[clang-tidy] NO.17 performance-for-range-copy (#57809)

* fix

* fix

.clang-tidy

@@ -189,7 +189,7 @@ modernize-use-override,
 modernize-use-transparent-functors,
 -modernize-use-uncaught-exceptions,
 performance-faster-string-find,
--performance-for-range-copy,
+performance-for-range-copy,
 -performance-implicit-conversion-in-loop,
 -performance-inefficient-algorithm,
 performance-inefficient-string-concatenation,

paddle/fluid/distributed/auto_parallel/dist_attr.cc

@@ -83,7 +83,7 @@ OperatorDistAttr& OperatorDistAttr::operator=(
 
 void OperatorDistAttr::initialize(const OpDesc* op) {
   if (op == nullptr) return;
-  for (std::string name : op->InputArgumentNames()) {
+  for (std::string const& name : op->InputArgumentNames()) {
     VarDesc* input = op->Block()->FindVarRecursive(name);
     VLOG(4) << "[OperatorDistAttr create input dist attr] " << name;
     if (input == nullptr || op->Type() == "create_py_reader") {
@@ -92,7 +92,7 @@ void OperatorDistAttr::initialize(const OpDesc* op) {
       input_dist_attrs_[name] = TensorDistAttr(get_tensor_shape(input));
     }
   }
-  for (std::string name : op->OutputArgumentNames()) {
+  for (std::string const& name : op->OutputArgumentNames()) {
     VarDesc* output = op->Block()->FindVarRecursive(name);
     VLOG(4) << "[OperatorDistAttr create output dist attr] " << name;
     if (output == nullptr) {

paddle/fluid/distributed/fleet_executor/carrier.cc

@@ -121,7 +121,7 @@ void Carrier::CopyParameters(
     const framework::ProgramDesc& program,
     const std::vector<std::string>& inference_root_scope_vars) {
   std::map<std::string, int> inference_root_scope_var_map;
-  for (auto var_name : inference_root_scope_vars) {
+  for (auto const& var_name : inference_root_scope_vars) {
     inference_root_scope_var_map.insert({var_name, 1});
   }
   for (size_t i = 0; i < program.Size(); ++i) {

paddle/fluid/distributed/fleet_executor/compute_interceptor.cc

@@ -85,7 +85,7 @@ InterceptorMessage ComputeInterceptor::PrepareVarsMsg() {
   ready_msg.set_message_type(DATA_WITH_VARS);
   ready_msg.set_scope_idx(cur_scope_id_);
   platform::DeviceContextPool& pool = platform::DeviceContextPool::Instance();
-  for (auto iter : node_->vars_to_dtype()) {
+  for (auto const& iter : node_->vars_to_dtype()) {
     VarList* vars = ready_msg.add_vars_list();
     const auto& var_name = iter.first;
     vars->set_name(var_name);

paddle/fluid/distributed/fleet_executor/dist_model.cc

@@ -295,7 +295,7 @@ void DistModel::InsertCommOp(std::string tmp_var_name,
      << ". The ring id is: " << ring_id << ". The group has: " << nranks
      << " ranks. Current rank in the group is: " << rank
      << ". The endpoint is: " << endpoint << ". Peer endpoints are: ";
-  for (auto ep : peer_endpoints) {
+  for (const auto &ep : peer_endpoints) {
     ss << ep << ", ";
   }
   VLOG(3) << ss.str();

paddle/fluid/distributed/fleet_executor/fleet_executor.cc

@@ -82,7 +82,7 @@ void PreventVarsDelete(
   for (const auto& pair : *unused_vars) {
     const framework::OperatorBase* op = pair.first;
     std::vector<std::string> cur_unused = pair.second;
-    for (auto name : vars_not_gc) {
+    for (auto const& name : vars_not_gc) {
       auto iter = std::find(cur_unused.begin(), cur_unused.end(), name);
       if (iter != cur_unused.end()) {
         VLOG(3) << "Removing var: [" << name
@@ -165,7 +165,7 @@ void FleetExecutor::Init(
       while_block_vars = GetUnusedVarsAfterWhile(
           program_desc, task_node, inference_root_scope_vars);
       VLOG(3) << "Vars will be gced after while op";
-      for (auto var : while_block_vars) {
+      for (auto const& var : while_block_vars) {
         VLOG(3) << var;
       }
       task_node->SetWhileBlockVars(while_block_vars);

