unify cpu context, part2 (#44012)

* fix init() * delete test_device_context * replace CPUDeviceContext with CPUContext * fix test_scalar * remove dot_op.cc * fix compile
PaddlePaddle · Jul 2, 2022 · 755438a · 755438a
1 parent 09096ae
commit 755438a
Show file tree

Hide file tree

Showing 289 changed files with 1,525 additions and 1,909 deletions.
diff --git a/paddle/fluid/distributed/collective/reducer.cc b/paddle/fluid/distributed/collective/reducer.cc
@@ -251,7 +251,7 @@ void EagerGroup::ConcatTensors(const platform::Place &place) {
         "Please recompile or reinstall Paddle with NCCL support."));
 #endif
   } else if (platform::is_cpu_place(place)) {
-    auto *default_ctx = static_cast<platform::CPUDeviceContext *>(
+    auto *default_ctx = static_cast<phi::CPUContext *>(
         platform::DeviceContextPool::Instance().Get(place));
     ConcatTensorsWithType(
         *default_ctx, dense_tensors_, &dense_contents_, dtype_);
@@ -274,7 +274,7 @@ void EagerGroup::SplitTensors(const platform::Place &place) {
         "Please recompile or reinstall Paddle with NCCL support."));
 #endif
   } else if (platform::is_cpu_place(place)) {
-    auto *default_ctx = static_cast<platform::CPUDeviceContext *>(
+    auto *default_ctx = static_cast<phi::CPUContext *>(
         platform::DeviceContextPool::Instance().Get(place));
     SplitTensorsWithType(
         *default_ctx, &dense_contents_, &dense_tensors_, dtype_);
@@ -891,7 +891,7 @@ void EagerReducer::AllReduceSparse(EagerGroup *group,
         "Please recompile or reinstall Paddle with NCCL support."));
 #endif
   } else if (platform::is_cpu_place(inner_place_)) {
-    dev_ctx = static_cast<platform::CPUDeviceContext *>(
+    dev_ctx = static_cast<phi::CPUContext *>(
         platform::DeviceContextPool::Instance().Get(inner_place_));
   } else {
     PADDLE_THROW(platform::errors::Unimplemented(

diff --git a/paddle/fluid/distributed/common/utils.h b/paddle/fluid/distributed/common/utils.h
@@ -31,9 +31,9 @@ namespace paddle {
 namespace distributed {
 
 template <typename T>
-inline phi::funcs::BlasT<paddle::platform::CPUDeviceContext, T> GetBlas() {
-  paddle::platform::CPUDeviceContext cpu_ctx;
-  return phi::funcs::GetBlas<paddle::platform::CPUDeviceContext, T>(cpu_ctx);
+inline phi::funcs::BlasT<phi::CPUContext, T> GetBlas() {
+  phi::CPUContext cpu_ctx;
+  return phi::funcs::GetBlas<phi::CPUContext, T>(cpu_ctx);
 }
 
 template <typename T>

diff --git a/paddle/fluid/distributed/ps/service/communicator/communicator.cc b/paddle/fluid/distributed/ps/service/communicator/communicator.cc
@@ -353,11 +353,12 @@ void Communicator::RpcRecvSparse(const std::string &varname,
 
   bool training = true;
 
-  auto status = _worker_ptr->PullSparseParam((float **)push_g_vec.data(),
-                                             table_id,  // NOLINT
-                                             sparse_push_keys.data(),
-                                             sparse_push_keys.size(),
-                                             training);
+  auto status =
+      _worker_ptr->PullSparseParam(static_cast<float **>(push_g_vec.data()),
+                                   table_id,
+                                   sparse_push_keys.data(),
+                                   sparse_push_keys.size(),
+                                   training);
   status.wait();
   return;
 }
@@ -1184,12 +1185,12 @@ void GeoCommunicator::SendDense(const CommContext &send_ctx) {
     auto &t_latest = var_latest->Get<framework::LoDTensor>();
     auto t_timestamp = var_timestamp->GetMutable<framework::LoDTensor>();
 
-    paddle::platform::CPUDeviceContext cpu_ctx;
+    phi::CPUContext cpu_ctx;
     auto *var_delta = delta_scope_->Var(varname);
     auto *t_delta = var_delta->GetMutable<framework::LoDTensor>();
     t_delta->mutable_data<float>(t_latest.dims(), cpu_ctx.GetPlace());
 
