Similarity aware context reduction for coarse grained reconfigurable architectures

Coarse-Grained Reconfigurable Architectures (CGRAs) are promising computing platform with better efficiency. However, too much cost is wasted on the path of context storage and transmitting. With statistics of real case, there is obvious similarity between context. We propose one kind of context reduction method based on similarity of context. By making context compress, transmit, and decompress, this method can reduce power and area of CGRA obviously, with little overhead on performance. We propose context dispensing and register model for homogeneous heterogeneous platform. With the optimize of compiler, we get 33% and 118% promotion on power efficiency to tradition architecture, 36% and 181% on area efficiency.