paddle/fluid/eager/auto_code_generator/eager_generator.cc

@@ -410,7 +410,7 @@ static std::pair<std::string, std::string> GetAttrType(
       ret = "std::vector<std::string>";
       if (is_arg) ret += "&";
       val += "{";
-      for (auto x : PADDLE_GET_CONST(std::vector<std::string>, attr)) {
+      for (auto const& x : PADDLE_GET_CONST(std::vector<std::string>, attr)) {
         val += "\"" + x + "\"" + ",";
       }
       if (val.size() > 1) val.pop_back();
@@ -1238,7 +1238,7 @@ static std::string GenerateGradNodeCreationContent(
       bool found_target_name = false;
       for (const auto& iter : op_base_infos) {
         const auto& grad_outs_slot_map = iter.GetGradOutsSlotnameMap();
-        for (auto iter : grad_outs_slot_map) {
+        for (auto const& iter : grad_outs_slot_map) {
           if ((!found_target_name) && (input_name == iter.second)) {
             const char* SET_GRAD_OUT_META_TEMPLATE =
                 "      grad_node->SetGradOutMeta(%s, %d);\n";
@@ -1256,7 +1256,7 @@ static std::string GenerateGradNodeCreationContent(
       bool found_target_name = false;
       for (const auto& iter : op_base_infos) {
         const auto& grad_outs_slot_map = iter.GetGradOutsSlotnameMap();
-        for (auto iter : grad_outs_slot_map) {
+        for (auto const& iter : grad_outs_slot_map) {
           if ((!found_target_name) && (input_name == iter.second)) {
             const char* SET_GRAD_OUT_META_TEMPLATE =
                 "      grad_node->SetGradOutMeta(%s, %d);\n";
@@ -2142,7 +2142,7 @@ static std::string GenerateSingleOpBase(
   // [Generation] Get Full Zero
   std::string fill_zero_str = "";
   if (ops_to_fill_zero_for_empty_grads.count(fwd_op_type)) {
-    for (auto iter : grad_ins) {
+    for (auto const& iter : grad_ins) {
       const std::string& grad_input_name = iter.first;
       if (grad_ins_grad_slotname_map.count(grad_input_name)) {
         size_t fwd_output_position = fwd_outputs_name_pos_map.at(
@@ -2189,7 +2189,7 @@ static std::string GenerateSingleOpBase(
       "backward_inplace_tensor" + std::to_string(*outs_size);
   bool process_backward_inplace = false;
   std::string ins_contents_str = "";
-  for (auto iter : grad_ins) {
+  for (auto const& iter : grad_ins) {
     const std::string& grad_input_name = iter.first;
 
     if (grad_ins_fwd_slotname_map.count(grad_input_name)) {
@@ -2293,7 +2293,7 @@ static std::string GenerateSingleOpBase(
       paddle::string::Sprintf(BWD_INS_MAP_TEMPLATE, ins_name, ins_contents_str);
   generated_grad_function_body += ins_map_str;
 
-  for (auto iter : grad_ins) {
+  for (auto const& iter : grad_ins) {
     const std::string& grad_input_name = iter.first;
 
     if (grad_ins_fwd_slotname_map.count(grad_input_name)) {
@@ -2335,7 +2335,7 @@ static std::string GenerateSingleOpBase(
   VLOG(6) << "Generated Ins Map";
   // [Generation] Get Outs Map
   std::string outs_contents_str = "";
-  for (auto iter : grad_outs) {
+  for (auto const& iter : grad_outs) {
     const std::string& grad_output_name = iter.first;
 
     if (grad_outs_slotname_map.count(grad_output_name)) {
@@ -2440,7 +2440,7 @@ static std::string GenerateSingleOpBase(
   generated_grad_function_body += outs_map_str;
   generated_grad_function_body += outs_contents_str;
   generated_grad_function_body += "\n";
-  for (auto iter : grad_outs) {
+  for (auto const& iter : grad_outs) {
     const std::string& grad_output_name = iter.first;
 
