Adaptive Systems for the Dynamic Run-time Optimization of Programs
Abstract
This thesis investigates adaptive compiler systems that perform, during program execution, code optimizations based on the dynamic behavior of the program as opposed to current approaches that employ a fixed code generation strategy, i.e., one in which a predetermined set of code optimizations are applied at compile-time to an entire program. The main problems associated with such adaptive systems are studied in general: which optimizations to apply to what parts of the program and when. Two different optimization strategies result: an ideal scheme which is not practical to implement, and a more basic scheme that is. The design of a practical system is discussed for the FORTRAN IV language. The system was implemented and tested with programs having different behavioral characteristics.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 01, 1974
- Accession Number
- AD0784880
Entities
People
- Gilbert J. Hansen
Organizations
- Carnegie Mellon University