diff --git a/paddle/fluid/framework/feed_fetch_method.cc b/paddle/fluid/framework/feed_fetch_method.cc
index 6338be75a4b1d3c4caf7a6f7add4d05fec690340..96530b2a3f9cfd9462627a42b2bb0fea98758f92 100644
--- a/paddle/fluid/framework/feed_fetch_method.cc
+++ b/paddle/fluid/framework/feed_fetch_method.cc
@@ -44,6 +44,7 @@ LoDTensor& GetFetchVariable(const Scope& scope, const std::string& var_name,
   // Since we want to fetch LodTensor from a variable, the variable must
   // be created alreadly.
   Variable* g_fetch_value = scope.FindVar(var_name);
+  PADDLE_ENFORCE_NOT_NULL(g_fetch_value, "%s is not found.", var_name);
   PADDLE_ENFORCE(g_fetch_value->IsType<FeedFetchList>(),
                  "Only %s can be invoked by GetFetchVariable",
                  typeid(FeedFetchList).name());
diff --git a/paddle/fluid/framework/operator.cc b/paddle/fluid/framework/operator.cc
index b22523e0f426942333640d9e8b14dcf70ce4a8dc..e15c838f4fbe44fa4f0b543021e97b6b6c70e757 100644
--- a/paddle/fluid/framework/operator.cc
+++ b/paddle/fluid/framework/operator.cc
@@ -989,11 +989,14 @@ void OperatorWithKernel::TransferInplaceVarsBack(
     const Scope& transfer_scope) const {
   for (auto& var_name : inplace_vars) {
     VLOG(3) << "share inplace var " + var_name + " back to it's original scope";
+    auto* origin_var = scope.FindVar(var_name);
+    PADDLE_ENFORCE_NOT_NULL(origin_var, "The var[%s] should not be nullptr.",
+                            var_name);
     auto* original_tensor =
-        GetMutableLoDTensorOrSelectedRowsValueFromVar(scope.FindVar(var_name));
+        GetMutableLoDTensorOrSelectedRowsValueFromVar(origin_var);
     auto* var = transfer_scope.FindVar(var_name);
-    PADDLE_ENFORCE(var != nullptr, "The var[%s] should not be nullptr",
-                   var_name);
+    PADDLE_ENFORCE_NOT_NULL(var, "The var[%s] should not be nullptr.",
+                            var_name);
     auto* transformed_tensor = GetLoDTensorOrSelectedRowsValueFromVar(*var);
     original_tensor->ShareDataWith(*transformed_tensor);
   }
diff --git a/paddle/fluid/memory/allocation/best_fit_allocator.cc b/paddle/fluid/memory/allocation/best_fit_allocator.cc
index 6f3e512fb0b68df5e86eba3e50a255c18f75214f..e3d6c2f511ef083ef9ecc1fe8df96051b2b85cc2 100644
--- a/paddle/fluid/memory/allocation/best_fit_allocator.cc
+++ b/paddle/fluid/memory/allocation/best_fit_allocator.cc
@@ -111,6 +111,8 @@ size_t BestFitAllocator::NumFreeChunks() const {
 }
 void BestFitAllocator::Free(Allocation* allocation) {
   auto* bf_allocation = dynamic_cast<BestFitAllocation*>(allocation);
+  PADDLE_ENFORCE_NOT_NULL(bf_allocation,
+                          "The input allocation is not BestFitAllocation.");
   auto chunk_it = bf_allocation->ChunkIterator();
   PADDLE_ENFORCE(!chunk_it->is_free);
   chunk_it->is_free = true;
diff --git a/paddle/fluid/operators/conv_op.cc b/paddle/fluid/operators/conv_op.cc
index bd788f03e7d666aad7ce6f0c63cea30f029e3491..fd9f156d070bdb1990a2fc9c63305933050e5524 100644
--- a/paddle/fluid/operators/conv_op.cc
+++ b/paddle/fluid/operators/conv_op.cc
@@ -222,7 +222,7 @@ void Conv2DOpMaker::Make() {
       .SetDefault(4096);
   AddAttr<bool>("exhaustive_search",
                 "(bool, default false) cuDNN has many algorithm to calculation "
-                "convolution, whether enable exhaustive search ",
+                "convolution, whether enable exhaustive search "
                 "for cuDNN convolution or not, defalut is False.")
       .SetDefault(false);
   AddComment(R"DOC(
@@ -341,7 +341,7 @@ void Conv3DOpMaker::Make() {
       .SetDefault(4096);
   AddAttr<bool>("exhaustive_search",
                 "(bool, default false) cuDNN has many algorithm to calculation "
-                "convolution, whether enable exhaustive search ",
+                "convolution, whether enable exhaustive search "
                 "for cuDNN convolution or not, defalut is False.")
       .SetDefault(false);
   AddComment(R"DOC(