Re: [Boost-bugs] [Boost C++ Libraries] #3408: gamma distribution's quantile performance

Subject: Re: [Boost-bugs] [Boost C++ Libraries] #3408: gamma distribution's quantile performance
From: Boost C++ Libraries (noreply_at_[hidden])
Date: 2009-10-05 17:52:10


#3408: gamma distribution's quantile performance
---------------------------------+------------------------------------------
 Reporter: chhenning@… | Owner: johnmaddock
     Type: Tasks | Status: assigned
Milestone: Boost 1.41.0 | Component: math
  Version: Boost 1.40.0 | Severity: Optimization
 Keywords: |
---------------------------------+------------------------------------------

Comment(by johnmaddock):

 Another update to Trunk and I think this is almost fixed now:

 {{{
 Testing igamma 5.849e-007
 Testing igamma-dcd 5.857e-007
 Testing igamma_inv 1.943e-006
 Testing igamma_inv-dcd 1.855e-006
 }}}

 The Boost inverse is somewhat slower as it computes more digits; changing
 the policy to 10 decimal places (which is what the Didonato and Morris
 code promises) then gives:

 {{{
 Testing igamma_inv 1.459e-006
 Testing igamma_inv-dcd 1.855e-006
 }}}

 And your original test code gives:

 {{{
 Boost 30.5s
 Boost (10 digits) 21.3s
 DCDFLIB 27s
 }}}

 Unfortunately, I've found a few other functions that appear to be slower
 than they should be, and the speedup achieved on Linux is very
 disappointing so far :-(

 So still more to do, but this basically fixes the original issue I
 believe. If you're able to give the new code in Trunk a test on your
 machine that would be much appreciated, as results will vary from machine
 to machine a bit...

 HTH, John.

-- 
Ticket URL: <https://svn.boost.org/trac/boost/ticket/3408#comment:9>
Boost C++ Libraries <http://www.boost.org/>
Boost provides free peer-reviewed portable C++ source libraries.

This archive was generated by hypermail 2.1.7 : 2017-02-16 18:50:01 UTC