x86 jitter? (One more try!)
Raab, Andreas
Andreas.Raab at disney.com
Tue Apr 10 03:53:30 UTC 2001
PhiHo,
You should use '0 tinyBenchmarks' to get a comparable result (also don't
remove -DDEBUG but replace it with -DNDEBUG). On my (Win98) machine this
gives roughly the same results Marcus sees on Linux:
BI: '56,387,665 bytecodes/sec; 1,719,372 sends/sec'
J3: '65,775,950 bytecodes/sec; 9,073,805 sends/sec'
Oh my dear - what a historic moment (btw, Marcus why are unrunnable settings
the default????) The last SqueakJ.exe I built was in ... let's see ...
November '98 or so... and here we go again ... itsa working ... :-)))
[BTW, for those curious and not afraid of crashes I've uploaded a copy of
the executable to
http://isgwww.cs.uni-magdeburg.de/~raab/squeak/alpha/SqueakJ3.zip]
Cheers,
- Andreas
> -----Original Message-----
> From: phiho.hoang [mailto:phiho.hoang at home.com]
> Sent: Monday, April 09, 2001 8:18 PM
> To: squeak at cs.uiuc.edu
> Cc: recipient list not shown
> Subject: Re: x86 jitter? (One more try!)
>
>
> Hi Marcus,
>
> Thanks for the info. I could build J3 from Andreas
> sources together with
> instructions from you and Ian. I just ran a couple bench marks
> (3.1.3898.image + WinME + AMD 1.2 MHz +
> 256 MB):
>
> (Time millisecondsToRun: [1000 benchmark])
> Interp: 4320
> J3: 3645
>
> (Time millisecondsToRun: [10000 benchmark])
> Interp: 43127
> J3: 36770
>
> There must be something wrong. This is non-debug (no
> -DDEBUG). I got
> assertion failed (at compile.cc: 1929) inMeth->includesVPC
> (endPC + 1),
> though (I just commented it out :-).
>
> Cheers,
>
> PhiHo
>
> ----- Original Message -----
> From: "Marcus Denker" <marcus at ira.uka.de>
> To: <squeak at cs.uiuc.edu>
> Cc: <phiho.hoang at home.com>; <ian.piumarta at inria.fr>
> Sent: Monday, April 09, 2001 8:35 AM
> Subject: Re: x86 jitter? (One more try!)
>
>
> > On Sat, Apr 07, 2001 at 07:26:33PM -0700, Raab, Andreas wrote:
> > > Hi PhiHo (and anyone else interested),
> > >
> > > I've just uploaded a variant of the Win32 sources to
> > >
> > >
> http://isgwww.cs.uni-magdeburg.de/~raab/squeak/alpha/w32-j3.11.tar.gz
> > >
> > > latest x86 tree from your site. It would be useful if you
> could run a
> diff
> > > on the two and integrate the modifications in the generic
> x86 tree.
> There's
> > > also a file COMPILING.win32 in the win32 subdirectory
> that has a list of
> the
> > > modifications I did].
> > >
> > Ok.
> >
> > > This version compiles out of the box (given that you've
> got the right
> tools
> > > of course), runs and then dies. But it turns out that with the
> alloca()ed
> > > stack we're getting way further - now we're getting to
> > >
> > > compile.cc: in `void gen_compile()'
> > > compile.cc:1111: phase error at sPC 26: 0084F18D r.d.. 1<
> 41 0084F1A8
> > > LdTemp 1
> > >
> > This looks good (:-)) ... it seems to be compiled with
> OPT_FULL, and I
> havn´t
> > yet looked at that... but OPT_MACHINE and OPT_INLINE should work:
> >
> > Please be sure that your #defines are:
> >
> > 1) optimise.cc
> > #define OPT_INLINE
> > #undef OPT_FULL
> >
> > 2)genArray.h
> > #undef INLINED_AT
> >
> > 3)machine.cc
> > #define INDIRECT_ICACHE false
> >
> >
> >
> > Marcus
> >
>
>
More information about the Squeak-dev
mailing list
|