Path: utzoo!utgpu!water!watmath!clyde!att!osu-cis!tut.cis.ohio-state.edu!mailrus!ames!oliveb!sun!chiba!khb
From: khb%chiba@Sun.COM (Keith Bierman - Sun Tactical Engineering)
Newsgroups: comp.lang.fortran
Subject: Re: FORTRAN 88
Message-ID: <75243@sun.uucp>
Date: 28 Oct 88 23:53:26 GMT
References: <5833@vdsvax.steinmetz.ge.com> <7540@megaron.arizona.edu> <5838@vdsvax.steinmetz.ge.com> <1128@microsoft.UUCP>
Sender: news@sun.uucp
Reply-To: khb@sun.UUCP (Keith Bierman - Sun Tactical Engineering)
Distribution: comp.lang.fortran
Organization: Sun Microsystems, Mountain View
Lines: 50

In article <1128@microsoft.UUCP> bobal@microsoft.UUCP (Bob Allison (uunet!microsoft!bobal)) writes:

>>be a problem.  Your opinion may differ, but I don't care about compile
>>time (within one order of magnitude) 
>>
>
>Well, you may not care about compile time, but what about execution time.
>I can personally guarantee that, if you use array expressions, running
>any reasonable 8X program will be at least 25% slower than an equivalent 
>program written in 77 (to avoid an often belabored point: if such a 
>program can be written in 77).  I can make this statement with confidence 
>for FORTRAN compilers on the PC, probably until at least 1995.  Many other 
>scalar machines will be in the same boat.

1) I doubt that it will take Lahey and his merrie crew anywhere near
   that long.

2) Those working in the realm of high performance are well aware of
   the fact that 5-25% of the code consumes 75-95% of the cycles (e.g.
   a famous structural code, in a slightly outdated version, spent 97%
   of its time in 7 lines of code on a cydra 5). If our vendor of
   choice (sun of course :>) can't optimize the array facilities, the
   user uses the nifty profiling tools to determine where the glitch
   is, and rewrites a few lines of code.

   Those who balk at such fun and games run slower, chose a different
   vendor, or buy different machines. But when you get to the world of
   high performance, this is ALWAYS necessary to get peak performance.


>I agree that machines will speed up by more than 25% in the interim, but
>most people don't buy a machine which runs 25% faster just so their software
>can run as fast as it used to.  They buy the machine so THEIR software runs
>25% faster.

True. But by the time f88 is widely available machines can be 2x as
fast. Most user's of PC's now have 80286's or 8088's, running at
moderate clocks. 25mhz+ 80386/7 machines are already
available and do pretty well. (* RISC chips do even better *).

The real point is will it reduce the cost of developing software. (*
yes it will *), this cost is much higher than the productively lost by
having to hand optimize a few modules, in the most critical applications.


>Bob Allison


Keith H. Bierman
It's Not My Fault ---- I Voted for Bill & Opus