diff --git a/paddle/phi/kernels/cpu/fill_grad_kernel.cc b/paddle/phi/kernels/cpu/fill_grad_kernel.cc
index 07448c85a57d6054cb82e15d3a64cac2505c0234..fda723ee2ba801480c6f83ee07626cdc69b20741 100644
--- a/paddle/phi/kernels/cpu/fill_grad_kernel.cc
+++ b/paddle/phi/kernels/cpu/fill_grad_kernel.cc
@@ -25,6 +25,6 @@ PD_REGISTER_KERNEL(fill_grad,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}
diff --git a/paddle/phi/kernels/cpu/fill_kernel.cc b/paddle/phi/kernels/cpu/fill_kernel.cc
index adca39e6ab95d36b1e5c8a90ac4abd2fc4512a26..7e8534a5fbeb86f845f4b949a45ca197213f71af 100644
--- a/paddle/phi/kernels/cpu/fill_kernel.cc
+++ b/paddle/phi/kernels/cpu/fill_kernel.cc
@@ -25,6 +25,6 @@ PD_REGISTER_KERNEL(fill,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}
diff --git a/paddle/phi/kernels/funcs/math_function.cc b/paddle/phi/kernels/funcs/math_function.cc
index 756fd8782e96042539d6a37d4fbcdc6b06468bb1..8c32f8bcd61f4fb69c458a55454cce4e294d65b4 100644
--- a/paddle/phi/kernels/funcs/math_function.cc
+++ b/paddle/phi/kernels/funcs/math_function.cc
@@ -27,9 +27,9 @@ limitations under the License. */
 #include <vector>
 
 #include "paddle/fluid/platform/bfloat16.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/phi/backends/cpu/cpu_context.h"
 #include "paddle/phi/common/data_type.h"
+#include "paddle/phi/common/float16.h"
 #include "paddle/phi/kernels/funcs/eigen/common.h"
 #include "paddle/phi/kernels/funcs/math_function_impl.h"
 #include "unsupported/Eigen/CXX11/Tensor"
diff --git a/paddle/phi/kernels/funcs/math_function.cu b/paddle/phi/kernels/funcs/math_function.cu
index 7e7a6c30c34fc3cc51155089517009d45ef925c8..decd668e3a358901aa6ad90483f5cdc25fd3c076 100644
--- a/paddle/phi/kernels/funcs/math_function.cu
+++ b/paddle/phi/kernels/funcs/math_function.cu
@@ -17,9 +17,9 @@ limitations under the License. */
 #include "paddle/fluid/memory/malloc.h"
 #include "paddle/fluid/memory/memcpy.h"
 #include "paddle/fluid/platform/bfloat16.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/common/data_type.h"
+#include "paddle/phi/common/float16.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 #include "paddle/phi/kernels/funcs/math_function_impl.h"
diff --git a/paddle/phi/kernels/gpu/bmm_grad_kernel.cu b/paddle/phi/kernels/gpu/bmm_grad_kernel.cu
index 1e51d373e097a0bb5ede62a1ef51eed6951c41a6..2c77d0e900461570f96cd6a3df41215afb310604 100644
--- a/paddle/phi/kernels/gpu/bmm_grad_kernel.cu
+++ b/paddle/phi/kernels/gpu/bmm_grad_kernel.cu
@@ -24,4 +24,4 @@ PD_REGISTER_KERNEL(bmm_grad,
                    phi::BmmGradKernel,
                    float,
                    double,
-                   paddle::platform::float16) {}
+                   phi::dtype::float16) {}
diff --git a/paddle/phi/kernels/gpu/bmm_kernel.cu b/paddle/phi/kernels/gpu/bmm_kernel.cu
index 36dfad3d6a037cfd17d16979e343ae409d927007..b7ec1ce33000e263e60f0c4537230bd03d309d30 100644
--- a/paddle/phi/kernels/gpu/bmm_kernel.cu
+++ b/paddle/phi/kernels/gpu/bmm_kernel.cu
@@ -18,10 +18,5 @@
 #include "paddle/phi/core/kernel_registry.h"
 #include "paddle/phi/kernels/impl/bmm_kernel_impl.h"
 
