Subject: Re: 1.5_Beta on 3100/M76 and MVII (hare and tortoise syndrome)
To: NetBSD Bob <nbsdbob@weedcon1.cropsci.ncsu.edu>
From: Todd Vierling <tv@wasabisystems.com>
List: port-vax
Date: 10/26/2000 15:47:42
On Thu, 26 Oct 2000, NetBSD Bob wrote:
: > > NetBSD-1.3 865218
: > > NetBSD-1.4 1015803
: > > NetBSD-1.5 1402255
:
: And now gcc and Feeping Creatures (or is that rats in the bit bucket?)
: For comparison, my minimal stripped MVII kernel of 1.5B1 is ~650000
: compared to the 1.5 generic kernel of 1402255
1.5 GENERIC on any architecture is pretty huge. We have filesystems and
pseudodevices galore these days. :)
It's almost all features. There's more you can do if you're strapped for
memory; try changing the "-O2" to "-Os" (that's a lower case s) in the
kernel Makefile. This will do -O2 *except* for optimizations that increase
code size. On a low memory machine, this may result in faster code.
One thing I've looked at but not yet really had time to implement is a set
of options that can hardwire sysctls and other typically dynamic values to
allow the compiler to optimize out if/then clauses in speed critical
code. I do plan on putting more work into this area.
Also, wasn't there a push to change the calling convention for vax between
1.3 and 1.4 to improve kernel speed? (Was this ever done?)
--
-- Todd Vierling <tv@wasabisystems.com> * http://www.wasabisystems.com/
-- Speed, stability, security, and support. Wasabi NetBSD: Run with it.