Page 1 of 1

Faster Incoherent Ray Traversal Using 8-Wide AVX Instruction

Posted: Thu Sep 05, 2013 10:59 am
by voxelium
http://voxelium.wordpress.com/2013/08/1 ... tructions/

Unfortunately the improvement is small compared to SSE, but at least it's consistently faster and easy to implement.

Re: Faster Incoherent Ray Traversal Using 8-Wide AVX Instruc

Posted: Fri Sep 06, 2013 7:51 am
by spectral
Looks nice,

Will go deeper... but do you think that such strategy can also help on the GPU ?

Re: Faster Incoherent Ray Traversal Using 8-Wide AVX Instruc

Posted: Fri Sep 06, 2013 10:58 am
by voxelium
Thanks! Not sure about the GPU, I think it's worth investigating, but going beyond 8-way branching is probably not a good idea.