[IR] New IR access InterpreterCore：add local scope logic (#55112)

* add local scope

* refine code

* refien code

* refine code

* support local scope for BuildFuncList

* fix bug

* add log

* fix bug

* polish code

* fix bug

paddle/fluid/framework/new_executor/interpreter/interpreter_util.cc

@@ -940,6 +940,7 @@ void BuildOpFuncList(
     ::ir::Block* block,
     std::vector<OpFuncNode>* vec_func_list,
     framework::Scope* scope,
+    framework::Scope* local_scope,
     const std::unordered_map<::ir::Value, std::string>& value_2_name_map,
     const ExecutionConfig& execution_config) {
   vec_func_list->reserve(block->size());
@@ -979,6 +980,7 @@ void BuildOpFuncList(
         false>((*it),
                value_2_name_map,
                scope,
+               local_scope,
                op_yaml_info_parser,
                &(op_func_node.infer_meta_context_));
 
@@ -1004,6 +1006,7 @@ void BuildOpFuncList(
                           true>((*it),
                                 value_2_name_map,
                                 scope,
+                                local_scope,
                                 op_yaml_info_parser,
                                 &(op_func_node.kernel_context_),
                                 &(op_func_node.input_index),

paddle/fluid/framework/new_executor/interpreter/interpreter_util.h

@@ -98,6 +98,7 @@ void BuildOpFuncList(
     ::ir::Block* block,
     std::vector<OpFuncNode>* vec_func_list,
     framework::Scope* scope,
+    framework::Scope* local_scope,
     const std::unordered_map<::ir::Value, std::string>& value_2_name_map,
     const ExecutionConfig& execution_config);
 

paddle/fluid/framework/new_executor/interpreter_base_impl.h

@@ -86,6 +86,8 @@ class InterpreterBaseImpl {
 
   virtual void reset_scope(Scope* new_scope) = 0;
 
+  virtual const Scope* local_scope() const = 0;
+
   virtual const platform::Place& GetPlace() const = 0;
 
   virtual void SetOutputHooks(const std::vector<HookFunc>& hookfuncs) = 0;

paddle/fluid/framework/new_executor/interpretercore.cc

@@ -101,6 +101,10 @@ void InterpreterCore::reset_scope(Scope* new_scope) {
   impl_->reset_scope(new_scope);
 }
 
+const Scope* InterpreterCore::local_scope() const {
+  return impl_->local_scope();
+}
+
 const platform::Place& InterpreterCore::GetPlace() const {
   return impl_->GetPlace();
 }

paddle/fluid/framework/new_executor/interpretercore.h

@@ -65,6 +65,8 @@ class InterpreterCore {
 
   void reset_scope(Scope* new_scope);
 
+  const Scope* local_scope() const;
+
   const platform::Place& GetPlace() const;
 
   void SetOutputHooks(const std::vector<HookFunc>& hookfuncs);

paddle/fluid/framework/new_executor/new_ir_interpreter.cc

@@ -49,12 +49,14 @@ NewIRInterpreter::NewIRInterpreter(const platform::Place& place,
       stream_analyzer_(place),
       execution_config_(execution_config),
       var_scope_(scope),
+      scope_(scope),
       ir_program_(std::move(ir_prog)) {
   VLOG(4) << "NewIRInterpreter(): " << this << " on " << place_;
   static_build_ = FLAGS_new_executor_static_build &&
                   !FLAGS_new_executor_use_cuda_graph &&
                   !execution_config.used_for_control_flow_op;
   //    &&interpreter::BlockCanBeStaticBuilt(block);
+  static_build_ = true;
 
