diff --git a/CMakeLists.txt b/CMakeLists.txt
index 4904e5d..c173eca 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -1,6 +1,6 @@
 cmake_minimum_required(VERSION 3.5)
 
-project (tkDNN CUDA)
+project (tkDNN)
 set(CMAKE_MODULE_PATH ${CMAKE_MODULE_PATH} ${CMAKE_CURRENT_SOURCE_DIR}/cmake)
 if(UNIX)
 set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -std=c++11 -fPIC -Wno-deprecated-declarations -Wno-unused-variable")
@@ -34,7 +34,7 @@ include_directories(${CUDNN_INCLUDE_DIR})
 # compile
 file(GLOB tkdnn_CUSRC "src/kernels/*.cu" "src/sorting.cu")
 cuda_include_directories(${CMAKE_CURRENT_SOURCE_DIR}/include ${CUDA_INCLUDE_DIRS} ${CUDNN_INCLUDE_DIRS})
-add_library(kernels SHARED ${tkdnn_CUSRC})
+cuda_add_library(kernels SHARED ${tkdnn_CUSRC})
 target_link_libraries(kernels ${CUDA_CUBLAS_LIBRARIES})
 
 
diff --git a/include/tkDNN/test.h b/include/tkDNN/test.h
index f1d44bd..e842269 100644
--- a/include/tkDNN/test.h
+++ b/include/tkDNN/test.h
@@ -40,7 +40,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
         TKDNN_TSTOP
         dim1.print();   
     }
-    for(int i=0; i<outputs.size(); i++) cudnn_out[i] = outputs[i]->dstData;
+    for(int i=0; i<outputs.size(); i++) cudnn_out.push_back(outputs[i]->dstData);
 
     if(netRT != nullptr) {
         tk::dnn::dataDim_t dim2 = net->input_dim;
@@ -51,7 +51,7 @@ int testInference(std::vector<std::string> input_bins, std::vector<std::string>
             TKDNN_TSTOP
             dim2.print();
         }
-        for(int i=0; i<outputs.size(); i++) rt_out[i] = (dnnType*)netRT->buffersRT[i+1];
+        for(int i=0; i<outputs.size(); i++) rt_out.push_back((dnnType*)netRT->buffersRT[i+1]);
     }
 
     int ret_cudnn = 0, ret_tensorrt = 0, ret_cudnn_tensorrt = 0;