libera/#sicl - IRC Chatlog
Search
2:25:18
moon-child
much though I love scans, I'm sceptical a binary search would be faster than a random-access data structure, at these orders of magnitude
2:27:06
moon-child
'It is very likely that the entry in the break table required to adjust a particular pointer is the same as the entry required to adjust the pointer immediately preceding it' is interesting, but needs to be quantified. I would guess that branch mispredicts kill you
2:29:08
moon-child
one important point: if objects are aligned to a multiple of two words, then you can represent an offset into a 1mb region using just 16 bits. That makes a direct lookup table seem somewhat attractive
2:45:04
moon-child
(I guess it might be possible to do the entry reuse branchlessly--with some fixed cost, of course--there is a 'branch buffering' scheme I keep running into, and you would probably want to make some such affordances anyway for immediate objects and the elderly)