[BC] MacroBenchmark fixes
Scott A Crosby
crosby at qwes.math.cmu.edu
Tue Feb 12 15:54:53 UTC 2002
On Mon, 11 Feb 2002, Anthony Hannan wrote:
> Scott A Crosby <crosby at qwes.math.cmu.edu> wrote:
> > Excellent news... I've integrated the new methodcache into the BC image.
>
> Thanks for working on this, Scott. Its great to see such speed
> improvements.
>
Yep, by the time everything thats already done gets integrated in, the VM
will be nearly twice the speed. (I've got a list of various outstanding
things for the VM, and their actual or estimated gains.)
> What do you mean? How is BC clobbering class definitions?
What Tim said. Thus I don't plan on doing too much integration work,
except for occasional testing, till BC is stabalized.
> > Any idea if/when the BC stuff will support running all of the macrobenchmarks?
>
> Attached are fixes for macroBenchmarks for BC. All work except the
> first benchmark (decompiling not implemented yet). I had to make some
> small changes to MessageTally and macroBenchmark3 so the comparison to
> the standard image may not be accurate, but I think it will be close
> enough.
Run and tested, but macrobenchmark3 (#4) really isn't comparable, and #2
isn't comparable either. So, there's nothing really new here. A few other
entries changed by a second, but thats noise. I nuked my BC-only VM, so I
won't rerun for that. (No point anyways.)
> STOCK MC BC BC+MC
> 1 54908 27235 - -
> 2 294773 219893 - 54632[*]
> 3 132697 56784 83456 41692
> 4 54756 30076 - 207758
> 5 - - - -
> 6 19057 14462 18267 11682
> 7 14544 11520 15518 8367
>
> These are all on a P2-450
[*] This number is wrong, it builds tiles for 9 methods, when the others
are run with 35.
Scott
More information about the Squeak-dev
mailing list
|