Why there is no "cudaDeviceSynchronize" after cufftExecC2C's call?

[zhuwanggg]

CUFFT_CALL(cufftExecC2C(plan, d_data, d_data, CUFFT_INVERSE));

CUDA_RT_CALL(cudaMemcpyAsync(data.data(), d_data, sizeof(data_type) * data.size(),
                                 cudaMemcpyDeviceToHost, stream));

CUDA_RT_CALL(cudaStreamSynchronize(stream));

But in [https://docs.nvidia.com/cuda/cufft/](cuFFT API Reference)，we can get a sample with cudaDeviceSynchronize just behind the call of "cufftExecC2C".

[ayushpareek2003]

there is no need for cudaDeviceSynchronize because cudaStreamSynchronize already ensures all operations in the specified stream are completed, including the cufftExecC2C call. adding cudaDeviceSynchronize would redundantly synchronize all streams, which is unnecessary and less efficient