Rename CopyFrom to Copy for tensors (#7292)

* Rename Tensor::CopyFrom to Tensor::Copy

* Fix CI

* Fix compile

paddle/framework/data_transform.h

@@ -88,7 +88,7 @@ struct CastDataType {
       trans(*context, in_begin, in_end, out_begin,
             CastDataTypeFunctor<InType, OutType>());
     } else {
-      // TODO(dzhwinter): enhance CopyFrom CPU<->GPU with different data type?
+      // TODO(dzhwinter): enhance Copy CPU<->GPU with different data type?
       PADDLE_THROW("Unsupport CPU <-> GPU!");
     }
   }

paddle/framework/device_data_transform.cc

@@ -37,7 +37,7 @@ Tensor* DeviceTransform(const Tensor& in, const platform::Place& dst_place) {
   Tensor* out = new Tensor();
   auto* dev_ctx = GetDeviceContext(in.place(), dst_place);
   dev_ctx->Wait();
-  CopyFrom(in, dst_place, *dev_ctx, out);
+  Copy(in, dst_place, *dev_ctx, out);
   dev_ctx->Wait();
   return out;
 }

paddle/framework/device_data_transform_test.cu

@@ -157,7 +157,7 @@ TEST(Operator, CPUtoGPU) {
   auto dev_ctx = pool.Get(cuda_place);
 
   paddle::framework::Tensor output_tensor;
-  CopyFrom(output2->Get<LoDTensor>(), paddle::platform::CPUPlace(), *dev_ctx,
+  Copy(output2->Get<LoDTensor>(), paddle::platform::CPUPlace(), *dev_ctx,
        &output_tensor);
 
   dev_ctx->Wait();

paddle/framework/lod_tensor.cc

@@ -232,7 +232,7 @@ std::vector<LoDTensor> LoDTensor::SplitLoDTensor(
     auto dst_ptr = dst.mutable_data(dst_place, src.type());
 
     // TODO(tonyyang-svail):
-    //   change the following to framework::CopyFrom
+    //   change the following to framework::Copy
     auto src_place = src.place();
     auto src_ptr = src.data<void>();
     auto size = src.numel() * SizeOfType(src.type());

paddle/framework/lod_tensor.h

@@ -147,7 +147,7 @@ LoDTensor LodExpand(const LoDTensor& source, const LoD& lod, size_t level,
   for (size_t ins = 0; ins < num_instances; ins++) {
     for (size_t elem = lod_level[ins]; elem < lod_level[ins + 1]; elem++) {
       auto slice = tensor.Slice(elem, elem + 1);
-      CopyFrom(source.Slice(ins, ins + 1), platform::CPUPlace(),
+      Copy(source.Slice(ins, ins + 1), platform::CPUPlace(),
            platform::CPUDeviceContext(), &slice);
     }
   }

paddle/framework/tensor_util.cc

@@ -69,7 +69,7 @@ struct AnyVisitor : public boost::static_visitor<bool> {
     tmp.mutable_data<bool>(cpu);
     auto gpuctx = platform::DeviceContextPool::Instance().Get(gpu);
     gpuctx->Wait();
-    CopyFrom(out, cpu, *gpuctx, &tmp);
+    Copy(out, cpu, *gpuctx, &tmp);
     gpuctx->Wait();
     return GetResult(tmp, cpu);
   }

paddle/framework/tensor_util.h

@@ -29,10 +29,10 @@ namespace framework {
  * @param[in] dst_place  The dst place.
  * @param[in] ctx        The device context contains device resources.
  *
- * @note    CopyFrom supports CPU <-> GPU, GPU <-> GPU.
+ * @note    Copy supports CPU <-> GPU, GPU <-> GPU.
  */
 
-inline void CopyFrom(const Tensor& src, const platform::Place& dst_place,
+inline void Copy(const Tensor& src, const platform::Place& dst_place,
                  const platform::DeviceContext& ctx, Tensor* dst) {
   src.check_memory_size();
 
@@ -88,9 +88,9 @@ inline void CopyFrom(const Tensor& src, const platform::Place& dst_place,
 }
 
 /**
- * @brief CopyFrom support CPU <-> CPU
+ * @brief Copy supports CPU <-> CPU
  */
-inline void CopyFrom(const Tensor& src, const platform::Place& dst_place,
+inline void Copy(const Tensor& src, const platform::Place& dst_place,
                  Tensor* dst) {
   src.check_memory_size();
   dst->Resize(src.dims());
@@ -316,7 +316,7 @@ inline void DeserializeFromStream(std::istream& is, Tensor* tensor,
           DeserializedDataFunctor(&buf, &cpu_tensor, ctx.GetPlace()));
       is.read(static_cast<char*>(buf), cpu_tensor.memory_size());
       auto cpu_place = new platform::CPUPlace();
-      framework::CopyFrom(cpu_tensor, *cpu_place, dev_ctx, tensor);
+      framework::Copy(cpu_tensor, *cpu_place, dev_ctx, tensor);
       delete cpu_place;
 #else
       PADDLE_THROW("Unexpected branch");

paddle/framework/tensor_util_test.cc

@@ -19,7 +19,7 @@
 namespace paddle {
 namespace framework {
 
-TEST(CopyFrom, Tensor) {
+TEST(Copy, Tensor) {
   Tensor src_tensor;
   Tensor dst_tensor;
   platform::CPUDeviceContext cpu_ctx((platform::CPUPlace()));
@@ -32,7 +32,7 @@ TEST(CopyFrom, Tensor) {
   src_tensor.set_layout(DataLayout::kAnyLayout);
 
   auto cpu_place = new platform::CPUPlace();
-  CopyFrom(src_tensor, *cpu_place, &dst_tensor);
+  Copy(src_tensor, *cpu_place, &dst_tensor);
 
   const int* dst_ptr = dst_tensor.data<int>();
   ASSERT_NE(src_ptr, dst_ptr);
@@ -43,7 +43,7 @@ TEST(CopyFrom, Tensor) {
   EXPECT_TRUE(dst_tensor.layout() == src_tensor.layout());
 
