Re: [Plib-devel] Cull Draw and many viewports

[Steve B. <sjb...@ai...>]

Brad Colbert wrote:

> If I have a couple of viewports that I want to render in the same window
> is it possible to cull for each viewport then draw in each viewport?

No.

SSG doesn't cull into some kind of list and then draw the
list - it culls down to a leaf node, then draws that node, then
goes back and does a bit more culling and so on.

You are probably thinking "that's different from Performer" (which
*can* do interleaved Cull and Draw like SSG - but by default does
a full cull into a "display list" (not an OpenGL display list) and
then draws everything in one fell swoop after that.

The Performer approach is interesting when you have multiple CPU's
and a heavy cull load because one CPU can be culling while another
is drawing the results of the previous cull.  However, 99% of PC
users are single-CPU and it's more efficient to interleave cull
and draw because the hardware rendering engine would be idle if
you spent a prolonged period of time doing a full cull at the
start of the frame.  In a single CPU environment, it's better
to get some polygons into the graphics card as soon as possible,
and while it's chewing away at rendering them, you can go back and
do a bit more culling.  Hopefully you produce some more polygons
before the hardware has finished and gone idle - but not so soon
that you fill up the input FIFO on your 3D card and stall the
CPU waiting for it to complete.

Hence, I believe that it's best to interleave culling and drawing.

One possible contrary viewpoint says that it would be better from
a CPU cache perspective to have the cull code resident for as
long as possible - and then the draw code resident for a continuous
stretch...I've heard that argument - but I don't buy it because
there is the issue of data coherency too - having visited a node
in order to cull test it, it's handily in cache ready for drawing.

Having said that SSG doesn't store a "display list" (bad name - but
that's what Performer calls it) for subsequent rendering, in fact
I *do* keep such a list so that I can render translucent leaf nodes
after all the opaque ones.

One other reason to cull to a display list is to allow sorting by
material property - which can (under some circumstances) improve
rendering times...it's really hard to do that well without greatly
increasing the amount of matrix pushing and popping that needs to
be done.

> Would this save any time or is ok to call ssgCullAndDraw for each
> viewport?

No - I don't think it would help much...use an ssgContext (new
feature - not yet documented) for each viewport and call ssgCullAndDraw
for each of them.

-- 
Steve Baker                  http://web2.airmail.net/sjbaker1
sjb...@ai... (home)  http://www.woodsoup.org/~sbaker
sj...@ht...      (work)