Use of C99 extra long double math functions after r236148
Stephen Montgomery-Smith
stephen at missouri.edu
Sun Aug 12 23:09:23 UTC 2012
On 07/19/2012 06:16 AM, Bruce Evans wrote:
> On Thu, 19 Jul 2012, Stephen Montgomery-Smith wrote:
>
>> I did a ULP test on clog. The test code is attached. (Not the
>> cleanest code, I know, but it does the job.) It needs the mpfr and
>> unuran ports installed.
>>
>> To my shock, I found that under certain circumstances, the ULP in the
>> real part was huge. The problem is when hypot(x,y) is close to 1,
>> because then the real part of clog is close to zero. I was seeing
>> ULPs in the thousands.
>
> Better than GULPs in the thousands :-).
>
> This is not the problem that I first thought it might be.
>
>> I struggled to find a solution, and now I think I have the ULP down to
>> about 2. I am going to work on it more tomorrow to see if I can get
>> ULP down even further.
I have the ULP down to about 1.2 now. I don't see how I can do better,
because I have to invoke log functions twice, and probably each one has
a ULP of about 0.6.
Also I decided to use 1/2 log(x*x+y*y) when x and y are not too large.
I am really rather proud of how I got around the large ULP when
hypot(x,y) is close to 1. I would be glad if any of you could look at
the code when you get a chance.
Also, now that I see how hard clog was, I have more appreciation of
Steve's objections.
More information about the freebsd-numerics
mailing list