Efficient High Performance Computing on Heterogeneous Platforms

• DOI: 10.4233/uuid:3efc8aae-e31f-47a4-a92c-b9da05917ada
• Publication Date: 2015-11-24
• Authors: Jie Shen
• Keywords: Symmetric multiprocessor system; Kernel (algebra)

Heterogeneous platforms are mixes of different processing units in a compute node (e.g., CPUs+GPUs, CPU+MICs) or chip package APUs). This type keeps gaining popularity various computer systems ranging from supercomputers to mobile devices. In this context, improving their efficiency and usability has become increasingly important. thesis, we develop systematic methods for large variety data parallel applications efficiently utilize heterogeneous platforms. Specifically, (1) evaluate the suitability OpenCL as unified programming model computing improve OpenCL's heterogenous platforms; (2) workload partitioning framework accelerate imbalanced on platforms, where match heterogeneity platform with imbalance workload; (3) propose model-based prediction method correctly quickly predict optimal partitioning, maximizing performance gain while speeding up process; (4) generalize approach which improves both balanced applications, datasets execution scenarios, hardware mixes; (5) design an application analyzer that analyzes kernel structures enables strategies accordingly obtain high wide applicability To summarize, thesis demonstrates right solution, performance-wise, many classes shows how can be achieved systematically.