garrett_damore@tadpole.com wrote:

> What I should do is _move_ these macros to below the #include's. 
> Probably I just stuck them where they are because there was already the
> cn_trap() macro there.  I will move them before I commit.

Ok, no problem.

> Thanks.  I wonder why there is a measurable difference.  There is a
> _very_ trivial extra register indirection for HEAD vs. gdamore-uart
> (without COM_REGMAP), but I would not have expected it to be noticeable.

Maybe I should check which part causes such difference,
but I haven't checked any compiled code yet.

> In any case, this hardware is so low end, then maybe it doesn't matter. 

As I wrote before, if indirection (and other changes) is really needed
to make a driver more generic MI (i.e. to support aucom.c which can't
be handled by MD code), I won't complain about such (negligible)
performance penalty.

> Does anyone run NetBSD seriously on hardware slower than this (25MHz)?

Maybe the possible slowest machine which has 16x50 com and can run
NetBSD is 12MHz 386SX PC or 16MHz 68020 hp300, but I'm not sure
if still there are any working ones. Most of my (obsolete) machines
are just running NetBSD to maintain NetBSD itself ;-)
---
Izumi Tsutsui