Workaround for Intel compiler issue mentioned in #215

[trb413]

Fixes #215 for Intel 17+. I tested with 16.0.3, 17.0.0, 17.0.2, and 18.0-beta, and life's (mostly) good. With 16.0.3 and 17.0.0, I ran into some other issues (see #177) so the compilation doesn't finish -- BUT! It gets past the matrix classes, which is the problem in #215 . 17.0.2 and 18.0-beta compiled just fine and the tests passed for me. Likewise, compiling with GCC 6.1.0 and 7.1.0 is still good.

[jeffhammond]

I will test this extensively and then merge.

[trb413]

Thanks. Admittedly, I just tested the Elemental configuration that I happen to use anyway on the suite of compilers I had in front of me, so it definitely isn't rigorous across configuration options or compiler versions. Hopefully the fix works. Meanwhile, I'm trying to come up with a MWE that illustrates this bug without all the preprocessor indirection (misdirection?) in Elemental. Hopefully that will be helpful to someone.

[jeffhammond]

I'm not seeing any issues here.

[jeffhammond]

I tested Intel 18.0.1, GCC 5.4.1, and Clang 4.0.0, which together with Travis covers everything we care about. Elemental deprecated support for IBM compilers years ago and has never attempted any other toolchains (e.g. PGI and Cray, although I added CMake support for CCE a while back).

I think it's perfectly reasonable to require the use of Intel 17.0.2 or later.

[jeffhammond]

And thanks to both @trb413 and @benson31 for fixing this! It's pretty sad that so much effort went into dealing with something that had such a simple solution. I wish I understood C++ well enough to fix stuff like this.