Re: [Squeak-VMdev] Unix sources in SF/CVS?

[Tim R. <ti...@su...>]

In message <Pin...@ti...>
          Ian Piumarta <ian...@in...> wrote:

> Want to send me the change set?  I'll file it in, rebuild and run the
> benchmarks.
It's supposed to be http://swiki.gsug.org:8080/sqfixes/2201.html but I
can't get to it right now.  I've dropped a copy on
http://sumeru.stanford.edu/tim/pooters/SqFiles/deltas/VMDevBcommonSendFi
xup.2.cs (and yes I realize that is a very confusing name but blame
scott not me!).

You'll note that there is a very odd and unused Interpreter class var by
the name of Must in there that I think is a stupid typo from a bodged
compile some where in a method with an improperly done comment saying
"Must be green " or somesuch. Now the really odd thing is that the first
time I filed this is the classvar CacheProbeShift went missing along
with Must. And it somehow corrupted the image in the process so I
couldn't restart it. Very odd.
> 
> >From what I remember of his posts he's talking about subtle effects (cache
> line invalidation cost) that are rather specific to his processor (some
> flavour of pentium clone) -- unless I'm reading _completely_ the wrong
> thread?
That seems to have been the thrust of his claims. Of course, it
completely ignores any other kind of cpu or device. Amongst other things
that irritated me considerably was the repeated claim that you _must_
add more memory and _of course_ everyone can afford (or even fit)
hundreds of MB ram. I'm all for trying things out and finding ways to
benefit common cases, but really, claims like that just piss me off.
PDA's anyone? Older machines? Bare hardware for embedded stuff? Bah.

Anyway, I'll be interested to see what results you get. I've run the
modified vm with both the Smalltalk macroBenchmarks and the old Green
book benchmarks
(http://sumeru.stanford.edu/tim/pooters/SqFiles/deltas/sqBenchmarks.cs)
with essentially identical results on the Acorn.
> 
> This is pretty inspirational stuff!
> 
> Unfortunately, what it inspires is a resounding "who the f**k cares?".  
> His cpu might have a broken cache architecture, mine doesn't.  (And given
> your results it would seem the ARM doesn't either.)
Not broken, but teeny-tiny. And it's embarrassing to have to admit to
having a tiny cache. Even more embarrassing than it would be to admit to
a miniscule love sausage. Which I'm not. I'll refer you to my wife and
mistresses.

tim
-- 
Tim Rowledge, ti...@su..., http://sumeru.stanford.edu/tim
Useful random insult:- When a thought crosses her mind, it's a long and lonely journey.