On Wednesday, July 16, 2014 6:57:38 AM UTC-4, Florian Oswald wrote:
>
> i'm looking for ways to improve the performance of ^ as in f(x) = x^1.4
>

A non-integer power like this is already calling the math-library `pow` 
function (assuming `x` is real). 

I've found that it is often faster to call exp(1.4 * log(x)) rather than 
x^1.4 (maybe the latter has to check for more corner cases, e.g. for zero 
powers?)

You could try to get a faster math library (e.g. maybe icc's?  some 
benchmarking would be required).

You could try to rearrange your algorithm to reduce the number of 
non-integer powers that need to be computed (e.g. by grouping terms).

The other thing you could do is to trade off accuracy for speed.  
Especially for a limited range of x, there are various approximations that 
you could use that are probably faster than exp or pow.

Reply via email to