Path: utzoo!attcan!uunet!cs.utexas.edu!usc!zaphod.mps.ohio-state.edu!uwm.edu!rutgers!att!dptg!ulysses!andante!alice!andrew
From: andrew@alice.UUCP (Andrew Hume)
Newsgroups: comp.lang.c
Subject: Re: A study in code optimization in C
Summary: reality break
Keywords: memcopy
Message-ID: <11097@alice.UUCP>
Date: 27 Jul 90 04:34:24 GMT
References: <133@smds.UUCP>
Organization: AT&T Bell Laboratories, Murray Hill NJ
Lines: 13



i hate to interject experimental results into a computer
science discussion but i implemented mr harter's `always faster'
memcpy just to see. the machine was an SGI 4D/240 running
irix 3.2.1. the timing is for 100000 copies of n bytes
(n uniform between 0 999). results:
	memcpy 2.81s
	harter 9.02s

in my experience, memcpy is osmething vendors REALLY work on;
to suggest that a simple portable implementation would beat
the system's memcpy implies you have only tried this on batshit systems.