Thread overview  


September 23, 2005 [Submission  std.math] tgamma, lgamma  

 
Attachments:   real tgamma(real x)  returns gamma(x) real lgamma(real x)  returns log(abs(gamma(x))  They come with unit tests and ddoc documentation. Both are based on Cephes, but with much cleaner code. Includes all the relevant unit tests from Cephes, and I've also tested them for consistency with each other and with factorials. Still, the coverage of the tests is not as good as I'd really like. There's also an unoptimised real poly(real x, real[] coeffs) which was necessary for implementation. this is all in statistical.d realtest.d includes a function used in the unit tests.  int consistencyTwoFuncs(real function (real) firstfunc, real function (real) secondfunc, real [] domain ...)  also a function which replaces the one in my previous submission. (Behaviour is identical, but code is much shorter).  int consistencyRealInverse(real function (real) forwardfunc, real function (real) inversefunc, real [] domain ...)  Contents of both files can be added as a cutandpaste job, but since the gamma implementation is so long, maybe it belongs in a file other than std.math. Enjoy! Don. 
October 06, 2005 Re: [Submission  std.math] tgamma, lgamma  

 
Posted in reply to Don Clugston  Thanks, I'm putting them in now. I still need your explicit permission, though. The lawyers like it that way <g>. 
October 06, 2005 Re: [Submission  std.math] tgamma, lgamma  

 
Posted in reply to Walter Bright  Walter Bright wrote:
> Thanks, I'm putting them in now. I still need your explicit permission,
> though. The lawyers like it that way <g>.
You have my full and unrestricted permission for all my submitted code. Please put it in.
(I guess this will actually be a DMC, not a DMD update? Wierd).
For the other functions I submitted, the comments are preDDoc, and the
creal cis(real theta) function should be renamed as creal exp(ireal theta).
A suggestion:
math.d has a lot of stuff in it, and will only get worse with time.
This problem is inherited from C, but since #include "math.h" has to
be changed to import std.math anyway, there does not seem to be any need to retain it.
How about:
math.d > elementary functions (sin, cos, sinh, exp, ln...)
+ lowlevel floating point functions which
rely on the actual bit representation (isnormal, etc)
mathspecial.d > mathematical special functions.
(gamma, erfc, Bessel functions, the productLog() I submitted earlier, etc).
mathstat.d or mathstatistic.d or statistical.d > mathematical statistical functions
(incomplete beta, normal, poisson, chisquared, F, student t distributions, etc)  I have most of these working now, currently improving the unit tests and docs.
By retaining 'math' in the first part of the name it would be easy to find them in the docs.
I even think it could be a good idea to split off the lowlevel bittwiddling functions into a seperate file (maybe ieee.d ?) which would be publicly imported by math.d. It would include the C99 functions that are in fenv.h as well as those from math.h. This would isolate the processor and OSspecific code from the strictly mathematical code, which would make porting easier.
Don

October 06, 2005 Re: [Submission  std.math] tgamma, lgamma  

 
Posted in reply to Don Clugston  On Thu, 06 Oct 2005 10:56:46 +0200, Don Clugston <dac@nospam.com.au> wrote:
> How about:
> math.d > elementary functions (sin, cos, sinh, exp, ln...)
> + lowlevel floating point functions which
> rely on the actual bit representation (isnormal, etc)
> mathspecial.d > mathematical special functions.
> (gamma, erfc, Bessel functions, the productLog() I submitted earlier, etc).
> mathstat.d or mathstatistic.d or statistical.d > mathematical statistical functions
> (incomplete beta, normal, poisson, chisquared, F, student t distributions, etc)  I have most of these working now, currently improving the unit tests and docs.
>
> By retaining 'math' in the first part of the name it would be easy to find them in the docs.
What you really want is:
import std.math;
import std.math.stat;
import std.math.spec;
and so on, and so forth.
But.. I don't think you can name them that way, can you?
Regan

Copyright © 19992016 by Digital Mars ®, All Rights Reserved