[clang-tidy] No.24 fix some performance-inefficient-vector-operation part:1 (#58028)

gouzil · web-flow · commit 60adf3a499df · 2023-10-12T14:31:56.000+08:00
diff --git a/paddle/fluid/distributed/collective/reducer.cc b/paddle/fluid/distributed/collective/reducer.cc
@@ -985,6 +985,7 @@ void EagerReducer::ProcessUnusedDenseVars() {
   opts.reduce_op = ReduceOp::SUM;
   std::vector<Tensor> reduce_tensors = {global_used_vars_};
   std::vector<phi::DenseTensor> in_out;
+  in_out.reserve(reduce_tensors.size());
   for (auto &t : reduce_tensors) {
     in_out.push_back(*std::dynamic_pointer_cast<phi::DenseTensor>(t.impl()));
   }
@@ -1081,6 +1082,7 @@ void EagerReducer::FusedAllReduceSchedule(EagerGroup *group,
   // all_reduce
   std::vector<Tensor> reduce_tensors = {group->dense_contents_};
   std::vector<phi::DenseTensor> in_out;
+  in_out.reserve(reduce_tensors.size());
   for (auto &t : reduce_tensors) {
     in_out.push_back(*std::dynamic_pointer_cast<phi::DenseTensor>(t.impl()));
   }
@@ -1166,6 +1168,7 @@ void EagerReducer::AllReduceSparse(EagerGroup *group,
   opts.reduce_op = ReduceOp::SUM;
   std::vector<Tensor> reduce_tensors = {rows_num_tensor};
   std::vector<phi::DenseTensor> in_out;
+  in_out.reserve(reduce_tensors.size());
   for (auto &t : reduce_tensors) {
     in_out.push_back(*std::dynamic_pointer_cast<phi::DenseTensor>(t.impl()));
   }
@@ -1214,6 +1217,8 @@ void EagerReducer::AllReduceSparse(EagerGroup *group,
     std::vector<Tensor> dst_rows_tensors = {dst_rows_tensor};
     std::vector<phi::DenseTensor> in;
     std::vector<phi::DenseTensor> out;
+    in.reserve(src_rows_tensors.size());
+    out.reserve(dst_rows_tensors.size());
     for (auto &t : src_rows_tensors) {
       in.push_back(*std::dynamic_pointer_cast<phi::DenseTensor>(t.impl()));
     }
@@ -1245,6 +1250,8 @@ void EagerReducer::AllReduceSparse(EagerGroup *group,
     std::vector<Tensor> dst_value_tensors = {dst_value_tensor};
     std::vector<phi::DenseTensor> src_dense;
     std::vector<phi::DenseTensor> dst_dense;
+    src_dense.reserve(src_value_tensors.size());
+    dst_dense.reserve(dst_value_tensors.size());
     for (auto &t : src_value_tensors) {
       src_dense.push_back(
           *std::dynamic_pointer_cast<phi::DenseTensor>(t.impl()));
diff --git a/paddle/fluid/distributed/fleet_executor/carrier.cc b/paddle/fluid/distributed/fleet_executor/carrier.cc
@@ -392,6 +392,7 @@ void Carrier::CreateInterceptors(
         }
       }
 
+      cores.reserve(microbatch_scopes_.size());
       for (framework::Scope* scope : microbatch_scopes_) {
         cores.push_back(std::make_shared<InterpreterCore>(
             place_, task_node->program()->Block(0), scope, execution_config));
diff --git a/paddle/fluid/framework/details/parallel_ssa_graph_executor.cc b/paddle/fluid/framework/details/parallel_ssa_graph_executor.cc
@@ -297,6 +297,7 @@ FetchResultType ParallelSSAGraphExecutor::Run(
         for (size_t i = 0; i < lodtensorarray_ptrs[0]->size(); ++i) {
           phi::DenseTensor var;
           std::vector<const phi::DenseTensor *> ptrs;
+          ptrs.reserve(lodtensor_ptrs.size());
           for (auto &lodtensorarray_ptr : lodtensorarray_ptrs) {
             ptrs.push_back(&(lodtensorarray_ptr->at(i)));
           }
diff --git a/paddle/fluid/framework/ir/delete_cast_op_pass_test.cc b/paddle/fluid/framework/ir/delete_cast_op_pass_test.cc
@@ -53,6 +53,7 @@ VarDesc* AddWriteToArray(BlockDesc* block,
   OpDesc* op = block->AppendOp();
   op->SetType("write_to_array");
   std::vector<std::string> x_names;
+  x_names.reserve(x.size());
   for (auto k : x) {
     x_names.push_back(k->Name());
   }
diff --git a/paddle/fluid/framework/ir/fuse_adamw_op_pass.cc b/paddle/fluid/framework/ir/fuse_adamw_op_pass.cc
@@ -24,6 +24,7 @@ namespace ir {
 
