diff --git a/mindspore/lite/nnacl/assembly/arm64/MatmulWinogradFp32.S b/mindspore/lite/nnacl/assembly/arm64/MatmulWinogradFp32.S
index 0c4356697f..e0437210d9 100644
--- a/mindspore/lite/nnacl/assembly/arm64/MatmulWinogradFp32.S
+++ b/mindspore/lite/nnacl/assembly/arm64/MatmulWinogradFp32.S
@@ -165,7 +165,7 @@ MatrixMultiplyWinograd:
             b LoopM
     EndLoopM:
         sub sp, sp, #48
-        st1 {v8.4s}, [sp], #16
+        ld1 {v8.4s}, [sp], #16
         ldp x19, x20, [sp], #16
         ldp x21, x22, [sp], #16
     ret
diff --git a/mindspore/lite/nnacl/assembly/fp16/MatmulWinogradFp16.S b/mindspore/lite/nnacl/assembly/fp16/MatmulWinogradFp16.S
index 882a33d851..38f869c8ee 100644
--- a/mindspore/lite/nnacl/assembly/fp16/MatmulWinogradFp16.S
+++ b/mindspore/lite/nnacl/assembly/fp16/MatmulWinogradFp16.S
@@ -199,7 +199,7 @@ MatrixMultiplyWinogradFp16:
 
     EndLoopM:
         sub sp, sp, #48
-        st1 {v8.8h}, [sp], #16
+        ld1 {v8.8h}, [sp], #16
         ldp x19, x20, [sp], #16
         ldp x21, x22, [sp], #16
     ret