   exception_notifier_ = main_thread_blocker_.RegisterEvent(kExceptionCaught);
   completion_notifier_ = main_thread_blocker_.RegisterEvent(kTaskCompletion);
@@ -62,21 +64,19 @@ NewIRInterpreter::NewIRInterpreter(const platform::Place& place,
   if (!FLAGS_new_executor_use_local_scope) {
     execution_config_.create_local_scope = false;
   }
-  execution_config_.AnalyzeThreadPoolConfig(place,
-                                            ir_program_->block()->size());
-  execution_config_.Log(/*log_level=*/8);
-
   if (execution_config_.create_local_scope) {
-    auto local_scope = &var_scope_.GetMutableScope()->NewScope();
+    auto local_scope = &scope_->NewScope();
     local_scope_ = local_scope;
+    VLOG(6) << "new ir interpretercore scope: " << scope_ << "\t"
+            << "; local scope: " << local_scope_;
   }
-
-  // force use outer scope for now
-  local_scope_ = scope;
-  static_build_ = true;
-
+  // TODO(zhangbo): delete var_scope
   var_scope_.SetLocalScope(local_scope_);
 
+  execution_config_.AnalyzeThreadPoolConfig(place,
+                                            ir_program_->block()->size());
+  execution_config_.Log(/*log_level=*/8);
+
   instruction_scheduling_priority_less = [this](size_t lhs, size_t rhs) {
     SchedulingPriority lhs_scheduling_priority =
         vec_instruction_[lhs].GetSchedulingPriority();
@@ -185,12 +185,13 @@ FetchList NewIRInterpreter::Run(const std::vector<std::string>& feed_names,
   if (!is_build_) {
     LOG_FIRST_N(INFO, 1) << "New Executor is Running.";
     ::ir::BuildScope(
-        ir_program_->block(), local_scope_, &value_2_var_name_map_);
+        ir_program_->block(), scope_, local_scope_, &value_2_var_name_map_);
 
     std::vector<paddle::framework::OpFuncNode> op_func_nodes;
     interpreter::BuildOpFuncList(place_,
                                  ir_program_->block(),
                                  &op_func_nodes,
+                                 scope_,
                                  local_scope_,
                                  value_2_var_name_map_,
                                  execution_config_);
@@ -212,8 +213,7 @@ FetchList NewIRInterpreter::Run(const std::vector<std::string>& feed_names,
   }
 
   // return Fetch Tensors
-  Scope* inner_scope =
-      HasLocalScope() ? local_scope_ : var_scope_.GetMutableScope();
+  Scope* inner_scope = HasLocalScope() ? local_scope_ : scope_;
   auto* fetch_var = inner_scope->FindVar(interpreter::kFetchVarName);
   if (fetch_var && need_fetch) {
     auto fetch_list = std::move(*fetch_var->GetMutable<framework::FetchList>());
@@ -287,6 +287,8 @@ void NewIRInterpreter::reset_scope(Scope* new_scope) {
   }
 }
 
+const Scope* NewIRInterpreter::local_scope() const { return local_scope_; }
+
 void NewIRInterpreter::ShareWorkQueueFrom(InterpreterBaseImpl* src) {
   async_work_queue_ = reinterpret_cast<NewIRInterpreter*>(src)->GetWorkQueue();
   VLOG(8) << "Share AsyncWorkQueue from InterpreterCore(" << src
@@ -321,8 +323,7 @@ std::shared_ptr<interpreter::AsyncWorkQueue> NewIRInterpreter::GetWorkQueue() {
 }
 
 void NewIRInterpreter::BuildAndCacheInstructionCtx(Instruction* instr_node) {
-  Scope* inner_scope =
-      HasLocalScope() ? local_scope_ : var_scope_.GetMutableScope();
+  Scope* inner_scope = HasLocalScope() ? local_scope_ : scope_;
   VariableValueMap ins_map;
   for (auto& var_name_item : instr_node->Inputs()) {
     std::vector<Variable*> input_vars;
@@ -349,8 +350,7 @@ void NewIRInterpreter::BuildAndCacheInstructionCtx(Instruction* instr_node) {
   if (instr_node->OpBase()->Type() == "cinn_launch" ||
       instr_node->OpBase()->Type() == "cinn_instruction_run") {  // OP use scope
                                                                  // in kernel
-    Scope* local_scope = HasLocalScope() ? var_scope_.GetMutableLocalScope()
-                                         : var_scope_.GetMutableScope();
+    Scope* local_scope = HasLocalScope() ? local_scope_ : scope_;
     instr_node->ResetContextWithScope(ins_map, outs_map, *local_scope);
   } else {
     instr_node->ResetContext(ins_map, outs_map);
@@ -380,8 +380,7 @@ void NewIRInterpreter::BuildInplace() {
     }
   }
 
