[new-exec] fix stream analysis (#37161)

* fix revord_event

* refine class Instruction

* refine Instruction and InterpreterCore

* make instruction and operator_base consistent

* support NoNeedBufferVar in stream_analyzer

* fix place of event

* add vlog before continue

paddle/fluid/framework/new_executor/event_manager.cc

@@ -41,6 +41,16 @@ void RecordEvent(const Instruction& instruction, const platform::Place& place) {
   }
 }
 
+void RecordEvent(const Instruction& instruction) {
+  // If InterpreterCore in on CPUPlace, do nothing.
+  if (platform::is_cpu_place(instruction.DeviceContext().GetPlace())) return;
+
+  for (auto& event : instruction.OutputEvents()) {
+    VLOG(3) << "Record event in out_var_id: " << event.var_id_;
+    event.event_->Record(&instruction.DeviceContext());
+  }
+}
+
 }  // namespace interpreter
 }  // namespace framework
 }  // namespace paddle

paddle/fluid/framework/new_executor/event_manager.h

@@ -20,6 +20,8 @@ namespace framework {
 namespace interpreter {
 void RecordEvent(const Instruction& instruction, const platform::Place& place);
 
+void RecordEvent(const Instruction& instruction);
+
 void WaitEvent(const Instruction& instruction, const platform::Place& place);
 
 }  // namespace interpreter

paddle/fluid/framework/new_executor/interpretercore.cc

@@ -77,20 +77,22 @@ paddle::framework::FetchList InterpreterCore::Run(
   return *(fetch_var->GetMutable<framework::FetchList>());
 }
 
-void InterpreterCore::Convert() {
+void InterpreterCore::Convert(
+    std::vector<paddle::framework::OpFuncNode>* op_func_nodes) {
   auto& vec_meta_info = global_scope_->MutableVecMetaInfo();
   auto var_nums = global_scope_->VarSize();
   input_var2op_info_.resize(var_nums);
+  auto nodes = *op_func_nodes;
 
-  auto op_nums = vec_func_list_.size();
+  auto op_nums = nodes.size();
   vec_instruction_.reserve(op_nums);
   dependecy_count_.resize(op_nums);
 
   for (size_t op_idx = 0; op_idx < op_nums; ++op_idx) {
-    auto& op_func_node = vec_func_list_[op_idx];
+    auto& op_func_node = nodes[op_idx];
     auto* dev_ctx_ = stream_analyzer_.ParseDeviceContext(op_func_node);
 
-    vec_instruction_.emplace_back(op_idx, op_func_node, *dev_ctx_);
+    vec_instruction_.emplace_back(op_idx, std::move(op_func_node), *dev_ctx_);
     auto& instr = vec_instruction_.back();
 
     OpInOutInfo info;
@@ -104,7 +106,7 @@ void InterpreterCore::Convert() {
         input_var2op_info_.at(id).push_back(op_idx);
         // var can be gc-ed
         if (!info.IsBuilt()) {
-          info.Build(op_func_node.operator_base_);
+          info.Build(op_func_node.operator_base_.get());
         }
         auto* var_desc = global_scope_->VarDesc(id);
         if (var_desc) {
@@ -522,11 +524,12 @@ void InterpreterCore::Prepare(
   if (!is_build_) {
     paddle::framework::interpreter::build_variable_scope(block_, global_scope_);
     FeedInput();
+    std::vector<paddle::framework::OpFuncNode> op_func_nodes;
     paddle::framework::interpreter::build_op_func_list(
-        place_, block_, &vec_func_list_, global_scope_);
+        place_, block_, &op_func_nodes, global_scope_);
     is_build_ = true;
     // convert vec func_list to graph
-    Convert();
+    Convert(&op_func_nodes);
   }
   // NOTE: Because feed_tensor will be GC after
   // paddle::framework::build_op_func_list, so we should

paddle/fluid/framework/new_executor/interpretercore.h

@@ -54,7 +54,7 @@ class InterpreterCore {
       const std::vector<framework::LoDTensor>& feed_tensors);
 
  private:
-  void Convert();
+  void Convert(std::vector<paddle::framework::OpFuncNode>* op_func_nodes);
 
   void BuildAndCacheInstructionCtx(Instruction* instr_node);
 
@@ -84,7 +84,6 @@ class InterpreterCore {
   const BlockDesc& block_;       // not owned
   VariableScope* global_scope_;  // not owned
 
