[Bf-cycles] AMD publishes FireRays source code
sergey.vfx at gmail.com
Wed May 18 20:17:47 CEST 2016
I've been checking the sources another day. Was mainly looking into
traversal parts of the code.
There's interesting to see stackless traversal implemented, however from a
quick investigation it only supports AABB nodes (unless i've missed
something?) and the ray-triangle intersection does not seem to be
implementing any of robust intersection checks. There is also no instancing
I hoped to see some vectorization happening in QBVH traversal, but there's
no such a thing, all the children are traversed one by one.
Main advantage f those kernels would be from bucketed ray-to-scene
intersectons. It might be better performance, but we do something really
similar in the split kernel (we use stack there tho). Would be interesting
to see how the FireRays kernels will behave after OBB/MoBlur/Hair support
On Wed, May 18, 2016 at 7:15 PM, Mohamed Sakr <3dsakr at gmail.com> wrote:
> looks interesting!
> I wonder if there is a method to use this into Cycles and test if it will
> improve "I expect Cycles BVH has many areas to improve".
> On Wed, May 18, 2016 at 7:05 PM, Stefan Werner <swerner at smithmicro.com>
>> AMD published the source code for their OpenCL ray tracing kernel and
>> sample renderer:
>> >From skimming the source code, they have embree 2.9 built-in as a
>> fallback for CPU rendering. Haven’t dug deeply into it yet, I haven’t seen
>> any references to motion blur or hair yet though.
>> Bf-cycles mailing list
>> Bf-cycles at blender.org
> Bf-cycles mailing list
> Bf-cycles at blender.org
With best regards, Sergey Sharybin
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Bf-cycles