|
 |
Descriptive Complexity Based Approach to Coarse Grained Reconfigurable Computing
Research Group
Soumyendu Raha and S K Nandy
Computational and Data Sciences, IISc
and Computational and Data Sciences, IISc
In experiments with implementation of matrix computation algorithms rich in matrix multiplication it has been observed that datapath based algorithm-architecture co-design results in parallelisms different than conventional computational based tuning. The objective of this project is to find a descriptive complexity based explanation for such results. We shall possibly use a resource bounded descriptive complexity (Kolmogorov Complexity) to characterize parallel computation on Coarse Grained Reconfigurable Architecture and exploit the results, if possible. to identify datapaths and task flows that give the most efficient (minimum entropy) parallelism for common algorithms, such as, the matrix operations. This research will help architect tensor processing pipelines for next generation processing cores.
|
Home 
