Revert "Refactor distribuetd to use absolute header path (#85780)"

This reverts commit 668082718aefce95ecc1b1c312ea6f127b2c662e. Reverted https://github.com/pytorch/pytorch/pull/85780 on behalf of https://github.com/huydhn due to Sorry for reverting your PR but it breaks build due to a missing file <c10d/Store.hpp>
2025-10-20 21:14:14 +08:00 · 2022-09-30 02:04:29 +00:00
parent 668082718a
commit a50d8864fc
77 changed files with 192 additions and 170 deletions
--- a/caffe2/CMakeLists.txt
+++ b/caffe2/CMakeLists.txt
@ -1317,6 +1317,25 @@ if(BUILD_LITE_INTERPRETER)
 endif()


+# For torch/csrc/distributed/c10d
+function(include_torch_lib_dir target)
+  target_include_directories(${target} PRIVATE $<BUILD_INTERFACE:${TORCH_SRC_DIR}/csrc/distributed>)
+endfunction()
+
+include_torch_lib_dir(torch_cpu)
+if(USE_ROCM)
+  include_torch_lib_dir(torch_hip)
+elseif(USE_CUDA)
+  if(BUILD_SPLIT_CUDA)
+    include_torch_lib_dir(torch_cuda_cpp)
+  else()
+    include_torch_lib_dir(torch_cuda)
+  endif()
+endif()
+if(BUILD_PYTHON)
+    include_torch_lib_dir(torch_python)
+endif()
+
 # Pass USE_DISTRIBUTED to torch_cpu, as some codes in jit/pickler.cpp and
 # jit/unpickler.cpp need to be compiled only when USE_DISTRIBUTED is set
 if(USE_DISTRIBUTED)
--- a/test/cpp/c10d/FileStoreTest.cpp
+++ b/test/cpp/c10d/FileStoreTest.cpp
@ -11,8 +11,8 @@

 #include <gtest/gtest.h>

-#include <torch/csrc/distributed/c10d/FileStore.hpp>
-#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
+#include <c10d/FileStore.hpp>
+#include <c10d/PrefixStore.hpp>

 #ifdef _WIN32
 std::string tmppath() {
--- a/test/cpp/c10d/HashStoreTest.cpp
+++ b/test/cpp/c10d/HashStoreTest.cpp
@ -6,8 +6,8 @@
 #include <iostream>
 #include <thread>

-#include <torch/csrc/distributed/c10d/HashStore.hpp>
-#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
+#include <c10d/HashStore.hpp>
+#include <c10d/PrefixStore.hpp>

 constexpr int64_t kShortStoreTimeoutMillis = 100;

--- a/test/cpp/c10d/ProcessGroupGlooAsyncTest.cpp
+++ b/test/cpp/c10d/ProcessGroupGlooAsyncTest.cpp
@ -2,9 +2,9 @@
 #include <c10/util/irange.h>

 #include <ATen/cuda/CUDAContext.h>
+#include <c10d/FileStore.hpp>
+#include <c10d/ProcessGroupGloo.hpp>
 #include <gtest/gtest.h>
-#include <torch/csrc/distributed/c10d/FileStore.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
 #include "CUDATest.hpp"
 #include "TestUtils.hpp"

--- a/test/cpp/c10d/ProcessGroupGlooTest.cpp
+++ b/test/cpp/c10d/ProcessGroupGlooTest.cpp
@ -17,8 +17,8 @@
 #include <torch/cuda.h>

 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/FileStore.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
+#include <c10d/FileStore.hpp>
+#include <c10d/ProcessGroupGloo.hpp>
 #include "TestUtils.hpp"

 using namespace c10d::test;
--- a/test/cpp/c10d/ProcessGroupMPITest.cpp
+++ b/test/cpp/c10d/ProcessGroupMPITest.cpp
@ -1,7 +1,7 @@
 #include <unistd.h>

 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/ProcessGroupMPI.hpp>
+#include <c10d/ProcessGroupMPI.hpp>

 #include <cstdlib>
 #include <iostream>
--- a/test/cpp/c10d/ProcessGroupNCCLErrorsTest.cpp
+++ b/test/cpp/c10d/ProcessGroupNCCLErrorsTest.cpp
@ -1,9 +1,9 @@
 #include <chrono>

 #include <c10/util/irange.h>
+#include <c10d/FileStore.hpp>
+#include <c10d/ProcessGroupNCCL.hpp>
 #include <torch/csrc/cuda/nccl.h>
-#include <torch/csrc/distributed/c10d/FileStore.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupNCCL.hpp>
 #include "CUDATest.hpp"
 #include "TestUtils.hpp"

--- a/test/cpp/c10d/ProcessGroupNCCLTest.cpp
+++ b/test/cpp/c10d/ProcessGroupNCCLTest.cpp
@ -1,8 +1,8 @@
 #include <chrono>
 #include <iostream>

