Using dynamic programming to generate optimized code in a Graham-Glanville style code generator

Christopher, Thomas W.; Hatcher, Philip J.; Kukuk, Ronald C.

doi:10.1145/502874.502877

Cited by 16 publications

(6 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Christopher et al [76] attempted to address this concern by using the concepts of the Graham-Glanville approach but extending the parser to produce all parse trees, and then select the one that yields the best assembly code. This was achieved by replacing the original LR parser with an implementation of Earley's algorithm [104], and although this scheme certainly improves code quality-at least in theory-it does so at the cost of enumerating all parse trees, which is often too expensive in practice.…”

Section: Maintaining Multiple Parse Trees For Better Code Qualitymentioning

confidence: 99%

“…As the target machines are becoming evermore complex, placing higher demands for more flexible and integrated code generation, the instruction selection problem may be in greater need of study than ever before. [255] ME L DMACS Wilcox [338] ME L Wasilew [331] TC L Donegan [101] ME L Tirrell [319] ME L Weingart [332] TC L Ammann et al [12,13] ME L Young [350] ME L Newcomer [263] TC L Simoneaux [309] ME L Snyder [310] ME L Fraser [140,141] ME L Ripken [294] TC L Glanville and Graham [158] TC L Johnson [191,192] TC L PCC Harrison [170] ME + L Cattell [67,70,234] TC L PQCC Auslander and Hopkins [33] ME + L Ganapathi and Fischer [146,147,148,149] TC L Krumme and Ackley [218] ME L Deutsch and Schiffman [96] ME L SMALLTALK-80 Christopher et al [76] TC L Davidson and Fraser [91] ME + L GCC, ACK, ZEPHYR/VPO Henry [177] TC L Aho et al [6,7,321] TC L TWIG Hatcher and Christopher [172] TC L Horspool [184] TC L Fraser and Wendt [135] ME + L Giegerich and Schmal [157] TC L PR SC OP MO DO IB IN KNOWN APPLICATIONS Hatcher and Tuller [174] TC L UNH-CODEGEN Pelegrí-Ll...…”

Section: Future Challengesmentioning

confidence: 99%

See 1 more Smart Citation

Instruction Selection

Blindell

2016

View full text Add to dashboard Cite

Section: Maintaining Multiple Parse Trees For Better Code Qualitymentioning

confidence: 99%

Section: Future Challengesmentioning

confidence: 99%

Instruction Selection

Blindell

2016

View full text Add to dashboard Cite

“…The resulting code generator, however, is simple and fast. In both the static and dynamic BUPMs, the cost analysis is usually carried out using dynamic programming [1,8,33]. For a comparison of the performance of static and dynamic BUPMs, see Henry and Damron [23,22] and Henry [20,21].…”

Section: Other Workmentioning

confidence: 99%

Code generation = A* + BURS

Nymeyer

Katoen

Westra

et al. 1996

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract.A system called BURS that is based on term rewrite systems and a search algorithm A are combined to produce a code generator that generates optimal code. The theory underlying BURS is re-developed, formalised and explained in this work. The search algorithm uses a cost heuristic that is derived from the term rewrite system to direct the search. The advantage of using a search algorithm is that we need to compute only those costs that may be part of an optimal rewrite sequence.Key words: compiler generators, code generation, term rewrite systems, search algorithms, formal techniquesCompiler building is a time-consuming and error-prone activity. Building the frontend (i.e. scanner, parser and intermediate-code generator) is straightforward-the theory is well established, and there is ample tool support. The main problem lies with the back-end, namely the code generator and optimiser-there is little theory and even less tool support. Generating a code generator from an abstract specification, also called automatic code generation, remains a very difficult problem.Pattern matching and selection is a general class of code-generation technique that has been studied in many forms. The most successful form uses a code generator that works predominantly bottom-up; a so-called bottom-up pattern matcher (BUPM). A variation of this technique is based on term rewrite systems. This technique, popularised under the name BURS, and developed by Pelegrí-Llopart and Graham [31], has arguably been considered the state of the art in automatic code generation. BURS, which stands for bottom-up rewrite system, has an underlying theory that is poorly understood. The theory has received virtually no attention in the literature since its initial publication [30]. The only research that has been carried out in this technique has been on improved tablecompression methods. Many researchers who claim to use BURS theory (e.g. [32,14]) generally use 'weaker' tree grammars instead of term rewrite systems, or equate BURS with a system that does a static cost analysis (e.g. [13]). We argue that a static cost analysis is neither necessary nor sufficient to warrant a BURS label, and that a system that is based on tree grammars cannot be a BURS.In this work we present an outline of formal BURS theory. Due to space restrictions, we present the full theory in [29]. This formalisation of BURS contrasts with the semiformal work of Pelegrí-Llopart and Graham. But there are other important differences in our work. We do not, for example, use instruction costs to do static pattern selection, and we do not use dynamic programming. Instead we use a heuristic search algorithm that only needs to dynamically compute costs for those patterns that may contribute to ?

show abstract

“…The disadvantage stems from the fact that cost computations associated with dynamic programming are performed at code generation time, thus slowing down the code generator. Dynamic programming using a top-down traversal was also used by Christopher et al [1984]. Weisberger and Wilhelm [1988] describe top-down and bottom-up techniques to generate code.…”

Section: Introductionmentioning

confidence: 99%

Extending Graham-Glanville techniques for optimal code generation

2000

ACM Trans. Program. Lang. Syst.

View full text Add to dashboard Cite

We propose a new technique for constructing code-generator generators, which combines the advantages of the Graham-Glanville parsing technique and the bottom-up tree parsing approach. Machine descriptions are similar to Yacc specifications. The construction effectively generates a pushdown automaton as the matching device. This device is able to handle ambiguous grammars, and can be used to generate locally optimal code without the use of heuristics. Cost computations are performed at preprocessing time. The class of regular tree grammars augmented with costs that can be handled by our system properly includes those that can be handled by bottom-up systems based on finite-state tree parsing automata. Parsing time is linear in the size of the subject tree. We have tested the system on specifications for some systems and report table sizes.

show abstract

Using dynamic programming to generate optimized code in a Graham-Glanville style code generator

Cited by 16 publications

References 12 publications

Instruction Selection

Instruction Selection

Code generation = A* + BURS

Extending Graham-Glanville techniques for optimal code generation

Contact Info

Product

Resources

About