Sorry to followup my own, but I've got a status update...

> Currently when I try building a solaris2.5.1 cross to sparc-current compiler,
> it gets a few important details wrong, like the underscore settting, and
> the local label naming, and quite a few other macros whose purpose I don't
> grok too well.

I tried copying over the netbsdelf.h and underscore.c bits from
src/gnu/usr.bin into my configure'd directory, and built a new cc1.
Now my "tag team" kernel builds succeed.

What used to take my (110mhz) Tadpole 3gx 40 minutes to build, now takes 20
minutes for both it and my (167mhz) Ultra 1 Solaris box at work.

Strangely enough, it seems that just using -pipe gives plenty of parallelism;
when I try adding -j2 to the mix it doesn't appear to improve the time. Then
again, I only have a couple measurements and nothing like a real sample set.

I still need to exhaustively verify that the cc1 output is identical to what
the native netbsd cc1 would have produced. I know it's pretty close, and some
other experiments with netbsd arm32+i386 yielded identical outputs on my test
file, so I am not too worried.

Call me reckless, but I'm running a make build with the special $(CC) to see
how it does. Something I noticed right away is that on a slow machine, the
cleandir, includes, and nroff portions of a make build consume quite a lot
of time...

Todd Whitesel
toddpw @ best.com