Petz, <SPEC organization decided on which code to use..>
Yes, that is the primary purpose of the organization,
to agree on limited set of codes from huge number
of candidates, otherwise the benchmark will run
forever. The applications are analyzed in terms
of computational classes, research is conducted on how
representative the particular class is in the
workstation marketplace, and the choice is made.
They do not change code details nor
functionality, but they may request the author
of chosen sub-benchmark to incorporate changes
for reasons of portability across various platforms.
The important part of spec.org is to decide on
what size of data set to use in computations,
to align it reasonably with expected memory
sizes for 3-5 years ahead. After that they
design a huge Perl script that controls
compillations, sets up working directories and
input data sets (up to 10-30MB in size), executes
sub-benchmarks, performs comparisons with
reference data sets, and reports results in
several output formats.
<Compaq Fortran compiler optimized for Athlon will result in higher SPECfp numbers for Athlon>
Somehow I feel it will :)
<the SPECint code is probably 'C'>
Yes it is, entirely. So AMD/Compaq needs to rely on
Microsoft C and their optimizations, with possible
use of third-party preprocessors. For SPECfp2000,
4 out of 14 sub-test are also written in C.
Regards,
- Ali |