-    auto blas = phi::funcs::GetBlas<platform::CPUDeviceContext, float>(cpu_ctx);
+    auto blas = phi::funcs::GetBlas<phi::CPUContext, float>(cpu_ctx);
     blas.VSUB(t_latest.numel(),
               t_latest.data<float>(),
               t_timestamp->data<float>(),
@@ -1218,7 +1219,7 @@ void GeoCommunicator::RecvDense(const CommContext &send_ctx) {
   RpcRecvDense(varnames, table_id, pserver_scope_.get());
 
   // 2.1 pserver - old => delta; 2.2 latest + old => latest 2.3 old => pserver
-  paddle::platform::CPUDeviceContext cpu_ctx;
+  phi::CPUContext cpu_ctx;
   for (auto &varname : varnames) {
     auto *var_latest = recv_scope_->FindVar(varname);
     auto t_latest = var_latest->GetMutable<framework::LoDTensor>();
@@ -1233,7 +1234,7 @@ void GeoCommunicator::RecvDense(const CommContext &send_ctx) {
     auto *t_delta = var_delta->GetMutable<framework::LoDTensor>();
     t_delta->mutable_data<float>(t_latest->dims(), cpu_ctx.GetPlace());
 
-    auto blas = phi::funcs::GetBlas<platform::CPUDeviceContext, float>(cpu_ctx);
+    auto blas = phi::funcs::GetBlas<phi::CPUContext, float>(cpu_ctx);
     blas.VSUB(t_latest->numel(),
               t_pserver.data<float>(),
               t_old->data<float>(),
@@ -1334,7 +1335,7 @@ void GeoCommunicator::SendSparse(const std::string &varname,
   auto *t_old = var_old->GetMutable<framework::LoDTensor>();
 
   auto dims1 = t_latest.dims()[1];
-  paddle::platform::CPUDeviceContext cpu_ctx;
+  phi::CPUContext cpu_ctx;
 
   auto *var_delta = delta_scope_->Var(varname);
   auto *t_delta = var_delta->GetMutable<phi::SelectedRows>();
@@ -1345,7 +1346,7 @@ void GeoCommunicator::SendSparse(const std::string &varname,
   t_delta->set_rows(sparse_ids);
   t_delta->set_height(t_latest.dims()[0]);
 
-  auto blas = phi::funcs::GetBlas<platform::CPUDeviceContext, float>(cpu_ctx);
+  auto blas = phi::funcs::GetBlas<phi::CPUContext, float>(cpu_ctx);
   float coefficient = 1.0 / static_cast<float>(trainers_);
 
   std::vector<float *> push_g_vec;
@@ -1419,8 +1420,8 @@ void GeoCommunicator::RecvSparse(const std::string &varname,
   std::vector<float> v_delta;
   v_delta.resize(numel);
 
-  paddle::platform::CPUDeviceContext cpu_ctx;
-  auto blas = phi::funcs::GetBlas<platform::CPUDeviceContext, float>(cpu_ctx);
+  phi::CPUContext cpu_ctx;
+  auto blas = phi::funcs::GetBlas<phi::CPUContext, float>(cpu_ctx);
 
   for (auto j = 0; j < static_cast<int>(keys.size()); ++j) {
     VLOG(5) << "DEBUG GeoCommunicator::RecvSparse recv sparse key" << keys[j]

diff --git a/paddle/fluid/distributed/ps/service/communicator/communicator.h b/paddle/fluid/distributed/ps/service/communicator/communicator.h
@@ -185,9 +185,8 @@ inline void MergeVars(const std::string &var_name,
     }
 
     // set output tensor to 0.
-    paddle::platform::CPUDeviceContext cpu_ctx;
-    phi::funcs::SetConstant<paddle::platform::CPUDeviceContext, T>
-        constant_functor;
+    phi::CPUContext cpu_ctx;
+    phi::funcs::SetConstant<phi::CPUContext, T> constant_functor;
     constant_functor(cpu_ctx, out_t, static_cast<T>(0));
     // sum all vars to out
     auto result = EigenVector<T>::Flatten(*out_t);
@@ -210,16 +209,13 @@ inline void MergeVars(const std::string &var_name,
     for (auto &var : vars) {
       inputs.push_back(&var->Get<phi::SelectedRows>());
     }
-    paddle::platform::CPUDeviceContext dev_ctx;
+    phi::CPUContext dev_ctx;
     if (merge_add) {
-      paddle::operators::math::scatter::
-          MergeAdd<paddle::platform::CPUDeviceContext, T>
-              merge_add;
+      paddle::operators::math::scatter::MergeAdd<phi::CPUContext, T> merge_add;
       merge_add(dev_ctx, inputs, out_slr);
     } else {
-      paddle::operators::math::scatter::
-          MergeAverage<paddle::platform::CPUDeviceContext, T>
-              merge_average;
+      paddle::operators::math::scatter::MergeAverage<phi::CPUContext, T>
+          merge_average;
       merge_average(dev_ctx, inputs, out_slr);
     }
 

