Some libraries don't support division or square root and in order to
compare all libraries the tests exclude these functions.
All libraries, except double-double is rounded to 28 decimal places during
intermediate calculations (otherwise they would be very slow!).
All libraries, except bigfloat (53-bit) returns visually correct results.
Clicking a button to test the first time is likely a cold interpreted
run. Clicking a second, third, etc time improves JIT compilation and
code should run hot (i.e. faster).