Open registration for c10 thread pool (pytorch#17788)

Summary:
1. Move ATen threadpool & open registration mechanism to C10
2. Move the `global_work_queue` to use this open registration mechanism, to allow users to substitute in their own
Pull Request resolved: pytorch#17788

Reviewed By: zdevito

Differential Revision: D14379707

Pulled By: jamesr66a

fbshipit-source-id: 949662d0024875abf09907d97db927f160c54d45

Author: James Reed

aten/src/ATen/core/thread_pool.cpp renamed to c10/core/thread_pool.cpp

@@ -1,5 +1,4 @@
-#include <ATen/core/ivalue.h>
-#include <ATen/core/thread_pool.h>
+#include <c10/core/thread_pool.h>
 
 namespace c10 {
 
@@ -125,9 +124,32 @@ void setNumThreads(size_t v) {
   }
 }
 
-ThreadPool& global_work_queue() {
-   static ThreadPool thread_pool(num_threads.exchange(-1));
-   return thread_pool;
+TaskThreadPoolBase& global_work_queue() {
+  static std::shared_ptr<TaskThreadPoolBase> pool =
+      ThreadPoolRegistry()->Create("C10", 0, num_threads.exchange(-1), false);
+  return *pool;
 }
 
+C10_DEFINE_SHARED_REGISTRY(
+    ThreadPoolRegistry,
+    TaskThreadPoolBase,
+    int,
+    int,
+    bool);
+
+namespace {
+
+std::shared_ptr<TaskThreadPoolBase> createC10ThreadPool(
+    int device_id,
+    int pool_size,
+    bool create_new) {
+  static std::shared_ptr<TaskThreadPoolBase> pool =
+      std::make_shared<ThreadPool>(pool_size);
+  return pool;
+}
+
+} // namespace
+
+C10_REGISTER_CREATOR(ThreadPoolRegistry, C10, createC10ThreadPool);
+
 } // namespace c10

aten/src/ATen/core/thread_pool.h renamed to c10/core/thread_pool.h

@@ -9,6 +9,8 @@
 
 #include <c10/util/Optional.h>
 #include <c10/util/intrusive_ptr.h>
+#include <c10/util/numa.h>
+#include <c10/util/thread_name.h>
 
 namespace c10 {
 
@@ -17,7 +19,7 @@ struct Future;
 } // namespace ivalue
 
 // TODO: move this to C10 and make it C10_API
-class CAFFE2_API TaskThreadPoolBase {
+class C10_API TaskThreadPoolBase {
  public:
   virtual void run(const std::function<void()>& func) = 0;
 
@@ -36,7 +38,7 @@ class CAFFE2_API TaskThreadPoolBase {
   virtual ~TaskThreadPoolBase() noexcept {}
 };
 
-class CAFFE2_API ThreadPool : public c10::TaskThreadPoolBase {
+class C10_API ThreadPool : public c10::TaskThreadPoolBase {
  protected:
   struct task_element_t {
     bool run_with_id;
@@ -100,8 +102,29 @@ class CAFFE2_API ThreadPool : public c10::TaskThreadPoolBase {
   void main_loop(std::size_t index);
 };
 
-CAFFE2_API void setNumThreads(size_t v);
+C10_API void setNumThreads(size_t v);
 
-CAFFE2_API ThreadPool& global_work_queue();
+C10_API TaskThreadPoolBase& global_work_queue();
+
+class C10_API TaskThreadPool : public c10::ThreadPool {
+ public:
+  explicit TaskThreadPool(
+      std::size_t pool_size,
+      int numa_node_id = -1)
+      : ThreadPool(pool_size, numa_node_id) {}
+
+  // TODO move this to ATen/core/thread_pool.h
+  void init_thread() override {
+    setThreadName("CaffeTaskThread");
+    NUMABind(numa_node_id_);
+  }
+};
+
+C10_DECLARE_SHARED_REGISTRY(
+    ThreadPoolRegistry,
+    TaskThreadPoolBase,
+    int,
+    int,
+    bool);
 
 } // namespace c10

c10/macros/Export.h

@@ -30,6 +30,20 @@
 // C10_BUILD_SHARED_LIB to check whether pytorch is building shared or static
 // libraries.
 