-PD_REGISTER_KERNEL(bmm,
-                   GPU,
-                   ALL_LAYOUT,
-                   phi::BmmKernel,
-                   float,
-                   double,
-                   paddle::platform::float16) {}
+PD_REGISTER_KERNEL(
+    bmm, GPU, ALL_LAYOUT, phi::BmmKernel, float, double, phi::dtype::float16) {}
diff --git a/paddle/phi/kernels/gpu/cholesky_kernel.cu b/paddle/phi/kernels/gpu/cholesky_kernel.cu
index 7c4a497703ecdecdf248fe92fb4146765a5e2dcd..41fc8387c4ea061fdc73c8c18753241167a80c72 100644
--- a/paddle/phi/kernels/gpu/cholesky_kernel.cu
+++ b/paddle/phi/kernels/gpu/cholesky_kernel.cu
@@ -23,10 +23,10 @@ limitations under the License. */
 #include <vector>
 
 #include "paddle/fluid/memory/memory.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/backends/dynload/cusolver.h"
 #include "paddle/phi/backends/gpu/gpu_context.h"
 #include "paddle/phi/core/kernel_registry.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 
@@ -130,7 +130,7 @@ void CholeskyKernel(const Context& dev_ctx,
       upper ? CUBLAS_FILL_MODE_LOWER : CUBLAS_FILL_MODE_UPPER;
   // portf is inplace, thus copy the triangular part of the input matrices to
   // the output and set the other triangular part to 0 firstly
-  paddle::platform::ForRange<GPUContext> for_range(dev_ctx, tensor_size);
+  phi::funcs::ForRange<GPUContext> for_range(dev_ctx, tensor_size);
   if (upper) {
     MatrixBandPartFunctor<T> matrix_band_part_functor(m,
                                                       m,
diff --git a/paddle/phi/kernels/gpu/fill_grad_kernel.cu b/paddle/phi/kernels/gpu/fill_grad_kernel.cu
index e18bb5c6dbb2446eae98969075a6da8ecd6fd4bd..512aef8c1c9841713af598e9c613cbd874bf1cfc 100644
--- a/paddle/phi/kernels/gpu/fill_grad_kernel.cu
+++ b/paddle/phi/kernels/gpu/fill_grad_kernel.cu
@@ -26,6 +26,6 @@ PD_REGISTER_KERNEL(fill_grad,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}
diff --git a/paddle/phi/kernels/gpu/fill_kernel.cu b/paddle/phi/kernels/gpu/fill_kernel.cu
index 3fedb4118ff9e111828ac832bd25b04101060be1..de46e678782f2d6d68a6cf78afdb8f0ea0368338 100644
--- a/paddle/phi/kernels/gpu/fill_kernel.cu
+++ b/paddle/phi/kernels/gpu/fill_kernel.cu
@@ -26,6 +26,6 @@ PD_REGISTER_KERNEL(fill,
                    double,
                    int64_t,
                    int,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::bfloat16,
                    bool) {}
diff --git a/paddle/phi/kernels/gpu/overlap_add_grad_kernel.cu b/paddle/phi/kernels/gpu/overlap_add_grad_kernel.cu
index 4957ade2c37bc98fcbfb44820ac171dc4eb3493e..5d68b503d89c84ed3c5808b95894cdc406b4def2 100644
--- a/paddle/phi/kernels/gpu/overlap_add_grad_kernel.cu
+++ b/paddle/phi/kernels/gpu/overlap_add_grad_kernel.cu
@@ -160,6 +160,6 @@ PD_REGISTER_KERNEL(overlap_add_grad,
                    int64_t,
                    float,
                    double,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::complex<float>,
                    paddle::platform::complex<double>) {}
diff --git a/paddle/phi/kernels/gpu/overlap_add_kernel.cu b/paddle/phi/kernels/gpu/overlap_add_kernel.cu
index 47989ebcda816514b3370dca074cd8a153564d0d..86d5920754172d4e112b23a4b2eae8fb524680d6 100644
--- a/paddle/phi/kernels/gpu/overlap_add_kernel.cu
+++ b/paddle/phi/kernels/gpu/overlap_add_kernel.cu
@@ -146,6 +146,6 @@ PD_REGISTER_KERNEL(overlap_add,
                    int64_t,
                    float,
                    double,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    paddle::platform::complex<float>,
                    paddle::platform::complex<double>) {}
diff --git a/paddle/phi/kernels/gpu/set_value_kernel.cu b/paddle/phi/kernels/gpu/set_value_kernel.cu
index 1a268c2f6b089b2dc77d1480acc0bf0ca56e0d76..2cd0e66675bfe2881768d81a89eca5e2983335eb 100644
--- a/paddle/phi/kernels/gpu/set_value_kernel.cu
+++ b/paddle/phi/kernels/gpu/set_value_kernel.cu
@@ -28,7 +28,7 @@ PD_REGISTER_KERNEL(set_value,
                    int,
                    int64_t,
                    bool,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    phi::dtype::complex<float>,
                    phi::dtype::complex<double>) {}
 PD_REGISTER_KERNEL(set_value_with_tensor,
@@ -40,6 +40,6 @@ PD_REGISTER_KERNEL(set_value_with_tensor,
                    int,
                    int64_t,
                    bool,
-                   paddle::platform::float16,
+                   phi::dtype::float16,
                    phi::dtype::complex<float>,
                    phi::dtype::complex<double>) {}
diff --git a/paddle/phi/kernels/gpudnn/conv_grad_kernel.cu b/paddle/phi/kernels/gpudnn/conv_grad_kernel.cu
index 43b0d0a81d43830aca62549d2408a0c8c5376d46..276480ed54dbd4b77b3e804b887e5cf58feeb978 100644
--- a/paddle/phi/kernels/gpudnn/conv_grad_kernel.cu
+++ b/paddle/phi/kernels/gpudnn/conv_grad_kernel.cu
@@ -24,7 +24,6 @@
 #endif
 