   Tensor slice_tensor = src_tensor.Slice(1, 2);
-  CopyFrom(slice_tensor, *cpu_place, &dst_tensor);
+  Copy(slice_tensor, *cpu_place, &dst_tensor);
   const int* slice_ptr = slice_tensor.data<int>();
   dst_ptr = dst_tensor.data<int>();
   ASSERT_NE(dst_ptr, slice_ptr);
@@ -67,11 +67,11 @@ TEST(CopyFrom, Tensor) {
     // CPU Tensor to GPU Tensor
     auto gpu_place = new platform::CUDAPlace(0);
     platform::CUDADeviceContext gpu_ctx(*gpu_place);
-    CopyFrom(src_tensor, *gpu_place, gpu_ctx, &gpu_tensor);
+    Copy(src_tensor, *gpu_place, gpu_ctx, &gpu_tensor);
 
     // GPU Tensor to CPU Tensor
     auto cpu_place = new platform::CPUPlace();
-    CopyFrom(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
+    Copy(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
 
     // Sync before Compare Tensors
     gpu_ctx.Wait();
@@ -84,10 +84,10 @@ TEST(CopyFrom, Tensor) {
     Tensor slice_tensor = src_tensor.Slice(1, 2);
 
     // CPU Slice Tensor to GPU Tensor
-    CopyFrom(slice_tensor, *gpu_place, gpu_ctx, &gpu_tensor);
+    Copy(slice_tensor, *gpu_place, gpu_ctx, &gpu_tensor);
 
     // GPU Tensor to CPU Tensor
-    CopyFrom(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
+    Copy(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
 
     // Sync before Compare Slice Tensors
     gpu_ctx.Wait();
@@ -155,7 +155,7 @@ TEST(CopyFromVector, Tensor) {
     CUDADeviceContext gpu_ctx(*gpu_place);
     CopyFromVector<int>(src_vec, gpu_ctx, &gpu_tensor);
     // Copy from GPU to CPU tensor for comparison
-    CopyFrom(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
+    Copy(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
 
     // Sync before Compare Tensors
     gpu_ctx.Wait();
@@ -175,7 +175,7 @@ TEST(CopyFromVector, Tensor) {
     CopyFromVector<int>(src_vec, cpu_ctx, &cpu_tensor);
     gpu_tensor.Resize(make_ddim({2, 2}));
     CopyFromVector<int>(src_vec, gpu_ctx, &gpu_tensor);
-    CopyFrom(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
+    Copy(gpu_tensor, *cpu_place, gpu_ctx, &dst_tensor);
 
     // Sync before Compare Tensors
     gpu_ctx.Wait();
@@ -287,7 +287,7 @@ TEST(Tensor, SerializeAndDeserialize) {
     auto gpu_place = new platform::CUDAPlace();
     platform::CUDADeviceContext gpu_ctx(*gpu_place);
 
-    CopyFrom(src_tensor, *gpu_place, gpu_ctx, &gpu_tensor);
+    Copy(src_tensor, *gpu_place, gpu_ctx, &gpu_tensor);
 
     std::ostringstream oss;
     SerializeToStream(oss, gpu_tensor, gpu_ctx);

paddle/operators/array_operator.h

@@ -42,7 +42,7 @@ class ArrayOp : public framework::OperatorBase {
     if (platform::is_gpu_place(i_tensor.place())) {
       // FIXME: Avoid copy from GPU to CPU
       framework::Tensor t;
-      framework::CopyFrom(i_tensor, platform::CPUPlace(), dev_ctx, &t);
+      framework::Copy(i_tensor, platform::CPUPlace(), dev_ctx, &t);
       dev_ctx.Wait();
       offset = static_cast<size_t>(*t.data<int64_t>());
     } else {

paddle/operators/array_to_lod_tensor_op.cc

@@ -110,7 +110,7 @@ class ArrayToLoDTensorOp : public framework::OperatorBase {
             platform::DeviceContextPool::Instance();
         auto &dev_ctx = *pool.Get(place);
 
-        framework::CopyFrom(x[x_idx].Slice(start_offset, end_offset), place,
+        framework::Copy(x[x_idx].Slice(start_offset, end_offset), place,
                         dev_ctx, &slice);
         out_offset += len;
       }

paddle/operators/assign_op.cc

@@ -45,7 +45,7 @@ class AssignFunctor {
     out_rows.set_height(rows.height());
     auto &t = rows.value();
     auto *m = out_rows.mutable_value();
-    framework::CopyFrom(t, t.place(), dev_ctx_, m);
+    framework::Copy(t, t.place(), dev_ctx_, m);
   }
 
   template <typename T>
@@ -57,7 +57,7 @@ class AssignFunctor {
   void copy_tensor(const framework::LoDTensor &lod_tensor,
                    framework::LoDTensor *out) const {
     auto &out_tensor = *out;
-    CopyFrom(lod_tensor, lod_tensor.place(), dev_ctx_, &out_tensor);
+    Copy(lod_tensor, lod_tensor.place(), dev_ctx_, &out_tensor);
     out_tensor.set_lod(lod_tensor.lod());
   }
 

