From 10895f0ea5ae2ed371cba1cd1250921418ad4a43 Mon Sep 17 00:00:00 2001
From: Robert Maynard <robert.maynard@kitware.com>
Date: Thu, 3 Jan 2019 15:03:35 -0500
Subject: [PATCH] RuntimeDeviceTracker DeepyCopy is better optimized.

The DeepCopy method is used when a ScopedGlobalRuntimeDeviceTracker
is constructed. This in turn causes the rebuilding of the device
names and states which isn't a free operation. Now we copy the already
computed information.

This was noticeable when using ArrayHandleTransform since it uses
ScopedGlobalRuntimeDeviceTracker when construction host side
portals.
---
 vtkm/cont/RuntimeDeviceTracker.cxx | 69 +++++++++++++++++-------------
 vtkm/cont/RuntimeDeviceTracker.h   |  8 ++--
 2 files changed, 45 insertions(+), 32 deletions(-)

diff --git a/vtkm/cont/RuntimeDeviceTracker.cxx b/vtkm/cont/RuntimeDeviceTracker.cxx
index 8a454758d..244afba0a 100644
--- a/vtkm/cont/RuntimeDeviceTracker.cxx
+++ b/vtkm/cont/RuntimeDeviceTracker.cxx
@@ -69,6 +69,10 @@ struct VTKM_NEVER_EXPORT GetDeviceNameFunctor
   }
 };
 
+#if !(defined(VTKM_CLANG) && (__apple_build_version__ < 8000000))
+thread_local static vtkm::cont::RuntimeDeviceTracker runtimeDeviceTracker;
+#endif
+
 } // end anon namespace
 
 namespace vtkm
@@ -101,7 +105,7 @@ struct RuntimeDeviceTrackerFunctor
 
 VTKM_CONT
 RuntimeDeviceTracker::RuntimeDeviceTracker()
-  : Internals(new detail::RuntimeDeviceTrackerInternals)
+  : Internals(std::make_shared<detail::RuntimeDeviceTrackerInternals>())
 {
   GetDeviceNameFunctor functor(this->Internals->DeviceNames);
   vtkm::ListForEach(functor, VTKM_DEFAULT_DEVICE_ADAPTER_LIST_TAG());
@@ -176,9 +180,7 @@ void RuntimeDeviceTracker::Reset()
 VTKM_CONT
 vtkm::cont::RuntimeDeviceTracker RuntimeDeviceTracker::DeepCopy() const
 {
-  vtkm::cont::RuntimeDeviceTracker dest;
-  dest.DeepCopy(*this);
-  return dest;
+  return vtkm::cont::RuntimeDeviceTracker(this->Internals);
 }
 
 VTKM_CONT
@@ -188,6 +190,15 @@ void RuntimeDeviceTracker::DeepCopy(const vtkm::cont::RuntimeDeviceTracker& src)
     src.Internals->RuntimeValid, VTKM_MAX_DEVICE_ADAPTER_ID, this->Internals->RuntimeValid);
 }
 
+VTKM_CONT
+RuntimeDeviceTracker::RuntimeDeviceTracker(
+  const std::shared_ptr<detail::RuntimeDeviceTrackerInternals>& internals)
+  : Internals(std::make_shared<detail::RuntimeDeviceTrackerInternals>())
+{
+  std::copy_n(internals->RuntimeValid, VTKM_MAX_DEVICE_ADAPTER_ID, this->Internals->RuntimeValid);
+  std::copy_n(internals->DeviceNames, VTKM_MAX_DEVICE_ADAPTER_ID, this->Internals->DeviceNames);
+}
+
 VTKM_CONT
 void RuntimeDeviceTracker::ForceDeviceImpl(vtkm::cont::DeviceAdapterId deviceId, bool runtimeExists)
 {
@@ -208,31 +219,6 @@ void RuntimeDeviceTracker::ForceDeviceImpl(vtkm::cont::DeviceAdapterId deviceId,
   this->Internals->RuntimeValid[deviceId.GetValue()] = runtimeExists;
 }
 
-VTKM_CONT
-vtkm::cont::RuntimeDeviceTracker GetGlobalRuntimeDeviceTracker()
-{
-#if defined(VTKM_CLANG) && (__apple_build_version__ < 8000000)
-  static std::mutex mtx;
-  static std::map<std::thread::id, vtkm::cont::RuntimeDeviceTracker> globalTrackers;
-  std::thread::id this_id = std::this_thread::get_id();
-
-  std::unique_lock<std::mutex> lock(mtx);
-  auto iter = globalTrackers.find(this_id);
-  if (iter != globalTrackers.end())
-  {
-    return iter->second;
-  }
-  else
-  {
-    vtkm::cont::RuntimeDeviceTracker tracker;
-    globalTrackers[this_id] = tracker;
-    return tracker;
-  }
-#else
-  return runtimeDeviceTracker;
-#endif
-}
-
 VTKM_CONT
 void RuntimeDeviceTracker::ForceDevice(DeviceAdapterId id)
 {
@@ -302,5 +288,30 @@ DeviceAdapterId RuntimeDeviceTracker::GetDeviceAdapterId(DeviceAdapterNameType n
 
   return vtkm::cont::DeviceAdapterTagUndefined{};
 }
+
+VTKM_CONT
+vtkm::cont::RuntimeDeviceTracker GetGlobalRuntimeDeviceTracker()
+{
+#if defined(VTKM_CLANG) && (__apple_build_version__ < 8000000)
+  static std::mutex mtx;
+  static std::map<std::thread::id, vtkm::cont::RuntimeDeviceTracker> globalTrackers;
+  std::thread::id this_id = std::this_thread::get_id();
+
+  std::unique_lock<std::mutex> lock(mtx);
+  auto iter = globalTrackers.find(this_id);
+  if (iter != globalTrackers.end())
+  {
+    return iter->second;
+  }
+  else
+  {
+    vtkm::cont::RuntimeDeviceTracker tracker;
+    globalTrackers[this_id] = tracker;
+    return tracker;
+  }
+#else
+  return runtimeDeviceTracker;
+#endif
+}
 }
 } // namespace vtkm::cont
diff --git a/vtkm/cont/RuntimeDeviceTracker.h b/vtkm/cont/RuntimeDeviceTracker.h
index a17bfb181..dd97e4e61 100644
--- a/vtkm/cont/RuntimeDeviceTracker.h
+++ b/vtkm/cont/RuntimeDeviceTracker.h
@@ -217,6 +217,11 @@ public:
 private:
   std::shared_ptr<detail::RuntimeDeviceTrackerInternals> Internals;
 
+  // Deep Copy constructor.
+  VTKM_CONT_EXPORT
+  VTKM_CONT
+  RuntimeDeviceTracker(const std::shared_ptr<detail::RuntimeDeviceTrackerInternals>& internals);
+
   VTKM_CONT_EXPORT
   VTKM_CONT
   void CheckDevice(vtkm::cont::DeviceAdapterId deviceId) const;
@@ -243,9 +248,6 @@ private:
 /// so that these choices are marked and shared.
 ///
 /// Xcode's clang only supports thread_local from version 8
-#if !(defined(VTKM_CLANG) && (__apple_build_version__ < 8000000))
-thread_local static vtkm::cont::RuntimeDeviceTracker runtimeDeviceTracker;
-#endif
 VTKM_CONT_EXPORT
 VTKM_CONT
 vtkm::cont::RuntimeDeviceTracker GetGlobalRuntimeDeviceTracker();