[sldev] An experiment in paleocomputing
Adam Frisby
adam at gwala.net
Thu Jul 5 02:54:22 PDT 2007
I think there's a good deal of room for improvement in the SL renderer
if it was rewritten again today - among other things geometry shaders
would be a real blessing (hell even doing the transforms in a vertex
shader today may be worth considering).
I'm actually contemplating somewhat, idea of benchmarking a few of these
ideas in a quick & dirty libsl based viewer. I'll post back the results
if I get around to building it.
Adam
Callum Lerwick wrote:
> On Wed, 2007-07-04 at 18:33 -0700, Dzonatas wrote:
>
>>Even still, Quake series still use prebaked lighting and a know set of
>>objects to work with.
>
>
> Also, the total polycount in an entire Quake level is probably less than
> a handful of SL prims. Quake only had to fill out a 320x200 screen.
> There's actually a very good series of articles out there about the
> development of the Quake 1 engine...
>
> http://www.gamedev.net/reference/list.asp?categoryid=40#221
>
> Really, SL is quite fast, except for avatars. Avatars are the killer.
> There was a thread a while back about optimizing the avatar transform,
> is there anyone within LL still working on that? It looks like to really
> gain speed from vectorized matrix multiplies, you have to unroll your
> loops and pipeline an entire array of vertexes at a time. To pull that
> off in SL it looks like some major class restructuring needs to be
> done...
>
>
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> Click here to unsubscribe or manage your list subscription:
> /index.html
More information about the SLDev
mailing list