Abstract
The phenomenal power of massively parallel supercomputers can facilitate significant advances in many areas of science and engineering. However, the effort required to develop high-performance applications using currently available parallel programming models is extremely high, and is often a limiting factor in the rate of scientific process. The challenges are compounded by a deepening memory hierarchy and architectural heterogeneity (e.g. multicore CPU's and GPU's). There is therefore interest in developing frameworks that ease the burden of development of high-performance scientific software. This talk will discuss some approaches being pursued towards this goal, the challenges, and potential future directions.