-#include <torch/csrc/distributed/c10d/FileStore.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupNCCL.hpp>
+#include <c10d/FileStore.hpp>
+#include <c10d/ProcessGroupNCCL.hpp>
 #include "CUDATest.hpp"
 #include "TestUtils.hpp"
 #include "c10d/Types.hpp"
--- a/test/cpp/c10d/StoreTestCommon.hpp
+++ b/test/cpp/c10d/StoreTestCommon.hpp
@ -1,6 +1,6 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/Store.hpp>
 #include "TestUtils.hpp"

 #include <gtest/gtest.h>
--- a/test/cpp/c10d/TCPStoreTest.cpp
+++ b/test/cpp/c10d/TCPStoreTest.cpp
@ -9,8 +9,8 @@

 #include <gtest/gtest.h>

-#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
-#include <torch/csrc/distributed/c10d/TCPStore.hpp>
+#include <c10d/PrefixStore.hpp>
+#include <c10d/TCPStore.hpp>

 constexpr int64_t kShortStoreTimeoutMillis = 100;
 constexpr int64_t kStoreCallbackTimeoutMillis = 5000;
--- a/test/cpp/c10d/example/allreduce.cpp
+++ b/test/cpp/c10d/example/allreduce.cpp
@ -1,6 +1,6 @@
 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/FileStore.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
+#include <c10d/FileStore.hpp>
+#include <c10d/ProcessGroupGloo.hpp>

 using namespace ::c10d;

--- a/test/cpp/rpc/e2e_test_base.h
+++ b/test/cpp/rpc/e2e_test_base.h
@ -1,10 +1,10 @@
 #include <gtest/gtest.h>

+#include <c10d/TCPStore.hpp>
 #include <torch/csrc/distributed/autograd/context/container.h>
 #include <torch/csrc/distributed/autograd/context/context.h>
 #include <torch/csrc/distributed/autograd/engine/dist_engine.h>
 #include <torch/csrc/distributed/autograd/utils.h>
-#include <torch/csrc/distributed/c10d/TCPStore.hpp>
 #include <torch/csrc/distributed/rpc/rref_context.h>
 #include <torch/csrc/distributed/rpc/script_call.h>
 #include <torch/csrc/distributed/rpc/script_remote_call.h>
--- a/test/cpp/rpc/test_e2e_tensorpipe.cpp
+++ b/test/cpp/rpc/test_e2e_tensorpipe.cpp
@ -2,7 +2,7 @@

 #include "e2e_test_base.h"

-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
+#include <c10d/ProcessGroupGloo.hpp>
 #include <torch/csrc/distributed/rpc/request_callback_no_python.h>
 #include <torch/csrc/distributed/rpc/tensorpipe_agent.h>
 #include <torch/torch.h>
--- a/test/cpp_extensions/cpp_c10d_extension.hpp
+++ b/test/cpp_extensions/cpp_c10d_extension.hpp
@ -12,11 +12,11 @@

 #include <pybind11/chrono.h>

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Work.hpp>
-#include <torch/csrc/distributed/c10d/Store.hpp>
-#include <torch/csrc/distributed/c10d/Types.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Work.hpp>
+#include <c10d/Store.hpp>
+#include <c10d/Types.hpp>
+#include <c10d/Utils.hpp>

 namespace c10d {

--- a/torch/CMakeLists.txt
+++ b/torch/CMakeLists.txt
@ -320,6 +320,9 @@ if(USE_DEPLOY)
    if(USE_TENSORPIPE)
      list(APPEND TORCH_PYTHON_COMPILE_DEFINITIONS USE_TENSORPIPE)
    endif()
+
+    # Set c10d-related include directories as well.
+    target_include_directories(torch_python_obj PRIVATE $<BUILD_INTERFACE:${TORCH_SRC_DIR}/csrc/distributed>)
  endif()
  target_compile_definitions(torch_python_obj PRIVATE "-DTHP_BUILD_MAIN_LIB -DUSE_DEPLOY")

--- a/torch/csrc/distributed/c10d/Backend.cpp
+++ b/torch/csrc/distributed/c10d/Backend.cpp
@ -1,6 +1,6 @@
 #include <c10/util/Logging.h>
+#include <c10d/Backend.hpp>
 #include <fmt/format.h>
-#include <torch/csrc/distributed/c10d/Backend.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/Backend.hpp
+++ b/torch/csrc/distributed/c10d/Backend.hpp
@ -10,12 +10,12 @@
 #include <ATen/ATen.h>
 #include <c10/macros/Macros.h>

