Distributing Data and Computation for Hybrid Execution in Heterogeneous Systems
Speaker: Dr. Yonghong Yan
March 28, 2016 - 4:00 p.m. to 5:00 p.m.
Location: DCC 318
Hosted By: Dr. Bulent Yener (x6907)
Successful Exascale application development and deployment requires applications to exploit significantly more concurrency from multiple computational devices including manycore accelerators and multicore processors. An effective solution to portable parallel programming becomes more urgent than ever in this era of architecture diversity and customization. The compiler and runtime must also innovate to automatically map both data and computation onto diverse hardware targets in either shared or disparate memory spaces.
In this talk, the speaker will present the ongoing efforts of developing programming models for distributing computation-intensive parallel loops across heterogeneous resources. The talk will highlight and demonstrate technical solutions to address programming challenges on heterogeneous systems: including portable interfaces for specifying data and computation distribution and alignment, approaches and algorithms to achieve load balance when assigning work to computational different devices, and the runtime system for hybrid executions on CPU, GPU, MIC and FPGA, and for unified memory management in discrete and shared address spaces. The speaker will conclude with discussions on the future computer system and architecture for parallel applications.
Dr. Yonghong Yan is an assistant professor from Oakland University, Rochester MI, a member of OpenMP Architectural Review Board and chair of OpenMP Interoperability Language Subcommittee. Dr. Yan is an expert in parallel computing, compiler technology and high performance computer architecture and systems. He has been working extensively on multiple compiler projects, including OpenMP and OpenACC compilers based on Open64, and the Habanero-C (X10-dialect) compiler and PACE compiler based on ROSE/LLVM. He had been awarded multiple NSF projects as PIs while working as research assistant professor. Dr. Yan received his Ph.D. degree from the University of Houston.
Research and development in Yan’s research group, Parallel Architecture and System Software (PASS) Laboratory, address performance, productivity and energy efficiency challenges in the area of parallel and high performance computing. His research team develop inter-/intra-node programming models, compiler, runtime systems and performance tools based on MPI, OpenMP and LLVM, explore conventional and advanced computer architectures including CPU, vector, GPU, MIC, FPGA, and dataflow system, and support applications ranging from classical HPC, to big data analysis and deep learning, and to medical imaging.
Last updated: March 24, 2016