alloca(), efficiency, 386 conventions

Article View: gnu.misc.discuss

Article #2209

#2209

From: David Vinayak Wallace
Date: Mon, 04 Jun 1990 06:02

15 lines
1539 bytes

Dick:

First of all, gnu.misc.discuss is for non-technical discussions.  I'm
pointing followups to bug-gcc.

The answer is that with a little compiler fiddling you can win.  You
make the compiler drop a frame pointer for functions which call
alloca.  Yes, the compiler has to know about alloca, but it's an
optimisation no worse (IMAO) than having the compiler unpack, say,
printf.  I asked rms about this; gcc should be able to do it now.

I don't think that it slows your program down; certainly a few
functions would have to init the fp on entry, but surely that's a lot
faster than calling malloc and free!  You'd have to get concrete
numbers from your application of course.

Technical Details

Message-ID: <GUMBY.90Jun3230213@Cygnus.COM>

Path: arc1-ca.pugleaf.net!gmdzi!unido!mcsun!uunet!shelby!neon!neon!gumby

References: <1990Jun1.200904.7850@ico.isc.com>

Show Full Headers

Complete Article Headers

X-Google-Language: ENGLISH,ASCII-7-bit
X-Google-Thread: 1025b4,f3ff59336294a7a3
X-Google-Attributes: gid1025b4,public
X-Google-Thread: 113d02,728128d361e8c4e4,start
X-Google-Attributes: gid113d02,public
X-Google-ArrivalTime: 1990-06-04 07:53:55 PST
Path: arc1-ca.pugleaf.net!gmdzi!unido!mcsun!uunet!shelby!neon!neon!gumby
From: gumby@Cygnus.COM (David Vinayak Wallace)
Newsgroups: gnu.misc.discuss,gnu.gcc.bug
Subject: alloca(), efficiency, 386 conventions
Message-ID: <GUMBY.90Jun3230213@Cygnus.COM>
Date: 4 Jun 90 06:02:13 GMT
References: <1990Jun1.200904.7850@ico.isc.com>
Sender: news@Neon.Stanford.EDU (USENET News System)
Reply-To: bug-gcc@prep.ai.mit.edu
Followup-To: gnu.gcc.bug
Distribution: gnu
Organization: Cygnus Support
Lines: 15
Posted: Mon Jun  4 07:02:13 1990
In-Reply-To: rcd@ico.isc.com's message of 1 Jun 90 20:09:04 GMT

Reply to Parent Message