Extended Linear Scan: An Alternate Foundation for Global Register Allocation

Abstract

Abstract. In this paper, we extend past work on Linear Scan register allocation, and propose two Extended Linear Scan (ELS) algorithms that retain the compiletime efficiency of past Linear Scan algorithms while delivering performance that can match or surpass that of Graph Coloring. Specifically, this paper makes the following contributions:-We highlight three fundamental theoretical limitations in using Graph Coloring as a foundation for global register allocation, and introduce a basic Extended Linear Scan algorithm, ELS 0 , which addresses all three limitations for the problem of Spill-Free Register Allocation. -We introduce the ELS 1 algorithm which extends ELS 0 to obtain a greedy algorithm for the problem of Register Allocation with Total Spills. -Finally, we present experimental results to compare the Graph Coloring and Extended Linear Scan algorithms. Our results show that the compile-time speedups for ELS 1 relative to GC were significant, and varied from 15× to 68×. In addition, the resulting execution time improved by up to 5.8%, with an average improvement of 2.3%.Together, these results show that Extended Linear Scan is promising as an alternate foundation for global register allocation, compared to Graph Coloring, due to its compile-time scalability without loss of execution time performance.

Register-allocated paging for big data calculations

“…Wimmer and Franz [31] found that the linear scan algorithm can be simplified if the program complies with a certain constraint, leading to improved compile speed with no increase in run time. Sarkar and Barik [32] developed Extended Linear Scan, providing convincing empirical evidence that this algorithm is more scalable than graph colouring, and competitive in terms of quality. [32] also gives a cogent critique of the graph colouring approach, picking up on Briggs's observation [13] that graph colouring only tackles a reduced form of the register allocation problem.…”

“…Sarkar and Barik [32] developed Extended Linear Scan, providing convincing empirical evidence that this algorithm is more scalable than graph colouring, and competitive in terms of quality. [32] also gives a cogent critique of the graph colouring approach, picking up on Briggs's observation [13] that graph colouring only tackles a reduced form of the register allocation problem.…”

“…My prototype for register-allocated paging has no provision for procedure calls. Although most register allocation techniques are global [32], this just means that they work across a whole procedure (in contrast to local, which works on a basic block, and interprocedural, which works across procedure calls). Thus my prototype's inputs to a global register allocation algorithm are likely to be considerably larger than the algorithm would encounter in its usual context; larger in both the number of instructions and in the number of pseudos.…”

Split Register Allocation: Linear Complexity Without the Performance Penalty

“…Notice that recent versions of the linear scan algorithm are capable of live range splitting [16,17]; they are implicitly based on this decoupled approach. This is not the case for the linear scan implemented in JikesRVM, and leads in practice to spurious spills (to our disadvantage), as we confirmed in our evaluation.…”

Preference-Guided Register Assignment

“…Furthermore, they can take register constraints into account. Recently, Sarkar and Barik [24] introduced more aggressive live-range splitting to linear scan allocators however without performing coalescing.…”