speed of pow() in c++

[Gord P 14]

The pow(x, y) function is a bit of a bottleneck for me.  My exponents have to be doubles so I can't trade it out for multiplication routines.  After searching around, I see that this is a common issue in c++ and there is no simple fix.  I'm just curious if others had come across something to help speed things up.  Accuracy is an issue for me, so approximate solutions aren't appropriate.

[Anders Melander 1815]

I don't use C++ anymore but I assume that like everybody else (Delphi for one) it's using pow(x, y) = exp(y * log(x)) and I doubt that you'll find anything faster than that.

Here's an implementation:

https://github.com/lattera/glibc/blob/master/sysdeps/ieee754/dbl-64/e_pow.c

Looks pretty optimized to me. Of course you'll need a decent compiler to turn it into something that actually takes advantage of the processor features.

[Anders Melander 1815]

Btw, you should probably check out Agner Fog's vector library.

Here's the pow implementation (but RTFM )

https://github.com/vectorclass/version2/blob/f4617df57e17efcd754f5bbe0ec87883e0ed9ce6/vectormath_exp.h#L1493

Edited October 15 by Anders Melander

[Gord P 14]

4 hours ago, Anders Melander said:

Btw, you should probably check out Agner Fog's vector library.

 

4 hours ago, Anders Melander said:

Here's an implementation:

https://github.com/lattera/glibc/blob/master/sysdeps/ieee754/dbl-64/e_pow.c

Thanks Anders, I'll check them out.