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Work.hpp>
-#include <torch/csrc/distributed/c10d/Types.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
-#include <torch/csrc/distributed/c10d/debug.h>
-#include <torch/csrc/distributed/c10d/sequence_num.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Work.hpp>
+#include <c10d/Types.hpp>
+#include <c10d/Utils.hpp>
+#include <c10d/debug.h>
+#include <c10d/sequence_num.hpp>

 constexpr auto kDefaultTimeout =
    std::chrono::milliseconds(30 * 60 * 1000);
--- a/torch/csrc/distributed/c10d/FileStore.cpp
+++ b/torch/csrc/distributed/c10d/FileStore.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/FileStore.hpp>
+#include <c10d/FileStore.hpp>

 #include <assert.h>
 #include <fcntl.h>
--- a/torch/csrc/distributed/c10d/FileStore.hpp
+++ b/torch/csrc/distributed/c10d/FileStore.hpp
@ -5,7 +5,7 @@
 #include <mutex>
 #include <unordered_map>

-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/Store.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/GlooDeviceFactory.cpp
+++ b/torch/csrc/distributed/c10d/GlooDeviceFactory.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/GlooDeviceFactory.hpp>
+#include <c10d/GlooDeviceFactory.hpp>

 #ifdef USE_C10D_GLOO

--- a/torch/csrc/distributed/c10d/HashStore.cpp
+++ b/torch/csrc/distributed/c10d/HashStore.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/HashStore.hpp>
+#include <c10d/HashStore.hpp>

 #include <errno.h>
 #include <stdint.h>
--- a/torch/csrc/distributed/c10d/HashStore.hpp
+++ b/torch/csrc/distributed/c10d/HashStore.hpp
@ -6,7 +6,7 @@
 #include <mutex>
 #include <unordered_map>

-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/Store.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/NCCLUtils.cpp
+++ b/torch/csrc/distributed/c10d/NCCLUtils.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/NCCLUtils.hpp>
+#include <c10d/NCCLUtils.hpp>

 #include <c10/util/CallOnce.h>

--- a/torch/csrc/distributed/c10d/Ops.hpp
+++ b/torch/csrc/distributed/c10d/Ops.hpp
@ -1,7 +1,7 @@
 #pragma once

 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>

 namespace c10d {
 namespace ops {
--- a/torch/csrc/distributed/c10d/OpsImpl.cpp
+++ b/torch/csrc/distributed/c10d/OpsImpl.cpp
@ -1,5 +1,5 @@
 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>
 #include <torch/csrc/distributed/c10d/Types.hpp>
 #include <torch/library.h>

--- a/torch/csrc/distributed/c10d/ParamCommsUtils.cpp
+++ b/torch/csrc/distributed/c10d/ParamCommsUtils.cpp
@ -1,6 +1,6 @@
 // (c) Facebook, Inc. and its affiliates. Confidential and proprietary.

-#include <torch/csrc/distributed/c10d/ParamCommsUtils.hpp>
+#include <c10d/ParamCommsUtils.hpp>

 namespace torch {

--- a/torch/csrc/distributed/c10d/PrefixStore.cpp
+++ b/torch/csrc/distributed/c10d/PrefixStore.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
+#include <c10d/PrefixStore.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/PrefixStore.hpp
+++ b/torch/csrc/distributed/c10d/PrefixStore.hpp
@ -1,6 +1,6 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/Store.hpp>
 #include <memory>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/ProcessGroup.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroup.cpp
@ -1,5 +1,5 @@
 #include <ATen/ThreadLocalState.h>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>

 #include <c10/util/Logging.h>
 #include <fmt/format.h>
--- a/torch/csrc/distributed/c10d/ProcessGroup.hpp
+++ b/torch/csrc/distributed/c10d/ProcessGroup.hpp
@ -10,11 +10,11 @@
 #include <ATen/ATen.h>
 #include <c10/macros/Macros.h>

-#include <torch/csrc/distributed/c10d/Types.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
-#include <torch/csrc/distributed/c10d/Work.hpp>
-#include <torch/csrc/distributed/c10d/debug.h>
-#include <torch/csrc/distributed/c10d/sequence_num.hpp>
+#include <c10d/Types.hpp>
+#include <c10d/Utils.hpp>
+#include <c10d/Work.hpp>
+#include <c10d/debug.h>
+#include <c10d/sequence_num.hpp>

 // *************************************************************************
 // PROCESS GROUP collective communication API IS BEING CHANGED BETWEEN
--- a/torch/csrc/distributed/c10d/ProcessGroupGloo.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupGloo.cpp
@ -1,9 +1,9 @@
 #include <c10/util/Exception.h>
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
+#include <c10d/ProcessGroupGloo.hpp>

 #ifdef USE_C10D_GLOO

-#include <torch/csrc/distributed/c10d/GlooDeviceFactory.hpp>
+#include <c10d/GlooDeviceFactory.hpp>
 #include <chrono>
 #include <exception>
 #include <ratio>
--- a/torch/csrc/distributed/c10d/ProcessGroupGloo.hpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupGloo.hpp
@ -18,10 +18,10 @@

