[NPU] reorganization for device API abstraction (#37110)

* [NPU] reorganization for device API abstraction

* [NPU] delete old files

* [NPU] fix npu_collective_helper

* [NPU] fix collective_helper

* [NPU] fix ut

* [NPU] mod memory allocation and hccl_helper

* [NPU] fix place_type

* [NPU] split enfoce.h

* move acl* call into npu_info

* merge conflict

* fix merge

* merge conflict

* merge conflict

paddle/fluid/eager/accumulation/gradient_accumulation.cc

@@ -34,7 +34,7 @@
 #include "xpu/refactor/math.h"
 #endif
 #ifdef PADDLE_WITH_ASCEND_CL
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 #endif
 
 namespace egr {

paddle/fluid/framework/details/nan_inf_utils_detail.cc

@@ -17,7 +17,7 @@
 #include "paddle/fluid/framework/op_proto_maker.h"
 
 #ifdef PADDLE_WITH_ASCEND_CL
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 #endif
 
 namespace paddle {

paddle/fluid/imperative/gradient_accumulator.cc

@@ -32,7 +32,7 @@
 #include "xpu/refactor/math.h"
 #endif
 #ifdef PADDLE_WITH_ASCEND_CL
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 #endif
 
 namespace paddle {

paddle/fluid/imperative/hccl_context.cc

@@ -22,7 +22,7 @@
 #include "paddle/fluid/platform/place.h"
 
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 
 namespace paddle {
 namespace framework {

paddle/fluid/imperative/hccl_context.h

@@ -19,8 +19,8 @@
 #include <vector>
 
 #include "paddle/fluid/imperative/parallel_context.h"
-#include "paddle/fluid/platform/dynload/hccl.h"
-#include "paddle/fluid/platform/npu_resource_pool.h"
+#include "paddle/fluid/platform/device/npu/dynload/hccl.h"
+#include "paddle/fluid/platform/device/npu/npu_resource_pool.h"
 
 namespace paddle {
 namespace framework {

paddle/fluid/inference/api/details/zero_copy_tensor.cc

@@ -290,7 +290,7 @@ void Tensor::CopyToCpuImpl(T *data, void *exec_stream, CallbackFunc cb,
     paddle::memory::Copy(paddle::platform::CPUPlace(),
                          static_cast<void *>(data), npu_place, t_data,
                          ele_num * sizeof(T), dev_ctx->stream());
-    aclrtSynchronizeStream(dev_ctx->stream());
+    paddle::platform::NPUStreamSync(dev_ctx->stream());
 #else
     PADDLE_THROW(paddle::platform::errors::Unavailable(
         "Can not create tensor with NPU place because paddle is not compiled "

paddle/fluid/memory/allocation/allocator_facade.cc

@@ -23,7 +23,6 @@
 #include "paddle/fluid/memory/allocation/naive_best_fit_allocator.h"
 #include "paddle/fluid/memory/allocation/retry_allocator.h"
 #include "paddle/fluid/platform/enforce.h"
-#include "paddle/fluid/platform/npu_info.h"
 #include "paddle/fluid/platform/place.h"
 
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)

paddle/fluid/memory/allocation/naive_best_fit_allocator.cc

@@ -22,7 +22,6 @@
 #include "paddle/fluid/memory/detail/system_allocator.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/gpu_info.h"
-#include "paddle/fluid/platform/npu_info.h"
 #include "paddle/fluid/platform/profiler.h"
 
 #include "paddle/fluid/string/printf.h"
@@ -33,6 +32,9 @@
 #ifdef PADDLE_WITH_XPU
 #include "paddle/fluid/platform/device/xpu/xpu_header.h"
 #endif
+#ifdef PADDLE_WITH_ASCEND_CL
+#include "paddle/fluid/platform/device/npu/npu_info.h"
+#endif
 
