Re: why gcj is slower than jvm

Re: why gcj is slower than jvm

[Jeff Sturm]

On Thu, 4 Sep 2003, guw wrote:
> Can you tell me  how demonstrate GCJ's runtime performance is better than JVM?

Are you looking for benchmarks?  A google.com search turns up many, one of
the more interesting ones is

http://www.shudo.net/jit/perf/

> and how incarnate the GCJ's runtime performance or compiler optimization?

I'm not sure I understand the question.

Jeff

Re: why gcj is slower than jvm

[Mark Wielaard]

Hi,

On Wed, Sep 03, 2003 at 02:34:50AM -0400, Jeff Sturm wrote:
> Your example is more of a comparison of runtime performance than compiler
> optimization.  From your results one might conclude that the JVM has a
> better MD-5 implementation than libjava, but little else.

The GNU-Crypto hackers have offered their MD5 and SHA-1 implementations.
They are noticably faster. It is on my list of things to merge into GNU
Classpath (which is the upstream library source of libgcj).

You can already use their implemeentation by installing a GCJ-friendly
GNU-Crypto <http://www.gnu.org/software/gnu-crypto/> and replacing the
security.privider.1 in lib/security/classpath.security with
gnu.crypto.jce.GnuCrypto.

Cheers,

Mark

Re: why gcj is slower than jvm

[guw]

[-- Attachment #1: Type: text/plain, Size: 818 bytes --]

Andrew Haley,hello！

Can you tell me  how demonstrate GCJ's runtime performance is better than JVM?
and how incarnate the GCJ's runtime performance or compiler optimization?

Had better give me a little example.

thanks.

======= 2003-09-03 11:17:00 =======

>guw writes:
> > Hello!
> > 	I write a JAVA programme and compile it with GCJ. then I compare GCJ's performance with JVM'performance,but I discover that the procedure to be GCJ compiled is slower than JVM.
> >   why?
>
>My guess would be that libgcj's implementation of SHA-1 is slower than
>that in whatever JVM you were using.  I expect that SHA-1 could be
>made to run very much faster.
>
>Andrew.

= = = = = = = = = = = = = = = = = = = =

				
　　　　　　　　guw
　　　　　　　　guw@koal.com
　　　　　　　　　　2003-09-04


[-- Attachment #2: Rabbit.GIF --]
[-- Type: image/gif, Size: 1488 bytes --]

Re: why gcj is slower than jvm

[guw]

[-- Attachment #1: Type: text/plain, Size: 818 bytes --]

Andrew Haley,hello！

Can you tell me  how demonstrate GCJ's runtime performance is better than JVM?
and how incarnate the GCJ's runtime performance or compiler optimization?

Had better give me a little example.

thanks.

======= 2003-09-03 11:17:00 =======

>guw writes:
> > Hello!
> > 	I write a JAVA programme and compile it with GCJ. then I compare GCJ's performance with JVM'performance,but I discover that the procedure to be GCJ compiled is slower than JVM.
> >   why?
>
>My guess would be that libgcj's implementation of SHA-1 is slower than
>that in whatever JVM you were using.  I expect that SHA-1 could be
>made to run very much faster.
>
>Andrew.

= = = = = = = = = = = = = = = = = = = =

				
　　　　　　　　guw
　　　　　　　　guw@koal.com
　　　　　　　　　　2003-09-04


[-- Attachment #2: Rabbit.GIF --]
[-- Type: image/gif, Size: 1488 bytes --]

why gcj is slower than jvm

[Andrew Haley]

guw writes:
 > Hello!
 > 	I write a JAVA programme and compile it with GCJ. then I compare GCJ's performance with JVM'performance,but I discover that the procedure to be GCJ compiled is slower than JVM.
 >   why?

My guess would be that libgcj's implementation of SHA-1 is slower than
that in whatever JVM you were using.  I expect that SHA-1 could be
made to run very much faster.

Andrew.

Re: why gcj is slower than jvm

[Jeff Sturm]

On Wed, 3 Sep 2003, guw wrote:
> I write a JAVA programme and compile it with GCJ. then I compare
> GCJ's performance with JVM'performance,but I discover that the procedure
> to be GCJ compiled is slower than JVM.

Your example is more of a comparison of runtime performance than compiler
optimization.  From your results one might conclude that the JVM has a
better MD-5 implementation than libjava, but little else.

Jeff

why gcj is slower than jvm

[guw]

[-- Attachment #1: Type: text/plain, Size: 2260 bytes --]

Hello!
	I write a JAVA programme and compile it with GCJ. then I compare GCJ's performance with JVM'performance,but I discover that the procedure to be GCJ compiled is slower than JVM.
  why?

thanks.