 #include <c10/util/hash.h>

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Store.hpp>
-#include <torch/csrc/distributed/c10d/Types.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Store.hpp>
+#include <c10d/Types.hpp>
+#include <c10d/Utils.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/ProcessGroupMPI.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupMPI.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/ProcessGroupMPI.hpp>
+#include <c10d/ProcessGroupMPI.hpp>

 #ifdef USE_C10D_MPI

--- a/torch/csrc/distributed/c10d/ProcessGroupMPI.hpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupMPI.hpp
@ -13,9 +13,9 @@
 #include <ATen/core/ivalue.h>
 #include <ATen/core/ivalue_inl.h>

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Types.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Types.hpp>
+#include <c10d/Utils.hpp>

 #include <c10/util/CallOnce.h>

--- a/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp
@ -1,6 +1,6 @@
-#include <torch/csrc/distributed/c10d/NCCLUtils.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupNCCL.hpp>
-#include <torch/csrc/distributed/c10d/UCCForNCCL.hpp>
+#include <c10d/NCCLUtils.hpp>
+#include <c10d/ProcessGroupNCCL.hpp>
+#include <c10d/UCCForNCCL.hpp>
 #include <sstream>

 #ifdef USE_C10D_NCCL
@ -21,9 +21,9 @@
 #include <c10/util/Logging.h>
 #include <c10/util/Optional.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/ParamCommsUtils.hpp>
-#include <torch/csrc/distributed/c10d/TraceUtils.h>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/ParamCommsUtils.hpp>
+#include <c10d/TraceUtils.h>
+#include <c10d/Utils.hpp>

 #include <torch/csrc/cuda/nccl.h>

--- a/torch/csrc/distributed/c10d/ProcessGroupNCCL.hpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupNCCL.hpp
@ -9,10 +9,10 @@
 #include <thread>
 #include <unordered_map>

-#include <torch/csrc/distributed/c10d/NCCLUtils.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Store.hpp>
-#include <torch/csrc/distributed/c10d/UCCForNCCL.hpp>
+#include <c10d/NCCLUtils.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Store.hpp>
+#include <c10d/UCCForNCCL.hpp>

 #include <ATen/DynamicLibrary.h>
 #include <ATen/cuda/CUDAContext.h>
--- a/torch/csrc/distributed/c10d/ProcessGroupRoundRobin.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupRoundRobin.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/ProcessGroupRoundRobin.hpp>
+#include <c10d/ProcessGroupRoundRobin.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/ProcessGroupRoundRobin.hpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupRoundRobin.hpp
@ -2,7 +2,7 @@

 #include <vector>

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/ProcessGroupUCC.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupUCC.cpp
@ -1,8 +1,8 @@
 #ifdef USE_C10D_UCC

-#include <torch/csrc/distributed/c10d/ProcessGroupUCC.hpp>
-#include <torch/csrc/distributed/c10d/UCCTracing.hpp>
-#include <torch/csrc/distributed/c10d/UCCUtils.hpp>
+#include <c10d/ProcessGroupUCC.hpp>
+#include <c10d/UCCTracing.hpp>
+#include <c10d/UCCUtils.hpp>
 #include <list>
 #include <memory>
 #include <unordered_map>
--- a/torch/csrc/distributed/c10d/ProcessGroupUCC.hpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupUCC.hpp
@ -2,7 +2,7 @@

 #ifdef USE_C10D_UCC

-#include <torch/csrc/distributed/c10d/UCCUtils.hpp>
+#include <c10d/UCCUtils.hpp>

 #include <exception>
 #include <memory>
@ -11,10 +11,10 @@
 #include <thread>
 #include <vector>

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Store.hpp>
-#include <torch/csrc/distributed/c10d/Types.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Store.hpp>
+#include <c10d/Types.hpp>
+#include <c10d/Utils.hpp>
 #ifdef USE_CUDA
 #include <ATen/cuda/CUDAEvent.h>
 #include <c10/cuda/CUDAStream.h>
--- a/torch/csrc/distributed/c10d/ProcessGroupWrapper.cpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupWrapper.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/ProcessGroupWrapper.hpp>
+#include <c10d/ProcessGroupWrapper.hpp>

 #ifdef USE_C10D_GLOO

@ -10,8 +10,8 @@
 #include <c10/util/Optional.h>
 #include <c10/util/intrusive_ptr.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroupGloo.hpp>
 #include <stdexcept>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/ProcessGroupWrapper.hpp
+++ b/torch/csrc/distributed/c10d/ProcessGroupWrapper.hpp
@ -2,10 +2,10 @@