 #include "paddle/fluid/platform/cudnn_workspace_helper.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/fluid/platform/profiler.h"
 #include "paddle/phi/common/bfloat16.h"
 #include "paddle/phi/common/float16.h"
diff --git a/paddle/phi/kernels/impl/atan2_grad_kernel_impl.h b/paddle/phi/kernels/impl/atan2_grad_kernel_impl.h
index 8b3ced7387ae7ac3ab32990be10312491a58b0ac..d0dd18298518ab351918aa2492eb48d11d3cf1d7 100644
--- a/paddle/phi/kernels/impl/atan2_grad_kernel_impl.h
+++ b/paddle/phi/kernels/impl/atan2_grad_kernel_impl.h
@@ -14,9 +14,9 @@
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/kernels/atan2_grad_kernel.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 
diff --git a/paddle/phi/kernels/impl/atan2_kernel_impl.h b/paddle/phi/kernels/impl/atan2_kernel_impl.h
index e80256b7254ccc7736ff7d445e425a7f6169fcf4..2cae914e2f61555377f7a41b3d89cdbb2b589247 100644
--- a/paddle/phi/kernels/impl/atan2_kernel_impl.h
+++ b/paddle/phi/kernels/impl/atan2_kernel_impl.h
@@ -14,9 +14,9 @@
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/core/dense_tensor.h"
 #include "paddle/phi/kernels/atan2_kernel.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 template <typename T>
diff --git a/paddle/phi/kernels/impl/cholesky_grad_kernel_impl.h b/paddle/phi/kernels/impl/cholesky_grad_kernel_impl.h
index 7ffd69e16ee85307266af3a1c379d0a28c652553..8e5fab281e795bf5f1e1f336725832f683265ae1 100644
--- a/paddle/phi/kernels/impl/cholesky_grad_kernel_impl.h
+++ b/paddle/phi/kernels/impl/cholesky_grad_kernel_impl.h
@@ -14,9 +14,9 @@ limitations under the License. */
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/kernels/cholesky_grad_kernel.h"
 #include "paddle/phi/kernels/funcs/blas/blas.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 
@@ -280,7 +280,7 @@ void CholeskyGradKernel(const Context& dev_ctx,
   blas.MatMul(l, trans_desc, l_grad, no_trans_desc, T(1), &middle, T(0));
 
