DIRECT_SEND_FROM_OUTPUT_ENTRY \ - -DENABLE_NPKIT_EVENT_DIRECT_SEND_FROM_OUTPUT_EXIT \ - -DENABLE_NPKIT_EVENT_RECV_ENTRY \ - -DENABLE_NPKIT_EVENT_RECV_EXIT \ - -DENABLE_NPKIT_EVENT_RECV_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_RECV_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_ENTRY \ - -DENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_EXIT \ - -DENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_RECV_REDUCE_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_RECV_REDUCE_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_RECV_REDUCE_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_SEND_FROM_OUTPUT_ENTRY \ - -DENABLE_NPKIT_EVENT_SEND_FROM_OUTPUT_EXIT \ - -DENABLE_NPKIT_EVENT_PRIM_SIMPLE_WAIT_PEER_ENTRY \ - -DENABLE_NPKIT_EVENT_PRIM_SIMPLE_WAIT_PEER_EXIT \ - -DENABLE_NPKIT_EVENT_PRIM_SIMPLE_REDUCE_OR_COPY_MULTI_ENTRY \ - -DENABLE_NPKIT_EVENT_PRIM_SIMPLE_REDUCE_OR_COPY_MULTI_EXIT \ - -DENABLE_NPKIT_EVENT_PRIM_LL_WAIT_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_PRIM_LL_WAIT_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_PRIM_LL_DATA_PROCESS_ENTRY \ - -DENABLE_NPKIT_EVENT_PRIM_LL_DATA_PROCESS_EXIT \ - -DENABLE_NPKIT_EVENT_PRIM_LL128_WAIT_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_PRIM_LL128_WAIT_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_PRIM_LL128_DATA_PROCESS_ENTRY \ - -DENABLE_NPKIT_EVENT_PRIM_LL128_DATA_PROCESS_EXIT \ - -DENABLE_NPKIT_EVENT_NET_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_NET_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_NET_TEST_ENTRY \ - -DENABLE_NPKIT_EVENT_NET_TEST_EXIT \ - -DENABLE_NPKIT_EVENT_NET_RECV_ENTRY \ - -DENABLE_NPKIT_EVENT_NET_RECV_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_RECV_REDUCE_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_RECV_REDUCE_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_REDUCE_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_REDUCE_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_RING_DIRECT_RECV_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_REDUCE_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_REDUCE_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_BROADCAST_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_UPDOWN_BROADCAST_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_BROADCAST_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_BROADCAST_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_REDUCE_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_BROADCAST_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_REDUCE_TREE_SPLIT_BROADCAST_EXIT \ - -DENABLE_NPKIT_EVENT_SEND_RECV_LOCAL_COPY_ENTRY \ - -DENABLE_NPKIT_EVENT_SEND_RECV_LOCAL_COPY_EXIT \ - -DENABLE_NPKIT_EVENT_SEND_RECV_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_SEND_RECV_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_SEND_RECV_RECV_ENTRY \ - -DENABLE_NPKIT_EVENT_SEND_RECV_RECV_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_RECV_COPY_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_RECV_COPY_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_DIRECT_RECV_ENTRY \ - -DENABLE_NPKIT_EVENT_ALL_GATHER_RING_DIRECT_RECV_EXIT \ - -DENABLE_NPKIT_EVENT_MSCCL_GENERIC_OP_ENTRY \ - -DENABLE_NPKIT_EVENT_MSCCL_GENERIC_OP_EXIT \ - -DENABLE_NPKIT_EVENT_MSCCL_REDUCE_ENTRY \ - -DENABLE_NPKIT_EVENT_MSCCL_REDUCE_EXIT \ - -DENABLE_NPKIT_EVENT_MSCCL_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_MSCCL_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_MSCCL_RECV_ENTRY \ - -DENABLE_NPKIT_EVENT_MSCCL_RECV_EXIT \ - -DENABLE_NPKIT_EVENT_MSCCL_RUN_ENTRY \ - -DENABLE_NPKIT_EVENT_MSCCL_RUN_EXIT \ - -DENABLE_NPKIT_EVENT_MSCCL_RECV_REDUCE_COPY_ENTRY \ - -DENABLE_NPKIT_EVENT_MSCCL_RECV_REDUCE_COPY_EXIT \ - -DENABLE_NPKIT_EVENT_MSCCL_INIT_ENTRY \ - -DENABLE_NPKIT_EVENT_MSCCL_INIT_EXIT \ - -DENABLE_NPKIT_EVENT_BROADCAST_RING_ENTRY \ - -DENABLE_NPKIT_EVENT_BROADCAST_RING_EXIT \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_ENTRY \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_EXIT \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_SEND_ENTRY \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_SEND_EXIT \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_COPY_ENTRY \ - -DENABLE_NPKIT_EVENT_REDUCE_SCATTER_RING_RECV_REDUCE_COPY_EXIT \ - -DENABLE_NPKIT_PRIM_COLLECT_DATA_PROCESS_TIME" + cmake_common_options="${cmake_common_options} -DENABLE_NPKIT=ON" fi check_exit_code "$?" @@ -420,9 +306,9 @@ if [[ "${build_tests}" == true ]] || ([[ "${run_tests}" == true ]] && [[ ! -x ./ fi # Initiate RCCL CMake -# Passing NPKIT_FLAGS separately (not as part of ${cmake_common_options}) as -# ${npkit_options} need to be passed "as-is" i.e. with `-D` to CMakeLists.txt -${cmake_executable} ${cmake_common_options} -DNPKIT_FLAGS="${npkit_options}" -DONLY_FUNCS="${ONLY_FUNCS}" ../../. +# Passing ONLY_FUNCS separately (not as part of ${cmake_common_options}) as +# ${ONLY_FUNCS} is a debug-only feature +${cmake_executable} ${cmake_common_options} -DONLY_FUNCS="${ONLY_FUNCS}" ../../. check_exit_code "$?" # Enable verbose output from Makefile