G. Puebla and M. Hermenegildo. Abstract multiple specialization and its application to program parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-316, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....and for automatic specializations based on the parameterized strategy presented in Section 2.5. The MAP system is described in Appendix A. The source code for the examples can be found in Appendix B. There are various methods for specializing programs w.r.t. properties of their input data [10, 16, 53, 61, 66]. Contextual specialization can be viewed as one of these methods. In this chapter we have presented a set of transformation rules and an automated strategy for the contextual specialization of constraint logic programs over a domain For our specialization strategy we have assumed the existence ....

....most specific generalization over finite terms. Among the many techniques for simplifying and manipulating constraints to get more efficient specialized programs, here we want to mention the following methods which are related to ours. C HAI I I 2. C I I) I UAL b I C IALIZA I II 1 C LI In [66] the authors propose a method based on abstract interpretation, for the implementation of multiple specialization of logic programs. Particular emphasis is given to program parallelization. Similarly to their work, our specialization strategy may produce several specialized versions of the same ....

PUEBLA, G., AND HERMENEGILDO, M. Abstract multiple specialization and its application to program parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs J1, 2&3 (November 1999), 279-316.

....redundant checks of constraint satisfiability, and reducing the number of arguments of predicates. In Section 5 we compare our contextual specialization technique to related work in the field of partial deduction [10,21] and specialization of programs w.r.t. properties of their context of use [3,19]. 2 Rules and Strategies for Partial Evaluation of CLP In this section we introduce a set of transformation rules and a strategy for partial evaluation of CLP programs over a domain D. Our rules differ from those studied in previous papers [1,4,13] in that they are tailored to partial ....

....of methods which can be used for the specialization of programs w.r.t. properties of their context of use. Among these methods we would like to mention: i) the methods for specializing logic programs presented in [2,15] ii) the method for multiple specialization of logic programs presented in [19], iii) the method for reducing the workload of the constraint solving engine presented in [14] and (iv) the method for parameterized partial evaluation of functional programs presented in [3] In [2,15] the authors present very general approaches to the specialization of definite logic programs ....

[Article contains additional citation context not shown here]

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-- 316, November 1999.

....their e ective application sometimes we need information on the operational or denotational semantics of the whole program. Various techniques which combine program transformation and abstract interpretation have been developed, especially for the task of program specialization (see, for instance, [28,43,60] in the case of logic programs) but also for the veri cation of concurrent systems (see [45] We believe that this line of research is very promising. Finally, we would like to notice that the program derivation techniques we have described in this chapter are essentially oriented to the ....

G. Puebla and M. Hermenegildo. Abstract multiple specialization and its application to program parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279316, November 1999.

....have been started in the last few years in different contexts. For instance, integration of abstract interpretation in termination analysis and in program specialization has received increasing attention. An integrated analysis tool, linked to an optimization tool, is under development in Madrid ([8,19]) Also, in the Mercury language, various types of information, partly declared, partly inferred, are combined in a powerful way to do both optimization and verification ( 24] Taking this a step further, it has been suggested by Maurice Bruynooghe to further integrate these tools in a ....

G. Puebla and M. Hermenegildo. Abstract multiple specialization and its application to program parallelization. Journal of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 1999. To appear.

....the specialization of programs w.r.t. properties of their context of use. Among these methods we would like to mention: i) the methods for specializing logic programs presented in [2, 14] ii) the method for multiple specialization of logic programs based on abstract interpretation presented in [18], and (iii) the method for parameterized partial evaluation of functional programs presented in [3] In [2, 14] the authors present very general approaches to the specialization of logic programs w.r.t. properties which, however, do not seem to be amenable to full mechanization. In particular, ....

....instead of predicates de ned by logic programs. By doing so, we were able to consider much more e ective transformation rules, whose applicability conditions are decidable, if we assume that the constraint theory is decidable, as it is often the case. A more practical approach is presented in [18], where a technique for the implementation of multiple specialization is proposed. This technique performs global analysis based on abstract interpretation and produces a polyvariant specialized program which contains di erent, optimized versions of the same predicate for di erent uses. The work ....

G. Puebla and M. Hermenegildo. Abstract multiple specialization and its application to program parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279316, November 1999.

No context found.

G. Puebla and M. Hermenegildo. Abstract multiple specialization and its application to program parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-316, 1999.

No context found.

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-316, November 1999.

No context found.

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279--316, November 1999.

No context found.

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-- 316, November 1999.

No context found.

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-316, 1999.

.... is accurate enough and the direction of approximation performed by the analysis is the appropriate one [37,10] this is the case for the properties var and does not fail in the example above) We say that the properties are abstractly reducible (to either true or false) or abstractly executable [39]. Note that, if the analysis is precise (in the sense that the abstract operations do not lose information beyond the abstraction implied by the abstraction function used [15] and, obviously, terminates, then the native properties will be decidable in all cases. However, since there may in ....

