[Bf-cycles] Cycles in Poser
sergey.vfx at gmail.com
Mon Nov 30 15:41:06 CET 2015
There's no exact roadmap for the split kernel yet. It needs some major
review and probably re-writing of some parts. It is behaving quite
unpredictable in certain configurations and doing some non-functional
changes might break existing configurations. I've got high expectance of
memory issues being involved there. This issues are the to be solved solved
first before moving forward IMO.
Implementing that for CPU and/or CUDA might help figuring out the issues,
that's for sure. In order to do that we might want to step back a few
steps, and reduce complexity:
- Disable features
- Disable auto-calculation of work group (this is likely to happen on a
higher level actually)
- Make sure diffuse BSDF scenes are rendering correct on all platforms
After that we can:
- Enable features back
- Add support for missing features in the path tracing (SSS, i.e.)
- Look into implementing branched path tracing
As for out-of-core rendering -- it's an interesting research project, but
wouldn't give it a high priority:
- We can still optimize quite a few of things memory-wise (there's some
redundancy in i.e. pre-aligned triangle storage)
- We can support half-float textures, normal maps storage etc
The issue of out-of-core rendering is that it complicates workflow quite a
bit, requires artists themselves to be more technical in order to prevent
render time being exploded due to lots of cache misses.
On Mon, Nov 30, 2015 at 4:29 PM, Stefan Werner <swerner at smithmicro.com>
> Thank you! We have more changes and fixes that we would like to submit, if
> there is interest in it. From the top of my head, we have texture
> coordinates for area and point light, spiral bucket order and optional
> usage of host (main) memory for CUDA renders. Note that we have only Cycles
> core code, our features are not integrated into Blender.
> Work in progress is the use of OIIO texture caching for CPU renders, using
> tiled mip maps. Still missing for this feature to be ready are missing
> differentials for samples that do not originate from camera ray hits, such
> as rays starting from a light source or SSS samples.
> Going forward, I was wondering what the road map is for the kernel split?
> We would eventually love to have a split kernel for CUDA, and I think a
> split kernel for the CPU could also allow massive out-of-core scenes*. I
> would not mind contributing to the development of the split kernel myself,
> but would like to coordinate the effort with the Blender team or any other
> interested developers before we end up with duplicate development efforts.
> Working with Cycles was a pleasure overall and in retrospect, we are happy
> with our choice of render engine. We are looking forward to exposing more
> of Cycles’ features in Poser and hope to contribute useful patches to
> Cycles in the future. Thanks go to all developers of Cycles.
> Stefan Werner | Senior Software Engineer; Productivity & Graphics
> Smith Micro Software, Inc. | http://www.smithmicro.com <
> On 11/29/15, 12:34 PM, "bf-cycles-bounces at blender.org on behalf of Thomas
> Dinges" <bf-cycles-bounces at blender.org on behalf of blender at dingto.org>
> >Poser Pro 11 has been released and ships with "SuperFly", which is based
> >on Cycles. (See this mail from Stefan from July:
> >It's really nice to see this, congratulations to the developers at Smith
> >Micro, and also thanks for contributing back some fixes to us!
> >Best regards,
> >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