Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Two broken benchmarks

6 views
Skip to first unread message

Joshua Gatcomb

unread,
Nov 20, 2004, 12:14:56 PM11/20/04
to perl6-i...@perl.org
All:
oo1-prop.pasm and oo2-prop.pasm are both broken. The
only way I knew this was that there times in the
benchmark graphs had dropped off the map. It was
previously proposed to make the benchmarks part of the
test suite being sure not to include anything that
took a substantial amount of time (Warnock applies).

I am suggesting, without the tuits to actually do it,
that we have a make testbench option which would not
get run during make test that would test all the
benchmarks. This requires modifying the Makefile
generation process as well as actually getting the
benchmarks in the test harness.

Thoughts?

Cheers
Joshua Gatcomb
a.k.a. Limbic~Region

__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.com

Leopold Toetsch

unread,
Nov 21, 2004, 5:18:15 AM11/21/04
to Joshua Gatcomb, perl6-i...@perl.org
Joshua Gatcomb <limbic_re...@yahoo.com> wrote:
> All:
> oo1-prop.pasm and oo2-prop.pasm are both broken.

Yep. First WRT calling conventions and - that's a bigger problem - again
WRT DOD/GC. The latter is permanently reocurring and needs fixing.

1) The main source of troubles in hash code are the moving chunk
buffers.
2) As every collection of buffer memory can move these buffers, buckets
can only be addressed by index, which causes one more lookup per hash
operation
3) This make hash code slower then it should be

I'd suggest to rewrite hash.c to use system memory.

Takers wanted.

Back to oo*-prop.pasm. I think they are useless as benchmarks, but the
loop of setting properties repeatedly would make a good gc.t test.

leo

0 new messages