Algorithms and software for increasing of the calculation speed on personal computers

Acceleration of time-consuming calculations is discussed.

Saved in:
Bibliographic Details
Published in:Вопросы атомной науки и техники
Date:2001
Main Authors: Shapoval, I.N., Krivorukov, E.V.
Format: Article
Language:English
Published: Національний науковий центр «Харківський фізико-технічний інститут» НАН України 2001
Subjects:
Online Access:https://nasplib.isofts.kiev.ua/handle/123456789/78452
Tags: Add Tag
No Tags, Be the first to tag this record!
Journal Title:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Cite this:Algorithms and software for increasing of the calculation speed on personal computers / I.N. Shapoval, E.V. Krivorukov // Вопросы атомной науки и техники. — 2001. — № 1. — С. 71-72. — Бібліогр.: 5 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
_version_ 1859914049252229120
author Shapoval, I.N.
Krivorukov, E.V.
author_facet Shapoval, I.N.
Krivorukov, E.V.
citation_txt Algorithms and software for increasing of the calculation speed on personal computers / I.N. Shapoval, E.V. Krivorukov // Вопросы атомной науки и техники. — 2001. — № 1. — С. 71-72. — Бібліогр.: 5 назв. — англ.
collection DSpace DC
container_title Вопросы атомной науки и техники
description Acceleration of time-consuming calculations is discussed.
first_indexed 2025-12-07T16:04:11Z
format Article
fulltext ALGORITHMS AND SOFTWARE FOR INCREASING OF THE CALCULATION SPEED ON PERSONAL COMPUTERS I.N. Shapoval, E.V. Krivorukov National Science Center “Kharkov Institute of Physics and Technology”, Kharkov, Ukraine Acceleration of time-consuming calculations is discussed. PACS: 61.20.-Ja. 1. INTRODUCTION When improvement of elemental base productivity began approaching to physical limits and a cost factor commenced to play a significant role the multiprocessor systems have started to develop by quick pace. But the world experience shows that at all steps of computing technique development also an important way of reduc- ing the time of obtaining calculation results is a choice of optimal decision algorithms and their programmed realization. Among these are the table-algorithmic methods (TAM) of function calculation being widely applied and at the same time the most effective. In the present paper considered are the results of ap- plication of TAM for increasing the speed of calcula- tions the essence of which consists in the use of a high- er-speed function approximation by reference values as compared to direct calculations. The results known in this field [1-5] were obtained mainly for universal com- puters processing a large memory. However, in connec- tion with significant increasing the operational memory capacity (as a rule exceeding 32 MB) in personal com- puters of latter generations and due to vigorous widen- ing a sphere of their application, of a great interest is the development and use of TAM for increasing the speed of calculations on PC depending on the features of their technological and system architecture. Let us consider briefly the problem statement (in more details this is given in authors’ paper [4]). As is well-known, a direct calculation of a wide class of functions in the vicinity of the point x0∈ [a, b] can be replaced by executing the approximation of a function by its reference values, e.g. by polynomials of a special form or by truncated series expansion of this function. The calculation of the function f(x) is realized by a set of operations {/, *, ±, sampling from the memo- ry} arranged in order of decreasing times of execution t1>t2>...≥/ t4 corresponding to the operation. Thus, the calculation time f(x) tf= Σ ti ni for the selected calcula- tion environment is determined by the set nf= {n1,n2,...,n4}, where n1 equals to the number of division operations, n2 - to the multiplication operations etc. Therefore, decrease of the calculation time can be achieved by reducing either all or first components at the expense of latter ones of the nf set. When selecting the algorithm of high-speed function calculation, the optimization by the nf set in propagated TAM realizations leads to such significant increase in volumes of function reference tables (RT), and so to in- crease in time expenditures for search of adequate func- tion nods in RT, that this can result in an effect being di- rectly contrary to the expected speed increasing. Tend- ing to obtain an acceptable accuracy of approximations without increasing table volumes in turn increases the time expenditures