Patterns and Practices for Future Architectures

v 1 Background and Motivation 1 1.1 Benchmarks 3 1.2 Breadth-First Search Framework 4 1.3 Approach 7 2 Implementations 8 2.1 Single CPU, List 8 2.1.1 Kernel 1 8 2.1.2 Kernel 2 10 2.2 Single CPU, Compressed Sparse Row 10 2.2.1 Kernel 1 10 2.2.2 Kernel 2 12 2.3 Single GPU, CSR 13 2.3.1 Kernel 1 13 2.3.2 Kernel 2 14 2.4 Single GPU, Out-of-Core CSR 15 2.4.1 Kernel 1 15 2.4.2 Kernel 2 17 2.5 Multi-CPU, CombBLAS 20 2.5.1 Kernel 1 20 2.5.2 Kernel 2 21 3 Results for All Implementations 24 3.1 Performance Comparison of Individual Approaches 24 3.1.1 Single CPU/GPU Performance 24 3.1.2 Parallel CPU/GPU Performance 26 3.2 Performance vs. Code Complexity 27 3.3 Summary 28 4 Conclusions and Future Work 30 Appendix Test Hardware Configuration 32