From: kero@...
Date: 2005-10-31T07:08:15+09:00
Subject: Re: Time for built-in Rational and Complex classes?

> > > I don't see why not.  But using GCD as an example the C version would
> > > have to handle fixnums, bignums and arbitrary types that respond to
> > > +,-,*,/,etc.  You could just do a straight translation into C using
> > > rb_funcall(), but if you wanted to see a speed up for the common case
> > > (fixnum and bignum) than it would require handling those cases specially
> > > (using arithmetic in C and the rb_big_*() functions).
> > > Is there a more straight forward solution?
> >
> > I can provide some C implementations of gcd for Fixnums.
> >
> > But first, is there anyone who has profiled Rational and can tell that gcd
> > really is a major time consumer ? What do commonly used denominators look
> > like and how do they affect the speed of gcd ?
> 
> I've no idea of the details, but a major performance boost for
> RubyGems came from using Time instead of Date.  The profile showed
> Integer#gcd as a key time sink, in the context of Date usage (i.e.
> basic usage, not complex manipulation).
> 
> I agree with people's comments that implementing GCD and LCM in C
> would be a good start.

From using Date, I found the same. for other applications, the snippet
below may not be a speed gain. YMMV. I use a rule of thumb that fewer
method calls is faster in Ruby.

I made this and it was faster for me (and then forgot to post it
somewhere)

  class Integer
    # Code without optimizations as in lib/rational.rb
    # this is faster for Date :)
    def gcd(other)
      a, b = self.abs, other.abs
      while b != 0
        a, b = b, a % b
      end
      a
    end
  end
			      
HtH,
Kero.

PS: I hope I get this onto this ML now.
I haven't subscribed for nothing ;)

+--- Kero ------------------------- kero@chello@nl ---+
|  all the meaningless and empty words I spoke        |
|                       Promises -- The Cranberries   |
+--- M38c --- http://members.chello.nl/k.vangelder ---+