+// For build systems that do not directly depend on CMake and directly build
+// from the source directory (such as Buck), one may not have a cmake_macros.h
+// file at all. In this case, the build system is responsible for providing
+// correct macro definitions corresponding to the cmake_macros.h.in file.
+//
+// In such scenarios, one should define the macro
+//     C10_USING_CUSTOM_GENERATED_MACROS
+// to inform this header that it does not need to include the cmake_macros.h
+// file.
+
+#ifndef C10_USING_CUSTOM_GENERATED_MACROS
+#include "c10/macros/cmake_macros.h"
+#endif // C10_USING_CUSTOM_GENERATED_MACROS
+
 #ifdef _WIN32
 #if defined(C10_BUILD_SHARED_LIBS)
 #define C10_EXPORT __declspec(dllexport)

caffe2/utils/thread_name.cc renamed to c10/util/thread_name.cpp

@@ -1,24 +1,24 @@
-#include "caffe2/utils/thread_name.h"
+#include "c10/util/thread_name.h"
 
 #include <algorithm>
 
 #if defined(__GLIBC__) && !defined(__APPLE__) && !defined(__ANDROID__)
-#define CAFFE2_HAS_PTHREAD_SETNAME_NP
+#define C10_HAS_PTHREAD_SETNAME_NP
 #endif
 
-#ifdef CAFFE2_HAS_PTHREAD_SETNAME_NP
+#ifdef C10_HAS_PTHREAD_SETNAME_NP
 #include <pthread.h>
 #endif
 
-namespace caffe2 {
+namespace c10 {
 
 void setThreadName(std::string name) {
-#ifdef CAFFE2_HAS_PTHREAD_SETNAME_NP
+#ifdef C10_HAS_PTHREAD_SETNAME_NP
   constexpr size_t kMaxThreadName = 15;
   name.resize(std::min(name.size(), kMaxThreadName));
 
   pthread_setname_np(pthread_self(), name.c_str());
 #endif
 }
 
-} // namespace caffe2
+} // namespace c10

c10/util/thread_name.h

@@ -0,0 +1,11 @@
+#pragma once
+
+#include <string>
+
+#include "c10/macros/Export.h"
+
+namespace c10 {
+
+C10_API void setThreadName(std::string name);
+
+} // namespace c10

caffe2/core/net.h

@@ -9,6 +9,7 @@
 #include <unordered_map>
 #include <vector>
 
+#include "c10/core/thread_pool.h"
 #include "c10/util/Registry.h"
 #include "caffe2/core/blob.h"
 #include "caffe2/core/common.h"
@@ -19,7 +20,6 @@
 #include "caffe2/core/workspace.h"
 #include "caffe2/proto/caffe2_pb.h"
 #include "caffe2/utils/simple_queue.h"
-#include "caffe2/utils/thread_pool.h"
 
 C10_DECLARE_string(caffe2_override_executor);
 

caffe2/core/net_async_base.cc

@@ -151,7 +151,7 @@ TaskThreadPoolBase* AsyncNetBase::poolGetter(
   std::unique_lock<std::mutex> pools_lock(pools_mutex_);
   auto pool = pools[device_id][pool_size];
   if (!pool) {
-    pool = ThreadPoolRegistry()->Create(
+    pool = c10::ThreadPoolRegistry()->Create(
         DeviceTypeName(device_type),
         device_id,
         pool_size,
@@ -478,26 +478,6 @@ AsyncNetBase::~AsyncNetBase() {
   }
 }
 
-C10_DEFINE_SHARED_REGISTRY(
-    ThreadPoolRegistry,
-    TaskThreadPoolBase,
-    int,
-    int,
-    bool);
-
-C10_REGISTER_CREATOR(
-    ThreadPoolRegistry,
-    CPU,
-    GetAsyncNetThreadPool<TaskThreadPool, PROTO_CPU>);
-C10_REGISTER_CREATOR(
-    ThreadPoolRegistry,
-    CUDA,
-    GetAsyncNetThreadPool<TaskThreadPool, PROTO_CUDA>);
-C10_REGISTER_CREATOR(
-    ThreadPoolRegistry,
-    HIP,
-    GetAsyncNetThreadPool<TaskThreadPool, PROTO_HIP>);
-
 ExecutionOptions::ExecutionOptions(
     const std::shared_ptr<const NetDef>& net_def) {
   static const std::string kDag = "dag";
@@ -558,3 +538,20 @@ ExecutionOptions::ExecutionOptions(
 }
 
 } // namespace caffe2
+
+namespace c10 {
+
+C10_REGISTER_CREATOR(
+    ThreadPoolRegistry,
+    CPU,
+    caffe2::GetAsyncNetThreadPool<TaskThreadPool, caffe2::PROTO_CPU>);
+C10_REGISTER_CREATOR(
+    ThreadPoolRegistry,
+    CUDA,
+    caffe2::GetAsyncNetThreadPool<TaskThreadPool, caffe2::PROTO_CUDA>);
+C10_REGISTER_CREATOR(
+    ThreadPoolRegistry,
+    HIP,
+    caffe2::GetAsyncNetThreadPool<TaskThreadPool, caffe2::PROTO_HIP>);
+
+} // namespace c10

caffe2/core/net_async_base.h

@@ -1,6 +1,7 @@
 #ifndef CAFFE2_CORE_NET_ASYNC_BASE_H_
 #define CAFFE2_CORE_NET_ASYNC_BASE_H_
 
