On Fri, Mar 22, 2002 at 02:01:53PM -0800, Ian Romanick wrote:
> That said, I personally believe that it is only a matter of time until this
> approval is given. I have been working for some weeks on this project, but
> I have been very quite about what I'm doing. The main reason for my silence
> is to avoid "when can we see your code?" type questions, because I would not
> have any answers.
>From where I sit, 'Texture management' is something which starts in Mesa
and goes through to the kernel.
Personally I'm more interested in what you aren't doing, in that
respect, keep what you are doing to yourself by all means :o)
> I have been working to factor out the /existing/ texture management
> code from the existing drivers. I've been using the MGA and Radeon
> drivers as my proof-of-concept cases. Once that is done, ANYONE can
> easilly experiment with different texture memory management schemes.
There just aren't that many schemes Ian, IMHO.
x mb of card space, n mb of textures this frame.
x > n - don't do anything.
x < n - buy a bigger card, use the slider in the app, use AGP or swap
(n-x)mb of textures this frame. Roughly in order of
quality/performance.
Which textures? NP complete I bet. You have a good general case for most
games where MRU probably gets as close to the optimum (n-x), LRU tends
to hit n mb and that's probably all that is wrong with the current
driver(s). For the non-general case, you're better off being the
application.
That does miss out a lot of improvements we could do, but it's difficult
to surmise whether you are looking at those or not.
(Not that I want to sound negative, I look forward to being proved wrong :))
--
Michael.
_______________________________________________
Dri-devel mailing list
[EMAIL PROTECTED]
https://lists.sourceforge.net/lists/listinfo/dri-devel