   /*! phi.tril_().diagonal(0, -2, -1).mul_(0.5) */
-  paddle::platform::ForRange<Context> for_range(dev_ctx, tensor_size);
+  phi::funcs::ForRange<Context> for_range(dev_ctx, tensor_size);
   MatrixBandPartScaleEndFunctor<T> matrix_band_part_scale_end_functor(
       m,
       m,
diff --git a/paddle/phi/kernels/impl/conv_cudnn_impl.h b/paddle/phi/kernels/impl/conv_cudnn_impl.h
index b66dd60279fcf0780b700884e2a4f553426d990c..cf419cd1fd4b38a26e8888b401899ff26b9c1405 100644
--- a/paddle/phi/kernels/impl/conv_cudnn_impl.h
+++ b/paddle/phi/kernels/impl/conv_cudnn_impl.h
@@ -24,9 +24,9 @@
 #endif
 
 #include "paddle/fluid/platform/cudnn_workspace_helper.h"
-#include "paddle/fluid/platform/float16.h"
 #include "paddle/fluid/platform/profiler.h"
 #include "paddle/phi/backends/dynload/cudnn.h"
+#include "paddle/phi/common/float16.h"
 #include "paddle/phi/kernels/cpu/conv_util.h"
 #include "paddle/phi/kernels/funcs/batch_norm_utils.h"
 #include "paddle/phi/kernels/funcs/padding.h"
diff --git a/paddle/phi/kernels/impl/eye_kernel_impl.h b/paddle/phi/kernels/impl/eye_kernel_impl.h
index 57b9ce73e89ce10c56ffdc88c573a9ae97f012fc..2d373f99a277af5225e01eb38a9f115239bdb4e6 100644
--- a/paddle/phi/kernels/impl/eye_kernel_impl.h
+++ b/paddle/phi/kernels/impl/eye_kernel_impl.h
@@ -14,8 +14,8 @@
 
 #pragma once
 
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/common/scalar.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 #include "paddle/phi/kernels/funcs/math_function.h"
 
 namespace phi {
@@ -48,7 +48,7 @@ void EyeKernel(const Context& ctx,
   phi::funcs::SetConstant<Context, T> set_zero;
   set_zero(ctx, out, static_cast<T>(0));
   int64_t num_eyes = (std::min)(rows, columns);
-  paddle::platform::ForRange<Context> for_range(ctx, num_eyes);
+  phi::funcs::ForRange<Context> for_range(ctx, num_eyes);
   EyeFunctor<T> functor(columns, out_data);
   for_range(functor);
 }
diff --git a/paddle/phi/kernels/impl/selu_grad_kernel_impl.h b/paddle/phi/kernels/impl/selu_grad_kernel_impl.h
index 4f6550b9bec24718ac6ab6e6397071f564f4a4b6..25d5e4e5112b23254bfd9d6fb860963adbab5625 100644
--- a/paddle/phi/kernels/impl/selu_grad_kernel_impl.h
+++ b/paddle/phi/kernels/impl/selu_grad_kernel_impl.h
@@ -28,7 +28,7 @@ void SeluGradKernel(const Context& dev_ctx,
   SeluGradFunctor<T> functor(
       out.data<T>(), dout.data<T>(), alpha, scale, dx_ptr);
   size_t limit = static_cast<size_t>(out.numel());
-  paddle::platform::ForRange<Context> for_range(dev_ctx, limit);
+  phi::funcs::ForRange<Context> for_range(dev_ctx, limit);
   for_range(functor);
 }
 }  // namespace phi
diff --git a/paddle/phi/kernels/impl/selu_kernel_impl.h b/paddle/phi/kernels/impl/selu_kernel_impl.h
index 0725b141252bcb3408ad501d504bab979dd3f59e..c5d756e6eb4fae0750f6fde256dfcf9ecbbbc687 100644
--- a/paddle/phi/kernels/impl/selu_kernel_impl.h
+++ b/paddle/phi/kernels/impl/selu_kernel_impl.h
@@ -16,8 +16,8 @@
 #include <string>
 