-  std::vector<paddle::framework::OpFuncNode> vec_func_list_;
   std::vector<Instruction> vec_instruction_;  // deconstruct before OpFuncNode
 
   std::vector<size_t> dependecy_count_;

paddle/fluid/framework/new_executor/interpretercore_util.cc

@@ -64,11 +64,12 @@ bool var_can_be_deleted(const std::string& name, const BlockDesc& block) {
 
 std::unordered_map<const paddle::framework::OperatorBase*,
                    std::vector<std::string>>
-get_unused_vars(const BlockDesc& block, const std::vector<OperatorBase*>& ops) {
+get_unused_vars(const BlockDesc& block,
+                const std::vector<std::shared_ptr<OperatorBase>>& ops) {
   std::unordered_map<std::string, size_t> var_op_idx_map;
 
   for (size_t i = 0; i < ops.size(); ++i) {
-    auto* op = ops[i];
+    const auto& op = ops[i];
 
     OpInOutInfo info;
     for (auto& name_pair : op->Inputs()) {
@@ -79,7 +80,7 @@ get_unused_vars(const BlockDesc& block, const std::vector<OperatorBase*>& ops) {
 
         // var can be gc-ed
         if (!info.IsBuilt()) {
-          info.Build(op);
+          info.Build(op.get());
         }
 
         if (info.IsInArgBufferNeeded(name)) {
@@ -107,7 +108,8 @@ get_unused_vars(const BlockDesc& block, const std::vector<OperatorBase*>& ops) {
   for (auto& name_op_idx_pair : var_op_idx_map) {
     auto& name = name_op_idx_pair.first;
     size_t op_idx = name_op_idx_pair.second;
-    result[ops[op_idx]].emplace_back(name);
+
+    result[ops[op_idx].get()].emplace_back(name);
   }
   return result;
 }
@@ -150,8 +152,8 @@ void build_variable_scope(const framework::BlockDesc& block,
   }
 }
 
-std::vector<OperatorBase*> create_all_ops(const framework::BlockDesc& block) {
-  std::vector<OperatorBase*> ops;
+void create_all_ops(const framework::BlockDesc& block,
+                    std::vector<std::shared_ptr<OperatorBase>>* ops) {
   for (auto& op : block.AllOps()) {
     VLOG(3) << "CreateOp from : " << op->Type();
 
@@ -166,9 +168,8 @@ std::vector<OperatorBase*> create_all_ops(const framework::BlockDesc& block) {
     }
     auto op_base =
         info.Creator()(op->Type(), inputs_names, outputs_names, op_attr_map);
-    ops.push_back(op_base);
+    ops->emplace_back(std::shared_ptr<OperatorBase>(op_base));
   }
-  return ops;
 }
 
 std::tuple<VariableValueMap, VariableIdMap> build_variable_map(
@@ -234,7 +235,8 @@ void apply_device_guard(const OperatorBase* op_base,
 }
 
 void deal_operator_base(const platform::Place& place,
-                        const VariableScope* var_scope, OperatorBase* op_base,
+                        const VariableScope* var_scope,
+                        std::shared_ptr<OperatorBase> op_base,
                         OpFuncNode* op_func_node) {
   platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
   auto* dev_ctx = pool.Get(place);
@@ -321,8 +323,9 @@ std::tuple<std::string, OpFuncNode> apply_place_transform_for_var(
                              var_name, kernel_type_for_var.place_, new_var_name,
                              expected_kernel_key.place_);
   auto& copy_info = OpInfoMap::Instance().Get(memcpy_op_type);
-  auto copy_op =
-      copy_info.Creator()(memcpy_op_type, copy_in_map, copy_out_map, attr_map);
+  auto copy_op = std::shared_ptr<OperatorBase>(
+      copy_info.Creator()(memcpy_op_type, copy_in_map, copy_out_map, attr_map));
+
   OpFuncNode copy_op_func_node;
   copy_op_func_node.input_index = copy_ins_name2id;
   copy_op_func_node.output_index = copy_out_name2id;
@@ -330,10 +333,10 @@ std::tuple<std::string, OpFuncNode> apply_place_transform_for_var(
   RuntimeContext copy_runtime_context({}, {});
   copy_runtime_context.inputs.swap(copy_ins_value_map);
   copy_runtime_context.outputs.swap(copy_outs_value_map);
-  InterpretercoreInferShapeContext copy_infer_shape_ctx(*copy_op,
+  InterpretercoreInferShapeContext copy_infer_shape_ctx(*copy_op.get(),
                                                         copy_runtime_context);
-  static_cast<const framework::OperatorWithKernel*>(copy_op)->InferShape(
-      &copy_infer_shape_ctx);
+  static_cast<const framework::OperatorWithKernel*>(copy_op.get())
+      ->InferShape(&copy_infer_shape_ctx);
 
