Subject: Re: 1.5_Beta on 3100/M76 and MVII (hare and tortoise syndrome)
To: NetBSD/vax Mailing List <port-vax@netbsd.org>
From: Lord Isildur <mrfusion@guildvax.guild.net>
List: port-vax
Date: 11/07/2000 15:42:15
hmm. That's under Ultrix 4.5. 
Under 4.3tahoe, on a KA650 (nominally 2.8 VUPS compared to the 8800's 
twelve) the same command gets far better time, compiled with pcc2
(the standard compiler in 4.3tahoe):

 14:23:53 osgiliath $ echo "2 9999 ^ 3 6308 ^ / p" | time dc
 2
        72.01 real        71.66 user         0.06 sys  

I doubt that gcc would even compile on that system to try it, but the 
8800 is supposed to be about 4 times as fast as the uvIII.. 
this, too, was compiled with -O

I remember vcc under ultrix 3 on a uvII generating noticeably faster code 
than cc. maybe things changed in later Ultrixes. I just tried under 
ultrix 4.4 on a vs3100/38 using the dc sources from 4.3tahoe, and vcc 
crashed on it. cc's output ran in 73 seconds on the 3100/38 , making it 
somewhat slower than pcc's which did it in 72 on a slower machine... 
I think, then, that pcc2 is a better compiler for the VAX than any other 
(in UNIX at least. anybody have VMS numbers? ) I think ultrix after 4.2 
got sorta slow.. 

where can i find the AT&T sources for dc so that i can use the same 
version you used? 

Isildur

ragge said:
> > To illustrate this I took some time and made a simple benchmark: On a
> > VAX8800 running Ultrix 4.5 I compiled the AT&T dc with those three
> > compilers, and made it calculate something that takes time. dc do not
> > use any library routines for its internal work so the differences do
> > fairly well mirror the compiler efficiency:
> > 
> > Running the command:
> > % echo "2 9999 ^ 3 6308 ^ / p" | /usr/bin/time ./dc
> > 
> > gave the following results:
> > 
> > cc:
> >        44.7 real        44.4 user         0.0 sys
> > 
> > vcc:
> >        51.0 real        50.8 user         0.0 sys  
> > 
> > gcc:
> >        31.3 real        31.1 user         0.0 sys  
> > 
> > 
> > This shows that there is a huge difference between the efficiency of
> > those compilers. 
> > All tests was compiled with -O on the compilers.