diff --git a/src/Function2D.cu b/src/Function2D.cu
index 2d64f99..ddf01c6 100644
--- a/src/Function2D.cu
+++ b/src/Function2D.cu
@@ -703,6 +703,7 @@ CudaMatrix Aurora::sum(const CudaMatrix &aMatrix, FunctionDirection direction ){
                 cudaMalloc((void**)&retData, sizeof(float)*2*fakeCol);
                 auto ret = CudaMatrix::fromRawData(retData,1,fakeCol,1,Complex);
                 sumZAllColKernel<<<fakeCol,256>>>(matData,retData, aMatrix.getDataSize());
+                cudaDeviceSynchronize();
                 float* result_data = nullptr;
                 cudaMalloc((void**)&result_data, sizeof(float)*2);
                 auto ret2 = CudaMatrix::fromRawData(result_data,1,1,1,Complex);
@@ -1349,12 +1350,13 @@ CudaMatrix Aurora::fft(const CudaMatrix &aMatrix, long aFFTSize){
     float* data = nullptr;
 
     cudaMalloc((void**)&data, sizeof(float)*2*bufferSize);
-if (aMatrix.isComplex()){
+    if (aMatrix.isComplex()){
         complexCopyKernel<<<aMatrix.getDimSize(1), 256>>>(aMatrix.getData(), data, needCopySize, aMatrix.getDimSize(0),ColEleCount);
     }
     else{
-    complexFillKernel<<<aMatrix.getDimSize(1), 256>>>(aMatrix.getData(), data, needCopySize, aMatrix.getDimSize(0),ColEleCount);
-}
+        complexFillKernel<<<aMatrix.getDimSize(1), 256>>>(aMatrix.getData(), data, needCopySize, aMatrix.getDimSize(0),ColEleCount);
+    }
+    cudaDeviceSynchronize();
     auto ret = Aurora::CudaMatrix::fromRawData(data,ColEleCount,aMatrix.getDimSize(1),1,Complex);
     ExecFFT(ret,0);
     return ret;
@@ -1373,6 +1375,7 @@ CudaMatrix Aurora::ifft(const CudaMatrix &aMatrix, long aFFTSize){
     cudaMalloc((void**)&data, sizeof(float)*2*bufferSize);
 
     complexCopyKernel<<<aMatrix.getDimSize(1), 256>>>(aMatrix.getData(), data, needCopySize, aMatrix.getDimSize(0),ColEleCount);
+    cudaDeviceSynchronize();
     auto ret = Aurora::CudaMatrix::fromRawData(data,ColEleCount,aMatrix.getDimSize(1),1,Complex);
     ExecFFT(ret,1);
     float colEleCountf = 1.f/ColEleCount;
@@ -1414,6 +1417,7 @@ void Aurora::fftshift(CudaMatrix &aMatrix){
     if (aMatrix.getDimSize(0) % 2 == 0) {
         fftshiftSwapKernel<<<aMatrix.getDimSize(1), 256>>>(
             aMatrix.getData(), aMatrix.getDimSize(0) * aMatrix.getValueType());
+        cudaDeviceSynchronize();  
     } else {
         int copySize = aMatrix.getDimSize(0) / 2 + 1;
         float *data = nullptr;
@@ -1422,11 +1426,13 @@ void Aurora::fftshift(CudaMatrix &aMatrix){
             aMatrix.getData(), data, copySize * aMatrix.getValueType(),
             aMatrix.getDimSize(0) * aMatrix.getValueType(),
             copySize * aMatrix.getValueType());
+        cudaDeviceSynchronize();
         memcpyColKernel<<<aMatrix.getDimSize(1), 256>>>(
             aMatrix.getData() + copySize* aMatrix.getValueType(), aMatrix.getData(),
             (copySize - 1) * aMatrix.getValueType(),
             aMatrix.getDimSize(0) * aMatrix.getValueType(),
             aMatrix.getDimSize(0) * aMatrix.getValueType());
+        cudaDeviceSynchronize();
         memcpyColKernel<<<aMatrix.getDimSize(1), 256>>>(
             data, aMatrix.getData() + (copySize - 1) * aMatrix.getValueType(),
             copySize * aMatrix.getValueType(),
@@ -1449,11 +1455,13 @@ void Aurora::ifftshift(CudaMatrix &aMatrix){
             data, copySize * aMatrix.getValueType(),
             aMatrix.getDimSize(0) * aMatrix.getValueType(),
             copySize * aMatrix.getValueType());
+        cudaDeviceSynchronize();
         memcpyColKernel<<<aMatrix.getDimSize(1), 256>>>(
             aMatrix.getData(), aMatrix.getData() + (copySize) * aMatrix.getValueType(),
             (copySize-1) * aMatrix.getValueType(),
             aMatrix.getDimSize(0) * aMatrix.getValueType(),
             aMatrix.getDimSize(0) * aMatrix.getValueType());
+        cudaDeviceSynchronize();
         memcpyColKernel<<<aMatrix.getDimSize(1), 256>>>(
             data, aMatrix.getData(),
             copySize * aMatrix.getValueType(),