eugene@eos.arc.nasa.gov (Eugene Miya) (01/05/89)
Some comments about the NIST lib posting.
I used some of these programs during the early stages of ETA-10
and Cray-YMP testing and am preparing some for "future" machines.
Several of the codes are extremely hardware and software specific.
It is existing supercomputer oriented. There are some Sequent and
Convex versions of code.
No makefiles. Little documentation (NAS kernels -- we gave a copy
of the TR in TeX). No shar files.
I offered the source to the Gabriel LISP benchmarks and was turned down.
The bibliography is a collection of the previous NBSLIB
set of references, and references collected by Alan Jay Smith,
myself, and Rafael Savaadera which I converted to refer format.
Several of the entries are pointers to other available databases.
John Rice's program has one error (A typo), I will correct this
momentarily. (I didn't document what I found, a simple typo).
Mendez's benchmarks are written for a 32- IBM type computer
(everything is declared double precision). It has IBM type vectorization
compiler directives embedded in the source.
Several benchmarks using CDC/ETA Q[78] style vector calls.
These are not portable and not fortran.
The LANL stuff expects CTSS operating system calls (you can
comment this out).
The LLNL and NAS stuff some time expects 64-bit precision (1E+50)
[You can tell a computer user by his or her perspective on word size.]
The NIST will not collect results, too risky. Liability. Slander.
The effort is a good effort at heart, but under funded and under-appreciated.
Another gross generalization from
--eugene miya, NASA Ames Research Center, eugene@aurora.arc.nasa.gov
resident cynic at the Rock of Ages Home for Retired Hackers:
"Mailers?! HA!", "If my mail does not reach you, please accept my apology."
{uunet,hplabs,ncar,decwrl,allegra,tektronix}!ames!aurora!eugene
"Send mail, avoid follow-ups. If enough, I'll summarize."