   auto kernels_iter = all_op_kernels.find(memcpy_op_type);
   PADDLE_ENFORCE_NE(kernels_iter, all_op_kernels.end(),
@@ -347,7 +350,7 @@ std::tuple<std::string, OpFuncNode> apply_place_transform_for_var(
   auto copy_exec_ctx =
       ExecutionContext(*copy_op, scope, *dev_ctx, copy_runtime_context);
   auto copy_expected_kernel_key =
-      dynamic_cast<const framework::OperatorWithKernel*>(copy_op)
+      dynamic_cast<const framework::OperatorWithKernel*>(copy_op.get())
           ->GetExpectedKernelType(copy_exec_ctx);
   auto kernel_iter = kernels.find(copy_expected_kernel_key);
   copy_op_func_node.kernel_func_ = OpKernelComputeFunc(kernel_iter->second);
@@ -362,19 +365,20 @@ std::tuple<std::string, OpFuncNode> apply_place_transform_for_var(
   return std::make_pair(new_var_name, copy_op_func_node);
 }
 
-std::vector<OpFuncNode> apply_data_transform(
-    const OpKernelType& expected_kernel_key, const platform::Place& place,
-    VariableValueMap* ins_map_temp, VariableScope* var_scope,
-    OpFuncNode* op_func_node) {
-  auto& op_base = op_func_node->operator_base_;
+void apply_data_transform(const OpKernelType& expected_kernel_key,
+                          const platform::Place& place,
+                          VariableValueMap* ins_map_temp,
+                          VariableScope* var_scope, OpFuncNode* op_func_node,
+                          std::vector<OpFuncNode>* copy_func_nodes) {
+  auto op_base = op_func_node->operator_base_.get();
   PADDLE_ENFORCE_NOT_NULL(op_base, platform::errors::PreconditionNotMet(
                                        "op_base is null, please pass a valid "
                                        "op_base in apply_data_transform."));
-  auto inputs_names = op_base->Inputs();
+
+  VariableNameMap new_ins(op_base->Inputs());
 
   std::unordered_set<int>
       no_data_transform_index;  // record the no need transform variable index.
-  std::vector<OpFuncNode> copy_func_nodes;  // return all the copy opfuncnode.
 
   for (auto& var_name_item : *ins_map_temp) {
     for (size_t i = 0; i < var_name_item.second.size(); ++i) {
@@ -382,7 +386,7 @@ std::vector<OpFuncNode> apply_data_transform(
       if (!(var->IsType<LoDTensor>() || var->IsType<SelectedRows>())) {
         continue;
       }
-      auto& var_name = inputs_names[var_name_item.first].at(i);
+      auto& var_name = new_ins[var_name_item.first].at(i);
       auto tensor_in = GetLoDTensorOrSelectedRowsValueFromVar(*var);
       if (!tensor_in->IsInitialized()) {
         continue;
@@ -404,8 +408,9 @@ std::vector<OpFuncNode> apply_data_transform(
                 var_name_item.first, *op_func_node, var, var_scope);
         op_func_node->input_index[var_name_item.first][i] =
             var_scope->VarId(new_var_name);
-        copy_func_nodes.push_back(copy_op_func_node);
+        copy_func_nodes->emplace_back(copy_op_func_node);
         var_name_item.second[i] = var_scope->Var(new_var_name);
+        new_ins[var_name_item.first][i] = new_var_name;
       } else if (need_dtype_transform_for_var(kernel_type_for_var,
                                               expected_kernel_key)) {
         // TODO(@xiongkun) add dtype judgement here
@@ -421,8 +426,14 @@ std::vector<OpFuncNode> apply_data_transform(
       }
     }
   }
+
+  // NOTE(zhiqiu): UPDATE the corresponding OeratorBase to make it consistent
+  // with instruction
+  // hot fix, it is not good design here
+  op_func_node->operator_base_ =
+      std::shared_ptr<OperatorBase>(framework::OpRegistry::CreateOp(
+          op_base->Type(), new_ins, op_base->Outputs(), op_base->Attrs()));
   op_func_node->no_data_transform_index = std::move(no_data_transform_index);
-  return copy_func_nodes;
 }
 
 void build_op_func_list(const platform::Place& place,
@@ -430,16 +441,16 @@ void build_op_func_list(const platform::Place& place,
                         std::vector<OpFuncNode>* vec_func_list,
                         VariableScope* var_scope) {
   auto& all_op_kernels = OperatorWithKernel::AllOpKernels();
-
+  std::vector<std::shared_ptr<OperatorBase>>
+      ops;  // its elements will be moved to vec_func_list
   // Step 1: create all ops for current block.
-  auto ops = create_all_ops(block);
+  create_all_ops(block, &ops);
   auto unused_var_map = get_unused_vars(block, ops);
 
