Recommended syntax depends on whether the reference distribution is continuous or discrete and on the sample data set. For both continuous and discrete distributions, there are two syntax options. Teradata recommends option 1 for large data sets stored across multiple nodes and option 2 for small data sets stored on a single node. However, performance ultimately depends on the data itself.
To run the CvM test on discrete distributions, you must use option 2 to avoid incorrect results.