 std::vector<std::string> GetNodeNames(const std::vector<Node *> &node_vector) {
   std::vector<std::string> out_vector;
+  out_vector.reserve(node_vector.size());
   for (auto i : node_vector) {
     out_vector.emplace_back(i->Name());
   }
diff --git a/paddle/fluid/framework/ir/fuse_gemm_epilogue_pass.cc b/paddle/fluid/framework/ir/fuse_gemm_epilogue_pass.cc
@@ -319,6 +319,8 @@ ir::Graph *FuseGemmEpiloguePass::FuseLinearBwd(ir::Graph *graph,
     auto ele_add_grad_op_role_val =
         details::GetOpRoleVarsOrEmpty(*(ele_add_grad_op->Op()));
     std::vector<std::string> fused_gemm_epilogue_grad_op_role_var;
+    fused_gemm_epilogue_grad_op_role_var.reserve(
+        matmul_grad_op_role_val.size());
     for (auto const &i : matmul_grad_op_role_val) {
       fused_gemm_epilogue_grad_op_role_var.push_back(i);
     }
@@ -455,6 +457,8 @@ ir::Graph *FuseGemmEpiloguePass::FuseLinearActBwd(
     auto ele_add_grad_op_role_val =
         details::GetOpRoleVarsOrEmpty(*(ele_add_grad_op->Op()));
     std::vector<std::string> fused_gemm_epilogue_grad_op_role_var;
+    fused_gemm_epilogue_grad_op_role_var.reserve(
+        matmul_grad_op_role_val.size());
     for (auto const &i : matmul_grad_op_role_val) {
       fused_gemm_epilogue_grad_op_role_var.push_back(i);
     }
diff --git a/paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_optimizer_op_pass.cc b/paddle/fluid/framework/ir/fuse_optimizer_ops_pass/fuse_optimizer_op_pass.cc
@@ -365,6 +365,7 @@ void FuseOptimizerOpPass::GradientsFilter(
     }
   }
   std::vector<Node *> sorted_ops;
+  sorted_ops.reserve(new_grad_idx.size());
   for (size_t i : new_grad_idx) {
     sorted_ops.emplace_back(opt_nodes->at(i));
   }
diff --git a/paddle/fluid/framework/ir/graph_helper.cc b/paddle/fluid/framework/ir/graph_helper.cc
@@ -502,6 +502,7 @@ static OpDesc *ReplaceScaleLossGradOp(const Node &node, OpDesc *desc) {
   // TODO(Ruibiao) : Set OpDeviceAttrName when needed
 
