完成低延迟接口功能

rocshmem_dir/rocshmem/lib/cmake/rocshmem-targets-release.cmake

-../../../lib/cmake/rocshmem/rocshmem-targets-release.cmake
+#----------------------------------------------------------------
\ No newline at end of file
+# Generated CMake target import file for configuration "Release".
+#----------------------------------------------------------------
+
+# Commands may need to know the format version.
+set(CMAKE_IMPORT_FILE_VERSION 1)
+
+# Import target "roc::rocshmem" for configuration "Release"
+set_property(TARGET roc::rocshmem APPEND PROPERTY IMPORTED_CONFIGURATIONS RELEASE)
+set_target_properties(roc::rocshmem PROPERTIES
+  IMPORTED_LINK_INTERFACE_LANGUAGES_RELEASE "CXX"
+  IMPORTED_LOCATION_RELEASE "${_IMPORT_PREFIX}/lib/librocshmem.a"
+  )
+
+list(APPEND _cmake_import_check_targets roc::rocshmem )
+list(APPEND _cmake_import_check_files_for_roc::rocshmem "${_IMPORT_PREFIX}/lib/librocshmem.a" )
+
+# Commands beyond this point should not need to know the version.
+set(CMAKE_IMPORT_FILE_VERSION)

rocshmem_dir/rocshmem/lib/cmake/rocshmem-targets.cmake

-../../../lib/cmake/rocshmem/rocshmem-targets.cmake
+# Generated by CMake
\ No newline at end of file
+
+if("${CMAKE_MAJOR_VERSION}.${CMAKE_MINOR_VERSION}" LESS 2.8)
+   message(FATAL_ERROR "CMake >= 2.8.0 required")
+endif()
+if(CMAKE_VERSION VERSION_LESS "2.8.12")
+   message(FATAL_ERROR "CMake >= 2.8.12 required")
+endif()
+cmake_policy(PUSH)
+cmake_policy(VERSION 2.8.12...3.27)
+#----------------------------------------------------------------
+# Generated CMake target import file.
+#----------------------------------------------------------------
+
+# Commands may need to know the format version.
+set(CMAKE_IMPORT_FILE_VERSION 1)
+
+# Protect against multiple inclusion, which would fail when already imported targets are added once more.
+set(_cmake_targets_defined "")
+set(_cmake_targets_not_defined "")
+set(_cmake_expected_targets "")
+foreach(_cmake_expected_target IN ITEMS roc::rocshmem)
+  list(APPEND _cmake_expected_targets "${_cmake_expected_target}")
+  if(TARGET "${_cmake_expected_target}")
+    list(APPEND _cmake_targets_defined "${_cmake_expected_target}")
+  else()
+    list(APPEND _cmake_targets_not_defined "${_cmake_expected_target}")
+  endif()
+endforeach()
+unset(_cmake_expected_target)
+if(_cmake_targets_defined STREQUAL _cmake_expected_targets)
+  unset(_cmake_targets_defined)
+  unset(_cmake_targets_not_defined)
+  unset(_cmake_expected_targets)
+  unset(CMAKE_IMPORT_FILE_VERSION)
+  cmake_policy(POP)
+  return()
+endif()
+if(NOT _cmake_targets_defined STREQUAL "")
+  string(REPLACE ";" ", " _cmake_targets_defined_text "${_cmake_targets_defined}")
+  string(REPLACE ";" ", " _cmake_targets_not_defined_text "${_cmake_targets_not_defined}")
+  message(FATAL_ERROR "Some (but not all) targets in this export set were already defined.\nTargets Defined: ${_cmake_targets_defined_text}\nTargets not yet defined: ${_cmake_targets_not_defined_text}\n")
+endif()
+unset(_cmake_targets_defined)
+unset(_cmake_targets_not_defined)
+unset(_cmake_expected_targets)
+
+
+# Compute the installation prefix relative to this file.
+get_filename_component(_IMPORT_PREFIX "${CMAKE_CURRENT_LIST_FILE}" PATH)
+get_filename_component(_IMPORT_PREFIX "${_IMPORT_PREFIX}" PATH)
+get_filename_component(_IMPORT_PREFIX "${_IMPORT_PREFIX}" PATH)
+get_filename_component(_IMPORT_PREFIX "${_IMPORT_PREFIX}" PATH)
+if(_IMPORT_PREFIX STREQUAL "/")
+  set(_IMPORT_PREFIX "")
+endif()
+
+# Create imported target roc::rocshmem
+add_library(roc::rocshmem STATIC IMPORTED)
+
+set_target_properties(roc::rocshmem PROPERTIES
+  INTERFACE_COMPILE_OPTIONS "-fgpu-rdc;-fgpu-rdc"
+  INTERFACE_INCLUDE_DIRECTORIES "${_IMPORT_PREFIX}/include;${_IMPORT_PREFIX}/include"
+  INTERFACE_LINK_LIBRARIES "IBVerbs::verbs;numa;\$<\$<BOOL:ON>:MPI::MPI_CXX>;Threads::Threads;hip::device;hip::host;dl;hsa-runtime64::hsa-runtime64;-fgpu-rdc"
+)
+
+# Load information for each installed configuration.
+file(GLOB _cmake_config_files "${CMAKE_CURRENT_LIST_DIR}/rocshmem-targets-*.cmake")
+foreach(_cmake_config_file IN LISTS _cmake_config_files)
+  include("${_cmake_config_file}")
+endforeach()
+unset(_cmake_config_file)
+unset(_cmake_config_files)
+
+# Cleanup temporary variables.
+set(_IMPORT_PREFIX)
+
+# Loop over all imported files and verify that they actually exist
+foreach(_cmake_target IN LISTS _cmake_import_check_targets)
+  if(CMAKE_VERSION VERSION_LESS "3.28"
+      OR NOT DEFINED _cmake_import_check_xcframework_for_${_cmake_target}
+      OR NOT IS_DIRECTORY "${_cmake_import_check_xcframework_for_${_cmake_target}}")
+    foreach(_cmake_file IN LISTS "_cmake_import_check_files_for_${_cmake_target}")
+      if(NOT EXISTS "${_cmake_file}")
+        message(FATAL_ERROR "The imported target \"${_cmake_target}\" references the file
+   \"${_cmake_file}\"
+but this file does not exist.  Possible reasons include:
+* The file was deleted, renamed, or moved to another location.
+* An install or uninstall procedure did not complete successfully.
+* The installation package was faulty and contained
+   \"${CMAKE_CURRENT_LIST_FILE}\"
+but not all the files it references.
+")
+      endif()
+    endforeach()
+  endif()
+  unset(_cmake_file)
+  unset("_cmake_import_check_files_for_${_cmake_target}")
+endforeach()
+unset(_cmake_target)
+unset(_cmake_import_check_targets)
+
+# This file does not depend on other imported targets which have
+# been exported from the same project but in a separate export set.
+
+# Commands beyond this point should not need to know the version.
+set(CMAKE_IMPORT_FILE_VERSION)
+cmake_policy(POP)

