remove -Wno-error=sign-compare, make warning as error (#21358)

* remove -Wno-error=sign-compare, make warning as error

test=develop test=document_fix

* fix exist compile warning

test=develop

cmake/flags.cmake

@@ -148,7 +148,6 @@ set(COMMON_FLAGS
     -Wno-unused-parameter
     -Wno-unused-function
     -Wno-error=literal-suffix
-    -Wno-error=sign-compare
     -Wno-error=unused-local-typedefs
     -Wno-error=parentheses-equality # Warnings in pybind11
     -Wno-error=ignored-attributes  # Warnings in Eigen, gcc 6.3
@@ -183,7 +182,6 @@ set(GPU_COMMON_FLAGS
     -Wdelete-non-virtual-dtor
     -Wno-unused-parameter
     -Wno-unused-function
-    -Wno-error=sign-compare
     -Wno-error=literal-suffix
     -Wno-error=unused-local-typedefs
     -Wno-error=unused-function  # Warnings in Numpy Header.

paddle/fluid/framework/data_feed.cc

@@ -391,7 +391,7 @@ void MultiSlotDataFeed::Init(
       use_slots_is_dense_.push_back(slot.is_dense());
       std::vector<int> local_shape;
       if (slot.is_dense()) {
-        for (size_t j = 0; j < slot.shape_size(); ++j) {
+        for (int j = 0; j < slot.shape_size(); ++j) {
           if (slot.shape(j) > 0) {
             total_dims_without_inductive_[i] *= slot.shape(j);
           }
@@ -400,7 +400,7 @@ void MultiSlotDataFeed::Init(
           }
         }
       }
-      for (size_t j = 0; j < slot.shape_size(); ++j) {
+      for (int j = 0; j < slot.shape_size(); ++j) {
         local_shape.push_back(slot.shape(j));
       }
       use_slots_shape_.push_back(local_shape);
@@ -736,7 +736,7 @@ void MultiSlotInMemoryDataFeed::Init(
       use_slots_is_dense_.push_back(slot.is_dense());
       std::vector<int> local_shape;
       if (slot.is_dense()) {
-        for (size_t j = 0; j < slot.shape_size(); ++j) {
+        for (int j = 0; j < slot.shape_size(); ++j) {
           if (slot.shape(j) > 0) {
             total_dims_without_inductive_[i] *= slot.shape(j);
           }
@@ -745,7 +745,7 @@ void MultiSlotInMemoryDataFeed::Init(
           }
         }
       }
-      for (size_t j = 0; j < slot.shape_size(); ++j) {
+      for (int j = 0; j < slot.shape_size(); ++j) {
         local_shape.push_back(slot.shape(j));
       }
       use_slots_shape_.push_back(local_shape);

paddle/fluid/framework/data_set.cc

@@ -638,7 +638,7 @@ void MultiSlotDataset::MergeByInsId() {
   }
   auto multi_slot_desc = data_feed_desc_.multi_slot_desc();
   std::vector<std::string> use_slots;
-  for (size_t i = 0; i < multi_slot_desc.slots_size(); ++i) {
+  for (int i = 0; i < multi_slot_desc.slots_size(); ++i) {
     const auto& slot = multi_slot_desc.slots(i);
     if (slot.is_used()) {
       use_slots.push_back(slot.name());
@@ -828,7 +828,7 @@ void MultiSlotDataset::SlotsShuffle(
   timeline.Start();
   auto multi_slot_desc = data_feed_desc_.multi_slot_desc();
   std::set<uint16_t> index_slots;
-  for (size_t i = 0; i < multi_slot_desc.slots_size(); ++i) {
+  for (int i = 0; i < multi_slot_desc.slots_size(); ++i) {
     std::string cur_slot = multi_slot_desc.slots(i).name();
     if (slots_to_replace.find(cur_slot) != slots_to_replace.end()) {
       index_slots.insert(i);

paddle/fluid/framework/data_set.h

@@ -215,7 +215,7 @@ class DatasetImpl : public Dataset {
   bool merge_by_insid_;
   bool parse_ins_id_;
   bool parse_content_;
-  int merge_size_;
+  size_t merge_size_;
   bool slots_shuffle_fea_eval_ = false;
   int preload_thread_num_;
   std::mutex global_index_mutex_;

paddle/fluid/framework/dist_multi_trainer.cc

@@ -152,7 +152,7 @@ void DistMultiTrainer::Finalize() {
   for (auto &th : threads_) {
     th.join();
   }
-  for (int i = 0; i < need_merge_var_names_.size(); i++) {
+  for (size_t i = 0; i < need_merge_var_names_.size(); i++) {
     Variable *root_var = root_scope_->FindVar(need_merge_var_names_[i]);
     if (root_var == nullptr) {
       continue;

paddle/fluid/framework/downpour_worker.cc

@@ -180,7 +180,7 @@ std::pair<int64_t, int64_t> GetTensorBound(LoDTensor* tensor, int index) {
   }
 }
 