   std::vector<std::string> output_names;
+  output_names.reserve(node.outputs.size());
   for (auto out : node.outputs) {
     output_names.emplace_back(out->Name());
   }
diff --git a/paddle/fluid/framework/ir/memory_optimize_pass/test_reference_count_pass_last_lived_ops.cc b/paddle/fluid/framework/ir/memory_optimize_pass/test_reference_count_pass_last_lived_ops.cc
@@ -123,6 +123,7 @@ class ReferenceCountPassTestHelper {
   std::vector<OperatorBase *> LastLivedOps(const std::string &name) const {
     auto &ops = last_live_ops_of_vars_[0].at(name).ops();
     std::vector<OperatorBase *> ret;
+    ret.reserve(ops.size());
     for (auto *op : ops) {
       ret.emplace_back(op->GetOp());
     }
diff --git a/paddle/fluid/framework/ir/multi_batch_merge_pass.cc b/paddle/fluid/framework/ir/multi_batch_merge_pass.cc
@@ -249,6 +249,7 @@ void BatchMergePass::ApplyImpl(ir::Graph* graph) const {
     sum_op.SetType("sum");
     std::vector<std::string> repeated_grad_names;
     std::vector<std::string> param_grad_op_role_var;
+    repeated_grad_names.reserve(kv.second.size());
     for (auto r : kv.second) {
       repeated_grad_names.push_back(r->Var()->Name());
     }
diff --git a/paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.cc b/paddle/fluid/framework/ir/multi_devices_graph_pass/multi_devices_graph_pass.cc
@@ -1137,6 +1137,7 @@ int DistSSAGraphBuilder::CreateRPCOp(ir::Graph *result, ir::Node *node) const {
             details::BuildStrategy::ReduceStrategy::kAllReduce &&
         node->inputs[0]->Name().find(".block") == std::string::npos) {
       std::vector<std::string> input_var_names;
+      input_var_names.reserve(node->inputs.size());
       for (ir::Node *n : node->inputs) {
         input_var_names.push_back(n->Name());
       }
@@ -1162,6 +1163,7 @@ int DistSSAGraphBuilder::CreateRPCOp(ir::Graph *result, ir::Node *node) const {
     }
   } else if (node->Op()->Type() == "recv") {
     std::vector<std::string> output_var_names;
+    output_var_names.reserve(node->inputs.size());
     for (ir::Node *n : node->outputs) {
       output_var_names.push_back(n->Name());
     }
@@ -1245,6 +1247,8 @@ int DistSSAGraphBuilder::CreateDistTrainOp(ir::Graph *result,
   int op_dev_id = -1;
   std::vector<std::string> input_var_names;
   std::vector<std::string> output_var_names;
+  input_var_names.reserve(node->inputs.size());
+  output_var_names.reserve(node->outputs.size());
   for (ir::Node *input : node->inputs) {
     input_var_names.push_back(input->Name());
   }
diff --git a/paddle/fluid/framework/ir/subgraph_detector.cc b/paddle/fluid/framework/ir/subgraph_detector.cc
@@ -234,6 +234,7 @@ void FlexibleDFS(const std::vector<BriefNode *> &source,
   } FNode;
 
   std::vector<FNode> stack;
+  stack.reserve(source.size());
   for (auto &node : source) {
     stack.push_back(FNode{node, false});
   }
diff --git a/paddle/fluid/framework/ir/transfer_layout_elim_pass.cc b/paddle/fluid/framework/ir/transfer_layout_elim_pass.cc
@@ -61,6 +61,7 @@ void TransferLayoutElimPass::PutTranferlayoutAfterOp(
   // group_norm has 3 inputs, but we do not need there is a transfer_layout
   // before Bias and Scale so we extract useful_var1s from op_node->inputs.
   std::vector<Node *> useful_var1s;
+  useful_var1s.reserve(op_node->inputs.size());
   for (auto var1 : op_node->inputs) {
     // if (var1->inputs.size() >= 1 &&
     //         var1->inputs[0]->Op()->Type() == "transfer_layout") {
diff --git a/paddle/fluid/framework/ir/transpose_flatten_concat_fuse_pass.cc b/paddle/fluid/framework/ir/transpose_flatten_concat_fuse_pass.cc
@@ -69,6 +69,7 @@ void TransposeFlattenConcatFusePass::RunTransposeFlattenConcatFuse(
 
   GraphPatternDetector gpd;
   std::vector<PDNode *> input_nodes;
+  input_nodes.reserve(times);
   for (int i = 0; i < times; i++) {
     input_nodes.push_back(gpd.mutable_pattern()
                               ->NewNode("x" + std::to_string(i))
@@ -166,6 +167,7 @@ void TransposeFlattenConcatFusePass::RunTransposeFlattenConcatFuse(
     int concat_axis = PADDLE_GET_CONST(int, concat_op->Op()->GetAttr("axis"));
     std::string output_name = concat_out->Name();
 
+    input_names.reserve(times);
     for (int i = 0; i < times; i++) {
       input_names.push_back(nodes[i * kNumFields]->Name());
     }
diff --git a/paddle/fluid/framework/ir/trt_delete_weight_dequant_linear_op_pass.cc b/paddle/fluid/framework/ir/trt_delete_weight_dequant_linear_op_pass.cc
@@ -254,6 +254,7 @@ void TrtDeleteWeightQuantDequantLinearOpPass::ApplyImpl(
     float* weight_scale_data = weight_scale_tensor->data<float>();
 
     auto weight_scale_nums = weight_scale_tensor->numel();
+    weight_scale.reserve(weight_scale_nums);
     for (int i = 0; i < weight_scale_nums; i++) {
       weight_scale.push_back(weight_scale_data[i] / static_cast<float>(range));
     }