     if (grad_outs_slotname_map.count(grad_output_name)) {
@@ -2498,7 +2498,7 @@ static std::string GenerateSingleOpBase(
         "%s[\"%s\"][0]);\n"
         "  };\n";
     std::string backward_inplace_map_str = "";
-    for (auto iter : backward_inplace_map) {
+    for (auto const& iter : backward_inplace_map) {
       std::string backward_inplace_input_name = iter.first;
       std::string backward_inplace_output_name = iter.second;
       backward_inplace_map_str += paddle::string::Sprintf(
@@ -2553,7 +2553,7 @@ static std::string GenerateSingleOpBase(
   // [Generation] Get Return
   std::string outputs_str = "";
   size_t num_appended_outputs = 0;
-  for (auto iter : grad_outs) {
+  for (auto const& iter : grad_outs) {
     const std::string& grad_out_name = iter.first;
     const std::string& fwd_name = grad_outs_slotname_map.at(grad_out_name);
 
@@ -2594,7 +2594,7 @@ static std::string GenerateSingleOpBase(
 
   /* Handle Special Case: "PullSparseOp", etc
      For returns, append "GradOut" to the very end of return list. */
-  for (auto iter : grad_outs) {
+  for (auto const& iter : grad_outs) {
     const std::string& grad_out_name = iter.first;
     const std::string& fwd_name = grad_outs_slotname_map.at(grad_out_name);
 

paddle/fluid/eager/backward.cc

@@ -85,7 +85,7 @@ void EnforceGradNodeHasInput(GradNodeBase* node) {
 void DuplicateCheck(const std::vector<paddle::Tensor>& inputs, bool is_input) {
   std::unordered_set<AutogradMeta*> visisted_ins;
   std::string msg = is_input ? "inputs" : "outputs";
-  for (auto in : inputs) {
+  for (auto const& in : inputs) {
     AutogradMeta* auto_grad_meta = EagerUtils::unsafe_autograd_meta(in);
     PADDLE_ENFORCE_EQ(
         visisted_ins.count(auto_grad_meta),

paddle/fluid/eager/custom_operator/custom_operator_node.cc

@@ -190,12 +190,12 @@ RunCustomOpNode::operator()(paddle::small_vector<std::vector<paddle::Tensor>,
     }
   }
 
-  for (auto it : fwd_outs) {
+  for (auto it : fwd_outs) {  // NOLINT
     VLOG(7) << "Insert fwd_outs to grad_inputs: " << it.first;
     tmp_ins[it.first] = RunCustomOpNode::Recover(&(it.second));
   }
 
-  for (auto it : fwd_ins) {
+  for (auto it : fwd_ins) {  // NOLINT
     // NOTE(HongyuJia): returned tensor maybe un-defined tensor when inputs
     // optional<Tensor>
     VLOG(7) << "Insert fwd_ins to grad_inputs: " << it.first;
@@ -406,12 +406,12 @@ RunCustomOpDoubleGradNode::operator()(
     }
   }
 
-  for (auto it : fwd_outs) {
+  for (auto it : fwd_outs) {  // NOLINT
     VLOG(7) << "Insert fwd_outs to grad_inputs: " << it.first;
     tmp_ins[it.first] = RunCustomOpDoubleGradNode::Recover(&(it.second));
   }
 
-  for (auto it : fwd_ins) {
+  for (auto it : fwd_ins) {  // NOLINT
     VLOG(7) << "Insert fwd_ins to grad_inputs: " << it.first;
     tmp_ins[it.first] = RunCustomOpDoubleGradNode::Recover(&(it.second));
   }

paddle/fluid/framework/data_set.cc

@@ -138,7 +138,7 @@ std::vector<std::string> DatasetImpl<T>::GetSlots() {
     }
   }
   std::cout << "dataset use slots: ";
-  for (auto s : use_slots_) {
+  for (auto const& s : use_slots_) {
     std::cout << s << " | ";
   }
   std::cout << " end " << std::endl;
@@ -216,7 +216,7 @@ template <typename T>
 std::vector<paddle::framework::DataFeed*> DatasetImpl<T>::GetReaders() {
   std::vector<paddle::framework::DataFeed*> ret;
   ret.reserve(readers_.size());
-  for (auto i : readers_) {
+  for (auto const& i : readers_) {
     ret.push_back(i.get());
   }
   return ret;

paddle/fluid/framework/downpour_worker.cc

@@ -135,7 +135,7 @@ void DownpourWorker::CollectLabelInfo(size_t table_idx) {
           static_cast<int>(table_idx)));
 