-  size_t ops_index = 0;
-  for (auto& op : block.AllOps()) {
+  for (size_t i = 0; i < ops.size(); ++i) {
+    auto op = ops[i].get();
     VLOG(6) << "Build OpFuncNode from : " << op->Type();
 
-    auto op_base = ops[ops_index++];
     auto inputs_names = op->Inputs();
     auto outputs_names = op->Outputs();
 
@@ -466,20 +477,18 @@ void build_op_func_list(const platform::Place& place,
     op_func_node.input_index = ins_name2id;
     op_func_node.output_index = outs_name2id;
 
-    if (dynamic_cast<const framework::OperatorWithKernel*>(op_base) ==
-        nullptr) {
+    if (dynamic_cast<const framework::OperatorWithKernel*>(op) == nullptr) {
       // op is not a operatorwithkernel, so direcly run OperatorBase::Run()
-      deal_operator_base(place, var_scope, op_base, &op_func_node);
+      deal_operator_base(place, var_scope, ops[i], &op_func_node);
     } else {
       // construct RuntimeContext and analysis KernelType
       RuntimeContext runtime_context({}, {});
       runtime_context.inputs.swap(ins_map);
       runtime_context.outputs.swap(outs_map);
-      InterpretercoreInferShapeContext infer_shape_ctx(*op_base,
-                                                       runtime_context);
+      InterpretercoreInferShapeContext infer_shape_ctx(*op, runtime_context);
       // TODO(Aurelius84): In case of control flow ops, they are NOT inheritted
       // from OperatorWithKernel.
-      static_cast<const framework::OperatorWithKernel*>(op_base)->InferShape(
+      static_cast<const framework::OperatorWithKernel*>(op)->InferShape(
           &infer_shape_ctx);
       auto kernels_iter = all_op_kernels.find(op->Type());
       PADDLE_ENFORCE_NE(
@@ -495,13 +504,13 @@ void build_op_func_list(const platform::Place& place,
       auto* dev_ctx = pool.Get(place);
       Scope scope;
       auto expected_kernel_key =
-          dynamic_cast<const framework::OperatorWithKernel*>(op_base)
+          dynamic_cast<const framework::OperatorWithKernel*>(op)
               ->GetExpectedKernelType(
-                  ExecutionContext(*op_base, scope, *dev_ctx, runtime_context));
+                  ExecutionContext(*op, scope, *dev_ctx, runtime_context));
 
       // consider device_guard()
       apply_device_guard(
-          op_base, place,
+          op, place,
           &expected_kernel_key);  // change device by the device_guard()
       VLOG(3) << "expected_kernel_key : " << expected_kernel_key;
 
@@ -510,14 +519,14 @@ void build_op_func_list(const platform::Place& place,
       std::vector<OpFuncNode> copy_op_to_insert;
       // NOTE(xiongkun03): assign op_base here to reduce parameter number of
       // apply_data_transform.
-      op_func_node.operator_base_ = op_base;
-      copy_op_to_insert = apply_data_transform(
-          expected_kernel_key, place, &ins_map_temp, var_scope, &op_func_node);
+      op_func_node.operator_base_ = ops[i];
+      apply_data_transform(expected_kernel_key, place, &ins_map_temp, var_scope,
+                           &op_func_node, &copy_op_to_insert);
       for (auto& item : copy_op_to_insert) {
         vec_func_list->push_back(item);
       }
       // step 4. Run op kernel
-      VLOG(3) << op_base->Type()
+      VLOG(3) << op->Type()
               << " : expected_kernel_key : " << expected_kernel_key;
 
       if (platform::is_gpu_place(expected_kernel_key.place_)) {
@@ -533,8 +542,7 @@ void build_op_func_list(const platform::Place& place,
       }
       op_func_node.dev_ctx_ = dev_ctx;
 
-      auto exec_ctx =
-          ExecutionContext(*op_base, scope, *dev_ctx, runtime_context);
+      auto exec_ctx = ExecutionContext(*op, scope, *dev_ctx, runtime_context);
 
       auto kernel_iter = kernels.find(expected_kernel_key);
       PADDLE_ENFORCE_NE(
@@ -547,9 +555,9 @@ void build_op_func_list(const platform::Place& place,
       op_func_node.kernel_func_(exec_ctx);
     }
 
