Section: New Results

An interval constrained memory allocator for the Givy GAS runtime

Participants : François Gindraud, Fabrice Rastello, Albert Cohen [ENS Ulm] , Francois Broquedis.

This work presents a memory allocator for a global address space (GAS) execution environment targeting manycore architectures with distributed memory. Among the family of Multi Processor System on Chip (MPSoC), these devices are composed of multiple nodes linked by an on-chip network; most nodes have multiple processors sharing a small local memory. An MPSoC has an excellent performance-per-Watt ratio, but it is hard to program due to multilevel parallelism, explicit resource and memory management, and hardware constraints (limited memory, network topology).

Practical programming frameworks let the programmer in charge of the hard, target-specific work (e.g., threads or node-local OpenMP plus explicit communications). Automatic, more abstract frameworks exist for specific (scientific) applications, but they target big systems and do not model the hardware constraints of MPSoC. Givy is a runtime system to execute dynamic task graphs on MPSoC. It has a focus on supporting irregular applications, and uses data-flow semantics to coordinate dynamic task scheduling and data transfer. To simplify the programmer's view of memory, both runtime and program data objects live in a GAS. To avoid address collisions when objects are dynamically allocated, and to maintain the consistency of these addresses across explicit data transfers and virtual memory remapping, a GAS-aware memory allocator is required. The allocator proposed in this work has the following properties: (1) it is free of inter-node synchronizations; (2) it is well suited for small memory systems; (3) its performances match that of existing state-of-the-art allocators.