Automatic transformation of a sequential program into a parallel form is a subject that presents a great intellectual challenge and promises a great practical award. There is a tremendous investment in existing sequential programs, and scientists and engineers continue to write their application programs in sequential languages (primarily in Fortran). The demand for higher speedups increases. The job of a restructuring compiler is to discover the dependence structure and the characteristics of the given machine. Much attention has been focused on the Fortran do loop. This is where one expects to find major chunks of computation that need to be performed repeatedly for different values of the index variable. Many loop transformations have been designed over the years, and several of them can be found in any parallelizing compiler currently in use in industry or at a university research facility. The book series on KappaLoop Transformations for Restructuring Compilerskappa provides a rigorous theory of loop transformations and dependence analysis. We want to develop the transformations in a consistent mathematical framework using objects like directed graphs, matrices, and linear equations. Then, the algorithms that implement the transformations can be precisely described in terms of certain abstract mathematical algorithms. The first volume, Loop Transformations for Restructuring Compilers: The Foundations, provided the general mathematical background needed for loop transformations (including those basic mathematical algorithms), discussed data dependence, and introduced the major transformations. The current volume, Loop Parallelization, builds a detailed theory of iteration-level loop transformations based on the material developed in the previous book.

This book introduces new compilation techniques, using the polyhedron model for the resource-adaptive parallel execution of loop programs on massively parallel processor arrays. The authors show how to compute optimal symbolic assignments and parallel schedules of loop iterations at compile time, for cases where the number of available cores becomes known only at runtime. The compile/runtime symbolic parallelization approach the authors describe reduces significantly the runtime overhead, compared to dynamic or just‐in-time compilation. The new, on‐demand fault‐tolerant loop processing approach described in this book protects loop nests for parallel execution against soft errors.

Loop tiling, as one of the most important compiler optimizations, is beneficial for both parallel machines and uniprocessors with a memory hierarchy. This book explores the use of loop tiling for reducing communication cost and improving parallelism for distributed memory machines. The author provides mathematical foundations, investigates loop permutability in the framework of nonsingular loop transformations, discusses the necessary machineries required, and presents state-of-the-art results for finding communication- and time-minimal tiling choices. Throughout the book, theorems and algorithms are illustrated with numerous examples and diagrams. The techniques presented in Loop Tiling for Parallelism can be adapted to work for a cluster of workstations, and are also directly applicable to shared-memory machines once the machines are modeled as BSP (Bulk Synchronous Parallel) machines. Features and key topics: Detailed review of the mathematical foundations, including convex polyhedra and cones; Self-contained treatment of nonsingular loop transformations, code generation, and full loop permutability; Tiling loop nests by rectangles and parallelepipeds, including their mathematical definition, dependence analysis, legality test, and code generation; A complete suite of techniques for generating SPMD code for a tiled loop nest; Up-to-date results on tile size and shape selection for reducing communication and improving parallelism; End-of-chapter references for further reading. Researchers and practitioners involved in optimizing compilers and students in advanced computer architecture studies will find this a lucid and well-presented reference work with numerous citations to original sources.

The LNCS series reports state-of-the-art results in computer science research, development, and education, at a high level and in both printed and electronic form. Enjoying tight cooperation with the R&D community, with numerous individuals, as well as with prestigious organizations and societies, LNCS has grown into the most comprehensive computer science research forum available. The scope of LNCS, including its subseries LNAI and LNBI, spans the whole range of computer science and information technology including interdisciplinary topics in a variety of application fields. The type of material published traditionally includes -proceedings (published in time for the respective conference) -post-proceedings (consisting of thoroughly revised final full papers) -research monographs (which may be based on outstanding PhD work, research projects, technical reports, etc.) More recently, several color-cover sublines have been added featuring, beyond a collection of papers, various added-value components; these sublines include -tutorials (textbook-like monographs or collections of lectures given at advanced courses) -state-of-the-art surveys (offering complete and mediated coverage of a topic) -hot topics (introducing emergent topics to the broader community)

Architecture-independent programming and automatic parallelisation have long been regarded as two different means of alleviating the prohibitive costs of parallel software development. Building on recent advances in both areas, Architecture-Independent Loop Parallelisation proposes a unified approach to the parallelisation of scientific computing code. This novel approach is based on the bulk-synchronous parallel model of computation, and succeeds in automatically generating parallel code that is architecture-independent, scalable, and of analytically predictable performance.

With the rise of manycore processors, parallelism is becoming a mainstream necessity. Unfortunately, parallel programming is inherently more difficult than sequential programming; therefore, techniques for automatic parallelisation will become indispensable. This doctoral thesis aims at extending the well-known polyhedron model, which promises this automation, beyond some of its current restrictions. Up to now, loop bounds and array subscripts in the modelled codes must be expressions linear in both the variables and the parameters. This restriction is lifted to allow certain polynomial expressions instead of linear ones. With these extensions, more programs can be handled in dependence analysis, in the transformation of the program model and in code generation.

We make a detailed examination of the performance achieved by a Krylov space sparse linear system solver that uses incompletely factored matrices for preconditioners. We compared two related mechanisms for parallelizing the computationally critical sparse triangular solves and sparse numeric incomplete factorizations on a range of test problems. From these comparisons we drew several interesting conclusions about methods that can be used to parallelize loops of the type found here. The performance we obtain is brought into perspective by comparison with timing results from a Cray X/MP supercomputer. Performance on an Encore Multimax/320 with relatively modest computational capabilities comes within a small factor of the performance on a comparable code run on a Cray X/MP. (KR).