Provided by: European Design and Automation Association
Date Added: Jan 2012
The need to improve the performance/energy ratio has caused a general trend towards increased heterogeneity in multi- and manycore systems, where general-purpose computing cores are complemented with energy-efficient special-purpose accelerators located either on or off-chip, as in, e.g., GPU-supported systems. However, this trend has also brought new, fundamental problems for the design, optimization and maintenance of software for such systems. Current programming approaches are either platform-specific, such as CUDA for Nvidia GPUs, or are portable but at a low level of abstraction, as with, e.g., OpenCL.