[rescue] VAXstation questions

[Peter Corlett]

Dave McGuire <mcguire at neurotica.com> wrote:
> On Thursday, May 15, 2003, at 06:11 AM, Peter Corlett wrote:
[...]
>> Sorry, I was sloppy in my usage. What I meant was that object size (which
>> does relate to instruction density) tells you nothing about the typical
>> memory usage of an OS running on that architecture.
> Oh yes yes...I agree with that statement completely. It does, however,
> tell you about the potency of the instruction set, which is the point I
> was trying to make.

I know very little about the VAX, alas. I'm told it's a lovely architecture
for writing assembly language on. Shame I'll probably never get to play with
it, really.

>> I'd hazard a guess that VAX Linux (is there such a beast?) would use much
>> more memory than VAX VMS.
> Well, Linux boots at least far enough on a VAX to identify some hardware
> and print a startup message, which seems to be enough to consider it
> "ported" to the Linux community. ;)

*snigger*

Even I've managed to get that far with a little side project to write a tiny
OS for an m68k system :)

[...]
>> I think this is probably correct, although I suspect that PPC is also an
>> important target these days.
> Agreed...Apple even uses it for OS X. I'd imagine they're putting some
> pressure (hopefully $$$) on the GCC development team.

It was exactly OSX I had in mind here...

>> I note that gcc3.3 does seem to generate much better m68k code than
>> earlier versions.
> Does it really? Wow, that is *cool*. Especially because I'm going to be
> starting a lot of m68k development this summer. :)

Well, I started $TOYOS with gcc-3.0, and it didn't seem terribly good at
holding stuff in registers, flushing them to the stack more than it really
had to. gcc-3.3 seemed a lot better at this.

I also discovered that declaring hardware registers "volatile" didn't always
do quite what was expected - I'd write a value to a write-only register, and
then the compiler would realise a bit later that it wanted that value for
something else, and read the register. This would cause the junk on the
bus to be re-written to that register and also give the wrong result.

I never did work out whether I'd screwed up the declaration or there was a
gcc bug, but the problem went away when I recently came back to the code
after an absence, built gcc-3.2.3 from the Debian toolchain-source, and
retested the code.

-- 
PGP key ID E85DC776 - finger abuse at mooli.org.uk for full key