G. Gupta and V. Santos Costa. Cut and Side-Effects in And-Or Parallel Prolog. In Proc. 4th IEEE Symposium on Parallel and Distributed Processing, Arlington, 1992.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....the operational semantics is recovered. 1 Introduction In recent years substantial progress has been reported on or parallel logic programming systems, on successful experiments with extensive test sets performed on it, as well as on and parallel extensions of these systems. See, e.g. [11, 13, 5, 10]. However, 7] discusses the lack of semantical consensus, at least for the Godel language, about what should be considered as acceptable implementations of pruning operators. In general, semantical methods, in particular compositional ones, seem somewhat lacking behind in the understanding of ....

G. Gupta and V. Santos Costa. Cuts and side-effects in and-or parallel Prolog. Journal of Logic Programming, 27:45--71, 1996.

....parallelism. Little work has been done on how to alleviate loss of parallelism while handling side effect problem. In addition to loss of parallelism, there is recomputation problem in AND OR parallel execution models systems, which results from the exploitation of AND parallelism. For example, in [8], to achieve the same side effect behaviors as those in sequential model, recomputation of AND parallel goals is adopted. While it is a simple solution for side effect problem in AND OR parallel execution models, another problem arises: too much re dundant recomputation, especially when there ....

....is adopted. To keep the sequential semantics, the recomputation of side effect built ins should not be ruled out. Therefore, if child trees are recomputed for multiple times according to the number of solutions in previous sibling trees, the sequential result will be achieved. The idea is used in [8]. However, the recomputation in [8] is even p(X) q(Y) r(Z) X=1 Y=3 Z=5 X=1 Y=3 Z=6 X=1 Y=4 Z=5 X=1 Y=4 Z=6 X=2 Y=3 Z=5 X=2 Y=3 Z=6 X=2 Y=4 Z=5 X=2 Y=4 Z=6 seed node normal node p(X) X=1,write(X) X=2,write(X) write(1) write(2) q(Y) Y=3,write(Y) Z=4,write(Y) write(3) write(4) 1 2 3 4 1 2 5 8 9 12 ....

[Article contains additional citation context not shown here]

G. Gupta, V. Santos Costa: "Cut and Side-Effects in And-Or Parallel Prolog," In Proc. 4th IEEE Symposium on Parallel and Distributed Processing, Arlington, 1992.

....Little work has been done on how to alleviate loss of parallelism while handling side effect problem. In addition to loss of parallelism, there is recomputation problem in AND OR parallel execution models systems, which results from the exploitation of AND parallelism. For example, in [10], to realize the same side effect results as those in sequential model, recomputation of AND parallel goals is adopted. While it is a simple solution for side effect problem in AND OR parallel execution models, another problem arises: too much redundant recomputation when there are speculative AND ....

....10 13 3 6 11 14 4 7 6 write(6) Figure 6: Side effects Orderings in OR forest be ruled out. Therefore, if child trees are recomputed for multiple times according to the number of solutions in previous sibling trees, the sequential result will be achieved. The idea of recomputation is used in [10]. However, the recomputation in [10] is even more than that in OR parallel execution discussed in previous section. For instance, the q(Y) in the example is recomputed for every branch of p(X) and r(Z) is recomputed for every combination of p(X) s branches and q(Y) s branches, no matter whether ....

[Article contains additional citation context not shown here]

G. Gupta, V. Santos Costa: "Cut and SideEffects in And-Or Parallel Prolog," In Proc. 4th IEEE Symposium on Parallel and Distributed Processing, Arlington, 1992.

....features but in a controlled way, e.g. PEPSys [17, 147] The disadvantage with both these approaches is that existing Prolog programs cannot be parallelized. Recently, however, newer approaches have been developed that allow support for Prolog s sequential semantics even during parallel execution [63, 60, 62, 121, 64]. Another issue that arises in systems that exploit independent and parallelism is whether to recompute solutions of independent goals, or to reuse them. For example, consider the following program for finding cousins at the same generation taken from [137] sg(X, X) person(X) sg(X, Y) ....

