I would like to ask if anyone here has ever tried to build Psi4 with MKL, and then tried to use Automatic Offloading to make use of Xeon Phi (MIC) accelerator cards? I have access to a small-ish MIC equipped cluster, but I am yet to find a way to put it to good use.
We have not tried this exactly. Literature did not illuminate this as a beneficial thing to do due to the limited bandwidth to and from the KNC cards.
We do have code coming that specifically targets KNL chips. However, thats a different structure than KNC accelerator cards.