PaddlePaddle
diff --git a/‎paddle/fluid/operators/beam_search_decode_op_xpu.h
Copy file name to clipboard
+11-10Lines changed: 11 additions & 10 deletions b/‎paddle/fluid/operators/beam_search_decode_op_xpu.h
Copy file name to clipboard
+11-10Lines changed: 11 additions & 10 deletions
diff --git a/‎paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op_xpu.cc
Copy file name to clipboard
+11-10Lines changed: 11 additions & 10 deletions b/‎paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op_xpu.cc
Copy file name to clipboard
+11-10Lines changed: 11 additions & 10 deletions
diff --git a/‎paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
Copy file name to clipboard
+25-24Lines changed: 25 additions & 24 deletions b/‎paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
Copy file name to clipboard
+25-24Lines changed: 25 additions & 24 deletions
diff --git a/‎paddle/fluid/operators/detection/generate_proposals_op.cu
Copy file name to clipboard
+25-24Lines changed: 25 additions & 24 deletions b/‎paddle/fluid/operators/detection/generate_proposals_op.cu
Copy file name to clipboard
+25-24Lines changed: 25 additions & 24 deletions
diff --git a/‎paddle/fluid/operators/fused/fused_dropout_helper.h
Copy file name to clipboard
+7-6Lines changed: 7 additions & 6 deletions b/‎paddle/fluid/operators/fused/fused_dropout_helper.h
Copy file name to clipboard
+7-6Lines changed: 7 additions & 6 deletions
diff --git a/‎paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h
Copy file name to clipboard
+7-6Lines changed: 7 additions & 6 deletions b/‎paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h
Copy file name to clipboard
+7-6Lines changed: 7 additions & 6 deletions
diff --git a/‎paddle/fluid/operators/fused/fused_residual_dropout_bias.h
Copy file name to clipboard
+13-12Lines changed: 13 additions & 12 deletions b/‎paddle/fluid/operators/fused/fused_residual_dropout_bias.h
Copy file name to clipboard
+13-12Lines changed: 13 additions & 12 deletions
diff --git a/‎paddle/fluid/operators/lookup_table_op.cu
Copy file name to clipboard
+13-12Lines changed: 13 additions & 12 deletions b/‎paddle/fluid/operators/lookup_table_op.cu
Copy file name to clipboard
+13-12Lines changed: 13 additions & 12 deletions
@@ -14,6 +14,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/fluid/operators/beam_search_decode_op_def.h"
+#include "paddle/phi/common/memory_utils.h"
 
 namespace paddle {
 namespace operators {
@@ -50,18 +51,18 @@ int CopyTensorByXPU(const phi::DenseTensor& srcTensor,
   if (flag == 0) {
     T* dstData =
         dstTensor->template mutable_data<T>(paddle::platform::CPUPlace());
-    paddle::memory::Copy(paddle::platform::CPUPlace(),
-                         dstData,
-                         place,
-                         srcData,
-                         srcTensor.numel() * sizeof(T));
+    phi::memory_utils::Copy(paddle::platform::CPUPlace(),
+                            dstData,
+                            place,
+                            srcData,
+                            srcTensor.numel() * sizeof(T));
   } else {
     T* dstData = dstTensor->template mutable_data<T>(place);
-    paddle::memory::Copy(place,
-                         dstData,
-                         paddle::platform::CPUPlace(),
-                         srcData,
-                         srcTensor.numel() * sizeof(T));
+    phi::memory_utils::Copy(place,
+                            dstData,
+                            paddle::platform::CPUPlace(),
+                            srcData,
+                            srcTensor.numel() * sizeof(T));
   }
 
   return xpu::Error_t::SUCCESS;
 
@@ -18,6 +18,7 @@ limitations under the License. */
 #include "paddle/fluid/platform/collective_helper.h"
 #include "paddle/fluid/platform/device/xpu/bkcl_helper.h"
 #include "paddle/phi/backends/xpu/xpu_context.h"
+#include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/funcs/axis_utils.h"
 #include "paddle/phi/kernels/funcs/cross_entropy.h"
@@ -248,11 +249,11 @@ struct CSoftmaxWithCrossEntropyProcessGroupFunctor<phi::XPUContext, T> {
         N * 1);
     PADDLE_ENFORCE_XDNN_SUCCESS(ret, "sub");
 
