Section: New Results

Studying Optimal Spilling in the Light of SSA

Participants : Florian Brandner, Quentin Colombet, Alain Darte.

Recent developments in register allocation, mostly linked to static single assignment (SSA) form, have shown that it is possible to decouple the problem in two successive phases: a first spilling phase places load and store instructions so that the register pressure at all program points is small enough, a second assignment and coalescing phase maps the remaining variables to physical registers and reduces the number of move instructions among registers. We focused on the first phase, for which many open questions remained: in particular, we studied the notion of optimal spilling (what can be expressed?) and the impact of SSA form (does it help?).

To identify the important features for optimal spilling on load-store architectures, we developed a new integer linear programming formulation, more accurate and expressive than previous approaches. Among other features, we can express SSA $\phi$-functions, memory-to-memory copies, and the fact that a value can be stored simultaneously in a register and in memory. We implemented this formulation in LAO and analyzed in details the static and dynamic results obtained for the SPEC INT 2000 and EEMBC 1.1 benchmarks. We can draw, among others, the following conclusions: a) rematerialization is extremely important, b) SSA complicates the formulation of optimal spilling, especially because of memory coalescing when the code is not in CSSA, c) micro-architectural features are significant and thus have to be accounted for, d) significant savings can be obtained in terms of static spill costs, cache miss rates, and dynamic instruction counts.

This work has been presented at the CASES'11 conference [8] .