Single precision builds fail test suite

[acmnpv]

A test of a binary build with prec = singleprecision fails the validation tests (both serial and parallel) with hot atoms in the outermost layer of the water sphere, right before causing a shake failure.
Up to this point the tests proceed as expected, with energy values inside the variation of builds with prec = doubleprecision. I have started to look into this and found some more cases of not using the defined precision, but I have not been able to find the root cause of the bug yet.

[esguerra]

What says Irek?

[iszeler]

The singleprecision is not yet finish. I have now more other work, but I will beck to it in short time.

[esguerra]

Yes. In my opinion this is a minor concern, though a concern. The majority of publications which do mention (not many) precision in molecular dynamics programs and no GPU usage do double precision (e.g. Nilsson, J.Comp.Chem, 2008, 15, 1490-1498). That is, checking that double precision is working well throughout is most important. Maybe an aesthetic cleanup of sizes.f90 and a write up in a Q developers manual could be useful for future developers. I'm starting a qdevelopers.tex manual in my development branch where such information could fit.
Perhaps it's high time for another Q-developers meeting?.

[acmnpv]

This issue should be fixed with the latest pull request on commit bda64ba
The issue was caused by the floating point comparison in the trigonometric routines, leading to a possibility of calculating 1/10^-12 as a force scaling factor.
I might still change the new values to be dependent on the build precision.