Move intraop_launch_future from Parallel.h (#64166)

Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/64166

Test Plan: Imported from OSS

Reviewed By: gchanan

Differential Revision: D30728585

Pulled By: dagitses

fbshipit-source-id: 75a41418ae9218bec9bac27597051295222b6eee

aten/src/ATen/Parallel-inl.h

@@ -1,5 +1,6 @@
 #pragma once
 
+#include <c10/util/Exception.h>
 #include <c10/util/SmallVector.h>
 
 namespace at {

aten/src/ATen/Parallel.h

@@ -1,7 +1,7 @@
 #pragma once
 #include <ATen/Config.h>
-#include <ATen/core/ivalue.h>
 #include <c10/macros/Macros.h>
+#include <functional>
 
 namespace at {
 
@@ -144,10 +144,6 @@ void launch_no_thread_state(std::function<void()> fn);
 // Launches intra-op parallel task
 TORCH_API void intraop_launch(std::function<void()> func);
 
-// Launches intra-op parallel task, returns a future
-TORCH_API c10::intrusive_ptr<c10::ivalue::Future> intraop_launch_future(
-    std::function<void()> func);
-
 // Returns number of intra-op threads used by default
 TORCH_API int intraop_default_num_threads();
 

aten/src/ATen/ParallelFuture.h

@@ -0,0 +1,13 @@
+#pragma once
+
+#include <ATen/core/ivalue.h>
+#include <c10/macros/Macros.h>
+#include <functional>
+
+namespace at {
+
+// Launches intra-op parallel task, returns a future
+TORCH_API c10::intrusive_ptr<c10::ivalue::Future> intraop_launch_future(
+    std::function<void()> func);
+
+}  // namespace at

aten/src/ATen/ParallelNative.cpp

@@ -1,6 +1,7 @@
 #include <ATen/Config.h>
 #if AT_PARALLEL_NATIVE
 #include <ATen/Parallel.h>
+#include <ATen/ParallelFuture.h>
 #include <ATen/PTThreadPool.h>
 
 #ifndef C10_MOBILE

aten/src/ATen/ParallelNative.h

@@ -4,6 +4,8 @@
 #include <cstddef>
 #include <exception>
 
+#include <c10/util/Exception.h>
+
 #define INTRA_OP_PARALLEL
 
 namespace at {

aten/src/ATen/ParallelNativeTBB.cpp

@@ -1,6 +1,7 @@
 #include <ATen/Config.h>
 #if AT_PARALLEL_NATIVE_TBB
 #include <ATen/Parallel.h>
+#include <ATen/ParallelFuture.h>
 #include <ATen/PTThreadPool.h>
 
 #include <atomic>

aten/src/ATen/ParallelNativeTBB.h

@@ -4,6 +4,8 @@
 #include <cstddef>
 #include <exception>
 
+#include <c10/util/Exception.h>
+
 #ifdef _WIN32
 #ifndef WIN32_LEAN_AND_MEAN
 #define WIN32_LEAN_AND_MEAN

aten/src/ATen/ParallelOpenMP.cpp

@@ -2,6 +2,7 @@
 #include <ATen/core/jit_type.h>
 #if AT_PARALLEL_OPENMP
 #include <ATen/Parallel.h>
+#include <ATen/ParallelFuture.h>
 
 #include <atomic>
 

aten/src/ATen/test/test_parallel.cpp

@@ -3,6 +3,7 @@
 #include <ATen/ATen.h>
 #include <ATen/DLConvertor.h>
 #include <ATen/Parallel.h>
+#include <ATen/ParallelFuture.h>
 
 #include <iostream>
 // NOLINTNEXTLINE(modernize-deprecated-headers)

binaries/at_launch_benchmark.cc

@@ -3,6 +3,7 @@
 #include "c10/util/Flags.h"
 #include "caffe2/core/init.h"
 
+#include <atomic>
 #include <chrono>
 #include <condition_variable>
 #include <iostream>