An Automatic Fusion Mechanism for Variable-Length List Skeletons in SkeTo

Abstract

Skeletal parallel programming is a promising approach to easy parallel programming in which users can build parallel programs simply by combining parts of a given set of ready-made parallel computation patterns called skeletons. There is a trade-off for this easiness in the form of an efficiency problem caused by the compositional style of the programming. One solution to this problem is fusion transformation that optimizes naively composed skeleton programs by eliminating redundant intermediate data structures. Several parallel skeleton libraries have automatic fusion mechanisms. However, there have been no automatic fusion mechanisms proposed for variable-length list (VLL) skeletons, even though such skeletons are useful for practical problems. The main difficulty is that previous fusion mechanisms are not applicable to VLL skeletons, and so the fusion cannot be completed. In this paper, we propose a novel fusion mechanism for VLL skeletons that can achieve both an easy programming interface and complete fusion. The proposed mechanism has been implemented in our skeleton library, SkeTo, by using the expression templates technique, experimental results have shown that it is very effective.

New List Skeletons for the Python Skeleton Library

“…In the first phase (lines 5-9), it is necessary to permute whole partitions of the current parallel list. In the second phase (lines [10][11][12][13][14], it is necessary to permute elements only in the same local partition. As PySke provides the same skeletons for both parallel lists and sequential lists, and that in a parallel list the local partitions are sequential lists, both phases are actually very similar.…”

“…PySke is related to SkeTo [13], [14], OSL [15], [16], and Muesli [17], [18], [19]. All these libraries are C++ libraries.…”

“…get_partition makes the distribution of the lists visible in the structure itself. For example, get_partition on the list of type PList in Figure 2 yields the PList [[0, 2, 4], [6,8], [10,12], [14,16]] (global view). Now each processor contains only one element (local size 1), but this element is a list.…”

“…Finally these intervals are used to get the slices of the local contents of the parallel list to send to other processors. After a call to the MPI Global intervals to send [(0, 4), (5,9), ∅, ∅] [∅, ∅, ∅, ∅] [∅, (10,14), (15,19), ∅] [∅, ∅, (20, 24), (25, 34)]…”

Dividing Huge XML Trees Using the m-bridge Technique over One-to-one Corresponding Binary Trees

“…Since the concrete order of computation is hidden behind the parallel tree skeletons, users can perform parallel computation just by passing the operators used in the computation. The authors developed a parallel skeleton library called SkeTo [12], [22], and it included efficient parallel tree skeletons for binary trees [20] and even for general trees [23]. In particular, we [23] implemented parallel general-tree skeletons based on a left-child right-sibling binary-tree representation and parallel binary-tree skeletons.…”