[GpuGraph] add enforce in alloc and free (PaddlePaddle#27)

* solve int overflow

* enforce for alloc and free

paddle/fluid/framework/fleet/heter_ps/heter_comm_inl.h

@@ -135,14 +135,14 @@ void HeterComm<KeyType, ValType, GradType>::create_storage(int start_index,
   auto& nodes = path_[start_index][end_index].nodes_;
   for (size_t i = 0; i < nodes.size(); ++i) {
     platform::CUDADeviceGuard guard(resource_->dev_id(nodes[i].dev_num));
-    allocator->DeviceAllocate(
+    PADDLE_ENFORCE_GPU_SUCCESS(allocator->DeviceAllocate(
         resource_->dev_id(nodes[i].dev_num),
         (void**)&(nodes[i].key_storage),  // NOLINT
-        keylen, resource_->remote_stream(nodes[i].dev_num, start_index));
-    allocator->DeviceAllocate(
+        keylen, resource_->remote_stream(nodes[i].dev_num, start_index)));
+    PADDLE_ENFORCE_GPU_SUCCESS(allocator->DeviceAllocate(
         resource_->dev_id(nodes[i].dev_num),
         (void**)&(nodes[i].val_storage),  // NOLINT
-        vallen, resource_->remote_stream(nodes[i].dev_num, start_index));
+        vallen, resource_->remote_stream(nodes[i].dev_num, start_index)));
     nodes[i].key_bytes_len = keylen;
     nodes[i].val_bytes_len = vallen;
   }
@@ -170,10 +170,10 @@ void HeterComm<KeyType, ValType, GradType>::destroy_storage(int start_index,
   for (size_t i = 0; i < nodes.size(); ++i) {
     platform::CUDADeviceGuard guard(resource_->dev_id(nodes[i].dev_num));
 
-    allocator->DeviceFree(resource_->dev_id(nodes[i].dev_num),
-                          nodes[i].key_storage);
-    allocator->DeviceFree(resource_->dev_id(nodes[i].dev_num),
-                          nodes[i].val_storage);
+    PADDLE_ENFORCE_GPU_SUCCESS(allocator->DeviceFree(
+        resource_->dev_id(nodes[i].dev_num), nodes[i].key_storage));
+    PADDLE_ENFORCE_GPU_SUCCESS(allocator->DeviceFree(
+        resource_->dev_id(nodes[i].dev_num), nodes[i].val_storage));
   }
 #endif
 }