A Guidebook to Fortran on Supercomputers presents in detail both the underlying architecture of supercomputers and the manner by which a compiler maps Fortran code onto that architecture. This book outlines the constructs preventing full optimization and provides specific strategies for restructuring a program. Organized into four chapters, this book begins with an overview of the basic concepts of vector, scalar, and parallel processing. This text then provides an in-depth look at the architectural features of a variety of existing machines, with emphasis to the features common to many of them. Other chapters consider the optimization techniques used by compilers and how a programmer can take advantage of this knowledge in restructuring existing programs and in the development of new applications. The final chapter presents examples of loops from real-world programs, with a discussion of the inherent problems. This book is a valuable resource for scientists, engineers, students, and research workers.Here we have eliminated the clock-cycle count in the abscissa to highlight the two distinct portions of the computation in which operations have been ... These a chained vector timesaquot; or achimesa are the dominant feature of the timing diagram.
|Title||:||A Guidebook to Fortran on Supercomputers|
|Author||:||John M. Levesque, Joel W. Williamson|
|Publisher||:||Academic Press - 2014-05-12|