Nvidia-docker: OpenCL support

[UniqueFool]

It seems, that the symlink to /usr/local/cuda/include/CL is missing, once it is manually set up, OpenCL examples build properly.

Equally, /usr/local/cuda/lib64 must be added to /etc/ld.so.conf (or a corresponding file), to ensure that libOpenCL* can be found

I guess, it might be a good idea to try to build the CUDA/OpenCL examples as part of the update/release process ?

Thanks

[3XX0]

We don't support OpenCL currently, we will probably add it in a future release though.

CUDA libraries are in fact included in ld.so.conf (see https://github.com/NVIDIA/nvidia-docker/blob/master/ubuntu-14.04/cuda/7.5/runtime/Dockerfile#L29)

We provide few CUDA samples images in samples

[UniqueFool]

thanks for clarifying, it would be great if OpenCL could be supported "out of the box", and it may make sense to run a few CUDA/OpenCL examples as regression tests when building/updating the docker image, i.e. as part of the release process

[flx42]

Yes, we definitely need to add more tests; I will try to do that simultaneously with #18

[UniqueFool]

regarding clinfo, I really suggest to check out this version, which seems to support all major ICDs, including parallel installations: https://github.com/Oblomov/clinfo

I have been using this, to test different ICDs (nvidia, amd and intel) with different hardware: https://github.com/elhigu/opencl-testsuite

You can see the results, which are published at: http://elhigu.github.io/opencl-testsuite/

There is a "send-report" make target which will automatically send a JSON request to update the test matrix

[lsb]

I'd like to see this too, but my workaround is minimal, I have success running
nvidia-docker run nvidia/cuda:7.5 /bin/bash -exc "apt-get update && apt-get install -y clinfo nvidia-opencl-icd-352 && clinfo"

[3XX0]

Here you go (pushed in the opencl branch):
https://github.com/NVIDIA/nvidia-docker/blob/opencl/ubuntu-14.04/opencl/runtime/Dockerfile

If you need devel files just install ocl-icd-opencl-dev

[UniqueFool]

Thank you very much for taking care of this, much appreciated !!

[UniqueFool]

we definitely need to add more tests;

@flx42, how about including some of the more popular OpenCL related libraries/platforms, e.g. things like pyOpenCL, numpy or arrayfire etc ?

Most of these already contain a bunch of unit tests using OpenCL, so that would be a straightforward way to grow a library of OpenCL unit tests without having to write them from scratch ?

[flx42]

@UniqueFool This should not be included in the base image, but feel free to build your own images :)

[mjmg]

It seems OpenCL support has not been included in the latest cuda-8.0 runtime dockerfiles for centos 6/7 and ubuntu-16.04.

[msalvaris]

@3XX0 : Any update on OpenCL support for the latest cuda-8.0 runtime dockerfiles?

[juliandroid]

@3XX0 : OpenCL support is missing. I was trying CentOS image and Davinci Resolve software, which requires both Cuda and OpenCL to run.

[bhalonen]

@3XX0 : I have the same request as @juliandroid . Would like nvidia-docker, Centos 6, and OpenCL.

Thank you

[3XX0]

We don't plan to publish OpenCL images for now. We will revisit after 2.0 is out.
In the meantime, creating your own Dockerfile should be too difficult, see:
https://gitlab.com/nvidia/opencl/blob/ubuntu14.04/runtime/Dockerfile

[flx42]

We just published the OpenCL images on DockerHub (finally!).
https://hub.docker.com/r/nvidia/opencl/

[gautam1858]

@flx42 PyOpenCL seems to be not working, this throwing some errors, here is the post describing the problem (https://devtalk.nvidia.com/default/topic/1047065/need-pyopencl-nvidia-docker-container/) can you please help us ?

[lagorsse]

Hi, I am trying to user nvidia opencl docker on GKE, clinfo doesn't detect plateform. The driver seems to be well instaled and the GPU seems detected.

/# /usr/local/nvidia/bin/nvidia-smi
Mon Jun 17 13:43:21 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.79 Driver Version: 410.79 CUDA Version: 10.0 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 Tesla P100-PCIE... Off | 00000000:00:04.0 Off | 0 |
| N/A 36C P0 27W / 250W | 0MiB / 16280MiB | 0% Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes: GPU Memory |
| GPU PID Type Process name Usage |
|=============================================================================|
| No running processes found |
+-----------------------------------------------------------------------------+

ldd /usr/bin/clinfo
linux-vdso.so.1 => (0x00007fff933e0000)
libOpenCL.so.1 => /usr/local/nvidia/lib64/libOpenCL.so.1 (0x00007fb60b652000)
libstdc++.so.6 => /usr/lib/x86_64-linux-gnu/libstdc++.so.6 (0x00007fb60b34e000)
libgcc_s.so.1 => /lib/x86_64-linux-gnu/libgcc_s.so.1 (0x00007fb60b138000)
libc.so.6 => /lib/x86_64-linux-gnu/libc.so.6 (0x00007fb60ad6f000)
libdl.so.2 => /lib/x86_64-linux-gnu/libdl.so.2 (0x00007fb60ab6b000)
libpthread.so.0 => /lib/x86_64-linux-gnu/libpthread.so.0 (0x00007fb60a94d000)
libm.so.6 => /lib/x86_64-linux-gnu/libm.so.6 (0x00007fb60a647000)
/lib64/ld-linux-x86-64.so.2 (0x00007fb60b859000)