Sure, it's interesting.
But is it correct? taylor(qq,x,0,120) gives the first term as
-%pi*x^42/2, which is different. Numerical evaluation with
fpprec:5000 seems to agree with that; for example, qq,x=1.0b-100 =>
-1.57b-4200. (Both of these take much more than 0.2 sec.)
Of course, the expression is constructed to cancel nastily, so both
Taylor and Bfloat may be wrong. Can you show that your result is
correct?
-s