home *** CD-ROM | disk | FTP | other *** search
- Xref: sparky comp.arch:10648 comp.lang.misc:3608
- Path: sparky!uunet!dtix!darwin.sura.net!spool.mu.edu!agate!linus!linus.mitre.org!gauss!bs
- From: bs@gauss.mitre.org (Robert D. Silverman)
- Newsgroups: comp.arch,comp.lang.misc
- Subject: Re: Hardware Support for Numeric Algorithms
- Message-ID: <1992Nov12.131856.12605@linus.mitre.org>
- Date: 12 Nov 92 13:18:56 GMT
- References: <1992Oct29.153514.22927@yrloc.ipsa.reuter.COM> <1992Nov5.202412.7266@linus.mitre.org> <1992Nov10.153705.27804@yrloc.ipsa.reuter.COM>
- Sender: news@linus.mitre.org (News Service)
- Organization: Research Computer Facility, MITRE Corporation, Bedford, MA
- Lines: 68
- Nntp-Posting-Host: gauss.mitre.org
-
- In article <1992Nov10.153705.27804@yrloc.ipsa.reuter.COM> rbe@yrloc.ipsa.reuter.COM (Robert Bernecky) writes:
- :In article <1992Nov5.202412.7266@linus.mitre.org> bs@gauss.mitre.org (Robert D. Silverman) writes:
- :>In article <1992Oct29.153514.22927@yrloc.ipsa.reuter.COM> rbe@yrloc.ipsa.reuter.COM (Robert Bernecky) writes:
- :>:
- :>:I write programs for correctness and maintainability first, THEN worry
- :>:about little details such as performance. If you create something that's
- :>
- :>You've probably never done any large scale computing.
- :>
- :>I measure the run time of my algorithms in MIPS-YEARS. A MIPS-YEAR
- :>is a 1 MIPS machine running for 1 year, or approximately 3.1 x 10^13
- :>instructions. Jobs that take several hundred MIPS-YEARS are not uncommon.
- :
- :Bad guess, unless Crays, 3090s (Yes, I know they barely count) and
- :a bit of CM-2 don't count.
-
- Several hundred MIPS-YEARS is a Cray running for a whole year or two. What jobs
- have you run on Cray-class computers that take a whole year?
-
- The COST of such a job is staggering.
-
- And I run many such jobs.
-
- :Perhaps your code REALLY is MIPS-years in complexity. OR, it could
- :be that failure to design properly leads to that complexity, whereas
- :careful thought could reduce it to months or less. Since I don't
-
- I suggest you read some of my papers in Math. Comp. (and elsewhere). Then you
- can decide whether I know what I am doing.
-
- :understand your problem, I can only speculate. However, I DO know
-
- Methods for computing discrete logs in large finite fields; methods for
- factoring large integers; other problems in computational number theory.
- See my survey article in the August 1991 issue of Notices of the AMS.
-
- :that algorithmic excellence is the way to get speed, NOT dinking
- :around with what you perceive to be a bottleneck at the bit-diddling
- :level. AFTER you have the algorithm right, THEN you start tweaking
-
- And what makes you think that Herman (or I) does not have the right
- algorithm? He knows far more about the mathematics of what he is doing than
- you (or I).
-
- It is assumed that the algorithm is right, in all of these discussions.
-
- :bits. Optimization before design is usually a bad idea.
-
- Spouting aphorisms adds little to this discussion. Especially since you
- are assuming facts not in evidence. (i.e. you are assuming that the
- optimization came before the design).
-
- :>Needless to say, there are people who have to worry about speed FIRST,
- :>and that other considerations are (almost) irrelevent.
- :
- :Are you saying that speed is more important than correct answers?
-
- Now you are being a pedant. It is always assumed that one wants the right
- answer. Noone has ever suggested otherwise. It is assumed that the algorithm
- has been designed and coded to work correctly. What we are discussing is
- the notion that some people seem to have that "correct style" is more
- important than speed.
-
- --
- Bob Silverman
- These are my opinions and not MITRE's.
- Mitre Corporation, Bedford, MA 01730
- "You can lead a horse's ass to knowledge, but you can't make him think"
-