Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: $Revision: 1.6.2.16 $; site pbear.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!bellcore!decvax!cca!pbear!peterb
From: peterb@pbear.UUCP
Newsgroups: net.arch
Subject: Re: Need 286 "C" benchmark
Message-ID: <600002@pbear.UUCP>
Date: Tue, 28-May-85 11:27:00 EDT
Article-I.D.: pbear.600002
Posted: Tue May 28 11:27:00 1985
Date-Received: Fri, 31-May-85 03:18:52 EDT
References: <426@oakhill.UUCP>
Lines: 51
Nf-ID: #R:oakhill:-42600:pbear:600002:000:1456
Nf-From: pbear!peterb    May 28 11:27:00 1985



There is no such thing as a single benchmark. There are many that exist.
The point it that more than one exists and that you can not test the merit of
a chip based on one benchmark. You have to run them all and analyse the results.
In the analysis you discover the better OVERALL chip, and also find the
strengths and weaknesses of each chip in small areas. These include:

	Register to register moves,
	Add/Sub/Mul/Div speeds (both signed and unsigned) (& BCD for the *86)
	memory to register
	register to memory
	memory to memory
	string ops,
	interupt response,
	i/o response time
	memory fault reload times

Also the benchmarks can not be done in a HLL since there you are not only
testing architecures, but compiler implementations. Take the previous
benchmark of moving arrays in C:

	int a[50000];
	main()
		{
			int i;
			for(i=0 ; i < 49999 ; i++)
				a[i+1]=a[i];
		}

When looking at the code you will see obvious deficiencies in the compiler
output. Some will recompute the address of a[i] twice, whereas others will
recognize the commons sub-expression and collapse it. A better benchmark would
be:

	unsigned int a[50000];

	main()
		{
			register int * j ;
			register int * k = a+50000;

			for ( ; j<k ; )
				*j = *(++j);
		}

This really goes beyond the idea of moving elements in an array and makes the
notion clearer that you are inserting a one element space into an array.

Peter Barada
{ihnp4!inmet|{harvard|cca}!ima}!pbear!peterb