diff --git a/paddle/fluid/operators/beam_search_decode_op_xpu.cc b/paddle/fluid/operators/beam_search_decode_op_xpu.cc
index c438070ce07f9d..f2bfd78dee74ab 100644
--- a/paddle/fluid/operators/beam_search_decode_op_xpu.cc
+++ b/paddle/fluid/operators/beam_search_decode_op_xpu.cc
@@ -111,7 +111,6 @@ class BeamSearchDecodeXPUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(beam_search_decode,
                           XPU,
diff --git a/paddle/fluid/operators/collective/alltoall_op.cc b/paddle/fluid/operators/collective/alltoall_op.cc
index bd99fdde2f2c24..8e7ff730931791 100644
--- a/paddle/fluid/operators/collective/alltoall_op.cc
+++ b/paddle/fluid/operators/collective/alltoall_op.cc
@@ -65,7 +65,6 @@ Scatter tensors from all participators to all participators.
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(alltoall,
                              ops::AllToAllBaseOp,
diff --git a/paddle/fluid/operators/collective/alltoall_op.cu.cc b/paddle/fluid/operators/collective/alltoall_op.cu.cc
index 93a44776851d4b..deee51a0983b6a 100644
--- a/paddle/fluid/operators/collective/alltoall_op.cu.cc
+++ b/paddle/fluid/operators/collective/alltoall_op.cu.cc
@@ -139,7 +139,6 @@ class AllToAllOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(alltoall,
                           GPU,
@@ -148,7 +147,7 @@ PD_REGISTER_STRUCT_KERNEL(alltoall,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           int64_t,
diff --git a/paddle/fluid/operators/collective/barrier_op.cc b/paddle/fluid/operators/collective/barrier_op.cc
index d73c215566d94e..f2ff97d37287a4 100644
--- a/paddle/fluid/operators/collective/barrier_op.cc
+++ b/paddle/fluid/operators/collective/barrier_op.cc
@@ -41,7 +41,6 @@ Barrier Operator - Barrier among all participators.)DOC");
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(barrier, ops::BarrierOp, ops::BarrierOpMaker);
 
diff --git a/paddle/fluid/operators/collective/barrier_op.cu.cc b/paddle/fluid/operators/collective/barrier_op.cu.cc
index dc6b701afee003..bbc4d146a16ca7 100644
--- a/paddle/fluid/operators/collective/barrier_op.cu.cc
+++ b/paddle/fluid/operators/collective/barrier_op.cu.cc
@@ -92,7 +92,6 @@ class BarrierOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(
     barrier, GPU, ALL_LAYOUT, ops::BarrierOpCUDAKernel, int) {}
diff --git a/paddle/fluid/operators/collective/c_allgather_op.cc b/paddle/fluid/operators/collective/c_allgather_op.cc
index e67a2cccc16e9e..cf39513c722358 100644
--- a/paddle/fluid/operators/collective/c_allgather_op.cc
+++ b/paddle/fluid/operators/collective/c_allgather_op.cc
@@ -68,7 +68,6 @@ reference: https://docs.nvidia.com/deeplearning/sdk/nccl-developer-guide/docs/us
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_allgather,
                              ops::CAllGatherOp,
diff --git a/paddle/fluid/operators/collective/c_allgather_op.cu.cc b/paddle/fluid/operators/collective/c_allgather_op.cu.cc
index 7b57e7af25f9b3..f555d4e560ccb6 100644
--- a/paddle/fluid/operators/collective/c_allgather_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_allgather_op.cu.cc
@@ -122,7 +122,6 @@ class CAllGatherOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_allgather,
                           GPU,
@@ -131,7 +130,7 @@ PD_REGISTER_STRUCT_KERNEL(c_allgather,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           uint8_t,
diff --git a/paddle/fluid/operators/collective/c_allgather_op_xpu.cc b/paddle/fluid/operators/collective/c_allgather_op_xpu.cc
index 48e965894a294d..c877024e6c47c4 100644
--- a/paddle/fluid/operators/collective/c_allgather_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_allgather_op_xpu.cc
@@ -116,7 +116,6 @@ class CAllGatherOpXPUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_allgather,
                           XPU,
diff --git a/paddle/fluid/operators/collective/c_allreduce_avg_op.cc b/paddle/fluid/operators/collective/c_allreduce_avg_op.cc
index 3343406a02b6cf..963ea26321bdbf 100644
--- a/paddle/fluid/operators/collective/c_allreduce_avg_op.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_avg_op.cc
@@ -37,7 +37,6 @@ DECLARE_INPLACE_OP_INFERER(AllreduceAvgInplaceInferer, {"X", "Out"});
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_allreduce_avg,
                              ops::CAllReduceOp,
diff --git a/paddle/fluid/operators/collective/c_allreduce_avg_op.cu.cc b/paddle/fluid/operators/collective/c_allreduce_avg_op.cu.cc
index e859145df8b734..021309c1fd5efd 100644
--- a/paddle/fluid/operators/collective/c_allreduce_avg_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_avg_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_ALLREDUCE_CUDA_KERNEL(CAllReduceAvg, kRedAvg)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_avg,
                           GPU,
@@ -32,4 +31,4 @@ PD_REGISTER_STRUCT_KERNEL(c_allreduce_avg,
                           int,
                           int64_t,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
diff --git a/paddle/fluid/operators/collective/c_allreduce_max_op.cc b/paddle/fluid/operators/collective/c_allreduce_max_op.cc
index d659be0f3d1411..ab174de1cec3c1 100644
--- a/paddle/fluid/operators/collective/c_allreduce_max_op.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_max_op.cc
@@ -41,7 +41,6 @@ DEFINE_C_ALLREDUCE_CPU_KERNEL(CAllReduceMax, kRedMax)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_allreduce_max,
                              ops::CAllReduceOp,
diff --git a/paddle/fluid/operators/collective/c_allreduce_max_op.cu.cc b/paddle/fluid/operators/collective/c_allreduce_max_op.cu.cc
index 012b280a9ab15b..7ea63be0cc44ab 100644
--- a/paddle/fluid/operators/collective/c_allreduce_max_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_max_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_ALLREDUCE_CUDA_KERNEL(CAllReduceMax, kRedMax)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_max,
                           GPU,
@@ -29,7 +28,7 @@ PD_REGISTER_STRUCT_KERNEL(c_allreduce_max,
                           ops::CAllReduceMaxCUDAKernel,
                           float,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           double,
                           int,
diff --git a/paddle/fluid/operators/collective/c_allreduce_max_op_xpu.cc b/paddle/fluid/operators/collective/c_allreduce_max_op_xpu.cc
index 943df02ad93e2c..92a9e8c07cbb13 100644
--- a/paddle/fluid/operators/collective/c_allreduce_max_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_max_op_xpu.cc
@@ -21,7 +21,7 @@ DEFINE_C_ALLREDUCE_XPU_KERNEL(CAllReduceMax, kRedMax)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_max,
                           XPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/collective/c_allreduce_min_op.cc b/paddle/fluid/operators/collective/c_allreduce_min_op.cc
index 2a9dd023cf1622..10dc4b9506b2c6 100644
--- a/paddle/fluid/operators/collective/c_allreduce_min_op.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_min_op.cc
