The Register Allocation and Instruction Scheduling Challenge

Abstract

Both register allocation and instruction scheduling are old and open issues in Computer Science, despite the efforts already made to address them separate or jointly. Register allocation may be seen as having two parts: allocation, which decides which values should be in registers, and assignment, which assigns a specific register to each value. Instruction scheduling aims at identifying and moving the instructions in the code, changing their original execution sequence, so that they may run in parallel. Register allocation and instruction scheduling attempt to minimize the execution time of the program, however, they are interdependent and are involved in a prioritization problem. This paper presents a Systematic Literature Review (SLR) related to this problem. From a total of 542 primary studies initially obtained on six databases, 25 studies closely related to this research theme were identified, 12 of them published between 2000 and October 2016. These studies were analyzed to answer the research questions proposed in this SLR, producing useful informations about this theme and about the approaches that, traditionally, have been used to solving this problem. An important finding of this research is the confirmation that this problem still has no definitive solution, and continues to be a relevant challenge for developers, since its solution is closely linked to the quality of the code generated by the compilers in general.