[Algorithms] re: vipm instancing

[Paul F. <pa...@ar...>]

> >I thought of maybe using some kind of linked triangle list with the extra
> >tris being paged in, stuck in a different chunk of ram then linked to the
> >last chunk. Bad fragmentation and cache trouble I fear, though...
>
> Why do you say that?  You could have a chunk of ram, say 2 MB, which you
> reserve for all your VIPM index lists.  Then you could write your own,
> simple memory manager for the VIPM index lists.  You actually wouldn't need
> linked lists, you could implement a "pushing" manager where you moved other
> lists around to make room for you to just grow your list linearly.

Wouldn't that be slow? If you have a mesh currently containing 50,000 tris
and you need to up it but havent the ram in the current location, you'd need
to move 50,000*3*2 (~300k) bytes of ram to some other location. Surely that
would ruin the cache etc?

How much allowance would you make in terms of padding each time you
allocated a new chunk of ram for the higher lod mesh?

Cheers, Paul.