@@ -41,7 +41,6 @@ DEFINE_C_ALLREDUCE_CPU_KERNEL(CAllReduceMin, kRedMin)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_allreduce_min,
                              ops::CAllReduceOp,
diff --git a/paddle/fluid/operators/collective/c_allreduce_min_op.cu.cc b/paddle/fluid/operators/collective/c_allreduce_min_op.cu.cc
index a3eec10051c52a..12dd874e552f37 100644
--- a/paddle/fluid/operators/collective/c_allreduce_min_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_min_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_ALLREDUCE_CUDA_KERNEL(CAllReduceMin, kRedMin)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_min,
                           GPU,
diff --git a/paddle/fluid/operators/collective/c_allreduce_min_op_xpu.cc b/paddle/fluid/operators/collective/c_allreduce_min_op_xpu.cc
index fb19a2924d1ebf..7033fb3af90ded 100644
--- a/paddle/fluid/operators/collective/c_allreduce_min_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_min_op_xpu.cc
@@ -21,7 +21,7 @@ DEFINE_C_ALLREDUCE_XPU_KERNEL(CAllReduceMin, kRedMin)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_min,
                           XPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/collective/c_allreduce_prod_op.cc b/paddle/fluid/operators/collective/c_allreduce_prod_op.cc
index 181b78b545e7cb..b9bcc0174b03fc 100644
--- a/paddle/fluid/operators/collective/c_allreduce_prod_op.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_prod_op.cc
@@ -41,7 +41,6 @@ DEFINE_C_ALLREDUCE_CPU_KERNEL(CAllReduceProd, kRedProd)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_allreduce_prod,
                              ops::CAllReduceOp,
diff --git a/paddle/fluid/operators/collective/c_allreduce_prod_op.cu.cc b/paddle/fluid/operators/collective/c_allreduce_prod_op.cu.cc
index e2c0a71a9ced42..21898cf970853d 100644
--- a/paddle/fluid/operators/collective/c_allreduce_prod_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_prod_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_ALLREDUCE_CUDA_KERNEL(CAllReduceProd, kRedProd)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_prod,
                           GPU,
diff --git a/paddle/fluid/operators/collective/c_allreduce_prod_op_xpu.cc b/paddle/fluid/operators/collective/c_allreduce_prod_op_xpu.cc
index d3696c2c5dfc19..ab27fd23b438d6 100644
--- a/paddle/fluid/operators/collective/c_allreduce_prod_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_prod_op_xpu.cc
@@ -21,7 +21,7 @@ DEFINE_C_ALLREDUCE_XPU_KERNEL(CAllReduceProd, kRedProd)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_prod,
                           XPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/collective/c_allreduce_sum_op.cc b/paddle/fluid/operators/collective/c_allreduce_sum_op.cc
index 80b97b2bc70cb8..0c9dc47feb2410 100644
--- a/paddle/fluid/operators/collective/c_allreduce_sum_op.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_sum_op.cc
@@ -62,7 +62,6 @@ DEFINE_C_ALLREDUCE_CPU_KERNEL(CAllReduceSum, kRedSum)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_allreduce_sum,
                              ops::CAllReduceOp,
diff --git a/paddle/fluid/operators/collective/c_allreduce_sum_op.cu.cc b/paddle/fluid/operators/collective/c_allreduce_sum_op.cu.cc
index 909bd23db2413d..1eed03f033ca8b 100644
--- a/paddle/fluid/operators/collective/c_allreduce_sum_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_sum_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_ALLREDUCE_CUDA_KERNEL(CAllReduceSum, kRedSum)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_sum,
                           GPU,
@@ -29,7 +28,7 @@ PD_REGISTER_STRUCT_KERNEL(c_allreduce_sum,
                           ops::CAllReduceSumCUDAKernel,
                           float,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           double,
                           int,
diff --git a/paddle/fluid/operators/collective/c_allreduce_sum_op_xpu.cc b/paddle/fluid/operators/collective/c_allreduce_sum_op_xpu.cc
index 21bedcff8774b4..282694a7c3e8c5 100644
--- a/paddle/fluid/operators/collective/c_allreduce_sum_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_allreduce_sum_op_xpu.cc
@@ -21,7 +21,7 @@ DEFINE_C_ALLREDUCE_XPU_KERNEL(CAllReduceSum, kRedSum)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(c_allreduce_sum,
                           XPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/collective/c_broadcast_op.cc b/paddle/fluid/operators/collective/c_broadcast_op.cc
index 27f3a1bcdc29f6..589fd51904e392 100644
--- a/paddle/fluid/operators/collective/c_broadcast_op.cc
+++ b/paddle/fluid/operators/collective/c_broadcast_op.cc
@@ -59,7 +59,6 @@ Reference: https://docs.nvidia.com/deeplearning/sdk/nccl-developer-guide/docs/us
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_broadcast,
                              ops::CBroadcastOp,
diff --git a/paddle/fluid/operators/collective/c_broadcast_op.cu.cc b/paddle/fluid/operators/collective/c_broadcast_op.cu.cc
index 98f9102f2d8f0b..d9ad9b0c252410 100644
--- a/paddle/fluid/operators/collective/c_broadcast_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_broadcast_op.cu.cc
@@ -90,7 +90,6 @@ class CBroadcastOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_broadcast,
                           GPU,
@@ -101,7 +100,7 @@ PD_REGISTER_STRUCT_KERNEL(c_broadcast,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           phi::dtype::float16) {
 }
diff --git a/paddle/fluid/operators/collective/c_broadcast_op_xpu.cc b/paddle/fluid/operators/collective/c_broadcast_op_xpu.cc
index ac7d9623e32415..84c3d2900c834f 100644
--- a/paddle/fluid/operators/collective/c_broadcast_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_broadcast_op_xpu.cc
@@ -129,7 +129,6 @@ class CBroadcastOpXPUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_broadcast,
                           XPU,
diff --git a/paddle/fluid/operators/collective/c_concat_op.cc b/paddle/fluid/operators/collective/c_concat_op.cc
index 75db7e9fad4279..0600a2a138884c 100644
--- a/paddle/fluid/operators/collective/c_concat_op.cc
+++ b/paddle/fluid/operators/collective/c_concat_op.cc
@@ -105,7 +105,6 @@ AllGather the tensors on different trainers and concat them along the last dimen
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OPERATOR(c_concat,
                   ops::CConcatOp,
diff --git a/paddle/fluid/operators/collective/c_concat_op.cu.cc b/paddle/fluid/operators/collective/c_concat_op.cu.cc
index 9ed68c7c6809bd..b3593cfa4706cc 100644
--- a/paddle/fluid/operators/collective/c_concat_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_concat_op.cu.cc
@@ -165,7 +165,6 @@ class CConcatOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_concat,
                           GPU,
@@ -176,7 +175,7 @@ PD_REGISTER_STRUCT_KERNEL(c_concat,
                           int,
