Website up again

> -----Original Message-----
> From: [hidden email]
> [mailto:[hidden email]] On
> Behalf Of Davide Varvello
> Sent: Monday, October 09, 2006 1:53 PM
> To: The general-purpose Squeak developers list
> Subject: Sieve of Eratosthenes
>
> Hi All,
> I did some performance tests comparing java and squeak. The
> java code is this:
> public class GeneratePrimes {
> public static double generate(int n) {
> long start = System.currentTimeMillis();
> boolean sieve[] = new boolean[n];
> Arrays.fill(sieve, true);
> sieve[0] = false;
> sieve[1] = false;
> for (int i = 2; i < n; i++) {
> if (sieve[i]) {
> for (int j = 2 * i; j < n; j += i) {
> sieve[j] = false;
> }
> }
> }
>
> return (System.currentTimeMillis() - start);
> }
>
> The Squeak one is:
>
> runOn: aNumber
> | sieve |
>
> sieve := Array new: aNumber withAll: true.
> sieve at: 1 put: false.
>
> (2 to: aNumber) do: [:i | (sieve at: i) ifTrue: [(i to:
> aNumber by: i) do: [:k | sieve at: k put: false] ]
>
> The Squeak code is better (of course) but on a number of
> 1000000 Java spends 94 msec instead Squeak 2650!
> Am I using the right approach or should I change something in
> my squeak code to make it perform better (but preserving clearness)?

> Hi All,
> I did some performance tests comparing java and squeak. The java code is this:
> public class GeneratePrimes {
> public static double generate(int n) {
> long start = System.currentTimeMillis();
> boolean sieve[] = new boolean[n];
> Arrays.fill(sieve, true);
> sieve[0] = false;
> sieve[1] = false;
> for (int i = 2; i < n; i++) {
> if (sieve[i]) {
> for (int j = 2 * i; j < n; j += i) {
> sieve[j] = false;
> }
> }
> }
>
> return (System.currentTimeMillis() - start);
> }
>
> The Squeak one is:
>
> runOn: aNumber
> | sieve |
>
> sieve := Array new: aNumber withAll: true.
> sieve at: 1 put: false.
>
> (2 to: aNumber) do: [:i | (sieve at: i) ifTrue: [(i to: aNumber by: i) do: [:k | sieve at: k put: false] ]
>
> The Squeak code is better (of course) but on a number of 1000000 Java spends 94 msec instead Squeak 2650!
> Am I using the right approach or should I change something in my squeak code to make it perform better (but preserving clearness)?
>
> Win xp, 3 ghz, Squeak 3.8, jdk 1.4.2
>
> Thanks
>  Davide
>
>
> ------------------------------------------------------------------------
>
>

> Hello Davide,
>
> Monday, October 9, 2006, 3:52:45 PM, you wrote:
>
> DV> The Squeak code is better (of course) but on a number of 1000000
> DV> Java spends 94 msec instead Squeak 2650!
>
> Your post made me very curious, so I ran some experiments in
> VisualWorks (no recent Squeak available here, sorry!).
>
> Your original code took about 1300ms to run.
>
> Not filling the array with true and using nil instead makes it run in
> about 1200ms (make sure that new:withAll: is using from:to:put:).
>
> Interestingly, using "== nil ifTrue:" instead of "isNil ifTrue:"
> offered no advantages.  Using ifNil: was much slower.
>
> Not creating intervals makes it run in 1000ms.
>
> If you want it to run even faster, skip even numbers after 2.  This
> change makes VW run in 900ms.  If you skip even numbers from the sieve
> altogether, then it runs slightly faster at about 870ms.
>
> Using a byte array instead of a regular array (and using 0 and 1 for
> flags) makes it even faster, running in about 720ms.  Using a
> ByteArray is actually closer to what the Java implementation is doing
> (as far as I could tell).
>
> Now, your machine has a 3ghz processor, and must be at least a p4 or
> something similar.  Mine is a p3/600 which is spending about 20% cpu
> playing stuff in Winamp with DSP plugins.  I think it's ok to assume
> that yours is about 10 times faster than mine.  This yields something
> comparable to Java's time, if not better.
>
> --
> Best regards,
>  Andres                            mailto:[hidden email]
>
>