 #ifdef USE_C10D_GLOO

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
-#include <torch/csrc/distributed/c10d/Types.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroupGloo.hpp>
+#include <c10d/Types.hpp>
+#include <c10d/Utils.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/PyProcessGroup.hpp
+++ b/torch/csrc/distributed/c10d/PyProcessGroup.hpp
@ -1,6 +1,6 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>
 #include <torch/csrc/utils/pybind.h>
 #include <torch/csrc/jit/python/pybind_utils.h>

--- a/torch/csrc/distributed/c10d/Store.cpp
+++ b/torch/csrc/distributed/c10d/Store.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/Store.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/TCPStore.cpp
+++ b/torch/csrc/distributed/c10d/TCPStore.cpp
@ -1,5 +1,5 @@
 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/TCPStore.hpp>
+#include <c10d/TCPStore.hpp>

 #include <fcntl.h>
 #include <algorithm>
@ -18,12 +18,12 @@
 #endif

 #ifdef _WIN32
-#include <torch/csrc/distributed/c10d/WinSockUtils.hpp>
+#include <c10d/WinSockUtils.hpp>
 #else
-#include <torch/csrc/distributed/c10d/UnixSockUtils.hpp>
+#include <c10d/UnixSockUtils.hpp>
 #endif

-#include <torch/csrc/distributed/c10d/socket.h>
+#include <c10d/socket.h>

 namespace c10d {
 namespace detail {
--- a/torch/csrc/distributed/c10d/TCPStore.hpp
+++ b/torch/csrc/distributed/c10d/TCPStore.hpp
@ -4,7 +4,7 @@
 #include <cstdint>
 #include <memory>

-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/Store.hpp>

 namespace c10d {
 namespace detail {
--- a/torch/csrc/distributed/c10d/TraceUtils.h
+++ b/torch/csrc/distributed/c10d/TraceUtils.h
@ -1,8 +1,8 @@
 #pragma once

 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/Store.hpp>
-#include <torch/csrc/distributed/c10d/Types.hpp>
+#include <c10d/Store.hpp>
+#include <c10d/Types.hpp>

 #include <sys/types.h>

--- a/torch/csrc/distributed/c10d/UCCTracing.cpp
+++ b/torch/csrc/distributed/c10d/UCCTracing.cpp
@ -1,9 +1,9 @@
 #ifdef USE_C10D_UCC

-#include <torch/csrc/distributed/c10d/UCCTracing.hpp>
-#include <torch/csrc/distributed/c10d/UCCUtils.hpp>
+#include <c10d/UCCTracing.hpp>
+#include <c10d/UCCUtils.hpp>

-#include <torch/csrc/distributed/c10d/ParamCommsUtils.hpp>
+#include <c10d/ParamCommsUtils.hpp>

 #include <sys/stat.h>
 #include <cstdlib>
--- a/torch/csrc/distributed/c10d/UCCTracing.hpp
+++ b/torch/csrc/distributed/c10d/UCCTracing.hpp
@ -2,7 +2,7 @@

 #ifdef USE_C10D_UCC

-#include <torch/csrc/distributed/c10d/UCCUtils.hpp>
+#include <c10d/UCCUtils.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/UCCUtils.cpp
+++ b/torch/csrc/distributed/c10d/UCCUtils.cpp
@ -1,7 +1,7 @@
 #ifdef USE_C10D_UCC

-#include <torch/csrc/distributed/c10d/UCCTracing.hpp>
-#include <torch/csrc/distributed/c10d/UCCUtils.hpp>
+#include <c10d/UCCTracing.hpp>
+#include <c10d/UCCUtils.hpp>

 #include <cctype>
 #include <string>
--- a/torch/csrc/distributed/c10d/UCCUtils.hpp
+++ b/torch/csrc/distributed/c10d/UCCUtils.hpp
@ -2,8 +2,8 @@

 #ifdef USE_C10D_UCC

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Store.hpp>
 #include <ucc/api/ucc.h>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/UnixSockUtils.hpp
+++ b/torch/csrc/distributed/c10d/UnixSockUtils.hpp
@ -1,6 +1,6 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/Utils.hpp>

 namespace c10d {
 namespace tcputil {
--- a/torch/csrc/distributed/c10d/Utils.cpp
+++ b/torch/csrc/distributed/c10d/Utils.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/Utils.hpp>

 #include <algorithm>
 #include <cstring>
--- a/torch/csrc/distributed/c10d/Utils.hpp
+++ b/torch/csrc/distributed/c10d/Utils.hpp
@ -3,7 +3,7 @@
 #include <ATen/ATen.h>
 #include <c10/util/accumulate.h>
 #include <c10/util/irange.h>
-#include <torch/csrc/distributed/c10d/Types.hpp>
+#include <c10d/Types.hpp>

