update simfactory files to use compilers new enough to avoid C++ math workarounds

[Roland Haas]

Cactus contains workarounds to handle old compilers whose C++ math library is not visible in a well defined manner. Ticket #2407 reports on issues this cases when trying it use regular C++ code.

Thus it would be good to find out what systems and compilers are still affected to see how feasible it is to retire the workarounds.

[Roland Haas]

QB2 uses new enough compiler as of git hash 3392106 "qb: use Intel 18 compiler" of simfactory2

[Roland Haas]

Also affected (according to #2407) are:

• comet
• mike
• shelob
• wheeler

‌

[Roland Haas]

Comet will be retired soon (but not in time for the next release). From the current motd on Comet:

The SDSC Comet supercomputer will be retired from XSEDE service on July 31, 2021.
Details available in the XSEDE news post:
https://portal.xsede.org/user-news/-/news/item/12761

If we remove the workarounds we should add a configure test for the expected behaviour and error out at configure time with a suggestion that gcc 6.X or newer is required for GNU and -gxx-name for the Intel compilers. Not sure what version of PGI or the Cray compiler would work. PGI will be still on some XSEDE systems and seems to be the one with best OpenACC support (so has something to speak for it). No idea about llvm (and hence AMD’s AOCC compiler) yet.

[Roland Haas]

Comet uses new enough compiler as of git hash 85631d3 "comet: use g++/7.2.0 for modern stdc++ library" of simfactory2

[Roland Haas]

Mike / Shelob use new enough compilers as of git hash b32f3e2 "shelob: use gcc 6.4 new enough for modern C++" of simfactory2.

[Roland Haas]

Wheeler uses new enough compilers as of git hash 11f8ac6 "wheeler: use g++/6.4.0 for modern stdc++ library" of simfactory2

[Roland Haas]

This should be all of them.

[Roland Haas]

Done as of as git hash 5739bd2b "smic: update intel to 19.0 and gcc to new version" of simfactory2 and will be in 2022_05.