Better input data for finding DEGs

Which is better input data for finding markers, log-normalized data or pearson-residual normalized data?
Additionally I want to know if log-normalized data would be used for input data, should the genes in data be selected?