-    memory::Copy(ctx.GetPlace(),
-                 softmax->data(),
-                 ctx.GetPlace(),
-                 softmax_2d.data(),
-                 N * D * sizeof(T));
+    phi::memory_utils::Copy(ctx.GetPlace(),
+                            softmax->data(),
+                            ctx.GetPlace(),
+                            softmax_2d.data(),
+                            N * D * sizeof(T));
   }
 };
 
@@ -513,11 +514,11 @@ struct CSoftmaxWithCrossEntropyFunctor<phi::XPUContext, T> {
         N * 1);
     PADDLE_ENFORCE_XDNN_SUCCESS(ret, "sub");
 
-    memory::Copy(ctx.GetPlace(),
-                 softmax->data(),
-                 ctx.GetPlace(),
-                 softmax_2d.data(),
-                 N * D * sizeof(T));
+    phi::memory_utils::Copy(ctx.GetPlace(),
+                            softmax->data(),
+                            ctx.GetPlace(),
+                            softmax_2d.data(),
+                            N * D * sizeof(T));
   }
 };
 
 
@@ -25,6 +25,7 @@ namespace cub = hipcub;
 #include "paddle/fluid/operators/detection/collect_fpn_proposals_op.h"
 #include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/backends/gpu/gpu_primitives.h"
+#include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/core/mixed_vector.h"
 #include "paddle/phi/kernels/funcs/concat_and_split_functor.h"
 #include "paddle/phi/kernels/funcs/gather.cu.h"
@@ -111,18 +112,18 @@ class GPUCollectFpnProposalsOpKernel : public framework::OpKernel<T> {
         }
       }
 
-      memory::Copy(place,
-                   concat_rois_data + roi_offset,
-                   place,
-                   roi_in->data<T>(),
-                   roi_in->numel() * sizeof(T),
-                   dev_ctx.stream());
-      memory::Copy(place,
-                   concat_scores_data + score_offset,
-                   place,
-                   score_in->data<T>(),
-                   score_in->numel() * sizeof(T),
-                   dev_ctx.stream());
+      phi::memory_utils::Copy(place,
+                              concat_rois_data + roi_offset,
+                              place,
+                              roi_in->data<T>(),
+                              roi_in->numel() * sizeof(T),
+                              dev_ctx.stream());
+      phi::memory_utils::Copy(place,
+                              concat_scores_data + score_offset,
+                              place,
+                              score_in->data<T>(),
+                              score_in->numel() * sizeof(T),
+                              dev_ctx.stream());
       roi_offset += roi_in->numel();
       score_offset += score_in->numel();
     }
@@ -233,12 +234,12 @@ class GPUCollectFpnProposalsOpKernel : public framework::OpKernel<T> {
     GetLengthLoD<<<blocks, threads, 0, dev_ctx.stream()>>>(
         real_post_num, out_id_data, length_lod_data);
     std::vector<int> length_lod_cpu(lod_size);
-    memory::Copy(platform::CPUPlace(),
-                 length_lod_cpu.data(),
-                 place,
-                 length_lod_data,
-                 sizeof(int) * lod_size,
-                 dev_ctx.stream());
+    phi::memory_utils::Copy(platform::CPUPlace(),
+                            length_lod_cpu.data(),
+                            place,
+                            length_lod_data,
+                            sizeof(int) * lod_size,
+                            dev_ctx.stream());
     dev_ctx.Wait();
 
     std::vector<size_t> offset(1, 0);
@@ -249,12 +250,12 @@ class GPUCollectFpnProposalsOpKernel : public framework::OpKernel<T> {
     if (ctx.HasOutput("RoisNum")) {
       auto* rois_num = ctx.Output<phi::DenseTensor>("RoisNum");
       int* rois_num_data = rois_num->mutable_data<int>({lod_size}, place);
-      memory::Copy(place,
-                   rois_num_data,
-                   place,
-                   length_lod_data,
-                   lod_size * sizeof(int),
-                   dev_ctx.stream());
+      phi::memory_utils::Copy(place,
+                              rois_num_data,
+                              place,
+                              length_lod_data,
+                              lod_size * sizeof(int),
+                              dev_ctx.stream());
     }
 
     framework::LoD lod;
 
