[Unify Tensors PR #8] Merged Tensor into DenseTensor, test=allcases

paddle/fluid/distributed/fleet.h

@@ -36,7 +36,6 @@ limitations under the License. */
 
 namespace paddle {
 namespace framework {
-class Tensor;
 class Scope;
 class SelectedRows;
 class Variable;

paddle/fluid/distributed/fleet_executor/dist_model.cc

@@ -145,8 +145,8 @@ bool DistModel::LoadParameters() {
   return true;
 }
 
-void DistModel::Run(const std::vector<framework::Tensor> &input_data,
-                    std::vector<framework::Tensor> *output_data) {
+void DistModel::Run(const std::vector<paddle::framework::Tensor> &input_data,
+                    std::vector<paddle::framework::Tensor> *output_data) {
   /* TODO(fleet exe dev): implement this funct */
 }
 

paddle/fluid/distributed/fleet_executor/dist_model.h

@@ -18,14 +18,14 @@
 #include <vector>
 
 #include "paddle/fluid/distributed/fleet_executor/fleet_executor_desc.pb.h"
+#include "paddle/fluid/framework/tensor.h"
 #include "paddle/fluid/platform/macros.h"
 #include "paddle/fluid/platform/place.h"
 
 namespace paddle {
 namespace framework {
 class ProgramDesc;
 class Scope;
-class Tensor;
 }
 
 namespace distributed {
@@ -45,8 +45,8 @@ class DistModel {
  public:
   explicit DistModel(const DistModelConfig& config) : config_(config) {}
   bool Init();
-  void Run(const std::vector<framework::Tensor>& input_data,
-           std::vector<framework::Tensor>* output_data);
+  void Run(const std::vector<paddle::framework::Tensor>& input_data,
+           std::vector<paddle::framework::Tensor>* output_data);
   ~DistModel() = default;
 
  private:

paddle/fluid/distributed/service/brpc_utils.cc

@@ -20,10 +20,13 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 class Variable;
-class Tensor;
 }  // namespace framework
 }  // namespace paddle
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace paddle {
 namespace distributed {
 

paddle/fluid/distributed/test/brpc_service_dense_sgd_test.cc

@@ -31,11 +31,14 @@ class PSClient;
 class PSServer;
 }  // namespace distributed
 namespace framework {
-class Tensor;
 class Variable;
 }  // namespace framework
 }  // namespace paddle
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace framework = paddle::framework;
 namespace platform = paddle::platform;
 namespace operators = paddle::operators;

paddle/fluid/distributed/test/brpc_service_sparse_sgd_test.cc

@@ -32,11 +32,14 @@ class PSClient;
 class PSServer;
 }  // namespace distributed
 namespace framework {
-class Tensor;
 class Variable;
 }  // namespace framework
 }  // namespace paddle
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace framework = paddle::framework;
 namespace platform = paddle::platform;
 namespace operators = paddle::operators;

paddle/fluid/eager/legacy/prepared_operator.h

@@ -29,14 +29,17 @@ DECLARE_bool(use_mkldnn);
 
 namespace paddle {
 namespace framework {
-class Tensor;
 class Variable;
 }  // namespace framework
 namespace platform {
 class DeviceContext;
 }  // namespace platform
 }  // namespace paddle
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace egr {
 namespace legacy {
 

paddle/fluid/framework/copy_same_tensor_test.cc

@@ -68,7 +68,7 @@ static bool CopySameTensorTestMain(const DDim &dims,
     if (sync_copy) {
       TensorCopySync(src_tensor, dst_place, &src_tensor);
     } else {
-      TensorCopy(src_tensor, dst_place, &src_tensor);
+      paddle::framework::TensorCopy(src_tensor, dst_place, &src_tensor);
       platform::DeviceContextPool::Instance().Get(src_place)->Wait();
       platform::DeviceContextPool::Instance().Get(dst_place)->Wait();
     }

paddle/fluid/framework/data_device_transform.cc

@@ -28,8 +28,9 @@ void TransDataDevice(const Tensor &in, const platform::Place &dst_place,
 
   // NOTE(zhiqiu): Special case for CPU->NPU, avoid stream sync.
   if (platform::is_cpu_place(in.place()) && platform::is_npu_place(dst_place)) {
-    TensorCopy(in, dst_place,
-               *platform::DeviceContextPool::Instance().Get(dst_place), out);
+    paddle::framework::TensorCopy(
+        in, dst_place, *platform::DeviceContextPool::Instance().Get(dst_place),
+        out);
     return;
   }
 

paddle/fluid/framework/data_device_transform.h

@@ -21,8 +21,6 @@ limitations under the License. */
 namespace paddle {
 namespace framework {
 
