昇腾和寒武纪相关代码退场 npu相关代码退场

paddle/fluid/framework/dlpack_tensor.cc

@@ -89,11 +89,6 @@ struct DLDeviceVisitor
         platform::errors::Unimplemented("platform::XPUPlace is not supported"));
   }
 
-  inline ::DLDevice operator()(const platform::NPUPlace &place) const {
-    PADDLE_THROW(
-        platform::errors::Unimplemented("platform::NPUPlace is not supported"));
-  }
-
   inline ::DLDevice operator()(const platform::NPUPinnedPlace &place) const {
     PADDLE_THROW(platform::errors::Unimplemented(
         "platform::NPUPinnedPlace is not supported"));

paddle/fluid/framework/ir/auto_mixed_precision_pass.cc

@@ -55,8 +55,6 @@ static phi::Backend ConvertPlaceToBackend(const phi::Place& place) {
       return phi::Backend::GPU;
     case phi::AllocationType::XPU:
       return phi::Backend::XPU;
-    case phi::AllocationType::NPU:
-      return phi::Backend::NPU;
     default:
       PADDLE_THROW(platform::errors::InvalidArgument(
           "Cannot convert place(%d).", static_cast<int>(place.GetType())));

paddle/fluid/framework/op_registry.h

@@ -374,9 +374,6 @@ struct OpKernelRegistrarFunctorEx<PlaceType,
 #define REGISTER_OP_XPU_KERNEL(op_type, ...) \
   REGISTER_OP_KERNEL(op_type, XPU, ::paddle::platform::XPUPlace, __VA_ARGS__)
 
-#define REGISTER_OP_NPU_KERNEL(op_type, ...) \
-  REGISTER_OP_KERNEL(op_type, NPU, ::paddle::platform::NPUPlace, __VA_ARGS__)
-
 #define REGISTER_OP_KERNEL_EX(op_type, library_type, place_class,  \
                               customized_name,                     \
                               customized_type_value,               \
@@ -413,12 +410,6 @@ struct OpKernelRegistrarFunctorEx<PlaceType,
       ::paddle::framework::OpKernelType::kDefaultCustomizedTypeValue, \
       __VA_ARGS__)
 
-#define REGISTER_OP_NPU_KERNEL_FUNCTOR(op_type, ...)                  \
-  REGISTER_OP_KERNEL_EX(                                              \
-      op_type, NPU, ::paddle::platform::NPUPlace, DEFAULT_TYPE,       \
-      ::paddle::framework::OpKernelType::kDefaultCustomizedTypeValue, \
-      __VA_ARGS__)
-
 #define REGISTER_OP_IPU_KERNEL_FUNCTOR(op_type, ...)                  \
   REGISTER_OP_KERNEL_EX(                                              \
       op_type, IPU, ::paddle::platform::IPUPlace, DEFAULT_TYPE,       \

paddle/fluid/framework/parallel_executor.cc

@@ -1327,8 +1327,6 @@ void ParallelExecutor::InitExecutorPrivateMemberInfo(
     device_name = "CPU";
   } else if (member_->use_device_ == p::kCUDA) {
     device_name = "CUDA";
-  } else if (member_->use_device_ == p::kNPU) {
-    device_name = "NPU";
   } else if (member_->use_device_ == p::kXPU) {
     device_name = "XPU";
   } else {

paddle/fluid/inference/api/analysis_predictor.cc

@@ -138,8 +138,6 @@ phi::Backend ConvertBackend(paddle_infer::PlaceType backend) {
     case paddle_infer::PlaceType::kGPU:
       // NOTE: phi also support phi::Backend::GPUDNN.
       return phi::Backend::GPU;
-    case paddle_infer::PlaceType::kNPU:
-      return phi::Backend::NPU;
     case paddle_infer::PlaceType::kXPU:
       return phi::Backend::XPU;
     case paddle_infer::PlaceType::kCPU:

paddle/fluid/inference/api/api_impl.cc

@@ -82,8 +82,6 @@ bool NativePaddlePredictor::Init(
     place_ = paddle::platform::CUDAPlace(config_.device);
   } else if (config_.use_xpu) {
     place_ = paddle::platform::XPUPlace(config_.device);
-  } else if (config_.use_npu) {
-    place_ = paddle::platform::NPUPlace(config_.device);
   } else {
     place_ = paddle::platform::CPUPlace();
   }