paddle/operators/detection_output_op.h

@@ -98,15 +98,15 @@ class DetectionOutputKernel : public framework::OpKernel<T> {
     T* conf_data = conf_tensor.data<T>();
     if (platform::is_gpu_place(context.GetPlace())) {
       loc_cpu.mutable_data<T>(loc_tensor.dims(), platform::CPUPlace());
-      framework::CopyFrom(loc_tensor, platform::CPUPlace(),
+      framework::Copy(loc_tensor, platform::CPUPlace(),
                       context.device_context(), &loc_cpu);
       loc_data = loc_cpu.data<T>();
       conf_cpu.mutable_data<T>(conf_tensor.dims(), platform::CPUPlace());
-      framework::CopyFrom(conf_tensor, platform::CPUPlace(),
+      framework::Copy(conf_tensor, platform::CPUPlace(),
                       context.device_context(), &conf_cpu);
       conf_data = conf_cpu.data<T>();
       priorbox_cpu.mutable_data<T>(in_priorbox->dims(), platform::CPUPlace());
-      framework::CopyFrom(*in_priorbox, platform::CPUPlace(),
+      framework::Copy(*in_priorbox, platform::CPUPlace(),
                       context.device_context(), &priorbox_cpu);
       priorbox_data = priorbox_cpu.data<T>();
     }
@@ -158,8 +158,8 @@ class DetectionOutputKernel : public framework::OpKernel<T> {
                                 batch_size, all_indices, all_decoded_bboxes,
                                 out_data);
     if (platform::is_gpu_place(context.GetPlace())) {
-      framework::CopyFrom(out_cpu, platform::CUDAPlace(),
-                          context.device_context(), out);
+      framework::Copy(out_cpu, platform::CUDAPlace(), context.device_context(),
+                      out);
     }
   }
 };

paddle/operators/expand_op.h

@@ -126,8 +126,7 @@ class ExpandGradKernel : public framework::OpKernel<T> {
       auto* in0 = context.Input<Tensor>(framework::GradVarName("Out"));
       auto* out0 = context.Output<Tensor>(framework::GradVarName("X"));
       out0->mutable_data<T>(context.GetPlace());
-      framework::CopyFrom(*in0, context.GetPlace(), context.device_context(),
-                          out0);
+      framework::Copy(*in0, context.GetPlace(), context.device_context(), out0);
     } else {
       switch (dims) {
         REP_EXPAND_GRAD_TEMPLATE(72)

paddle/operators/feed_op.cc

@@ -52,7 +52,7 @@ class FeedOp : public framework::OperatorBase {
     platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
     auto &dev_ctx = *pool.Get(place);
 
-    framework::CopyFrom(feed_item, place, dev_ctx, out_item);
+    framework::Copy(feed_item, place, dev_ctx, out_item);
     out_item->set_lod(feed_item.lod());
   }
 };

paddle/operators/fetch_op.cc

@@ -55,7 +55,7 @@ class FetchOp : public framework::OperatorBase {
     platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
     auto &dev_ctx = *pool.Get(src_item.place());
 
-    CopyFrom(src_item, platform::CPUPlace(), dev_ctx, &dst_item);
+    Copy(src_item, platform::CPUPlace(), dev_ctx, &dst_item);
     dev_ctx.Wait();
     dst_item.set_lod(src_item.lod());
 

paddle/operators/fill_op.cc

@@ -72,7 +72,7 @@ class FillOp : public framework::OperatorBase {
       platform::DeviceContextPool &pool =
           platform::DeviceContextPool::Instance();
       auto &dev_ctx = *pool.Get(place);
-      framework::CopyFrom(tensor, place, dev_ctx, &out);
+      framework::Copy(tensor, place, dev_ctx, &out);
     }
   }
 };

paddle/operators/linear_chain_crf_op.h

@@ -196,7 +196,7 @@ class LinearChainCRFOpKernel : public framework::OpKernel<T> {
     auto copyLoDTensor = [](const platform::DeviceContext& ctx,
                             const LoDTensor& src, LoDTensor* dst) {
       dst->mutable_data<T>(src.dims(), platform::CPUPlace());
-      framework::CopyFrom(src, platform::CPUPlace(), ctx, dst);
+      framework::Copy(src, platform::CPUPlace(), ctx, dst);
     };
 
     copyLoDTensor(ctx, emission_weights_src, emission_weights_dst);
@@ -204,7 +204,7 @@ class LinearChainCRFOpKernel : public framework::OpKernel<T> {
 
     transition_weights_dst->mutable_data<T>(transition_weights_src.dims(),
                                             platform::CPUPlace());
-    framework::CopyFrom(transition_weights_src, platform::CPUPlace(), ctx,
+    framework::Copy(transition_weights_src, platform::CPUPlace(), ctx,
                     transition_weights_dst);
   }
 
