With this branch, I was able to robot-build HPL-2.1-foss-2016b.eb on our BG/Q frontend (POWER7) with EASYBUILD_OPTARCH=GENERIC set in the environment. I only had to tweak the FFTW easyconfig a bit (due to the hardcoded --enable-sse2 configure option and a missing libquadmath). The resulting xhpl binary ran fine, though I don't have the possibility to test whether it would also run on a different POWER-based system (i.e., whether it is "generic").

@geimer please sync with develop now that #1974 is merged?

Seems like you'll also need to deal with a merge conflict in systemtools.py...

@boegel: Synced. And git didn't report any conflicts...

@geimer OK, great. GitHub mentioned there was a conflict, but I guess it just got confused.

@JackPerdue Can you take a look at these changes, and provide some feedback?

Looks good to me, but this may impact you (even though the changes look backwards-compatible to me).

A quick test:
gcc -mcpu=native hello.c
gcc -mcpu=powerpc64 hello.c
xlc -qtune=auto -qmaxmem=1 hello.c
all work on our Power7 cluster (our BG/Q is gone).

I didn't do a full test with EB, but it looks good to me at this point..

OK, thanks for the feedback @JackPerdue!

I see no reason to hold this back any further, thanks a lot for tackling this @geimer, going in!