                           int64_t,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           phi::dtype::float16) {
 }
diff --git a/paddle/fluid/operators/collective/c_concat_op_xpu.cc b/paddle/fluid/operators/collective/c_concat_op_xpu.cc
index fcd3c8b33f8b97..25db70f4002cd6 100644
--- a/paddle/fluid/operators/collective/c_concat_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_concat_op_xpu.cc
@@ -160,7 +160,6 @@ class CConcatOpXPUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_concat,
                           XPU,
@@ -169,5 +168,5 @@ PD_REGISTER_STRUCT_KERNEL(c_concat,
                           float,
                           int,
                           int64_t,
-                          plat::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::float16,
+                          phi::dtype::bfloat16) {}
diff --git a/paddle/fluid/operators/collective/c_embedding_op.cc b/paddle/fluid/operators/collective/c_embedding_op.cc
index 0bbd64abb10d59..d51358b2269e23 100644
--- a/paddle/fluid/operators/collective/c_embedding_op.cc
+++ b/paddle/fluid/operators/collective/c_embedding_op.cc
@@ -176,7 +176,7 @@ class CEmbeddingOpGradVarTypeInference : public framework::VarTypeInference {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 REGISTER_OPERATOR(c_embedding,
                   ops::CEmbeddingOp,
                   ops::CEmbeddingOpMaker,
diff --git a/paddle/fluid/operators/collective/c_reduce_avg_op.cc b/paddle/fluid/operators/collective/c_reduce_avg_op.cc
index 53ce6e221a9f8c..8c38d9efebf363 100644
--- a/paddle/fluid/operators/collective/c_reduce_avg_op.cc
+++ b/paddle/fluid/operators/collective/c_reduce_avg_op.cc
@@ -37,7 +37,6 @@ class CReduceAvgOpMaker : public CReduceOpMaker {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_reduce_avg,
                              ops::CReduceOp,
diff --git a/paddle/fluid/operators/collective/c_reduce_avg_op.cu.cc b/paddle/fluid/operators/collective/c_reduce_avg_op.cu.cc
index 1dcd5a2c6489c4..cc4ffa735527db 100644
--- a/paddle/fluid/operators/collective/c_reduce_avg_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_avg_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_CUDA_KERNEL(CReduceAvg, kRedAvg);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_reduce_avg,
                           GPU,
@@ -32,4 +31,4 @@ PD_REGISTER_STRUCT_KERNEL(c_reduce_avg,
                           int,
                           int64_t,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
diff --git a/paddle/fluid/operators/collective/c_reduce_max_op.cc b/paddle/fluid/operators/collective/c_reduce_max_op.cc
index a0181c9f0e7afc..569b9733aa6a18 100644
--- a/paddle/fluid/operators/collective/c_reduce_max_op.cc
+++ b/paddle/fluid/operators/collective/c_reduce_max_op.cc
@@ -39,7 +39,6 @@ DEFINE_C_REDUCE_CPU_KERNEL(CReduceMax, kRedMax)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_reduce_max,
                              ops::CReduceOp,
diff --git a/paddle/fluid/operators/collective/c_reduce_max_op.cu.cc b/paddle/fluid/operators/collective/c_reduce_max_op.cu.cc
index 24f3dffd0517e9..1679ee828a624b 100644
--- a/paddle/fluid/operators/collective/c_reduce_max_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_max_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_CUDA_KERNEL(CReduceMax, kRedMax);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_reduce_max,
                           GPU,
diff --git a/paddle/fluid/operators/collective/c_reduce_max_op_xpu.cc b/paddle/fluid/operators/collective/c_reduce_max_op_xpu.cc
index 6712a6eb500ee4..74d7cbed3216b7 100644
--- a/paddle/fluid/operators/collective/c_reduce_max_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_max_op_xpu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_XPU_KERNEL(CReduceMax, kRedMax);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(
     c_reduce_max, XPU, ALL_LAYOUT, ops::CReduceMaxXPUKernel, float) {}
diff --git a/paddle/fluid/operators/collective/c_reduce_min_op.cc b/paddle/fluid/operators/collective/c_reduce_min_op.cc
index 621272895fe4c8..cacbc1a66e832a 100644
--- a/paddle/fluid/operators/collective/c_reduce_min_op.cc
+++ b/paddle/fluid/operators/collective/c_reduce_min_op.cc
@@ -38,7 +38,6 @@ DEFINE_C_REDUCE_CPU_KERNEL(CReduceMin, kRedMin)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_reduce_min,
                              ops::CReduceOp,
diff --git a/paddle/fluid/operators/collective/c_reduce_min_op.cu.cc b/paddle/fluid/operators/collective/c_reduce_min_op.cu.cc
index c7d979bd932b61..7fbe143eb44bf8 100644
--- a/paddle/fluid/operators/collective/c_reduce_min_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_min_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_CUDA_KERNEL(CReduceMin, kRedMin);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_reduce_min,
                           GPU,
diff --git a/paddle/fluid/operators/collective/c_reduce_min_op_xpu.cc b/paddle/fluid/operators/collective/c_reduce_min_op_xpu.cc
index 440c2b85acde35..c0605b02aba490 100644
--- a/paddle/fluid/operators/collective/c_reduce_min_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_min_op_xpu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_XPU_KERNEL(CReduceMin, kRedMin);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(
     c_reduce_min, XPU, ALL_LAYOUT, ops::CReduceMinXPUKernel, float) {}
diff --git a/paddle/fluid/operators/collective/c_reduce_prod_op.cc b/paddle/fluid/operators/collective/c_reduce_prod_op.cc
index c34e799f5d8e15..47f55bdaa5b19b 100644
--- a/paddle/fluid/operators/collective/c_reduce_prod_op.cc
+++ b/paddle/fluid/operators/collective/c_reduce_prod_op.cc
@@ -39,7 +39,6 @@ DEFINE_C_REDUCE_CPU_KERNEL(CReduceProd, kRedProd)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_reduce_prod,
                              ops::CReduceOp,
diff --git a/paddle/fluid/operators/collective/c_reduce_prod_op.cu.cc b/paddle/fluid/operators/collective/c_reduce_prod_op.cu.cc
index b8b562031bc4e6..7a36dea1a3c3a5 100644
--- a/paddle/fluid/operators/collective/c_reduce_prod_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_prod_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_CUDA_KERNEL(CReduceProd, kRedProd);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_reduce_prod,
                           GPU,
diff --git a/paddle/fluid/operators/collective/c_reduce_prod_op_xpu.cc b/paddle/fluid/operators/collective/c_reduce_prod_op_xpu.cc
index 1541918396d077..bc90ef95f2edf4 100644
--- a/paddle/fluid/operators/collective/c_reduce_prod_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_prod_op_xpu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_XPU_KERNEL(CReduceProd, kRedProd);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(
     c_reduce_prod, XPU, ALL_LAYOUT, ops::CReduceProdXPUKernel, float) {}
diff --git a/paddle/fluid/operators/collective/c_reduce_sum_op.cc b/paddle/fluid/operators/collective/c_reduce_sum_op.cc
