Re: round-off error in std::pow(std::complex<T>, double) in C++11

[Marc Glisse <marc dot glisse at inria dot fr>]

On Fri, 25 Jan 2013, Gabriel Dos Reis wrote:
[ pow(z,4) ]
> Why can't we be both fast and correct when the second
> argument is an integer?

I just meant that squaring twice (rounding in between) probably doesn't 
yield the correct result (i.e. computing with infinite precision and only 
rounding once at the end). It is probably still possible to do something 
better than the general case when the second argument is an integer, but 
it isn't obvious.

Now if all platforms are like glibc and provide an imprecise pow 
implementation, we are not losing anything by adding a simple integer 
implementation.

> The standard replaced code with prose/comments -- e.g. the
> contrary of what we would like to do.

On the other hand, we can interpret it as giving us all the freedom we 
need to write those integer overloads.

--
Marc Glisse