@@ -220,7 +220,7 @@ class LinearChainCRFOpKernel : public framework::OpKernel<T> {
     auto copyTensor = [](const platform::DeviceContext& ctx, const Tensor& src,
                          Tensor* dst) {
       dst->mutable_data<T>(platform::CUDAPlace());
-      framework::CopyFrom(src, platform::CUDAPlace(), ctx, dst);
+      framework::Copy(src, platform::CUDAPlace(), ctx, dst);
     };
     copyTensor(ctx, emission_exps_src, emission_exps_dst);
     copyTensor(ctx, transition_exps_src, transition_exps_dst);
@@ -410,12 +410,12 @@ class LinearChainCRFGradOpKernel : public framework::OpKernel<T> {
     // Copy the inputs from GPU memory to CPU memory when this operators runs on
     // GPU device.
     label_dst->mutable_data<T>(label_src.dims(), platform::CPUPlace());
-    framework::CopyFrom(label_src, platform::CPUPlace(), ctx, label_dst);
+    framework::Copy(label_src, platform::CPUPlace(), ctx, label_dst);
 
     auto copyTensor = [](const platform::DeviceContext& ctx, const Tensor& src,
                          Tensor* dst) {
       dst->mutable_data<T>(src.dims(), platform::CPUPlace());
-      framework::CopyFrom(src, platform::CPUPlace(), ctx, dst);
+      framework::Copy(src, platform::CPUPlace(), ctx, dst);
     };
     copyTensor(ctx, emission_exps_src, emission_exps_dst);
     copyTensor(ctx, transition_exps_src, transition_exps_dst);
@@ -434,7 +434,7 @@ class LinearChainCRFGradOpKernel : public framework::OpKernel<T> {
                          Tensor* dst) {
       if (src && dst) {
         dst->mutable_data<T>(platform::CUDAPlace());
-        framework::CopyFrom(*src, platform::CUDAPlace(), ctx, dst);
+        framework::Copy(*src, platform::CUDAPlace(), ctx, dst);
       }
     };
     copyTensor(ctx, emission_grad_src, emission_grad_dst);

paddle/operators/load_op.cc

@@ -53,7 +53,7 @@ class LoadOp : public framework::OperatorBase {
       out_var->Clear();
       tensor = out_var->GetMutable<framework::LoDTensor>();
       tensor->set_lod(cpu_tensor.lod());
-      CopyFrom(cpu_tensor, place, dev_ctx, tensor);
+      Copy(cpu_tensor, place, dev_ctx, tensor);
     }
   }
 };

paddle/operators/lod_reset_op.h

@@ -33,7 +33,7 @@ class LoDResetKernel : public framework::OpKernel<T> {
       auto* lod = lod_t->data<int>();
       if (platform::is_gpu_place(ctx.GetPlace())) {
         framework::Tensor lod_cpu;
-        framework::CopyFrom(*lod_t, platform::CPUPlace(), ctx.device_context(),
+        framework::Copy(*lod_t, platform::CPUPlace(), ctx.device_context(),
                         &lod_cpu);
         lod = lod_cpu.data<int>();
       }

paddle/operators/lod_tensor_to_array_op.cc

@@ -92,7 +92,7 @@ class LoDTensorToArrayOp : public framework::OperatorBase {
             platform::DeviceContextPool::Instance();
         auto &dev_ctx = *pool.Get(place);
 
-        framework::CopyFrom(x.Slice(static_cast<int>(each_range.begin),
+        framework::Copy(x.Slice(static_cast<int>(each_range.begin),
                                 static_cast<int>(each_range.end)),
                         x.place(), dev_ctx, &slice);
         offset += len;

paddle/operators/math/context_project.h

@@ -149,7 +149,7 @@ class ContextProjectFunctor {
             Tensor out_t_sub = out_t.Slice(k * context_length,
                                            k * context_length + padding_size);
             Tensor w_sub = padding_data.Slice(k, k + padding_size);
-            framework::CopyFrom(w_sub, context.GetPlace(), context, &out_t_sub);
+            framework::Copy(w_sub, context.GetPlace(), context, &out_t_sub);
           }
         }
         if (down_pad > 0) {  // add down pad
@@ -179,7 +179,7 @@ class ContextProjectFunctor {
                 (down_pad_begin_row + t) * context_length);
             Tensor w_sub = padding_data.Slice(
                 up_pad + padding_idx, up_pad + padding_idx + padding_size);
-            framework::CopyFrom(w_sub, context.GetPlace(), context, &out_t_sub);
+            framework::Copy(w_sub, context.GetPlace(), context, &out_t_sub);
           }
         }
         out_t.Resize({sequence_height, context_length * sequence_width});

paddle/operators/math/im2col_test.cc

@@ -63,7 +63,7 @@ void testIm2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     input = input_tmp;
   } else {
-    CopyFrom(input_tmp, *place, *context, &input);
+    Copy(input_tmp, *place, *context, &input);
   }
   output_cfo.mutable_data<float>(
       {1, filter_size, filter_size, output_height, output_width}, *place);
@@ -88,7 +88,7 @@ void testIm2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     out_cfo_ptr = output_cfo.data<float>();
   } else {
-    CopyFrom(output_cfo, paddle::platform::CPUPlace(), *context, &output_tmp);
+    Copy(output_cfo, paddle::platform::CPUPlace(), *context, &output_tmp);
     out_cfo_ptr = output_tmp.data<float>();
   }
   for (int i = 0; i < 6; ++i) {
@@ -99,7 +99,7 @@ void testIm2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     out_ocf_ptr = output_ocf.data<float>();
   } else {
-    CopyFrom(output_ocf, paddle::platform::CPUPlace(), *context, &output_tmp);
+    Copy(output_ocf, paddle::platform::CPUPlace(), *context, &output_tmp);
     out_ocf_ptr = output_tmp.data<float>();
   }
   for (int i = 0; i < 6; ++i) {
@@ -119,7 +119,7 @@ void testIm2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     input = input_tmp;
   } else {
-    CopyFrom(input_tmp, *place, *context, &input);
+    Copy(input_tmp, *place, *context, &input);
   }
 