index 5bf5c1c2f8b9f5..ccc73d50d258ed 100644
--- a/paddle/fluid/operators/collective/c_reduce_sum_op.cc
+++ b/paddle/fluid/operators/collective/c_reduce_sum_op.cc
@@ -39,7 +39,6 @@ DEFINE_C_REDUCE_CPU_KERNEL(CReduceSum, kRedSum)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_reduce_sum,
                              ops::CReduceOp,
diff --git a/paddle/fluid/operators/collective/c_reduce_sum_op.cu.cc b/paddle/fluid/operators/collective/c_reduce_sum_op.cu.cc
index 56fd0e12933895..39fcda979355f7 100644
--- a/paddle/fluid/operators/collective/c_reduce_sum_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_sum_op.cu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_CUDA_KERNEL(CReduceSum, kRedSum);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_reduce_sum,
                           GPU,
@@ -32,4 +31,4 @@ PD_REGISTER_STRUCT_KERNEL(c_reduce_sum,
                           int,
                           int64_t,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
diff --git a/paddle/fluid/operators/collective/c_reduce_sum_op_xpu.cc b/paddle/fluid/operators/collective/c_reduce_sum_op_xpu.cc
index 230dca3503538e..c7c722e557adb6 100644
--- a/paddle/fluid/operators/collective/c_reduce_sum_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_reduce_sum_op_xpu.cc
@@ -21,7 +21,6 @@ DEFINE_C_REDUCE_XPU_KERNEL(CReduceSum, kRedSum);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(
     c_reduce_sum, XPU, ALL_LAYOUT, ops::CReduceSumXPUKernel, float) {}
diff --git a/paddle/fluid/operators/collective/c_reducescatter_op.cc b/paddle/fluid/operators/collective/c_reducescatter_op.cc
index 7726c3bf5ca41d..47f87bcf4b7f70 100644
--- a/paddle/fluid/operators/collective/c_reducescatter_op.cc
+++ b/paddle/fluid/operators/collective/c_reducescatter_op.cc
@@ -67,7 +67,6 @@ Reference: https://docs.nvidia.com/deeplearning/sdk/nccl-developer-guide/docs/us
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_reducescatter,
                              ops::CReduceScatterOp,
diff --git a/paddle/fluid/operators/collective/c_reducescatter_op.cu.cc b/paddle/fluid/operators/collective/c_reducescatter_op.cu.cc
index e00433ad7b4d60..ecbc5ce5f9a738 100644
--- a/paddle/fluid/operators/collective/c_reducescatter_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_reducescatter_op.cu.cc
@@ -127,7 +127,6 @@ class CReduceScatterOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_reducescatter,
                           GPU,
@@ -136,7 +135,7 @@ PD_REGISTER_STRUCT_KERNEL(c_reducescatter,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           int64_t,
diff --git a/paddle/fluid/operators/collective/c_scatter_op.cc b/paddle/fluid/operators/collective/c_scatter_op.cc
index d3caf134850360..0991e4a8ae8dd8 100644
--- a/paddle/fluid/operators/collective/c_scatter_op.cc
+++ b/paddle/fluid/operators/collective/c_scatter_op.cc
@@ -84,7 +84,6 @@ Scatter the source to all participators.
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_scatter, ops::CScatterOp, ops::CScatterOpMaker);
 
diff --git a/paddle/fluid/operators/collective/c_scatter_op.cu.cc b/paddle/fluid/operators/collective/c_scatter_op.cu.cc
index 7cfe5b6785b5a0..9507ed05a72048 100644
--- a/paddle/fluid/operators/collective/c_scatter_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_scatter_op.cu.cc
@@ -167,7 +167,6 @@ class CScatterOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_scatter,
                           GPU,
diff --git a/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.cc b/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.cc
index 496733759adb38..e1b85f95a2b18b 100644
--- a/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.cc
+++ b/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op.cc
@@ -189,7 +189,6 @@ DECLARE_INPLACE_OP_INFERER(CSoftmaxWithCrossEntropyGradInplaceInferer,
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OPERATOR(
     c_softmax_with_cross_entropy,
diff --git a/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op_xpu.cc b/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op_xpu.cc
index 65329ccd8b2695..bc0f01fcc1d785 100644
--- a/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_softmax_with_cross_entropy_op_xpu.cc
@@ -577,7 +577,6 @@ class CSoftmaxWithCrossEntropyGrad : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_softmax_with_cross_entropy,
                           XPU,
diff --git a/paddle/fluid/operators/collective/c_split_op.cc b/paddle/fluid/operators/collective/c_split_op.cc
index f684c6fe35cf99..a0cb920ac403d1 100644
--- a/paddle/fluid/operators/collective/c_split_op.cc
+++ b/paddle/fluid/operators/collective/c_split_op.cc
@@ -113,7 +113,6 @@ Split the tensor evenly according to its rank.
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OPERATOR(c_split,
                   ops::CSplitOp,
diff --git a/paddle/fluid/operators/collective/c_sync_calc_stream_op.cc b/paddle/fluid/operators/collective/c_sync_calc_stream_op.cc
index 3a75775e7a98f1..a0aceea268a816 100644
--- a/paddle/fluid/operators/collective/c_sync_calc_stream_op.cc
+++ b/paddle/fluid/operators/collective/c_sync_calc_stream_op.cc
@@ -32,7 +32,6 @@ Call calculation stream synchronization.
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_sync_calc_stream,
                              ops::CSyncCalcStreamOp,
diff --git a/paddle/fluid/operators/collective/c_sync_calc_stream_op.cu.cc b/paddle/fluid/operators/collective/c_sync_calc_stream_op.cu.cc
index 8d1134be70de17..cfa23dd4f49d74 100644
--- a/paddle/fluid/operators/collective/c_sync_calc_stream_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_sync_calc_stream_op.cu.cc
@@ -14,7 +14,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/collective/c_sync_calc_stream_op.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_sync_calc_stream,
                           GPU,
@@ -25,4 +24,4 @@ PD_REGISTER_STRUCT_KERNEL(c_sync_calc_stream,
                           int,
                           int64_t,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
diff --git a/paddle/fluid/operators/collective/c_sync_calc_stream_op_xpu.cc b/paddle/fluid/operators/collective/c_sync_calc_stream_op_xpu.cc
index 3053a415524904..6fb9e5ac7f8dd0 100644
--- a/paddle/fluid/operators/collective/c_sync_calc_stream_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_sync_calc_stream_op_xpu.cc
@@ -15,7 +15,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/collective/c_sync_calc_stream_op.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_sync_calc_stream,
                           XPU,
diff --git a/paddle/fluid/operators/collective/c_sync_comm_stream_op.cc b/paddle/fluid/operators/collective/c_sync_comm_stream_op.cc
index 935de19b948dcd..5caa4947c15cff 100644
--- a/paddle/fluid/operators/collective/c_sync_comm_stream_op.cc
+++ b/paddle/fluid/operators/collective/c_sync_comm_stream_op.cc
@@ -48,7 +48,6 @@ Call communication stream synchronization.
