diff --git a/paddle/fluid/operators/load_combine_op.cu b/paddle/fluid/operators/load_combine_op.cu
index f23f4ea5aaa2bd20bb0f2f6a24d0abcea94b2d0a..2a42c0daa7fc58165e85d851c602a65ec287c905 100644
--- a/paddle/fluid/operators/load_combine_op.cu
+++ b/paddle/fluid/operators/load_combine_op.cu
@@ -21,4 +21,5 @@ REGISTER_OP_CUDA_KERNEL(
     ops::LoadCombineOpKernel<paddle::platform::CUDADeviceContext, float>,
     ops::LoadCombineOpKernel<paddle::platform::CUDADeviceContext, double>,
     ops::LoadCombineOpKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::LoadCombineOpKernel<paddle::platform::CUDADeviceContext, int8_t>,
     ops::LoadCombineOpKernel<paddle::platform::CUDADeviceContext, int64_t>);
diff --git a/paddle/fluid/operators/load_op.cu b/paddle/fluid/operators/load_op.cu
index dcbef0b592275644ea99c4d93ae12f88cf2047ec..90f78110f8f349ebc834570c4fb9f15af24b144d 100644
--- a/paddle/fluid/operators/load_op.cu
+++ b/paddle/fluid/operators/load_op.cu
@@ -20,4 +20,5 @@ REGISTER_OP_CUDA_KERNEL(
     load, ops::LoadOpKernel<paddle::platform::CUDADeviceContext, float>,
     ops::LoadOpKernel<paddle::platform::CUDADeviceContext, double>,
     ops::LoadOpKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::LoadOpKernel<paddle::platform::CUDADeviceContext, int8_t>,
     ops::LoadOpKernel<paddle::platform::CUDADeviceContext, int64_t>);
diff --git a/paddle/fluid/operators/save_combine_op.cu b/paddle/fluid/operators/save_combine_op.cu
index 71476fd802bdd9e617e69d422b1e931178f05322..bc4478b51b111518439fe250a70b8dee0df53ad9 100644
--- a/paddle/fluid/operators/save_combine_op.cu
+++ b/paddle/fluid/operators/save_combine_op.cu
@@ -21,4 +21,5 @@ REGISTER_OP_CUDA_KERNEL(
     ops::SaveCombineOpKernel<paddle::platform::CUDADeviceContext, float>,
     ops::SaveCombineOpKernel<paddle::platform::CUDADeviceContext, double>,
     ops::SaveCombineOpKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::SaveCombineOpKernel<paddle::platform::CUDADeviceContext, int8_t>,
     ops::SaveCombineOpKernel<paddle::platform::CUDADeviceContext, int64_t>);
diff --git a/paddle/fluid/operators/save_op.cu b/paddle/fluid/operators/save_op.cu
index ebb7e1832cbcff350b29e5b48518d8a1509e1d9f..0a778a694e52f146b6cceddb969b8af08f40ef9e 100644
--- a/paddle/fluid/operators/save_op.cu
+++ b/paddle/fluid/operators/save_op.cu
@@ -21,6 +21,7 @@ REGISTER_OP_CUDA_KERNEL(
     save, ops::SaveOpKernel<paddle::platform::CUDADeviceContext, float>,
     ops::SaveOpKernel<paddle::platform::CUDADeviceContext, double>,
     ops::SaveOpKernel<paddle::platform::CUDADeviceContext, int>,
+    ops::SaveOpKernel<paddle::platform::CUDADeviceContext, int8_t>,
     ops::SaveOpKernel<paddle::platform::CUDADeviceContext, int64_t>,
     ops::SaveOpKernel<paddle::platform::CUDADeviceContext,
                       paddle::platform::float16>);