===============================================================================
Output data:
JVM:
SHA1: Hashing 100000 blocks of 500 bytes each: time = 5.276, speed = 9254.762 KB/s
MD5: Hashing 100000 blocks of 500 bytes each: time = 3.578, speed = 13646.765 KB/s

GCJ:
SHA1: Hashing 100000 blocks of 500 bytes each: time = 8.719, speed = 5600.1978 KB/s
MD5: Hashing 100000 blocks of 500 bytes each: time = 14.337, speed = 3405.7422 KB/s

===============================================================================
Compiled command:
gcj -O2 --main=HashSpeed HashSpeed.class -o HashSpeed.exe



===============================================================================	
Source code:

import java.security.MessageDigest;
import java.util.Iterator;

public class HashSpeed {

	public static void main(String[] args) {
		 if (args == null || args.length == 0) {
			speed("SHA-1");
		 }
		 else{
		 	int i=0;
			while(i<args.length){
				speed(args[i++]);	
			}					
		 }		 
	  }

	  private static void speed(String name) {
		 try {
			MessageDigest hash = MessageDigest.getInstance(name);
			speed(hash);
		 } catch (Exception x) {
			System.out.println(name+": Failed self-test...:"+x.getLocalizedMessage());
		 }
	  }

	  private static void speed(MessageDigest hash) {
		 try {
			int iterations = 100000;
			int blocksize = 500;
			byte[] data = new byte[blocksize];
			int i;
			for (i = 0; i < blocksize; i++) {
			   data[i] = (byte) i;
			}

			System.out.print(hash.getAlgorithm()+": Hashing "+String.valueOf(iterations)
				  +" blocks of "+String.valueOf(blocksize)+" bytes each: ");
			long elapsed = -System.currentTimeMillis();
			for (i = 0; i < iterations; i++) {
			   hash.update(data, 0, blocksize);
			}
			elapsed += System.currentTimeMillis();
			float secs = (elapsed > 1) ? (float) elapsed / 1000 : 1;
			float speed = (float) iterations * blocksize / 1024 / secs;

			System.out.println("time = "+secs+", speed = "+speed+" KB/s");
		 } catch (Exception x) {
			x.printStackTrace(System.err);
		 }
	  }
}

archer
guw@koal.com


[-- Attachment #2: Rabbit.GIF --]
[-- Type: image/gif, Size: 1488 bytes --]

why gcj is slower than jvm

[guw]

[-- Attachment #1: Type: text/plain, Size: 2260 bytes --]

Hello!
	I write a JAVA programme and compile it with GCJ. then I compare GCJ's performance with JVM'performance,but I discover that the procedure to be GCJ compiled is slower than JVM.
  why?

thanks.

===============================================================================
Output data:
JVM:
SHA1: Hashing 100000 blocks of 500 bytes each: time = 5.276, speed = 9254.762 KB/s
MD5: Hashing 100000 blocks of 500 bytes each: time = 3.578, speed = 13646.765 KB/s

GCJ:
SHA1: Hashing 100000 blocks of 500 bytes each: time = 8.719, speed = 5600.1978 KB/s
MD5: Hashing 100000 blocks of 500 bytes each: time = 14.337, speed = 3405.7422 KB/s

===============================================================================
Compiled command:
gcj -O2 --main=HashSpeed HashSpeed.class -o HashSpeed.exe



===============================================================================	
Source code:

import java.security.MessageDigest;
import java.util.Iterator;

public class HashSpeed {

	public static void main(String[] args) {
		 if (args == null || args.length == 0) {
			speed("SHA-1");
		 }
		 else{
		 	int i=0;
			while(i<args.length){
				speed(args[i++]);	
			}					
		 }		 
	  }

	  private static void speed(String name) {
		 try {
			MessageDigest hash = MessageDigest.getInstance(name);
			speed(hash);
		 } catch (Exception x) {
			System.out.println(name+": Failed self-test...:"+x.getLocalizedMessage());
		 }
	  }

	  private static void speed(MessageDigest hash) {
		 try {
			int iterations = 100000;
			int blocksize = 500;
			byte[] data = new byte[blocksize];
			int i;
			for (i = 0; i < blocksize; i++) {
			   data[i] = (byte) i;
			}

			System.out.print(hash.getAlgorithm()+": Hashing "+String.valueOf(iterations)
				  +" blocks of "+String.valueOf(blocksize)+" bytes each: ");
			long elapsed = -System.currentTimeMillis();
			for (i = 0; i < iterations; i++) {
			   hash.update(data, 0, blocksize);
			}
			elapsed += System.currentTimeMillis();
			float secs = (elapsed > 1) ? (float) elapsed / 1000 : 1;
			float speed = (float) iterations * blocksize / 1024 / secs;

			System.out.println("time = "+secs+", speed = "+speed+" KB/s");
		 } catch (Exception x) {
			x.printStackTrace(System.err);
		 }
	  }
}

archer
guw@koal.com


[-- Attachment #2: Rabbit.GIF --]
[-- Type: image/gif, Size: 1488 bytes --]