Unix VM and MacIntel
piumarta at gmail.com
Mon Oct 16 21:28:42 UTC 2006
On Oct 16, 2006, at 1:44 PM, John M McIntosh wrote:
> Ah *slower* you mean
1. '229390681 bytecodes/sec; 7759029 sends/sec'
2. '291904218 bytecodes/sec; 8411337 sends/sec'
(1) is Unix 3.9-9 built from source on Mac Book Pro with register
variables disabled (i.e., the thing I said was slower).
(2) is Unix 3.9-9 built from source on Mac Book Pro with register
variables enabled (i.e., the thing I said was faster and the thing I
dumped in the binary archive in consideration of one person's
hesitation to build from source).
> mac carbon 3.8.13b3u on pb 17 osx 10.4.8 gives
If the carbon VM is faster on PowerBook then maybe you use some
clever XCode PPC optimisation flags that Apple doesn't bother to tell
command-line users about? In any case, it has no relation to either
of the things I was comparing when I claimed one was faster than the
> On testing on macIntel the downloaded application, and the build
> application both would not open the squeak window
> when given a 2.8 image or a 3.8.latest image. Squeak CPU usage
> would run at 100%, not userful. No message to the system log or
> www.squeakvm.org unix page only shows 3.9-8
I though I explained that 3.9-8 is the latest release. 3.9-9 doesn't
exist yet. Anything you might have seen called 3.9-9 is a figment of
my imagination that I might do someone a favour by making a binary VM
available to them. (Maybe I've learned that I should never copy the
list in such situations.)
> So if someone has a *working* 3.9-9 macintel VM
I have one. In fact, I have two of them now. (One of which is
faster than the other. ;-)
> Lastly the VM application 3.9-9 posted is macintel only, not powerpc.
I know. I don't make universal binaries.
FWIW, the 'hang' sounds like a problem with a broken lock on i386. I
fixed this and posted new 3.9-9 sources (which don't really exist yet
sice 3.9-8 is the latest release) shortly after someone found it.
(FWIW #2, the slow refresh some people reported is probably because I
reverted to the crappy OS lock on .386 instead of writing a real
spinlock. The latter is #13489 on today's 'to-do' list.)
More information about the Squeak-dev