Re: [Boost-bugs] [Boost C++ Libraries] #12066: Accuracy of modified Bessel functions

Subject: Re: [Boost-bugs] [Boost C++ Libraries] #12066: Accuracy of modified Bessel functions
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2016-03-13 18:15:24

#12066: Accuracy of modified Bessel functions
------------------------------------------+--------------------------
  Reporter:  Pavel Holoborodko <pavel@…>  |      Owner:  johnmaddock
      Type:  Library Submissions          |     Status:  new
 Milestone:  To Be Determined             |  Component:  math
   Version:  Boost 1.61.0                 |   Severity:  Optimization
Resolution:                               |   Keywords:
------------------------------------------+--------------------------
Changes (by johnmaddock):

 * cc: pbristow@… (added)

Comment:

 Yes we're obviously interested, and yes a PR would be fine.

 We already have 35-digit test data and the tools to measure both
 performance and accuracy, so a before and after shoot-out is easy enough.
 I do like your scatter plots for presenting this kind of information BTW.

 We don't use GSL to compute test values BTW - all the other libraries I've
 tested tend to have some serious issues over some parts of their domain.
 Rather we do comparative tests of both our lib and various others against
 our high-precision reference data.

 If you have a list of test cases that generate the outlier results that
 could be useful to (obviously I could go hunting for them, but I'm hoping
 you can see them in your data right away).  I suspect the difference
 between our results (and this is probably a weakness in our data), is that
 since we test at float precision as well as double and long double, and
 also insist on binary-exact input values, none of our input test values
 have more than 23-bit precision.

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/12066#comment:4>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.