Original file line number	Diff line number	Diff line change
`@@ -392,6 +392,7 @@ void Carrier::CreateInterceptors(`
`392`	`392`	`}`
`393`	`393`	`}`
`394`	`394`
	`395`	`+ cores.reserve(microbatch_scopes_.size());`
`395`	`396`	`for (framework::Scope* scope : microbatch_scopes_) {`
`396`	`397`	`cores.push_back(std::make_shared<InterpreterCore>(`
`397`	`398`	`place_, task_node->program()->Block(0), scope, execution_config));`
Original file line number	Diff line number	Diff line change
`@@ -297,6 +297,7 @@ FetchResultType ParallelSSAGraphExecutor::Run(`
`297`	`297`	`for (size_t i = 0; i < lodtensorarray_ptrs[0]->size(); ++i) {`
`298`	`298`	`phi::DenseTensor var;`
`299`	`299`	`std::vector<const phi::DenseTensor *> ptrs;`
	`300`	`+ ptrs.reserve(lodtensor_ptrs.size());`
`300`	`301`	`for (auto &lodtensorarray_ptr : lodtensorarray_ptrs) {`
`301`	`302`	`ptrs.push_back(&(lodtensorarray_ptr->at(i)));`
`302`	`303`	`}`
Original file line number	Diff line number	Diff line change
`@@ -53,6 +53,7 @@ VarDesc* AddWriteToArray(BlockDesc* block,`
`53`	`53`	`OpDesc* op = block->AppendOp();`
`54`	`54`	`op->SetType("write_to_array");`
`55`	`55`	`std::vector<std::string> x_names;`
	`56`	`+ x_names.reserve(x.size());`
`56`	`57`	`for (auto k : x) {`
`57`	`58`	`x_names.push_back(k->Name());`
`58`	`59`	`}`
Original file line number	Diff line number	Diff line change
`@@ -24,6 +24,7 @@ namespace ir {`
`24`	`24`
`25`	`25`	`std::vector<std::string> GetNodeNames(const std::vector<Node *> &node_vector) {`
`26`	`26`	`std::vector<std::string> out_vector;`
	`27`	`+ out_vector.reserve(node_vector.size());`
`27`	`28`	`for (auto i : node_vector) {`
`28`	`29`	`out_vector.emplace_back(i->Name());`
`29`	`30`	`}`
Original file line number	Diff line number	Diff line change
`@@ -365,6 +365,7 @@ void FuseOptimizerOpPass::GradientsFilter(`
`365`	`365`	`}`
`366`	`366`	`}`
`367`	`367`	`std::vector<Node *> sorted_ops;`
	`368`	`+ sorted_ops.reserve(new_grad_idx.size());`
`368`	`369`	`for (size_t i : new_grad_idx) {`
`369`	`370`	`sorted_ops.emplace_back(opt_nodes->at(i));`
`370`	`371`	`}`
Original file line number	Diff line number	Diff line change
`@@ -502,6 +502,7 @@ static OpDesc ReplaceScaleLossGradOp(const Node &node, OpDesc desc) {`
`502`	`502`	`// TODO(Ruibiao) : Set OpDeviceAttrName when needed`
`503`	`503`
`504`	`504`	`std::vector<std::string> output_names;`
	`505`	`+ output_names.reserve(node.outputs.size());`
`505`	`506`	`for (auto out : node.outputs) {`
`506`	`507`	`output_names.emplace_back(out->Name());`
`507`	`508`	`}`
Original file line number	Diff line number	Diff line change
`@@ -123,6 +123,7 @@ class ReferenceCountPassTestHelper {`
`123`	`123`	`std::vector<OperatorBase *> LastLivedOps(const std::string &name) const {`
`124`	`124`	`auto &ops = last_live_ops_of_vars_[0].at(name).ops();`
`125`	`125`	`std::vector<OperatorBase *> ret;`
	`126`	`+ ret.reserve(ops.size());`
`126`	`127`	`for (auto *op : ops) {`
`127`	`128`	`ret.emplace_back(op->GetOp());`
`128`	`129`	`}`
Original file line number	Diff line number	Diff line change
`@@ -249,6 +249,7 @@ void BatchMergePass::ApplyImpl(ir::Graph* graph) const {`
`249`	`249`	`sum_op.SetType("sum");`
`250`	`250`	`std::vector<std::string> repeated_grad_names;`
`251`	`251`	`std::vector<std::string> param_grad_op_role_var;`
	`252`	`+ repeated_grad_names.reserve(kv.second.size());`
`252`	`253`	`for (auto r : kv.second) {`
`253`	`254`	`repeated_grad_names.push_back(r->Var()->Name());`
`254`	`255`	`}`