diff --git a/paddle/fluid/eager/nan_inf_utils.cc b/paddle/fluid/eager/nan_inf_utils.cc
@@ -48,8 +48,7 @@ void CheckTensorHasNanOrInf(const std::string& api_name, const Tensor& tensor) {
 #endif
       return;
     }
-    paddle::framework::details::tensor_check<
-        paddle::platform::CPUDeviceContext>(
+    paddle::framework::details::tensor_check<phi::CPUContext>(
         api_name, tensor_name, *dense_tensor, place);
   }
 }

diff --git a/paddle/fluid/framework/data_device_transform_test.cu b/paddle/fluid/framework/data_device_transform_test.cu
@@ -90,9 +90,8 @@ REGISTER_OP_WITHOUT_GRADIENT(
     test_op,
     paddle::framework::TestOpWithKernel,
     paddle::framework::OpKernelTestProtoAndCheckerMaker);
-REGISTER_OP_CPU_KERNEL(
-    test_op,
-    paddle::framework::TestKernel<paddle::platform::CPUDeviceContext, float>);
+REGISTER_OP_CPU_KERNEL(test_op,
+                       paddle::framework::TestKernel<phi::CPUContext, float>);
 REGISTER_OP_CUDA_KERNEL(
     test_op,
     paddle::framework::TestKernel<paddle::platform::CUDADeviceContext, float>);

diff --git a/paddle/fluid/framework/data_layout_transform.cc b/paddle/fluid/framework/data_layout_transform.cc
@@ -44,8 +44,8 @@ void CastDataLayout::apply() {
   auto place = ctx_->GetPlace();
 
   if (platform::is_cpu_place(place)) {
-    phi::funcs::Transpose<platform::CPUDeviceContext, T, 4> trans4;
-    auto* context = static_cast<const platform::CPUDeviceContext*>(ctx_);
+    phi::funcs::Transpose<phi::CPUContext, T, 4> trans4;
+    auto* context = static_cast<const phi::CPUContext*>(ctx_);
     trans4(*context, in_, out_, axis_);
   } else {
     PADDLE_THROW(platform::errors::PreconditionNotMet(

diff --git a/paddle/fluid/framework/data_type_transform.cc b/paddle/fluid/framework/data_type_transform.cc
@@ -94,8 +94,8 @@ struct CastDataType {
     auto* out_begin = out_->mutable_data<OutType>(in_.place());
 
     if (platform::is_cpu_place(in_.place())) {
-      platform::Transform<platform::CPUDeviceContext> trans;
-      auto* context = static_cast<const platform::CPUDeviceContext*>(ctx_);
+      platform::Transform<phi::CPUContext> trans;
+      auto* context = static_cast<const phi::CPUContext*>(ctx_);
       trans(*context,
             in_begin,
             in_end,

diff --git a/paddle/fluid/framework/details/broadcast_op_handle_test.h b/paddle/fluid/framework/details/broadcast_op_handle_test.h
@@ -117,7 +117,7 @@ struct TestBroadcastOpHandle {
       for (int i = 0; i < count; ++i) {
         auto p = p::CPUPlace();
         place_list_.push_back(p);
-        ctxs_.emplace_back(new p::CPUDeviceContext(p));
+        ctxs_.emplace_back(new phi::CPUContext(p));
       }
 #if defined(PADDLE_WITH_XPU_BKCL)
       bkcl_ctxs_.reset(nullptr);

diff --git a/paddle/fluid/framework/details/gather_op_handle_test.cc b/paddle/fluid/framework/details/gather_op_handle_test.cc
@@ -69,7 +69,7 @@ struct TestGatherOpHandle {
       for (int i = 0; i < count; ++i) {
         auto p = p::CPUPlace();
         gpu_list_.push_back(p);
-        ctxs_.emplace_back(new p::CPUDeviceContext(p));
+        ctxs_.emplace_back(new phi::CPUContext(p));
       }
     }
   }

diff --git a/paddle/fluid/framework/details/nan_inf_utils_detail.cc b/paddle/fluid/framework/details/nan_inf_utils_detail.cc
@@ -316,7 +316,7 @@ template <>
 
