of floating point numbers.
The assumption that a processor's floats use the same order as its integers
is probably unwarranted, but it is better than requiring non-national CPUs
to #define national, as that drags in national-specific code too.
(According to a comment in the code, the national FPU does not do infinity)