being reciprocal to the expected gain. So, for example, it is well known that the use of the best Chebyshev approximation provides the best uniform ap- proximation among polynomials of the set exponent. However, then a significant time is spent for search of alternance points and for calculation of polynomial co- efficients. Thus, we have before us the problem of find- ing a TAM realization with which the effect of increas- ing the speed of a directly approximating calculation does not lead to the competitive increase of time expen- ditures to support this approximation. In this respect adaptive aspects of RT applying emerge as a decisive factor. Let us introduce the set n*={n1 *,n2 *,...,n5 *}, represent- ing the function approximation operations by the opti- mally selected series : certainly, this set and the time of its realization t*=Σ ti ni * should not be dependent on the choice of a function. Let tт f is the time necessary for construction by the reference points df of the RT func- tion f(x) , which should be constructed and used taking in mind assurance of the acceptable accuracy of calcula- tion approximation. To estimate the effectiveness of TAM realizations developed we determine the coefficients Ka- of reducing the time of proper calculation and Kf - of reducing the time of full calculation. Then for a calculation session in which the function is calculated Mf times, Mf >> df , the coefficients introduced take the form t*)) d - M( t /( tM K/ t*, t K ff f mfff fa +== (1) And with fixed tf, tт f, df, t*, it is evident aK→fK (2) together with the increase of Mf . Such a situation takes place, of course, in calculation sessions where the num- ber of function calls is accompanied by the essential lo- calization of the function argument values. A specific feature of universal computers designed for a wide range of problems has in consequence forma- tion of universal software. Therefore, the programmed realization of TAM on a universal computer inevitably increases RT dimensions while embracing classes of problems or extending the range of function determina- tion. This observation is confirmed also by TAM real- ization for set functions that have been performed in leading computer firms. In [4] one describes the realiza- tion allowing obtaining an appreciable acceleration of calculation. The mentioned above realization means in PROBLEMS OF ATOMIC SCIENCE AND TECHNOLOGY. 2001, № 1. Series: Nuclear Physics Investigations (37), p. 71-72. 71 essence a personalization of calculations on universal computers. In connection with an extensive application of PC and increase of their memory resource it is seem natu- rally to transfer and develop further TAM on PC. The present paper is aimed at presenting such a direction of calculation speed increase. 2. ADAPTIVE TAM REALIZATIONS The calculation speed increase is based, according to formula (1), on reducing calculation for a wide class of functions to the standard approximation of functions over its self-stored reference values, so the effect of cal- culation time reducing within large sessions localized by calculation argument values approaches to the coeffi- cient Ka in correspondence with (2). In this case the speed of this approximation is determined only by the character of RT construction. The table gives the estimations for Ka obtained in a numerical experiment on PC being analogous to coeffi- cients of calculation time reducing on universal comput- ers obtained earlier. Their realization in С++ language, under conditions when a user has at its disposal practi- cally unlimited resources of the operative memory and PC on the whole, of course, allows one to solve optimal- ly the problem of direct approximation of calculations executed in short time tт f. Function Ka x16/(1-x8) 7. Integral Sine 4. Gamma- function 2.5 Complete elliptic integral of the first kind 6.5 Generalized complete elliptic integral 8.0 Bessel function 70. An important constituent part of TAM for increasing the speed of calculation are the operations with RT: their intrinsic organization, technique of finding nods in RT being the most suitable for approximation, ways of completing RT with new nodal values of a given func- tion. As effective are the following mode of adaptive TAM realizations: −table with intervals (is constructed immediately after as-determined interval), −discrete RT (RT is completed with as-calculated nods, −RT with one reference point (when the function argu- ment changes monotonically). Each of above listed modes of