Optimistic parallelism benefits from data partitioning

Abstract

Recent studies of irregular applications such as finite-element mesh generators and data-clustering codes have shown that these applications have a generalized data parallelism arising from the use of iterative algorithms that perform computations on elements of worklists. In some irregular applications, the computations on different elements are independent. In other applications, there may be complex patterns of dependences between these computations.The Galois system was designed to exploit this kind of irregular data parallelism on multicore processors. Its main features are (i) two kinds of set iterators for expressing worklist-based data parallelism, and (ii) a runtime system that performs optimistic parallelization of these iterators, detecting conflicts and rolling back computations as needed. Detection of conflicts and rolling back iterations requires information from class implementors.In this paper, we introduce mechanisms to improve the execution efficiency of Galois programs: data partitioning, data-centric work assignment, lock coarsening, and over-decomposition. These mechanisms can be used to exploit locality of reference, reduce mis-speculation, and lower synchronization overhead. We also argue that the design of the Galois system permits these mechanisms to be used with relatively little modification to the user code. Finally, we present experimental results that demonstrate the utility of these mechanisms.

Exploiting inter-thread temporal locality for chip multithreading

“…For tasks with regular access patterns, Tiling or Blocking can be employed to partition the workload into consecutive chunks that each map to a thread [10], [20], [22], [23], [27], [29], [31], [36], [42], [45] When data accesses are irregular, the underlying system can still reorder independent sections of code according to userinstrumented address hints that indicate data locality among code sections [41] or runtime discovery.…”

“…Therefore, the number of tiles by default equals the number of cores in S-tile. However, in the case of those applications that may have unbalanced workloads among tiles, programmers can overdecompose the parallel loop into smaller tiles with a larger quantity than the number of cores [31]. Load-balancing is then achieved by having idle cores fetch remaining tiles.…”

On the Scalability of an Automatically Parallelized Irregular Application

“…Alternatively, conflict detection can be performed on partitioned data structures [9]. For example, a Delaunay mesh can easily be partitioned.…”

“…When a thread accesses a partition of a data structure, it owns all elements in that partition, and the other threads are not allowed to access them. Assigning multiple partitions to a core increases the probability that a thread can continue to perform useful work even if other threads have temporarily locked some of its partitions [9].…”

“…While we believe our system to be practical, our previous studies [8], [9], [10] have not investigated the scalability beyond small multicore systems. The goal of this paper is to study the performance of an application that has been automatically parallelized by the Galois system on a large-scale shared-memory multiprocessor.…”