eugene@eos.arc.nasa.gov (Eugene Miya) (11/29/90)
I've had problems trying to email copies of the source to people
who have asked for Frank's latest copy of the Livermore Loops,
aka the MFLOPS program, the Livermore Fortran Kernels (LFK), etc.
There are a couple of problems:
1) the source, mostly comments, is nearly 300KB in size.
2) The comments are mostly the results of runs. Frank did a neat thing,
the results of various runs are place back into the source code
making the source a database of results.
3) UUCP has a default limit of 100KB. I cannot always insure a remote
site has uuencode/uudecode, compress/uncompress.
4) I have to check NISTLIB as to why the server there is not working.
You might also try netlib, but you might have the same limit (Dongarra
may have split(1) the file into three parts.
I personally like Frank's Loops alot, even if they do not run very long
on a Cray. They have their problems, but the source is well thought out.
If you still want me to send them, keep trying, yes I know about mailers....
Another rambling later this evening before I go skiing in Utah.
--e.n. miya, NASA Ames Research Center, eugene@eos.arc.nasa.gov
{uunet,mailrus,most gateways}!ames!eugene
AMERICA: CHANGE IT OR LOSE IT.mccalpin@perelandra.cms.udel.edu (John D. McCalpin) (11/29/90)
>>>>> On 29 Nov 90 01:11:56 GMT, eugene@eos.arc.nasa.gov (Eugene Miya) said:
Eugene> [....] Frank's latest copy of the Livermore
Eugene> Loops, aka the MFLOPS program, the Livermore Fortran Kernels
Eugene> (LFK), etc.
Eugene> I personally like Frank's Loops alot, even if they do not run
Eugene> very long on a Cray. They have their problems, but the source
Eugene> is well thought out.
Unfortunately, they run *too* well on cached machines. Even the
longest loops are mostly cacheable on my RS/6000-320, and *none* of my
applications are that small.
I posted the full results to comp.unix.aix a few weeks ago (before we
got comp.benchmarks turned on here), and the top quartile was running
at speeds of 15-22 MFLOPS, which is not characteristic of larger jobs
on the same hardware (by a factor of about 3).
--
John D. McCalpin mccalpin@perelandra.cms.udel.edu
Assistant Professor mccalpin@brahms.udel.edu
College of Marine Studies, U. Del. J.MCCALPIN/OMNETabe@mace.cc.purdue.edu (Vic Abell) (11/30/90)
In article <7628@eos.arc.nasa.gov>, eugene@eos.arc.nasa.gov (Eugene Miya) writes: > I've had problems trying to email copies of the source to people > who have asked for Frank's latest copy of the Livermore Loops, > aka the MFLOPS program, the Livermore Fortran Kernels (LFK), etc. A 1986 Fortran version of MFLOPS is available from netlib@ornl.gov in seven parts. Send mail to netlib with the following message: send livermore from benchmarks Netlib also has a C version of the kernels, dated 22 Aug 88, authored by F. H. McMahon. Obtain it via mail with the message: send livermorec from benchmarks Vic Abell <abe@mace.cc.purdue.edu>
eugene@eos.arc.nasa.gov (Eugene Miya) (11/30/90)
In article <6202@mace.cc.purdue.edu> abe@mace.cc.purdue.edu (Vic Abell) writes: >A 1986 Fortran version of MFLOPS is available from netlib@ornl.gov in This is a version of the loops which have repetition loops (Do this kernel 100 times [or what ever] because this system may have a poor clock). To a smart optimizing compiler it says, "the result of a pass thru this loop really isn't used, I (the compiler) will just optimize this out....." 8^) I will ask Jack to update this and send him the 1990 version. --e.n. miya, NASA Ames Research Center, eugene@eos.arc.nasa.gov {uunet,mailrus,most gateways}!ames!eugene AMERICA: CHANGE IT OR LOSE IT.