Advertisement

Mozart : Efficient Composition of Library Functions for Heterogeneous Execution

  • Rajkishore BarikEmail author
  • Tatiana Shpeisman
  • Hongbo Rong
  • Chunling Hu
  • Victor W. Lee
  • Todd A. Anderson
  • Greg Henry
  • Hai Liu
  • Youfeng Wu
  • Paul Petersen
  • Geoff Lowney
Conference paper
Part of the Lecture Notes in Computer Science book series (LNCS, volume 11403)

Abstract

Current processor trend is to couple a commodity processor with a GPU, a co-processor, or an accelerator. To unleash the full computational power of such heterogeneous systems is a daunting task: programmers often resort to heterogeneous scheduling runtime frameworks that use device specific library routines. However, highly-tuned libraries do not compose very well across heterogeneous architectures. That is, important performance-oriented optimizations such as data locality and reuse “across” library calls is not fully exploited. In this paper, we present a framework, called Mozart, to extend existing library frameworks to efficiently compose a sequence of library calls for heterogeneous execution. Mozart consists of two components: library description (LD) and library composition runtime. We advocate library writers to wrap existing libraries using LD in order to provide their performance parameters on heterogeneous cores, no programmer intervention is necessary. Our runtime performs composition of libraries via task-fission, load balances among heterogeneous cores using information from LD, and automatically adapts to runtime behavior of an application. We evaluate Mozart on a Xeon + 2 Xeon Phi system using the High Performance Linpack benchmark which is the most popular benchmark to rank supercomputers in TOP500 and show GFLOPS improvement of 31.7 % over MKL with Automatic Offload and 6.7 % over hand-optimized ninja code.

