doubles and cos(pi / 2)

[Henry Cejtin]

re cos(pi/2), I'm pretty sure that IEEE does not require sin or cos to be done
`good to the last bit' as they do with +, -, * and /.