This has two reasons:
	- It already has to be done like this for C++
	- flang doesn't implement the OpenMP reduction operation yet

This reverts commit 0e1919cd.

This makes the code simpler and it is closer to what the other programs
(including the reference) already do.

Whoops

We are using floops in nas-ft, so it makes sense to use it everywhere

We remove the automatic alignment that fortran does to make the output
more similar to the one from the C / C++ programs.

This function is GNU specific, but it returns floating points directly.
By using it, we don't have to copy or proxy RAND_MAX to the non-C code.

We don't care about compiler options, and some of these parameters are
not portable anyways.