[Bf-cycles] AMD OpenCL

Sergey Sharybin sergey.vfx at gmail.com
Wed Jul 22 10:12:09 CEST 2015


I can't tell on behalf of AMD team about plans for porting improvements
from OSX to Windows/Linux. We are in contact with the team and when things
becomes more clear we'll let the community to know. All i can say from
myself here -- i'm really willing to have those improvements ported to
other platforms as well.

The split kernel we'll continue improving/fixing. There are some issue in
it which makes it unstable on certain platforms (some bad memory access or
so, makes kernel to stuck). This thing is to be investigated and fixed
there first. After that we'll be looking into supporting wider variety
range of features there (SSS, Volume, branched path tracing). And
eventually we'll need to switch to OpenCL 2.0 and use dynamic kernel
scheduling, which will eliminate roundtrip to CPU for every kernel

As for CUDA -- it should also benefit from split kernel. In fact, one of
the first ever tasks i did after picking op development of the initial
patch was to organice the code in a way which will be most simple for reuse
on CUDA.

it's also to be mentioned here that the work which was done by AMD OSX team
will help us a lot improving performance of the split kernel and remove
some workarounds in the code. So even in the scope of split kernel their
work is really welcome and useful.

As for feature difference, you can check this
https://blender.org/manual/render/cycles/features.html Roughly speaking
currently it's only transparent shadows which is a difference, but it
should be possible to enable volume/SSS. I'm looking into this to happen
for the 2.76.

On Wed, Jul 22, 2015 at 9:49 AM, Nazim Mer <nazim.mer at gmail.com> wrote:

> Hello Blender heads,
> Its great that Blender Cycles is finally working on AMD GPUs.
> However, I got to ask.
> Are the osx amd opencl developers gonna move those improvements made in el
> capitan to the official amd driver for other users on windows and linux?
> If the megakernel works in el Capitan, whilst also be better than the
> split kernel, where is the future for the split kernel?
> I thought the split kernel was going to make its way into the cycles cuda
> code?
> Especially since DUDE from luxrender has seen massive improvements from
> splitting the luxrender kernel?
> Is the split kernel going to be shelfed, or is it going to be the new code
> face of cycles, as i heard that even nvidia gpus are starting struggle with
> the megakernal
> What are the plans for the split kernel....
> "AMD team who's working on OSX drivers for El Capitan (OS X 10.11) did
> really nice work on improving the driver which is now capable of compiling
> and running OpenCL megakernel. The following features are supported:
>    - Hard and rough surface BSDF
>    - Transparent shadows
>    - Motion blur (camera, object, deformation)
>    - Hair
> "
> Are these features working for the split kernel ?
> Kind Regards,
> Nazim
> _______________________________________________
> Bf-cycles mailing list
> Bf-cycles at blender.org
> http://lists.blender.org/mailman/listinfo/bf-cycles

With best regards, Sergey Sharybin
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.blender.org/pipermail/bf-cycles/attachments/20150722/29375c8f/attachment.htm 

More information about the Bf-cycles mailing list