WebCL News

>> Tuesday, April 23, 2013

Here are two items of interest related to WebCL, the browser-based implementation of OpenCL:

  1. I wrote an article that explains how to execute WebCL kernels from Firefox.
  2. The Khronos Group has a Request for Quote (PDF) for a tool capable of validating WebCL kernels. The goal is to prevent bad kernels from crashing the user's computer.
I considered submitting a quote, but I just don't have the time. And I'd imagine validating a kernel is a tricky task.

Read more...

Ten Reasons Why Android Should Support OpenCL

>> Wednesday, April 3, 2013

In an earlier post, I explained why a Google engineer's grievance with OpenCL was mistaken. Now here are ten reasons why Google should make OpenCL the high-performance language of choice on Android:

10. GPU vendors are the driving force behind OpenCL and their tools provide stable, high-speed kernel execution. When GPU technology improves, OpenCL improves immediately.

9. If Android becomes a desktop OS, it should be able to access desktop GPUs and mobile GPUs. OpenCL has broad support on both platforms.

8. Aparapi makes it easy to launch OpenCL kernels from Java. It's open-source and GSS Mahadevan has successfully used it on Android.

7. In addition to CPUs and GPUs, OpenCL kernels can be executed on DSPs and FPGAs. Future high-performance devices will be more likely to support OpenCL than any other language.

6. When OpenCL devices are added to a context, they can work together to execute kernels. With OpenCL, embedded devices have the potential of accessing more powerful systems to crunch data.

5. One of OpenCL's chief advantages is OpenCL-OpenGL interoperability, which allows OpenCL kernels to process OpenGL buffer data before rendering starts. It would take a lot of work to add a similar capability for Renderscript.

4. Currently, native Android developers need to learn x86 and ARM/NEON instructions to ensure high-performance. Intel and ARM are both strong supporters of OpenCL, so if Android adopted OpenCL, native developers would only have to learn one language.

3. Google has put aside O3D in favor of WebGL and the Renderscript graphics engine in favor of OpenGL. If history is any guide, Google will choose OpenCL over Renderscript computation.

2. The general-purpose GPU (GPGPU) community is small and fragmented. It's unlikely that a new, OS-specific language will attract a developer base large enough to justify its existence.

1. If iOS supports OpenCL and Android doesn't, GPU-accelerated apps will run faster on iOS. High-performance mobile computing isn't a big deal yet, but there's no telling what the future may bring.

Read more...

  © Blogger template Werd by Ourblogtemplates.com 2009

Back to TOP