   col2im(*context, output_cfo, dilation, stride, padding, &input);
@@ -128,7 +128,7 @@ void testIm2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     in_ptr = input.data<float>();
   } else {
-    CopyFrom(input, paddle::platform::CPUPlace(), *context, &input_tmp);
+    Copy(input, paddle::platform::CPUPlace(), *context, &input_tmp);
     in_ptr = input_tmp.data<float>();
   }
   for (int i = 0; i < 6; ++i) {
@@ -140,7 +140,7 @@ void testIm2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     input = input_tmp;
   } else {
-    CopyFrom(input_tmp, *place, *context, &input);
+    Copy(input_tmp, *place, *context, &input);
   }
 
   col2im_ocf(*context, output_ocf, dilation, stride, padding, &input);
@@ -148,7 +148,7 @@ void testIm2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     in_ptr = input.data<float>();
   } else {
-    CopyFrom(input, paddle::platform::CPUPlace(), *context, &input_tmp);
+    Copy(input, paddle::platform::CPUPlace(), *context, &input_tmp);
     in_ptr = input_tmp.data<float>();
   }
   for (int i = 0; i < 6; ++i) {

paddle/operators/math/math_function_test.cu

@@ -16,15 +16,15 @@ TEST(math_function, notrans_mul_trans) {
   auto* gpu_place = new paddle::platform::CUDAPlace(0);
   paddle::platform::CUDADeviceContext context(*gpu_place);
 
-  paddle::framework::CopyFrom(input1, *gpu_place, context, &input1_gpu);
-  paddle::framework::CopyFrom(input1, *gpu_place, context, &input2_gpu);
+  paddle::framework::Copy(input1, *gpu_place, context, &input1_gpu);
+  paddle::framework::Copy(input1, *gpu_place, context, &input2_gpu);
 
   out_gpu.mutable_data<float>({2, 2}, *gpu_place);
 
   paddle::operators::math::matmul<paddle::platform::CUDADeviceContext, float>(
       context, input1_gpu, false, input2_gpu, true, 1, &out_gpu, 0);
 
-  paddle::framework::CopyFrom(out_gpu, *cpu_place, context, &out);
+  paddle::framework::Copy(out_gpu, *cpu_place, context, &out);
 
   float* out_ptr = out.data<float>();
   context.Wait();
@@ -50,15 +50,15 @@ TEST(math_function, trans_mul_notrans) {
   auto* gpu_place = new paddle::platform::CUDAPlace(0);
   paddle::platform::CUDADeviceContext context(*gpu_place);
 
-  paddle::framework::CopyFrom(input1, *gpu_place, context, &input1_gpu);
-  paddle::framework::CopyFrom(input1, *gpu_place, context, &input2_gpu);
+  paddle::framework::Copy(input1, *gpu_place, context, &input1_gpu);
+  paddle::framework::Copy(input1, *gpu_place, context, &input2_gpu);
 
   out_gpu.mutable_data<float>({3, 3}, *gpu_place);
 
   paddle::operators::math::matmul<paddle::platform::CUDADeviceContext, float>(
       context, input1_gpu, true, input2_gpu, false, 1, &out_gpu, 0);
 
-  paddle::framework::CopyFrom(out_gpu, *cpu_place, context, &out);
+  paddle::framework::Copy(out_gpu, *cpu_place, context, &out);
 
   float* out_ptr = out.data<float>();
   context.Wait();
@@ -99,9 +99,9 @@ TEST(math_function, gemm_notrans_cublas) {
   auto* gpu_place = new paddle::platform::CUDAPlace(0);
   paddle::platform::CUDADeviceContext context(*gpu_place);
 
-  paddle::framework::CopyFrom(input1, *gpu_place, context, &input1_gpu);
-  paddle::framework::CopyFrom(input2, *gpu_place, context, &input2_gpu);
-  paddle::framework::CopyFrom(input3, *gpu_place, context, &input3_gpu);
+  paddle::framework::Copy(input1, *gpu_place, context, &input1_gpu);
+  paddle::framework::Copy(input2, *gpu_place, context, &input2_gpu);
+  paddle::framework::Copy(input3, *gpu_place, context, &input3_gpu);
   float* a = input1_gpu.data<float>();
   float* b = input2_gpu.data<float>();
   float* c = input3_gpu.mutable_data<float>(*gpu_place);
@@ -109,7 +109,7 @@ TEST(math_function, gemm_notrans_cublas) {
   paddle::operators::math::gemm<paddle::platform::CUDADeviceContext, float>(
       context, false, false, m, n, k, 1, a, 3, b + 1, 4, 1, c + 1, 4);
 
-  paddle::framework::CopyFrom(input3_gpu, *cpu_place, context, &input3);
+  paddle::framework::Copy(input3_gpu, *cpu_place, context, &input3);
 
   // numpy code:
   // a = np.arange(6).reshape(2, 3)
@@ -154,9 +154,9 @@ TEST(math_function, gemm_trans_cublas) {
   auto* gpu_place = new paddle::platform::CUDAPlace(0);
   paddle::platform::CUDADeviceContext context(*gpu_place);
 
-  paddle::framework::CopyFrom(input1, *gpu_place, context, &input1_gpu);
-  paddle::framework::CopyFrom(input2, *gpu_place, context, &input2_gpu);
-  paddle::framework::CopyFrom(input3, *gpu_place, context, &input3_gpu);
+  paddle::framework::Copy(input1, *gpu_place, context, &input1_gpu);
+  paddle::framework::Copy(input2, *gpu_place, context, &input2_gpu);
+  paddle::framework::Copy(input3, *gpu_place, context, &input3_gpu);
   float* a = input1_gpu.data<float>();
   float* b = input2_gpu.data<float>();
   float* c = input3_gpu.mutable_data<float>(*gpu_place);
@@ -164,7 +164,7 @@ TEST(math_function, gemm_trans_cublas) {
   paddle::operators::math::gemm<paddle::platform::CUDADeviceContext, float>(
       context, false, true, m, n, k, 1, a, 3, b + 3, 3, 1, c + 1, 4);
 
