Overview

Chai is a benchmark suite of Collaborative Heterogeneous Applications for Integrated-architectures. The Chai benchmarks are designed to use the latest features of heterogeneous architectures such as shared virtual memory and system-wide atomics to achieve efficient simultaneous collaboration between host and accelerator devices.

Each benchmark has multiple implementations: OpenCL-U, OpenCL-D, CUDA-U, CUDA-D, CUDA-U-Sim, CUDA-D-Sim, and C++AMP. The versions suffixed with -U use unified memory and system-wide atomics while the versions suffixed with -D use the traditional communication techniques of discrete architectures. The versions suffixed with -Sim run on the gem5-gpu simulator.

Check out our Getting Started page for instructions on how to download and use the benchmarks.

Please cite the following paper if you find our benchmark suite useful:

• J. Gómez-Luna, I. El Hajj, L.-W. Chang, V. Garcia-Flores, S. Garcia de Gonzalo, T. Jablin, A. J. Peña, W.-M. Hwu. Chai: Collaborative Heterogeneous Applications for Integrated-architectures. In Proceedings of IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS), 2017. [pdf] [slides] [bibtex]