Address issues with IBM XL builds

[quellyn]

Description
This PR addresses #606, by adding correct detection for the IBM XL Fortran (xlf) recursive option (-qrecur).

Note that the current logic will test ALL possible recursive flags against a target compiler; in the case of xlf this unexpectedly "succeeds" when the Intel-specific flag -recursive is passed in. Actually it doesn't really succeed, but xlf does not correctly report failure, so cmake CheckFortranCompilerFlag ends up tacking -recursive onto CMAKE_Fortran_FLAGS. The resulting attempt to compile with this bogus flag fails.

You can see the errant compiler behavior with a simple example "Hello World" program:

$ xlf_r -recursive -o hello-dumb hello.f90
** helloworld   === End of Compilation 1 ===
1501-510  Compilation successful for file hello.f90.
ld: cannot find -lgcc_s
ld: cannot find -lgcc_s

$ ls
hello.f90  hello.o

$ file hello.o
hello.o: ELF 64-bit LSB relocatable, 64-bit PowerPC or cisco 7500, version 1 (SYSV), not stripped

The xlf compiler produces an intermediate object code that will not link.

P.S. I've sent this patch to the Spack project. They've accepted it but of course it would be best to implement the fixes upstream.

Checklist

• The documentation has been updated.
• If the PR solves a specific issue, it is set to be closed on merge.

[codecov]

Codecov Report

Merging #621 (26a81d1) into master (85ef8de) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##           master     #621   +/-   ##
=======================================
  Coverage    0.00%    0.00%           
=======================================
  Files        1894     1894           
  Lines      184021   184021           
=======================================
  Misses     184021   184021           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 85ef8de...26a81d1. Read the comment docs.

[quellyn]

I should mention @martin-frbg, who helped me debug this issue.

[weslleyspereira]

I think it is a good idea to have a non-empty else() for other compiler IDs. I would considers using either:

else()
  check_fortran_compiler_flag("-recursive" _recursiveFlag)
  check_fortran_compiler_flag("-frecursive" _frecursiveFlag)
  check_fortran_compiler_flag("-Mrecursive" _MrecursiveFlag)
  check_fortran_compiler_flag("-qrecur" _qrecurFlag)
endif()

or:

else()
  message( WARNING "Fortran local arrays should be allocated on the stack."
    " Please assure your compiler guarantees that."
    " See https://github.com/Reference-LAPACK/lapack/pull/188 and references therein." )
endif()

[quellyn]

Sounds good! I like the second choice better; the shotgunning approach didn't work so well with XL ...

[weslleyspereira]

Thanks!