diff --git a/lite/kernels/mlu/bridges/graph.h b/lite/kernels/mlu/bridges/graph.h
index 6cf40675397eef2db96a51e21b76296e6695e317..b927883b6fad570ec4da3dace342929bbd3aab92 100644
--- a/lite/kernels/mlu/bridges/graph.h
+++ b/lite/kernels/mlu/bridges/graph.h
@@ -110,8 +110,8 @@ class Graph {
     if (!disable_batch_size_changeable) {
       constexpr int input_dimNb = 4;
       bool input_dim_mutable[4] = {true, false, false, false};
-      cnmlSetTensorDimMutable(
-          tensor->mlu_tensor(), input_dim_mutable, input_dimNb);
+      CNML_CALL(cnmlSetTensorDimMutable(
+          tensor->mlu_tensor(), input_dim_mutable, input_dimNb));
     }
   }
 
diff --git a/lite/kernels/mlu/subgraph_compute.h b/lite/kernels/mlu/subgraph_compute.h
index 72ab7cd2703d6834f1ab21082feaa71cdea8b27d..5d696a7402a33b3532a3c6c59a2cac3ccfe3daa0 100644
--- a/lite/kernels/mlu/subgraph_compute.h
+++ b/lite/kernels/mlu/subgraph_compute.h
@@ -179,6 +179,8 @@ class SubgraphEngine : public subgraph::Engine {
       auto op = inst.op();
       CHECK(op);
       std::string op_type = op->op_info()->Type();
+      // since cnml's compile api will not return error now, we simply check
+      // op's type
       if (!disable_batch_size_changeable_ &&
           std::find(unsupport_batch_size_changeable_op_type_.begin(),
                     unsupport_batch_size_changeable_op_type_.end(),