 #include "paddle/fluid/operators/math.h"
-#include "paddle/fluid/platform/for_range.h"
 #include "paddle/phi/core/dense_tensor.h"
+#include "paddle/phi/kernels/funcs/for_range.h"
 
 namespace phi {
 
@@ -86,7 +86,7 @@ void SeluKernel(const Context& dev_ctx,
   auto out_ptr = dev_ctx.template Alloc<T>(out);
   SeluFunctor<T> functor(x.data<T>(), alpha, scale, out_ptr);
   size_t limit = static_cast<size_t>(x.numel());
-  paddle::platform::ForRange<Context> for_range(dev_ctx, limit);
+  phi::funcs::ForRange<Context> for_range(dev_ctx, limit);
   for_range(functor);
 }
 }  // namespace phi
diff --git a/paddle/phi/kernels/xpu/bmm_grad_kernel.cc b/paddle/phi/kernels/xpu/bmm_grad_kernel.cc
index 246da888d2ca572ef64ffdebd0c3b8b0983f6ec2..3d19c8cecd7d4f09300cc3676a8510fa82a0a8bc 100644
--- a/paddle/phi/kernels/xpu/bmm_grad_kernel.cc
+++ b/paddle/phi/kernels/xpu/bmm_grad_kernel.cc
@@ -27,7 +27,7 @@ void MatMul(const Context& dev_ctx,
             DenseTensor* out) {
   dev_ctx.template Alloc<T>(out);
   xpu::Context* xpu_ctx = dev_ctx.x_context();
-  if (std::is_same<paddle::platform::float16, T>::value) {
+  if (std::is_same<phi::dtype::float16, T>::value) {
     MatMulXPUFunction<T, int16_t>(a, b, out, trans_a, trans_b, xpu_ctx);
   } else {
     if (std::getenv("XPU_PADDLE_FC_INT32") != nullptr) {
@@ -99,9 +99,6 @@ void BmmGradKernel(const Context& dev_ctx,
 
 }  // namespace phi
 
-PD_REGISTER_KERNEL(bmm_grad,
-                   XPU,
-                   ALL_LAYOUT,
-                   phi::BmmGradKernel,
-                   float,
-                   paddle::platform::float16) {}
+PD_REGISTER_KERNEL(
+    bmm_grad, XPU, ALL_LAYOUT, phi::BmmGradKernel, float, phi::dtype::float16) {
+}
diff --git a/paddle/phi/kernels/xpu/bmm_kernel.cc b/paddle/phi/kernels/xpu/bmm_kernel.cc
index b75383bbaa5aa15522d29e28f6ad2abd64c94e02..c927b7e30758898d53730e289983a0f14b317256 100644
--- a/paddle/phi/kernels/xpu/bmm_kernel.cc
+++ b/paddle/phi/kernels/xpu/bmm_kernel.cc
@@ -62,7 +62,7 @@ void BmmKernel(const Context& dev_ctx,
           y_dims[1]));
 
   xpu::Context* xpu_ctx = dev_ctx.x_context();
-  if (std::is_same<paddle::platform::float16, T>::value) {
+  if (std::is_same<phi::dtype::float16, T>::value) {
     MatMulXPUFunction<T, int16_t>(x, y, out, trans_x, trans_y, xpu_ctx);
   } else {
     if (std::getenv("XPU_PADDLE_FC_INT32") != nullptr) {
@@ -77,4 +77,4 @@ void BmmKernel(const Context& dev_ctx,
 }  // namespace phi
 
 PD_REGISTER_KERNEL(
-    bmm, XPU, ALL_LAYOUT, phi::BmmKernel, float, paddle::platform::float16) {}
+    bmm, XPU, ALL_LAYOUT, phi::BmmKernel, float, phi::dtype::float16) {}
diff --git a/paddle/phi/kernels/xpu/xpu_api_wrapper.h b/paddle/phi/kernels/xpu/xpu_api_wrapper.h
index 8fefbd84c65d35b31fc7339a28ae30abf23825b2..8433c6b421eed026a3db3c6e7f2afa04d788737b 100644
--- a/paddle/phi/kernels/xpu/xpu_api_wrapper.h
+++ b/paddle/phi/kernels/xpu/xpu_api_wrapper.h
@@ -34,7 +34,7 @@ enum XPUFCCalcType {
 
 template <typename T>
 XPUFCCalcType FCCalcType() {
-  if (std::is_same<paddle::platform::float16, T>::value ||
+  if (std::is_same<phi::dtype::float16, T>::value ||
       std::is_same<float16, T>::value) {
     return XPUFCCalcType::FC_INT16;
   } else if (std::getenv("XPU_PADDLE_FC_INT32") != nullptr) {