-    vec_func_list->push_back(op_func_node);
+    vec_func_list->emplace_back(op_func_node);
     // gc---------------------------------------------------------------------------
-    auto iter = unused_var_map.find(op_base);
+    auto iter = unused_var_map.find(op);
     if (iter == unused_var_map.end()) {
       continue;
     }
@@ -586,7 +594,7 @@ void build_op_func_list(const platform::Place& place,
 
     delete garbages;  // free mem
 
-    VLOG(3) << "run " << op_base->Type() << " done.";
+    VLOG(3) << "run " << op->Type() << " done.";
   }
 }
 

paddle/fluid/framework/new_executor/new_executor_defs.cc

@@ -629,5 +629,109 @@ void VariableScopeListener::onCreateScope(Scope* Scope) {}
 void VariableScopeListener::onDeleteScope(Scope* Scope) {}
 void VariableScopeListener::onClear() {}
 
+Instruction::Instruction(size_t id, OpFuncNode&& op_func_node,
+                         const platform::DeviceContext& dev_ctx)
+    : id_(id), op_func_node_(op_func_node), dev_ctx_(dev_ctx) {
+  PADDLE_ENFORCE_GE(id, 0, platform::errors::PreconditionNotMet(
+                               "Required id >= 0, but received id = %d", id));
+}
+
+size_t Instruction::Id() const { return id_; }
+
+const std::map<std::string, std::vector<int>>& Instruction::Inputs() const {
+  return op_func_node_.input_index;
+}
+
+const std::map<std::string, std::vector<int>>& Instruction::Outputs() const {
+  return op_func_node_.output_index;
+}
+
+const std::unordered_set<int>& Instruction::NoDataTransformVars() const {
+  return op_func_node_.no_data_transform_index;
+}
+
+OpKernelComputeFunc Instruction::KernelFunc() const {
+  return op_func_node_.kernel_func_;
+}
+
+OpFuncType Instruction::KernelType() const { return op_func_node_.type_; }
+
+OperatorBase* Instruction::OpBase() const {
+  auto op_base = op_func_node_.operator_base_;
+  PADDLE_ENFORCE_NOT_NULL(op_base, platform::errors::PreconditionNotMet(
+                                       "op_base shall not be nullptr."));
+  return op_base.get();
+}
+
+NextInstruction& Instruction::NextInstructions() { return next_instruction_; }
+
+const NextInstruction& Instruction::NextInstructions() const {
+  return next_instruction_;
+}
+
+void Instruction::AddGCCheckVar(size_t id) { gc_check_var_list_.push_back(id); }
+
+const std::vector<size_t>& Instruction::GCCheckVars() const {
+  return gc_check_var_list_;
+}
+
+void Instruction::ResetContext(const VariableValueMap& in_vars,
+                               const VariableValueMap& out_vars) {
+  runtime_ctx_.reset(new RuntimeContext(in_vars, out_vars));
+  infershape_ctx_.reset(
+      new InterpretercoreInferShapeContext(*OpBase(), *runtime_ctx_.get()));
+  // NOTE: Because execution_ctx_ is constructed by `scope&`, so we fake an
+  // empty here to avoid illegal local reference.
+  static framework::Scope scope_;
+  execution_ctx_.reset(
+      new ExecutionContext(*OpBase(), scope_, dev_ctx_, *runtime_ctx_.get()));
+}
+
+std::shared_ptr<RuntimeContext> Instruction::InnerRuntimeContext() const {
+  return runtime_ctx_;
+}
+
+std::shared_ptr<InterpretercoreInferShapeContext>
+Instruction::InnerInferShapeContext() const {
+  return infershape_ctx_;
+}
+
+std::shared_ptr<ExecutionContext> Instruction::InnerExecutionContext() const {
+  return execution_ctx_;
+}
+
+const platform::DeviceContext& Instruction::DeviceContext() const {
+  return dev_ctx_;
+}
+
+const std::vector<std::pair<Variable*, Variable*>>& Instruction::InplaceInfo()
+    const {
+  return vec_inplace_in_to_out_;
+}
+
+void Instruction::AddInplace(Variable* in, Variable* out) {
+  vec_inplace_in_to_out_.emplace_back(in, out);
+}
+
+const std::vector<EventInter>& Instruction::InputEvents() const {
+  return intput_events_;
+}
+
+const std::vector<EventInter>& Instruction::OutputEvents() const {
+  return output_events_;
+}
+
+void Instruction::AddInputEvent(size_t var_id,
+                                std::shared_ptr<platform::DeviceEvent> event,
+                                platform::DeviceType waiter_type) {
+  intput_events_.emplace_back(var_id, event, waiter_type);
+}
+
+void Instruction::AddOutputEvent(size_t var_id,
+                                 std::shared_ptr<platform::DeviceEvent> event,
+                                 platform::DeviceType waiter_type) {
+  output_events_.emplace_back(var_id, event, waiter_type);
+}
+
 }  // namespace framework
 }  // namespace paddle