 PADDLE_DEFINE_EXPORTED_bool(
     init_allocated_mem, false,
@@ -327,8 +329,8 @@ void *Alloc<platform::NPUPlace>(const platform::NPUPlace &place, size_t size) {
     size_t avail, total;
     platform::NPUMemoryUsage(&avail, &total);
     PADDLE_THROW(platform::errors::ResourceExhausted(
-        "Cannot allocate %s in GPU %d, avaliable %s, total %s, GpuMinChunkSize "
-        "%s, GpuMaxChunkSize %s, GPU memory used: %s.",
+        "Cannot allocate %s in NPU %d, avaliable %s, total %s, NpuMinChunkSize "
+        "%s, NpuMaxChunkSize %s, NPU memory used: %s.",
         string::HumanReadableSize(size), place.device,
         string::HumanReadableSize(avail), string::HumanReadableSize(total),
         string::HumanReadableSize(buddy_allocator->GetMinChunkSize()),
@@ -336,7 +338,7 @@ void *Alloc<platform::NPUPlace>(const platform::NPUPlace &place, size_t size) {
         string::HumanReadableSize(Used<platform::NPUPlace>(place))));
   } else {
     if (FLAGS_init_allocated_mem) {
-      aclrtMemset(ptr, size, 0xEF, size);
+      platform::NPUMemsetSync(ptr, 0xEF, size, size);
     }
   }
   VLOG(10) << "Allocate " << size << " bytes on " << platform::Place(place);
@@ -387,8 +389,7 @@ void *Alloc<platform::NPUPinnedPlace>(const platform::NPUPinnedPlace &place,
   void *ptr = buddy_allocator->Alloc(size);
 
   if (ptr == nullptr) {
-    LOG(WARNING) << "aclrtMallocHost Cannot allocate " << size
-                 << " bytes in NPUPinnedPlace";
+    LOG(WARNING) << "Cannot allocate " << size << " bytes in NPUPinnedPlace";
   }
   if (FLAGS_init_allocated_mem) {
     memset(ptr, 0xEF, size);

paddle/fluid/memory/allocation/npu_allocator.cc

@@ -14,8 +14,8 @@
 
 #include "paddle/fluid/memory/allocation/npu_allocator.h"
 #include <string>
+#include "paddle/fluid/platform/device/npu/npu_info.h"
 #include "paddle/fluid/platform/enforce.h"
-#include "paddle/fluid/platform/npu_info.h"
 
 namespace paddle {
 namespace memory {

paddle/fluid/memory/allocation/npu_pinned_allocator.cc

@@ -23,7 +23,7 @@ void NPUPinnedAllocator::ProcessEventsAndFree() {
   for (auto it = npu_events_.begin(); it != npu_events_.end();) {
     aclrtEvent event = it->second;
     aclrtEventStatus status = ACL_EVENT_STATUS_COMPLETE;
-    PADDLE_ENFORCE_NPU_SUCCESS(aclrtQueryEvent(event, &status));
+    platform::NPUEventQuery(event, &status);
 
     if (status == ACL_EVENT_STATUS_COMPLETE) {
       Allocation *allocation = it->first;
@@ -31,7 +31,7 @@ void NPUPinnedAllocator::ProcessEventsAndFree() {
       free(ptr);
       npu_events_.erase(it++);
       delete allocation;
-      PADDLE_ENFORCE_NPU_SUCCESS(aclrtDestroyEvent(event));
+      platform::NPUEventDestroy(event);
     } else {
       ++it;
     }
@@ -67,12 +67,12 @@ void NPUPinnedAllocator::FreeImpl(Allocation *allocation) {
 
   aclrtEvent event = iter->second;
   aclrtEventStatus status = ACL_EVENT_STATUS_COMPLETE;
-  PADDLE_ENFORCE_NPU_SUCCESS(aclrtQueryEvent(event, &status));
+  platform::NPUEventQuery(event, &status);
   if (status == ACL_EVENT_STATUS_COMPLETE) {
     free(ptr);
     npu_events_.erase(allocation);
     delete allocation;
-    PADDLE_ENFORCE_NPU_SUCCESS(aclrtDestroyEvent(event));
+    platform::NPUEventDestroy(event);
   }
   return;
 }
@@ -87,8 +87,8 @@ void NPUPinnedAllocator::RecordEvent(Allocation *allocation,
                                      aclrtStream stream) {
   std::lock_guard<std::mutex> lock(mtx_);
   aclrtEvent event = nullptr;
-  PADDLE_ENFORCE_NPU_SUCCESS(aclrtCreateEvent(&event));
-  PADDLE_ENFORCE_NPU_SUCCESS(aclrtRecordEvent(event, stream));
+  platform::NPUEventCreate(&event);
+  platform::NPUEventRecord(event, stream);
   npu_events_.insert({allocation, event});
 }
 