 #ifdef _WIN32
 #include <winsock2.h>
--- a/torch/csrc/distributed/c10d/WinSockUtils.hpp
+++ b/torch/csrc/distributed/c10d/WinSockUtils.hpp
@ -1,6 +1,6 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/Utils.hpp>

 namespace c10d {
 namespace tcputil {
--- a/torch/csrc/distributed/c10d/Work.cpp
+++ b/torch/csrc/distributed/c10d/Work.cpp
@ -1,6 +1,6 @@
 #include <ATen/ThreadLocalState.h>

-#include <torch/csrc/distributed/c10d/Work.hpp>
+#include <c10d/Work.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/comm.cpp
+++ b/torch/csrc/distributed/c10d/comm.cpp
@ -1,11 +1,11 @@
-#include <torch/csrc/distributed/c10d/comm.hpp>
+#include <c10d/comm.hpp>

 #include <deque>

 #include <ATen/core/functional.h>
 #include <c10/util/irange.h>
+#include <c10d/reducer.hpp>
 #include <torch/csrc/distributed/c10d/Ops.hpp>
-#include <torch/csrc/distributed/c10d/reducer.hpp>
 #include <torch/csrc/utils/tensor_flatten.h>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/comm.hpp
+++ b/torch/csrc/distributed/c10d/comm.hpp
@ -2,7 +2,7 @@

 #include <ATen/ATen.h>
 #include <ATen/core/ivalue.h>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>
 #include <torch/csrc/Export.h>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/debug.cpp
+++ b/torch/csrc/distributed/c10d/debug.cpp
@ -4,15 +4,15 @@
 // This source code is licensed under the BSD-style license found in the
 // LICENSE file in the root directory of this source tree.

-#include <torch/csrc/distributed/c10d/debug.h>
+#include <c10d/debug.h>

 #include <algorithm>
 #include <cctype>
 #include <cstdlib>
 #include <string>

-#include <torch/csrc/distributed/c10d/exception.h>
-#include <torch/csrc/distributed/c10d/logging.h>
+#include <c10d/exception.h>
+#include <c10d/logging.h>

 namespace c10d {
 namespace detail {
--- a/torch/csrc/distributed/c10d/default_comm_hooks.cpp
+++ b/torch/csrc/distributed/c10d/default_comm_hooks.cpp
@ -1,10 +1,10 @@
 #include <c10/core/ScalarType.h>
 #include <c10/util/Exception.h>
-#include <torch/csrc/distributed/c10d/default_comm_hooks.hpp>
+#include <c10d/default_comm_hooks.hpp>

+#include <c10d/ProcessGroup.hpp>
+#include <c10d/comm.hpp>
 #include <torch/csrc/distributed/c10d/Ops.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/comm.hpp>
 #include <torch/torch.h>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/default_comm_hooks.hpp
+++ b/torch/csrc/distributed/c10d/default_comm_hooks.hpp
@ -1,7 +1,7 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/comm.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/comm.hpp>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/exception.cpp
+++ b/torch/csrc/distributed/c10d/exception.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/exception.h>
+#include <c10d/exception.h>

 namespace c10d {

--- a/torch/csrc/distributed/c10d/init.cpp
+++ b/torch/csrc/distributed/c10d/init.cpp
@ -1,42 +1,42 @@
 #include <torch/csrc/python_headers.h>

 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/distributed/c10d/FileStore.hpp>
-#include <torch/csrc/distributed/c10d/TCPStore.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/FileStore.hpp>
+#include <c10d/TCPStore.hpp>
+#include <c10d/Utils.hpp>
 #ifndef _WIN32
-#include <torch/csrc/distributed/c10d/HashStore.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupRoundRobin.hpp>
+#include <c10d/HashStore.hpp>
+#include <c10d/ProcessGroupRoundRobin.hpp>
 #endif
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/PyProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/PyProcessGroup.hpp>

 #ifdef USE_C10D_GLOO
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupWrapper.hpp>
+#include <c10d/ProcessGroupGloo.hpp>
+#include <c10d/ProcessGroupWrapper.hpp>
 #endif

 #ifdef USE_C10D_NCCL
-#include <torch/csrc/distributed/c10d/NCCLUtils.hpp>
-#include <torch/csrc/distributed/c10d/ProcessGroupNCCL.hpp>
+#include <c10d/NCCLUtils.hpp>
+#include <c10d/ProcessGroupNCCL.hpp>
 #endif

 #ifdef USE_C10D_MPI
-#include <torch/csrc/distributed/c10d/ProcessGroupMPI.hpp>
+#include <c10d/ProcessGroupMPI.hpp>
 #endif

 #ifdef USE_C10D_UCC
-#include <torch/csrc/distributed/c10d/ProcessGroupUCC.hpp>
+#include <c10d/ProcessGroupUCC.hpp>
 #endif