@@ -21,6 +21,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/memory/memory.h"
 #include "paddle/fluid/operators/detection/bbox_util.cu.h"
+#include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/core/mixed_vector.h"
 #include "paddle/phi/kernels/funcs/gather.cu.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
@@ -79,12 +80,12 @@ static std::pair<phi::DenseTensor, phi::DenseTensor> ProposalForOneImage(
                                               keep_index.data<int>());
   int keep_num;
   const auto gpu_place = ctx.GetPlace();
-  memory::Copy(platform::CPUPlace(),
-               &keep_num,
-               gpu_place,
-               keep_num_t.data<int>(),
-               sizeof(int),
-               ctx.stream());
+  phi::memory_utils::Copy(platform::CPUPlace(),
+                          &keep_num,
+                          gpu_place,
+                          keep_num_t.data<int>(),
+                          sizeof(int),
+                          ctx.stream());
   ctx.Wait();
   keep_index.Resize({keep_num});
 
@@ -221,18 +222,18 @@ class CUDAGenerateProposalsKernel : public framework::OpKernel<T> {
       phi::DenseTensor &proposals = box_score_pair.first;
       phi::DenseTensor &scores = box_score_pair.second;
 
-      memory::Copy(place,
-                   rpn_rois_data + num_proposals * 4,
-                   place,
-                   proposals.data<T>(),
-                   sizeof(T) * proposals.numel(),
-                   dev_ctx.stream());
-      memory::Copy(place,
-                   rpn_roi_probs_data + num_proposals,
-                   place,
-                   scores.data<T>(),
-                   sizeof(T) * scores.numel(),
-                   dev_ctx.stream());
+      phi::memory_utils::Copy(place,
+                              rpn_rois_data + num_proposals * 4,
+                              place,
+                              proposals.data<T>(),
+                              sizeof(T) * proposals.numel(),
+                              dev_ctx.stream());
+      phi::memory_utils::Copy(place,
+                              rpn_roi_probs_data + num_proposals,
+                              place,
+                              scores.data<T>(),
+                              sizeof(T) * scores.numel(),
+                              dev_ctx.stream());
       num_proposals += proposals.dims()[0];
       offset.emplace_back(num_proposals);
       tmp_num.push_back(proposals.dims()[0]);
@@ -241,12 +242,12 @@ class CUDAGenerateProposalsKernel : public framework::OpKernel<T> {
       auto *rpn_rois_num = context.Output<phi::DenseTensor>("RpnRoisNum");
       rpn_rois_num->mutable_data<int>({num}, context.GetPlace());
       int *num_data = rpn_rois_num->data<int>();
-      memory::Copy(place,
-                   num_data,
-                   cpu_place,
-                   &tmp_num[0],
-                   sizeof(int) * num,
-                   dev_ctx.stream());
+      phi::memory_utils::Copy(place,
+                              num_data,
+                              cpu_place,
+                              &tmp_num[0],
+                              sizeof(int) * num,
+                              dev_ctx.stream());
       rpn_rois_num->Resize({num});
     }
     framework::LoD lod;
 
@@ -18,6 +18,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/fused/fused_dropout_act_bias.h"
 #include "paddle/fluid/operators/fused/fused_layernorm_residual_dropout_bias.h"
 #include "paddle/fluid/operators/fused/fused_residual_dropout_bias.h"
+#include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/core/generator.h"
 #include "paddle/phi/kernels/funcs/dropout_impl_util.h"
 #include "paddle/phi/kernels/funcs/functors.h"
@@ -193,12 +194,12 @@ class FusedDropoutHelper {
         d_bias,
         ctx);
     if (d_residual) {
-      memory::Copy(ctx.GetPlace(),
-                   d_residual,
-                   ctx.GetPlace(),
-                   d_out,
-                   rows_ * cols_ * sizeof(T),
-                   ctx.stream());
+      phi::memory_utils::Copy(ctx.GetPlace(),
+                              d_residual,
+                              ctx.GetPlace(),
+                              d_out,
+                              rows_ * cols_ * sizeof(T),
+                              ctx.stream());
     }
   }
 
