[kaffe] Re: INTRP, JIT1 & JIT3 on m68k platform
Kevin D. Kissell
kevink at mips.com
Tue May 6 08:00:02 PDT 2003
These are really interesting results. While the best
"real" Java benchmarks have to be paid for, would
it be possible for you to repeat the experiment running
the free "Caffeinemark" (or "embedded Caffeinemark
if you don't have AWT working) benchmarks? These
are available for free download, if I recall correctly
from the site of pendragon.com. A google search
will turn them up if I don't recall correctly.
----- Original Message -----
From: "Tony Wyatt" <wyattaw at optushome.com.au>
To: "Kiyo Inaba" <inaba at src.ricoh.co.jp>
Cc: <inaba at src.ricoh.co.jp>; <kaffe at kaffe.org>
Sent: Tuesday, May 06, 2003 11:39 AM
Subject: [kaffe] Re: INTRP, JIT1 & JIT3 on m68k platform
> Hi Kiyo,
> On 06-May-03, you wrote:
> > Sounds good. If possible, could you please make (very rough) performance
> > measurement for JIT and JIT3 on Amiga. The easiest is just measure the
> > execution time for HelloWorldApp by 'time' command.
> I've spent all day compiling all three versions and creating three
> installation environments (/usr/local/kaffe-intrp, -jit, -jit3).
> Now I can switch from one to the other.
> I can not detect any difference in speed between the interpreter,
> jit1 or jit3, using the test classes, they just don't run for long enough.
> So I got hold of a benchmark called "Linpack" from the web and
> compiled and ran that. I compiled it using JIT1 and ran the same compiled
> on all three VMs.
> Measurements taken on my Amiga 68060/50 MHz, using the dreaded ixemul
> library (terribly slow). I'll run the same tests on native Linux on the
> hardware when I get JIT3 going under m68k Linux.
> Interpreter: 0.065 Mflops/s, time taken 10.56 sec
> JIT (1): 0.618 Mflops/s, time taken 1.11 sec
> JIT (3): 0.457 Mflops/s, time taken 1.5 sec
> So, I agree with you that JIT3 is slower then JIT1.
> kaffe mailing list
> kaffe at kaffe.org
More information about the kaffe