-  paddle::framework::CopyFrom(input3_gpu, *cpu_place, context, &input3);
+  paddle::framework::Copy(input3_gpu, *cpu_place, context, &input3);
   context.Wait();
 
   EXPECT_EQ(input3_ptr[0], 0);
@@ -205,14 +205,14 @@ void GemvTest(int m, int n, bool trans) {
   }
 
   paddle::platform::CUDADeviceContext context(*gpu_place);
-  paddle::framework::CopyFrom(mat_a, *gpu_place, context, &g_mat_a);
-  paddle::framework::CopyFrom(vec_b, *gpu_place, context, &g_vec_b);
+  paddle::framework::Copy(mat_a, *gpu_place, context, &g_mat_a);
+  paddle::framework::Copy(vec_b, *gpu_place, context, &g_vec_b);
 
   paddle::operators::math::gemv<paddle::platform::CUDADeviceContext, T>(
       context, trans, static_cast<int>(m), static_cast<int>(n), 1., g_data_a,
       g_data_b, 0., g_data_c);
 
-  paddle::framework::CopyFrom(g_vec_c, paddle::platform::CPUPlace(), context,
+  paddle::framework::Copy(g_vec_c, paddle::platform::CPUPlace(), context,
                           &vec_c);
 
   if (!trans) {

paddle/operators/math/selected_rows_functor_test.cu

@@ -67,7 +67,7 @@ TEST(selected_rows_functor, gpu_add) {
   EXPECT_EQ(out_rows[6], 9);
 
   Tensor out_cpu;
-  CopyFrom(*out_value, cpu_place, ctx, &out_cpu);
+  Copy(*out_value, cpu_place, ctx, &out_cpu);
   ctx.Wait();
 
   auto* out_cpu_data = out_cpu.data<float>();
@@ -94,7 +94,7 @@ TEST(selected_rows_functor, gpu_add) {
   add_tensor_functor(ctx, *output, *tensor1, tensor2.get());
 
   Tensor tensor2_cpu;
-  CopyFrom(*tensor2, cpu_place, ctx, &tensor2_cpu);
+  Copy(*tensor2, cpu_place, ctx, &tensor2_cpu);
   ctx.Wait();
 
   auto* tensor2_cpu_data = tensor2_cpu.data<float>();
@@ -167,7 +167,7 @@ TEST(selected_rows_functor, gpu_add_to) {
   EXPECT_EQ(out_rows[6], 9);
 
   Tensor out_cpu;
-  CopyFrom(*out_value, cpu_place, ctx, &out_cpu);
+  Copy(*out_value, cpu_place, ctx, &out_cpu);
   ctx.Wait();
 
   auto* out_cpu_data = out_cpu.data<float>();
@@ -191,7 +191,7 @@ TEST(selected_rows_functor, gpu_add_to) {
   add_to_tensor_functor(ctx, *output, tensor1.get());
 
   Tensor tensor1_cpu;
-  CopyFrom(*tensor1, cpu_place, ctx, &tensor1_cpu);
+  Copy(*tensor1, cpu_place, ctx, &tensor1_cpu);
   ctx.Wait();
 
   auto* tensor1_cpu_data = tensor1_cpu.data<float>();

paddle/operators/math/vol2col_test.cc

@@ -71,7 +71,7 @@ void testVol2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     input = input_tmp;
   } else {
-    CopyFrom(input_tmp, *place, *context, &input);
+    Copy(input_tmp, *place, *context, &input);
   }
   output.mutable_data<float>({1, filter_size, filter_size, filter_size,
                               output_depth, output_height, output_width},
@@ -85,7 +85,7 @@ void testVol2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     out_cfo_ptr = output.data<float>();
   } else {
-    CopyFrom(output, paddle::platform::CPUPlace(), *context, &output_tmp);
+    Copy(output, paddle::platform::CPUPlace(), *context, &output_tmp);
     out_cfo_ptr = output_tmp.data<float>();
   }
 
@@ -99,7 +99,7 @@ void testVol2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     input = input_tmp;
   } else {
-    CopyFrom(input_tmp, *place, *context, &input);
+    Copy(input_tmp, *place, *context, &input);
   }
 
   paddle::operators::math::Col2VolFunctor<DeviceContext, float> col2vol;
@@ -109,7 +109,7 @@ void testVol2col() {
   if (paddle::platform::is_cpu_place(*place)) {
     in_ptr = input.data<float>();
   } else {
-    CopyFrom(input, paddle::platform::CPUPlace(), *context, &input_tmp);
+    Copy(input, paddle::platform::CPUPlace(), *context, &input_tmp);
     in_ptr = input_tmp.data<float>();
   }
 

paddle/operators/merge_lod_tensor_op.cc

@@ -49,7 +49,7 @@ class MergeLoDTensorOp : public framework::OperatorBase {
       cpu_mask->ShareDataWith(mask);
     } else if (platform::is_gpu_place(mask.place())) {
 #ifdef PADDLE_WITH_CUDA
-      framework::CopyFrom(mask, platform::CPUPlace(), dev_ctx, cpu_mask.get());
+      framework::Copy(mask, platform::CPUPlace(), dev_ctx, cpu_mask.get());
 #else
       PADDLE_THROW("Not supported GPU, Please compile WITH_GPU option");
 #endif
@@ -104,8 +104,8 @@ class MergeLoDTensorOp : public framework::OperatorBase {
         continue;
       }
       auto slice = out->Slice(out_offset, out_offset + len);
-      framework::CopyFrom(input->Slice(start_offset, end_offset), place,
-                          dev_ctx, &slice);
+      framework::Copy(input->Slice(start_offset, end_offset), place, dev_ctx,
+                      &slice);
       out_offset += len;
       (*in_idx) += 1;
     }