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

@@ -124,9 +124,6 @@ T *Tensor::mutable_data(PlaceType place) {
     case static_cast<int>(PlaceType::kXPU): {
       return tensor->mutable_data<T>(paddle::platform::XPUPlace(device_));
     }
-    case static_cast<int>(PlaceType::kNPU): {
-      return tensor->mutable_data<T>(paddle::platform::NPUPlace(device_));
-    }
     case static_cast<int>(PlaceType::kCUSTOM): {
       return tensor->mutable_data<T>(
           paddle::platform::CustomPlace(device_type_, device_));

paddle/fluid/inference/api/paddle_tensor.h

@@ -67,7 +67,7 @@ enum DataType {
   // TODO(Inference): support more data types if needed.
 };
 
-enum class PlaceType { kUNK = -1, kCPU, kGPU, kXPU, kNPU, kIPU, kCUSTOM };
+enum class PlaceType { kUNK = -1, kCPU, kGPU, kXPU, kIPU, kCUSTOM };
 
 enum class DataLayout { kUNK = -1, kAny, kNHWC, kNCHW };
 

paddle/fluid/operators/collective/c_sync_comm_stream_op.cc

@@ -53,6 +53,3 @@ namespace plat = paddle::platform;
 REGISTER_OP_WITHOUT_GRADIENT(c_sync_comm_stream,
                              ops::CSyncCommStreamOp,
                              ops::CSyncCommStreamOpMaker);
-
-REGISTER_OP_NPU_KERNEL(c_sync_comm_stream,
-                       ops::CSyncCommStreamKernel<float, plat::NPUPlace>);

paddle/fluid/operators/fill_constant_op.cc

@@ -95,9 +95,6 @@ class FillConstantOp : public framework::OperatorWithKernel {
         case 3:
           kt.set_backend(phi::Backend::XPU);
           break;
-        case 4:
-          kt.set_backend(phi::Backend::NPU);
-          break;
         default:
           PADDLE_THROW(platform::errors::Unimplemented(
               "Could NOT determine the place of variable, place_type = %d .",
@@ -161,8 +158,7 @@ class FillConstantOpMaker : public framework::OpProtoAndCheckerMaker {
                  "0: CPUPlace. "
                  "1: CUDAPlace. "
                  "2: CUDAPinnedPlace. "
-                 "3: XPUPlace. "
-                 "4: NPUPlace. ")
+                 "3: XPUPlace. ")
         .SetDefault(-1);
     AddOutput("Out",
               "(Tensor) Tensor of specified shape will be filled "

paddle/fluid/operators/memcpy_d2h_op.cc

@@ -86,16 +86,15 @@ class MemcpyD2HOpProtoMaker : public framework::OpProtoAndCheckerMaker {
     AddOutput("Out",
               "(phi::DenseTensor) The type of output "
               "is the same as input X.");
-    AddAttr<int>(
-        "dst_place_type",
-        "Determine the dst place of tensor copy. "
-        "By Now it ONLY support XPU/NPUPlace/CUDAPlace <-> CUDAPinnedPlace/CPU"
-        "Other place type is Unimplemented and will cause ERROR."
-        "0: dst is on CPUPlace. "
-        "1: dst is on CUDAPinnedPlace. ");
+    AddAttr<int>("dst_place_type",
+                 "Determine the dst place of tensor copy. "
+                 "By Now it ONLY support XPU/CUDAPlace <-> CUDAPinnedPlace/CPU"
+                 "Other place type is Unimplemented and will cause ERROR."
+                 "0: dst is on CPUPlace. "
+                 "1: dst is on CUDAPinnedPlace. ");
     AddComment(R"DOC(
     MemcpyD2H Operator.
-    By now, it ONLY supports the memcopy between NPUPlace/CUDAPlace <-> CUDAPinnedPlace/CPU.
+    By now, it ONLY supports the memcopy between CUDAPlace <-> CUDAPinnedPlace/CPU.
     You would have to update it if you want other more capacities.
 Out = X,  when type in [phi::DenseTensor]
 raise error if the type is not listed above.