paddle/fluid/framework/new_executor/new_executor_defs.h

@@ -271,7 +271,8 @@ enum class OpFuncType {
 class RuntimeInferShapeContext;
 
 struct OpFuncNode {
-  OperatorBase* operator_base_;
+  // TODO(zhiqiu): Better make it unique_ptr
+  std::shared_ptr<OperatorBase> operator_base_;
   std::map<std::string, std::vector<int>> input_index;
   std::map<std::string, std::vector<int>> output_index;
   std::unordered_set<int> no_data_transform_index;
@@ -283,100 +284,62 @@ struct OpFuncNode {
 
 class Instruction {
  public:
-  Instruction(size_t id, const OpFuncNode& op_func_node,
-              const platform::DeviceContext& dev_ctx)
-      : id_(id), op_func_node_(op_func_node), dev_ctx_(dev_ctx) {
-    PADDLE_ENFORCE_GE(id, 0, platform::errors::PreconditionNotMet(
-                                 "Required id >= 0, but received id = %d", id));
-  }
+  Instruction(size_t id, OpFuncNode&& op_func_node,
+              const platform::DeviceContext& dev_ctx);
 
-  size_t Id() const { return id_; }
+  size_t Id() const;
 
-  const std::map<std::string, std::vector<int>>& Inputs() const {
-    return op_func_node_.input_index;
-  }
+  const std::map<std::string, std::vector<int>>& Inputs() const;
 
-  const std::map<std::string, std::vector<int>>& Outputs() const {
-    return op_func_node_.output_index;
-  }
+  const std::map<std::string, std::vector<int>>& Outputs() const;
 
-  const std::unordered_set<int>& NoDataTransformVars() const {
-    return op_func_node_.no_data_transform_index;
-  }
+  const std::unordered_set<int>& NoDataTransformVars() const;
 
-  OpKernelComputeFunc KernelFunc() const { return op_func_node_.kernel_func_; }
+  OpKernelComputeFunc KernelFunc() const;
 
-  OpFuncType KernelType() const { return op_func_node_.type_; }
+  OpFuncType KernelType() const;
 
-  OperatorBase* OpBase() const {
-    auto* op_base = op_func_node_.operator_base_;
-    PADDLE_ENFORCE_NOT_NULL(op_base, platform::errors::PreconditionNotMet(
-                                         "op_base shall not be nullptr."));
-    return op_base;
-  }
+  OperatorBase* OpBase() const;
 
-  NextInstruction& NextInstructions() { return next_instruction_; }
+  NextInstruction& NextInstructions();
 
-  const NextInstruction& NextInstructions() const { return next_instruction_; }
+  const NextInstruction& NextInstructions() const;
 
-  void AddGCCheckVar(size_t id) { gc_check_var_list_.push_back(id); }
+  void AddGCCheckVar(size_t id);
 
-  const std::vector<size_t>& GCCheckVars() const { return gc_check_var_list_; }
+  const std::vector<size_t>& GCCheckVars() const;
 
   void ResetContext(const VariableValueMap& in_vars,
-                    const VariableValueMap& out_vars) {
-    runtime_ctx_.reset(new RuntimeContext(in_vars, out_vars));
-    infershape_ctx_.reset(
-        new InterpretercoreInferShapeContext(*OpBase(), *runtime_ctx_.get()));
-    // NOTE: Because execution_ctx_ is constructed by `scope&`, so we fake an
-    // empty here to avoid illegal local reference.
-    static framework::Scope scope_;
-    execution_ctx_.reset(
-        new ExecutionContext(*OpBase(), scope_, dev_ctx_, *runtime_ctx_.get()));
-  }
+                    const VariableValueMap& out_vars);
 
-  std::shared_ptr<RuntimeContext> InnerRuntimeContext() const {
-    return runtime_ctx_;
-  }
+  std::shared_ptr<RuntimeContext> InnerRuntimeContext() const;
 
   std::shared_ptr<InterpretercoreInferShapeContext> InnerInferShapeContext()
-      const {
-    return infershape_ctx_;
-  }
+      const;
 
