[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: About performance, hash-tables and garbage collection
From: |
Stefan Monnier <address@hidden> |
Subject: |
Re: About performance, hash-tables and garbage collection |
Date: |
20 Sep 2002 17:07:38 -0400 |
User-agent: |
Gnus/5.09 (Gnus v5.9.0) Emacs/21.3.50 |
> 1. I have read that hash-tables have a little overhead and that
> association lists are a little bit faster, when they contain not
> more than about a dozen of key-value pairs. It is _possible_ that I
> have to maintain several hundreds or thousand of key/value pairs,
> but in most cases it will be only a dozen or so. So my question is:
> how much overhead has a hashtable? Makes it sense to add a
> condition, like this:
> (if (> ncolours 15)
> (gethash colour colour-map)
> (cdr (assoc (colour colour-map))))
The `if' overhead will probably dwarf the difference, so I'd say,
just use the hash-table. Most likely any performance problem you'll have
won't come from that anyway.
> 2. I noticed that the garbage collector hits in a little bit too often,
What's too often ? Why do you care ?
Does it use a significant percentage of the running time ?
If yes, look for things like `append' and things like that to see where
you do most allocation.
> though I do not know why. I have a lot of `setq'-statements
> like `(setq pointer (1+ pointer))'. Is it possible that they are
> the cause?
If the code is interpreted, maybe, tho I don't think so.
> If so, what could I do?
If you've byte-compiled your code, then an increment (as shown above) will
not allocate any memory and will thus have no influence on GC.
Stefan