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(c_sync_comm_stream,
                              ops::CSyncCommStreamOp,
diff --git a/paddle/fluid/operators/collective/c_sync_comm_stream_op.cu.cc b/paddle/fluid/operators/collective/c_sync_comm_stream_op.cu.cc
index 4ae16d8ca62f7b..7f6f962762568c 100644
--- a/paddle/fluid/operators/collective/c_sync_comm_stream_op.cu.cc
+++ b/paddle/fluid/operators/collective/c_sync_comm_stream_op.cu.cc
@@ -14,7 +14,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/collective/c_sync_comm_stream_op.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(
     c_sync_comm_stream, GPU, ALL_LAYOUT, ops::CSyncCommStreamKernel, float) {}
diff --git a/paddle/fluid/operators/collective/c_sync_comm_stream_op_xpu.cc b/paddle/fluid/operators/collective/c_sync_comm_stream_op_xpu.cc
index e42cca6c329999..08887af8dbc2b9 100644
--- a/paddle/fluid/operators/collective/c_sync_comm_stream_op_xpu.cc
+++ b/paddle/fluid/operators/collective/c_sync_comm_stream_op_xpu.cc
@@ -15,7 +15,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/collective/c_sync_comm_stream_op.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(c_sync_comm_stream,
                           XPU,
diff --git a/paddle/fluid/operators/collective/global_gather_op.cc b/paddle/fluid/operators/collective/global_gather_op.cc
index 1b74fc6bde5f77..7d16b8764d2869 100644
--- a/paddle/fluid/operators/collective/global_gather_op.cc
+++ b/paddle/fluid/operators/collective/global_gather_op.cc
@@ -104,7 +104,7 @@ class GlobalGatherOpGradMaker : public framework::SingleGradOpMaker<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 REGISTER_OPERATOR(global_gather,
                   ops::GlobalGatherOp,
                   ops::GlobalGatherOpMaker,
diff --git a/paddle/fluid/operators/collective/global_gather_op.cu.cc b/paddle/fluid/operators/collective/global_gather_op.cu.cc
index 8c0285cba049d0..5f3c8095fb22ef 100644
--- a/paddle/fluid/operators/collective/global_gather_op.cu.cc
+++ b/paddle/fluid/operators/collective/global_gather_op.cu.cc
@@ -337,7 +337,6 @@ class GlobalGatherOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(global_gather,
                           GPU,
diff --git a/paddle/fluid/operators/collective/global_scatter_op.cc b/paddle/fluid/operators/collective/global_scatter_op.cc
index e6b1bb8295bde8..4efea416fc504b 100644
--- a/paddle/fluid/operators/collective/global_scatter_op.cc
+++ b/paddle/fluid/operators/collective/global_scatter_op.cc
@@ -108,7 +108,7 @@ class GlobalScatterOpGradMaker : public framework::SingleGradOpMaker<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 REGISTER_OPERATOR(global_scatter,
                   ops::GlobalScatterOp,
                   ops::GlobalScatterOpMaker,
diff --git a/paddle/fluid/operators/collective/global_scatter_op.cu.cc b/paddle/fluid/operators/collective/global_scatter_op.cu.cc
index 1eeb23fa602e23..c405f623e1df21 100644
--- a/paddle/fluid/operators/collective/global_scatter_op.cu.cc
+++ b/paddle/fluid/operators/collective/global_scatter_op.cu.cc
@@ -344,7 +344,6 @@ class GlobalScatterOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(global_scatter,
                           GPU,
diff --git a/paddle/fluid/operators/collective/mp_allreduce_sum_op.cc b/paddle/fluid/operators/collective/mp_allreduce_sum_op.cc
index d30d52821e74e1..283826a5a31fc0 100644
--- a/paddle/fluid/operators/collective/mp_allreduce_sum_op.cc
+++ b/paddle/fluid/operators/collective/mp_allreduce_sum_op.cc
@@ -79,7 +79,6 @@ DEFINE_C_ALLREDUCE_CPU_KERNEL(MpAllReduceSum, kRedSum);
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OPERATOR(mp_allreduce_sum,
                   ops::MpAllReduceSumOp,
diff --git a/paddle/fluid/operators/collective/mp_allreduce_sum_op.cu.cc b/paddle/fluid/operators/collective/mp_allreduce_sum_op.cu.cc
index fc856ea04e6f2d..75289a71531b38 100644
--- a/paddle/fluid/operators/collective/mp_allreduce_sum_op.cu.cc
+++ b/paddle/fluid/operators/collective/mp_allreduce_sum_op.cu.cc
@@ -22,7 +22,7 @@ DEFINE_C_ALLREDUCE_CUDA_KERNEL(MpAllReduceSum, kRedSum)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(mp_allreduce_sum,
                           GPU,
                           ALL_LAYOUT,
@@ -32,7 +32,7 @@ PD_REGISTER_STRUCT_KERNEL(mp_allreduce_sum,
                           int,
                           int64_t,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           phi::dtype::float16) {
 }
diff --git a/paddle/fluid/operators/collective/mp_allreduce_sum_op_xpu.cc b/paddle/fluid/operators/collective/mp_allreduce_sum_op_xpu.cc
index 323d39f62092e5..ad2c99858eb120 100644
--- a/paddle/fluid/operators/collective/mp_allreduce_sum_op_xpu.cc
+++ b/paddle/fluid/operators/collective/mp_allreduce_sum_op_xpu.cc
@@ -21,7 +21,7 @@ DEFINE_C_ALLREDUCE_XPU_KERNEL(CAllReduceSum, kRedSum)
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(mp_allreduce_sum,
                           XPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/collective/partial_allgather_op.cc b/paddle/fluid/operators/collective/partial_allgather_op.cc
index 3ae33ecd9eeba2..4dfaf3f37d4b74 100644
--- a/paddle/fluid/operators/collective/partial_allgather_op.cc
+++ b/paddle/fluid/operators/collective/partial_allgather_op.cc
@@ -75,7 +75,6 @@ DECLARE_INPLACE_OP_INFERER(PartialAllGatherOpInplaceInferer, {"X", "Out"});
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OPERATOR(
     partial_allgather,
diff --git a/paddle/fluid/operators/collective/partial_allgather_op.cu.cc b/paddle/fluid/operators/collective/partial_allgather_op.cu.cc
index 2ed198f7ba773d..e33c30152a502e 100644
--- a/paddle/fluid/operators/collective/partial_allgather_op.cu.cc
+++ b/paddle/fluid/operators/collective/partial_allgather_op.cu.cc
@@ -147,7 +147,6 @@ class PartialAllGatherOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(partial_allgather,
                           GPU,
@@ -156,7 +155,7 @@ PD_REGISTER_STRUCT_KERNEL(partial_allgather,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           int64_t,
diff --git a/paddle/fluid/operators/collective/partial_recv_op.cc b/paddle/fluid/operators/collective/partial_recv_op.cc
