A Thought about Multiple Cores

Michael Rueger m.rueger at acm.org
Sat Jan 27 09:39:46 UTC 2007


tim Rowledge wrote:
> The Heeg port of objectworks to some dual cpu 68k workstation - back in 
> the day when 68k cpus were the dogs nadgers - did this core/blt split. 
> Mike Reuger can probably remember the details. It did the bitblt and 
> scanchars benchmarks *very* fast.

The PCS Cadmus had a second 68k processor integrated into the bitmap 
display. I don't remember who did the coding (Hans Martin?), but we did 
what Dan described, offloading bitblt and char scan to the display. Made 
for an amazing Smalltalk experience back then :-) (250% Dorado)

> I guess we could claim that using things like the Rome/Cairo plugin is 
> doing a modern equivalent since it is supposed to pass off most 
> rendering chores to the GPU. If Squeak made effective use of such 

The main bottleneck is that in most cases we still have at least one 
extra bitblt as Squeak doesn't support a native surface as Display yet. 
(at least to my knowledge and what I got from discussions with Bert, he 
can explain that in more detail than I can).

Michael



More information about the Squeak-dev mailing list