-bool CheckValidOutput(LoDTensor* tensor, int batch_size) {
+bool CheckValidOutput(LoDTensor* tensor, size_t batch_size) {
   auto& dims = tensor->dims();
   if (dims.size() != 2) return false;
   if (tensor->lod().size() != 0) {
@@ -189,7 +189,7 @@ bool CheckValidOutput(LoDTensor* tensor, int batch_size) {
       return false;
     }
   } else {
-    if (dims[0] != batch_size) {
+    if (dims[0] != static_cast<int>(batch_size)) {
       return false;
     }
   }
@@ -329,7 +329,8 @@ void DownpourWorker::FillSparseValue(size_t table_idx) {
         }
         memcpy(ptr + table.emb_dim() * index, fea_value[fea_idx].data(),
                sizeof(float) * table.emb_dim());
-        if (is_nid && index == tensor->lod()[0][nid_ins_index]) {
+        if (is_nid &&
+            static_cast<size_t>(index) == tensor->lod()[0][nid_ins_index]) {
           nid_show_.push_back(fea_value[fea_idx][0]);
           ++nid_ins_index;
         }
@@ -346,7 +347,8 @@ void DownpourWorker::FillSparseValue(size_t table_idx) {
         }
         memcpy(ptr + table.emb_dim() * index, fea_value[fea_idx].data() + 2,
                sizeof(float) * table.emb_dim());
-        if (is_nid && index == tensor->lod()[0][nid_ins_index]) {
+        if (is_nid &&
+            static_cast<size_t>(index) == tensor->lod()[0][nid_ins_index]) {
           nid_show_.push_back(fea_value[fea_idx][0]);
           ++nid_ins_index;
         }
@@ -402,7 +404,7 @@ void DownpourWorker::AdjustInsWeight() {
   int64_t nid_adjw_num = 0;
   double nid_adjw_weight = 0.0;
   size_t ins_index = 0;
-  for (int i = 0; i < len; ++i) {
+  for (size_t i = 0; i < len; ++i) {
     float nid_show = nid_show_[i];
     VLOG(3) << "nid_show " << nid_show;
     if (nid_show < 0) {
@@ -970,7 +972,7 @@ void DownpourWorker::TrainFiles() {
       }
     }
     if (need_dump_field_) {
-      int batch_size = device_reader_->GetCurBatchSize();
+      size_t batch_size = device_reader_->GetCurBatchSize();
       std::vector<std::string> ars(batch_size);
       for (auto& ar : ars) {
         ar.clear();
@@ -990,7 +992,7 @@ void DownpourWorker::TrainFiles() {
         if (!CheckValidOutput(tensor, batch_size)) {
           continue;
         }
-        for (int i = 0; i < batch_size; ++i) {
+        for (size_t i = 0; i < batch_size; ++i) {
           auto output_dim = tensor->dims()[1];
           std::string output_dimstr =
               boost::lexical_cast<std::string>(output_dim);

paddle/fluid/framework/pull_dense_worker.cc

@@ -32,8 +32,8 @@ void PullDenseWorker::Initialize(const TrainerDesc& param) {
   threshold_ = param_.threshold();
   thread_num_ = param_.device_num();
   sleep_time_ms_ = param_.sleep_time_ms();
-  for (size_t i = 0;
+  for (int i = 0; i < dwp_param_.program_config(0).pull_dense_table_id_size();
-       i < dwp_param_.program_config(0).pull_dense_table_id_size(); ++i) {
+       ++i) {
     uint64_t tid = static_cast<uint64_t>(
         dwp_param_.program_config(0).pull_dense_table_id(i));
     TableParameter table;
@@ -67,7 +67,7 @@ void PullDenseWorker::Wait(std::vector<::std::future<int32_t>>* status_vec) {
     }
   }
 
