diff --git a/CMakeLists.txt b/CMakeLists.txt
index 83da232f9f56565028ca95876f7ecfbbf355a253..1007e46bf593449373c4b279c0c949024bb445eb 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -7,7 +7,7 @@ option(USE_EXCEPTION "use std exception" ON)
 option(LOG_PROFILE "log profile" ON)
 # select the platform to build
 option(CPU "cpu" ON)
-option(MALI_GPU "mali gpu" OFF)
+option(MALI_GPU "mali gpu" ON)
 option(FPGA "fpga" OFF)
 
 if (CPU)
diff --git a/src/io/io.cpp b/src/io/io.cpp
index 7dc2e3cb0cea5a08a31a00440862678a43c239cc..91f8c5727c91a07a8dceb6f2ac9b848989ec830c 100644
--- a/src/io/io.cpp
+++ b/src/io/io.cpp
@@ -160,6 +160,8 @@ const framework::Program<Dtype, P> Loader<Dtype, P>::LoadProgram(
 }
 
 template class Loader<CPU, Precision::FP32>;
+template class Loader<FPGA, Precision::FP32>;
+template class Loader<GPU_MALI, Precision::FP32>;
 
 #pragma mark - executor
 
@@ -413,5 +415,7 @@ std::vector<typename Executor<Dtype, P>::Ptype> Executor<Dtype, P>::Predict(
 }
 
 template class Executor<CPU, Precision::FP32>;
+template class Executor<FPGA, Precision::FP32>;
+template class Executor<GPU_MALI, Precision::FP32>;
 
 }  // namespace paddle_mobile