Improved register allocator for x86 and x86_64

by karan for Jikes RVM

The current Linear Scan register allocation algorithm is efficient and simple in implementation. It is a good approach for JIT and dynamic compilers. The algorithm assumes large and rich register sets architecture but this assumption can result in degraded performance on architectures with fewer available register sets. The goal of project is to study and provide modification to linear scan algorithm with assumption that registers are limited and fewer in number.