Subject: Re: Compiler timings on varous MVII NetBSDs etc.
To: David Brownlee <abs@netbsd.org>
From: Lord Isildur <mrfusion@umbar.vaxpower.org>
List: port-vax
Date: 01/25/2001 15:30:58
one note about the libc growth- some of it has gotten a LOT slower.
for example, on bob's machien he put up for testing yesterday, i ran some
test program that among other things calledprintf a little under 300
thousand time. compiled with dynamic linking, with -O2, it took the
printfs about 980 seconds to run. Compiled statically, about 910 seconds.
On the ka650 at home, with a much older libc, the same thing took about
85 seconds. (the program did about 90 seconds of other work)..
isildur
On Thu, 25 Jan 2001, David Brownlee wrote:
> On Thu, 25 Jan 2001, NetBSD Bob wrote:
>
> > > > Offhand, can I compile a NetBSD-1.2 kernel on NetBSD-1.5
> > > > by unrolling into a separate tree, and not break anything?
> > >
> > > i dont know if things would work quite right- config has certainly
> > > changed betwen 1.3 and 1.4, and probably there some gcc idiosyncracies
> > > that will make it barf on the 1.2 stuff..
> >
> > But, I would expect the config from 1.5 to be backwards compatible
> > with 1.2, but, the reverse, touch and go.
> >
> We went around a similar loop a month or so ago - there
> have been many changes to the format. You would want to
> compile a 1.2 config on 1.5.
>
> > That was exactly my thought. It does not have to be all that great a
> > compile, but it just has to finish, and link into a completed kernel.
> > If there was no great size change, then gcc was doing its job and
> > a lot of misc gago has not crept in with gcc. If it grew by
> > say more than 10%, then gcc is throwing in things, because the
> > code was the same.
>
> You might be better of compiling some large application
> such as ghostscript or pine, then compare the final binary
> and sizes of all the .o files generated. Link it statically
> on 1.5 also.
>
> I'd expect final binary to increase as libc has grown (locale
> support, extra features, etc), but the .o files should be
> pretty much the same. Smaller in -current with Matt's changes
> to gcc :)
>
> David/absolute -- www.netbsd.org: No hype required --
>
>
>
>
>