References

  1. 1.
  2. 2.
  3. 3.
    Intel Math Kernel Library Automatic Offload for Intel Xeon Phi Coprocessor. https://software.intel.com/en-us/articles/math-kernel-library-automatic-offload-for-intel-xeon-phi-coprocessor
  4. 4.
    AlSaber, N., Kulkarni, M.: Semcache: semantics-aware caching for efficient GPU offloading. In: Proceedings of the 27th International ACM Conference on International Conference on Supercomputing, ICS 2013, pp. 421–432. ACM, New York (2013)Google Scholar
  5. 5.
    Augonnet, C., Thibault, S., Namyst, R., Wacrenier, P.-A.: StarPU: a unified platform for task scheduling on heterogeneous multicore architectures. Concurr. Comput.: Pract. Exp. 23(2), 187–198 (2011)CrossRefGoogle Scholar
  6. 6.
    Barik, R., et al.: Efficient mapping of irregular C++ applications to integrated GPUs. In: IEEE/ACM International Symposium on Code Generation and Optimization (CGO) (2014)Google Scholar
  7. 7.
    Belviranli, M.E., Bhuyan, L.N., Gupta, R.: A dynamic self-scheduling scheme for heterogeneous multiprocessor architectures. ACM Trans. Archit. Code Optim. 9(4), 57:1–57:20 (2013)CrossRefGoogle Scholar
  8. 8.
    Bezanson, J., Karpinski, S., Shah, V.B., Edelman, A.: Julia: a fast dynamic language for technical computing. CoRR, abs/1209.5145 (2012)Google Scholar
  9. 9.
    Bueno, J., Martorell, X., Badia, R.M., Ayguadé, E., Labarta, J.: Implementing OmpSs support for regions of data in architectures with multiple address spaces. In: Proceedings of the 27th International ACM Conference on International Conference on Supercomputing, ICS 2013, pp. 359–368. ACM, New York (2013)Google Scholar
  10. 10.
    Cederman, D., Tsigas, P.: On dynamic load balancing on graphics processors. In: Proceedings of the 23rd ACM SIGGRAPH/EUROGRAPHICS Symposium on Graphics Hardware, GH 2008, Aire-la-Ville, Switzerland, pp. 57–64 (2008)Google Scholar
  11. 11.
    Chatterjee, S., Grossman, M., Sbîrlea, A., Sarkar, V.: Dynamic task parallelism with a GPU work-stealing runtime system. In: Rajopadhye, S., Mills Strout, M. (eds.) LCPC 2011. LNCS, vol. 7146, pp. 203–217. Springer, Heidelberg (2013).  https://doi.org/10.1007/978-3-642-36036-7_14CrossRefGoogle Scholar
  12. 12.
    Chen, L., Villa, O., Krishnamoorthy, S., Gao, G.R.: Dynamic load balancing on single- and multi-GPU systems. In: IEEE International Symposium on Parallel and Distributed Processing (IPDPS), pp. 1–12 (2010)Google Scholar
  13. 13.
    Chronaki, K., Rico, A., Badia, R.M., Ayguadé, E., Labarta, J., Valero, M.: Criticality-aware dynamic task scheduling for heterogeneous architectures. In: Proceedings of the 29th ACM on International Conference on Supercomputing, ICS 2015, New York, NY, USA, pp. 329–338 (2015)Google Scholar
  14. 14.
    Grewe, D., Wang, Z., O’Boyle, M.F.P.: OpenCL task partitioning in the presence of GPU contention. In: Caşcaval, C., Montesinos, P. (eds.) LCPC 2013. LNCS, vol. 8664, pp. 87–101. Springer, Cham (2014).  https://doi.org/10.1007/978-3-319-09967-5_5CrossRefGoogle Scholar
  15. 15.
    Guo, Z., Pierce, M., Fox, G., Zhou, M.: Automatic task re-organization in MapReduce. In: 2011 IEEE International Conference on Cluster Computing (CLUSTER), pp. 335–343 (2011)Google Scholar
  16. 16.
    Guyer, S.Z., Lin, C.: Broadway: a software architecture for scientific computing. In: IFIPS Working Group 2.5: Software Architecture for Scientific Computing (2000)Google Scholar
  17. 17.
    Haidar, A., Tomov, S., Arturov, K., Guney, M., Story, S., Dongarra, J.: LU, QR, and Cholesky factorizations: programming model, performance analysis and optimization techniques for the Intel Knights Landing Xeon Phi. In: 2016 IEEE High Performance Extreme Computing Conference (HPEC), pp. 1–7, September 2016Google Scholar
  18. 18.
    Harris, T., Maas, M., Marathe, V.J.: Callisto: co-scheduling parallel runtime systems. In: Proceedings of the Ninth European Conference on Computer Systems, EuroSys 2014, pp. 24:1–24:14. ACM, New York (2014)Google Scholar
  19. 19.