....as argument) introduced in Section 1.5, is that if an error is detected, a more informative message can be printed than if only the postcondition responsible for the error is available. Note also that the definition of predicate sorted num list has been optimized by the abstract specializer [38,39] by eliminating the number tests. This is possible by taking advantage of type analysis which tells us that on success the second argument of qsort is a list of numbers. 16 how properties which are not natively understood by the analysis can also be useful for detecting bugs at compile time) ....

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. Journal of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 1999. To appear.

....can then be used to optimize the program by specializing the program procedures for particular cases. If analysis is multivariant then several variants may be generated for a given procedure [6] allowing di erent optimizations to be performed, a technique which has shown quite relevant in practice [3, 2]. Separating programs into modules is a well known technique which has proven very useful in program development and maintenance. Unfortunately, though analysis and specialization of modular programs has recently received some attention [4, 5] most existing analysis and specialization ....

....procedure from another module, which is reachable in the call dependency graph of M from P : CP , the entry P : CP P 2 : CP 2 is added to the inter module dependency graph. Example 1 Consider a simple program consisting of the three modules below: module main. import rev. main(X) A = [1,2,3], rev(A,B) module rev. import app. rev(X,Y) X = Y = rev(X,Y) X = A X1] rev(X1,Y1) B = A] app(Y1,B,Y) module app. app(X,Y,Z) X = Y = Z. app(X,Y,Z) X = A X1] Z = A Z1] app(X1,Y1,Z1) Consider compiling module rev using groundness analysis domain P ....

[Article contains additional citation context not shown here]

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-316, November 1999.

....Specialization [JGS93, Gal93, Leu97] aims at optimizing programs by specializing the code to particular cases. Though much of the discussion we present could apply to other kinds of specialization, we will focus on abstract multiple specialization [JLW90, GH91, Bru91, Win92, PH95, KMM 96, PH99] which directly uses the results of global analysis in order to optimize the program introducing multivariant specialization if required. Example 5.1 Consider the length predicate in Figure 2. This is a good example of a reversible predicate which can be used in several modes. For example, in ....

....di erent issues which appear when considering (abstract) specialization of programs split into modules. We assume that the program has already been analyzed according to the scenarios and algorithms presented in Section 4. An important feature of the multiple specialization algorithm in [PH95, PH99] is that it allows minimizing the number of versions implemented in the nal program. For this, there is a ow of information among modules which propagates information bottom up and corresponds to the potential optimizations which are possible in di erent versions of a predicate (if they were ....

[Article contains additional citation context not shown here]

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279-316, November 1999.

....of queries Q i E(A; P; Q) E(A;P; Q) If A and A 0 are equivalent but A 0 is simpler then obviously A 0 should be used instead for run time checking. Generating equivalent Compile time simpli cation of assertions can be done using techniques such as abstract specialization (see, e.g. [28, 27]) However, space limitations prevent us from discussing further this interesting issue. 6 Implementation We have implemented the schema of Figure 1 as a generic framework. This genericity means that di erent instances of the tools involved in the schema for di erent CLP dialects can be ....

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. Journal of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 1999. In press.

....problem is overcome by means of multiple program specialization where di erent versions of the predicate are generated for each use. Each version is then optimized for the particular subset of input values with which it is to be used. The abstract multiple specialization technique used in Ciaopp [24] has the advantage that it can be incorporated with little or no modi cation of some existing abstract interpreters, provided they are multivariant (PLAI and similar frameworks have this property) This specialization can be used for example to improve automatic parallelization in those cases ....

....mode (the only one which will succeed in Prolog) will be quickly directed to the optimized versions of the predicates and iterate on them. This is because the specializer has been able to detect this optimization as an invariant of the loop. The complete code for this example can be found in [24]. The multiple specialization implemented incorporates a minimization algorithm which keeps in the nal program as few versions as possible while not losing opportunities for optimization. For example, eight versions of predicate vmul (for vector multiplication) would be generated if no ....

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. J. of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 41(2&3):279{ 316, November 1999. In press.

....as argument) introduced in Section 5, is that if an error is detected, a more informative message can be printed than if only the postcondition responsible for the error is available. Note also that the definition of predicate sorted num list has been optimized by the abstract specializer [33, 34] by eliminating the number tests. This is possible by taking advantage of type analysis which tells us that on success the second argument of qsort is a list of numbers. 13 If we run the program with run time checks in order to sort, say, list [1,2] the CIAO system generates the following error ....

G. Puebla and M. Hermenegildo. Abstract Multiple Specialization and its Application to Program Parallelization. Journal of Logic Programming. Special Issue on Synthesis, Transformation and Analysis of Logic Programs, 1999. To appear.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC