[Bug math/28472] pow(10, i) accuracy

["wdijkstr at arm dot com" <sourceware-bugzilla@sourceware.org>]

https://sourceware.org/bugzilla/show_bug.cgi?id=28472

--- Comment #25 from Wilco <wdijkstr at arm dot com> ---
(In reply to b. from comment #23)
> @Wilco, thanks for your comment, would you mind sharing  
> a.) a code pointer,  
> b.) a hint which method / tool is used / would be accepted  
> to compare performance?  
> c.) 'just use a correctly rounded math library',  
> a hint which and how to?  
> ( I'm not! a pro in coding and 'system', just good in  
> spotting weak points. )

The implementation [1] comes from Arm Optimized Routines and is believed to be
the best in the industry. It has been ported to GLIBC, Android and MUSL.

Note pow() doesn't at any point deal with integers. It literally implements pow
(x, y) as exp (log (x) * y) with higher internal precision. You'd have to
increase internal precision further with larger tables and polynomials - but
even that is no guarantee it rounds all cases the way you want. You need
correctly rounded math for that.

Both Optimized Routines and GLIBC have internal tests and benchmarks.

[1] https://github.com/ARM-software/optimized-routines/blob/master/math/pow.c

-- 
You are receiving this mail because:
You are on the CC list for the bug.