The “Polyhedron Fortran Benchmarks” Suite

The “Polyhedron Fortran Benchmarks” Suite is a set of 17 standard Fortran 90 programs designed to compare the performance of executable files produced by different compilers on a single computer system.

In addition to the benchmark source code and input files, the suite comprises a test harness, an output validator, and parameter files which are used by the harness to control the benchmarking process.  The test harness compiles each benchmark source, and runs it repeatedly, recording the elapsed time for each run, until the estimated timing error is below a specified value.  It is important that the benchmark machine is not running any other significant processes, such as screen-savers, while this happens.  Before finishing, the harness normally invokes the output validator to check that the output matches target values within specified tolerances, and archives output and executable files for future reference.  The whole process may take several hours for each compiler tested.

External Use

You may download and use the benchmarks for your own purposes, provided that, in any published material:-

  • the benchmarks are referred to as the “Polyhedron Fortran Benchmarks”
  • there is a clear link to www.fortran.uk.  Links to individual pages are not recommended, as the target pages may be moved or deleted without notice.
  • if individual tests are quoted, then all tests in the relevant suite must be quoted and given equal prominence. If a single figure is quoted, it must be the Geometric mean execution time. All quoted times should be as reported by the Polyhedron Harness program, with the “standard” timing parameters (7200.0 0.1 10 60). Screen savers, and other tasks which may compete for CPU time should be disabled.
  • the specification of the benchmark machine should be detailed, including the Processor type, speed, cache size etc., main memory amount, type and speed, and Operating System details.
  • the date of the test, and the name of the organization which performed the tests must be specified.

You may also reproduce the current benchmark results tables from www.fortran.uk with suitable attributions and dates.  A link to www.fortran.uk must also be included.

New Benchmarks

If you have a program which you think would be a suitable addition to these benchmarks, please email us.We prefer non-trivial programs that take a significant time to solve a real problem.  Permission to publish the source code is essential.

Publication of Benchmark Data for Other Platforms

If you are willing to collaborate with Polyhedron Solutions, with a view to publishing benchmark data for a different and interesting benchmark machine, please email us.

Download Polyhedron Fortran Benchmarks

View Instructions for Installation and Use