    Heinecke, A., et al.: Design and implementation of the Linpack benchmark for single and multi-node systems based on Intel Xeon Phi coprocessor. In: 2013 IEEE 27th International Symposium on Parallel and Distributed Processing (IPDPS), Washington, DC, USA, pp. 126–137 (2013)Google Scholar
  20. 20.
    Hugo, A.-E., Guermouche, A., Wacrenier, P.-A., Namyst, R.: Composing multiple StarPU applications over heterogeneous machines: a supervised approach. IJHPCA 28(3), 285–300 (2014)Google Scholar
  21. 21.
    Kennedy, K., et al.: Telescoping languages: a strategy for automatic generation of scientific problem-solving systems from annotated libraries. J. Parallel Distrib. Comput. 61(12), 1803–1826 (2001)CrossRefGoogle Scholar
  22. 22.
    Kim, J., Kim, H., Lee, J.H., Lee, J.: Achieving a single compute device image in OpenCL for multiple GPUs. In: Proceedings of the 16th ACM Symposium on Principles and Practice of Parallel Programming, PPoPP 2011, NY, USA, pp. 277–288 (2011)Google Scholar
  23. 23.
    Kim, J., Seo, S., Lee, J., Nah, J., Jo, G., Lee, J.: SnuCL: an OpenCL framework for heterogeneous CPU/GPU clusters. In: Proceedings of the 26th ACM International Conference on Supercomputing, ICS 2012, pp. 341–352 (2012)Google Scholar
  24. 24.
    Komoda, T., Miwa, S., Nakamura, H.: Communication library to overlap computation and communication for OpenCL application. In: Proceedings of the 2012 IEEE 26th International Parallel and Distributed Processing Symposium Workshops and Ph.D. Forum, IPDPSW 2012, Washington, DC, USA, pp. 567–573 (2012)Google Scholar
  25. 25.
    Lee, J., Samadi, M., Park, Y., Mahlke, S.: Transparent CPU-GPU collaboration for data-parallel kernels on heterogeneous systems. In: Proceedings of the 22nd International Conference on Parallel Architectures and Compilation Techniques, PACT (2013)Google Scholar
  26. 26.
    Luk, C.-K., Hong, S., Kim, H.: Qilin: exploiting parallelism on heterogeneous multiprocessors with adaptive mapping. In: Proceedings of the 42nd Annual IEEE/ACM International Symposium on Microarchitecture, MICRO, NY, USA, pp. 45–55 (2009)Google Scholar
  27. 27.
    Majo, Z., Gross, T.R.: A library for portable and composable data locality optimizations for NUMA systems. ACM Trans. Parallel Comput. 3(4), 20:1–20:32 (2017) CrossRefGoogle Scholar
  28. 28.
    Marjanović, V., Labarta, J., Ayguadé, E., Valero, M.: Overlapping communication and computation by using a hybrid MPI/SMPSs approach. In: Proceedings of the 24th ACM International Conference on Supercomputing, ICS 2010, New York, NY, USA, pp. 5–16 (2010)Google Scholar
  29. 29.
    Ogata, Y., Endo, T., Maruyama, N., Matsuoka, S.: An efficient, model-based CPU-GPU heterogeneous FFT library. In: IEEE International Symposium on Parallel and Distributed Processing, IPDPS, pp. 1–10 (2008)Google Scholar
  30. 30.
    Pan, H., Hindman, B., Asanović, K.: Lithe: enabling efficient composition of parallel libraries. In: Proceedings of the First USENIX Conference on Hot Topics in Parallelism, HotPar 2009, p. 11. USENIX Association, Berkeley (2009)Google Scholar
  31. 31.
    Pandit, P., Govindarajan, R.: Fluidic kernels: cooperative execution of OpenCL programs on multiple heterogeneous devices. In: Proceedings of Annual IEEE/ACM International Symposium on Code Generation and Optimization, CGO 2014, NY, USA, pp. 273:273–273:283 (2014)Google Scholar
  32. 32.
    Phothilimthana, P.M., Ansel, J., Ragan-Kelley, J., Amarasinghe, S.: Portable performance on heterogeneous architectures. In: Proceedings of the Eighteenth International Conference on Architectural Support for Programming Languages and Operating Systems, ASPLOS 2013, NY, USA, pp. 431–444 (2013)Google Scholar
  33. 33.
    Planas, J., Badia, R.M., Ayguadé, E., Labarta, J.: SSMART: smart scheduling of multi-architecture tasks on heterogeneous systems. In: Proceedings of the Second Workshop on Accelerator Programming Using Directives, WACCPD 2015, pp. 1:1–1:11 (2015)Google Scholar
  34. 34.
