[rescue] Sun 3/60

[Greg A. Woods]

[ On Thursday, August 16, 2001 at 01:26:07 (-0400), Dave McGuire wrote: ]
> Subject: Re: [rescue] Sun 3/60
>
>   Then why is it that X on an M68k built with gcc (circa early gcc 2.x
> days) runs easily 2-3 times faster when built with GCC as compared
> with Sun's cc?  Not to be argumentative, but...

I've always found the opposite to be true!

Now I've not run the exact same X11 code on the same OS revision and
hardware.  However I did run X11R5 built by SunOS-4's native CC on
sun3's, and NetBSD's port of X11R6 on NetBSD 1.2, built by GCC, and all
I can say is that the latter sucked in all ways (much bigger, far
slower, and far less reliable).  Indeed found a diskless 3/60 to be
faster running SunOS-4 and X11R5 compiled by Sun's cc, than a better
endowed (32MB RAM vs. 16MB) 3/280 with local disks running NetBSD
w/X11R6.  As far as I know the only difference in the graphics hardware
in those two cases was the P4-bus on the 3/60 vs. the VME bus on the
3/280.  Unfortunately it's hard to make a fair comparison there because
of the radical differences in SunOS-4 and NetBSD-1.2.  It might be
better to compare NetBSD-1.5W or newer (with the new VM and unified
buffer cache) with SunOS-4 (and these days I think you could compare the
same X11 server code too).

In other venues I've found the X11R6 server to be faster than X11R5, but
I'm not sure the comparison's fair because so many other variables
change there.

Indeed there are several code constructs that GCC could/can generate
better m68k code for, but as anyone familiar with compiler benchmarks
knows, it's always possible to make any given compiler shine better than
all the rest if you know how to tweak the code you feed it (and manage
the often plentiful optimisation control parameters).  However given the
relatively easy readability of m68k code, it's not too hard to spot
constructs where GCC is abysmal too!

The benchmarks that convinced me to stick with plain old SunOS-4's cc
were real-world applications in the target domain of my production
machines.  Things like web/news/ftp servers, script interpreters (mawk,
perl, etc.), MTAs, etc.

Let me assure you that I wouldn't have gone to all the work I did at
back-porting code to be strictly K&R conforming if I'd not felt the
result would be well worth the effort!

-- 
							Greg A. Woods

+1 416 218-0098      VE3TCP      <gwoods at acm.org>     <woods at robohack.ca>
Planix, Inc. <woods at planix.com>;   Secrets of the Weird <woods at weird.com>