....i.e. g 1 : g i Gamma1 , have finished) The execution of the alternative in g i is begun, and when it finishes, the goals g i 1 : g n are started again so that their solutions can be recomputed. Because of recomputation of independent goals ACE can support sequential Prolog semantics [61, 60, 64]. Paged Binding Array based Model ACE can be seen as subsuming Prolog with MUSE while preserving Prolog semantics. In a similar vain, one can combine Prolog with Aurora while preserving Prolog semantics. However, as in the case of AO WAM, the binding array technique has to be extended to ....

[Article contains additional citation context not shown here]

G. Gupta and V. Santos Costa. Cut and Side-Effects in And-Or Parallel Prolog. In Proc. 4th IEEE Symposium on Parallel and Distributed Processing, Arlington, 1992.

....System [HCC88, LWH90, AK91] and the Prolog system [HG90, MH89, CC89] We contend that the same can be done for systems that combine independent and and or parallelism. We present techniques for incorporating side effects in and or parallel systems to execute full Prolog in a companion paper [GS91]. In this paper we present a concrete model which facilitates the inclusion of side effects. We discuss issues such as environment representation, scheduling, and memory management for this model. Our approach is two pronged, first we present an abstract model of and or parallelism for logic ....

....then that goal has to be sequentialised too, and we have to carry on this sequentialisation process right up to the top level query. If, however, the goals are recomputed then this sequentialisation can be avoided, and parallelism exploited even in the presence of cuts and side effects [GS91]. Hence, there is a strong argument for recomputing non deterministic and parallel goals, especially, if they are not pure, and even more so if we want to support Prolog as the user languagey. Additionally, recent simulations done on and or parallelism [SH91] show that typical Prolog programs ....

[Article contains additional citation context not shown here]

G. Gupta, V. Santos-Costa, "Cut and Side Effects in And-Or Parallel Prolog," Technical Report TR-91-26, Department of Computer Science, University of Bristol, Oct. 1991.

....our goal we have devised techniques for preserving Prolog s sequential semantics in systems combining or and independent and parallelism. To ensure that sequential semantics of Prolog is preserved during parallel execution in a system based on C trees, the following rules have to be observed [GS91] when an execution order sensitive predicate is reached in a goal inside a CGE. Consider an extralogical predicate e encountered in one of the alternatives of the goal g i in the CGE (true = g 1 : g n ) that is encountered during execution of a program. During combined independent and ....

....g 1 through g i Gamma1 have finished execution, and; iii) The CGE containing g i recursively satisfies conditions (i) and (ii) with respect to the subtree and outer CGEs that may contain it. In the case of a cut, the conditions above can be somewhat relaxed and the cut can be executed earlier [GS91]. This is because a cut does not affect the solutions of a query, rather it affects the number of solutions reported and the amount of space searched. See [GS91] for more details. 4. Adding Dependent And Parallelism Exploiting dependent and parallelism in presence of non determinism ....

[Article contains additional citation context not shown here]

G. Gupta, V. Santos Costa, "Cut and Side Effects in And-Or Parallel Prolog," Technical Report TR-91-26, Department of Computer Science, University of Bristol, Oct. 1991.

....Both MUSE and Prolog have developed techniques for supporting full Prolog. In ACE we need to extend these techniques to support sequential Prolog semantics in presence of both or and independent andparallelism. Here we can benefit from the principles designed for the C tree abstraction [10]. In designing the solutions to these problems, our aim is to obtain a full Parallel Prolog system that will have low sequential overheads and good parallel speedups. Also, we try to follow the techniques that have been used for Prolog and MUSE as much as possible, as they have proven to be ....

G. Gupta and V. Santos Costa. Cut and SideEffects in And-Or Parallel Prolog. In 4th IEEE Symp. on Parallel and Distr. Processing, '92.

....ideas from previous implementations of other and system and or systems. ffl Systems that are designed to support full Prolog need to know whether side effects builtins can go ahead. The concept of a team allows us to clearly divide the leftmost search into and leftmost and or leftmost searches [8]. One can see a team as an address space shared by several workers and composed of two parts, i) the shared system memory pool, which is in fact shared by every team in the system, and (ii) the SBA, which is private to a team. One can also see the team as formed by and agents (in the spirit of ....

G. Gupta and V. S. Costa. Cuts and Side-Effects in And-Or Parallel Prolog. Journal of Logic Programming, 27(1):45--71, April 1996.

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