add_op kernel implementation

paddle/framework/operator.cc

@@ -17,6 +17,18 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+template <>
+DeviceType* KernelContext::get_eigen_device<CPUPlace>() {
+  return device_context_.get_eigen_device<DeviceType>();
+}
+
+#ifndef PADDLE_ONLY_CPU
+template <>
+DeviceType* KernelContext::get_eigen_device<GPUPlace>() {
+  return device_context_.get_eigen_device<DeviceType>();
+}
+#endif
+
 std::string OperatorBase::DebugString() const {
   std::stringstream ss;
   ss << "=================\n";

paddle/framework/operator.h

@@ -29,6 +29,21 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
+template <typename T>
+struct EigenDeviceConverter;
+
+template <>
+struct EigenDeviceConverter<CPUPlace> {
+  using EigenDeviceType = Eigen::DefaultDevice;
+};
+
+#ifndef PADDLE_ONLY_CPU
+template <>
+struct EigenDeviceConverter<GPUPlace> {
+  using EigenDeviceType = Eigen::GpuDevice;
+};
+#endif
+
 class OperatorBase;
 
 /**
@@ -72,33 +87,39 @@ class OperatorBase {
   AttributeMap attrs_;
 };
 
-class OpKernel {
+/**
+ * KernelContext is the only parameter of Kernel Run function.
+ * Run will get input/output variables, state such as momentum and
+ * device resource such as CUDA stream, cublas handle, etc. from
+ * KernelContext. User should construct it before run the Operator.
+ */
+class KernelContext {
  public:
-  /**
-   * KernelContext is the only parameter of Kernel Run function.
-   * Run will get input/output variables, state such as momentum and
-   * device resource such as CUDA stream, cublas handle, etc. from
-   * KernelContext. User should construct it before run the Operator.
-   */
-  class KernelContext {
-   public:
-    KernelContext(const OperatorBase* op, const std::shared_ptr<Scope>& scope,
-                  const platform::DeviceContext& device_context)
-        : op_(*op), scope_(scope), device_context_(device_context) {}
-
-    const Variable* Input(int index) const {
-      return scope_->GetVariable(op_.inputs_[index]);
-    }
+  KernelContext(const OperatorBase* op, const std::shared_ptr<Scope>& scope,
+                const platform::DeviceContext& device_context)
+      : op_(*op), scope_(scope), device_context_(device_context) {}
 
-    Variable* Output(int index) const {
-      return scope_->GetVariable(op_.outputs_[index]);
-    }
+  const Variable* Input(int index) const {
+    return scope_->GetVariable(op_.inputs_[index]);
+  }
 
-    const OperatorBase& op_;
-    const std::shared_ptr<Scope>& scope_;
-    const platform::DeviceContext& device_context_;
-  };
+  Variable* Output(int index) const {
+    return scope_->GetVariable(op_.outputs_[index]);
+  }
+
+  platform::DeviceContext& device_context() const { return device_context_; }
 
+  template <typename PlaceType, typename DeviceType = EigenDeviceConverter<
+                                    PlaceType>::EigenDeviceType>
+  DeviceType* get_eigen_device();
+
+  const OperatorBase& op_;
+  const std::shared_ptr<Scope>& scope_;
+  const platform::DeviceContext& device_context_;
+};
+
+class OpKernel {
+ public:
   virtual void Compute(const KernelContext& context) const = 0;
 
   virtual ~OpKernel() {}

paddle/framework/tensor.h

@@ -35,7 +35,7 @@ class Tensor {
 
   template <typename T>
 
-  T* data() const {
+  const T* data() const {
     PADDLE_ENFORCE(
         holder_ != nullptr,
         "Tenosr has not been initialized. Call Tensor::mutable_data first.");
@@ -58,6 +58,20 @@ class Tensor {
                                 offset_);
   }
 
+  template <typename T,  // must be POD types
+            typename std::enable_if<std::is_pod<T>::value>::type* = nullptr>
+  T* mutable_data(paddle::platform::Place place) {
+    if (holder_ == nullptr ||
+        !(holder_->Place() ==
+          place) /* some versions of boost::variant don't have operator!= */
+        || holder_->Size() < product(dims_) * sizeof(T) + offset_) {
+      holder_.reset(new PlaceholderImpl<T>(place, product(dims_) * sizeof(T)));
+      offset_ = 0;
+    }
+    return reinterpret_cast<T*>(reinterpret_cast<uintptr_t>(holder_->Ptr()) +
+                                offset_);
+  }
+
   size_t NumElements() const { return product(dims_); }
 
   template <typename T, size_t NDIMS>

paddle/operators/add_op.cc

@@ -1,6 +1,6 @@
-#include <paddle/framework/op_registry.h>
-#include <paddle/framework/tensor.h>
-#include <paddle/operators/add_op.h>
+#include "paddle/operators/add_op.h"
+#include "paddle/framework/op_registry.h"
+#include "paddle/framework/tensor.h"
 
 namespace paddle {
 namespace operators {
@@ -36,9 +36,10 @@ The equation is: Out = X + Y
 )DOC");
   }
 };
-}  // namespace op
+}  // namespace operators
 }  // namespace paddle
 
 REGISTER_OP(add_two, paddle::operators::AddOp, paddle::operators::AddOpMaker);
 REGISTER_OP_CPU_KERNEL(
-    add_two, ::paddle::operators::AddKernel<::paddle::platform::CPUPlace>);
+    add_two,
+    ::paddle::operators::AddKernel<::paddle::platform::CPUPlace, float>);

paddle/operators/add_op.cu

@@ -1,5 +1,7 @@
-#include <paddle/operators/add_op.h>
-#include <paddle/framework/op_registry.h>
+#define EIGEN_USE_GPU
+
+#include "paddle/operators/add_op.h"
+#include "paddle/framework/op_registry.h"
 
 REGISTER_OP_GPU_KERNEL(add_two,
-                       paddle::operators::AddKernel<paddle::platform::GPUPlace>);
+                       paddle::operators::AddKernel<paddle::platform::GPUPlace, float>);

paddle/operators/add_op.h

@@ -1,17 +1,26 @@
 #pragma once
-#include <glog/logging.h>
-#include <paddle/framework/operator.h>
+#include "glog/logging.h"
+#include "paddle/framework/operator.h"
+//#include "paddle/operators/add_op_functor.h"
 
 namespace paddle {
 namespace operators {
 
-template <typename Place>
+// Place can be CPUPlace or GPUPlace
+template <typename Place, typename DataType>
 class AddKernel : public framework::OpKernel {
 public:
-  void Compute(const KernelContext &context) const override {
-    LOG(INFO) << "Add kernel in " << typeid(Place).name();
+  void Compute(const KernelContext& context) const override {
+    auto* input0 = context.Input(0);
+    auto* input1 = context.Input(1);
+
+    auto* output = context.Output(0);
+    output->mutable_data<DataType>(Place());
+
+    output->flat<T>().device(*(context.get_eigen_device<Place>())) =
+        input0->flat<T>() + input1->flat<T>();
   }
 };
 
-}  // namespace op
+}  // namespace operators
 }  // namespace paddle