pycuda updates (Re: testing packages against nvidia-cuda-toolkit 7.0)

[Tomasz Rybak]

W dniu 18.01.2016, pon o godzinie 02∶03 +0100, użytkownik Andreas
Beckmann napisał:
> On 2016-01-15 22:44, Andreas Beckmann wrote:
> > On 2016-01-09 20:53, Tomasz Rybak wrote:
> > > I've tested PyCUDA and it works without a problem with CUDA 7.0.
> > > What's the plan now? Will you upload CUDA 7.0 to unstable,
> > > or (based on lumin's work) wait for 7.5?
> > 
> > I have not decided, yet, but neither will happen before I have
> > caught up
> > with the drivers.
> > 
> > > I'm asking because there's Python 3.5 transition (for which I'll
> > > need
> > > to build and upload PyCUDA myself), so I'd like to plan
> > > accordingly.
> 
> If you are going to do a manual upload, you should drop the -dbg
> packages (in favor of autogenerated -dbgsyms ones)

Not yet - I've checked this and current setup cannot generate
symbols for Python interpreter. I asked for this particularly
on debian-devel (on 2015-09-20) and Niels Thykier responded
that it's planned but not yet ready.

> 
> > That transition is ongoing. If all you need is a binNMU to pick up
> > the
> > updated dependencies, I can do that, you don't have to do a
> > sourceful
> > upload unless you have other changes or a new upstream as well.
> > (I would build the binNMU locally (with the +b1 binNMU suffix),
> > like I
> > do for other non-free packages from time to time, since these
> > cannot be
> > built on the buildd network.)
> 
> I plan to backport nvidia-cuda-toolkit 6.5 for jessie-backports once
> driver 352 and the newer 6.5 reach testing. Would it be useful to
> have
> pycuda backports as well?

Do you mean just NMU of PyCUDA rebuilt with CUDA 6.5, or uploading
new version of PyCUDA? As for the former, it'd be good. For latter,
I've changed PyCUDA (and PyOpenCL, but it's not yet fully working)
to use pybuilder, so backporting it might be hard.

> 
> PS: don't forget, 7.0 drops i386 support

I know - I was waiting with removal of this to see whether I'd need
to upload PyCUDA during Python transition.

Best regards.

-- 
Tomasz Rybak  GPG/PGP key ID: 2AD5 9860
Fingerprint A481 824E 7DD3 9C0E C40A  488E C654 FB33 2AD5 9860
http://member.acm.org/~tomaszrybak


-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 819 bytes
Desc: This is a digitally signed message part
URL: <http://lists.alioth.debian.org/pipermail/pkg-nvidia-devel/attachments/20160118/4c9382be/attachment-0001.sig>