[rescue] Sun 3/60

[Dave McGuire]

On August 16, Greg A. Woods wrote:
> > Quote from X11R4 release notes:
> > 
> > 7.   If you are running on a VAX  or  680x0  processor,  you
> >      should consider using the GNU C compiler (available via
> >      anonymous ftp from the machine prep.ai.mit.edu) to com-
> >      pile  the server.  It can result in up to a factor of 2
> >      improvement in performance.  See the  HasGcc  parameter
> >      in  the files sun.cf, ultrix.cf, macII.cf, and site.def
> >      in mit/config/.
> 
> Ah, very interesting!
> 
> I've not got an X11R4 tree around any more -- in fact I don't think I've
> run it for so long now that I can't even remember where I might have had
> it last.
> 
> This smells very VERY much like custom hacking of the X11 code to tweak
> it explicitly for GCC though -- if you have to turn on a flag like that.

  The flag in question simply sets the CC makefile variable to "gcc"
and turns on both "-g" and "-O", which (as you know) don't coexist
with many compilers.

> I happened to check in the world.log file the last time I built X11R5 on
> sun3 (1995, ~ 12 hours to build! :-)) and I see I built it with
> '-f68881' and as I recall that made a very noticable difference both to
> the feel of things as well as to x11perf...  I see I also used '-target
> sun3', but I don't remember why.

  Oh yes, hardware floating point is wonderful for X. :-)

> I'm betting they did some early inlining experiments or some such.

  That's certainly possible.  It might be interesting to look through
the sources to see if that flag conditionally compiles any code or
just affects the build process.

> I think it was about X11R4 when work by folks at the UofToronto to
> improve code locality in the Xserver might have made it out to the
> general release too.

  Yes...R4 was primarily a speed-enhancement release, better tuned to
the cache sizes of the day, and optimized for unit-stride memory
references and such.  The performance improvement over R3 was most
impressive.

> It's interesting that this advice didn't make it into the X11R5 notes or
> beyond (maybe they thought sun3 & vax were dead?).  I do note that in

  In the R5 days?  No way.  That was like 1991, if memory serves.  My
company was still getting close to $10K for Sun3/280 systems then.

> X11R5 there are some farily specific compiler options enabled for GCC,
> espeically on m68k and i386 platforms -- some apparently to work around
> compiler differences but others to enable special optimisations.

  Compiler optimizations, or compiler-specific code?

> >   If you actually want to do this, let me know...I'm interested in
> > your results, and I might do it here as well for kicks.  I miss
> > hacking on Sun3 systems.
> 
> It's not looking like it's going to cool down anytime soon, but I'll try
> to remember!  ;-)

  Hey, it'll be fun! :-)

> (right now my basement computer room is hovering around 80-85 at mid-day
> on any sunny day, and that's with an 8,000 BTU AC blasting away in the
> window!)

  Ouch. :-/

    -Dave

-- 
Dave McGuire
Laurel, MD