Does this actually enable the testing?

No, I gave up with testing under -ffast-math because it breaks a lot of function and I'm notsure we want to actually support that. This PR is just to help for #550 . Just renamed it.

As with the other fix based on volatile, my opinion, independently of it works or not here and now, is that I don't know how prone is to silently breaking in future versions of any compiler.

For me to introduce a fix like this I'd have to be sure to know why it works now and why it didn't before, which means to do some reading of the C++ standard (a boring chore). This doesn't mean that the code is incorrect. Just that I see it as risky code because I don't really know why this is good and the other version was bad.

If this would my codebase, I'd probably explain what is going on with the reorderings either as a comment on the code or on the commit message body.