paddle/fluid/memory/allocation/npu_pinned_allocator.h

@@ -21,7 +21,7 @@
 
 #include "acl/acl.h"
 #include "paddle/fluid/memory/allocation/allocator.h"
-#include "paddle/fluid/platform/npu_info.h"
+#include "paddle/fluid/platform/device/npu/npu_info.h"
 #include "paddle/fluid/platform/place.h"
 
 namespace paddle {

paddle/fluid/memory/detail/buddy_allocator.h

@@ -25,8 +25,8 @@ limitations under the License. */
 #include "paddle/fluid/memory/detail/memory_block.h"
 #include "paddle/fluid/memory/detail/system_allocator.h"
 #include "paddle/fluid/platform/cpu_info.h"
+#include "paddle/fluid/platform/device/npu/npu_info.h"
 #include "paddle/fluid/platform/gpu_info.h"
-#include "paddle/fluid/platform/npu_info.h"
 
 namespace paddle {
 namespace memory {

paddle/fluid/memory/detail/buddy_allocator_test.cc

@@ -24,8 +24,8 @@ limitations under the License. */
 
 #include "gflags/gflags.h"
 #include "gtest/gtest.h"
+#include "paddle/fluid/platform/device/npu/npu_info.h"
 #include "paddle/fluid/platform/gpu_info.h"
-#include "paddle/fluid/platform/npu_info.h"
 
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP) || \
     defined(PADDLE_WITH_ASCEND_CL)

paddle/fluid/memory/detail/system_allocator.cc

@@ -27,9 +27,9 @@ limitations under the License. */
 #include "gflags/gflags.h"
 #include "paddle/fluid/memory/allocation/allocator.h"
 #include "paddle/fluid/platform/cpu_info.h"
+#include "paddle/fluid/platform/device/npu/npu_info.h"
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/gpu_info.h"
-#include "paddle/fluid/platform/npu_info.h"
 
 #if defined(PADDLE_WITH_CUDA) || defined(PADDLE_WITH_HIP)
 #include "paddle/fluid/platform/cuda_device_guard.h"
@@ -326,14 +326,14 @@ void* NPUPinnedAllocator::Alloc(size_t* index, size_t size) {
 
   void* p;
   // PINNED memory is visible to all NPU contexts.
-  auto result = aclrtMallocHost(&p, size);
+  auto result = platform::NPUHostMalloc(&p, size);
 
   if (result == ACL_ERROR_NONE) {
     *index = 1;  // PINNED memory
     npu_pinnd_alloc_size_ += size;
     return p;
   } else {
-    LOG(WARNING) << "aclrtMallocHost failed.";
+    LOG(WARNING) << "NPUHostMalloc failed.";
     return nullptr;
   }
 
@@ -351,14 +351,13 @@ void NPUPinnedAllocator::Free(void* p, size_t size, size_t index) {
                         "allocated npu pinned memory (%d)",
                         size, npu_pinnd_alloc_size_));
   npu_pinnd_alloc_size_ -= size;
-  err = aclrtFreeHost(p);
+  err = platform::NPUHostFree(p);
 
   if (err != ACL_ERROR_NONE) {
     PADDLE_ENFORCE_EQ(
         err, 0,
         platform::errors::Fatal(
-            "aclrtFreeHost failed in NPUPinnedAllocator, error code is %d",
-            err));
+            "NPUHostFree failed in NPUPinnedAllocator, error code is %d", err));
   }
 }
 

paddle/fluid/operators/CMakeLists.txt

@@ -164,7 +164,6 @@ endif()
 
 if (WITH_ASCEND_CL)
   cc_test(assign_op_npu_test SRCS assign_op_npu_test.cc DEPS assign_op)
-  cc_library(npu_op_runner SRCS npu_op_runner.cc DEPS operator npu_info)
   set(COMMON_OP_DEPS ${COMMON_OP_DEPS} npu_op_runner)
 endif()
 

paddle/fluid/operators/abs_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the Licnse. */
 