index 2a512260a792d0..5a8ed36eff6ac3 100644
--- a/paddle/fluid/operators/collective/partial_recv_op.cc
+++ b/paddle/fluid/operators/collective/partial_recv_op.cc
@@ -123,7 +123,6 @@ Reference: https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/usage/p2p.h
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(partial_recv,
                              ops::PartialRecvOp,
diff --git a/paddle/fluid/operators/collective/partial_recv_op.cu.cc b/paddle/fluid/operators/collective/partial_recv_op.cu.cc
index 7e623706b20373..0bffb761511ae5 100644
--- a/paddle/fluid/operators/collective/partial_recv_op.cu.cc
+++ b/paddle/fluid/operators/collective/partial_recv_op.cu.cc
@@ -172,7 +172,6 @@ class PartialRecvOpCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(partial_recv,
                           GPU,
@@ -181,7 +180,7 @@ PD_REGISTER_STRUCT_KERNEL(partial_recv,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           int64_t,
diff --git a/paddle/fluid/operators/collective/partial_send_op.cc b/paddle/fluid/operators/collective/partial_send_op.cc
index 388ece7f4ba12a..cf2a0ece1a7ab6 100644
--- a/paddle/fluid/operators/collective/partial_send_op.cc
+++ b/paddle/fluid/operators/collective/partial_send_op.cc
@@ -88,7 +88,6 @@ Reference: https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/usage/p2p.h
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(partial_send,
                              ops::PartialSendOp,
diff --git a/paddle/fluid/operators/collective/partial_send_op.cu.cc b/paddle/fluid/operators/collective/partial_send_op.cu.cc
index eef547eefa5106..6450d22a097790 100644
--- a/paddle/fluid/operators/collective/partial_send_op.cu.cc
+++ b/paddle/fluid/operators/collective/partial_send_op.cu.cc
@@ -168,7 +168,6 @@ class PartialSendCUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(partial_send,
                           GPU,
@@ -177,7 +176,7 @@ PD_REGISTER_STRUCT_KERNEL(partial_send,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           int64_t,
diff --git a/paddle/fluid/operators/collective/recv_v2_op.cc b/paddle/fluid/operators/collective/recv_v2_op.cc
index 1448aad5f9bfa4..e71037a5c81f16 100644
--- a/paddle/fluid/operators/collective/recv_v2_op.cc
+++ b/paddle/fluid/operators/collective/recv_v2_op.cc
@@ -110,7 +110,6 @@ Reference: https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/usage/p2p.h
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(recv_v2, ops::RecvOpV2, ops::RecvOpV2Maker);
 
diff --git a/paddle/fluid/operators/collective/recv_v2_op.cu.cc b/paddle/fluid/operators/collective/recv_v2_op.cu.cc
index be849d7e6c53b5..f4dbdeca95551a 100644
--- a/paddle/fluid/operators/collective/recv_v2_op.cu.cc
+++ b/paddle/fluid/operators/collective/recv_v2_op.cu.cc
@@ -291,7 +291,6 @@ class RecvOpV2CUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(recv_v2,
                           GPU,
@@ -300,7 +299,7 @@ PD_REGISTER_STRUCT_KERNEL(recv_v2,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           int64_t,
diff --git a/paddle/fluid/operators/collective/send_v2_op.cc b/paddle/fluid/operators/collective/send_v2_op.cc
index c1763a5cd64780..cc41558804d6f1 100644
--- a/paddle/fluid/operators/collective/send_v2_op.cc
+++ b/paddle/fluid/operators/collective/send_v2_op.cc
@@ -82,7 +82,6 @@ Reference: https://docs.nvidia.com/deeplearning/nccl/user-guide/docs/usage/p2p.h
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_WITHOUT_GRADIENT(send_v2, ops::SendOpV2, ops::SendOpV2Maker);
 
diff --git a/paddle/fluid/operators/collective/send_v2_op.cu.cc b/paddle/fluid/operators/collective/send_v2_op.cu.cc
index 6938f413b05487..345783992a5f00 100644
--- a/paddle/fluid/operators/collective/send_v2_op.cu.cc
+++ b/paddle/fluid/operators/collective/send_v2_op.cu.cc
@@ -266,7 +266,6 @@ class SendOpV2CUDAKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(send_v2,
                           GPU,
@@ -275,7 +274,7 @@ PD_REGISTER_STRUCT_KERNEL(send_v2,
                           float,
                           double,
 #if NCCL_VERSION_CODE >= 21000 && CUDA_VERSION >= 11000
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
 #endif
                           int,
                           int64_t,
diff --git a/paddle/fluid/operators/controlflow/fetch_v2_op.cc b/paddle/fluid/operators/controlflow/fetch_v2_op.cc
index 591d3bed324d3c..ae306e7b1b93b9 100644
--- a/paddle/fluid/operators/controlflow/fetch_v2_op.cc
+++ b/paddle/fluid/operators/controlflow/fetch_v2_op.cc
@@ -245,6 +245,6 @@ PD_REGISTER_STRUCT_KERNEL(fetch_v2,
                           uint8_t,
                           bool,
                           phi::dtype::float16,
-                          plat::bfloat16,
-                          plat::complex<float>,
-                          plat::complex<double>) {}
+                          phi::dtype::bfloat16,
+                          phi::dtype::complex<float>,
+                          phi::dtype::complex<double>) {}
diff --git a/paddle/fluid/operators/cross_entropy_op.cu b/paddle/fluid/operators/cross_entropy_op.cu
index e4e2420d152bc9..6dafe597afa47d 100644
--- a/paddle/fluid/operators/cross_entropy_op.cu
+++ b/paddle/fluid/operators/cross_entropy_op.cu
@@ -15,7 +15,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/cross_entropy_op.h"
 #include "paddle/phi/common/float16.h"
 
-namespace plat = paddle::platform;
 namespace ops = paddle::operators;
 
 PD_REGISTER_STRUCT_KERNEL(cross_entropy,
diff --git a/paddle/fluid/operators/flatten_op_xpu.cc b/paddle/fluid/operators/flatten_op_xpu.cc
index ec54a8f815ab42..df09294156cee2 100644
--- a/paddle/fluid/operators/flatten_op_xpu.cc
+++ b/paddle/fluid/operators/flatten_op_xpu.cc
@@ -17,7 +17,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/flatten_op.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_XPU_KERNEL(
     flatten2,
diff --git a/paddle/fluid/operators/fused/fused_elemwise_activation_op.cu b/paddle/fluid/operators/fused/fused_elemwise_activation_op.cu
index e712b78c426697..d231bbff9b93b7 100644
--- a/paddle/fluid/operators/fused/fused_elemwise_activation_op.cu
