update

Chao1Han · Sep 4, 2024 · 076db36 · 076db36
1 parent c9ef78f
commit 076db36
Show file tree

Hide file tree

Showing 6 changed files with 11 additions and 6 deletions.
diff --git a/build_variables.bzl b/build_variables.bzl
@@ -542,7 +542,6 @@ libtorch_distributed_extra_sources = [
     "torch/csrc/distributed/autograd/rpc_messages/rref_backward_req.cpp",
     "torch/csrc/distributed/autograd/rpc_messages/rref_backward_resp.cpp",
     "torch/csrc/distributed/c10d/HashStore.cpp",
-    "torch/csrc/distributed/c10d/ProcessGroupXCCL.cpp",
     "torch/csrc/distributed/rpc/agent_utils.cpp",
     "torch/csrc/distributed/rpc/message.cpp",
     "torch/csrc/distributed/rpc/profiler/remote_profiler_manager.cpp",
@@ -787,6 +786,7 @@ libtorch_python_cuda_sources = libtorch_python_cuda_core_sources + [
 
 libtorch_python_xpu_sources = [
     "torch/csrc/xpu/xccl.cpp",
+    "torch/csrc/distributed/c10d/ProcessGroupXCCL.cpp",
     "torch/csrc/xpu/Event.cpp",
     "torch/csrc/xpu/Module.cpp",
     "torch/csrc/xpu/Stream.cpp",

diff --git a/caffe2/CMakeLists.txt b/caffe2/CMakeLists.txt
@@ -1015,13 +1015,12 @@ endif()
 
 if(USE_XPU)
   if(USE_XCCL)
-  list(APPEND Caffe2_XPU_SRCS
-    ${TORCH_SRC_DIR}/csrc/xpu/xccl.cpp)
+    list(APPEND Caffe2_XPU_SRCS
+      ${TORCH_SRC_DIR}/csrc/xpu/xccl.cpp)
   endif()
   add_library(torch_xpu ${Caffe2_XPU_SRCS})
   torch_compile_options(torch_xpu)  # see cmake/public/utils.cmake
   target_compile_definitions(torch_xpu PRIVATE USE_XPU)
-
   # ATen XPU implementation
   set(TORCH_XPU_OPS_DIR ${TORCH_ROOT}/third_party/torch-xpu-ops)
   set(TORCH_XPU_OPS_REPO_URL https://github.com/intel/torch-xpu-ops.git)

diff --git a/cmake/Summary.cmake b/cmake/Summary.cmake
@@ -155,6 +155,7 @@ function(caffe2_print_configuration_summary)
   message(STATUS "  USE_ITT               : ${USE_ITT}")
   message(STATUS "  USE_XCCL              : ${USE_XCCL}")
   if(${USE_XCCL})
+    message(STATUS "    USE_C10D_XCCL       : ${USE_C10D_XCCL}")
     message(STATUS "    XCCL include path   : ${XCCL_INCLUDE_DIR}")
     message(STATUS "    XCCL library        : ${XCCL_LIBRARY}")
   endif()

diff --git a/torch/CMakeLists.txt b/torch/CMakeLists.txt
@@ -165,6 +165,9 @@ if(USE_XPU)
     append_filelist("libtorch_python_xpu_sources" TORCH_PYTHON_SRCS)
 
     list(APPEND TORCH_PYTHON_COMPILE_DEFINITIONS USE_XPU)
+    # if(USE_XCCL)
+    list(APPEND TORCH_PYTHON_LINK_LIBRARIES torch::xpurt)
+    # endif()
 endif()
 
 if(USE_CUDNN OR USE_ROCM)
@@ -419,6 +422,8 @@ endif()
 target_compile_definitions(torch_python PRIVATE "-DTHP_BUILD_MAIN_LIB")
 
 target_link_libraries(torch_python PRIVATE ${TORCH_LIB} ${TORCH_PYTHON_LINK_LIBRARIES})
+target_link_libraries(torch_python PRIVATE torch::xpurt)
+target_link_libraries(torch_python PRIVATE c10_xpu)
 
 target_compile_definitions(torch_python PRIVATE ${TORCH_PYTHON_COMPILE_DEFINITIONS})
 

diff --git a/torch/csrc/distributed/c10d/ProcessGroupXCCL.hpp b/torch/csrc/distributed/c10d/ProcessGroupXCCL.hpp
@@ -11,8 +11,6 @@
 
 #include <oneapi/ccl.hpp>
 #include <torch/csrc/xpu/xccl.h>
-#include <torch/csrc/xpu/Stream.h>
-#include <torch/csrc/xpu/Event.h>
 #include <exception>
 #include <memory>
 #include <vector>

diff --git a/torch/csrc/xpu/xccl.h b/torch/csrc/xpu/xccl.h
@@ -6,6 +6,8 @@
 #include <optional>
 #include <vector>
 #include <torch/csrc/distributed/c10d/Store.hpp>
+#include <torch/csrc/xpu/Stream.h>
+#include <torch/csrc/xpu/Event.h>
 
 namespace torch::xpu::xccl {