 #include "paddle/fluid/operators/abs_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/activation_op_npu.cc

@@ -19,7 +19,7 @@ limitations under the Licnse. */
 #include "paddle/fluid/framework/framework.pb.h"
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/activation_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/amp/alloc_float_status_op_npu.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include <cmath>
 #include <vector>
 #include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/amp/check_finite_and_unscale_op_npu.cc

@@ -17,7 +17,7 @@ limitations under the License. */
 
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/amp/check_finite_and_unscale_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/amp/clear_float_status_op_npu.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include <cmath>
 #include <vector>
 #include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/amp/get_float_status_op_npu.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include <cmath>
 #include <vector>
 #include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/amp/update_loss_scaling_op_npu.cc

@@ -17,7 +17,7 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/framework/data_type.h"
 #include "paddle/fluid/framework/op_registry.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 DECLARE_int32(min_loss_scaling);
 

paddle/fluid/operators/arg_max_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the Licnse. */
 
 #include "paddle/fluid/operators/arg_min_max_op_base.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/arg_min_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/arg_min_max_op_base.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/argsort_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/argsort_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/assign_op_npu.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/operators/assign_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 #include "paddle/fluid/platform/float16.h"
 
 namespace paddle {

paddle/fluid/operators/batch_norm_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/batch_norm_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/bce_loss_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/bce_loss_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/cast_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/operators/cast_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/clip_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/clip_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/coalesce_tensor_op.cc

@@ -21,7 +21,7 @@
 #include "paddle/fluid/operators/math/math_function.h"
 #include "paddle/fluid/platform/device_memory_aligment.h"
 #ifdef PADDLE_WITH_ASCEND_CL
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 #endif
 
 namespace paddle {

paddle/fluid/operators/collective/c_allgather_op_npu.cc

@@ -18,7 +18,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {

paddle/fluid/operators/collective/c_allgather_op_npu_test.cc

@@ -38,7 +38,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/c_allreduce_max_op_npu_test.cc

@@ -38,7 +38,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/c_allreduce_op.h

@@ -21,7 +21,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/memory/memcpy.h"
 #include "paddle/fluid/memory/memory.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 #if defined(PADDLE_WITH_NCCL) || defined(PADDLE_WITH_RCCL) || \
     defined(PADDLE_WITH_ASCEND_CL) || defined(PADDLE_WITH_XPU_BKCL)
@@ -42,7 +42,7 @@ limitations under the License. */
 #endif
 
 #if defined(PADDLE_WITH_ASCEND_CL)
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 #if defined(PADDLE_WITH_ASCEND_CL)

paddle/fluid/operators/collective/c_allreduce_sum_op_npu_test.cc

@@ -35,7 +35,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 // Node1: HCCL_WHITELIST_DISABLE=1 FLAGS_selected_npus=1 GLOG_v=4 RANK_ID=1

paddle/fluid/operators/collective/c_broadcast_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {

paddle/fluid/operators/collective/c_broadcast_op_npu_test.cc

@@ -35,7 +35,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/c_comm_init_hccl_op.cc

@@ -22,11 +22,10 @@ class Scope;
 }  // namespace framework
 }  // namespace paddle
 #if defined(PADDLE_WITH_ASCEND_CL)
-#include "acl/acl.h"
 #include "hccl/hccl.h"
 #include "hccl/hccl_types.h"
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {
@@ -69,12 +68,11 @@ class CCommInitOpAscend : public framework::OperatorBase {
     for (int32_t idx = 0; idx < size; idx++) {
       input[idx] = 1.0;
     }
-    PADDLE_ENFORCE_NPU_SUCCESS(aclrtMalloc(reinterpret_cast<void**>(&buff),
-                                           size * sizeof(float),
-                                           ACL_MEM_MALLOC_HUGE_FIRST));
-    PADDLE_ENFORCE_NPU_SUCCESS(aclrtMemcpy(
-        reinterpret_cast<void*>(buff), size * sizeof(float), input.data(),
-        size * sizeof(float), ACL_MEMCPY_HOST_TO_DEVICE));
+    PADDLE_ENFORCE_NPU_SUCCESS(platform::RecordedNPUMalloc(
+        reinterpret_cast<void**>(&buff), size * sizeof(float), device_id));
+    platform::NPUMemcpySync(reinterpret_cast<void*>(buff), input.data(),
+                            size * sizeof(float), ACL_MEMCPY_HOST_TO_DEVICE,
+                            size * sizeof(float));
     VLOG(3) << "Build buff data successful.";
 
