[sharikone]

I think that you will find that for subnormal numbers any math library will use the identity function for sin(x) and 1 for cos(x)

[ThePadawan]

Right, but the largest subnormal number in single-precision floats is ~ 10^-38.

That the sin(x) approximation still works well for 10^-1 (with an error of ~0.01%) is the really cool thing!