Citation Context

...loads and stores, except the simple but inefficients“spill-everywhere” approach. Even worse, it can happen that some spilling is done even if this actually does not help to make the graph k-colorable =-=[5]-=-. For coalescing, although simple and appealing, conservative coalescing is sometimes not aggressive enough and too many moves may remain in the code. Finally, even if “splitting”, i.e., adding regist...

Citation Context

...eered in coalescing on SSA-form programs already using many properties that SSA-based register allocation relies on. However, they are only concerned with aggressive coalescing. In 2005, three groups =-=[26, 2, 4]-=- independently from each other discovered that the interference graphs of SSA-form programs are chordal. All yet published coalescing techniques tailored to SSA-based register allocation use interfere...

Citation Context

...beyond the scope of this paper; we simply assume that swap instructions are available. The Lost Copy Problem, which is also beyond the scope of this paper, can be rectified by critical edge splitting =-=[2, 4]-=-. 2.2 Chordal Graphs Let G = (V, E) be an undirected graph. For v∈V, N(v) is the set of vertices adjacent to v. An Elimination Order (EO) is a function σ : V → {1, 2, …, |V|} that assigns a unique val...

Citation Context

...n algorithms, as opposed to classical annotation-driven JIT compilation. A fully decoupled approach has been used by Appel and George [13], and studied in the context of SSA-based register allocation =-=[7, 14, 15]-=-. 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 ...

Citation Context

...move to the same register. It has been shown that inserting moves at every program point is unnecessary; it is sufficient that the program be in SSA form and the assignment problem becomes polynomial =-=[5, 8, 15, 23]-=-. Additional work has embraced the coalescing view of register assignment [4, 13, 16, 31, 32]. These approaches perform spill code generation as a separate phase, insert sufficient moves to render the...

Citation Context

...industrial and academic compilers, static or just-in-time, use SSA in their backends, e.g., LLVM [24], Java HotSpot [21], LAO [13], LibFirm [23, 10], Mono [27]. Recent research on register allocation =-=[6, 14, 29]-=- even allows to retain SSA form until the very end of the code generation process. This work investigates the use of SSA properties to simplify and accelerate liveness analysis, i.e., an analysis that...

Citation Context

...ation of spilling costs (if k registers are not enough) and of coalescing costs, i.e., which live-ranges should be fused while keeping the graph k-colorable. These results have been presented at WDDD =-=[7]-=- and LCPC [8]. Critical edges A critical edge is an edge of the control flow graph whose predecessor block has more than one successor and whose successor block has more than one predecessor. Forbiddi...