Standard Performance Evaluation Corporation

SPEC CFP95 and Compiler/Parallelizing Research

SUIF COMPILER-PARALLELIZED CFP95 BENCHMARKS

Researchers from the Stanford SUIF compiler group (contact: Mary Hall, USC Information Sciences Institute, [email protected] ) have provided SPEC with a compiler-parallelized version of the SPEC CFP95 source code. This source code was produced automatically by the Stanford SUIF compiler, and has been used for extensive experimentation, as described by articles in _IEEE Computer_ (12/96 issue) and ASPLOS '96. It is anticipated these benchmarks will be useful for two purposes:

• for technology transfer, providing insight into how the compiler transforms the applications to yield the reported results; and,
• for further experimentation, such as in architecture simulation studies and analysis of run-time system features.

The SUIF compiler was used in these experiments as a source-to-source translator, taking Fortran as input, and producing as output C and Fortran with calls to a run-time library and other supporting libraries. The release contains the C and Fortran output of the compiler, along with source code for all of the accompanying libraries. Note that the source code is provided for research purposes only. Measurements taken on this source code are not compliant with the SPEC CPU95 Run and Reporting Rules (by virtue of being different source code from what SPEC provides and not using the SPEC CPU95 tool set) and by SPEC license agreement should not be portrayed as such or compared to results which are.

SPEC licensees can contact SPEC for information about this distribution at info@spec.org. This distribution may not be redistributed.