-  Scope* local_scope = HasLocalScope() ? var_scope_.GetMutableLocalScope()
-                                       : var_scope_.GetMutableScope();
+  Scope* local_scope = HasLocalScope() ? local_scope_ : scope_;
   std::vector<std::vector<size_t>> input_var2op(var_scope_.VarSize());
   for (Instruction& instr : vec_instruction_) {
     for (auto& item : instr.Inputs()) {
@@ -799,8 +798,7 @@ void NewIRInterpreter::BuildSkipShareLoDInfo() {
 void NewIRInterpreter::RunOperator(const Instruction& instr_node) {
   auto* op = instr_node.OpBase();
   auto place = instr_node.DeviceContext().GetPlace();
-  Scope* local_scope = HasLocalScope() ? var_scope_.GetMutableLocalScope()
-                                       : var_scope_.GetMutableScope();
+  Scope* local_scope = HasLocalScope() ? local_scope_ : scope_;
   VLOG(4) << "Start run " << place << " " << op->DebugStringEx(local_scope);
 
   auto op_with_kernel = dynamic_cast<const framework::OperatorWithKernel*>(op);
@@ -1047,7 +1045,7 @@ void NewIRInterpreter::ExecuteInstructionList(
             if (cancel) {
               break;
             }
-            VLOG(0) << "deps:\n" << GetDepsString();
+            VLOG(6) << "deps:\n" << GetDepsString();
             times++;
           }
           return times;

paddle/fluid/framework/new_executor/new_ir_interpreter.h

@@ -60,6 +60,8 @@ class NewIRInterpreter : public InterpreterBaseImpl {
 
   void reset_scope(Scope* new_scope) override;
 
+  const Scope* local_scope() const override;
+
   const platform::Place& GetPlace() const override { return place_; }
 
   void SetOutputHooks(const std::vector<HookFunc>& hookfuncs) override {
@@ -143,6 +145,7 @@ class NewIRInterpreter : public InterpreterBaseImpl {
   ExecutionConfig execution_config_;
 
   VariableScope var_scope_;
+  Scope* scope_{nullptr};
   Scope* local_scope_{nullptr};  // not owned
 
   EventsWaiter main_thread_blocker_;

paddle/fluid/framework/new_executor/program_interpreter.cc

@@ -275,6 +275,7 @@ void ProgramInterpreter::reset_scope(Scope* new_scope) {
   }
 }
 
+const Scope* ProgramInterpreter::local_scope() const { return local_scope_; }
 void ProgramInterpreter::ShareWorkQueueFrom(InterpreterBaseImpl* src) {
   async_work_queue_ =
       reinterpret_cast<ProgramInterpreter*>(src)->GetWorkQueue();

paddle/fluid/framework/new_executor/program_interpreter.h

@@ -62,6 +62,8 @@ class ProgramInterpreter : public InterpreterBaseImpl {
 
   void reset_scope(Scope* new_scope) override;
 
+  const Scope* local_scope() const override;
+
   const platform::Place& GetPlace() const override { return place_; }
 
   void SetOutputHooks(const std::vector<HookFunc>& hookfuncs) override {

paddle/fluid/ir/phi_kernel_adaptor/phi_kernel_adaptor.h

@@ -56,7 +56,7 @@ class PhiKernelAdaptor {
   void run_kernel_prog(ir::Program* program) {
     auto block = program->block();
     std::unordered_map<ir::Value, std::string> name_map;
-    BuildScope(block, scope_, &name_map);
+    BuildScope(block, scope_, nullptr, &name_map);
     ir::IrContext* ctx = ir::IrContext::Instance();
 
     ctx->GetOrRegisterDialect<paddle::dialect::PaddleDialect>();
@@ -87,7 +87,7 @@ class PhiKernelAdaptor {
           phi::MetaTensor,
           phi::MetaTensor,
           paddle::small_vector<phi::MetaTensor, phi::kInputSmallVectorSize>,
-          false>((*it), name_map, scope_, op_yaml_info_parser, &ctx);
+          false>((*it), name_map, scope_, nullptr, op_yaml_info_parser, &ctx);
 
       infer_meta_impl->infer_meta_(&ctx);
 
