Harnessing OpenCL in Modern Coprocessors
Download as PPT, PDF0 likes703 views
The document discusses the use of OpenCL for developing efficient parallel implementations of relational joins in heterogeneous systems, emphasizing performance modeling and optimization techniques. It covers previous work, challenges in parallel computing, and details an algorithmic approach to optimize relational joins for modern coprocessors. The conclusions highlight the importance of device-specific performance tuning and the ongoing need for advancements in cost modeling and multi-device execution support.
Harnessing OpenCL in Modern Coprocessors
- 1. Harnessing OpenCL in modern coprocessors Unai Lopez-Novoa unai.lopez@ehu.es 06 Aug 2014 Intelligent Systems Group University of the Basque Country UPV/EHU
- 2. Outline • Previous work • Work @ UniMan: Relational Join 1.Motivation 2.Algorithm 3.Results 4.Conclusions 2
- 3. About Myself • PhD Student @ Intelligent Systems Group: 2011 – Now • Research interest: Efficient use of Modern coprocessors • Performance modeling • Code acceleration • Development of parallel implementations • Molecular Dynamics simulation code (MSc thesis) • Kernel Density Estimation (Under review) • Relational Join (Work @ UniMan) 3
- 4. Kernel Density Estimation • Estimate the Probability Density Function of a population • Our use case: Climate models • Challenge: large volumes of data 4 Histogram: KDE:
- 5. Kernel Density Estimation • 1st : Algorithmic rework • 2nd : Parallel implementation: multi/many core processors • Compared to R+MKL and CUDA implementations Naive approach for each evaluation_point e for each sample s d = distance(e,s) e += density (d) Our approach B = computeBoundingBox() for each sample s b = fitBoundingBox(B,s) for each e_point e in b d = distance(e,s) e += density (d) 5
- 7. Join Slide based on: Wu, Lisa, et al. "Navigating big data with high-throughput, energy-efficient data partitioning." Proc. of the 40th Annual International Symposium on Computer Architecture. ACM, 2013. Do sunblock sales correlate with weather? Sales Weather Join-Date(Sales,Weather) Join-Date 7
- 8. Join •Join is everyday operation 8
- 9. Join Goal: Develop a parallel implementation of relational join targeting nowadays heterogeneous systems 9
- 10. Heterogeneous systems • Performance depends on the nature of the application Multi-core •16 cores •250 GFLOP/s Many-core •61 cores •1 TFLOP/s GPU •2880 cores •1.3 TFLOP/s Complex control flow Number crunchingComplex control flow Number crunching 10
- 11. • Wide variety of programming environments in HPC • OpenMP, CUDA, MPI, TBB,… • Our choice: OpenCL Heterogeneous systems NVIDIA SDKIntel SDKAMD SDK Write once Compile Run many 11
- 12. Heterogeneous systems • Cross-platform portability != Performance portability • OpenCL: Abstraction layer • Solution 1: per-device hand-made tuning • Not portable at all • Solution 2: auto tuning • Rely on performance models 12
- 13. Previous work • Collection of performance modeling proposals for latest GPUs and Intel Xeon Phi • Comprehensive analysis of the literature since ~2007 • Organized as: Unai Lopez-Novoa et al. A Survey of Performance Modeling and Simulation Techniques for Accelerator-based Computing IEEE Transactions on Parallel and Distributed Computing, DOI: 10.1109/TPDS.2014.2308216 Execution time estimation Bottleneck highlighting Power cons. estimation Simulators 13
- 14. Types of Join 100 103 104 100 102 Inner Left Outer Right Outer Full Outer 100 100 100 100 103 - 104 - 100 100 - 102 100 100 103 - 104 - - 102 Table A Table B 14
- 15. Algorithm • Biggest debate: Sort or Hash? Hash-join Complexity: Limitation: Extensive use of atomics prevent efficient parallelization O(n + m) Procedure: 1. Hash smaller table 2. Scan larger table Sort-join Sorting increases complexity O(n·log(n)) 1. Sort keys 2. Scan interleaved 15
- 16. Algorithm • Step 1: Sort keys in both tables • Radix sort: speed/scalability sweet spot 100 104 103 103 102 100 100 102 101 102 100 100 102 103 103 104 100 101 102 102 Sort 16
- 17. Algorithm • Step 2: Merge • Add non matching keys for outer joins 100 100 102 103 103 104 100 101 102 102 100 100 100 100 102 102 102 102 Table A Table B Result – Inner Join 17
- 18. Implementation • Steps: 1)Develop a naive OpenCL implementation 2)Optimize per device type 3)Add a cost model for load balancing and partitioning • Experimental setup: • M1: 4 (x2 SMT) Cores Xeon + Xeon Phi + 384 Cores GPU • M2: 12 (x2 SMT) Cores Xeon + Xeon Phi + 2496 Cores GPU • Baseline: ModernGPU (CUDA) 18
- 19. Results 19
- 20. Per-device tuning • Optimizations: • Thread scheduling • Memory management • Overheads: • Compilation • Memory allocation 20
- 21. Optimizations • Per device thread scheduling OpenCL Kernel Threads: Groups: OpenCL Devices Four core CPU 0 1 2 3 61 core Xeon Phi 21 2 3 4 600 1
- 22. • Per device memory management Optimizations Private Local Global OpenCL Device Memory Hierarchy Thread Thread-group Any thread 22 Scope: Registers On-chip RAM Registers RAM RAMRegisters RAM RAM
- 23. Overheads • Compilation • Online compilation: X% of runtime (without I/O) • Memory allocation • Intel SDK: Y % of Merge Step in Xeon Phi OpenCL Program Host code Device code Compilation: Offline (gcc) Online (SDK) 23
- 24. Results 24
- 25. Future work 1) Finish tuning per device code 2) Test join in FPGA 3) Revisit partitioning strategy 4) Support multi-device execution • Develop a cost model that characterizes Join • Split the workload in runtime among existing devices 25
- 26. Conclusions • Performance: device specific code • Performance portability: a) Platform specific code b) Parameterizable code • High OpenCL SDK dependence • Only portable debugging tool: printf • …but still the only portable framework • Future: OpenACC / OpenMP 4.0 ? 26
- 27. Harnessing OpenCL in modern coprocessors Unai Lopez-Novoa unai.lopez@ehu.es 06 Aug 2014 Intelligent Systems Group University of the Basque Country UPV/EHU