-  int MAX_FAIL_NUM = 20;
+  size_t MAX_FAIL_NUM = 20;
   if (pull_dense_fail_times_ > MAX_FAIL_NUM) {
     LOG(FATAL) << "Pull Dense Failed Times More Than " << MAX_FAIL_NUM
                << " Times";
@@ -85,8 +85,8 @@ void PullDenseWorker::Stop() {
 
 void PullDenseWorker::PullDense(bool force_update) {
   pull_dense_status_.resize(0);
-  for (size_t i = 0;
+  for (int i = 0; i < dwp_param_.program_config(0).pull_dense_table_id_size();
-       i < dwp_param_.program_config(0).pull_dense_table_id_size(); ++i) {
+       ++i) {
     uint64_t tid = static_cast<uint64_t>(
         dwp_param_.program_config(0).pull_dense_table_id(i));
     if (force_update || CheckUpdateParam(tid)) {
@@ -127,7 +127,8 @@ bool PullDenseWorker::CheckUpdateParam(uint64_t table_id) {
   auto& version = training_versions_[table_id];
   current_version_[table_id] =
       *(std::min_element(version.begin(), version.end()));
-  if (current_version_[table_id] - last_versions_[table_id] < threshold_) {
+  if (current_version_[table_id] - last_versions_[table_id] <
+      static_cast<size_t>(threshold_)) {
     return false;
   }
   return true;

paddle/fluid/imperative/tests/test_tracer.cc

@@ -69,7 +69,7 @@ TEST(test_tracer, test_trace_op) {
   mul_attr_map["use_mkldnn"] = false;
   tracer.TraceOp("mul", ins, outs, mul_attr_map, place, true);
   const auto& out_tensor = vout->Var().Get<framework::LoDTensor>();
-  for (size_t i = 0; i < vout->Var().Get<framework::LoDTensor>().numel(); i++) {
+  for (int i = 0; i < vout->Var().Get<framework::LoDTensor>().numel(); i++) {
     ASSERT_EQ(out_tensor.data<float>()[i], 20.0);
   }
 }
@@ -108,7 +108,7 @@ TEST(test_tracer, test_trace_op_with_backward) {
   mul_attr_map["use_mkldnn"] = false;
   tracer.TraceOp("mul", ins, outs, mul_attr_map, place, true);
   const auto& out_tensor = vout->Var().Get<framework::LoDTensor>();
-  for (size_t i = 0; i < vout->Var().Get<framework::LoDTensor>().numel(); i++) {
+  for (int i = 0; i < vout->Var().Get<framework::LoDTensor>().numel(); i++) {
     ASSERT_EQ(out_tensor.data<float>()[i], 20.0);
   }
 }
@@ -239,14 +239,14 @@ TEST(test_tracer, test_trace_op_with_multi_device_inputs) {
   framework::LoDTensor rlt;
   framework::TensorCopySync(vout->Var().Get<framework::LoDTensor>(), place,
                             &rlt);
-  for (size_t i = 0; i < rlt.numel(); i++) {
+  for (int i = 0; i < rlt.numel(); i++) {
     ASSERT_EQ(rlt.data<float>()[i], 4.0);
   }
 
   framework::LoDTensor out_grad;
   framework::TensorCopySync(vout->GradVar().Get<framework::LoDTensor>(), place,
                             &out_grad);
-  for (size_t i = 0; i < out_grad.numel(); ++i) {
+  for (int i = 0; i < out_grad.numel(); ++i) {
     ASSERT_EQ(out_grad.data<float>()[i], 1.0);
   }
 
@@ -254,7 +254,7 @@ TEST(test_tracer, test_trace_op_with_multi_device_inputs) {
   framework::TensorCopySync(x_in->GradVar().Get<framework::LoDTensor>(), place,
                             &x_grad);
 
-  for (size_t i = 0; i < x_grad.numel(); ++i) {
+  for (int i = 0; i < x_grad.numel(); ++i) {
     ASSERT_EQ(x_grad.data<float>()[i], 1.0);
   }
 
@@ -262,7 +262,7 @@ TEST(test_tracer, test_trace_op_with_multi_device_inputs) {
   framework::TensorCopySync(y_in->GradVar().Get<framework::LoDTensor>(), place,
                             &y_grad);
 
