diff --git a/paddle/fluid/operators/cross_entropy_op.h b/paddle/fluid/operators/cross_entropy_op.h
index d3ba49ce9b3b90..a997720ae184c4 100644
--- a/paddle/fluid/operators/cross_entropy_op.h
+++ b/paddle/fluid/operators/cross_entropy_op.h
@@ -14,10 +14,10 @@ limitations under the License. */
 
 #pragma once
 #include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/core/tensor_utils.h"
 #include "paddle/phi/kernels/funcs/cross_entropy.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
@@ -142,7 +142,7 @@ class CrossEntropyGradientOpKernel : public framework::OpKernel<T> {
                                         x->data<T>(),
                                         label->data<T>(),
                                         static_cast<size_t>(class_num));
-      platform::ForRange<DeviceContext> for_range(
+      phi::funcs::ForRange<DeviceContext> for_range(
           ctx.template device_context<DeviceContext>(),
           static_cast<size_t>(dx->numel()));
       for_range(functor);
@@ -153,7 +153,7 @@ class CrossEntropyGradientOpKernel : public framework::OpKernel<T> {
                                label->data<int64_t>(),
                                static_cast<size_t>(class_num),
                                static_cast<size_t>(ignore_index));
-      platform::ForRange<DeviceContext> for_range(
+      phi::funcs::ForRange<DeviceContext> for_range(
           ctx.template device_context<DeviceContext>(),
           static_cast<size_t>(dy->numel()));
       for_range(functor);
@@ -259,7 +259,7 @@ class CrossEntropyOpKernel2 : public framework::OpKernel<T> {
 
     auto ignore_index = ctx.Attr<int>("ignore_index");
 
-    platform::ForRange<DeviceContext> for_range(
+    phi::funcs::ForRange<DeviceContext> for_range(
         ctx.template device_context<DeviceContext>(), batch_size);
     for_range(HardLabelCrossEntropyForwardFunctor<T>(
         p_x, p_y, p_match_x, p_label, ignore_index, feature_size));
@@ -285,7 +285,7 @@ class CrossEntropyGradientOpKernel2 : public framework::OpKernel<T> {
     int64_t feature_size = dx->dims()[rank - 1];
     int64_t batch_size = common::product(dx->dims()) / feature_size;
 
-    platform::ForRange<DeviceContext> for_range(
+    phi::funcs::ForRange<DeviceContext> for_range(
         ctx.template device_context<DeviceContext>(),
         batch_size * feature_size);
     for_range(HardLabelCrossEntropyBackwardFunctor<T>(
diff --git a/paddle/fluid/operators/detection/collect_fpn_proposals_op.cu b/paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
index e599bf319317bb..a600b97ef0f330 100644
--- a/paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
+++ b/paddle/fluid/operators/detection/collect_fpn_proposals_op.cu
@@ -23,11 +23,11 @@ namespace cub = hipcub;
 #include "paddle/fluid/memory/memcpy.h"
 #include "paddle/fluid/operators/detection/bbox_util.h"
 #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/for_range.h"
 #include "paddle/phi/kernels/funcs/gather.cu.h"
 #include "paddle/phi/kernels/funcs/strided_memcpy.h"
 
@@ -139,7 +139,8 @@ class GPUCollectFpnProposalsOpKernel : public framework::OpKernel<T> {
     phi::DenseTensor index_in_t;
     int* idx_in =
         index_in_t.mutable_data<int>({total_roi_num}, dev_ctx.GetPlace());
-    platform::ForRange<phi::GPUContext> for_range_total(dev_ctx, total_roi_num);
+    phi::funcs::ForRange<phi::GPUContext> for_range_total(dev_ctx,
+                                                          total_roi_num);
     for_range_total(RangeInitFunctor{0, 1, idx_in});
 
     phi::DenseTensor keys_out_t;
@@ -188,7 +189,8 @@ class GPUCollectFpnProposalsOpKernel : public framework::OpKernel<T> {
     phi::DenseTensor batch_index_t;
     int* batch_idx_in =
         batch_index_t.mutable_data<int>({real_post_num}, dev_ctx.GetPlace());
-    platform::ForRange<phi::GPUContext> for_range_post(dev_ctx, real_post_num);
+    phi::funcs::ForRange<phi::GPUContext> for_range_post(dev_ctx,
+                                                         real_post_num);
     for_range_post(RangeInitFunctor{0, 1, batch_idx_in});
 
     phi::DenseTensor out_id_t;
diff --git a/paddle/fluid/operators/detection/generate_proposals_op.cu b/paddle/fluid/operators/detection/generate_proposals_op.cu
index f7f6246d7724c7..69b598be7e0419 100644
--- a/paddle/fluid/operators/detection/generate_proposals_op.cu
+++ b/paddle/fluid/operators/detection/generate_proposals_op.cu
@@ -57,7 +57,7 @@ static std::pair<phi::DenseTensor, phi::DenseTensor> ProposalForOneImage(
   proposals.mutable_data<T>({pre_nms_num, 4}, ctx.GetPlace());
 
   {
-    platform::ForRange<phi::GPUContext> for_range(ctx, pre_nms_num);
+    phi::funcs::ForRange<phi::GPUContext> for_range(ctx, pre_nms_num);
     for_range(BoxDecodeAndClipFunctor<T>{anchors.data<T>(),
                                          bbox_deltas.data<T>(),
                                          variances.data<T>(),
diff --git a/paddle/fluid/operators/elementwise/elementwise_op_function.h b/paddle/fluid/operators/elementwise/elementwise_op_function.h
index dc6d23a106926e..085022c76af840 100644
--- a/paddle/fluid/operators/elementwise/elementwise_op_function.h
+++ b/paddle/fluid/operators/elementwise/elementwise_op_function.h
@@ -48,7 +48,7 @@ limitations under the License. */
 
