> Some even more interesting observations follow.
>
> After talking to Charles about this a fair bit, and testing certain things,
> we found that if I link in umul.o and strcmp.o statically to dhry.c, I get a
> *better* benchmark than the fully static version!

Interesting stuff!  Now we know why Theo's snapshots were all linked statically
- he was prescient  (-:

2 possibly useless suggestions:

(1) Seeing "umul.o", has anyone looked into using David Hough's freely-usable
    math library replacement under NetBSD/SPARC?  Wonder if that might provide
    some performance improvements or at least insights.

(2) I noted the following in the gcc 2.6.1 announcement:

Noteworthy changes in GCC version 2.6.1:
...
The following new configurations are supported:

        GNU on x86 (instead of treating it like MACH)
        NetBSD on SPARC and Motorola 68k
	...

Anyone tried building 2.6.1 on a NetBSD/SPARC system yet?  I wonder what new
results for Mathew's tests might come out of that ... especially given Jason's
comment:

> On a side note, -static makes no difference on my hp300 or i486 ... My guess
> is the way gcc 2.4.5 generates code for a SPARC ... Under SunOS, I've seen
> non-trivial performance increases with gcc 2.5.x and 2.6.

	- Greg