     aclrtStream stream = nullptr;
@@ -88,7 +86,7 @@ class CCommInitOpAscend : public framework::OperatorBase {
     PADDLE_ENFORCE_NPU_SUCCESS(platform::dynload::HcclBroadcast(
         buff, size, HCCL_DATA_TYPE_FP32, 0, comm->comm(), stream));
     // Synchronize stream to find hccl error in time.
-    PADDLE_ENFORCE_NPU_SUCCESS(aclrtSynchronizeStream(stream));
+    platform::NPUStreamSync(stream);
     VLOG(3) << "Build connection successful.";
 #else
     PADDLE_THROW(platform::errors::PreconditionNotMet(

paddle/fluid/operators/collective/c_embedding_op_npu.cc

@@ -17,8 +17,8 @@ limitations under the License. */
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/collective/c_embedding_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
-#include "paddle/fluid/platform/npu_info.h"
+#include "paddle/fluid/platform/device/npu/npu_info.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {
@@ -136,11 +136,10 @@ void NPUGetIdsEmbedding(const framework::ExecutionContext &context) {
 
   uint8_t *pad_data = reinterpret_cast<uint8_t *>(
       table_t_pad.mutable_data<T>(pad_shape, context.GetPlace()));
-  PADDLE_ENFORCE_NPU_SUCCESS(
-      aclrtMemcpyAsync(pad_data, mem_size, table_t->data<T>(), mem_size,
-                       ACL_MEMCPY_DEVICE_TO_DEVICE, stream));
-  PADDLE_ENFORCE_NPU_SUCCESS(aclrtMemsetAsync(
-      pad_data + mem_size, line_mem_size, 0, line_mem_size, stream));
+  platform::NPUMemcpyAsync(pad_data, table_t->data<T>(), mem_size,
+                           ACL_MEMCPY_DEVICE_TO_DEVICE, stream, mem_size);
+  platform::NPUMemsetAsync(pad_data + mem_size, 0, line_mem_size, stream,
+                           line_mem_size);
 
   output_t->mutable_data<T>(context.GetPlace());
   NpuOpRunner runner;
@@ -202,8 +201,8 @@ void NPUUpdateEmbedding(const framework::ExecutionContext &context) {
       table_t_pad.mutable_data<T>(pad_shape, context.GetPlace()));
   size_t table_t_pad_mem_size =
       table_t_pad.numel() * framework::SizeOfType(table_t_pad.type());
-  PADDLE_ENFORCE_NPU_SUCCESS(aclrtMemsetAsync(pad_data, table_t_pad_mem_size, 0,
-                                              table_t_pad_mem_size, stream));
+  platform::NPUMemsetAsync(pad_data, 0, table_t_pad_mem_size, stream,
+                           table_t_pad_mem_size);
 
   // NOTE(zhiqiu): It seems in cann 20.1, the first input and output
   // can be different tensor, but in cann 20.2+, it does inplace operation.
@@ -225,8 +224,8 @@ void NPUUpdateEmbedding(const framework::ExecutionContext &context) {
                     platform::errors::InvalidArgument(
                         "NPU only accept the second dim must align by 64"));
 
-  PADDLE_ENFORCE_NPU_SUCCESS(aclrtMemcpyAsync(
-      dst, mem_size, pad_data, mem_size, ACL_MEMCPY_DEVICE_TO_DEVICE, stream));
+  platform::NPUMemcpyAsync(dst, pad_data, mem_size, ACL_MEMCPY_DEVICE_TO_DEVICE,
+                           stream, mem_size);
 }
 
 template <typename T>

paddle/fluid/operators/collective/c_gen_hccl_id_op.cc

@@ -23,7 +23,7 @@ limitations under the License. */
 #include "paddle/fluid/platform/enforce.h"
 #include "paddle/fluid/platform/place.h"
 
