This paper describes an algorithm for the code generation of determinacy testing for nondeterminate flat concurrent logic programming languages. Languages such as Andorra and Pandora require that procedure invocations suspend if there is more than one candidate clause potentially satisfying the goal. The algorithm described has been developed specifically for a variant of flat Pandora based on FGHC, although the concepts are general. We have extended Kliger and Shapiro's decision-graph construction algorithm to compile “don't-know” procedures that must suspend for nondeterminate goal invocation. The determinacy test is compiled into a decision graph quite different from those of committed-choice procedures: They are more similar to decision trees optimized by code sharing. We present both empirical data of compilation results (code size and graph characteristics), and a correctness proof for our code-generation algorithm.

A parallel copying garbage collection algorithm for symbolic languages executing on shared-memory multiprocessors is proposed. The algorithm is an extension of Baker's sequential algorithm with a novel method of heap allocation to prevent fragmentation and facilitate load distribution during garbage collection. An implementation of the algorithm within a concurrent logic programming system, VPIM, has been evaluated and the results, for a wide selection of benchmarks, are analyzed here. The authors show 1) how much the algorithm reduces the contention for critical sections during garbage collection, 2) how well the load-balancing strategy works and its expected overheads, and 3) the expected speedup achieved by the algorithm.>

Granularity collecting (e.g., in OR-parallel Prolog) and removal of synchronization points (e.g., in AND-parallel committed-choice languages), are in general useful techniques for speeding up parallel logic programs. In this paper we explain these techniques with a unifying example: the Semigroup Problem, the calculation of the members of a semigroup given an initial set of generators. Performance measurements on a Sequent Symmetry multiprocessor are presented as evidence of the utility of the paradigms.

This paper describes the design and performance of a tightly-coupled shared-memory coherent cache optimized for the execution of parallel logic programming architectures. The cache utilizes a copy-back write-allocation protocol having five states and a hardware lock mechanism. Optimizations for logic programming are introduced in four software-controlled memory access commands: direct-write, exclusive-read, read-purge, and read-invalidate. In this paper we describe these operations and present simulated measurements showing their performance advantage for an architecture of the committed-choice language KL1. The cache optimizations also improve the performance of non-committed-choice languages, such as OR-parallel Prolog. A version of the cache design described here is currently being implemented for ICOT's Parallel Inference Machine (PIM).

Several local data buffers are proposed and measurements are presented for variations of the Warren abstract machine (WAM) architecture for Prolog. Choice-point buffers, stack buffers, split-stack buffers, multiple-register sets, copyback caches, and smart caches are examined. Statistics collected from four benchmark programs indicate that small conventional local memories perform quite well because of the WAM's high locality. The data memory performance results are equally valid for native code and reduced instruction set implementations of Prolog.>