-  std::shared_ptr<ExecutionContext> InnerExecutionContext() const {
-    return execution_ctx_;
-  }
+  std::shared_ptr<ExecutionContext> InnerExecutionContext() const;
 
-  const platform::DeviceContext& DeviceContext() const { return dev_ctx_; }
+  const platform::DeviceContext& DeviceContext() const;
 
-  const std::vector<std::pair<Variable*, Variable*>>& InplaceInfo() const {
-    return vec_inplace_in_to_out_;
-  }
+  const std::vector<std::pair<Variable*, Variable*>>& InplaceInfo() const;
 
-  void AddInplace(Variable* in, Variable* out) {
-    vec_inplace_in_to_out_.emplace_back(in, out);
-  }
+  void AddInplace(Variable* in, Variable* out);
 
-  const std::vector<EventInter>& InputEvents() const { return intput_events_; }
+  const std::vector<EventInter>& InputEvents() const;
 
-  const std::vector<EventInter>& OutputEvents() const { return output_events_; }
+  const std::vector<EventInter>& OutputEvents() const;
 
   void AddInputEvent(size_t var_id,
                      std::shared_ptr<platform::DeviceEvent> event,
-                     platform::DeviceType waiter_type) {
-    intput_events_.emplace_back(var_id, event, waiter_type);
-  }
+                     platform::DeviceType waiter_type);
 
   void AddOutputEvent(size_t var_id,
                       std::shared_ptr<platform::DeviceEvent> event,
-                      platform::DeviceType waiter_type) {
-    output_events_.emplace_back(var_id, event, waiter_type);
-  }
+                      platform::DeviceType waiter_type);
 
  private:
   size_t id_;
-  const OpFuncNode& op_func_node_;          // not owned
+  OpFuncNode op_func_node_;
   const platform::DeviceContext& dev_ctx_;  // not owned
 
   std::shared_ptr<RuntimeContext> runtime_ctx_;
@@ -403,6 +366,11 @@ static bool IsMemcpyH2D(const Instruction& instr) {
 static bool IsMemcpyD2H(const Instruction& instr) {
   return instr.OpBase()->Type() == kMemcpyD2H;
 }
+
+static bool IsCpuOp(const Instruction& instr) {
+  return platform::is_cpu_place(instr.DeviceContext().GetPlace());
+}
+
 }  // namespace interpreter
 
 }  // namespace framework

paddle/fluid/framework/new_executor/stream_analyzer.cc

@@ -27,33 +27,60 @@ namespace framework {
  *
  * For example: matmul(gpu) -> out_var -> memcpy_d2h
  * out_var should be associated with an event.
+ *
+ * NOTE(zhiqiu): There are two special case that no event is needed:
+ *  1. the variable is marked as NoDataTransformVar
+ *  2. the variable is marked as NoNeedDataBuffer
  */
-std::vector<size_t> StreamAnalyzer::ParseEventVarIds(
+std::vector<size_t> StreamAnalyzer::GetNeedEventVarIds(
     const Instruction& cur_instr, const Instruction& next_instr) {
   std::unordered_set<size_t> unique_var_ids;
   for (auto& item : cur_instr.Outputs()) {
     unique_var_ids.insert(item.second.begin(), item.second.end());
   }
 
-  std::vector<size_t> new_event_var_ids;
+  auto is_no_need_buffer = [&next_instr](std::string name) {
+    auto* op = next_instr.OpBase();
+    auto& inferer = op->Info().NoNeedBufferVarsInferer();
+    if (inferer) {
+      auto no_need_buffer_ins =
+          inferer(op->Inputs(), op->Outputs(), op->Attrs());
+      return no_need_buffer_ins.count(name) != 0;
+    }
+    return false;
+  };
+
+  std::vector<size_t> need_event_var_ids;
   for (auto& item : next_instr.Inputs()) {
     for (auto var_id : item.second) {
-      if (unique_var_ids.count(var_id) > 0 &&
-          next_instr.NoDataTransformVars().count(var_id) == 0) {
-        new_event_var_ids.push_back(var_id);
+      if (unique_var_ids.count(var_id) > 0) {
+        if (next_instr.NoDataTransformVars().count(var_id)) {
+          VLOG(4) << "Skip inserting event at variable " << item.first
+                  << " of operator " << next_instr.OpBase()->Type()
+                  << " since it is NoDataTransform";
+          continue;
         }
+        if (is_no_need_buffer(item.first)) {
+          VLOG(4) << "Skip inserting event at variable " << item.first
+                  << " of operator " << next_instr.OpBase()->Type()
+                  << " since it is NoNeedBufferVar";
+          continue;
         }
+
+        need_event_var_ids.push_back(var_id);
       }
-  return new_event_var_ids;
+    }
+  }
+  return need_event_var_ids;
 }
 
