[VM] NCM performance report
Alan Lovejoy
sourcery at pacbell.net
Fri Mar 26 06:32:16 UTC 1999
Ian Piumarta wrote:
> Helge,
>
> > I wonder what a J2-NCM VM could do...
> [snip]
> What does cost (a *lot*) is block activation, since I have to go
> through the maps to find the initial nPC based on the intial vPC
> stored in the closure. (For compiled methods this cost is only paid
> when flushing a context out of the context cache, which happens
> relatively infrequently.) My #1 request would therefore be to have
> block bodies compiled into their own CompiledMethods, separate from
> the home method. (Or does NCM do this already??? I've not been
> following the thread very closely.)
I haven't looked at Tim's NCM implementation yet (Hi, Tim!), butI
know that the VisualWorks implementation has something called
a CompiledBlock, which shares a common superclass with
CompiledMethod (namely, CompiledCode). I think this is essentially
what you requested.
Tim and I worked together on a project (that shall remain nameless)
where we modified the Squeak VM in a way very similar to Tim's
current NCM implementation (well, actually Tim and Don Charnley
reworked the VM, and I reworked the CompiledMethod class).
It **is** cleaner and more elegant, and makes CompiledMethods
more easily extensible.
--Alan
Content-Type: text/x-vcard; charset=us-ascii; name="vcard.vcf"
Content-Transfer-Encoding: 7bit
Content-Description: Card for Alan Lovejoy
Content-Disposition: attachment; filename="vcard.vcf"
Attachment converted: Anon:vcard.vcf 12 (TEXT/ttxt) (00007A78)
More information about the Squeak-dev
mailing list
|