TitleOptimization of Software using OpenTuner and Machine Learning
Number of students 1
Language English
Supervisor Juan J. Durillo
Description In this thesis a Support Vector Machine based approach to Software Autotuning shall be implemented and compared to state of the art approaches. The main idea is to move the time consuming process of recompiling a program with different running parameters and evaluating their goodness in terms of an objective function away from the active tuning phase and into a separate training phase. In addition to the implementation of this strategy as an extension to the Opentuner Framework, it shall be compared to classical tuning strategies provided by the framework in terms of a set of various Stencil Codes.
  • Study Support Vector Machines
  • Review of Parallelism
  • Get familiar with OpenTuner
  • Implement several algorithms within OpenTuner
  • Benchmarking analysis different Stencil Codes
Additional Information