-  for (size_t i = 0; i < y_grad.numel(); ++i) {
+  for (int i = 0; i < y_grad.numel(); ++i) {
     ASSERT_EQ(y_grad.data<float>()[i], 1.0);
   }
 }

paddle/fluid/operators/concat_op.h

@@ -26,7 +26,7 @@ namespace paddle {
 namespace operators {
 static inline framework::DDim ComputeAndCheckShape(
     const bool is_runtime, const std::vector<framework::DDim>& inputs_dims,
-    const int axis) {
+    const size_t axis) {
   const size_t n = inputs_dims.size();
   auto out_dims = inputs_dims[0];
   size_t in_zero_dims_size = out_dims.size();

paddle/fluid/operators/conv_op.cc

@@ -114,7 +114,7 @@ void ConvOp::InferShape(framework::InferShapeContext* ctx) const {
   if (!channel_last) {
     output_shape.push_back(filter_dims[0]);
   }
-  for (size_t i = 0; i < in_data_dims.size(); ++i) {
+  for (int i = 0; i < in_data_dims.size(); ++i) {
     if ((!ctx->IsRuntime()) &&
         (in_data_dims[i] <= 0 || filter_dims[i + 2] <= 0)) {
       output_shape.push_back(-1);

paddle/fluid/operators/conv_op.h

@@ -72,8 +72,8 @@ inline void UpdatePaddingAndDilation(std::vector<int>* paddings,
                                      const std::vector<int>& ksize) {
   // set padding size == data_dims.size() * 2
   auto data_shape = framework::vectorize<int>(data_dims);
-  if (paddings->size() == data_dims.size()) {
+  if (static_cast<int>(paddings->size()) == data_dims.size()) {
-    for (size_t i = 0; i < data_dims.size(); ++i) {
+    for (int i = 0; i < data_dims.size(); ++i) {
       int copy_pad = *(paddings->begin() + 2 * i);
       paddings->insert(paddings->begin() + 2 * i + 1, copy_pad);
     }
@@ -85,7 +85,7 @@ inline void UpdatePaddingAndDilation(std::vector<int>* paddings,
 
   // when padding_algorithm is "VALID" or "SAME"
   if (padding_algorithm == "SAME") {
-    for (size_t i = 0; i < data_dims.size(); ++i) {
+    for (int i = 0; i < data_dims.size(); ++i) {
       int out_size = (data_dims[i] + strides[i] - 1) / strides[i];
       int pad_sum =
           std::max((out_size - 1) * strides[i] + ksize[i] - data_shape[i], 0);

paddle/fluid/operators/distributed/communicator.cc

@@ -192,8 +192,8 @@ void AsyncCommunicator::SendThread() {
         auto send_task = [this, &var_name, &var_queue] {
           VLOG(3) << var_name << " merge and send";
           std::vector<std::shared_ptr<Variable>> vars;
-          size_t merged_var_num = 0;
+          int merged_var_num = 0;
-          size_t wait_times = 0;
+          int wait_times = 0;
           while (merged_var_num < FLAGS_communicator_max_merge_var_num) {
             if (var_queue->Size() == 0) {
               VLOG(3) << "wait_times -> " << wait_times;
@@ -254,7 +254,7 @@ void AsyncCommunicator::SendThread() {
 void AsyncCommunicator::RecvThread() {
   VLOG(3) << "RecvThread start!";
   while (running_) {
-    auto grad_num = grad_num_.load();
+    int grad_num = grad_num_.load();
     if (grad_num > FLAGS_communicator_min_send_grad_num_before_recv) {
       VLOG(1) << "current grad num " << grad_num;
       RecvAll();
@@ -538,7 +538,7 @@ void GeoSgdCommunicator::Send(const std::vector<std::string> &sparse_var_names,
     int element_number = var_tensor.numel();
     int *var_mutable_data = var_tensor.mutable_data<int>(var_tensor.place());
     // insert ids which has not been record
-    for (size_t j = 0; j < element_number; j++) {
+    for (int j = 0; j < element_number; j++) {
       auto ep_idx = GetSectionIndex(var_mutable_data[j],
                                     absolute_section_[sparse_var_tables[i]]);
       ids_table->at(sparse_var_tables[i])[ep_idx].insert(var_mutable_data[j]);
@@ -559,7 +559,7 @@ void GeoSgdCommunicator::SendThread() {
     std::vector<std::future<void>> task_futures;
     task_futures.reserve(send_varname_to_ctx_.size());
 
