From 2a03e3808d48257a71366f5802aeec052914e1cc Mon Sep 17 00:00:00 2001
From: qijun <qijun1994@hotmail.com>
Date: Mon, 17 Jul 2017 16:45:42 +0800
Subject: [PATCH] set correct place for output tensor

---
 paddle/framework/operator.cc | 4 ++--
 paddle/framework/operator.h  | 4 +++-
 paddle/operators/add_op.h    | 4 ++--
 3 files changed, 7 insertions(+), 5 deletions(-)

diff --git a/paddle/framework/operator.cc b/paddle/framework/operator.cc
index 946bde5734b..1a7e3322278 100644
--- a/paddle/framework/operator.cc
+++ b/paddle/framework/operator.cc
@@ -18,14 +18,14 @@ namespace paddle {
 namespace framework {
 
 template <>
-Eigen::DefaultDevice* OpKernel::KernelContext::get_eigen_device<
+Eigen::DefaultDevice* OpKernel::KernelContext::GetEigenDevice<
     platform::CPUPlace, Eigen::DefaultDevice>() const {
   return device_context_.get_eigen_device<Eigen::DefaultDevice>();
 }
 
 #ifndef PADDLE_ONLY_CPU
 template <>
-Eigen::GpuDevice* OpKernel::KernelContext::get_eigen_device<
+Eigen::GpuDevice* OpKernel::KernelContext::GetEigenDevice<
     platform::GPUPlace, Eigen::GpuDevice>() const {
   return device_context_.get_eigen_device<Eigen::GpuDevice>();
 }
diff --git a/paddle/framework/operator.h b/paddle/framework/operator.h
index e6cae9c32b3..b8c5098e49a 100644
--- a/paddle/framework/operator.h
+++ b/paddle/framework/operator.h
@@ -109,7 +109,9 @@ class OpKernel {
     template <typename PlaceType,
               typename DeviceType =
                   typename EigenDeviceConverter<PlaceType>::EigenDeviceType>
-    DeviceType* get_eigen_device() const;
+    DeviceType* GetEigenDevice() const;
+
+    platform::Place GetPlace() const { return device_context_.GetPlace(); }
 
     const OperatorBase& op_;
     const ScopePtr& scope_;
diff --git a/paddle/operators/add_op.h b/paddle/operators/add_op.h
index e8c718669a9..e9a793d23b3 100644
--- a/paddle/operators/add_op.h
+++ b/paddle/operators/add_op.h
@@ -27,9 +27,9 @@ public:
     auto input1 = context.Input(1)->Get<framework::Tensor>();
     auto* output = context.Output(0)->GetMutable<framework::Tensor>();
 
-    output->mutable_data<T>(Place());
+    output->mutable_data<T>(context.GetPlace());
 
-    output->flat<T>().device(*(context.get_eigen_device<Place>())) =
+    output->flat<T>().device(*(context.GetEigenDevice<Place>())) =
         input0.flat<T>() + input1.flat<T>();
   }
 };
-- 
GitLab