On Mon, 15 Oct 2001, Brian S. Julin wrote:
[...]
> > [2] E.g.: texture allocation deserves to be done cleanly
>
> Here you hit the nail on the head.
Yep, allocation is a very annoying issue when you want to manage all the
details (pitch != width, and the like).
[...]
> Hopefully we can finish the
> chopping job soon.
Maybe we should try a "divide and conqueer" strategy: split the problem
into smaller problems.
E.g. wrt texture storage management, I guess many chipsets allow to
specify per-texture pitch (at least the Matrox does). Then, of course, the
easiest way is to set (pitch == width) so that any texture uses a
continuous memory area. Then you fall back to a conventional memory
allocation issue (like malloc). Not so easy to tackle (fragmentation,
etc.) but at least some well known algorithms exist.
Maybe we can find some specific solutions to such specific (and hopefully
common) situations - and have a (slow and inefficient and hopefully not
used often) last-resort solution for the really general case.
That's pretty much "sidestepping the problem" as you say - but I'm still
wondering if a general algorithm has been published to address the problem
of "bidimensional" memory allocation in one shot.
Rodolphe