-    size_t wait_times = 0;
+    int wait_times = 0;
     while (ids_send_vec_.size() < geo_need_push_nums_) {
       VLOG(4) << "ids_send_vec_ Size: " << ids_send_vec_.size();
       if (need_push_queue_->Size() > 0) {
@@ -747,7 +747,7 @@ void GeoSgdCommunicator::SendUpdateSparseVars(
   auto cpu_ctx = paddle::platform::CPUDeviceContext();
   auto blas = math::GetBlas<paddle::platform::CPUDeviceContext, float>(cpu_ctx);
   float avg = 1 / static_cast<float>(trainer_nums_);
-  for (int y = 0; y < new_rows.size(); y++) {
+  for (size_t y = 0; y < new_rows.size(); y++) {
     auto ids = new_rows[y];
 
     float *x_val = x_value + ids * row_numel;
@@ -876,7 +876,7 @@ void GeoSgdCommunicator::RecvUpdateSparseVars(
 
   auto cpu_ctx = paddle::platform::CPUDeviceContext();
   auto blas = math::GetBlas<paddle::platform::CPUDeviceContext, float>(cpu_ctx);
-  for (int y = 0; y < new_rows.size(); y++) {
+  for (size_t y = 0; y < new_rows.size(); y++) {
     std::vector<float> row_delta(row_numel, 0);
 
     auto ids = new_rows[y];

paddle/fluid/operators/distributed/communicator.h

@@ -418,7 +418,7 @@ class GeoSgdCommunicator : public Communicator {
 
  private:
   int trainer_nums_ = 1;
-  int geo_need_push_nums_ = 100;
+  size_t geo_need_push_nums_ = 100;
   bool is_geo_sgd_ = false;
   Scope* training_scope_;
   std::shared_ptr<Scope> delta_scope_;  // parameter local delta: recv - old

paddle/fluid/operators/distributed/communicator_test.cc

@@ -75,7 +75,7 @@ TEST(communicator, merge_selected_rows) {
     auto dims =
         framework::make_ddim({static_cast<int64_t>(rows.size()), width});
     auto *data = slr->mutable_value()->mutable_data<float>(dims, cpu_place);
-    for (auto i = 0; i < rows.size(); ++i) {
+    for (size_t i = 0; i < rows.size(); ++i) {
       for (auto j = 0; j < width; ++j) {
         data[i * width + j] = static_cast<float>(rows[i]);
       }
@@ -97,7 +97,7 @@ TEST(communicator, merge_selected_rows) {
   for (auto i = 0; i < 10; ++i) {
     out_values.push_back(static_cast<float>(i * (10 - i)));
   }
-  for (auto i = 0; i < out_slr.rows().size(); ++i) {
+  for (size_t i = 0; i < out_slr.rows().size(); ++i) {
     ASSERT_EQ(out_slr.rows()[i], i);
     for (auto j = 0; j < width; ++j) {
       ASSERT_EQ(out_data[i * width + j], out_values[i]);

paddle/fluid/operators/distributed/parameter_prefetch.cc

@@ -215,7 +215,7 @@ void prefetchs(const std::vector<std::string>& id_var_names,
   std::unordered_set<int64_t> s(ids_union.begin(), ids_union.end());
   ids_union.assign(s.begin(), s.end());
 
-  for (int i = 0; i < table_names.size(); i++) {
+  for (size_t i = 0; i < table_names.size(); i++) {
     tables.push_back(std::make_pair(table_names[i], endpoints[i]));
   }
 
@@ -230,7 +230,7 @@ void prefetchs(const std::vector<std::string>& id_var_names,
   }
 
   // copy vectors to out vars
-  for (int i = 0; i < out_var_names.size(); i++) {
+  for (size_t i = 0; i < out_var_names.size(); i++) {
     auto& ids = ids_group[i];
     auto* out_t =
         scope.FindVar(out_var_names[i])->GetMutable<framework::LoDTensor>();
@@ -240,7 +240,7 @@ void prefetchs(const std::vector<std::string>& id_var_names,
 
     auto* out_d = out_t->mutable_data<float>(place);
 
-    for (int idx = 0; idx < ids.size(); idx++) {
+    for (size_t idx = 0; idx < ids.size(); idx++) {
       const auto& id = ids[idx];
 
       if (padding_idx != distributed::kNoPadding && id == padding_idx) {

paddle/fluid/operators/distributed/parameter_send.cc

@@ -50,12 +50,12 @@ inline EP_SPLIT_TABLE_PAIRS GetMultiFieldRpcContext(
     PADDLE_ENFORCE_GT(multi_parts, 0, "multi_parts must >=1");
 
