Robert Osfield pisze:
Hi Marcin,
I'll leave it until after 2.8 to do a full review.
On reading your post I do wonder if use of doubles internally will
slow down the intersections significantly (due to conversion of Vec3f
to Vecd + the slower calcs), and since most people will just need
single precision it not a price that most people will be happy to pay.
Robert,
I'm not an expert on this but as far as I remember you won't be able to
see any speed difference for float and double calculations for current
CPUs most of the time. If you think that performance lost from
conversions might be signifacant for many users out there, perhaps some
kind of conditional code (perhaps CMake configurable) might be an option?
I also wonder why the present system of using double Matrices and
LineSegmentIntersectors is not sufficient, as the during traversal the
IntersectionVisitor transforms the LineSegmentIntersector into the
local coords of the geometry where the final intersection against
triangles is done, this final stage should have a local origin so
should have reasonable precision.
Is it that your data doesn't have a local origin?
Robert.
Our data have local origin but we need to intersect with pretty long
line segments - I suspect it's where the precision problem might come
from. I tried scaling line segment down just for testing but for 10km
length I still couldn't get any result and we cannot accept it being
shorter.
It looks like I put too much automation during file changes and made a
mistake with vertex pointers for index calculation so you may ignore
last file.
I'll try to reduce number of data conversions and will send you correct
version soon.
Best,
Marcin
_______________________________________________
osg-submissions mailing list
[email protected]
http://lists.openscenegraph.org/listinfo.cgi/osg-submissions-openscenegraph.org