    Ravi, V.T., Ma, W., Chiu, D., Agrawal, G.: Compiler and runtime support for enabling generalized reduction computations on heterogeneous parallel configurations. In: Proceedings of the 24th ACM International Conference on Supercomputing, ICS 2010, NY, USA, pp. 137–146 (2010)Google Scholar
  35. 35.
    Ravi, V.T., Agrawal, G.: A dynamic scheduling framework for emerging heterogeneous systems. In: 2011 18th International Conference on High Performance Computing (HiPC), pp. 1–10, December 2011Google Scholar
  36. 36.
    Rey, A., Igual, F.D., Prieto-Matías, M.: HeSP: a simulation framework for solving the task scheduling-partitioning problem on heterogeneous architectures. In: Dutot, P.-F., Trystram, D. (eds.) Euro-Par 2016. LNCS, vol. 9833, pp. 183–195. Springer, Cham (2016).  https://doi.org/10.1007/978-3-319-43659-3_14CrossRefGoogle Scholar
  37. 37.
    Satish, N., Kim, C., Chhugani, J., Dubey, P.: Large-scale energy-efficient graph traversal: a path to efficient data-intensive supercomputing. In: Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis, SC 2012, Los Alamitos, CA, USA, pp. 14:1–14:11 (2012)Google Scholar
  38. 38.
    Satish, N.: Can traditional programming bridge the ninja performance gap for parallel computing applications? In: Proceedings of the 39th Annual International Symposium on Computer Architecture, ISCA 2012, Washington, DC, USA, pp. 440–451 (2012)Google Scholar
  39. 39.
    Schaa, D., Kaeli, D.: Exploring the multiple-GPU design space. In: IEEE International Symposium on Parallel Distributed Processing, IPDPS, pp. 1–12 (2009)Google Scholar
  40. 40.
    Song, F., Dongarra, J.: A scalable framework for heterogeneous GPU-based clusters. In: Proceedings of the 24th ACM Symposium on Parallelism in Algorithms and Architectures, SPAA 2012, NY, USA, pp. 91–100 (2012)Google Scholar
  41. 41.
    Song, L., Feng, M., Ravi, N., Yang, Y., Chakradhar, S.: COMP: compiler optimizations for manycore processors. In: Proceedings of the 47th Annual IEEE/ACM International Symposium on Microarchitecture, MICRO-47, Washington, DC, USA, pp. 659–671 (2014)Google Scholar
  42. 42.
    Thomas, N., Tanase, G., Tkachyshyn, O., Perdue, J., Amato, N.M., Rauchwerger, L.: A framework for adaptive algorithm selection in STAPL. In: Proceedings of the Tenth ACM SIGPLAN Symposium on Principles and Practice of Parallel Programming, PPoPP 2005, pp. 277–288. ACM, New York (2005)Google Scholar
  43. 43.
    Tzenakis, G., Papatriantafyllou, A., Vandierendonck, H., Pratikakis, P., Nikolopoulos, D.S.: BDDT: block-level dynamic dependence analysis for task-based parallelism. In: Wu, C., Cohen, A. (eds.) APPT 2013. LNCS, vol. 8299, pp. 17–31. Springer, Heidelberg (2013).  https://doi.org/10.1007/978-3-642-45293-2_2CrossRefGoogle Scholar
  44. 44.
    Wu, W., Bouteiller, A., Bosilca, G., Faverge, M., Dongarra, J.: Hierarchical DAG scheduling for hybrid distributed systems. In: 29th IEEE International Parallel and Distributed Processing Symposium (IPDPS), Hyderabad, India, May 2015Google Scholar
  45. 45.
    Yu, H., Rauchwerger, L.: An adaptive algorithm selection framework for reduction parallelization. IEEE Trans. Parallel Distrib. Syst. 17(10), 1084–1096 (2006)CrossRefGoogle Scholar
  46. 46.
    Zandifar, M., Jabbar, M.A., Majidi, A., Keyes, D., Amato, N.M., Rauchwerger, L.: Composing algorithmic skeletons to express high-performance scientific applications. In: Proceedings of the 29th ACM on International Conference on Supercomputing, ICS 2015, pp. 415–424. ACM, New York (2015)Google Scholar

Copyright information

© Springer Nature Switzerland AG 2019

Authors and Affiliations

  • Rajkishore Barik
    • 1
    Email author
  • Tatiana Shpeisman
    • 1
  • Hongbo Rong
    • 1
  • Chunling Hu
    • 1
  • Victor W. Lee
    • 1
  • Todd A. Anderson
    • 1
  • Greg Henry
    • 1
  • Hai Liu
    • 1
  • Youfeng Wu
    • 1
  • Paul Petersen
    • 1
  • Geoff Lowney
    • 1
  1. 1.Intel CorporationSanta ClaraUSA

Personalised recommendations