Cosmin E. Oancea

Cosmin
Assistant Professor
HIPERFIT Centre, DIKU , Office 01-0-017
University of Copenhagen
Nørre Campus,
Universitetsparken 5,
DK-2100 Copenhagen Ø
Denmark

Phone:   +45 23 82 80 86
E-mail:   cosmin.oancea@diku.dk
 

Research Interests

I am interested in a variaty of topics from computer systems field, including: static, dynamic and hybrid techniques for automatic parallelization (optimising compilers), programming language design and implementation, memory management, computer algebra, etc.

2011-Present @DIKU, University of Copenhagen

Working in collaboration with Troels Henriksen, Martin Elsman, Fritz Henglein, Ken Friis Larsen on designing and implementing Futhark: a pure-functional, map-reduce language supporting (parallel) bulk operation on regular arrays. Futhark currently targets efficient execution on GPGPUs and seeks a common ground between the functional and imperative approaches. Results show that Futhark has efficiencies comparable with hand-written GPGPU code on a number of Rodinia benchmarks and on several real-world applications from financial domain.

2009-2011 collaborating with Lawrence Rauchwerger at Parasol Lab, Texas A&M University

2007-2009 collaborating with Alan Mycroft at the Computer Laboratory, University of Cambridge

2000-2006 PhD studies, Advisor Stephen M. Watt, The University of Western Ontario



Organization of Conferences (Selected)

Co-Chair FHPC'17.
PC Member: CHIUW'17, PPoPP'17, PACT'16, PACT'15, FHPC'15.

Academic Awards and Honors (Selected)

DIKU Teacher of the Year Award (2015 and 2013)

Supervised Students

  • Troels Henriksen. "Exploiting functional invariants to optimise parallelism: a dataflow approach", MSc Thesis, DIKU, University of Copenhagen, February 2014. PDF



  • Selected Refereed Technical Papers

    1. T. Henriksen, N. G. W. Serup, M. Elsman, F. Henglein and C. E. Oancea. "Futhark: Purely Functional GPU-programming with Nested Parallelism and In-place Array Updates", Int. Conf. Programming Languages Design and Implementation (PLDI), Barcelona, Spain, 2017.

    2. T. Henriksen, M. Dybdal, H. Urms, A. S. Kiehn, D. Gavin, H. Abelskov, M. Elsman and C. E. Oancea. "APL on GPUs: A TAIL from the Past, Scribbled in Futhark", 5th Int. Workshop on Functional High Performance Computing (FHPC), Nara, Japan, 2016. PDF

    3. T. Henriksen, K. F. Larsen and C. E. Oancea. "Design and GPGPU Performance of Futhark’s Redomap Construct", 3rd Int. Workshop on Libraries, Languages and Compilers for Programming (ARRAY), pp. 17-24, Santa Barbara, US, 2016. PDF

    4. C. Andreetta, V. Begot, J. Berthold, M. Elsman, F. Henglein, T. Henriksen, M. Nordfang and C. E. Oancea. "FinPar: A Parallel Financial Benchmark", ACM Journal Trans. Archit. Code Optim. (TACO), vol. 13(2), pp. 18.1–18.27, 2016.

    5. F. C. Gieseke, C. E. Oance, A. Mahabal, C. Igle, Tom Heskes. "Bigger buffer k-d trees on multi-many-core systems", Big Data Deep Learning in High Performance Computing, Springer, pp. 172-180, 2016.

    6. C. E. Oancea and L. Rauchwerger. "Scalable conditional-induction variables (CIV) analysis", 13th IEEE/ACM International Symposium on Code Generation and Optimization (CGO'15), San Francisco, USA, pp. 213-224, 2015. PDF

    7. Fabian C. Gieseke, Justin Heinermann, Cosmin E. Oancea and Christian Igel. "Buffer k-d trees: processing massive nearest neighbor queries on GPUs", 31st Int. Conf. on Machine Learning (ICML'14), Beijing, China, 2014. PDF

    8. Fabian C. Gieseke, Kai L. Polsterer, Cosmin E. Oancea and Christian Igel. "Speedy Greedy Feature Selection: Better Redshift Estimation via Massive Parallelism.", 22nd European Symposium on Artificial Neural Networks (ESANN), pp. 87-92, Belgium, 2014.

