[Bf-committers] CUDA compiler open-sourced

[Dalai Felinto]

>> but we still can't compile the full kernel.
http://wiki.blender.org/index.php/Dev:2.6/Source/Render/Cycles/OpenCL
note: __TRANSPARENT_SHADOWS__ is actually working in some ATIs [win and
osx].

Thus the ones that seems to fail to compile in all ATIs are: __TEXTURE__
and __MULTI_CLOSURE__ .
That doesn't mean they work in all ATIs though. e.g. in OSX __SVM__ works
in my computer but not in Jens' ATI.

--
Dalai


2011/12/15 Brecht Van Lommel <brechtvanlommel at pandora.be>

> Here are the details for the AMD issue. It has been reported to AMD,
> and they fixed the first issue we had, but we still can't compile the
> full kernel.
> http://wiki.blender.org/index.php/Dev:2.6/Source/Render/Cycles/OpenCL
>
> For Cycles, I don't really care about OpenCL vs. CUDA languages
> really, the code is such that it can compile for both and no specific
> CUDA feature is needed. The difference is in the compiler
> implementations.
>
> Brecht.
>
> On Thu, Dec 15, 2011 at 10:16 AM, Erwin Coumans <erwin.coumans at gmail.com>
> wrote:
> > AMD open sourced their OpenCL GPU LLVM backend a few days earlier,
> > http://lists.cs.uiuc.edu/pipermail/llvmdev/2011-December/046136.html
> >
> > No matter how hard NVIDIA pushes their CUDA tech,
> > I think it is unlikely that Intel, AMD and others will dropping
> > support for the open standard OpenCL in favor of CUDA.
> >
> > By the way, I hear comments by Brecht that AMD OpenCL is lacking
> > for Cycles. Where are the details on that?
> >
> > Thanks!
> > Erwin
> >
> >
> >
> > On 14 December 2011 23:54, Reuben Martin <reuben.m at gmail.com> wrote:
> >
> >> The devil is in the details. Nobody really knows yet, as they have only
> >> made
> >> the announcement. I don't think you can actually get ahold of anything
> yet.
> >> And they haven't indicated what licenese it is to be released under.
> >>
> >> -Reuben
> >>
> >> On Thursday, December 15, 2011 12:47:32 AM Davis Sorenson wrote:
> >> > To quote from the linked article:
> >> > "NVIDIA today announced that *it will provide the source code* for the
> >> new
> >> > NVIDIA® CUDA® LLVM-based compiler *to academic researchers and
> >> > software-tool vendors*, enabling them to more easily add GPU support
> for
> >> > more programming languages and support CUDA applications on
> alternative
> >> > processor architectures."
> >> >
> >> > While It does mention opening (In the title) and source code, the
> wording
> >> > is a bit strange. Do they mean that it will be "Shared-source" like
> some
> >> > Microsoft products, or do they mean that it will be under an
> open-source
> >> > license? Just thought I would point this out. Either way this is good
> >> news,
> >> > but if it's real open-source it's much better news.
> >> >
> >> > Davis
> >> >
> >> > On Thu, Dec 15, 2011 at 12:42 AM, Reuben Martin <reuben.m at gmail.com>
> >> wrote:
> >> > > With all the headaches of trying to make Cycles work properly with
> >> > > OpenCL, I
> >> > > thought it was interesting that Nvidia has now open sourced with
> CUDA
> >> > > compiler
> >> > > as well as the documentation of the intermediate representation.
> >> > >
> >> > >
> >> > >
> >>
> http://pressroom.nvidia.com/easyir/customrel.do?easyirid=A0D622CE9F579F0
> >> > > 9&version=live&releasejsp=release_157&xhtml=true&prid=831864
> >> > >
> >> > > In theory, this could mean that CUDA could eventually be ported to
> >> > > non-nvidia
> >> > > architectures.
> >> > >
> >> > >
> >> > >
> >> > > -Reuben
> >>
> >> _______________________________________________
> >> Bf-committers mailing list
> >> Bf-committers at blender.org
> >> http://lists.blender.org/mailman/listinfo/bf-committers
> >>
> > _______________________________________________
> > Bf-committers mailing list
> > Bf-committers at blender.org
> > http://lists.blender.org/mailman/listinfo/bf-committers
> _______________________________________________
> Bf-committers mailing list
> Bf-committers at blender.org
> http://lists.blender.org/mailman/listinfo/bf-committers
>