Granularity analysis and optimization for multiprocessor systems
B.S. Carlson, Oscar Roberto López-Bonilla · 1998
This dissertation presents a design tool that is used to perform analysis and optimization of the granularity of programs for multiprocessor systems. The tool is capable of analyzing and optimizing granularity for arbitrary software programs and arbitrary heterogeneous multiprocessor architectures. The presented tool was developed to work with the Ptolemy system. The data flow graph specification capability of Ptolemy is used to capture the software specification. Code was written to use Ptolemy's user interface to specify multiprocessor systems. The granularity analysis/optimization software performs the analysis/optimization of a program specified as a data flow graph with respect to a multiprocessor architecture specified in Ptolemy. The software can be used interactively to explore the performance of various architectures and software programs, or can be used to automatically optimize a given flow graph for execution on a specified architecture. The tool is intended to help users increase their programming productivity when using and programming parallel computers. It can also be used as a test-bed for designing and comparing the different elements of parallel computer architectures, e.g. static mapping algorithms, routers, interconnection network topologies, etc.