    9. Troels Henriksen, Martin Elsman, and Cosmin E. Oancea. "A Hybrid Approach to Size Inference in Futhark", 3rd ACM-SIGPLAN Workshop on Functional High-Performance Computing (FHPC), Guthenburg, Sweden, September 2014. PDF

    10. Troels Henriksen and Cosmin E. Oancea. "Bounds Checking: An Instance of Hybrid Analysis", ACM SIGPLAN Int. Workshop on Libraries, Languages and Compilers for Array Programming (ARRAY). Edinburgh, UK, June 2014. PDF

    11. Troels Henriksen and Cosmin E. Oancea. "A T2 Graph-Reduction Approach To Fusion", 2nd ACM SIGPLAN Workshop on Functional High-Performance Computing. Boston, Massachusetts. September 2013. PDF

    12. Cosmin E. Oancea, Christian Andreetta, Jost Berthold, Alain Frisch, and Fritz Henglein. "Financial software on GPUs: between Haskell and Fortran.", 1st ACM SIGPLAN workshop on Functional high-performance computing (FHPC ‘12). Copenhagen 2012. PDF

    13. Cosmin E. Oancea and Lawrence Rauchwerger. "Logical Inference Techniques for Loop Parallelization", 33rd ACM-SIGPLAN Conf. on Prog. Lang. Design and Implem. (PLDI'12), pp 509-520, June 2012. PDF

    14. Cosmin E. Oancea and Lawrence Rauchwerger. "A Hybrid Approach to Proving Memory Reference Monotonicity", 24th Int. Lang. and Compilers for Parallel Computing (LCPC'11), LNCS, Vol 7146, pp 61-75, Sept 2013. PDF

    15. Cosmin E. Oancea and Stephen M. Watt. "An Architecture for Generic Extensions", Science of Computer Programming Journal, Elsevier, Vol 76(4), pp 258-277, doi:10.1016/j.scico.2009.09.008, 2011.

    16. Cosmin E. Oancea, Alan Mycroft and Tim Harris. "A Lightweight In-Place Implementation for Software Thread-Level Speculation", 21st ACM Symposium on Parallelism in Algorithms and Architectures (SPAA'09), August 2009, Calgary, Canada. PDF

    17. Cosmin E. Oancea, Alan Mycroft and Stephen M. Watt. "A New Approach to Parallelising Tracing Algorithms", International Symposium on Memory Management (ISMM'09), June 2009, Dublin, Ireland. PDF

    18. Cosmin E. Oancea and Alan Mycroft. "Set-Congruence Dynamic Analysis for Software Thread-Level Speculation", Languages and Compilers for Parallel Computing (LCPC'08) 21st Annual Workshop, August 2008, Edmonton, Canada. PDF

    19. Cosmin E. Oancea and Alan Mycroft. "Software Thread-Level Speculation: an Optimistic Library Implementation", International Workshop on Multicore Software Engineering (IWMSE'08), pp 23-32 (ACM Digital Library), May 2008, Leipzig, Germany. PDF

    20. Cosmin E. Oancea and Stephen M. Watt. "Generic Library Extension in a Heterogeneous Environment", Library-Centric Software Design LCSD'06, pp. 25-35, October 2006, Portland, USA. PDF

    21. Cosmin E. Oancea. "Parametric Polymorphism for Software Component Architectures and Related Optimizations", PhD Thesis, Department of Computer Science, The University of Western Ontario, Advisor: Stephen M. Watt, July 2005, London, ON, Canada. PDF

    22. Cosmin E. Oancea and Stephen M. Watt. "Parametric Polymorphism for Software Component Architectures", ACM Object-Oriented Programming, Systems, Languages and Applications OOPSLA'05, pp. 147 - 166, October 2005, San Diego, USA. PDF

    23. Cosmin E. Oancea and Stephen M. Watt. "Domains and Expressions: An Interface Between Two Approaches to Computer Algebra," ACM International Symposium on Symbolic and Algebraic Computation ISSAC'05, pp. 261 - 269, July 2005, Beijing, China. PDF

    24. Cosmin E. Oancea, Jason W. Selby, Mark Giesbrecht, and Stephen M. Watt. "Distributed Models of Thread Level Speculation", International Conference on Parallel and Distributed Processing Techniques and Applications PDPTA'05, pp. 920-927, June 2005, Las Vegas, USA. PDF

    25. Cosmin E. Oancea, Clare So, and Stephen M. Watt. "Generalization in Maple", Maple Conference, pp. 277-382, July 2005, Waterloo, Canada.

