Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site petrus.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!bellcore!petrus!hammond
From: hammond@petrus.UUCP
Newsgroups: net.lang.c
Subject: Vax VMS C compiler
Message-ID: <331@petrus.UUCP>
Date: Tue, 23-Apr-85 09:19:18 EST
Article-I.D.: petrus.331
Posted: Tue Apr 23 09:19:18 1985
Date-Received: Wed, 24-Apr-85 03:55:33 EST
Distribution: net
Organization: Bell Communications Research, Inc
Lines: 28

Since no one else has done this, I will.  Someone (the article has expired
on our system) posted an article claiming that compiling the UNIX kernel
and utilities with the VAX VMS C compiler resulted in a system which
was 20% faster than the current unix (I assume 4.2 BSD).

I DON"T BELIEVE THIS!!!!!!  If they could get such a performance improvement
their salespersons should be beating down our doors to try and convince us to
run vaxen with ULTRIX rather than other brands.  Paying for ULTRIX as
opposed to free 4.2 BSD would probably be worthwhile if we got a 20%
improvement.

Possible reasons for the claim not being true: 
1) The code is faster, but the system/commands crash (i.e. overzealous
   optimization introduces new bugs)  That is much more believable in
   C than any other language except assembler.
2) Since our vaxen are often I/O bound, speeding things up by 20% merely
   made 20% more idle cycles, not jobs which completed in 80% of their
   former running time.
3) Some things run 20% faster, but you don't see much difference on an
   average workload.  I can believe a smart compiler could make code which
   ran much faster, particularly for floating point.
4) A combination of all of the above.

I am not saying it isn't possible to optimize C code, just that I don't
believe the claims in the absence of any supporting data (i.e. salespersons
bringing those claims to us.

Rich Hammond	{ allegra | decvax | ucbvax} !bellcore!hammond