Scientific Computing on Vector Computers
The goal of this volume is to gradually guide the reader from his usual base of general purpose computer knowledge to the highly specialized knowledge necessary for the efficient use of vector computers. The basic rules for the selection of optimal data structures and algorithms for vector computers are presented. The properties of the hardware and software of the following vector computers are discussed in the context of measurements: CRAY-1, CRAY X-MP, CRAY-2, CYBER 205, ETA 10, Fujutsu VP 200, IBM VF, and CONVEX C1. The FIDISOL program package, developed by the author's research group, is presented as an example of the full vectorization. The advantages and the deficiencies of the most relevant vector computers are stressed. Related questions of a large general purpose software package for vector computers are also discussed.
What people are saying - Write a review
We haven't found any reviews in the usual places.
THE FIRST COMMERCIALLY SIGNIFICANT VECTOR COMPUTERS
THE ARITHMETIC PERFORMANCE OF THE FIRST COMMERCIALLY
16 other sections not shown
algorithm apply arithmetic assume bank basic better cache coefficients column compiler consider contiguous corresponding CRAY X-MP CYBER cycle data structure diagonal difference direct discuss efficient elements elimination equations error example executed expected formulae Fortran grid hardware holds Illustration increase innermost instruction iteration linear system linked triad load loop machine main memory matrix matrix multiplication means measurements mentioned method MFLOPS MFLOPS-rate Mwords Note nsec obtain operands operations optimal parallel PDE's peak performance performance pipes pivot possible presented problem processing processors Progr purpose computer question reason reduction relation result scalar selected similar simple single solution solve speed step storage stored supercomputer Table theoretical transfer unit unknowns usually values vector computer vector length vector operations vectorizable whole