@david_chisnall@infosec.exchange
Post #3788838
2026-07-12 16:54 UTC
@zwol@masto.hackers.town This one I would vote no for because everyone has pretty much given up on big endian because it doesn’t really have any benefits (and does have some minor down sides) for ABIs and because adding a byte-swap instruction after a load or before a store costs basically nothing on a modern system, but burning half of the load-store encoding space (which is an enormous part of the total on well-designed ISAs) on a rarely-used feature is incredibly wasteful.
Replies (1)
-
@zwol@masto.hackers.town 2026-07-12 18:34
@david_chisnall@infosec.exchange I have had a *lot* of bad experiences with software that blindly assumes CPU endianness is interchange endianness, enough that I'm tempted to burn that bit of instruction encoding space, just to force low level programmers to make an explicit decision up front. The real counterargument is that for this to help you'd also have to force people to annotate the endianness of every non-register variable in every language besides assembly. And still I think it might be worth it.