Merge pull request #5484 from qingqing01/cmake_speed

Remove fill_constant_batch_size_like_op.h and clean some operator codes.

doc/howto/dev/new_op_cn.md

@@ -214,7 +214,7 @@ MulOp(const std::string &type, const framework::VariableNameMap &inputs,
 
     ```cpp
     // if use Eigen unsupported module before include head files
-    #define EIGEN_USE_GPU
+    // #define EIGEN_USE_GPU
 
     namespace ops = paddle::operators;
     REGISTER_OP_GPU_KERNEL(mul, ops::MulKernel<paddle::platform::GPUPlace, float>);

paddle/operators/accuracy_op.h

@@ -14,7 +14,6 @@ limitations under the License. */
 
 #pragma once
 #include <algorithm>
-#include "paddle/framework/eigen.h"
 #include "paddle/framework/op_registry.h"
 
 namespace paddle {
@@ -22,18 +21,6 @@ namespace operators {
 
 using Tensor = framework::Tensor;
 
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;
-
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using EigenVector = framework::EigenVector<T, MajorType, IndexType>;
-
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using EigenScalar = framework::EigenScalar<T, MajorType, IndexType>;
-
 template <typename Place, typename T>
 class AccuracyKernel : public framework::OpKernel<T> {
  public:

paddle/operators/batch_norm_op.cc

@@ -19,9 +19,6 @@ namespace operators {
 
 using Tensor = framework::Tensor;
 using LoDTensor = framework::LoDTensor;
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;
 
 template <typename T>
 using EigenArrayMap =

paddle/operators/fill_constant_batch_size_like_op.cu

@@ -12,7 +12,6 @@
    See the License for the specific language governing permissions and
    limitations under the License. */
 
-#define EIGEN_USE_GPU
 #include "paddle/framework/op_registry.h"
 #include "paddle/operators/fill_constant_batch_size_like_op.h"
 

paddle/operators/fill_constant_batch_size_like_op.h

@@ -13,8 +13,8 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #pragma once
-#include "paddle/framework/eigen.h"
 #include "paddle/framework/op_registry.h"
+#include "paddle/operators/math/math_function.h"
 
 namespace paddle {
 namespace operators {
@@ -27,9 +27,8 @@ class FillConstantBatchSizeLikeOpKernel : public framework::OpKernel<T> {
     out->mutable_data<T>(ctx.GetPlace());
     auto value = ctx.Attr<float>("value");
 
-    auto out_eigen = framework::EigenVector<T>::Flatten(*out);
-    auto place = ctx.GetEigenDevice<Place>();
-    out_eigen.device(place) = out_eigen.constant(static_cast<T>(value));
+    math::SetConstant<Place, T> setter;
+    setter(ctx.device_context(), out, static_cast<T>(value));
   }
 };
 

paddle/operators/fill_zeros_like_op.cu

@@ -12,7 +12,6 @@
    See the License for the specific language governing permissions and
    limitations under the License. */
 
-#define EIGEN_USE_GPU
 #include "paddle/framework/op_registry.h"
 #include "paddle/operators/fill_zeros_like_op.h"
 

paddle/operators/fill_zeros_like_op.h

@@ -13,8 +13,8 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #pragma once
-#include "paddle/framework/eigen.h"
 #include "paddle/framework/op_registry.h"
+#include "paddle/operators/math/math_function.h"
 
 namespace paddle {
 namespace operators {
@@ -23,10 +23,11 @@ template <typename Place, typename T>
 class FillZerosLikeKernel : public framework::OpKernel<T> {
  public:
   void Compute(const framework::ExecutionContext& context) const override {
-    auto* output = context.Output<framework::Tensor>("Y");
-    output->mutable_data<T>(context.GetPlace());
-    auto t = framework::EigenVector<T>::Flatten(*output);
-    t.device(context.GetEigenDevice<Place>()) = t.constant(static_cast<T>(0));
+    auto* out = context.Output<framework::Tensor>("Y");
+    out->mutable_data<T>(context.GetPlace());
+
+    math::SetConstant<Place, T> setter;
+    setter(context.device_context(), out, static_cast<T>(0));
   }
 };
 

paddle/operators/mul_op.cu

@@ -12,7 +12,6 @@
    See the License for the specific language governing permissions and
    limitations under the License. */
 
-#define EIGEN_USE_GPU
 #include "paddle/operators/mul_op.h"
 
 namespace ops = paddle::operators;

paddle/operators/mul_op.h

@@ -16,16 +16,12 @@
 
 #include "paddle/operators/math/math_function.h"
 
-#include "paddle/framework/eigen.h"
 #include "paddle/framework/op_registry.h"
 
 namespace paddle {
 namespace operators {
 
 using Tensor = framework::Tensor;
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;
 
 template <typename Place, typename T>
 class MulKernel : public framework::OpKernel<T> {

paddle/operators/nccl_op_test.cu

@@ -26,7 +26,6 @@
 #include "paddle/framework/op_registry.h"
 #include "paddle/framework/program_desc.h"
 #include "paddle/framework/var_desc.h"
-#include "paddle/operators/math/math_function.h"
 #include "paddle/operators/nccl/nccl_gpu_common.h"
 #include "paddle/platform/device_context.h"
 #include "paddle/platform/enforce.h"

paddle/operators/sequence_concat_op.cu

@@ -12,8 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#define EIGEN_USE_GPU
-
 #include "paddle/operators/sequence_concat_op.h"
 
 namespace ops = paddle::operators;

paddle/operators/sequence_softmax_op.cu

@@ -12,8 +12,6 @@ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 See the License for the specific language governing permissions and
 limitations under the License. */
 
-#define EIGEN_USE_GPU
-
 #include "paddle/operators/sequence_softmax_op.h"
 
 namespace ops = paddle::operators;

paddle/operators/sequence_softmax_op.h

@@ -14,7 +14,6 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/framework/eigen.h"
 #include "paddle/framework/op_registry.h"
 #include "paddle/operators/math/softmax.h"
 

paddle/operators/softmax_op.cu

@@ -12,7 +12,6 @@
    See the License for the specific language governing permissions and
    limitations under the License. */
 
-#define EIGEN_USE_GPU
 #include "paddle/operators/softmax_op.h"
 
 namespace ops = paddle::operators;

paddle/operators/softmax_op.h

@@ -13,7 +13,6 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #pragma once
-#include "paddle/framework/eigen.h"
 #include "paddle/framework/op_registry.h"
 #include "paddle/operators/math/softmax.h"
 
@@ -21,9 +20,6 @@ namespace paddle {
 namespace operators {
 
 using Tensor = framework::Tensor;
-template <typename T, int MajorType = Eigen::RowMajor,
-          typename IndexType = Eigen::DenseIndex>
-using EigenMatrix = framework::EigenMatrix<T, MajorType, IndexType>;
 
 template <typename Place, typename T>
 class SoftmaxKernel : public framework::OpKernel<T> {