ddU in return_check comes out as NaN or silly large when running without SSE with GCC

[kostrzewa]

1. running without explicit SSE is now faster when compiled with GCC than if explicit SSE stuff is enabled (strange, huh?) (GCC 7.4.0)
2. in return_check.data, ddU comes out as NaN or some insanely large number in the return check for reasons which are not clear to me. When compiled with SSE, this is not the case.

Everything else seems to work correctly. In particular, there don't seem to be any problems in the clover derivative or the clover term where _su3_minus_su3 is used as well.

I cannot identify where the problem originates, but it does not seem to affect anything else. Still worth investigating!

[kostrzewa]

I haven't tested other architectures, nor do I really have the time to do so...

[kostrzewa]

seems to be related to threading...