HP Exemplar SPP
SPECseis96_SM
The SPECseis96_SM benchmark was run by
Hewlett-Packard Co.
(SPEC license #HPG-012) in Richardson, Texas on a
Exemplar SPP 1600/XA
on 3-Sep-1996. This code was run using the message programming model.
Benchmark Results
System configuration
Disclosure notes
Benchmark Results
| Number CPUs |
SPECseis96_SM* |
Elapsed Time (secs) |
| 2 |
118.2 |
731 |
| 4 |
226.8 |
381 |
| 8 |
411.4 |
210 |
* The SPECseis96_SM metric is 84000/elapased seconds.
Source code base was SPEChpc96.001.
System Configuration
Disclosure Notes
- Source code modifications included changes to file seifft.f to call
the C2DFFT function from the ConvexMLIB math library as a substitute for
the 2D FFT included in the benchmark source base. Also, modifications to
vrfy.f as per the SPEC/HPG code sponsor for correct validation.
- All Fortran files were compiled with the above flags except for file
src/seis/mg3d.f which was compiled without +Oaggressive. The $ALIAS
directive was used in src/util/seifft.f and src/seis/vrfy.f to resolve
symbol names from the ConvexMLIB library (they do not have underscores).
ConvexMLIB V3.0 was linked in. Routines used were: C2DFFT (2-dimensional,
complex FFT) in src/util/seifft.f as a substitute for the source code version;
ccopy and scopy in src/util/seifft.f to replace the source code versions;
and sscal in src/seis/vrfy as a replacement for vsmul (scale an array by
a scalar value).
SPEC High Performance Group
info@specbench.org