From: Charles Hixson <charleshixsn@...>
Date: 2012-10-27T02:34:17+09:00
Subject: Re: bit count or bit set

Kaspar Schiess wrote:
> Hei Charles,
>
> Any number of these bit twiddling hacks here:
> 	http://www-graphics.stanford.edu/~seander/bithacks.html
>
> will translate straight to Ruby. I guess the lookup table approach is
> easy to implement and reasonably fast.
>
> Another strategy would be to implement this in C and load it as an
> extension. Make sure this is really the hotspot of your code before
> optimizing...
>
> kaspar
Right now I'm trying to do the simple optimizations that happen while 
designing.  The reason for the question is that I *don't* want to drop 
into C, so I'm trying to design something that's "fast enough" in a 
higher level language.  And I'm skeptical about bithacks being very fast 
in a high level language.  They are a way to do something if you must, 
but they are really MUCH more appropriate in C.  Or C++, Ada, D, any of 
those.  I'm not even sure, though, that they are appropriate in Java.  
In Java one should probably look for something that has been implemented 
in the JVM.

So.  Part of the design optimization is choosing the implementation 
language.  You need to pick one that has support for the primitive 
operations that you'll be using a lot.  Which means you need to design 
the general approach at the same time that you pick the language.  A 
part of my design turned out to be that I need long lists that will 
support multiple kinds of entries.  This let out, e.g., Java.  (Well, I 
COULD use a list of objects, but that is fighting with the language, a 
thing I try to avoid.)  So this narrowed the choices to a few, Ruby, 
Python, Smalltalk, etc.  So then I go through the other requirements.  
So far Ruby1.9 and Python3 are the leading candidates.  I must admit 
that Ruby's fake parallelism bothers me, but I don't like the way Java 
handles unicode, so JRuby is out.  (I prefer utf8 or 32bit unicode.  
Utf8, because that's what the files are, and it takes less space in 
RAM.  32bit unicode because it's easy to handle.  16bit unicode seems to 
blend the worse characteristics of both.)

P.S.:  ARE Threads going to eventually get real parallelism?  Multiple 
CPUs are becoming much more common.  I understand that originally the 
reason was because many libraries weren't thread-safe, but is that being 
addressed?