A general approach for run-time specialization and its application to C 论文

摘要

Specializing programs with respect to run-time invariants is an optimization technique that has shown to improve the performance of programs substantially. It allows a program to adapt to execution contexts that are valid for a limited time.Run-time specialization is being actively investigated in a variety of areas. For example, recently, major operating system research projects have been focusing on run-time specialization as a means to obtain efficiency from highly extensible and parameterized systems.This paper describes a general approach to run-time specialization. For a given program and a declaration of its run-time invariants, it automatically produces source templates at compile time, and transforms them so that they can be processed by a standard compiler. At run time, only minor operations need to be performed: selecting and copying templates, filling holes with run-time values, and relocating jump targets. As a consequence, run-time specialization is performed very efficiently and thus does not require the specialized code to be executed many times before its cost is amortized.Our approach improves on previous work in that: (1) templates are automatically produced from the source program and its invariants, (2) the approach is not machine dependent, (3) it is formally defined and proved correct, (4) it is efficient, as shown by our implementation for the C language.