Depending on how closely related your genomes are, removing sites that have any gaps in them may be removing a ton of truly informative sites. I typically do phylogenies via phylogenomics or multi-gene alignments myself, but we wouldn't have much data if we removed all of the gap containing sites. Can the programs based on SNP input alone handle missing data? If they can (and they should) you'll have better overall resolution by leaving gaps in. Removing regions that are poorly aligned is a different story of course, because you don't want to introduce noise and false-positive signal to the analysis, and the sites at the edge of large "gappy" regions are often misaligned. Similarly for N containing sites, it depends on if your downstream analysis program can handle ambiguous data. If it can then leaving them in doesn't really hurt. If the program doesn't handle N's but does handle gap characters you can typically replace N's with -'s in your alignment. Most phylogeny programs treat the two in the same way when reconstructing the tree.