+++ b/paddle/fluid/operators/fused/fused_elemwise_activation_op.cu
@@ -15,7 +15,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/fused/fused_elemwise_activation_op.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 PD_REGISTER_STRUCT_KERNEL(fused_elemwise_activation,
                           GPU,
diff --git a/paddle/fluid/operators/fused/fused_gate_attention_op.cu b/paddle/fluid/operators/fused/fused_gate_attention_op.cu
index 78202f70bcffbb..cf25fa843f8df3 100644
--- a/paddle/fluid/operators/fused/fused_gate_attention_op.cu
+++ b/paddle/fluid/operators/fused/fused_gate_attention_op.cu
@@ -616,7 +616,7 @@ class FusedGateAttentionGradKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 #ifdef PADDLE_WITH_HIP
 PD_REGISTER_STRUCT_KERNEL(fused_gate_attention,
                           GPU,
@@ -624,14 +624,14 @@ PD_REGISTER_STRUCT_KERNEL(fused_gate_attention,
                           ops::FusedGateAttentionOpKernel,
                           float,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
 PD_REGISTER_STRUCT_KERNEL(fused_gate_attention_grad,
                           GPU,
                           ALL_LAYOUT,
                           ops::FusedGateAttentionGradKernel,
                           float,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
 #else
 PD_REGISTER_STRUCT_KERNEL(fused_gate_attention,
                           GPU,
@@ -640,7 +640,7 @@ PD_REGISTER_STRUCT_KERNEL(fused_gate_attention,
                           float,
                           double,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
 PD_REGISTER_STRUCT_KERNEL(fused_gate_attention_grad,
                           GPU,
                           ALL_LAYOUT,
@@ -648,5 +648,5 @@ PD_REGISTER_STRUCT_KERNEL(fused_gate_attention_grad,
                           float,
                           double,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
 #endif
diff --git a/paddle/fluid/operators/fused/fused_multi_transformer_int8_op.cu b/paddle/fluid/operators/fused/fused_multi_transformer_int8_op.cu
index 5893024c0e958d..b696a183170c33 100644
--- a/paddle/fluid/operators/fused/fused_multi_transformer_int8_op.cu
+++ b/paddle/fluid/operators/fused/fused_multi_transformer_int8_op.cu
@@ -662,7 +662,7 @@ class FusedMultiTransformerINT8OpKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(fused_multi_transformer_int8,
                           GPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/fused/resnet_basic_block_op_xpu.cc b/paddle/fluid/operators/fused/resnet_basic_block_op_xpu.cc
index 50a3b3c46137d1..a674ef722c2da2 100644
--- a/paddle/fluid/operators/fused/resnet_basic_block_op_xpu.cc
+++ b/paddle/fluid/operators/fused/resnet_basic_block_op_xpu.cc
@@ -993,7 +993,7 @@ class ResNetBasicBlockGradXPUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(resnet_basic_block,
                           XPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/fused/resnet_unit_op.cu b/paddle/fluid/operators/fused/resnet_unit_op.cu
index 6afe03a67ceab1..2955fd3b453b4d 100644
--- a/paddle/fluid/operators/fused/resnet_unit_op.cu
+++ b/paddle/fluid/operators/fused/resnet_unit_op.cu
@@ -418,7 +418,7 @@ class ResNetUnitGradKernel : public framework::OpKernel<T> {
 
 #if CUDNN_VERSION >= 8000
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(
     resnet_unit, GPU, ALL_LAYOUT, ops::ResNetUnitKernel, phi::dtype::float16) {}
 PD_REGISTER_STRUCT_KERNEL(resnet_unit_grad,
diff --git a/paddle/fluid/operators/fused/resnet_unit_op_xpu.cc b/paddle/fluid/operators/fused/resnet_unit_op_xpu.cc
index f50d452d6c2855..91de3c067a0c70 100644
--- a/paddle/fluid/operators/fused/resnet_unit_op_xpu.cc
+++ b/paddle/fluid/operators/fused/resnet_unit_op_xpu.cc
@@ -358,7 +358,7 @@ class ResNetUnitGradXPUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(resnet_unit,
                           XPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/isfinite_op.cu b/paddle/fluid/operators/isfinite_op.cu
index 71aaa66a5ad0d9..2ea6bbb45b9a85 100755
--- a/paddle/fluid/operators/isfinite_op.cu
+++ b/paddle/fluid/operators/isfinite_op.cu
@@ -26,7 +26,9 @@ REGISTER_OP_CUDA_KERNEL(
     ops::OverflowKernel<phi::GPUContext,
                         phi::dtype::float16,
                         ops::InfinityFunctor>,
-    ops::OverflowKernel<phi::GPUContext, plat::bfloat16, ops::InfinityFunctor>);
+    ops::OverflowKernel<phi::GPUContext,
+                        phi::dtype::bfloat16,
+                        ops::InfinityFunctor>);
 
 REGISTER_OP_CUDA_KERNEL(
     isnan,
@@ -34,4 +36,5 @@ REGISTER_OP_CUDA_KERNEL(
     ops::OverflowKernel<phi::GPUContext, float, ops::NANFunctor>,
     ops::OverflowKernel<phi::GPUContext, double, ops::NANFunctor>,
     ops::OverflowKernel<phi::GPUContext, phi::dtype::float16, ops::NANFunctor>,
-    ops::OverflowKernel<phi::GPUContext, plat::bfloat16, ops::NANFunctor>);
+    ops::
+        OverflowKernel<phi::GPUContext, phi::dtype::bfloat16, ops::NANFunctor>);
diff --git a/paddle/fluid/operators/load_combine_op.cc b/paddle/fluid/operators/load_combine_op.cc
index 3f5b85ecc434ba..fe63f19166a10a 100644
--- a/paddle/fluid/operators/load_combine_op.cc
+++ b/paddle/fluid/operators/load_combine_op.cc
@@ -89,7 +89,7 @@ PD_REGISTER_STRUCT_KERNEL(load_combine,
                           ops::LoadCombineOpKernel,
                           float,
                           double,
-                          plat::bfloat16,
+                          phi::dtype::bfloat16,
                           int,
                           int8_t,
                           int64_t) {}
diff --git a/paddle/fluid/operators/lod_reset_op.cc b/paddle/fluid/operators/lod_reset_op.cc
index 654bc669c7504c..21c5bba66e3e06 100644
--- a/paddle/fluid/operators/lod_reset_op.cc
+++ b/paddle/fluid/operators/lod_reset_op.cc
@@ -235,7 +235,6 @@ DECLARE_NO_NEED_BUFFER_VARS_INFERER(LoDResetGradNoNeedBufferVarInferer, "X");
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 REGISTER_OPERATOR(lod_reset,
                   ops::LoDResetOp,
                   ops::LoDResetOpMaker,
diff --git a/paddle/fluid/operators/matmul_op_xpu.cc b/paddle/fluid/operators/matmul_op_xpu.cc
