Optimizing Scientific Application Loops on Stream Processors

In this paper, the authors describe a graph coloring compiler framework to allocate on-chip SRF (Stream Register File) storage for optimizing scientific applications on stream processors. Their framework consists of first applying enabling optimizations such as loop unrolling to expose stream reuse and opportunities for maximizing parallelism, i.e., overlapping kernel execution and memory transfers. Then the three SRF management tasks are solved in a unified manner via graph coloring: placing streams in the SRF, exploiting stream use and maximizing parallelism.

Provided by: Association for Computing Machinery Topic: Hardware Date Added: Jun 2008 Format: PDF

Find By Topic