@@ -107,7 +107,7 @@ class PhiKernelAdaptor {
                           phi::TensorBase*,
                           paddle::small_vector<const phi::TensorBase*>,
                           true>(
-          (*it), name_map, scope_, op_yaml_info_parser, &kernel_ctx);
+          (*it), name_map, scope_, nullptr, op_yaml_info_parser, &kernel_ctx);
       kernel_fn(&kernel_ctx);
 
       auto out_value = (*it)->result(0);

paddle/fluid/ir/phi_kernel_adaptor/phi_kernel_util.cc

@@ -43,114 +43,160 @@
 
 namespace ir {
 
-void BuildScope(ir::Block* block,
-                paddle::framework::Scope* scope,
-                std::unordered_map<ir::Value, std::string>* name_map) {
-  std::unordered_map<ir::Value, int> map_test;
-
-  int count = name_map->size();
-  for (auto it = block->begin(); it != block->end(); ++it) {
-    size_t input_num = (*it)->num_operands();
-    auto attr_map = (*it)->attributes();
-    std::string op_name = (*it)->name();
-    if (attr_map.count("op_name")) {
-      op_name = attr_map.at("op_name").dyn_cast<ir::StrAttribute>().data();
-    }
-    if (op_name == "pd.fetch") {
-      // fetch is a very special op, with no output
-      for (size_t i = 0; i < input_num; ++i) {
-        auto var = scope->Var("fetch");
-        auto fetch_list = var->GetMutable<paddle::framework::FetchList>();
-        int index =
-            (*it)->attributes().at("col").dyn_cast<ir::Int32Attribute>().data();
-        fetch_list->resize(index + 1);
-      }
-      continue;
+paddle::framework::Variable* CreateVar(ir::Value value,
+                                       std::string name,
+                                       paddle::framework::Scope* scope,
+                                       paddle::framework::Scope* local_scope) {
+  Operation* def_op = value.GetDefiningOp();
+  bool is_persisable = false;
+  if (def_op->attributes().count("is_persisable")) {
+    is_persisable = def_op->attributes()
+                        .at("is_persisable")
+                        .dyn_cast<ir::BoolAttribute>()
+                        .data();
+  }
+  if (is_persisable) {
+    const paddle::framework::Scope* ancestor_scope = scope;
+    while (ancestor_scope->parent()) {
+      ancestor_scope = ancestor_scope->parent();
     }
+    VLOG(6) << "Create var: " << name << " in scope " << ancestor_scope;
+    return const_cast<paddle::framework::Scope*>(ancestor_scope)->Var(name);
+  } else {
+    VLOG(6) << "Create var: " << name << " in scope " << local_scope;
+    return local_scope->Var(name);
+  }
+}
 