-void StreamAnalyzer::AssociateInputWithEvents(
+void StreamAnalyzer::ConstructEventForVar(
     const std::vector<size_t>& new_event_var_id, Instruction* next_instr,
-    platform::DeviceType waiter_type) {
+    platform::DeviceType waiter_type, const platform::Place& place) {
   for (auto var_id : new_event_var_id) {
     if (var_id2event_.count(var_id) == 0) {
       auto device_event = std::make_shared<platform::DeviceEvent>(
-          place_, platform::GenerateDeviceEventFlag());
+          place, platform::GenerateDeviceEventFlag());
       var_id2event_.emplace(var_id, std::move(device_event));
     }
     // Add events for next_instr.inputs
@@ -69,21 +96,27 @@ void StreamAnalyzer::Schedule(const std::vector<size_t>& downstream_ops,
   std::vector<size_t> event_var_ids;
   for (auto next_op_id : downstream_ops) {
     auto& next_instr = instructions->at(next_op_id);
-
     if (IsDirectRun(cur_instr, next_instr)) {
+      VLOG(4) << "DirectRun: " << cur_instr.OpBase()->Type() << "->"
+              << next_instr.OpBase()->Type();
       next_instruction.AddDirectRun(next_op_id);
     } else {
       // Always insert events between different stream
-      auto new_event_var_ids = ParseEventVarIds(cur_instr, next_instr);
-      event_var_ids.insert(event_var_ids.end(), new_event_var_ids.begin(),
-                           new_event_var_ids.end());
+      auto need_event_var_ids = GetNeedEventVarIds(cur_instr, next_instr);
+      event_var_ids.insert(event_var_ids.end(), need_event_var_ids.begin(),
+                           need_event_var_ids.end());
 
       auto waiter_type = GetWaiterType(next_instr);
-      AssociateInputWithEvents(new_event_var_ids, &next_instr, waiter_type);
+      ConstructEventForVar(need_event_var_ids, &next_instr, waiter_type,
+                           cur_instr.DeviceContext().GetPlace());
 
       if (waiter_type == platform::kCPU) {  // GPU -> CPU
+        VLOG(4) << "SyncRun: " << cur_instr.OpBase()->Type() << "->"
+                << next_instr.OpBase()->Type();
         next_instruction.AddSyncRun(next_op_id);
       } else {  // GPU -> GPU(different stream)
+        VLOG(4) << "EventRun: " << cur_instr.OpBase()->Type() << "->"
+                << next_instr.OpBase()->Type();
         next_instruction.ADDEventRun(next_op_id);
       }
     }
@@ -116,12 +149,15 @@ platform::DeviceContext* StreamAnalyzer::ParseDeviceContext(
  * NOTE(dev): The following cases are considered as directly run:
  *
  *  1. with same dev_ctx_, such as: CPU -> CPU, GPU -> GPU
- *  2. D2H -> CPU
- *  3. CPU -> H2D
+ *  2. CPU -> any (it is possible: CPU op->VAR->GPU op, when var is no need
+ * buffer or no need data transform)
+ *  3. D2H -> CPU
+ *  4. CPU -> H2D
  */
 bool StreamAnalyzer::IsDirectRun(Instruction& cur_instr,
                                  const Instruction& next_instr) {
   return (&cur_instr.DeviceContext() == &next_instr.DeviceContext() ||
+          interpreter::IsCpuOp(cur_instr) ||
           interpreter::IsMemcpyD2H(cur_instr) ||
           interpreter::IsMemcpyH2D(next_instr));
 }

paddle/fluid/framework/new_executor/stream_analyzer.h

@@ -35,12 +35,13 @@ class StreamAnalyzer {
   platform::DeviceContext* ParseDeviceContext(const OpFuncNode& op_func_node);
 
  private:
-  std::vector<size_t> ParseEventVarIds(const Instruction& cur_instr,
+  std::vector<size_t> GetNeedEventVarIds(const Instruction& cur_instr,
                                          const Instruction& next_instr);
 
-  void AssociateInputWithEvents(const std::vector<size_t>& new_event_var_id,
+  void ConstructEventForVar(const std::vector<size_t>& new_event_var_id,
                             Instruction* next_instr,
-                                platform::DeviceType waiter_type);
+                            platform::DeviceType waiter_type,
+                            const platform::Place& place);
 
   bool IsDirectRun(Instruction& cur_instr,  // NOLINT
                    const Instruction& next_instr);