Improve memset/memcpy
About 10% was spend in memset according to profiles. This vectorizes it with uint64s making it only 1.5% in my profiles. Also using the builtin copy improves performance of memcpy enough to make it disappear in profiles.
About 10% was spend in memset according to profiles. This vectorizes it with uint64s making it only 1.5% in my profiles. Also using the builtin copy improves performance of memcpy enough to make it disappear in profiles.