-    if (op_name == "builtin.set_parameter") {
-      auto param_name = (*it)
-                            ->attributes()
-                            .at("parameter_name")
-                            .dyn_cast<ir::StrAttribute>()
-                            .data();
-
-      auto in_ptr = (*it)->operand(0);
-      // change opreand name to param_name
-
-      auto orig_name = name_map->at(in_ptr);
-      (*name_map)[in_ptr] = param_name;
-      scope->Rename(orig_name, param_name);
-      continue;
+void HandleForSpecialOp(ir::Operation* op,
+                        paddle::framework::Scope* scope,
+                        paddle::framework::Scope* local_scope,
+                        std::unordered_map<ir::Value, std::string>* name_map,
+                        int& count) {  // NOLINT
+  std::string op_name = op->name();
+  if (op->attributes().count("op_name")) {
+    op_name =
+        op->attributes().at("op_name").dyn_cast<ir::StrAttribute>().data();
+  }
+  size_t input_num = op->num_operands();
+
+  if (op_name == "pd.fetch") {
+    // fetch is a very special op, with no output
+    VLOG(6) << "Handle for pd.fetch:";
+    for (size_t i = 0; i < input_num; ++i) {
+      auto var = scope->Var("fetch");
+      VLOG(6) << "Create var: fetch in scope " << scope;
+      auto fetch_list = var->GetMutable<paddle::framework::FetchList>();
+      int index =
+          op->attributes().at("col").dyn_cast<ir::Int32Attribute>().data();
+      fetch_list->resize(index + 1);
     }
+  }
 
-    if (op_name == "builtin.get_parameter") {
-      auto param_name = (*it)
-                            ->attributes()
-                            .at("parameter_name")
-                            .dyn_cast<ir::StrAttribute>()
-                            .data();
-
-      auto out_ptr = (*it)->result(0);
+  if (op_name == "pd.feed") {
+    VLOG(6) << "Handle for pd.feed:";
+    auto ptr = op->result(0);
+    std::string name = "inner_var_" + std::to_string(count++);
+    name_map->emplace(ptr, name);
+    auto var = CreateVar(ptr, name, scope, local_scope);
+    // TODO(phlrain): need to update here, support StringTensor
+    auto out_tensor = var->GetMutable<phi::DenseTensor>();
+
+    auto feed_var = scope->Var("feed");
+    VLOG(6) << "Create var: feed in scope " << scope;
+    int index =
+        op->attributes().at("col").dyn_cast<ir::Int32Attribute>().data();
+    auto feed_list = feed_var->Get<paddle::framework::FeedList>();
+    auto& in_tensor = (PADDLE_GET(phi::DenseTensor, feed_list.at(index)));
+    out_tensor->ShareDataWith(in_tensor);
+  }
 
-      name_map->emplace(out_ptr, param_name);
-      continue;
+  if (op_name == "builtin.combine") {
+    VLOG(6) << "Handle for builtin.combine:";
+    auto out_value = op->result(0);
+    std::string name;
+    if (name_map->find(out_value) != name_map->end()) {
+      name = name_map->at(out_value);
+    } else {
+      name = "inner_var_" + std::to_string(count++);
+      name_map->emplace(out_value, name);
     }
 
-    if (op_name == "pd.feed") {
-      auto ptr = (*it)->result(0);
-      std::string name = "inner_var_" + std::to_string(count++);
-      name_map->emplace(ptr, name);
-      auto var = scope->Var(name);
-      // TODO(phlrain): need to update here, support StringTensor
-      auto out_tensor = var->GetMutable<phi::DenseTensor>();
-
-      auto feed_var = scope->Var("feed");
-      int index =
-          (*it)->attributes().at("col").dyn_cast<ir::Int32Attribute>().data();
-      auto feed_list = feed_var->Get<paddle::framework::FeedList>();
-      auto& in_tensor = (PADDLE_GET(phi::DenseTensor, feed_list.at(index)));
+    auto var = CreateVar(out_value, name, scope, local_scope);
+    auto tensor_array = var->GetMutable<paddle::framework::TensorRefArray>();
 
-      out_tensor->ShareDataWith(in_tensor);
+    for (size_t i = 0; i < input_num; ++i) {
+      auto ptr = op->operand(i);
 
-      continue;
+      PADDLE_ENFORCE_EQ(
+          name_map->count(ptr),
+          true,
+          phi::errors::PreconditionNotMet("can not found input of combine op"));
+      tensor_array->emplace_back(
+          &(CreateVar(ptr, name_map->at(ptr), scope, local_scope)
+                ->Get<phi::DenseTensor>()));
     }
+  }
 
