cexp error

[Stephen Montgomery-Smith]

> % rcexp: max_er = 0x1658ca3e2ce252 11716177.9430, avg_er = 0.197,
> #>=1:0.5 = 15750254:115895992
>
> Everything except this works fairly well.  cexp() passed tests on only
> 2**12 x 2**12 args.

Do you have any idea why you are getting this huge error?

All I could think of is that cos(y) or sin(y) is subnormal, and exp(x) 
is large.  Is it possible for sin(y) or cos(y) to be subnormal if, say, 
y is close to a non-zero multiple of PI/2?  In other words, can rem_pio2 
give subnormal result (unless of course y is subnormal, but I would 
think that wouldn't generate the above error).