paddle/fluid/operators/memcpy_h2d_op.cc

@@ -91,13 +91,12 @@ class MemcpyH2DOpProtoMaker : public framework::OpProtoAndCheckerMaker {
                  "Determine the dst place of tensor copy. "
                  "By Now it support:"
                  "0. CUDAPinnedPlace/CPU <->CUDAPlace"
-                 "1. NPUPinnedPlace/CPU <-> NPUPlace"
-                 "2. CPU <->XPUPlace"
-                 "3. CPU <->IPUPlace"
+                 "1. CPU <->XPUPlace"
+                 "2. CPU <->IPUPlace"
                  "Other place type is Unimplemented and will cause ERROR.");
     AddComment(R"DOC(
     MemcpyD2H Operator.
-    By now, it ONLY supports the memcopy between CUDAPinnedPlace/CPU <-> NPUPlace/CUDAPlace.
+    By now, it ONLY supports the memcopy between CUDAPinnedPlace/CPU <-> CUDAPlace.
     You would have to update it if you want other more capacities.
 Out = X,  when type in [phi::DenseTensor]
 raise error if the type is not listed above.

paddle/fluid/operators/memcpy_op.cc

@@ -105,20 +105,17 @@ class MemcpyOpProtoMaker : public framework::OpProtoAndCheckerMaker {
               "is the same as input X.");
     AddAttr<int>("dst_place_type",
                  "Determine the dst place of tensor copy. "
-                 "By Now it ONLY support CUDAPlace <-> CUDAPinnedPlace or "
-                 "NPUPlace <-> CPUPlace. "
+                 "By Now it ONLY support CUDAPlace <-> CUDAPinnedPlace."
                  "Other place type is Unimplemented and will cause ERROR."
                  "0: dst is on CPUPlace. "
                  "1: dst is on CUDAPlace. "
                  "2: dst is on CUDAPinnedPlace. "
                  "3: dst is on XPUPlace. "
-                 "4: dst is on NPUPlace. "
-                 "5: dst is on NPUPinnerPlace. "
-                 "6: dst is on CustomDevicePlace");
+                 "4: dst is on NPUPinnerPlace. "
+                 "5: dst is on CustomDevicePlace");
     AddComment(R"DOC(
     Memcpy Operator.
-    By now, it ONLY supports the memcopy between CUDAPinnedPlace <-> CUDAPlace or
-    NPUPlace <-> CPUPlace, and used as an internal op by Recompute-Offload.
+    By now, it ONLY supports the memcopy between CUDAPinnedPlace <-> CUDAPlace, and used as an internal op by Recompute-Offload.
     You would have to update it if you want other more capacities.
 
 Out = X,  when type in [phi::DenseTensor]

paddle/fluid/platform/device_context.h

@@ -106,7 +106,6 @@ DeviceType Place2DeviceType(const platform::Place& place);
 constexpr DeviceType kCPU = DeviceType::CPU;
 constexpr DeviceType kCUDA = DeviceType::CUDA;
 constexpr DeviceType kXPU = DeviceType::XPU;
-constexpr DeviceType kNPU = DeviceType::NPU;
 constexpr DeviceType kIPU = DeviceType::IPU;
 constexpr DeviceType kCUSTOM_DEVICE = DeviceType::CUSTOM_DEVICE;
 

paddle/fluid/platform/device_event.h

@@ -26,7 +26,6 @@
 using ::paddle::platform::kCPU;
 using ::paddle::platform::kCUDA;
 using ::paddle::platform::kCUSTOM_DEVICE;
-using ::paddle::platform::kNPU;
 using ::paddle::platform::kXPU;
 
