Path: csiph.com!x330-a1.tempe.blueboxinc.net!newsfeed.hal-mli.net!feeder1.hal-mli.net!feeder.news-service.com!newsfeed.xs4all.nl!newsfeed6.news.xs4all.nl!xs4all!post.news.xs4all.nl!not-for-mail
MIME-Version: 1.0
In-Reply-To: <1313968634.3135.14.camel@thegeorge>
References: <4e513ceb$0$23863$e4fe514c@news2.news.xs4all.nl> <1313947658.3424.3.camel@thegeorge> <mailman.282.1313951079.27778.python-list@python.org> <roy-D16F14.14525021082011@news.panix.com> <1313968634.3135.14.camel@thegeorge>
Date: Mon, 22 Aug 2011 00:37:45 +0100
Subject: Re: relative speed of incremention syntaxes (or "i=i+1" VS "i+=1")
From: Chris Angelico <rosuav@gmail.com>
To: python-list@python.org
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: quoted-printable
Precedence: list
Newsgroups: comp.lang.python
Message-ID: <mailman.290.1313969874.27778.python-list@python.org>
Lines: 37
NNTP-Posting-Host: 2001:888:2000:d::a6
Xref: x330-a1.tempe.blueboxinc.net comp.lang.python:11970

2011/8/22 Andreas L=F6scher <andreas.loescher@s2005.tu-chemnitz.de>:
> How would such an jump table work to behave the same liek a
> switch-case-statement?

If your switch statement uses a simple integer enum with sequential
values, then it can be done quite easily. Take this as an example:

        switch (argc)
        {
            case 0: printf("No args at all, this is weird\n"); break;
            case 1: printf("No args\n"); break;
            case 2: printf("Default for second arg\n");
            case 3: printf("Two args\n"); break;
            default: printf("Too many args\n"); break;
        }

I compiled this using Open Watcom C, looked at the disassembly, and
hereby translate it into pseudocode (I'll email/post the full 80x86
disassembly if you like):

1) Check if argc > 3 (unsigned comparison), if so jump to default case.
2) Left shift argc two places, add a constant offset, fetch a pointer
from there, and jump to it - that's the jump table. One JMP statement.
3) Code follows for each case.

Incidentally, the Open Watcom compiler actually turned several of the
cases into offset-load of the appropriate string pointer, and then a
jump to the single call to printf. The fall-through from 'case 2' to
'case 3' works fine, although it means that 'case 2' has to be
de-optimized from that one simplification.

This type of optimization works best when the case values are
sequential. (If I remove the 'case 0', the compiler decrements argc
and proceeds to continue as above.) Otherwise, the jump table has to
have a lot of copies of the "default" pointer.

Chris Angelico