 template <>
 template <typename T>
-void TensorCheckerVisitor<platform::CPUDeviceContext>::apply(
+void TensorCheckerVisitor<phi::CPUContext>::apply(
     typename std::enable_if<
         std::is_floating_point<T>::value ||
         std::is_same<T, ::paddle::platform::complex<float>>::value ||
@@ -329,11 +329,11 @@ void TensorCheckerVisitor<platform::CPUDeviceContext>::apply(
 }
 
 template <>
-void tensor_check<platform::CPUDeviceContext>(const std::string& op_type,
-                                              const std::string& var_name,
-                                              const framework::Tensor& tensor,
-                                              const platform::Place& place) {
-  TensorCheckerVisitor<platform::CPUDeviceContext> vistor(
+void tensor_check<phi::CPUContext>(const std::string& op_type,
+                                   const std::string& var_name,
+                                   const framework::Tensor& tensor,
+                                   const platform::Place& place) {
+  TensorCheckerVisitor<phi::CPUContext> vistor(
       op_type, var_name, tensor, place);
   VisitDataType(framework::TransToProtoVarType(tensor.dtype()), vistor);
 }
@@ -439,7 +439,7 @@ void CheckVarHasNanOrInf(const std::string& op_type,
 #endif
     return;
   }
-  tensor_check<platform::CPUDeviceContext>(op_type, var_name, *tensor, place);
+  tensor_check<phi::CPUContext>(op_type, var_name, *tensor, place);
 }
 
 void CheckVarHasNanOrInf(const std::string& op_type,

diff --git a/paddle/fluid/framework/details/reduce_op_handle_test.cc b/paddle/fluid/framework/details/reduce_op_handle_test.cc
@@ -81,7 +81,7 @@ struct TestReduceOpHandle {
       for (int i = 0; i < count; ++i) {
         auto p = p::CPUPlace();
         gpu_list_.push_back(p);
-        ctxs_.emplace_back(new p::CPUDeviceContext(p));
+        ctxs_.emplace_back(new phi::CPUContext(p));
       }
 #if defined(PADDLE_WITH_NCCL) || defined(PADDLE_WITH_RCCL)
       nccl_ctxs_.reset(nullptr);

diff --git a/paddle/fluid/framework/lod_tensor.h b/paddle/fluid/framework/lod_tensor.h
@@ -144,7 +144,7 @@ LoDTensor LodExpand(const LoDTensor& source,
       auto slice = tensor.Slice(elem, elem + 1);
       TensorCopy(source.Slice(ins, ins + 1),
                  platform::CPUPlace(),
-                 platform::CPUDeviceContext(),
+                 phi::CPUContext(),
                  &slice);
     }
   }

diff --git a/paddle/fluid/framework/op_registry_test.cc b/paddle/fluid/framework/op_registry_test.cc
@@ -232,9 +232,8 @@ class OpKernelTest : public paddle::framework::OpKernel<T> {
 REGISTER_OP_WITHOUT_GRADIENT(op_with_kernel,
                              paddle::framework::OpWithKernelTest,
                              paddle::framework::OpKernelTestMaker);
-REGISTER_OP_CPU_KERNEL(
-    op_with_kernel,
-    paddle::framework::OpKernelTest<paddle::platform::CPUDeviceContext, float>);
+REGISTER_OP_CPU_KERNEL(op_with_kernel,
+                       paddle::framework::OpKernelTest<phi::CPUContext, float>);
 
 REGISTER_OP_CUDA_KERNEL(
     op_with_kernel,
@@ -264,10 +263,9 @@ TEST(OperatorRegistrar, CUDA) {
 }
 
 static int op_test_value = 0;
-
-using paddle::platform::CPUDeviceContext;
 using paddle::platform::CUDADeviceContext;
 using paddle::platform::DeviceContext;
+using phi::CPUContext;
 
 namespace paddle {
 namespace framework {
@@ -295,8 +293,7 @@ class OpMultiKernelTest : public paddle::framework::OpKernel<T> {
 };
 
 template <typename T>
-class OpMultiKernelTest<CPUDeviceContext, T>
-    : public paddle::framework::OpKernel<T> {
+class OpMultiKernelTest<CPUContext, T> : public paddle::framework::OpKernel<T> {
  public:
   void Compute(const paddle::framework::ExecutionContext& ctx) const {
     ++op_test_value;
@@ -319,7 +316,7 @@ class OpMultiKernelTest2 : public paddle::framework::OpKernel<T> {
 };
 