   TableParameter table;
-  for (auto i : param_.sparse_table()) {
+  for (auto const& i : param_.sparse_table()) {
     if (i.table_id() == table_id) {
       table = i;
       break;
@@ -191,7 +191,7 @@ void DownpourWorker::FillSparseValue(size_t table_idx) {
           static_cast<int>(table_idx)));
 
   TableParameter table;
-  for (auto i : param_.sparse_table()) {
+  for (auto const& i : param_.sparse_table()) {
     if (i.table_id() == table_id) {
       table = i;
       break;
@@ -513,7 +513,7 @@ void DownpourWorker::TrainFilesWithProfiler() {
       uint64_t tid = static_cast<uint64_t>(
           param_.program_config(0).pull_sparse_table_id(i));
       TableParameter table;
-      for (auto j : param_.sparse_table()) {
+      for (auto const& j : param_.sparse_table()) {
         if (j.table_id() == tid) {
           table = j;
           break;
@@ -599,7 +599,7 @@ void DownpourWorker::TrainFilesWithProfiler() {
         uint64_t tid = static_cast<uint64_t>(
             param_.program_config(0).push_sparse_table_id(i));
         TableParameter table;
-        for (auto i : param_.sparse_table()) {
+        for (auto const& i : param_.sparse_table()) {
           if (i.table_id() == tid) {
             table = i;
             break;
@@ -819,7 +819,7 @@ void DownpourWorker::TrainFiles() {
       uint64_t tid = static_cast<uint64_t>(
           param_.program_config(0).pull_sparse_table_id(i));
       TableParameter table;
-      for (auto j : param_.sparse_table()) {
+      for (auto const& j : param_.sparse_table()) {
         if (j.table_id() == tid) {
           table = j;
           break;
@@ -936,7 +936,7 @@ void DownpourWorker::TrainFiles() {
         uint64_t tid = static_cast<uint64_t>(
             param_.program_config(0).push_sparse_table_id(i));
         TableParameter table;
-        for (auto i : param_.sparse_table()) {
+        for (auto const& i : param_.sparse_table()) {
           if (i.table_id() == tid) {
             table = i;
             break;

paddle/fluid/framework/downpour_worker_opt.cc

@@ -262,7 +262,7 @@ void DownpourWorkerOpt::CreateThreadOperatorsWithRerank(
     uint64_t tid =
         static_cast<uint64_t>(param_.program_config(0).pull_sparse_table_id(i));
     TableParameter table;
-    for (auto j : param_.sparse_table()) {
+    for (auto const& j : param_.sparse_table()) {
       if (j.table_id() == tid) {
         table = j;
         break;
@@ -315,7 +315,7 @@ void DownpourWorkerOpt::TrainFiles() {
     uint64_t tid =
         static_cast<uint64_t>(param_.program_config(0).pull_sparse_table_id(i));
     TableParameter table;
-    for (auto j : param_.sparse_table()) {
+    for (auto const& j : param_.sparse_table()) {
       if (j.table_id() == tid) {
         table = j;
         break;
@@ -344,7 +344,7 @@ void DownpourWorkerOpt::TrainFiles() {
       uint64_t tid = static_cast<uint64_t>(
           param_.program_config(0).pull_sparse_table_id(i));
       TableParameter table;
-      for (auto j : param_.sparse_table()) {
+      for (auto const& j : param_.sparse_table()) {
         if (j.table_id() == tid) {
           table = j;
           break;
@@ -455,7 +455,7 @@ void DownpourWorkerOpt::TrainFiles() {
         uint64_t tid = static_cast<uint64_t>(
             param_.program_config(0).push_sparse_table_id(i));
         TableParameter table;
-        for (auto i : param_.sparse_table()) {
+        for (auto const& i : param_.sparse_table()) {
           if (i.table_id() == tid) {
             table = i;
             break;

