Andreas.Ekstrand++at++saab.se
Thu, 24 Sep 1998 11:29:04 +0200
Thanks for your fast reply.
> > One would think that
> > these tiles were to be loaded after a while, but they aren't.
> are you paging in multiple ASD nodes or
> paging in parts of a single ASD node?
> paging in multiple ASD nodes should work
> just like regular node paging.
> there is a memory leak or fragmentation problem
> that needs to be investigated.
The phenomenon is visible both when I load multiple paged ASD nodes
and when I load a fairly large single paged ASD node. I use the
pfdBuildASD (i.e. pfdBreakTiles) to generate the paging files.
To view the terrain I use asdfly to load a config-file (Multiple
config-files when multiple ASD nodes).
> if you are paging in ASD tiles, then i
> would like to know more about what
> your configuration is, and on which level
> you see missing tiles. we will work on
> fixing these bugs in the next a couple of
> months.
I use a Onyx with Infinite Reality, IRIX 6.2 and Performer 2.2.
I have tried to generate the tiles with different lookaheads,
up to 6 in both directions. I have also tried different number
of lod levels. For example, when I generate one ASD node from a
1001 x 1001 elevation data (50x50 km in 50m resolution) in 9 levels,
some tiles in level 8 and many tiles in level 9 are missing. Different
tiles every time. When I generate four ASD nodes covering the same
area, but with 8 levels each, tiles in level 6, 7 and 8 are missing.
> > I wonder if the Compute process that obviously takes care of both
> > the morphing and the paging (true?) is separated from the App, Cull and
Draw
> > processes
>
> you have to specifically tell performer that you want compute process to
> be forked. the MultiProcess bit mask for computer is 32.
> paging is done in a pfQueue process like the clipmap.
I figured the fork business out for myself, I should have realized it before,
sorry. But do you mean the Compute process has nothing to do with the paging?
Because its length increases with a paging terrain, compared to the same
terrain without paging.
> > But I have memory problems when creating this huge area.
> are you using pfdBuildASD?
> it doesn't manage memory cleverly.
> clipGen does a great job creating large clipmaps.
> we are committed to provide similar support
> to ASD terrain creation.
Yes, I'm using pfdBuildASD. I have converted it a bit though. I changed the
estimated numverts to be (2^(inputlods-1) + 1)^2 instead (wich should be a
correct figure, not an estimate, as I see it). I found the original estimate
to be too large, and the result was that when the pfdBreakTiles calculated
the max and min coordinates for the tile, it found plenty of zero-coordinates
(0,0,0). This, in turn, implied that the bounding box became too large and
the position and movement resolution in asdfly was all wrong. Was this a
correct modification do you think?
I have also messed around a bit with the pruning algorithm, but I don't think
that would have anything to do with this. The pfdBuildASD works fine without
paging.
I haven't looked into clipmapping that much yet, but I'm planning on combining
my large ASD terrain with a large clipmap later. Is clipGen a tool? Does it
come with the Performer 2.2 release?
I have a couple of further questions. Does the evaluation callback function
work? I haven't been able to create my own since I can't understand the
parameters to it. It seems the faceid and the refvertid is the same every
time the function is called. This can't be right, can it? Furthermore,
wouldn't it be nice to have some more parameters to the evaluation function,
such as the current ASD node? Have I gotten it all wrong? I haven't managed
to use the example callback function either.
I'd appreciate any comment, answer or ideas.
Regards,
Andreas Ekstrand
-- -------------------------------------------------------- Andreas Ekstrand |E-mail: Andreas.Ekstrand++at++saab.se Saab Simulation Centre |Phone: +46-13-184042 Linkoping | Sweden | --------------------------------------------------------
This archive was generated by hypermail 2.0b2 on Thu Sep 24 1998 - 02:29:19 PDT