I’m posting this here since some people may be interested to play around with Kwant and the following:
Intel has just released the “Intel Distribution for Python”. That’s an anaconda-based variant of the scientific Python stack that has been optimized with Intel’s libraries like MKL and TBB.
It seems that Intel’s “distribution” can be installed as a single big download on Linux, Mac & Windows.
Building Kwant on top of this “distribution” should work just fine (without MUMPS). It should be also possible to use it with MUMPS.
What’s perhaps the most interesting is that the “distribution” includes Intel’s MKL which in term includes a variant of the PARDISO direct linear solver. If someone is interested in working on adding a MKL PARDISO solver to Kwant, the new “distribution” could be an easy way to install all the required bits in one go.
It would be also interesting to see how much faster Kwant is with MUMPS+MKL and PARDISO+MKL, and how much better it scales on multiple cores.
Please let us know of any results with regard to this.