On Mon 22 Feb, Peter Crowther wrote:

> Tim, I hadn't realised you'd cleaned up the headers so much; I'll download
> one of your VMs and images forthwith and restart my half-built
> implementation of this.  Having even one spare bit in the CompiledMethod
> header will be just *so* handy...
If you are using w95 or (some chance!) Acorn, there are built VMs on my
website. Otherwise, you'll need to use the new 'interp.c' to build a new
VM, which shouldn't be too tricky for someone already fiddling with things
like this.

Personally, I think the execution time cost would come to about three
instructions during each uncached lookup and cache setup - and none for
cache hits, since the hard work is already done and the primitive
dispatching mechanism already there would do the job. Since the NCM stuff
cleaned up the spread-out primitive index and saved three or four cycles
per uncached send, I think we have cosmic balance.

tim

-- 
Useful random insult:- The twinkle in his eyes is actually the sun shining between his ears.
Tim Rowledge:  rowledge at interval.com (w)  +1 (650) 842-6110 (w)
 tim at sumeru.stanford.edu (h)  <http://sumeru.stanford.edu/tim>