 template <typename T>
-class OpMultiKernelTest2<CPUDeviceContext, T>
+class OpMultiKernelTest2<CPUContext, T>
     : public paddle::framework::OpKernel<T> {
  public:
   void Compute(const paddle::framework::ExecutionContext& ctx) const {
@@ -342,16 +339,14 @@ class OpMultiKernelTest2<CUDADeviceContext, T>
 REGISTER_OP_WITHOUT_GRADIENT(op_with_multi_kernel,
                              paddle::framework::OpWithMultiKernelTest,
                              paddle::framework::OpKernelTestMaker);
-REGISTER_OP_KERNEL(
-    op_with_multi_kernel,
-    CPU,
-    paddle::platform::CPUPlace,
-    paddle::framework::OpMultiKernelTest<CPUDeviceContext, float>);
-REGISTER_OP_KERNEL(
-    op_with_multi_kernel,
-    MKLDNN,
-    paddle::platform::CPUPlace,
-    paddle::framework::OpMultiKernelTest2<CPUDeviceContext, float>);
+REGISTER_OP_KERNEL(op_with_multi_kernel,
+                   CPU,
+                   paddle::platform::CPUPlace,
+                   paddle::framework::OpMultiKernelTest<CPUContext, float>);
+REGISTER_OP_KERNEL(op_with_multi_kernel,
+                   MKLDNN,
+                   paddle::platform::CPUPlace,
+                   paddle::framework::OpMultiKernelTest2<CPUContext, float>);
 REGISTER_OP_KERNEL(
     op_with_multi_kernel,
     CUDA,

diff --git a/paddle/fluid/framework/operator_test.cc b/paddle/fluid/framework/operator_test.cc
@@ -420,16 +420,13 @@ REGISTER_OP_WITHOUT_GRADIENT(
 
 REGISTER_OP_CPU_KERNEL(
     indicate_lod_tensor_data_type_test,
-    paddle::framework::EmptyTestKernel<paddle::platform::CPUDeviceContext,
-                                       int>);
+    paddle::framework::EmptyTestKernel<phi::CPUContext, int>);
 REGISTER_OP_CPU_KERNEL(
     indicate_selected_rows_data_type_test,
-    paddle::framework::EmptyTestKernel<paddle::platform::CPUDeviceContext,
-                                       int>);
+    paddle::framework::EmptyTestKernel<phi::CPUContext, int>);
 REGISTER_OP_CPU_KERNEL(
     indicate_other_data_type_test,
-    paddle::framework::EmptyTestKernel<paddle::platform::CPUDeviceContext,
-                                       int>);
+    paddle::framework::EmptyTestKernel<phi::CPUContext, int>);
 
 TEST(IndicateVarDataTypeTest, lodtensor) {
   paddle::framework::InitDevices();
@@ -599,16 +596,14 @@ REGISTER_OP_WITHOUT_GRADIENT(get_lod_level_test,
                              paddle::framework::GetSetLoDLevelTestMaker);
 REGISTER_OP_CPU_KERNEL(
     get_lod_level_test,
-    paddle::framework::EmptyTestKernel<paddle::platform::CPUDeviceContext,
-                                       float>);
+    paddle::framework::EmptyTestKernel<phi::CPUContext, float>);
 
 REGISTER_OP_WITHOUT_GRADIENT(set_lod_level_test,
                              paddle::framework::SetLoDLevelTest,
                              paddle::framework::GetSetLoDLevelTestMaker);
 REGISTER_OP_CPU_KERNEL(
     set_lod_level_test,
-    paddle::framework::EmptyTestKernel<paddle::platform::CPUDeviceContext,
-                                       float>);
+    paddle::framework::EmptyTestKernel<phi::CPUContext, float>);
 
 void SetGetLoDLevelTestMain(std::string op_type) {
   paddle::framework::InitDevices({});

diff --git a/paddle/fluid/framework/phi_utils.h b/paddle/fluid/framework/phi_utils.h
@@ -66,7 +66,7 @@ struct ConvertToPhiContext {
 };
 
 template <>
-struct ConvertToPhiContext<platform::CPUDeviceContext> {
+struct ConvertToPhiContext<phi::CPUContext> {
   using TYPE = phi::CPUContext;
 };
 

diff --git a/paddle/fluid/framework/selected_rows_utils_test.cc b/paddle/fluid/framework/selected_rows_utils_test.cc
@@ -53,7 +53,7 @@ TEST_F(SelectedRowsTester, complete_dims) {
 
 TEST_F(SelectedRowsTester, SerializeAndDeseralize) {
   phi::SelectedRows dst_tensor;
-  platform::CPUDeviceContext cpu_ctx(place_);
+  phi::CPUContext cpu_ctx(place_);
   std::ostringstream oss;
 
   SerializeToStream(oss, *selected_rows_, cpu_ctx);