Global code motion/global value numbering

Click, Cliff

doi:10.1145/207110.207154

Cited by 71 publications

(25 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Global code motion has been solved both in isolation [15,22] as well as in integration with register allocation [6,31]. Moreover, attempts at exploiting fine-grained instruction level parallelism using SIMD instructions [35,42] suggest that the selection of such instructions should be done together with global code motion.…”

Section: Related Workmentioning

confidence: 99%

Complete and Practical Universal Instruction Selection

Blindell

Carlsson

Lozano

et al. 2017

ACM Trans. Embed. Comput. Syst.

View full text Add to dashboard Cite

In code generation, instruction selection chooses processor instructions to implement a program under compilation where code quality crucially depends on the choice of instructions. Using methods from combinatorial optimization, this paper proposes an expressive model that integrates global instruction selection with global code motion. The model introduces (1) handling of memory computations and function calls, (2) a method for inserting additional jump instructions where necessary, (3) a dependency-based technique to ensure correct combinations of instructions, (4) value reuse to improve code quality, and (5) an objective function that reduces compilation time and increases scalability by exploiting bounding techniques. The approach is demonstrated to be complete and practical, competitive with LLVM, and potentially optimal (w.r.t. the model) for medium-sized functions. The results show that combinatorial optimization for instruction selection is well-suited to exploit the potential of modern processors in embedded systems.

show abstract

Section: Related Workmentioning

confidence: 99%

Complete and Practical Universal Instruction Selection

Blindell

Carlsson

Lozano

et al. 2017

ACM Trans. Embed. Comput. Syst.

View full text Add to dashboard Cite

show abstract

“…In this algorithm, the only modification is that ψ operations have to be handled with the same rules as the φ operations. We have also ported dead code elimination [9] and global value numbering [4] algorithms to this representation, and we expect that partial redundancy elimination [2], and induction variable analysis [14] should be easy to adapt.…”

Section: The Psi-ssa Representationmentioning

confidence: 99%

Improvements to the Psi-SSA representation

Ferrière

2007

Proceedingsof the 10th International Workshop on Software &Amp; Compilers for Embedded Systems - SCOPES '07

View full text Add to dashboard Cite

show abstract

“…Although SSA provides the freedom to think outside the lexical box because each assignment essentially becomes its own variable, even SSAPRE does not take advantage of this but regards expressions in terms of the source variables from which their operands come; in fact, it makes stronger assumptions about the namespace than basic SSA [34]. The fact that many redundant results in a program do not come from lexically identical expressions (nor do all lexically identical expressions produce the same results) has motivated research to make PRE more effective [5] and led Click to assert that "in practice, GVN finds a larger set of congruences than PRE" [9].…”

Section: Related Workmentioning

confidence: 99%

“…SSA is a major boon for GVN, since (SSA versions of) variables are always in the same class, and therefore so are expressions on those versions in the same class, statically and globally. Congruences can be found by hashing [9,6] or by first assuming all expressions to be equal and splitting classes when incongruences are found [3,16,30]. Since identifying an operation as redundant depends on the operation having the same global number as an earlier operation, in traditional GVN such an earlier operation must dominate (i.e., occur on all paths to) the later operation.…”

Section: Gvnmentioning

confidence: 99%

Value-Based Partial Redundancy Elimination

VanDrunen

Hosking

2004

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract. Partial redundancy elimination (PRE) is a program transformation that identifies and eliminates expressions that are redundant on at least one (but not necessarily all) execution paths. Global value numbering (GVN) is a program analysis and transformation that identifies operations that compute the same value and eliminates operations that are redundant. A weakness of PRE is that it traditionally considers only expressions that are lexically equivalent. A weakness of GVN is that it traditionally considers only operations that are fully redundant. In this paper, we examine the work that has been done on PRE and GVN and present a hybrid algorithm that combines the strengths of each. The contributions of this work are a framework for thinking about expressions and values without source-level lexical constraints, a system of data-flow equations for determining insertion points, and a practical algorithm for extending a simple hash-based GVN for PRE. Our implementation subsumes GVN statically and, on most benchmarks, in terms of performance.

show abstract

Global code motion/global value numbering

Cited by 71 publications

References 15 publications

Complete and Practical Universal Instruction Selection

Complete and Practical Universal Instruction Selection

Improvements to the Psi-SSA representation

Value-Based Partial Redundancy Elimination

Contact Info

Product

Resources

About