-#include "paddle/fluid/platform/dynload/hccl.h"
+#include "paddle/fluid/platform/device/npu/dynload/hccl.h"
 #include "paddle/fluid/platform/gen_comm_id_helper.h"
 
 namespace paddle {

paddle/fluid/operators/collective/c_reduce_op.h

@@ -43,7 +43,7 @@ limitations under the License. */
 #endif
 
 #if defined(PADDLE_WITH_ASCEND_CL)
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {

paddle/fluid/operators/collective/c_reduce_sum_op_npu_test.cc

@@ -35,7 +35,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/c_reducescatter_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {

paddle/fluid/operators/collective/c_reducescatter_op_npu_test.cc

@@ -38,7 +38,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/c_sync_calc_stream_op.cc

@@ -69,7 +69,7 @@ class CSyncCalcStreamKernel : public framework::OpKernel<T> {
 
     auto dev_ctx = static_cast<platform::NPUDeviceContext*>(
         platform::DeviceContextPool::Instance().Get(place));
-    PADDLE_ENFORCE_NPU_SUCCESS(aclrtSynchronizeStream(dev_ctx->stream()));
+    platform::NPUStreamSync(dev_ctx->stream());
 
 #else
     PADDLE_THROW(platform::errors::PreconditionNotMet(

paddle/fluid/operators/collective/c_sync_comm_stream_op.cc

@@ -21,7 +21,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {
@@ -80,7 +80,7 @@ class CSyncCommStreamKernel : public framework::OpKernel<T> {
     int ring_id = ctx.Attr<int>("ring_id");
     auto stream =
         platform::HCCLCommContext::Instance().Get(ring_id, place)->stream();
-    PADDLE_ENFORCE_NPU_SUCCESS(aclrtSynchronizeStream(stream));
+    platform::NPUStreamSync(stream);
 
 #else
     PADDLE_THROW(platform::errors::PreconditionNotMet(

paddle/fluid/operators/collective/c_sync_comm_stream_op_npu_test.cc

@@ -35,7 +35,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/checknumeric_npu_test.cc

@@ -36,7 +36,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/gen_hccl_id_op.cc

@@ -21,9 +21,9 @@ limitations under the License. */
 #include "paddle/fluid/framework/operator.h"
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/framework/var_type_traits.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #include "paddle/fluid/platform/device_context.h"
 #include "paddle/fluid/platform/enforce.h"
-#include "paddle/fluid/platform/hccl_helper.h"
 #include "paddle/fluid/platform/place.h"
 #include "paddle/fluid/string/split.h"
 

paddle/fluid/operators/collective/gen_hccl_id_op_helper.cc

@@ -31,7 +31,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 DECLARE_int32(get_host_by_name_time);

paddle/fluid/operators/collective/partial_allgather_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 #include <memory>
 
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/collective/partial_recv_op_npu.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/collective/partial_recv_op.h"
 
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/collective/partial_send_op_npu.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/collective/send_v2_op.h"
 
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/collective/recv_v2_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {

paddle/fluid/operators/collective/recv_v2_op_npu_test.cc

@@ -35,7 +35,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/collective/send_v2_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace paddle {

paddle/fluid/operators/collective/send_v2_op_npu_test.cc

@@ -34,7 +34,7 @@ limitations under the License. */
 
 #if defined(PADDLE_WITH_ASCEND_CL)
 #include "paddle/fluid/platform/collective_helper.h"
-#include "paddle/fluid/platform/hccl_helper.h"
+#include "paddle/fluid/platform/device/npu/hccl_helper.h"
 #endif
 
 namespace f = paddle::framework;

paddle/fluid/operators/concat_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/concat_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/controlflow/compare_op_npu.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/controlflow/compare_op.h"
 #include "paddle/fluid/framework/op_version_registry.h"
 #include "paddle/fluid/operators/elementwise/elementwise_op_function.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/controlflow/logical_op_npu.cc

@@ -10,7 +10,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/controlflow/logical_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/conv_op_npu.cc