+#include <c10d/PrefixStore.hpp>
 #include <fmt/format.h>
 #include <pybind11/chrono.h>
-#include <torch/csrc/distributed/c10d/PrefixStore.hpp>

-#include <torch/csrc/distributed/c10d/comm.hpp>
-#include <torch/csrc/distributed/c10d/debug.h>
-#include <torch/csrc/distributed/c10d/logger.hpp>
-#include <torch/csrc/distributed/c10d/reducer.hpp>
+#include <c10d/comm.hpp>
+#include <c10d/debug.h>
+#include <c10d/logger.hpp>
+#include <c10d/reducer.hpp>

 #include <torch/csrc/Exceptions.h>
 #include <torch/csrc/distributed/c10d/Ops.hpp>
--- a/torch/csrc/distributed/c10d/logger.cpp
+++ b/torch/csrc/distributed/c10d/logger.cpp
@ -1,14 +1,14 @@
 #include <c10/util/StringUtil.h>
+#include <c10d/Utils.hpp>
+#include <c10d/debug.h>
+#include <c10d/logger.hpp>
 #include <fmt/format.h>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
-#include <torch/csrc/distributed/c10d/debug.h>
-#include <torch/csrc/distributed/c10d/logger.hpp>
 #include <string>

 #include <c10/util/CallOnce.h>

 #ifdef USE_C10D_GLOO
-#include <torch/csrc/distributed/c10d/ProcessGroupGloo.hpp>
+#include <c10d/ProcessGroupGloo.hpp>
 #endif

 namespace c10d {
--- a/torch/csrc/distributed/c10d/logger.hpp
+++ b/torch/csrc/distributed/c10d/logger.hpp
@ -1,5 +1,5 @@
 #include <c10/util/Logging.h>
-#include <torch/csrc/distributed/c10d/reducer.hpp>
+#include <c10d/reducer.hpp>

 #include <mutex>

--- a/torch/csrc/distributed/c10d/logging.cpp
+++ b/torch/csrc/distributed/c10d/logging.cpp
@ -4,9 +4,9 @@
 // This source code is licensed under the BSD-style license found in the
 // LICENSE file in the root directory of this source tree.

-#include <torch/csrc/distributed/c10d/logging.h>
+#include <c10d/logging.h>

-#include <torch/csrc/distributed/c10d/debug.h>
+#include <c10d/debug.h>

 namespace c10d {
 namespace detail {
--- a/torch/csrc/distributed/c10d/python_comm_hook.cpp
+++ b/torch/csrc/distributed/c10d/python_comm_hook.cpp
@ -1,7 +1,7 @@
 #include <torch/csrc/distributed/c10d/python_comm_hook.h>

 #include <ATen/core/functional.h>
-#include <torch/csrc/distributed/c10d/reducer.hpp>
+#include <c10d/reducer.hpp>
 #include <torch/csrc/jit/python/pybind_utils.h>
 #include <torch/csrc/utils/tensor_flatten.h>

--- a/torch/csrc/distributed/c10d/python_comm_hook.h
+++ b/torch/csrc/distributed/c10d/python_comm_hook.h
@ -1,10 +1,10 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/comm.hpp>
+#include <c10d/comm.hpp>

 #include <ATen/ATen.h>
 #include <ATen/core/ivalue.h>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
+#include <c10d/ProcessGroup.hpp>
 #include <torch/csrc/utils/pybind.h>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/quantization/quantization_gpu.cu
+++ b/torch/csrc/distributed/c10d/quantization/quantization_gpu.cu
@ -1,5 +1,5 @@
 #include <c10/cuda/CUDAGuard.h>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
+#include <c10d/Utils.hpp>
 #include <torch/csrc/distributed/c10d/quantization/quantization_gpu.h>
 #include <torch/csrc/distributed/c10d/quantization/quantization_utils.h>
 #include <torch/library.h>
--- a/torch/csrc/distributed/c10d/reducer.cpp
+++ b/torch/csrc/distributed/c10d/reducer.cpp
@ -1,7 +1,7 @@
-#include <torch/csrc/distributed/c10d/reducer.hpp>
+#include <c10d/reducer.hpp>

-#include <torch/csrc/distributed/c10d/Utils.hpp>
-#include <torch/csrc/distributed/c10d/default_comm_hooks.hpp>
+#include <c10d/Utils.hpp>
+#include <c10d/default_comm_hooks.hpp>

