[brlcad-tracker] [ brlcad-Bugs-1812560 ] Cutaway view 210x slower

[SourceForge.net <no...@so...>]

Bugs item #1812560, was opened at 2007-10-12 21:15
Message generated for change (Comment added) made by clock3
You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=640802&aid=1812560&group_id=105292

Please note that this message will contain a full copy of the comment thread,
including the initial issue submission, for this request,
not just the latest update.
Category: Raytrace
Group: unexpected behavior
Status: Open
Resolution: None
Priority: 5
Private: No
Submitted By: Karel Kulhavy (clock3)
Assigned to: Nobody/Anonymous (nobody)
Summary: Cutaway view 210x slower

Initial Comment:
OpenBSD 4.0 Pentium M 1500MHz BRL-CAD 7.10.3 from the CVS
rt ronja.g head
rt ronja.g headcut
The headcut, cutaway view of head is 210x slower on my machine to render than the other.


----------------------------------------------------------------------

>Comment By: Karel Kulhavy (clock3)
Date: 2007-10-13 09:33

Message:
Logged In: YES 
user_id=1194787
Originator: YES

File Added: bad.txt

----------------------------------------------------------------------

Comment By: Karel Kulhavy (clock3)
Date: 2007-10-13 09:33

Message:
Logged In: YES 
user_id=1194787
Originator: YES

Did profiling. gprof -E mcount -E _mcount. OK.txt is the fast one, bad.txt
is the slow one.
File Added: OK.txt

----------------------------------------------------------------------

Comment By: John Anderson (johnranderson)
Date: 2007-10-13 05:29

Message:
Logged In: YES 
user_id=1185553
Originator: NO

Typically cutaway images are produced using rt by subtracting (or
intersecting) an RPP with a top level object. Rt performs this operation by
pushing the subtraction (or intersection) down into each region. The RPP
ends up being referenced by every region and also listed in every space
partition box. The effect of this is that every region gets evaluated on
every ray (because every ray hits the RPP which is part of every region).
This significantly slows down the raytrace. In the ronja.g model, a bit
better performance can be obtained by mirroring the cutaway RPP in the Y
direction and intersecting with that rather than doing the subtraction
(this produces smaller region bounding boxes, and so, better performance).
Much better performance could be obtained with a special option to rt for
doing cutaways that specified a cut plane. The calculated ray result could
be modified according to the cut plane before being passed to the shader.

----------------------------------------------------------------------

You can respond by visiting: 
https://sourceforge.net/tracker/?func=detail&atid=640802&aid=1812560&group_id=105292