 #endif
 
-#include "paddle/fluid/platform/for_range.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 #define DIVUP(x, y) (((x) + (y)-1) / (y))
@@ -474,7 +474,7 @@ void FusedElemwiseAndActComputeNoBroadcast(
     phi::DenseTensor *intermediate_out) {
   size_t N = static_cast<size_t>(common::product(x_dim));
 
-  platform::ForRange<DeviceContext> for_range(
+  phi::funcs::ForRange<DeviceContext> for_range(
       ctx.template device_context<DeviceContext>(), N);
 
   for_range(
@@ -654,7 +654,7 @@ void FusedElemwiseAndActGradComputeNoBroadcast(
     DY_OP dy_op,
     DIntermediate_OP dintermediate_op) {
   size_t N = static_cast<size_t>(common::product(x_dim));
-  platform::ForRange<DeviceContext> for_range(
+  phi::funcs::ForRange<DeviceContext> for_range(
       ctx.template device_context<DeviceContext>(), N);
   const T *x_data = nullptr;
   const T *y_data = nullptr;
diff --git a/paddle/fluid/operators/optimizers/ftrl_op.h b/paddle/fluid/operators/optimizers/ftrl_op.h
index 6909a2b5e7c4a5..39f191aef57590 100644
--- a/paddle/fluid/operators/optimizers/ftrl_op.h
+++ b/paddle/fluid/operators/optimizers/ftrl_op.h
@@ -14,8 +14,8 @@ limitations under the License. */
 
 #pragma once
 #include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/selected_rows_functor.h"
 
 namespace paddle {
@@ -202,7 +202,7 @@ class FTRLOpKernel : public framework::OpKernel<T> {
       auto row_numel = static_cast<int64_t>(merged_grad->value().dims()[1]);
       auto row_height = static_cast<int64_t>(merged_grad->rows().size());
 
-      platform::ForRange<DeviceContext> for_range(
+      phi::funcs::ForRange<DeviceContext> for_range(
           static_cast<const DeviceContext&>(ctx.device_context()),
           row_numel * row_height);
 
diff --git a/paddle/fluid/operators/optimizers/sparse_momentum_op.h b/paddle/fluid/operators/optimizers/sparse_momentum_op.h
index 55e8717928b64b..b23295e64c1970 100644
--- a/paddle/fluid/operators/optimizers/sparse_momentum_op.h
+++ b/paddle/fluid/operators/optimizers/sparse_momentum_op.h
@@ -20,10 +20,10 @@
 
 #include "paddle/fluid/framework/convert_utils.h"
 #include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/common/amp_type_traits.h"
 #include "paddle/phi/common/float16.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 #ifdef __NVCC__
 #include "cub/cub.cuh"
@@ -436,7 +436,7 @@ class SparseMomentumOpKernel : public framework::OpKernel<T> {
 
     auto grad = ctx.Input<phi::DenseTensor>("Grad");
 
-    platform::ForRange<DeviceContext> for_range(
+    phi::funcs::ForRange<DeviceContext> for_range(
         static_cast<const DeviceContext&>(ctx.device_context()),
         param->numel());
 
@@ -465,7 +465,7 @@ class SparseMomentumOpKernel : public framework::OpKernel<T> {
       auto sort_value_ptr =
           sort_value.mutable_data<IndexT>({num_index}, ctx.GetPlace());
 
-      platform::ForRange<DeviceContext> for_range_index(
+      phi::funcs::ForRange<DeviceContext> for_range_index(
           static_cast<const DeviceContext&>(ctx.device_context()), num_index);
       RangeFunctor<IndexT> range_functor(sort_value_ptr);
       for_range_index(range_functor);
diff --git a/paddle/fluid/operators/svd_helper.h b/paddle/fluid/operators/svd_helper.h
index aa9a7f439121b1..4f56efc7422a9c 100644
--- a/paddle/fluid/operators/svd_helper.h
+++ b/paddle/fluid/operators/svd_helper.h
@@ -26,10 +26,10 @@
 #include "paddle/fluid/operators/diag_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_op_function.h"
 #include "paddle/fluid/platform/device_context.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/complex_functors.h"
 #include "paddle/phi/kernels/funcs/eigen/eigen_function.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/lapack/lapack_function.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
@@ -666,7 +666,7 @@ struct DeviceIndependenceTensorOperations {
                             const phi::DenseTensor& input) {
     phi::DenseTensor out;
     auto& dev_ctx = context.template device_context<DeviceContext>();
-    platform::ForRange<DeviceContext> for_range(dev_ctx, input.numel());
+    phi::funcs::ForRange<DeviceContext> for_range(dev_ctx, input.numel());
     DiagAndFillFunctor<T, ValueType> diag_and_copy_functor(
         m,
         n,
@@ -685,9 +685,9 @@ struct DeviceIndependenceTensorOperations {
     auto& dev_ctx = context.template device_context<DeviceContext>();
     return phi::funcs::GetBlas<DeviceContext, T>(dev_ctx);
   }
-  platform::ForRange<DeviceContext> GetForRange(int numel) {
+  phi::funcs::ForRange<DeviceContext> GetForRange(int numel) {
     auto& dev_ctx = context.template device_context<DeviceContext>();
-    return platform::ForRange<DeviceContext>(dev_ctx, numel);
+    return phi::funcs::ForRange<DeviceContext>(dev_ctx, numel);
   }
   template <size_t D>
   void EigenSliceWrapper(const phi::DenseTensor* in,