OnAnalyzingGenerativeandDenoisingCapabilities ofDiffusion-basedDeepGenerativeModels