Path: utzoo!attcan!uunet!snorkelwacker!tut.cis.ohio-state.edu!att!dptg!ulysses!andante!alice!ark
From: ark@alice.UUCP (Andrew Koenig)
Newsgroups: comp.lang.eiffel
Subject: Re: Benchmark article - Eiffel vs. C++ vs. Smalltalk-80
Keywords: benchmark, Eiffel, C++, performance
Message-ID: <10617@alice.UUCP>
Date: 23 Mar 90 18:58:55 GMT
References: <274@eiffel.UUCP>
Organization: AT&T Bell Laboratories, Liberty Corner NJ
Lines: 39

In article <274@eiffel.UUCP>, kimr@eiffel.UUCP (Kim Rochat) writes:

> Note that Eiffel is just as fast as C++ on the Sun and faster than C++
> on the Tektronix workstation.  This seems ironic since C++ is touted as
> being efficient and the mosaic program exercises features which should
> be efficiently implemented in C.

I thought it would be interesting to see what could be done trivially
to speed up the C++ version of this program, so I profiled it.
My machine does not have the Berkeley random() function, so I
substituted an equivalent function, nrand() from our local C library.

I found the following front-runners:

	%time	function
	20	Row::contrastColorCanAppearAt
	17	nrand
	15	Row::subsequentRow
	11	Row::contrastColor
	10	mcount	(the C profiler itself!)
	10	Row::addRandomColorAt
	8	Row::addBaseColorAt
	3	Row::addContrastColorAt

Of the six C++ functions shown here, four of them were so short that
many C++ programmers would have written them as inline as a matter of
routine, namely Row::contrastColor, Row::addRandomColorAt,
Row::addBaseColorAt, and Row::addContrastColorAt.  Row::subsequentRow
was not called many times and Row::contrastColorCanAppearAt was fairly
large, so inlining wouldn't help either of these much.

With these trivial changes, the run time on my machine (a MicroVax II)
went from 161 to 100 seconds (timed without profiling).  Profiling the
revised program showed 30% spent in nrand, which suggests that it
would be hard to obtain further dramatic improvement without first
obtaining a faster random number generator.
-- 
				--Andrew Koenig
				  ark@europa.att.com