[Debian-med-packaging] Help needed to build libhmsbeagle using OpenCL

[Ghislain Vaillant]

On 21/10/15 15:25, Andreas Tille wrote:
> Hi Ghislain,
>
> On Wed, Oct 21, 2015 at 02:33:37PM +0100, Ghislain Vaillant wrote:
>> Hi Andreas, my answer inline below
>
> sure. ;-)
>
>>> advantages when using OpenCL.  I personally do not have any experience
>>> with OpenCL and simply added
>>>
>>> 	libpoclu-dev,
>>> 	ocl-icd-opencl-dev
>>>
>>
>> The correct build dependency to build with any OpenCL environment is:
>>
>> ocl-icd-opencl-dev | opencl-dev,
>>
>> And for testing with any OpenCL ICD:
>>
>> pocl-opencl-icd | opencl-icd
>
> OK, I implemented this.
>
>> Regarding solutions, you may want to just disable the subset of the test
>> suite that requires OpenCL for now, or at least surround the execution of
>> the OpenCL tests with OpenCL capability detection. Please have a look at the
>> discussion here for a nice implementation [1].
>>
>> [1] https://lists.debian.org/debian-mentors/2015/08/msg00081.html
>
> Sounds sensible - so I implemented it as well.

Double-checked, yes you have everything in place.

> Unfortunately this also did not worked:
>
> diff --git a/debian/control b/debian/control
> index 8ead827..c05aed6 100644
> --- a/debian/control
> +++ b/debian/control
> @@ -15,9 +15,9 @@ Build-Depends: debhelper (>= 9),
>                  dh-linktree,
>                  libjs-jquery,
>                  libpoclu-dev,
> -               ocl-icd-opencl-dev,
> -               beignet-dev,
> -               mesa-opencl-icd
> +               ocl-icd-opencl-dev | opencl-dev,
> +               pocl-opencl-icd | opencl-icd,
> +               clinfo
>   Standards-Version: 3.9.6
>   Vcs-Browser: https://anonscm.debian.org/cgit/debian-med/libhmsbeagle.git
>   Vcs-Git: git://anonscm.debian.org/debian-med/libhmsbeagle.git
>
> ...
>
> # cat tmp/buildd/libhmsbeagle-2.1.2+20150609/examples/tinytest/tinytest.log
>
> OpenCL error: CL_DEVICE_NOT_FOUND from file <GPUInterfaceOpenCL.cpp>, line 118.
> FAIL tinytest (exit status: 255)

The error comes from their GPU interface abstraction. At that point, it 
may be worth asking whether upstream have tested it with POCL. Chances 
are that they did test with Beignet (since it worked on your machine?) 
but not on POCL. It could also well be that POCL v0.10 is insufficient.

Ghis