[comp.benchmarks] cached LLNL Loops

eugene@eos.arc.nasa.gov (Eugene Miya) (11/30/90)

John writes about the Loops fitting into cache.
This problem is not unique, as first noted in Linpack when 100x100 was
the system size.

Yes, benchmarks have a problem with scale.  I'll get to that in disucssion.
But the solution to this problem occured to me some time ago.
We have to start making machine-readable header files (*.h files
for C programs).  Using a macro processor like Unix m4 (easy to use)
one can subtitute important literals (constants or strings) into
benchmark programs (this avoids [or maybe we should use] the C pre-processor).
These machine header files MUST be made public domain.  They will
increase portability of programs in general.  But that's something worth
discussing later.

But still caches vary in size and function.....

--e.n. miya, NASA Ames Research Center, eugene@eos.arc.nasa.gov
  {uunet,mailrus,most gateways}!ames!eugene
  AMERICA: CHANGE IT OR LOSE IT.