Bug#873468: CUDA 9.0 transition

2018-01-12 Thread Samuel Thibault
Andreas Beckmann, on jeu. 11 janv. 2018 16:47:57 +0100, wrote:
> On 2018-01-11 09:35, Graham Inggs wrote:
> > On 10/01/2018 06:19, Andreas Beckmann wrote:
> >> CUDA 9.0 finally reached experimental. Did someone get around testing
> >> it, yet?
> > 
> > I sync'd it into Ubuntu as soon as it cleared NEW, and have been testing
> > since then.
> 
> > Pycuda 2017.1.1-1 built without modification,
> 
> > I also was able to build eztrace-contrib, hwloc-contrib and
> > starpu-contrib, but haven't done any testing of them.
> 
> Thanks, just uploaded to unstable. I'll take care of the pycuda binNMU
> tomorrow.

{hwloc,eztrace,starpu}-contrib uploaded.

Samuel



Bug#873468: CUDA 9.0 transition

2018-01-11 Thread Andreas Beckmann
On 2018-01-11 09:35, Graham Inggs wrote:
> On 10/01/2018 06:19, Andreas Beckmann wrote:
>> CUDA 9.0 finally reached experimental. Did someone get around testing
>> it, yet?
> 
> I sync'd it into Ubuntu as soon as it cleared NEW, and have been testing
> since then.

> Pycuda 2017.1.1-1 built without modification,

> I also was able to build eztrace-contrib, hwloc-contrib and
> starpu-contrib, but haven't done any testing of them.

Thanks, just uploaded to unstable. I'll take care of the pycuda binNMU
tomorrow.


Andreas



Bug#873468: CUDA 9.0 transition

2018-01-11 Thread Graham Inggs

On 10/01/2018 06:19, Andreas Beckmann wrote:

CUDA 9.0 finally reached experimental. Did someone get around testing
it, yet?


I sync'd it into Ubuntu as soon as it cleared NEW, and have been testing 
since then.  As usual, the Ubuntu drivers require modification [1], so I 
built them locally for now.


Pycuda 2017.1.1-1 built without modification, however I needed to 
uninstall cython and cython3 to avoid a FTBFS.  To run the examples, I 
needed to install g++-6.  I also think I need a new graphics card, as my 
one at work complains that value 'sm_21' is not defined for option 
'gpu-architecture'.


I also was able to build eztrace-contrib, hwloc-contrib and 
starpu-contrib, but haven't done any testing of them.



[1] https://github.com/tseliot/nvidia-graphics-drivers/pull/16



Bug#873468: CUDA 9.0 transition

2018-01-10 Thread Samuel Thibault
Hello,

Andreas Beckmann, on mer. 10 janv. 2018 05:19:29 +0100, wrote:
> I didn't test the other rdepends that are in sid only. But they will
> need sourceful uploads anyway for switching the compiler (to gcc-6/g++-6
> - not the default, but the best we can get and still better than clang-3.x).

starpu-contrib and eztrace-contrib seem to build fine with the switch to
gcc-6.

Samuel



Bug#873468: CUDA 9.0 transition

2018-01-09 Thread Andreas Beckmann
Hi,

CUDA 9.0 finally reached experimental. Did someone get around testing
it, yet?

I'd like to start the transition soon. There are only two rdepends left
in testing (and going to be autorm'ed in 5 days) - both pycuda and
hwloc-contrib seem to be (manually) binNMUable.
I didn't test the other rdepends that are in sid only. But they will
need sourceful uploads anyway for switching the compiler (to gcc-6/g++-6
- not the default, but the best we can get and still better than clang-3.x).


Andreas