SC²S Colloquium - August 10, 2016

From Sccswiki
Jump to navigation Jump to search
Date: August 10, 2016
Room: 02.07.023
Time: 3:00 pm, s.t.

Nathaniel Knapp: Parasite Local Scalability Profiling for Parallelization

The Lehrstuhl für Rechnertechnik und Rechnerorganisation at TUM has developed an interactive parallelization tool, Parceive, which analyzes data dependencies and profiling information collected from a trace of a sequential program produced using dynamic binary instrumentation. Using Parceive's visualization, programmers can easily identify areas of the code that are parallelization candidates, at multiple granularity levels. The object of this thesis is to design and implement Parasite, a local scalability profiling tool that will extend the capabilities of Parceive. Parasite takes a parallel program or a sequential program with parallel annotations as input, and ranks function call sites in the program for the user in order of their potential speedup from parallelization. The Parceive visualization will also be extended to add a feature where users can select shared memory locations which could require lock synchronization, and then see the effect of these synchronizations on scalability.