Subject: Re: Compiler timings on varous MVII NetBSDs etc.
To: Matthias Buelow <>
From: NetBSD Bob <>
List: port-vax
Date: 01/23/2001 10:04:57
> >> this trival case) that NetBSD-1.4.3 and NetBSD-1.5 w/gcc 2.91 seem to be
> >> about on par with eachother.
> >
> >Well, yes and no.  They are supposedly the same compiler.  Yet,
> Are you sure of that?  I am too lazy to check CVS logs now but
> I think I remember reading "egcs-1.1.1 release" on 1.4 instead
> of the following on 1.5:
> gcc version egcs-2.91.66 19990314 (egcs-1.1.2 release)

I will have to check that.  Folks had been telling me they were
essentially the same.

> and indeed, on 1.5 it doesn't show one particular compiler error
> which it tripped over on 1.4 (on VAX).
> Of course such a minor change (if it is one) "usually" doesn't
> affect performance that much so maybe it's something with the
> system still...


> Another opinion thrown in, I don't think digging up the rotten pcc
> corpse from its well-deserved peace is actually a good idea --


Well, I agree with you in principle, except on the slow boxes.

I would take the position (Inet Flak Suit) in hand, that IFF
something is not done about the situation of what I would consider
bloat or excess featuritis, in 1.5, then we have effectively
priced the 1.0VUP and slower machines out of the market.  They
just can't handle the load anymore.

It is OK to relegate them to the retirement pasture, ane let 1.4.x
be their final hurrah.  But, if we are not careful, the same thing
will happen shortly to the 3100 class machines.  I can feel the
slowness on those, too.  IF we maintain that we are actively going
to support the historic VAX machines, then, ``Houston,  we have a

I don't know what the solution should be, or can be, but, I am looking
for options.  For example, if I want to do a lot of compiling, 1.5
is not usable anymore.  Ultrix or 1.2 are clearly the best.  It is
bad enough that I have taken to running big 1.5 compiles on my M76
box, and then port things over to the MVII and MV3100 boxes.  For
example, perl took 36 hours compilation time and was still making
its makefiles, while the M76 finished in 3 hours.  That just sucks.
A kernel compile takes over 24 hours on the MVII, and only 9 hours
on the M76.... etc.  That kind of slowness hurts.  Compare that
with a kernel compile using pcc on 4.3BSD on an even smaller
memory MVII, at .... 45 MINUTES.  Clearly, Houston, we do have
a problem.

> Maybe writing a new compiler suite from scratch would be the best
> idea but I don't currently know anybody who'd undertake such a huge
> effort, when another working (mostly) and free (somewhat) compiler
> is already available and maintained, even if it's huge and slow
> and has a rich history of bugs.

Pcc works fine, within its limitations, and I think gcc-1.4.2 will,
too.  IFF things like kernels have taken to high-gccisms, contrary
to the classical unchanging monolithic single-purpose, according
to Ken Thompson (I think that was who made that statement about
kernels), then we are in trouble, and again, we have priced the
low-end VAXen out of the market, and, ``Houston, we have a problem!''

I know it is heresy to say this, but I will, again..... IF it really
is reaching that point in time where the 1.0VUP boxes are becoming
priced out of the market, then we need to separate out a code tree
for VAX 1.0 VUP machines and under, and then optimize it as a final
hurrah for these machines, freeze it, and keep it as the Classical
VAX NetBSD.  There is nothing wrong with that, and 1.5 is getting
mostly useless on these machines.  It hurts, too, cuz I really do
like the ol' MVII critters, and am actively looking for a 730 or
750 box to try.

Is there a practical solution?  I dunno, but keep hoping there is.
NetBSD-1.5 on the slow boxes is just not practical, anymore.