-    if (op_name == "builtin.combine") {
-      auto out_value = (*it)->result(0);
+  if (op_name == "builtin.set_parameter") {
+    VLOG(6) << "Handle for builtin.set_parameter:";
+    auto param_name = op->attributes()
+                          .at("parameter_name")
+                          .dyn_cast<ir::StrAttribute>()
+                          .data();
 
-      VLOG(5) << "process builtin combine";
-      std::string name;
-      if (name_map->find(out_value) != name_map->end()) {
-        name = name_map->at(out_value);
-      } else {
-        name = "inner_var_" + std::to_string(count++);
-        name_map->emplace(out_value, name);
-      }
+    auto in_ptr = op->operand(0);
+    // change opreand name to param_name
 
-      auto var = scope->Var(name);
-      auto tensor_array = var->GetMutable<paddle::framework::TensorRefArray>();
+    auto orig_name = name_map->at(in_ptr);
+    (*name_map)[in_ptr] = param_name;
+    scope->Rename(orig_name, param_name);
+  }
 
-      for (size_t i = 0; i < input_num; ++i) {
-        auto ptr = (*it)->operand(i);
+  if (op_name == "builtin.get_parameter") {
+    VLOG(6) << "Handle for builtin.get_parameter:";
+    auto param_name = op->attributes()
+                          .at("parameter_name")
+                          .dyn_cast<ir::StrAttribute>()
+                          .data();
+    auto out_ptr = op->result(0);
+    name_map->emplace(out_ptr, param_name);
+  }
+}
+
+void BuildScope(ir::Block* block,
+                paddle::framework::Scope* scope,
+                paddle::framework::Scope* local_scope,
+                std::unordered_map<ir::Value, std::string>* name_map) {
+  // NOTE(zhiqiu): if use local_scope (local_scope != nullptr), the persistable
+  // is created in scope , and other is created in local_scope.
+  auto inner_local_scope = local_scope != nullptr ? local_scope : scope;
+  VLOG(6) << "Build: scope [" << scope << "] inner_local_scope ["
+          << inner_local_scope << "]";
 
-        PADDLE_ENFORCE_EQ(name_map->count(ptr),
-                          true,
-                          phi::errors::PreconditionNotMet(
-                              "can not found input of combine op"));
+  // int count = name_map->size();
+  int count = name_map->size();
+  for (auto it = block->begin(); it != block->end(); ++it) {
+    ir::Operation* op = *it;
 
-        tensor_array->emplace_back(
-            &(scope->Var(name_map->at(ptr))->Get<phi::DenseTensor>()));
-      }
+    auto attr_map = op->attributes();
+    std::string op_name = op->name();
+    if (attr_map.count("op_name")) {
+      op_name = attr_map.at("op_name").dyn_cast<ir::StrAttribute>().data();
+    }
 
+    if (op_name == "pd.feed" || op_name == "pd.fetch" ||
+        op_name == "builtin.combine" || op_name == "builtin.set_parameter" ||
+        op_name == "builtin.get_parameter") {
+      VLOG(6) << "HandleForSpecialOp: " << op_name;
+      HandleForSpecialOp(op, scope, inner_local_scope, name_map, count);
       continue;
     }
 
-    // TODO(zhangbo): support builtin.slice
-
+    size_t input_num = op->num_operands();
     if (input_num > 0) {
       for (size_t i = 0; i < input_num; ++i) {
-        auto ptr = (*it)->operand(i);
+        auto ptr = op->operand(i);
         if (ptr) {
           PADDLE_ENFORCE_NE(
               name_map->find(ptr),
@@ -163,11 +209,10 @@ void BuildScope(ir::Block* block,
       }
     }
 