+#include "c10/core/thread_pool.h"
 #include "c10/util/Registry.h"
 #include "caffe2/core/common.h"
 #include "caffe2/core/net.h"
@@ -12,7 +13,6 @@
 #include "caffe2/proto/caffe2_pb.h"
 #include "caffe2/proto/prof_dag.pb.h"
 #include "caffe2/utils/proto_utils.h"
-#include "caffe2/utils/thread_pool.h"
 
 C10_DECLARE_int(caffe2_streams_per_gpu);
 C10_DECLARE_int(caffe2_net_async_max_gpus);
@@ -167,13 +167,6 @@ class CAFFE2_API AsyncNetBase : public NetBase {
   friend class tracing::Tracer;
 };
 
-C10_DECLARE_SHARED_REGISTRY(
-    ThreadPoolRegistry,
-    TaskThreadPoolBase,
-    int,
-    int,
-    bool);
-
 class AsyncNetExecutorHelper : public ExecutorHelper {
  public:
   explicit AsyncNetExecutorHelper(AsyncNetBase* net) : net_(net) {}

caffe2/core/net_parallel.cc

@@ -132,7 +132,7 @@ TaskThreadPoolBase* ParallelNet::poolGetter(
   std::unique_lock<std::mutex> pools_lock(pools_mutex_);
   auto pool = pools[device_id][pool_size];
   if (!pool) {
-    pool = ThreadPoolRegistry()->Create(
+    pool = c10::ThreadPoolRegistry()->Create(
         DeviceTypeName(device_type),
         device_id,
         pool_size,

caffe2/image/image_input_op.h

@@ -7,14 +7,14 @@
 #include <iostream>
 #include <algorithm>
 
+#include "c10/core/thread_pool.h"
 #include "caffe2/core/common.h"
 #include "caffe2/core/db.h"
+#include "caffe2/image/transform_gpu.h"
+#include "caffe2/operators/prefetch_op.h"
 #include "caffe2/proto/caffe2_legacy.pb.h"
 #include "caffe2/utils/cast.h"
 #include "caffe2/utils/math.h"
-#include "caffe2/utils/thread_pool.h"
-#include "caffe2/operators/prefetch_op.h"
-#include "caffe2/image/transform_gpu.h"
 
 namespace caffe2 {
 

caffe2/utils/CMakeLists.txt

@@ -13,7 +13,6 @@ list(APPEND Caffe2_CPU_SRCS
   utils/signal_handler.cc
   utils/smart_tensor_printer.cc
   utils/string_utils.cc
-  utils/thread_name.cc
   utils/threadpool/ThreadPool.cc)
 
 # ---[ threadpool/pthreadpool* is a local modification of the NNPACK

caffe2/utils/thread_name.h

@@ -3,9 +3,9 @@
 #include <atomic>
 #include <condition_variable>
 #include <thread>
+#include "c10/util/thread_name.h"
 #include "caffe2/core/common.h"
 #include "caffe2/core/logging.h"
-#include "caffe2/utils/thread_name.h"
 
 #if defined(_MSC_VER)
 #include <intrin.h>
@@ -263,7 +263,7 @@ class alignas(kGEMMLOWPCacheLineSize) Worker {
 
   // Thread entry point.
   void ThreadFunc() {
-    setThreadName("CaffeWorkersPool");
+    c10::setThreadName("CaffeWorkersPool");
     ChangeState(State::Ready);
 
     // Thread main loop

caffe2/utils/thread_pool.h

@@ -6,11 +6,11 @@
 #include <random>
 #include <string>
 
+#include <c10/core/thread_pool.h>
 #include <caffe2/core/db.h>
 #include <caffe2/core/logging.h>
 #include <caffe2/operators/prefetch_op.h>
 #include <caffe2/utils/math.h>
-#include <caffe2/utils/thread_pool.h>
 #include <caffe2/video/video_io.h>
 
 namespace caffe2 {

caffe2/utils/threadpool/WorkersPool.h

@@ -13,7 +13,7 @@
 #include <ATen/core/ivalue.h>
 #include <torch/csrc/jit/operator.h>
 #include <torch/csrc/jit/script/jit_exception.h>
-#include <ATen/core/thread_pool.h>
+#include <c10/core/thread_pool.h>
 
 #include <exception>
 #include <iostream>

caffe2/video/video_input_op.h

@@ -14,7 +14,7 @@
 #include <ATen/ExpandUtils.h>
 #include <ATen/WrapDimUtils.h>
 #include <ATen/core/ivalue.h>
-#include <ATen/core/thread_pool.h>
+#include <c10/core/thread_pool.h>
 #include <c10/util/SmallVector.h>
 
 #include <algorithm>