 USE_EVENT(kCPU)

paddle/fluid/platform/place.cc

@@ -102,8 +102,6 @@ Place PlaceHelper::CreatePlace(const std::string &dev_type, size_t dev_id) {
     return platform::CPUPlace();
   } else if (dev_type == "gpu") {
     return platform::CUDAPlace(dev_id);
-  } else if (dev_type == "npu") {
-    return platform::NPUPlace(dev_id);
   } else if (dev_type == "xpu") {
     return platform::XPUPlace(dev_id);
   } else {

paddle/fluid/platform/place.h

@@ -28,7 +28,6 @@ using Place = phi::Place;
 using CPUPlace = phi::CPUPlace;
 using CUDAPlace = phi::GPUPlace;
 using CUDAPinnedPlace = phi::GPUPinnedPlace;
-using NPUPlace = phi::NPUPlace;
 using NPUPinnedPlace = phi::NPUPinnedPlace;
 using XPUPlace = phi::XPUPlace;
 using IPUPlace = phi::IPUPlace;
@@ -88,11 +87,6 @@ typename Visitor::result_type VisitPlace(const Place &place,
       return typename Visitor::result_type();
 #endif
     }
-    case phi::AllocationType::NPU: {
-      PADDLE_THROW(platform::errors::Unavailable(
-          "Paddle is not compiled with NPU. Cannot visit npu_pinned"));
-      return typename Visitor::result_type();
-    }
     case phi::AllocationType::NPUPINNED: {
       PADDLE_THROW(platform::errors::Unavailable(
           "Paddle is not compiled with NPU. Cannot visit npu_pinned"));

paddle/fluid/pybind/eager.cc

@@ -159,7 +159,7 @@ void InitTensorWithNumpyValue(TensorObject* self,
   } else {
     PADDLE_THROW(platform::errors::InvalidArgument(
         "Place should be one of "
-        "CPUPlace/XPUPlace/CUDAPlace/CUDAPinnedPlace/NPUPlace/CustomPlace"));
+        "CPUPlace/XPUPlace/CUDAPlace/CUDAPinnedPlace/CustomPlace"));
   }
 }
 

paddle/fluid/pybind/eager_math_op_patch.cc

@@ -108,7 +108,7 @@ void InitTensorWithNumpyValue(const py::object& array,
   } else {
     PADDLE_THROW(platform::errors::InvalidArgument(
         "Place should be one of "
-        "CPUPlace/XPUPlace/CUDAPlace/CUDAPinnedPlace/NPUPlace/CustomPlace"));
+        "CPUPlace/XPUPlace/CUDAPlace/CUDAPinnedPlace/CustomPlace"));
   }
 }
 

paddle/fluid/pybind/eager_utils.cc

@@ -52,7 +52,6 @@ extern PyTypeObject* g_place_pytype;
 extern PyTypeObject* g_cudaplace_pytype;
 extern PyTypeObject* g_cpuplace_pytype;
 extern PyTypeObject* g_xpuplace_pytype;
-extern PyTypeObject* g_npuplace_pytype;
 extern PyTypeObject* g_cudapinnedplace_pytype;
 extern PyTypeObject* g_customplace_pytype;
 extern PyTypeObject* g_framework_tensor_pytype;
@@ -529,9 +528,6 @@ platform::Place CastPyArg2Place(PyObject* obj, ssize_t arg_pos) {
   } else if (PyObject_IsInstance(
                  obj, reinterpret_cast<PyObject*>(g_xpuplace_pytype))) {
     place = ::pybind11::handle(obj).cast<platform::XPUPlace>();
-  } else if (PyObject_IsInstance(
-                 obj, reinterpret_cast<PyObject*>(g_npuplace_pytype))) {
-    place = ::pybind11::handle(obj).cast<platform::NPUPlace>();
   } else if (PyObject_IsInstance(
                  obj, reinterpret_cast<PyObject*>(g_cudapinnedplace_pytype))) {
     place = ::pybind11::handle(obj).cast<platform::CUDAPinnedPlace>();
@@ -542,7 +538,7 @@ platform::Place CastPyArg2Place(PyObject* obj, ssize_t arg_pos) {
     PADDLE_THROW(platform::errors::InvalidArgument(
         "argument (position %d) must be "
         "one "
-        "of(Place,CUDAPlace,CPUPlace,XPUPlace,NPUPlace,CUDAPinnedPlace,"
+        "of(Place,CUDAPlace,CPUPlace,XPUPlace,CUDAPinnedPlace,"
         "CustomPlace), "
         "but got %s",
         arg_pos + 1,