The build did seem to run, despite memset using fillchar without the explicit casting, so I assumed it did an implicit casting, but the original casting can be kept of course. With this build, my sample tests for 1-byte, 2-byte and 4-byte chars also ran ok, so again, I assumed memset did the casting.

As for the benchmarks, I haven't run them, but yes, it's likely the best way to find out the performance considerations I was asking about.