paddle/fluid/framework/hogwild_worker.cc

@@ -126,7 +126,7 @@ void HogwildWorker::SetZero(phi::DenseTensor *tensor,
 void HogwildWorker::BindingDataFeedMemory() {
   const std::vector<std::string> &input_feed =
       device_reader_->GetUseSlotAlias();
-  for (auto name : input_feed) {
+  for (auto const &name : input_feed) {
     device_reader_->AddFeedVar(thread_scope_->FindVar(name), name);
   }
 }

paddle/fluid/framework/ir/add_support_int8_pass.cc

@@ -61,8 +61,8 @@ void AddSupportInt8Pass::ApplyImpl(ir::Graph* graph) const {
     // scale for one output
     for (auto out_node : quant_op->outputs) {
       for (auto out_op_node : out_node->outputs) {
-        for (auto name : out_op_node->Op()->InputNames()) {
-          for (auto input_name : out_op_node->Op()->Input(name)) {
+        for (auto const& name : out_op_node->Op()->InputNames()) {
+          for (auto const& input_name : out_op_node->Op()->Input(name)) {
             if (out_op_node->Op()->HasAttr("Input_scale_" + input_name)) {
               for (size_t i = 0; i < quanted_op_desc->OutputNames().size();
                    i++) {

paddle/fluid/framework/ir/coalesce_grad_tensor_pass.cc

@@ -189,7 +189,7 @@ class CoalesceGradTensorPass : public ir::Pass {
       const {
     if (params_grads.empty()) return true;
     auto dtype = GetDtypeOfVar(vars_info, params_grads.front().second);
-    for (auto p_g : params_grads) {
+    for (auto const &p_g : params_grads) {
       auto next_dtype = GetDtypeOfVar(vars_info, p_g.second);
       if (next_dtype != dtype) {
         return false;

paddle/fluid/framework/ir/constant_folding_pass.cc

@@ -93,7 +93,7 @@ void ConstantFoldingPass::ApplyImpl(ir::Graph *graph) const {
       map[out_node->Name()] = 0;
     }
     // Forbid other node in graph having the same name with nodes in map
-    for (auto iter : map) {
+    for (auto const &iter : map) {
       for (auto node : graph->Nodes()) {
         if (node->IsVar() && node->Name() == iter.first) {
           map[node->Name()]++;

paddle/fluid/framework/ir/delete_cast_op_pass.cc

@@ -84,7 +84,7 @@ static std::vector<Node*> FindOpNodeWithInputName(
     if (!node->IsOp()) continue;
     auto inputs = node->Op()->Inputs();
     bool find_input = false;
-    for (auto input : inputs) {
+    for (auto const& input : inputs) {
       auto input_names = input.second;
       if (std::count(input_names.begin(), input_names.end(), input_name) > 0) {
         find_input = true;
@@ -103,7 +103,7 @@ static std::vector<Node*> FindOpNodeWithOutputName(
     if (!node->IsOp()) continue;
     auto outputs = node->Op()->Outputs();
     bool find_output = false;
-    for (auto output : outputs) {
+    for (auto const& output : outputs) {
       auto output_names = output.second;
       if (std::count(output_names.begin(), output_names.end(), output_name) >
           0) {

paddle/fluid/framework/ir/delete_repeated_ops_pass.cc

@@ -36,7 +36,7 @@ namespace ir {
 bool HasOutVarName(Node* op_node, std::string name) {
   auto* op_desc = op_node->Op();
   auto outputs = op_desc->Outputs();
-  for (auto iter : outputs) {
+  for (auto const& iter : outputs) {
     auto out_names = iter.second;
     if (std::count(out_names.begin(), out_names.end(), name) > 0) {
       return true;
@@ -155,7 +155,7 @@ void DeleteRepeatedOpsPass::DeleteRepeatedOps(
       }
     }
 
-    for (auto iter : ops_map) {
+    for (auto const& iter : ops_map) {
       auto ops = iter.second;
       auto* first_op_out = ops[0]->outputs[0];
       auto first_op_out_name = first_op_out->Name();