Merge pull request #1156 from smilejames/opencl

update fetch kernel

src/framework/executor.cpp

@@ -922,12 +922,7 @@ void Executor<GPU_CL, Precision::FP32>::InitMemory() {
       if (var_desc->Persistable()) {
         CLImage *cl_image = nullptr;
         if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
-          auto tensor = var->template GetMutable<framework::LoDTensor>();
-          if (var_desc->Name() == "fetch") {
-            const framework::TensorDesc &desc = var_desc->Tensor_desc();
-            framework::DDim ddim = framework::make_ddim(desc.Dims());
-            tensor->mutable_data<float>(ddim);
-          }
+          var->template GetMutable<framework::LoDTensor>();
           continue;
         } else {
           cl_image = var->template GetMutable<framework::CLImage>();
@@ -991,12 +986,7 @@ void Executor<GPU_CL, Precision::FP32>::InitCombineMemory() {
       if (var_desc->Persistable()) {
         CLImage *cl_image = nullptr;
         if (var_desc->Name() == "feed" || var_desc->Name() == "fetch") {
-          auto tensor = var->template GetMutable<framework::LoDTensor>();
-          if (var_desc->Name() == "fetch") {
-            const framework::TensorDesc &desc = var_desc->Tensor_desc();
-            framework::DDim ddim = framework::make_ddim(desc.Dims());
-            tensor->mutable_data<float>(ddim);
-          }
+          var->template GetMutable<framework::LoDTensor>();
           continue;
         } else {
           cl_image = var->template GetMutable<framework::CLImage>();

src/framework/operator.cpp

@@ -95,7 +95,7 @@ void OperatorBase<Dtype>::Run() {
         if (type_ == "fetch") {
           Tensor *tensor = vari->template GetMutable<framework::LoDTensor>();
           if (tensor) {
-            DLOG << type_ << " output- " << key << "=" << tensor->dims();
+            DLOG << type_ << " output- " << key << "=" << *tensor;
           }
         } else {
           CLImage *cl_image = vari->template GetMutable<framework::CLImage>();

src/operators/kernel/cl/fetch_kernel.cpp

@@ -21,6 +21,8 @@ namespace operators {
 template <>
 bool FetchKernel<GPU_CL, float>::Init(FetchParam<GPU_CL> *param) {
   this->cl_helper_.AddKernel("fetch", "fetch_kernel.cl");
+  auto *out = param->Out();
+  out->mutable_data<float>();
   return true;
 }