OpenCL Support for Jestson TK1

I have been looking for OpenCL drivers for Jetson TK1 board.
Does anybody know, when Nvidia will support OpenCL drivers for TK1??

I don’t think NVIDIA plans to support it: [url]http://www.bdti.com/InsideDSP/2014/04/29/NVIDIA[/url]

Apparently the Tegra chip is capable of supporting it, but the Jetson TK1’s OS, Linux for Tegra, is not.

What a shame. I had high hopes, but I’ll never get budget to port our OpenCL codebase to CUDA.

I was asking this, and was told:

This is unofficial, but OpenCL could be supported by Q2 2015.
Some of the Khronos employees are also Nvidia employees, which should help push the issue.

Sorry but I doubt you heard that from an NVIDIA employee. NVIDIA supports CUDA on the Jetson TK1 platform, CUDA is a powerful and flexible platform and programming model for developing parallel applications on the GPU. At this time OpenCL is not supported on Jetson TK1.

Well it is not 2015 yet, is it?

There are libraries to link to under /usr/lib/arm-linux-gnueabihf/libOpenCL.so… on my board after doing a apt-get upgrade recently.

And there is the effort to use Clang/LLVM to front end OpenCL to produce ptx output which is loaded via clCreateProgramWithBinary() or some such.

http://blog.jholewinski.org/llvm-3-0-ptx-backend/

Hello all, any updates on this topic? I also need to have the OpenCL support on Tegra board for some research projects…

Noveau on tk1 has opencl ?

If Nouveau on tk1 has OpenCL, then it won’t use the GPU.

Hello all,
Are there any updates to this topic? I am wondering if it is possible to use OpenCL with the Jetson TK1. I’ve read conflicting info about this in different forums.

Thanks,
Carl

OpenCL is not supported on TK1.

Thanks for the reply, DaneLLL! As a follow up question, we have existing OpenCL baselines that we were hoping to test on the TK1. Is there an easy way to port them to CUDA, or would it be a manual process?

Thanks,
Carl

Hi CarlDohrman,
OpenCL on TK1 is not in plan, so it is also uncertain to us. On TK1, we still suggest do direct CUDA programming.