paddle/operators/multiplex_op.cu

@@ -33,7 +33,7 @@ class MultiplexGPUKernel : public framework::OpKernel<T> {
     auto cols = ins[0]->numel() / rows;
     // copy index to cpu
     Tensor index_t_cpu;
-    CopyFrom(*ids, platform::CPUPlace(), ctx.device_context(), &index_t_cpu);
+    Copy(*ids, platform::CPUPlace(), ctx.device_context(), &index_t_cpu);
     auto* index = index_t_cpu.data<int32_t>();
     auto stream = ctx.cuda_device_context().stream();
     platform::CUDAPlace place = boost::get<platform::CUDAPlace>(ctx.GetPlace());
@@ -69,7 +69,7 @@ class MultiplexGradGPUKernel : public framework::OpKernel<T> {
     auto cols = ins[0]->numel() / rows;
     // copy index to cpu
     Tensor index_t_cpu;
-    CopyFrom(*ids, platform::CPUPlace(), ctx.device_context(), &index_t_cpu);
+    Copy(*ids, platform::CPUPlace(), ctx.device_context(), &index_t_cpu);
     auto* index = index_t_cpu.data<int32_t>();
 
     auto stream = ctx.cuda_device_context().stream();

paddle/operators/parallel_do_op.cc

@@ -211,7 +211,7 @@ class ParallelDoGradOp : public OperatorBase {
         auto &tt = sub_scopes[place_idx]->FindVar(s)->Get<LoDTensor>();
         VLOG(3) << place_idx;
         VLOG(3) << tt;
-        framework::CopyFrom(tt, places[0], t_buf);
+        framework::Copy(tt, places[0], t_buf);
 
         auto sum_op = framework::OpRegistry::CreateOp(
             "sum", {{"X", {s, s_buf}}}, {{"Out", {s}}},
@@ -220,7 +220,7 @@ class ParallelDoGradOp : public OperatorBase {
       }
 
       VLOG(3) << t;
-      framework::CopyFrom(t, place, scope.FindVar(s)->GetMutable<LoDTensor>());
+      framework::Copy(t, place, scope.FindVar(s)->GetMutable<LoDTensor>());
     }
   }
 };

paddle/operators/recurrent_op.cc

@@ -290,7 +290,7 @@ class RecurrentOp : public RecurrentBase {
             auto dst_out = dst_tensor->Slice(seq_offset, seq_offset + 1);
             // Explicit copy output since the local RNN scope can be destroyed
             // early.
-            framework::CopyFrom(src_tensor, place, dev_ctx, &dst_out);
+            framework::Copy(src_tensor, place, dev_ctx, &dst_out);
           });
 
       scopes.Next();
@@ -376,7 +376,7 @@ class RecurrentGradOp : public RecurrentBase {
           auto *cur_grad_var = cur_scope.Var(cur_grad);
           auto cur_grad_tensor =
               cur_grad_var->GetMutable<framework::LoDTensor>();
-          framework::CopyFrom(ex_tensor, place, dev_ctx, cur_grad_tensor);
+          framework::Copy(ex_tensor, place, dev_ctx, cur_grad_tensor);
         }
       }
 
@@ -450,7 +450,7 @@ class RecurrentGradOp : public RecurrentBase {
             }
 
             auto dst = outside->Slice(seq_offset, seq_offset + 1);
-            framework::CopyFrom(inside, place, dev_ctx, &dst);
+            framework::Copy(inside, place, dev_ctx, &dst);
           });
       VLOG(5) << "Link outside gradient finished ";
 
@@ -463,7 +463,7 @@ class RecurrentGradOp : public RecurrentBase {
                 framework::LoDTensor *outside) {
               outside->Resize(inside.dims());
               outside->mutable_data(place, inside.type());
-              framework::CopyFrom(inside, place, dev_ctx, outside);
+              framework::Copy(inside, place, dev_ctx, outside);
             });
         VLOG(5) << "Link initialize state gradient finished ";
       }

paddle/operators/reorder_lod_tensor_by_rank_op.cc

@@ -146,7 +146,7 @@ class ReorderLoDTensorByRankTableBase : public framework::OperatorBase {
 
     platform::DeviceContextPool &pool = platform::DeviceContextPool::Instance();
     auto &dev_ctx = *pool.Get(place);
-    framework::CopyFrom(x_sliced, out_sliced.place(), dev_ctx, &out_sliced);
+    framework::Copy(x_sliced, out_sliced.place(), dev_ctx, &out_sliced);
     out_offset += len;
     return out_offset;
   }

paddle/operators/reshape_op.h

@@ -28,7 +28,7 @@ class ReshapeKernel : public framework::OpKernel<T> {
     auto* in = ctx.Input<framework::Tensor>("X");
     auto out_dims = out->dims();
     out->mutable_data<T>(ctx.GetPlace());
-    framework::CopyFrom(*in, ctx.GetPlace(), ctx.device_context(), out);
+    framework::Copy(*in, ctx.GetPlace(), ctx.device_context(), out);
     out->Resize(out_dims);
   }
 };
@@ -42,7 +42,7 @@ class ReshapeGradKernel : public framework::OpKernel<T> {
     d_x->mutable_data<T>(ctx.GetPlace());
 
