Methods and means of parallel high-precision arithmetics on the base of vertical processing systems

Parallel computing systems, getting over rounding-off errors by means of the use of the multiply precision arithmetics are investigated in the paper aiming at the creation of superprecision parallel arithmetics methods and means on the base of associative parallel processors. As a result general principles of the design of computing systems, getting over rounding-off errors with the help of the multiply precision arithmetics and dynamic lenght of words have been developed as well as the description language and the obtaining methodology of refined estimations of the algorithm execution time for vertical processing systems. Algorithms of the superprecision arithmetics have been also developed. The parallel programming system of superprecision parallel arithmetic calculations on the base of ES-2720 for the "Sibir" system has been realized. The parallel programming methods of these calculations and parallel algorithms and programs for some standerd problems have been developed. The system created is the component of the "Sibir" high-production computing system (Russian Academy of Sciences, Siberian Department, Computer Center, Novosibirsk)Available from VNTIC / VNTIC - Scientific & Technical Information Centre of RussiaSIGLERURussian Federatio