Text archives Help

---

It was my impression that they used masks for SSE'ed packets (sized 4) and ray packet frustrum culling on the kd-tree for much larger packets. I think I will try the latter idea first.

I haven't quite gotten my head around all of the implications of performing this culling beyond primary rays or rays with the same origin--although I haven't read the paper yet. It's also not entirely clear from the language that was used during the talk if they prune their tree's or not, I can't imagine that pruning would scale very well. It also should be noted that their demo only used primary and one shadow ray. (Similar to our boeing demo but without transparency.)

Abe


Peter Shirley wrote:

Hi guys.  I did finally do my refraction code a couple of weeks agao (now
entirely rewritten by Solomon and 4x faster).  He built new ray packets
for reflection and refraction rays and used an index array to asscoiate
the new rays with parent rays (because rays can die due to attenuation,
the second refraction ray may correspond to the 3rd primary ray for example). Sounds like the Intel folks use masking vectors instead to
kill some members of a packet.  That sounds easier to code to me with
unclear performance implications.  Opinions?

Pete

---

• [MANTA] ray packet question, Peter Shirley, 08/04/2005
  • Re: [MANTA] ray packet question, Abe Stephens, 08/05/2005

---

Archive powered by MHonArc 2.6.16.