[mrcode007]

On x86-64 you are free to manipulate the 16 MSB bits of an address however you like. The only micro-architectural restriction is that the 16 msb bits are all either 0 or 1. Otherwise you get a GPF. One way to tag pointers is mask off 15 higher bits and do zero or sign extend based on 16msb bit prior to memory dereference with MOVZX or MOVSX instruction. No need for shifts and other magic. I think in this scenario the unsigned numbers can retain their plain arithmetic semantics after masking off of a single MSB bit without a need for shifts etc. alternatively when loading/operating on arrays its type can be encoded in the pointer to the array by the above scheme while data can be stored in POD format again with no extra shift etc.