Efficient memory management in a merged heap/stack prolog machine

Abstract

Traditional Prolog implementations are based on the stack/heap memory architecture: the stack holds local variables and control information, whereas the heap stores data objects which outlive procedure activations. A s t a c k frame can be deallocated when an activation ends while heap space can only be reclaimed on backtracking or by garbage collection. Conventional garbage collection methods may yield poor performance. In this paper, I present a n o vel memory management approach used in the implementation of Logic Virtual Machine (LVM). The LVM combines the stack and the heap into a single memory block for all dynamical memory requirements, supports coarse-grain two-stream uni cation, and embeds an e cient garbage collection algorithm, the Chronological Garbage Collection (CGC), to reclaim useless memory cells. An experimental LVM emulator has been implemented. Our experimental results show t h a t the proposed approach h a s l o w runtime overhead, good virtual memory and cache performance, and very short, evenly distributed pause times during garbage collection. Some benchmarks even revealed that the CGC not only improves the program's cache performance by more than enough to pay its own cost, but also improves the program execution performance which is competitive with the SICStus fastcode.

Incremental copying garbage collection for WAM-based Prolog systems

“…Those differences are mainly optimisations which are possible in the system of Touati and Hama because of the fixed generational policy. Chronological garbage collection, described in (Li 2000), is a generational garbage collection algorithm used in the implementation of the Logic Virtual Machine (LVM). Just like the WAM (Warren 1983;Aït-Kaci 1990), the LVM is an execution model that can be used as a basis for Prolog implementations.…”

“…mqueens was also used in Demoen et al (2002). The other benchmarks are taken from (Li 2000). They are classical benchmarks, but have extra parameters to increase the size of the benchmark.…”