-    int out_num = (*it)->num_results();
-
+    int out_num = op->num_results();
     if (out_num > 0) {
       for (int i = 0; i < out_num; ++i) {
-        ir::Value ptr = (*it)->result(i);
+        ir::Value ptr = op->result(i);
         std::string name;
         if (name_map->find(ptr) != name_map->end()) {
           name = name_map->at(ptr);
@@ -175,7 +220,7 @@ void BuildScope(ir::Block* block,
           name = "inner_var_" + std::to_string(count++);
           name_map->emplace(ptr, name);
         }
-        auto var = scope->Var(name);
+        auto var = CreateVar(ptr, name, scope, inner_local_scope);
         // Only support DenseTensor or Vector<DenseTensor>
         if (!ptr.type()) {
           var->GetMutable<phi::DenseTensor>();
@@ -195,7 +240,7 @@ void BuildScope(ir::Block* block,
                     "Element of VectorType output only support "
                     "DenseTensorType"));
             std::string name_i = "inner_var_" + std::to_string(count++);
-            auto var_i = scope->Var(name_i);
+            auto var_i = CreateVar(ptr, name_i, scope, inner_local_scope);
             tensor_array->emplace_back(var_i->GetMutable<phi::DenseTensor>());
           }
         } else {

paddle/fluid/ir/phi_kernel_adaptor/phi_kernel_util.h

@@ -25,6 +25,7 @@
 #include "paddle/ir/core/utils.h"
 #include "paddle/phi/core/meta_tensor.h"
 
+#include "paddle/fluid/framework/new_executor/interpreter/execution_config.h"
 #include "paddle/fluid/framework/scope.h"
 #include "paddle/fluid/framework/tensor.h"
 #include "paddle/fluid/framework/variable.h"
@@ -39,9 +40,20 @@
 #include "glog/logging.h"
 
 namespace ir {
+paddle::framework::Variable* CreateVar(ir::Value value,
+                                       std::string name,
+                                       paddle::framework::Scope* scope,
+                                       paddle::framework::Scope* local_scope);
+
+void HandleForSpecialOp(ir::Operation* op,
+                        paddle::framework::Scope* scope,
+                        paddle::framework::Scope* local_scope,
+                        std::unordered_map<ir::Value, std::string>* name_map,
+                        int& count);  // NOLINT
 
 void BuildScope(ir::Block* block,
                 paddle::framework::Scope* scope,
+                paddle::framework::Scope* local_scope,
                 std::unordered_map<ir::Value, std::string>* name_map);
 
 template <typename Context,
@@ -53,10 +65,15 @@ void BuildPhiContext(
     ir::Operation* op,
     const std::unordered_map<ir::Value, std::string>& name_map,
     paddle::framework::Scope* scope,
+    paddle::framework::Scope* local_scope,
     const paddle::dialect::OpYamlInfoParser& op_yaml_info,
     Context* ctx,
     std::map<std::string, std::vector<int>>* input_map = nullptr,
     std::map<std::string, std::vector<int>>* output_map = nullptr) {
+  paddle::framework::Scope* inner_scope =
+      local_scope != nullptr ? local_scope : scope;
+  VLOG(6) << "BuildPhiContext in scope[" << scope << "] inner_scope["
+          << inner_scope << "]";
   // inputs include input and mutable attributes
 
   auto attr_map = op->attributes();
@@ -80,11 +97,10 @@ void BuildPhiContext(
     auto in_var_name = name_map.at(ptr);
     VLOG(6) << "ctx->EmplaceBackInput: " << t << "\t" << in_var_name;
 
-    PADDLE_ENFORCE_NOT_NULL(scope->FindLocalVar(in_var_name),
+    PADDLE_ENFORCE_NOT_NULL(inner_scope->FindLocalVar(in_var_name),
                             phi::errors::PreconditionNotMet(
                                 "can not find var[%s] in scope", in_var_name));
-
-    auto var = scope->Var(in_var_name);
+    auto var = inner_scope->FindVar(in_var_name);
     if (var->IsType<phi::DenseTensor>()) {
       const phi::TensorBase* tensor_in = &(var->Get<phi::DenseTensor>());
       ctx->EmplaceBackInput(InType(tensor_in));
@@ -123,12 +139,12 @@ void BuildPhiContext(
       auto& tensor_attr_type = op_yaml_info.TensorAttrTypeName(t);
       VLOG(6) << "ctx->EmplaceBack mutable attr: " << t << "\t" << in_var_name;
       if (tensor_attr_type == "paddle::dialect::IntArrayAttribute") {
-        phi::Attribute r1 =
-            phi::TensorRef(&(scope->Var(in_var_name)->Get<phi::DenseTensor>()));
+        phi::Attribute r1 = phi::TensorRef(
+            &(inner_scope->FindVar(in_var_name)->Get<phi::DenseTensor>()));
         ctx->EmplaceBackAttr(r1);
       } else if (tensor_attr_type == "paddle::dialect::ScalarAttribute") {
-        phi::Attribute r1 =
-            phi::TensorRef(&(scope->Var(in_var_name)->Get<phi::DenseTensor>()));
+        phi::Attribute r1 = phi::TensorRef(
+            &(inner_scope->FindVar(in_var_name)->Get<phi::DenseTensor>()));
 