tests/test_internode.py

@@ -159,7 +159,6 @@ def test_main(args: argparse.Namespace, num_sms: int,
 
                     # Test combine
                     # torch.cuda.synchronize()
-                    # print("lijian test dipatch end and combine start.")
                     bias_0 = torch.ones((num_tokens, hidden), dtype=torch.bfloat16, device='cuda')
                     bias_1 = torch.randn((num_tokens, hidden), dtype=torch.bfloat16, device='cuda')
                     combine_args = {'x': recv_x, 'handle': handle, 'config': config, 'async_finish': async_mode}
@@ -264,7 +263,7 @@ def test_loop(local_rank: int, num_local_ranks: int, args: argparse.Namespace):
     if args.test_ll_compatibility:
         ll_num_tokens, ll_hidden, ll_num_experts, ll_num_topk = 16, 5120, 256, 9
 
-    num_sms = 24
+    num_sms = 30
     num_qps_per_rank = max(num_sms, ll_num_experts // num_ranks if args.test_ll_compatibility else 0)
 
     hidden_bytes = get_hidden_bytes(args)
@@ -274,7 +273,7 @@ def test_loop(local_rank: int, num_local_ranks: int, args: argparse.Namespace):
         num_rdma_bytes = max(config.get_rdma_buffer_size_hint(hidden_bytes, group.size()), num_rdma_bytes)
 
     buffer = deep_ep.Buffer(group, num_nvl_bytes, num_rdma_bytes, low_latency_mode=args.test_ll_compatibility,
-                            num_qps_per_rank=num_qps_per_rank, explicitly_destroy=True, use_default_stream_as_comm_stream=False)
+                            num_qps_per_rank=num_qps_per_rank, explicitly_destroy=True)
     assert num_local_ranks == 8 and num_ranks > 8
 
     for seed in range(int(1e9)):