Text archives Help
- From: Thiago Ize <
>
- To:
- Cc:
- Subject: [Manta] Re: GPU parallelization of manta
- Date: Tue, 26 Mar 2013 14:06:31 -0600
Hi Doug,
The main time consuming parts of ray tracing are building the
acceleration structure, traversing rays through the acceleration
structure, and then intersecting the actual primitives. For this to
help Manta you'd probably want to minimize the communication back and
forth from CPU to GPU, so you'd probably want to either do the build on
the GPU or all of the traversal and primitive intersection on the GPU.
I think the building might be more self contained and easier to see a
speedup. The "Simpler and faster HLBVH with work queues" paper
(
https://dl.acm.org/citation.cfm?id=2018333) would be really cool to
implement. You can probably get some code from
https://code.google.com/p/slash-sandbox/wiki/NIH to help you out with
parts of the code. There are other papers out there for how to build
different acceleration structures. If you want something even easier
you could just take the binned SAH BVH build from NIH and port it to
Manta. While easier, it will still require some work and will also be
very useful and interesting. In fact, I personally would find the
regular binned SAH BVH builder to be more useful since then the BVH that
gets built is identical to what Manta builds.
Thiago
On 3/26/13 8:59 AM, Doug Gelley wrote:
Hello,
For part of a senior project, my group would like to potentially increase
manta performance by using CUDA and an NVIDIA GPU. We've already been able to
use the CUDA makefiles written by Abe Stephens to compile manta with the nvcc
compiler. What we would like to know is if there are any recommendations of
code that can be pushed to the gpu. We are all new to parallelization and
programming GPUs so any other advice would be greatly appreciated.
Thank you.
Best regards,
Doug Gelley
Archive powered by MHonArc 2.6.16.