-class Tensor;
-
 void TransDataDevice(const Tensor& in, const platform::Place& dst_place,
                      Tensor* out);
 

paddle/fluid/framework/data_feed.h

@@ -50,12 +50,15 @@ DECLARE_bool(enable_slotrecord_reset_shrink);
 namespace paddle {
 namespace framework {
 class DataFeedDesc;
-class Tensor;
 class Scope;
 class Variable;
 }  // namespace framework
 }  // namespace paddle
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace paddle {
 namespace framework {
 

paddle/fluid/framework/data_layout_transform.h

@@ -25,7 +25,6 @@
 namespace paddle {
 namespace framework {
 class OpKernelType;
-class Tensor;
 }  // namespace framework
 }  // namespace paddle
 

paddle/fluid/framework/data_transform.h

@@ -31,7 +31,6 @@ namespace paddle {
 namespace framework {
 
 class OpKernelType;
-class Tensor;
 class Variable;
 
 void TransformData(const OpKernelType &expected_kernel_type,

paddle/fluid/framework/data_type_transform.h

@@ -25,7 +25,6 @@ namespace paddle {
 namespace framework {
 
 class OpKernelType;
-class Tensor;
 
 using KernelTypePair = std::pair<OpKernelType, OpKernelType>;
 

paddle/fluid/framework/details/async_ssa_graph_executor.cc

@@ -169,7 +169,7 @@ FetchResultType AsyncSSAGraphExecutor::Run(
       std::vector<const LoDTensor *> lodtensor_ptrs;
       lodtensor_ptrs.push_back(&(BOOST_GET(LoDTensor, val.at(fetch_idx))));
       LoDTensor var;
-      var.MergeLoDTensor(lodtensor_ptrs, platform::CPUPlace());
+      MergeLoDTensor(&var, lodtensor_ptrs, platform::CPUPlace());
       ret.emplace_back(var);
     } else {
       auto array = BOOST_GET(LoDTensorArray, val.at(fetch_idx));
@@ -179,7 +179,8 @@ FetchResultType AsyncSSAGraphExecutor::Run(
         std::vector<const LoDTensor *> lodtensor_ptrs;
         lodtensor_ptrs.push_back(&array[i]);
         item_array.emplace_back();
-        item_array.back().MergeLoDTensor(lodtensor_ptrs, platform::CPUPlace());
+        MergeLoDTensor(&(item_array.back()), lodtensor_ptrs,
+                       platform::CPUPlace());
       }
       ret.emplace_back(item_array);
     }

paddle/fluid/framework/details/fetch_async_op_handle.h

@@ -22,14 +22,18 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/device_context.h"
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace paddle {
 namespace framework {
-class Tensor;
 
 namespace ir {
 class Node;
 }  // namespace ir
 }  // namespace framework
+
 namespace platform {
 class DeviceContext;
 }  // namespace platform

paddle/fluid/framework/details/fetch_op_handle.cc

@@ -81,7 +81,7 @@ void FetchOpHandle::WaitAndMergeCPUFetchVars() const {
       }
       auto &val = BOOST_GET(FetchList, *data_);
       LoDTensor var;
-      var.MergeLoDTensor(tensors_ptr, platform::CPUPlace());
+      MergeLoDTensor(&var, tensors_ptr, platform::CPUPlace());
       val.at(offset_) = std::move(var);
     } else {
       auto &array = BOOST_GET_CONST(LoDTensorArray, tensors_[0]);
@@ -99,7 +99,7 @@ void FetchOpHandle::WaitAndMergeCPUFetchVars() const {
           tensors_ptr.push_back(&element[i]);
         }
         tmp_array.emplace_back();
-        tmp_array.back().MergeLoDTensor(tensors_ptr, platform::CPUPlace());
+        MergeLoDTensor(&(tmp_array.back()), tensors_ptr, platform::CPUPlace());
       }
       auto &val = BOOST_GET(FetchList, *data_);
       val.at(offset_) = std::move(tmp_array);

paddle/fluid/framework/details/gather_op_handle.cc

@@ -16,11 +16,10 @@
 
 #include "paddle/fluid/framework/details/container_cast.h"
 #include "paddle/fluid/framework/details/variable_visitor.h"
-namespace paddle {
-namespace framework {
-class Tensor;
-}  // namespace framework
-}  // namespace paddle
+
+namespace pten {
+class DenseTensor;
+}  // namespace pten
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/details/nan_inf_utils_detail.h