     auto in_dims = d_x->dims();
-    framework::CopyFrom(*d_out, ctx.GetPlace(), ctx.device_context(), d_x);
+    framework::Copy(*d_out, ctx.GetPlace(), ctx.device_context(), d_x);
     d_x->Resize(in_dims);
   }
 };

paddle/operators/sequence_slice_op.h

@@ -66,12 +66,12 @@ class SequenceSliceOpKernel : public framework::OpKernel<T> {
 
     if (platform::is_gpu_place(ctx.GetPlace())) {
       offset_cpu.mutable_data<T>(offset->dims(), platform::CPUPlace());
-      framework::CopyFrom(*offset, platform::CPUPlace(), ctx.device_context(),
+      framework::Copy(*offset, platform::CPUPlace(), ctx.device_context(),
                       &offset_cpu);
       offset_data = offset_cpu.data<int64_t>();
 
       length_cpu.mutable_data<T>(length->dims(), platform::CPUPlace());
-      framework::CopyFrom(*length, platform::CPUPlace(), ctx.device_context(),
+      framework::Copy(*length, platform::CPUPlace(), ctx.device_context(),
                       &length_cpu);
       length_data = length_cpu.data<int64_t>();
     }
@@ -127,12 +127,12 @@ class SequenceSliceGradOpKernel : public framework::OpKernel<T> {
 
     if (platform::is_gpu_place(ctx.GetPlace())) {
       offset_cpu.mutable_data<T>(offset->dims(), platform::CPUPlace());
-      framework::CopyFrom(*offset, platform::CPUPlace(), ctx.device_context(),
+      framework::Copy(*offset, platform::CPUPlace(), ctx.device_context(),
                       &offset_cpu);
       offset_data = offset_cpu.data<int64_t>();
 
       length_cpu.mutable_data<T>(length->dims(), platform::CPUPlace());
-      framework::CopyFrom(*length, platform::CPUPlace(), ctx.device_context(),
+      framework::Copy(*length, platform::CPUPlace(), ctx.device_context(),
                       &length_cpu);
       length_data = length_cpu.data<int64_t>();
     }

paddle/operators/shrink_rnn_memory_op.cc

@@ -115,7 +115,7 @@ class ShrinkRNNMemoryGradOp : public ArrayOp {
       auto &dout_tensor = dout_var->Get<framework::LoDTensor>();
       auto height = dout_tensor.dims()[0];
       auto slice = dx_tensor.Slice(0, static_cast<int>(height));
-      framework::CopyFrom(dout_tensor, dout_tensor.place(), dev_ctx, &slice);
+      framework::Copy(dout_tensor, dout_tensor.place(), dev_ctx, &slice);
       if (dx_tensor.dims()[0] > height) {
         auto rest_tensor = dx_tensor.Slice(
             static_cast<int>(height), static_cast<int>(dx_tensor.dims()[0]));

paddle/operators/split_lod_tensor_op.cc

@@ -53,7 +53,7 @@ class SplitLoDTensorOp : public framework::OperatorBase {
       cpu_mask->ShareDataWith(mask);
     } else if (platform::is_gpu_place(mask.place())) {
 #ifdef PADDLE_WITH_CUDA
-      framework::CopyFrom(mask, platform::CPUPlace(), dev_ctx, cpu_mask.get());
+      framework::Copy(mask, platform::CPUPlace(), dev_ctx, cpu_mask.get());
 #else
       PADDLE_THROW("Not supported GPU, Please compile WITH_GPU option");
 #endif
@@ -111,7 +111,7 @@ class SplitLoDTensorOp : public framework::OperatorBase {
         // out[offset: offset+len] = x[each_range.begin: each_range.end]
         auto slice = out->Slice(static_cast<int>(offset),
                                 static_cast<int>(offset + len));
-        framework::CopyFrom(x.Slice(static_cast<int>(each_range.begin),
+        framework::Copy(x.Slice(static_cast<int>(each_range.begin),
                                 static_cast<int>(each_range.end)),
                         x.place(), dev_ctx, &slice);
         offset += len;

paddle/operators/sum_op.h

@@ -107,7 +107,7 @@ class SumKernel : public framework::OpKernel<T> {
               out_array.resize(i + 1);
             }
             if (out_array[i].numel() == 0) {
-              framework::CopyFrom(in_array[i], in_array[i].place(),
+              framework::Copy(in_array[i], in_array[i].place(),
                               context.device_context(), &out_array[i]);
               out_array[i].set_lod(in_array[i].lod());
             } else {

paddle/operators/tensor_array_read_write_op.cc

@@ -44,7 +44,7 @@ class WriteToArrayOp : public ArrayOp {
           platform::DeviceContextPool::Instance();
       auto &dev_ctx = *pool.Get(place);
 
-      CopyFrom(x_tensor, place, dev_ctx, out_tensor);
+      Copy(x_tensor, place, dev_ctx, out_tensor);
       out_tensor->set_lod(x_tensor.lod());
     } else {
       VLOG(10) << "WARNING: The input tensor 'x_tensor' holds no memory, so "
@@ -135,7 +135,7 @@ class ReadFromArrayOp : public ArrayOp {
       platform::DeviceContextPool &pool =
           platform::DeviceContextPool::Instance();
       auto &dev_ctx = *pool.Get(place);
-      framework::CopyFrom(x_array[offset], place, dev_ctx, out_tensor);
+      framework::Copy(x_array[offset], place, dev_ctx, out_tensor);
       out_tensor->set_lod(x_array[offset].lod());
     } else {
       VLOG(10) << "offset " << offset << " >= " << x_array.size();