    26. Yannis Chicha, Michael Lloyd, Cosmin E. Oancea, and Stephen M. Watt. "Parametric Polymorphism for Computer Algebra Software Components", 6th International Symposium on Symbolic and Numeric Algorithms for Scientific Computing SYNASC'04, pp. 119 - 130, Sept. 2004, Timisoara, Romania. PDF

    27. Cosmin E. Oancea and Stephen M. Watt. "A Framework for Using Aldor Libraries with Maple", EACA'04 satellite conference to the ISSAC'04, pp. 219 - 224, June 2004, Universidad de Santander, Spain. PDF

    28. Francois Boulier, Marc Moreno Maza, and Cosmin E. Oancea. "A new henselian construction and its application to polynomial GCDs over direct products of fields", EACA'04 satellite conference to the ISSAC'04, pp. 47 - 52, June 2004, Universidad de Santander, Spain

    29. Radu G. Belu, Cosmin E. Oancea, and A. C. Belu "A Wavelet-based Simulation for Identification and Classification of Short-Term Power Disturbances and Transients as a Teaching Aid", 2003 Annual ASEE Conference, pp. 1380-4. June 2003, Nashville, TN, USA.

    30. Radu G. Belu, Cosmin E. Oancea, and A. C. Belu "A 2-D indoor radio propagation modeling", FIE (Frontiers in Education) Confference, pp. 1780-4. November 2003, Boulder, CO, USA.






    2009-2011 @Parasol Laboratory, Department of Computer Science and Engineering, Texas A&M University

    In the imperative context, entirely static analysis has been successfully applied to (auto-) parallelizing loops with simple control flow and affine array indexing, but is often conservative when these assumptions do not hold. On the other hand, entirely dynamic analysis of memory-reference traces, e.g., thread-level speculation, can be applied aggressively, but incurs significant (and often non-scalable) overheads.

    In collaboration with Prof. Lawrence Rauchwerger, I have studied how to unify static and dynamic analysis into a hybrid compiler framework for the Fortran77 language, which succeeds in detecting and optimizing parallelism for a large class of difficult loops at negligible (or in the worst case scalable) runtime overheads.

    The idea has been to use static, interprocedural analysis (i) to model loop parallelism as an equation on abstract sets (of array references), and furthermore (ii) to extract a cascade of sufficient conditions for parallelism, of increasing time complexity, that are verified at runtime until one succeeds. An evaluation of ~30 benchmarks from SPEC and Perfect-Club suites (~1000 loops) demonstrates that the approach is viable and outperforms commercial compilers by a significant margin.




    2007-2009 @Computer Laboratory, University of Cambridge

    In collaboration with Prof. Alan Mycroft, I have studied topics related to dynamic extraction and optimization of parallelism. In particular:




    2000-2006 @ the Computer Science Department, The University of Western Ontario, ORCCA Lab

    As a PhD student, advised by Prof. Stephen M. Watt, I have studied various aspects related to language interoperability. The original motivation for my research was the observation that although parametric polymorphism was already mainstream, software component architectures of that time, such as CORBA, JNI, DCOM, were lagging behind the advances in common programming practice and were not supporting parametric polymorphism across languages boundaries.

    In this context, I investigated how to resolve different binding times and parametrization semantics in a range of representative languages, such as C++, Java, Aldor, Maple, and have identified a common ground that could be suitable mapped to different language bindings. This work has resulted into two frameworks: Alma that allows interoperability between two very different computer algebra systems (Aldor and Maple), and GIDL (Generic Interface Definition Language), a more systematic solution for parametric polymorphism that is designed as a generic extension framework, that can be easily adapted to work on top of various component architectures.