Benchmark suite for heterogeneous computing infrastructures
Clone or download

README.md

Rodinia Benchmark Suite

This repository hosts a fork of the Rodinia benchmark suite, version 3.1. The fork features a clean-up of the code and its build system, as well as additional ports of select benchmarks to the Julia programming language.

For more information, refer to the original home page.

Usage

Execute make in any of the suite's main directories, or in any of the benchmark subfolders. At every level, you can create a Make.user overriding specific settings (eg. cuda/Make.user defining CUDA_DIR).

Every benchmark should contain a run script to start execution. Some benchmarks might also provide a verify script, to generate output and check it against known-good outputs in the results/ folder.

Sometimes, a profile script is provided as well. This script outputs for every computational kernel a comma-separated string denoting the name of the benchmark suite, the name of the kernel, and the execution time in microseconds. This information can then be parsed by a script, see eg. tools/profile.

NOTE: the current focus of development is on a subset of benchmarks, mostly limited to the CUDA and Julia+CUDA suites. If you have interest in other benchmarks, or running other suites, do check the revision history of said benchmark in one of the actively-maintained suite subfolders (ie. cuda/ or julia_cuda/).