diff --git a/paddle/fluid/distributed/ps/service/brpc_ps_server.cc b/paddle/fluid/distributed/ps/service/brpc_ps_server.cc
index 4ca5f9c8207fe5b06fe2dc51f4c791a996e31b3a..129e78d677d5adc75fe6c828c343dbe32f188633 100644
--- a/paddle/fluid/distributed/ps/service/brpc_ps_server.cc
+++ b/paddle/fluid/distributed/ps/service/brpc_ps_server.cc
@@ -136,7 +136,7 @@ std::future<int32_t> BrpcPsServer::SendPServer2PServerMsg(
     int msg_type, int to_pserver_id, const std::string &msg) {
   auto promise = std::make_shared<std::promise<int32_t>>();
   std::future<int> fut = promise->get_future();
-  if (to_pserver_id >= _pserver_channels.size()) {
+  if (static_cast<size_t>(to_pserver_id) >= _pserver_channels.size()) {
     LOG(FATAL) << "to_pserver_id is out of range pservers, which size is "
                << _pserver_channels.size();
     promise->set_value(-1);
diff --git a/paddle/fluid/distributed/ps/service/communicator/communicator.cc b/paddle/fluid/distributed/ps/service/communicator/communicator.cc
index edbfd06d55a5435f3416dd977c4f3aa8bfc01bf7..cf8d289d4c876fad4561ba9a4a039089918abb74 100644
--- a/paddle/fluid/distributed/ps/service/communicator/communicator.cc
+++ b/paddle/fluid/distributed/ps/service/communicator/communicator.cc
@@ -727,7 +727,7 @@ void AsyncCommunicator::PushSparseFromTensorAsync(
         ++input_idx;
       }
     }
-    CHECK(static_cast<size_t>(output_len) == g_tensor->numel());
+    CHECK(static_cast<int64_t>(output_len) == g_tensor->numel());
   }
 
   std::vector<float *> push_g_vec(input_idx, nullptr);
diff --git a/paddle/fluid/distributed/ps/table/barrier_table.cc b/paddle/fluid/distributed/ps/table/barrier_table.cc
index b9d0345313cc3728f828ffecd439f3890a436cb8..f94c095d28d64e9cc52e8543be59495920a9feda 100644
--- a/paddle/fluid/distributed/ps/table/barrier_table.cc
+++ b/paddle/fluid/distributed/ps/table/barrier_table.cc
@@ -42,7 +42,7 @@ int32_t BarrierTable::Barrier(const uint32_t trainer_id,
             << " add trainer id: " << trainer_id;
   }
 
-  if (trainer_ids_.size() < trigger_.load()) {
+  if (static_cast<int>(trainer_ids_.size()) < trigger_.load()) {
     std::vector<uint32_t> diffs(trainer_all_.size());
     auto iter = std::set_difference(trainer_all_.begin(), trainer_all_.end(),
                                     trainer_ids_.begin(), trainer_ids_.end(),
diff --git a/paddle/fluid/distributed/ps/table/ctr_double_accessor.cc b/paddle/fluid/distributed/ps/table/ctr_double_accessor.cc
index 44c672eff61c04511a0e7e4f965b5558a7fddd40..791d3f2f95c6e766d97316c897f1042990977e8a 100644
--- a/paddle/fluid/distributed/ps/table/ctr_double_accessor.cc
+++ b/paddle/fluid/distributed/ps/table/ctr_double_accessor.cc
@@ -234,7 +234,7 @@ int32_t CtrDoubleAccessor::Merge(float** update_values,
         update_value[i] += other_update_value[i];
     }*/
     for (size_t i = 0; i < total_dim; ++i) {
-      if (i != CtrDoublePushValue::SlotIndex()) {
+      if (static_cast<int>(i) != CtrDoublePushValue::SlotIndex()) {
         update_value[i] += other_update_value[i];
       }
     }
diff --git a/paddle/fluid/distributed/ps/table/memory_dense_table.cc b/paddle/fluid/distributed/ps/table/memory_dense_table.cc
index 857850ce50b6a7e93fb83eed813a401c426eab59..3a581b671c4b845e51406711891b6afb6e7a9cd1 100644
--- a/paddle/fluid/distributed/ps/table/memory_dense_table.cc
+++ b/paddle/fluid/distributed/ps/table/memory_dense_table.cc
@@ -276,7 +276,7 @@ int32_t MemoryDenseTable::Load(const std::string& path,
           CHECK(str_len == param_col_ids_.size())
               << "expect " << param_col_ids_.size() << " float, but got "
               << str_len;
-          for (int col_idx = 0; col_idx < str_len; ++col_idx) {
+          for (size_t col_idx = 0; col_idx < str_len; ++col_idx) {
             if (param_col_ids_[col_idx] < 0) {
               continue;
             }
diff --git a/paddle/fluid/distributed/ps/table/memory_sparse_table.cc b/paddle/fluid/distributed/ps/table/memory_sparse_table.cc
index 171853f96672d6eafd015eb9d98703e898bf2904..f6a2ad0e02e44bc2aab16343d00c6829036e9226 100644
--- a/paddle/fluid/distributed/ps/table/memory_sparse_table.cc
+++ b/paddle/fluid/distributed/ps/table/memory_sparse_table.cc
@@ -53,7 +53,8 @@ int32_t MemorySparseTable::InitializeValue() {
   _avg_local_shard_num =
       sparse_local_shard_num(_sparse_table_shard_num, _shard_num);
   _real_local_shard_num = _avg_local_shard_num;
-  if (_real_local_shard_num * (_shard_idx + 1) > _sparse_table_shard_num) {
+  if (static_cast<int>(_real_local_shard_num * (_shard_idx + 1)) >
+      _sparse_table_shard_num) {
     _real_local_shard_num =
         _sparse_table_shard_num - _real_local_shard_num * _shard_idx;
     _real_local_shard_num =
diff --git a/paddle/fluid/distributed/ps/table/sparse_accessor.cc b/paddle/fluid/distributed/ps/table/sparse_accessor.cc
index 081a77cedf792672df0ed554f90279f005ec3d02..a5435064c6bf91c52b6780484e34b5e964b4478e 100644
--- a/paddle/fluid/distributed/ps/table/sparse_accessor.cc
+++ b/paddle/fluid/distributed/ps/table/sparse_accessor.cc
@@ -202,7 +202,7 @@ int32_t SparseAccessor::Merge(float** update_values,
     float* update_value = update_values[value_item];
     const float* other_update_value = other_update_values[value_item];
     for (size_t i = 0; i < total_dim; ++i) {
-      if (i != SparsePushValue::SlotIndex()) {
+      if (static_cast<int>(i) != SparsePushValue::SlotIndex()) {
         update_value[i] += other_update_value[i];
       }
     }
diff --git a/paddle/fluid/distributed/ps/table/ssd_sparse_table.cc b/paddle/fluid/distributed/ps/table/ssd_sparse_table.cc
index dc77a6c6c51e2f223f69789429952c6099fa96c9..76d00f2f017d9975067ac268177a1fa3c1639414 100644
--- a/paddle/fluid/distributed/ps/table/ssd_sparse_table.cc
+++ b/paddle/fluid/distributed/ps/table/ssd_sparse_table.cc
@@ -637,8 +637,9 @@ int32_t SSDSparseTable::Load(size_t start_idx, size_t end_idx,
   size_t mf_value_size =
       _value_accesor->GetAccessorInfo().mf_size / sizeof(float);
 