     if (multi_parts == 1) {
-      for (int i = 0; i < rpc_ctx.splited_var_names.size(); i++) {
+      for (size_t i = 0; i < rpc_ctx.splited_var_names.size(); i++) {
         table_pairs.push_back(
             std::make_pair(rpc_ctx.epmap[i], rpc_ctx.splited_var_names[i]));
       }
     } else {
-      for (int i = 0; i < rpc_ctx.splited_var_names.size(); i++) {
+      for (size_t i = 0; i < rpc_ctx.splited_var_names.size(); i++) {
         for (int x = 0; x < multi_parts; x++) {
           auto table =
               string::Sprintf("%s@%d@PIECE", rpc_ctx.splited_var_names[i], x);

paddle/fluid/operators/distributed/request_handler_impl.cc

@@ -171,7 +171,7 @@ bool RequestGetHandler::Handle(const std::string& varname,
         auto* data = out_slr->mutable_value()->mutable_data<float>(
             out_dims, origin_tensor.place());
         auto width = dims[1];
-        for (auto i = 0; i < updated_rows.size(); ++i) {
+        for (size_t i = 0; i < updated_rows.size(); ++i) {
           PADDLE_ENFORCE_LT(updated_rows[i], dims[0]);
           memcpy(data + i * width, origin_tensor_data + updated_rows[i] * width,
                  sizeof(float) * width);

paddle/fluid/operators/fused/conv_fusion_op.cc

@@ -137,7 +137,7 @@ class Conv2DFusionOpInferShape : public framework::InferShapeBase {
     std::vector<int64_t> output_shape({in_dims[0]});
     output_shape.push_back(filter_dims[0]);
 
-    for (size_t i = 0; i < in_data_dims.size(); ++i) {
+    for (int i = 0; i < in_data_dims.size(); ++i) {
       if ((!ctx->IsRuntime()) &&
           (in_data_dims[i] <= 0 || filter_dims[i + 2] <= 0)) {
         output_shape.push_back(-1);

paddle/fluid/operators/jit/gen/matmul.cc

@@ -41,7 +41,7 @@ void MatMulJitCode::genCode() {
   for (size_t g = 0; g < groups.size(); ++g) {
     size_t x_offset = 0;
     size_t wgt_offset_tmp = 0;
-    for (int i = 0; i < g; ++i) {
+    for (size_t i = 0; i < g; ++i) {
       wgt_offset_tmp += groups[i] * block_len;
     }
     for (int k = 0; k < k_; ++k) {

paddle/fluid/operators/math/padding.h

@@ -122,8 +122,8 @@ void PaddingGradFunctor(int rank, const framework::ExecutionContext& context,
 inline bool IsSymmetricPadding(const std::vector<int>& pads,
                                const int data_dim) {
   bool is_sys_pad = true;
-  if (pads.size() == data_dim * 2) {
+  if (static_cast<int>(pads.size()) == data_dim * 2) {
-    for (size_t i = 0; i < data_dim; ++i) {
+    for (int i = 0; i < data_dim; ++i) {
       if (pads[2 * i] != pads[2 * i + 1]) {
         is_sys_pad = false;
         return is_sys_pad;

paddle/fluid/operators/pool_op.h

@@ -65,8 +65,8 @@ inline void UpdatePadding(std::vector<int>* paddings, const bool global_pooling,
                           const std::vector<int>& ksize) {
   // set padding size == data_dims.size() * 2
   auto data_shape = framework::vectorize<int>(data_dims);
-  if (paddings->size() == data_dims.size()) {
+  if (static_cast<int>(paddings->size()) == data_dims.size()) {
-    for (size_t i = 0; i < data_dims.size(); ++i) {
+    for (int i = 0; i < data_dims.size(); ++i) {
       int copy_pad = *(paddings->begin() + 2 * i);
       paddings->insert(paddings->begin() + 2 * i + 1, copy_pad);
     }

paddle/fluid/operators/split_op.h

@@ -94,7 +94,7 @@ static inline std::vector<framework::DDim> UpdateOutsDims(
             framework::make_ddim(sections), in_dims, axis);
       }
     }
-    for (size_t i = 0; i < outs_number; ++i) {
+    for (int i = 0; i < outs_number; ++i) {
       outs_dims[i][axis] = sections[i];
     }
   }

paddle/fluid/operators/uniform_random_op.h

@@ -44,7 +44,7 @@ inline std::vector<int64_t> GetNewDataFromShapeTensor(
                      &cpu_starts_tensor);
       new_data = cpu_starts_tensor.data<int32_t>();
     }
-    for (size_t i = 0; i < new_data_tensor->numel(); ++i) {
+    for (int i = 0; i < new_data_tensor->numel(); ++i) {
       vec_new_data.push_back(static_cast<int64_t>(*(new_data + i)));
     }
     return vec_new_data;