diff --git a/src/operators/kernel/cl/depthwise_conv_kernel.cpp b/src/operators/kernel/cl/depthwise_conv_kernel.cpp
index ad0c9958f817f96139fb93f2a33df0cdec833d35..b84a56ca974fa3aca0f2fe23bb3b9b283eff7a7e 100644
--- a/src/operators/kernel/cl/depthwise_conv_kernel.cpp
+++ b/src/operators/kernel/cl/depthwise_conv_kernel.cpp
@@ -47,7 +47,7 @@ void DepthwiseConvKernel<GPU_CL, float>::Compute(
   int nh = default_work_size[2];
   auto input = param.Input()->GetCLImage();
   auto filter = param.Filter()->GetCLImage();
-  auto output = param.Output();
+  auto output = param.Output()->GetCLImage();
   int stride = param.Strides()[0];
   int offset = param.Offset();
   int input_c = param.Input()->CBlock();
diff --git a/src/operators/kernel/cl/fetch_kernel.cpp b/src/operators/kernel/cl/fetch_kernel.cpp
index fcf087aade2ade7a73924f710349672f213b9307..87a892ee9c2a3ef9fa9584bc5e358ba4b06f7577 100644
--- a/src/operators/kernel/cl/fetch_kernel.cpp
+++ b/src/operators/kernel/cl/fetch_kernel.cpp
@@ -58,7 +58,7 @@ void FetchKernel<GPU_CL, float>::Compute(const FetchParam<GPU_CL> &param) {
   clEnqueueNDRangeKernel(this->cl_helper_.CLCommandQueue(), kernel, 3, NULL,
                          default_work_size.data(), NULL, 0, NULL, NULL);
 
-  memcpy(out->data(), out_cl_tensor.Data(), out->memory_size());
+  memcpy(out->data<float>(), out_cl_tensor.Data<float>(), out->memory_size());
 }
 
 template class FetchKernel<GPU_CL, float>;