 #include <functional>

@ -11,6 +11,8 @@
 #include <c10/util/Logging.h>
 #include <c10/util/hash.h>
 #include <c10/util/irange.h>
+#include <c10d/comm.hpp>
+#include <c10d/logger.hpp>
 #include <torch/csrc/autograd/engine.h>
 #include <torch/csrc/autograd/function_hook.h>
 #include <torch/csrc/autograd/functions/accumulate_grad.h>
@ -18,8 +20,6 @@
 #include <torch/csrc/autograd/utils/grad_layout_contract.h>
 #include <torch/csrc/autograd/utils/lambda_post_hook.h>
 #include <torch/csrc/distributed/c10d/Ops.hpp>
-#include <torch/csrc/distributed/c10d/comm.hpp>
-#include <torch/csrc/distributed/c10d/logger.hpp>
 #include <torch/csrc/utils/memory.h>

 namespace c10d {
--- a/torch/csrc/distributed/c10d/reducer.hpp
+++ b/torch/csrc/distributed/c10d/reducer.hpp
@ -10,12 +10,12 @@
 #include <ATen/core/ivalue_inl.h>
 #include <c10/macros/Macros.h>
 #include <c10/util/intrusive_ptr.h>
-#include <torch/csrc/distributed/c10d/ProcessGroup.hpp>
-#include <torch/csrc/distributed/c10d/Utils.hpp>
-#include <torch/csrc/distributed/c10d/comm.hpp>
-#include <torch/csrc/distributed/c10d/debug.h>
-#include <torch/csrc/distributed/c10d/reducer_timer.hpp>
-#include <torch/csrc/distributed/c10d/default_comm_hooks.hpp>
+#include <c10d/ProcessGroup.hpp>
+#include <c10d/Utils.hpp>
+#include <c10d/comm.hpp>
+#include <c10d/debug.h>
+#include <c10d/reducer_timer.hpp>
+#include <c10d/default_comm_hooks.hpp>
 #include <torch/csrc/autograd/function.h>
 #include <torch/csrc/autograd/profiler.h>
 #include <torch/csrc/autograd/variable.h>
--- a/torch/csrc/distributed/c10d/reducer_cuda.cpp
+++ b/torch/csrc/distributed/c10d/reducer_cuda.cpp
@ -1,4 +1,4 @@
-#include <torch/csrc/distributed/c10d/reducer_timer.hpp>
+#include <c10d/reducer_timer.hpp>

 #include <ATen/cuda/CUDAEvent.h>
 #include <c10/core/DeviceGuard.h>
--- a/torch/csrc/distributed/c10d/sequence_num.cpp
+++ b/torch/csrc/distributed/c10d/sequence_num.cpp
@ -1,6 +1,6 @@
 #include <ATen/ThreadLocalState.h>
 #include <c10/util/Optional.h>
-#include <torch/csrc/distributed/c10d/sequence_num.hpp>
+#include <c10d/sequence_num.hpp>

 #include <c10/util/Logging.h>

--- a/torch/csrc/distributed/c10d/socket.cpp
+++ b/torch/csrc/distributed/c10d/socket.cpp
@ -4,7 +4,7 @@
 // This source code is licensed under the BSD-style license found in the
 // LICENSE file in the root directory of this source tree.

-#include <torch/csrc/distributed/c10d/socket.h>
+#include <c10d/socket.h>

 #include <cstring>
 #include <system_error>
@ -30,9 +30,9 @@
 #include <fmt/chrono.h>
 #include <fmt/format.h>

-#include <torch/csrc/distributed/c10d/error.h>
-#include <torch/csrc/distributed/c10d/exception.h>
-#include <torch/csrc/distributed/c10d/logging.h>
+#include <c10d/error.h>
+#include <c10d/exception.h>
+#include <c10d/logging.h>

 #include <c10/util/CallOnce.h>

--- a/torch/csrc/distributed/c10d/socket.h
+++ b/torch/csrc/distributed/c10d/socket.h
@ -12,7 +12,7 @@
 #include <string>

 #include <c10/macros/Macros.h>
-#include <torch/csrc/distributed/c10d/exception.h>
+#include <c10d/exception.h>

 namespace c10d {
 namespace detail {
--- a/torch/csrc/distributed/rpc/agent_utils.h
+++ b/torch/csrc/distributed/rpc/agent_utils.h
@ -1,6 +1,6 @@
 #pragma once

-#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
+#include <c10d/PrefixStore.hpp>
 #include <torch/csrc/distributed/rpc/utils.h>

 namespace torch {
--- a/torch/csrc/distributed/rpc/tensorpipe_agent.h
+++ b/torch/csrc/distributed/rpc/tensorpipe_agent.h
@ -6,8 +6,8 @@
 #include <thread>

 #include <c10/core/thread_pool.h>
-#include <torch/csrc/distributed/c10d/PrefixStore.hpp>
-#include <torch/csrc/distributed/c10d/Store.hpp>
+#include <c10d/PrefixStore.hpp>
+#include <c10d/Store.hpp>
 #include <torch/csrc/distributed/rpc/rpc_agent.h>

 // Forward-declare the TensorPipe classes we need, to avoid including its