@@ -19,11 +19,9 @@
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/platform/place.h"
 
-namespace paddle {
-namespace framework {
-class Tensor;
-}  // namespace framework
-}  // namespace paddle
+namespace pten {
+class DenseTensor;
+}  // namespace pten
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/details/parallel_ssa_graph_executor.cc

@@ -275,7 +275,7 @@ FetchResultType ParallelSSAGraphExecutor::Run(
       }
       if (lodtensor_ptrs.size() != 0) {
         LoDTensor var;
-        var.MergeLoDTensor(lodtensor_ptrs, platform::CPUPlace());
+        MergeLoDTensor(&var, lodtensor_ptrs, platform::CPUPlace());
         ret.emplace_back(var);
       } else {
         LoDTensorArray var_array(lodtensorarray_ptrs[0]->size());
@@ -285,7 +285,7 @@ FetchResultType ParallelSSAGraphExecutor::Run(
           for (size_t j = 0; j < lodtensorarray_ptrs.size(); ++j) {
             ptrs.push_back(&(lodtensorarray_ptrs[j]->at(i)));
           }
-          var.MergeLoDTensor(ptrs, platform::CPUPlace());
+          MergeLoDTensor(&var, ptrs, platform::CPUPlace());
           var_array[i] = std::move(var);
         }
         ret.emplace_back(var_array);

paddle/fluid/framework/details/scale_loss_grad_op_handle.cc

@@ -18,11 +18,9 @@
 
 #include "paddle/fluid/platform/profiler.h"
 
-namespace paddle {
-namespace framework {
-class Tensor;
-}  // namespace framework
-}  // namespace paddle
+namespace pten {
+class DenseTensor;
+}  // namespace pten
 
 namespace paddle {
 namespace framework {

paddle/fluid/framework/details/share_tensor_buffer_functor.cc

@@ -22,7 +22,6 @@
 namespace paddle {
 namespace framework {
 class Scope;
-class Tensor;
 class Variable;
 
 namespace ir {
@@ -31,6 +30,10 @@ class MemOptVarInfo;
 }  // namespace framework
 }  // namespace paddle
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace paddle {
 namespace framework {
 namespace details {

paddle/fluid/framework/details/variable_visitor.cc

@@ -16,9 +16,12 @@
 
 #include "paddle/fluid/framework/selected_rows.h"
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace paddle {
 namespace framework {
-class Tensor;
 class Variable;
 }  // namespace framework
 }  // namespace paddle

paddle/fluid/framework/details/variable_visitor.h

@@ -19,7 +19,6 @@
 
 namespace paddle {
 namespace framework {
-class Tensor;
 class Variable;
 }  // namespace framework
 }  // namespace paddle

paddle/fluid/framework/device_worker.cc

@@ -14,10 +14,13 @@ limitations under the License. */
 
 #include "paddle/fluid/framework/device_worker.h"
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace paddle {
 namespace framework {
 
-class Tensor;
 class Scope;
 
 void DeviceWorker::SetRootScope(Scope* root_scope) { root_scope_ = root_scope; }

paddle/fluid/framework/device_worker.h

@@ -43,7 +43,6 @@ limitations under the License. */
 
 namespace paddle {
 namespace framework {
-class Tensor;
 class ProgramDesc;
 class Scope;
 }  // namespace framework

paddle/fluid/framework/dlpack_tensor.h

@@ -21,8 +21,6 @@
 namespace paddle {
 namespace framework {
 
-class Tensor;
-
 class DLPackTensor {
  public:
   using LaneType = decltype(::DLTensor::dtype.lanes);  // uint16_t

paddle/fluid/framework/downpour_worker.cc

@@ -15,9 +15,12 @@ limitations under the License. */
 #include "paddle/fluid/framework/device_worker.h"
 #include "paddle/fluid/platform/cpu_helper.h"
 
+namespace pten {
+class DenseTensor;
+}  // namespace pten
+
 namespace paddle {
 namespace framework {
-class Tensor;
 class Variable;
 }  // namespace framework
 }  // namespace paddle