The NAS Parallel Benchmarks are developed by NASA Ames Research Center and include kernel codes derived from typical CFD applications. There exists a original (sequential) version, an MPI version, and an HPF version of each code. This paper evaluates the HPF versions and the sequential versions which we originally rewrite in HPF.
The APR Benchmarks are the benchmark collection including famous benchmarks such as tomcatv, shallow, and grid. They were originally collected and rewritten in HPF by Applied Parallel Research, Inc. for the purpose of evaluating their own HPF compiler (xhpf). They were distributed at the Supercomputing 94 Exhibition. The GENESIS Benchmarks were developed by University of Southampton for benchmarking distributed-memory parallel computers, and are distributed on their web page. The suite also includes some HPF codes. This paper shows the benchmarking results of these HPF codes, modified to suitable to vector parallel machines.
HPFBench was developed by Thinking Machines Corp. with Yale University and Syracuse University and now is distributed by Rice University. It includes some linear-algebraic kernels and application benchmarks. This paper presents the evaluation results for the "as is" codes besides the results for tuned code.
This paper measures overall performance, investigates computation and communication costs, and evaluates scalability for each code. It also presents some comparison with the performance of the equivalent MPI codes. Statistical analysis for the amount of code modification is also provided.
This paper discusses effectiveness of the HPF2 approved extensions, the HPF/JA extensions, and our original extensions in HPF/SX V2 by comparing the benchmarking results of the codes using (or not using) these features. It also studies the possible improvements of HPF/SX V2, based on the benchmarking results.