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/fluid/operators/conv_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/conv_transpose_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/conv_transpose_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/crop_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/crop_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/cumsum_op_npu.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/framework/tensor.h"
 #include "paddle/fluid/operators/cum_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/detection/box_coder_op_npu.cc

@@ -10,7 +10,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/detection/box_coder_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/detection/density_prior_box_op_npu.cc

@@ -10,7 +10,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/detection/density_prior_box_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/detection/iou_similarity_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/detection/iou_similarity_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/detection/prior_box_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/detection/prior_box_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/dropout_op_npu.cc

@@ -18,7 +18,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/ddim.h"
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/dropout_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_add_op_npu.cc

@@ -18,7 +18,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/elementwise/elementwise_add_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_div_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/operators/elementwise/elementwise_div_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_floordiv_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/operators/elementwise/elementwise_div_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_max_op_npu.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/elementwise/elementwise_max_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_min_op_npu.cc

@@ -18,7 +18,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/elementwise/elementwise_min_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_mod_op_npu.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/elementwise/elementwise_mod_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_mul_op_npu.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/elementwise/elementwise_mul_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_npu.h

@@ -16,7 +16,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/elementwise/elementwise_op.h"
 #include "paddle/fluid/operators/elementwise/elementwise_op_function.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_pow_op_npu.cc

@@ -17,7 +17,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/elementwise/elementwise_npu.h"
 #include "paddle/fluid/operators/elementwise/elementwise_pow_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/elementwise/elementwise_sub_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/operators/elementwise/elementwise_sub_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/expand_as_v2_op_npu.cc

@@ -12,7 +12,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/expand_as_v2_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/expand_op_npu.cc

@@ -17,7 +17,7 @@ limitations under the License. */
 
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/operators/expand_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/expand_v2_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/expand_v2_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/eye_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/eye_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/fill_any_like_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/fill_any_like_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/fill_constant_batch_size_like_op_npu.cc

@@ -13,8 +13,8 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/fill_constant_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
 #include "paddle/fluid/operators/utils.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/fill_constant_op_npu.cc

@@ -13,8 +13,8 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/fill_constant_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
 #include "paddle/fluid/operators/utils.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/fill_zeros_like_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/fill_zeros_like_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/flatten_op_npu.cc

@@ -12,7 +12,7 @@
 // See the License for the specific language governing permissions and
 // limitations under the License.
 #include "paddle/fluid/operators/flatten_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/gather_nd_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/gather_nd_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/gather_op_npu.cc

@@ -18,8 +18,8 @@ limitations under the License. */
 #include <vector>
 #include "paddle/fluid/framework/tensor_util.h"
 #include "paddle/fluid/operators/kron_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
-#include "paddle/fluid/platform/npu_info.h"
+#include "paddle/fluid/platform/device/npu/npu_info.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/gelu_op_npu.cc

@@ -16,7 +16,7 @@ limitations under the License. */
 #include <string>
 
 #include "paddle/fluid/operators/gelu_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/group_norm_op_npu.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/group_norm_op.h"
 #include <vector>
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/huber_loss_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/huber_loss_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/increment_op_npu.cc

@@ -13,7 +13,7 @@
 // limitations under the License.
 
 #include "paddle/fluid/operators/increment_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/index_sample_op_npu.cc

@@ -14,7 +14,7 @@ limitations under the License. */
 
 #include "paddle/fluid/operators/index_sample_op.h"
 
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/index_select_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/index_select_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/interpolate_op_npu.cc

@@ -15,7 +15,7 @@ limitations under the License. */
 #include "paddle/fluid/operators/interpolate_op.h"
 #include <string>
 #include <vector>
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/interpolate_v2_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the License. */
 
 #include "paddle/fluid/operators/interpolate_v2_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {

paddle/fluid/operators/kldiv_loss_op_npu.cc

@@ -13,7 +13,7 @@ See the License for the specific language governing permissions and
 limitations under the Licnse. */
 
 #include "paddle/fluid/operators/kldiv_loss_op.h"
-#include "paddle/fluid/operators/npu_op_runner.h"
+#include "paddle/fluid/platform/device/npu/npu_op_runner.h"
 
 namespace paddle {
 namespace operators {