On Dec 20, 2010, at 9:30 AM, Matthew Knepley wrote:
> CUDA config/install should really work. This means clearing up auto download
> of Thrust/Cusp,
Is this actually feasible? I am not sure. The problem is we currently need
to (for the preconditioners) use the dev version obtained via Mecurial. Do we
(1) take a snapshot of that as a tarball and stick it at the PETSc ftp site to
autodownload or do we (2) autodownload the Mecurial version.
The problem with (1) is who is going to roll the new snapshot all the time
(we already had problems with this autodownload being out of date).
The problem with (2) is if the thrust/cusp guys change something that
breaks us then we are autodownloading broken stuff.
I think we should just require the user to install the appropriate CUDA,
thrust and cusp themselves. The installation instructions now tell the user
exactly where to get the packages.
> fixing/understanding 64-bit checks,
I think these checks are ok, please report problems.
> and getting a nightly test that checks it.
Agreed. Satish?
Barry
>
> Matt
>
> On Sun, Dec 19, 2010 at 2:47 PM, Jed Brown <jed at 59a2.org> wrote:
> Decent preallocation for DMComposite.
>
> CUDA for Dense. I've seen two requests for this, I assume it's almost
> trivially easy to support and is an obvious performance win.
>
> Move linear DMMG into PCMG. (I don't have a sense for how long this will
> take.)
>
> Grid sequencing in SNES, this might be too much for 3.2.
>
> Jed
>
>
>> On Dec 19, 2010 10:05 AM, "Barry Smith" <bsmith at mcs.anl.gov> wrote:
>>
>>
>> What do people want to have working in petsc-dev before we do a release?
>> Both big and little things. Let's generate a list and see if it makes sense
>> to do a release soon.
>>
>> Barry
>>
>>
>
>
>
>
> --
> What most experimenters take for granted before they begin their experiments
> is infinitely more interesting than any results to which their experiments
> lead.
> -- Norbert Wiener
