Hi Anthony,

> I fixed my VM bug which was showing up in continuations.  You can get
> the new VM from the VI4 website: http://minnow.cc.gatech.edu/squeak/VI4.
>  You will need the new image as well.

    Thank you and MobVM now also supports VI4.

> To answer Stephane Ducasse question about macroBenchmark speed.  The
> only one that is still comparable - ie. doesn't involve code that has
> changed - is #6, open and close browsers.  It reports about a 30%
> increase in speed on both a Pentium 2 and Pentium 4.

    I know that macroBenchmarks is not very reliable, but just for fun,

    I perform a set of bench mark on an AMD 1.1 GHz, 768MB WindowsXP for the
following VMs:

    1/- VI4 MegaVM as downlaoded from VI4 page.
    2/- VI4 MobVM MegaInterp (with combo of InterpreterPlugin,
ObjectMemoryPlugin and PrimitivesPlugin)
    3/- VI4 MobVM MiniInterp (with separate ObjectMemoryPlugin and
PrimitivesPlugin)
    4/- VI3 MobVM MegaInterp (to comapre against VI4)
    5/- VI3 MobVM MiniInterp   (to comapre against VI4)

    It looks like the partition of InterpreterPlugin, ObjectMemoryPlugin and
PrimitivesPlugin
    for VI4 MobVM is _far_ _far_ from optimum more work need be done for
this.

    That for VI3 is not too bad. Still, can be improved.

    MobVM MegaInterp for both VI3 and VI4 is comparable to MegaVM.
    (Hey, for tinyBenchmarks, MobVM beats MegaVM ;-)

    Please find appended the summaries of  macroBenchmarks and
tinyBenchMarks.

    NB. Test #4 (mini.image) is skipped for VI4 because of  imcompatible
image format.

    Cheers,

    PhiHo.

VI4 MegaVM (downloaded from VI4 page)

Smalltalk macroBenchmarks

#(20,465 17,781 31,653 0 7,460 5,170)  Total time = 82,529 milliseconds.
#(20,500 17,892 31,566 0 7,402 4,989)  Total time = 82,349 milliseconds.
#(21,004 17,947 31,651 0 7,440 5,075)  Total time = 83,117 milliseconds.

0 tinyBenchmarks

'128,320,802 bytecodes/sec; 5,217,731 sends/sec'
'128,000,000 bytecodes/sec; 5,130,389 sends/sec'
'128,128,128 bytecodes/sec; 5,248,812 sends/sec'

VI4 MobVM MegaInterp

Smalltalk macroBenchmarks

#(20,928 17,877 32,422 0 7,518 5,224)  Total time = 83,969 milliseconds.
#(20,802 17,971 31,860 0 7,550 5,160)  Total time = 83,343 milliseconds.
#(21,472 17,938 31,800 0 7,600 5,211)  Total time = 84,021 milliseconds.

0 tinyBenchmarks

'130,879,345 bytecodes/sec; 5,418,259 sends/sec'
'130,745,658 bytecodes/sec; 5,405,793 sends/sec'
'130,612,244 bytecodes/sec; 5,418,259 sends/sec'

VI4 MobVM MiniInterp

Smalltalk macroBenchmarks

#(22,564 19,489 34,667 0 8,095 5,491)  Total time = 90,306 milliseconds.
#(22,621 19,492 34,508 0 8,154 5,804)  Total time = 90,579 milliseconds.
#(23,074 19,642 34,533 0 8,119 5,498)  Total time = 90,866 milliseconds.

0 tinyBenchmarks

'127,744,510 bytecodes/sec; 5,100,691 sends/sec'
'128,837,443 bytecodes/sec; 4,908,882 sends/sec'
'127,617,148 bytecodes/sec; 5,104,384 sends/sec'

VI3 MobVM MegaInterp

Smalltalk macroBenchmarks

#(10,489 82,754 32,497 11,639 2,047 10,501 4,979)  Total time = 154,906
milliseconds.
#(10,482 83,150 35,500 11,645 2,045 10,533 4,943)  Total time = 158,298
milliseconds.
#(11,534 88,292 38,972 11,665 2,063 10,599 5,003)  Total time = 168,128
milliseconds.

0 tinyBenchmarks
'106,048,053 bytecodes/sec; 3,273,190 sends/sec'
'104,746,317 bytecodes/sec; 3,273,190 sends/sec'
'105,349,794 bytecodes/sec; 3,280,585 sends/sec'

VI3 MobVM MiniInterp

Smalltalk macroBenchmarks

#(11,347 89,065 35,081 12,679 0 11,185 5,244)  Total time = 164,601
milliseconds.
#(11,346 89,883 37,596 12,684 0 11,129 5,250)  Total time = 167,888
milliseconds.
#(11,582 89,935 38,179 12,715 0 11,217 5,264)  Total time = 168,892
milliseconds.

0 tinyBenchmarks

'101,667,990 bytecodes/sec; 2,981,941 sends/sec'
'101,829,753 bytecodes/sec; 2,961,670 sends/sec'
'101,829,753 bytecodes/sec; 2,86,0549 sends/sec'


----- Original Message -----
From: "Anthony Hannan" <ajh18 at cornell.edu>
To: <squeak-dev at lists.squeakfoundation.org>
Sent: Monday, October 28, 2002 12:59 AM
Subject: [VI4]


> Hello all,
>
> I fixed my VM bug which was showing up in continuations.  You can get
> the new VM from the VI4 website: http://minnow.cc.gatech.edu/squeak/VI4.
>  You will need the new image as well.
>
> To answer Stephane Ducasse question about macroBenchmark speed.  The
> only one that is still comparable - ie. doesn't involve code that has
> changed - is #6, open and close browsers.  It reports about a 30%
> increase in speed on both a Pentium 2 and Pentium 4.
>
> Cheers,
> Anthony
>