 
@@ -15,6 +15,7 @@ limitations under the License. */
 #pragma once
 
 #include "paddle/fluid/operators/fused/fused_residual_dropout_bias.h"
+#include "paddle/phi/common/memory_utils.h"
 
 namespace paddle {
 namespace operators {
@@ -847,12 +848,12 @@ void LaunchLayernormResidualDropoutBias(
   // NOTE(minghaoBD): OutType should be T if drop_out_rate == 1.0
   if (std::abs(dropout_prob - 1.0f) < 1e-5) {
     auto cuda_place = ctx.GetPlace();
-    memory::Copy(cuda_place,
-                 dst,
-                 cuda_place,
-                 residual,
-                 rows * cols * sizeof(T),
-                 ctx.stream());
+    phi::memory_utils::Copy(cuda_place,
+                            dst,
+                            cuda_place,
+                            residual,
+                            rows * cols * sizeof(T),
+                            ctx.stream());
     if (mask_data != nullptr) {
       PADDLE_ENFORCE_GPU_SUCCESS(cudaMemsetAsync(
           mask_data, 0, rows * cols * sizeof(MaskType), ctx.stream()));
 
@@ -16,6 +16,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/fused/fused_dropout_common.h"
 #include "paddle/phi/common/amp_type_traits.h"
+#include "paddle/phi/common/memory_utils.h"
 
 namespace paddle {
 namespace operators {
@@ -253,12 +254,12 @@ void LaunchResidualDropoutBias(const uint32_t rows,
     // NOTE(minghaoBD): OutType should be T if dropout_prob == 1.0
     if (residual == dst) return;
     if (residual) {
-      memory::Copy(ctx.GetPlace(),
-                   dst,
-                   ctx.GetPlace(),
-                   residual,
-                   rows * cols * sizeof(T),
-                   ctx.stream());
+      phi::memory_utils::Copy(ctx.GetPlace(),
+                              dst,
+                              ctx.GetPlace(),
+                              residual,
+                              rows * cols * sizeof(T),
+                              ctx.stream());
     } else {
       SetZero<T>(ctx, dst, rows * cols);
     }
@@ -475,12 +476,12 @@ void LaunchResidualDropoutBiasGrad(const T *dout,
         if (dx == nullptr || dx == dout) {                                    \
           return;                                                             \
         }                                                                     \
-        memory::Copy(ctx.GetPlace(),                                          \
-                     dx,                                                      \
-                     ctx.GetPlace(),                                          \
-                     dout,                                                    \
-                     rows *cols * sizeof(T),                                  \
-                     ctx.stream());                                           \
+        phi::memory_utils::Copy(ctx.GetPlace(),                               \
+                                dx,                                           \
+                                ctx.GetPlace(),                               \
+                                dout,                                         \
+                                rows *cols * sizeof(T),                       \
+                                ctx.stream());                                \
       } else {                                                                \
         const uint64_t n = rows * cols;                                       \
         platform::GpuLaunchConfig config =                                    \
 
@@ -16,6 +16,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/phi/backends/gpu/gpu_primitives.h"
 #include "paddle/phi/common/float16.h"
+#include "paddle/phi/common/memory_utils.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 
 namespace paddle {
@@ -175,12 +176,12 @@ class LookupTableGradCUDAKernel : public framework::OpKernel<T> {
 
       // TODO(yuyang18): Strange code here.
       phi::MixVector<int64_t> mixv_new_rows(&new_rows);
-      memory::Copy(gpu_place,
-                   mixv_new_rows.CUDAMutableData(context.GetPlace()),
-                   gpu_place,
-                   ids_data,
-                   ids_num * sizeof(int64_t),
-                   stream);
+      phi::memory_utils::Copy(gpu_place,
+                              mixv_new_rows.CUDAMutableData(context.GetPlace()),
+                              gpu_place,
+                              ids_data,
+                              ids_num * sizeof(int64_t),
+                              stream);
       mixv_new_rows.CopyToCPU();
       d_table->set_rows(new_rows);
 
@@ -202,12 +203,12 @@ class LookupTableGradCUDAKernel : public framework::OpKernel<T> {
                             "output@Grad's shape = [%s].",
                             d_table_value->dims(),
                             d_output_dims_2d));
-      memory::Copy(gpu_place,
-                   d_table_data,
-                   gpu_place,
-                   d_output_data,
-                   d_output->numel() * sizeof(T),
-                   stream);
+      phi::memory_utils::Copy(gpu_place,
+                              d_table_data,
+                              gpu_place,
+                              d_output_data,
+                              d_output->numel() * sizeof(T),
+                              stream);
 
     } else {
       auto ids_t = context.Input<phi::DenseTensor>("Ids");