Accelerate is open source!

[Stanley Seibert]

Hi all,

We just posted an announcement to our blog that Accelerate has been open sourced.  You can go here for the details:

https://www.continuum.io/blog/developer-blog/open-sourcing-anaconda-accelerate

However, I suspect most people on this list are interested in the CUDA library wrappers, which we are now calling pyculib (to mirror the name of pyroclib, our equivalent for AMD GPUs):

https://github.com/numba/pyculib

This is literally the accelerate.cuda library, renamed to pyculib and BSD licensed like Numba.  The biggest change is that this library is now part of the Numba project (like llvmlite), so we can update it much more easily and quickly than before.

We have pyculib packages in the numba channel:

conda install -c numba pyculib

Note that the 1.0.1 release is pinned to cudatoolkit 7.5.  We believe we can lift that restriction, so you will see a 1.0.2 release that can be installed with cudatoolkit 7.5 or 8.0 in a week or so.  These packages will also make their way to the default conda channel at that time.

I know a lot of you have been asking for this, and I'm happy we can finally make this transition.  Please let me know if you have any questions!

[Rémi Lehe]

This is really great!! Many thanks to Continuum for open-sourcing this!

Man, with prange just a few weeks ago, and now this! Numba is more powerful than ever!