Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site cadovax.UUCP
Path: utzoo!watmath!clyde!bonnie!akgua!sdcsvax!sdcrdcf!trwrba!cadovax!keithd
From: keithd@cadovax.UUCP (Keith Doyle)
Newsgroups: net.micro
Subject: BYTE benchmark bogosity
Message-ID: <286@cadovax.UUCP>
Date: Mon, 12-Nov-84 17:46:42 EST
Article-I.D.: cadovax.286
Posted: Mon Nov 12 17:46:42 1984
Date-Received: Thu, 15-Nov-84 03:57:28 EST
Organization: Contel Cado, Torrance, CA
Lines: 20


Has anyone noticed that the original BYTE article and all of the 
followups on the Sieve of Eras... (forgot how to spell it) Primes
program have been perpetuating a BUG?  At one time, I wanted to see
the primes that it generated.  The algorithm works by initializing
an array of flags, 1-8192 etc. each used to indicate whether or not
the numbers 1-8192 are prime.  The program scans the array, first by
2's and marks all multiples of 2 as non-prime.  It then scans the
array for the next prime, 3, and then scans the array by 3's marking
all multiples as non-prime.   The original program misses one of these
scans (the first one I think) leaving several non-primes marked as
prime.  The count of primes is then incorrect, and this error has
been perpetuated ever since the initial benchmark program was
published.  Not that it matters really, I suppose it's still a
reasonable comparison test, but it does give the wrong answer for
the number of primes from 1-8192 (there are 1029).

Keith Doyle
{ucbvax,decvax}!trwrb!cadovax!keithd