A Graph-based Model for Building Optimization Sequences: A Study Case on Code Size Reduction

• Nilton Luiz Queiroz Junior UEM
• Anderson Faustino da Silva UEM

Resumo

Embedded Systems applications have several limitations, one of these limitations is the memory size. Modern compilers provide optimization sequences that reduce the code size, contributing to solve this memory issue. This paper presents a new approach to predict optimization sequences for unseen programs. Our strategy builds a graph-based model, which describes the relationship between two optimization, and an engine to traversal the graph and build an optimization sequence. We propose four traverse algorithms. The best traversal algorithm overcomes OPT Oz in 4%. Moreover, our algorithms have a performance similar to state-of-the-art algorithms, through a single evaluation.

Palavras-chave: Compiler, Code Size, Program reasoning