Xref: utzoo comp.lang.fortran:4210 comp.lang.c:34358
Path: utzoo!utgpu!watserv1!watmath!att!linac!pacific.mps.ohio-state.edu!zaphod.mps.ohio-state.edu!wuarchive!psuvax1!hsdndev!cmcl2!lanl!ttw
From: ttw@lanl.gov (Tony Warnock)
Newsgroups: comp.lang.fortran,comp.lang.c
Subject: Re: Fortran vs. C for numerical work (SUMMARY)
Message-ID: <7339@lanl.gov>
Date: 29 Nov 90 20:44:15 GMT
References: <9458:Nov2721:51:5590@kramden.acf.nyu.edu> <2392:Nov2902:59:0590@kramden.acf.nyu.edu>
Organization: Los Alamos Natl Lab, Los Alamos, N.M.
Lines: 39



Dan Bernstein asks:
 
RE: >In article <7200@lanl.gov> ttw@lanl.gov (Tony Warnock) writes:
    >>     With respect to speed, almost all machines that I have
    >>     used during the last 25 or so years have had faster
    >>     multiplications than memory accesses.
 
>Hmmm. What machines do you use? In my experience (mostly mainframes and
>supers, some micros, and only recently a bit with minis) local memory
>access is up to several times as fast as integer multiplication. (I
>don't like this situation; converting to floating point just to multiply
>quickly on a Cray seems rather silly.)
 
 
      Model        Multiplication Time     Memory Latency
 
      YMP          5  clock periods         18 clock periods
      XMP          4  clock periods         14 clock periods
      CRAY-1       6  clock periods         11 clock periods
 
      Compaq       25 clock periods         4  clock periods
      Zenith      120 clock periods        30  clock periods
 
    The times on the PC-clones are approximate depending on the type
    of variables being accessed and the sizes of the indices.
 
    Most of my work has been on CRAY type computers. It is always a
    win to avoid memory accesses. On the PC-type machines, memory
    access is faster, but in the particular cases that I have been
    programming, one does many more constant-stride access than random
    accesses, usually in a ratio of many thousand to one. For an LU
    decompositon with partial pivoting, one does rougly N/3 constant
    stride memory accesses for each "random" access. For small N, say
    100 by 100 size matrices or so, one would do about 30
    strength-reduced operations for each memory access. For medium
    (1000 by 1000) problems, the ratio is about 300 and for large
    (10000 by 10000) it is about 30000.