Hyper-Heuristics to Automatically Target Code to Computer Architectures
- Sandia National Lab. (SNL-NM), Albuquerque, NM (United States)
With each new generation of High Performance Computing (HPC) architecture, the gap between peak theoretical performance and the observed performance is growing. The goal of this research is to develop a tool to utilize hyper-heuristics to target code to a computational environment. In order to test this, sorting algorithms will be evolved on several different architectures. The final solutions will then migrate to all other architectures and their fitnesses compared. If the natively-evolved algorithms out-perform all others, then it can be concluded that the tool successfully targeted its solutions to the architecture of origin. This is the first step towards creating a program-agnostic tool for optimizing code to the native environment. The results are pending testing on the high performance cluster. If it can be shown that the tool is able to optimize solutions for the environment, then the door opens to automatically optimizing entire programs.
- Research Organization:
- Sandia National Laboratories (SNL-NM), Albuquerque, NM (United States)
- Sponsoring Organization:
- USDOE National Nuclear Security Administration (NNSA)
- DOE Contract Number:
- AC04-94AL85000
- OSTI ID:
- 1618239
- Report Number(s):
- SAND--2016-2954R; 637555
- Country of Publication:
- United States
- Language:
- English
Similar Records
Visualization for Hyper-Heuristics: Back-End Processing
Visualization for Hyper-Heuristics: Front-End Graphical User Interface