-  end_idx =
-      end_idx < _sparse_table_shard_num ? end_idx : _sparse_table_shard_num;
+  end_idx = static_cast<int>(end_idx) < _sparse_table_shard_num
+                ? end_idx
+                : _sparse_table_shard_num;
   int thread_num = (end_idx - start_idx) < 20 ? (end_idx - start_idx) : 20;
   omp_set_num_threads(thread_num);
 #pragma omp parallel for schedule(dynamic)
diff --git a/paddle/fluid/distributed/ps/wrapper/fleet.cc b/paddle/fluid/distributed/ps/wrapper/fleet.cc
index bddda8f8fff8acf16d69131162e4b2af23b2b351..d08e0e518e80894e4ca645f2c812778d9e0d5455 100644
--- a/paddle/fluid/distributed/ps/wrapper/fleet.cc
+++ b/paddle/fluid/distributed/ps/wrapper/fleet.cc
@@ -555,10 +555,12 @@ void FleetWrapper::PushSparseFromTensorAsync(
             // in
             // ctr_accessor.h
             push_values.back()[0] = 2;  // TODO(zhaocaibei123): slot
-            push_values.back()[1] =
-                (i >= show_size ? 1 : static_cast<float>(show_tensor[i]));
-            push_values.back()[2] =
-                (i >= clk_size ? 0 : static_cast<float>(clk_tensor[i]));
+            push_values.back()[1] = (static_cast<int>(i) >= show_size
+                                         ? 1
+                                         : static_cast<float>(show_tensor[i]));
+            push_values.back()[2] = (static_cast<int>(i) >= clk_size
+                                         ? 0
+                                         : static_cast<float>(clk_tensor[i]));
             float* data = push_values.back().data() + 3;
             memcpy(data, g + output_len, sizeof(float) * fea_dim);
           }
@@ -582,10 +584,12 @@ void FleetWrapper::PushSparseFromTensorAsync(
           // slot show clk grad... consistent with CtrCommonPushValue defined in
           // ctr_accessor.h
           push_values.back()[0] = 2;  // TODO(zhaocaibei123): slot
-          push_values.back()[1] =
-              (i >= show_size ? 1 : static_cast<float>(show_tensor[i]));
-          push_values.back()[2] =
-              (i >= clk_size ? 0 : static_cast<float>(clk_tensor[i]));
+          push_values.back()[1] = (static_cast<int>(i) >= show_size
+                                       ? 1
+                                       : static_cast<float>(show_tensor[i]));
+          push_values.back()[2] = (static_cast<int>(i) >= clk_size
+                                       ? 0
+                                       : static_cast<float>(clk_tensor[i]));
           float* data = push_values.back().data() + 3;
           memcpy(data, g + output_len, sizeof(float) * fea_dim);
         }
diff --git a/paddle/fluid/distributed/test/ctr_accessor_test.cc b/paddle/fluid/distributed/test/ctr_accessor_test.cc
index bb25fd699166575a15f9e901a2b33c19948b0d16..0e0acec164e47b4fa053f6863ee90f50e7be85fb 100644
--- a/paddle/fluid/distributed/test/ctr_accessor_test.cc
+++ b/paddle/fluid/distributed/test/ctr_accessor_test.cc
@@ -239,7 +239,7 @@ TEST(downpour_feature_value_accessor_test, test_update) {
     push_v.show = grad[i][1];
     push_v.click = grad[i][2];
     push_v.embed_g = grad[i][3];
-    for (int j = 0; j < parameter.embedx_dim(); ++j) {
+    for (unsigned int j = 0; j < parameter.embedx_dim(); ++j) {
       push_v.embedx_g.push_back(grad[i][4 + j]);
     }