Optimizations for Automatic Extraction of Parallelism

概要

論文の詳細を見る
In scientific research world, massively parallel computers are considered to be one of the most promising answers to the remarkable demand increase in computational power. The continuos improvements in hardware technologies had made possible to produce computers based on articulated and complex architecture, indeed capable of a high system performance on a potential level. However, a frontier challenge is to effectively exploit available resources in order to bridge the gap between actual and achievable performance. The process of efficient parallel software development becomes a critical factor, being extremely error-prone because of the inherent complexity of the programming environment. Parallelizing compilers for imperative languages aim at automatically discovering as much parallelism as possible from sequential source code, and at mapping the original program to a modified parallel one with improved execution performance. Reconstruction techniques have been extensively developed for vector computers, but in the field of massively parallel machines relevant questions still need analysis.
一般社団法人情報処理学会の論文
1997-09-24