Can We Achieve Efficient Diffusion without Self-Attention? Distilling Self-Attention into Convolutions.