ABSTRACT
Retargetable compilation has posed many challenges to researchers. The ultimate goal of the field is to develop a universal compiler that generates high-quality code for all known architectures. Among all existing compilers, GNU Compiler Collection (GCC) comes closest to this goal by using a myriad of formal algorithms and engineering hacks. Currently, GCC officially supports more than 30 architectures commonly used around the world [12]. Still, for many less known architectures, there is no GCC support. Many recent application-specific instruction set processors (ASIPs) are designed with such specialized architectural features that it is not a straightforward task to create customized compilers for them, let alone retarget GCC. Despite the difficulties, good retargetable compilers are highly desirable for these architectures, especially during their development stages in which many candidate domain-specific features need to be evaluated. If a retargetable compiler is capable of utilizing these candidate features and generating optimized code accordingly, their effectiveness can be quickly evaluated by running the resulting code on an equally retargetable simulator. In this regard, retargetable compilation and retargetable simulation are closely related problems, but retargetable compilation is a much harder one.