index 095a90737f9ad3..ee7327705e07ad 100644
--- a/paddle/fluid/operators/matmul_op_xpu.cc
+++ b/paddle/fluid/operators/matmul_op_xpu.cc
@@ -150,19 +150,19 @@ class MatMulGradXPUKernel : public framework::OpKernel<T> {
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OP_XPU_KERNEL(
     matmul,
     ops::MatMulXPUKernel<paddle::platform::XPUDeviceContext, float>,
-    ops::MatMulXPUKernel<paddle::platform::XPUDeviceContext, plat::bfloat16>,
+    ops::MatMulXPUKernel<paddle::platform::XPUDeviceContext,
+                         phi::dtype::bfloat16>,
     ops::MatMulXPUKernel<paddle::platform::XPUDeviceContext,
                          phi::dtype::float16>);
 REGISTER_OP_XPU_KERNEL(
     matmul_grad,
     ops::MatMulGradXPUKernel<paddle::platform::XPUDeviceContext, float>,
     ops::MatMulGradXPUKernel<paddle::platform::XPUDeviceContext,
-                             plat::bfloat16>,
+                             phi::dtype::bfloat16>,
     ops::MatMulGradXPUKernel<paddle::platform::XPUDeviceContext,
                              phi::dtype::float16>);
 #endif
diff --git a/paddle/fluid/operators/optimizers/sparse_momentum_op.cu b/paddle/fluid/operators/optimizers/sparse_momentum_op.cu
index 0a98ee4b3e5de9..2c8bf1ac616456 100644
--- a/paddle/fluid/operators/optimizers/sparse_momentum_op.cu
+++ b/paddle/fluid/operators/optimizers/sparse_momentum_op.cu
@@ -17,7 +17,7 @@
 #include "paddle/phi/common/float16.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(sparse_momentum,
                           GPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/partial_concat_op.cu b/paddle/fluid/operators/partial_concat_op.cu
index a597cb11f08ff3..5778b5c8e2d9e8 100644
--- a/paddle/fluid/operators/partial_concat_op.cu
+++ b/paddle/fluid/operators/partial_concat_op.cu
@@ -16,8 +16,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/partial_concat_op.h"
 #include "paddle/phi/common/float16.h"
 
-namespace plat = paddle::platform;
-
 namespace paddle {
 namespace operators {
 
diff --git a/paddle/fluid/operators/partial_sum_op.cu b/paddle/fluid/operators/partial_sum_op.cu
index 25758cfde4870b..ec8945cccad892 100644
--- a/paddle/fluid/operators/partial_sum_op.cu
+++ b/paddle/fluid/operators/partial_sum_op.cu
@@ -16,8 +16,6 @@ limitations under the License. */
 #include "paddle/fluid/operators/partial_sum_op.h"
 #include "paddle/phi/common/float16.h"
 
-namespace plat = paddle::platform;
-
 namespace paddle {
 namespace operators {
 
diff --git a/paddle/fluid/operators/reshape_op.cc b/paddle/fluid/operators/reshape_op.cc
index d984edc4c41724..677385f4698f34 100644
--- a/paddle/fluid/operators/reshape_op.cc
+++ b/paddle/fluid/operators/reshape_op.cc
@@ -700,7 +700,6 @@ DECLARE_NO_NEED_BUFFER_VARS_INFERER(ReshapeDoubleGradOpNoNeedBufferVarInferer,
 }  // namespace operators
 }  // namespace paddle
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
 
 REGISTER_OPERATOR(
     reshape,
@@ -776,7 +775,7 @@ REGISTER_OP_CUDA_KERNEL_FUNCTOR(reshape,
                                 ops::ReshapeKernel,
                                 phi::dtype::float16,
                                 ops::ReshapeKernel,
-                                plat::bfloat16,
+                                phi::dtype::bfloat16,
                                 ops::ReshapeKernel);
 REGISTER_OP_CUDA_KERNEL_FUNCTOR(reshape_grad,
                                 float,
@@ -793,6 +792,6 @@ REGISTER_OP_CUDA_KERNEL_FUNCTOR(reshape_grad,
                                 ops::ReshapeGradKernel,
                                 phi::dtype::float16,
                                 ops::ReshapeGradKernel,
-                                plat::bfloat16,
+                                phi::dtype::bfloat16,
                                 ops::ReshapeGradKernel);
 #endif
diff --git a/paddle/fluid/operators/share_data_op.cc b/paddle/fluid/operators/share_data_op.cc
index 4accee24e17faa..074ca142c95678 100644
--- a/paddle/fluid/operators/share_data_op.cc
+++ b/paddle/fluid/operators/share_data_op.cc
@@ -62,7 +62,7 @@ Return a tensor $Out$ that shares data with the input tensor $X$ and without ten
 }  // namespace paddle
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 REGISTER_OPERATOR(
     share_data,
     ops::ShareDataOp,
diff --git a/paddle/fluid/operators/share_data_op.cu b/paddle/fluid/operators/share_data_op.cu
index 2b1c32d655b80f..dd369bfdd41e58 100644
--- a/paddle/fluid/operators/share_data_op.cu
+++ b/paddle/fluid/operators/share_data_op.cu
@@ -15,7 +15,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/share_data_op.h"
 
 namespace ops = paddle::operators;
-namespace plat = paddle::platform;
+
 PD_REGISTER_STRUCT_KERNEL(share_data,
                           GPU,
                           ALL_LAYOUT,
diff --git a/paddle/fluid/operators/soft_relu_op.cu b/paddle/fluid/operators/soft_relu_op.cu
index e4273c73530f6c..f3802ec10c0662 100644
--- a/paddle/fluid/operators/soft_relu_op.cu
+++ b/paddle/fluid/operators/soft_relu_op.cu
@@ -40,7 +40,7 @@ PD_REGISTER_STRUCT_KERNEL(soft_relu,
                           float,
                           double,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
 PD_REGISTER_STRUCT_KERNEL(soft_relu_grad,
                           GPU,
                           ALL_LAYOUT,
@@ -48,4 +48,4 @@ PD_REGISTER_STRUCT_KERNEL(soft_relu_grad,
                           float,
                           double,
                           phi::dtype::float16,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}
diff --git a/paddle/fluid/operators/uniform_random_batch_size_like_op.cc b/paddle/fluid/operators/uniform_random_batch_size_like_op.cc
index bcff52e1af6d7d..4a11057de9539b 100644
--- a/paddle/fluid/operators/uniform_random_batch_size_like_op.cc
+++ b/paddle/fluid/operators/uniform_random_batch_size_like_op.cc
@@ -194,4 +194,4 @@ PD_REGISTER_STRUCT_KERNEL(uniform_random_batch_size_like,
                           ops::CPUUniformRandomKernel,
                           float,
                           double,
-                          plat::bfloat16) {}
+                          phi::dtype::bfloat16) {}