OpenCL and DirectCompute

[Andrew Whitworth]

Unfortunately there are relatively few places in the Parrot core that
would benefit from GPU acceleration. GPUs tend to focus on extremely
fast floating-point operations, which are mostly important for
math-heavy applications.

We definitely would like to use acceleration where possible (we would
like to follow any path that leads to improved performance), so we are
definitely keeping things like OpenCL in mind. I suspect the best use
for this would be in mathematics-based extensions. I think LLVM has
some features for doing hardware acceleration as well, so once we get
the LLVM-based JIT added to Parrot we will be able to explore this
area in more detail.


--Andrew Whitworth



On Thu, Jun 24, 2010 at 8:14 AM, Tim Bunce <Tim.Bunce at pobox.com> wrote:
> I hope someone's keeping an eye on OpenCL and DirectCompute
> with a view to how Parrot can make use of them:
>
> http://en.wikipedia.org/wiki/OpenCL (Apple,IBM,Intel,Nvidia,AMD)
> http://en.wikipedia.org/wiki/DirectCompute (Microsoft)
>
> Tim.
> _______________________________________________
> http://lists.parrot.org/mailman/listinfo/parrot-dev
>