Progressive spill code placement

Abstract

Register allocation has gained renewed attention in the recent past. Several authors propose a separation of the problem into decoupled sub-tasks including spilling, allocation, assignment, and coalescing. This approach is largely motivated by recent advances in SSA-based register allocation that suggest that a decomposition does not significantly degrade the overall allocation quality.The algorithmic challenges of intra-procedural spilling have been neglected so far and very crude heuristics were employed. In this work, (1) we introduce the constrained mincut (CMC) problem for solving the spilling problem, (2) we provide an integer linear program formulation for computing an optimal solution of CMC, and (3) we devise a progressive Lagrangian solver that is viable for production compilers. Our experiments with Spec2k and MiBench show that optimal solutions are feasible, even for very large programs, and that heuristics leave significant potential behind for small register files.

Studying Optimal Spilling in the Light of SSA

“…In particular, all heuristics that use a spill-everywhere approach, from linear scan to puzzle solving through graph coloring [Chaitin 1982;George and Appel 1996;Pereira and Palsberg 2008;Poletto and Sarkar 1999], use that simplification. This is also true for more recent SSA-based spilling approaches [Ebner et al 2009;Hack 2007;Braun and Hack 2009]. Moreover, this simplification may be combined with the previous assumption, as in the progressive spill-code placement of Ebner et al [2009].…”

“…This is also true for more recent SSA-based spilling approaches [Ebner et al 2009;Hack 2007;Braun and Hack 2009]. Moreover, this simplification may be combined with the previous assumption, as in the progressive spill-code placement of Ebner et al [2009].…”

“…However, a variable may only be assigned to a single allocation class at any given program point. Ebner et al [2009] address the spilling problem for SSA programs using a series of network-flow problems, one for each variable. Nodes correspond to instructions and 47:4 Q. Colombet et al edges to program points where loads can be placed.…”

“…This model is used in the heuristic of Braun and Hack [2009] under the precondition that the program is in CSSA, which guarantees that no actual memory copies are required (how Ebner et al [2009] capture φ-functions is not explained). Indeed, in CSSA, variables connected by φ-functions do not interfere and can be spilled to the same memory location.…”

Register allocation and spilling using the expected distance heuristic

“…Minimizing the costs of spill instructions on the control flow graph can also be posed as a minimum-cut problem. Ebner et al described a separate spill code optimization pass based on the minimum-cut algorithm [30]. They combine all of the load placement problems for spilled virtual registers in a function and solve them simultaneously using ILP.…”