         ctx->EmplaceBackAttr(r1);
       } else {
@@ -239,7 +255,7 @@ void BuildPhiContext(
       (op->attributes().at("op_name").dyn_cast<ir::StrAttribute>().data() ==
        "pd.fetch")) {
     // process fetch op
-    auto fetch_var = scope->Var("fetch");
+    auto fetch_var = inner_scope->FindVar("fetch");
     auto* fetch_list = fetch_var->GetMutable<paddle::framework::FetchList>();
     int index =
         op->attributes().at("col").dyn_cast<ir::Int32Attribute>().data();
@@ -251,7 +267,7 @@ void BuildPhiContext(
       auto name = name_map.at(out_ptr);
       if (out_ptr.type()) {
         ctx->EmplaceBackOutput(OutType(const_cast<phi::DenseTensor*>(
-            &(scope->Var(name)->Get<phi::DenseTensor>()))));
+            &(inner_scope->FindVar(name)->Get<phi::DenseTensor>()))));
       } else {
         phi::DenseTensor* ptr = nullptr;
         OutType out_ptr(ptr);

test/cpp/new_executor/standalone_executor_new_ir_test.cc

@@ -73,7 +73,11 @@ TEST(StandaloneExecutor, run) {
 
   test_core.Run({});
 
-  auto out_tensor = scope.Var("inner_var_2")->Get<phi::DenseTensor>();
+  auto out_tensor = test_core.local_scope() == nullptr
+                        ? scope.FindVar("inner_var_2")->Get<phi::DenseTensor>()
+                        : test_core.local_scope()
+                              ->FindVar("inner_var_2")
+                              ->Get<phi::DenseTensor>();
 
   bool res0 = simple_cmp(out_tensor.data<float>()[0], 2.0);
   bool res1 = simple_cmp(out_tensor.data<float>()[1], 2.0);
@@ -142,7 +146,11 @@ TEST(StandaloneExecutor, run_2) {
 
   test_core.Run({});
 
-  auto out_tensor = scope.Var("inner_var_10")->Get<phi::DenseTensor>();
+  auto out_tensor = test_core.local_scope() == nullptr
+                        ? scope.FindVar("inner_var_10")->Get<phi::DenseTensor>()
+                        : test_core.local_scope()
+                              ->FindVar("inner_var_10")
+                              ->Get<phi::DenseTensor>();
 
   bool res0 = simple_cmp(out_tensor.data<float>()[0], 1.80721);
   bool res1 = simple_cmp(out_tensor.data<float>()[1], 1.70047);
@@ -213,7 +221,11 @@ TEST(StandaloneExecutor, data_transfer) {
 
   test_core.Run({});
 
-  auto out_tensor = scope.Var("inner_var_9")->Get<phi::DenseTensor>();
+  auto out_tensor = test_core.local_scope() == nullptr
+                        ? scope.FindVar("inner_var_9")->Get<phi::DenseTensor>()
+                        : test_core.local_scope()
+                              ->FindVar("inner_var_9")
+                              ->Get<phi::DenseTensor>();
 
   auto& pool = phi::DeviceContextPool::Instance();
   phi::DenseTensor out;