Maximizing Parallelism in the Construction of BVHs...

[spectral]

Here is a new paper from NVidia, once again. A very good research work from Tero Karras

It is about build the BVH (and other structures) on the GPU.

http://research.nvidia.com/publication/ ... -k-d-trees

[trierman]

This is good stuff, very clear writing, good figures and the pseudo code is nice.
So great thanks to Tero for this nice work. I implemented it in a couple of days and it work very nice i
will be back with some timings.
I did some work in this area previously, and i implemented Pantaleonis hlbvh paper, which i think was very hard to implement.
And last year i made a version of Kirill Garanzha's simpler and faster, which is also very nice, but uses some low level cuda stuff which might
make it hard to port to opencl.
But Karras idea is very good and relativly easy to implement compared to the other ones. So i really recommend this and
i am sorry i will miss HPG this year.

[spectral]

Yes gives us numbers

And what about the tree quality... because it is fine to build it quickly... but what are the performance of the BVH then ?
Do you think that it is possible to use it to build SBVH on the GPU then ?

Thanks

Krys