RT use has its own domain of preference determined by the function form and character of distributions required in the course of calculating the function argument values. If in the calculation mix dominating in time are the parts being accelerated, then the total acceleration in the limit of large sessions, approaching to Ka, can be signifi- cant. The degree of approximation depends, as is shown in [4], on the spatial organization of RT, the order of ap- pearance of values in the calculation session completing RT, and on the calculation volume. A representative bank of results of TAM application for acceleration of calculations and effective algorithms of their realization are formed during development and generalization of experience in their service. 3. DEVELOPMENT OF TAM On the base of tested algorithms used for increasing the speed of calculation of one-variable functions the following methods should be developed: −direct methods for table-algorithmic accelerating the calculation of several-variables functions; −methods of dimensionality model reduction for the available functional dependence or for its required do- main of argument changes, and development, on the base of these methods, of TAM for increasing the speed of calculation; −construction of space filling curves (Peano curves) for many-dimensional domains of function determining. Reducing of TAM for increasing the speed of calcula- tion of many- variable functions to the table-algorithmic accelerating the calculation of one-variable function via superposition of the starting function and Peano map- ping; −TAM for increasing the speed of calculating the «by measure» functions i.e. approximating the given func- tion with an acceptable accuracy, excepting a set of ar- gument values with the measure of a set small value in which the function calculated directly and its “TAM equivalent” can differ considerably. 4. CONCLUSIONS The necessity of application and development in the PC environment of executed TAM with an experience of using them for universal computers is substantiated. On the base of the expected acceleration it is shown that one should hope for considerable acceleration of time- consuming calculations when localizing domains of their determination. REFERENCES 1. B.A. Popov, G.S.Tesler Approximation of functions for technological applications- Kiev: “Naukova dumka” (in Russian). 2. A.B. Vasil’ev at al. Table-algorithmic method for function calculation for universal microcomputers. L.: Publ. H. of Inst. of precise mechanics and optics, 1985, v. 3, p. 621 (in Russian). 3. V.I. Potapov, A.I. Florensov Table−algorithmic cal- culation of functions on computers. Irkutsk: Publ. House, 1985, p. 107 (in Russian). 4. E.V. Krivorukov, I.N. Shapoval. Programmed real- ization of table-algorithmic acceleration of function calculation. Programming. 1989, №3, p.78-87 (R). 5. E.V. Krivorukov, I.N. Shapoval. Development of al- gorithmic methods of increasing the speed of calcu- lations // VANT. Ser. Yad. fiz. issledovaniya, №1 (33), 1999, p. 123-124 (in Russian). 72 I.N. Shapoval, E.V. Krivorukov National Science Center “Kharkov Institute of Physics and Technology”, Kharkov, Ukraine Acceleration of time-consuming calculations is discussed. 1. INTRODUCTION 2.  ADAPTIVE TAM REALIZATIONS The necessity of application and development in the PC environment of executed TAM with an experience of using them for universal computers is substantiated. On the base of the expected acceleration it is shown that one should hope for considerable acceleration of time-consuming calculations when localizing domains of their determination. REFERENCES
id nasplib_isofts_kiev_ua-123456789-78452
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
issn 1562-6016
language English
last_indexed 2025-12-07T16:04:11Z
publishDate 2001
publisher Національний науковий центр «Харківський фізико-технічний інститут» НАН України
record_format dspace
spelling Shapoval, I.N.
Krivorukov, E.V.
2015-03-17T12:16:25Z
2015-03-17T12:16:25Z
2001
Algorithms and software for increasing of the calculation speed on personal computers / I.N. Shapoval, E.V. Krivorukov // Вопросы атомной науки и техники. — 2001. — № 1. — С. 71-72. — Бібліогр.: 5 назв. — англ.
1562-6016
PACS: 61.20.-Ja.
https://nasplib.isofts.kiev.ua/handle/123456789/78452
Acceleration of time-consuming calculations is discussed.
en
Національний науковий центр «Харківський фізико-технічний інститут» НАН України
Вопросы атомной науки и техники
Experimental methods and computations
Algorithms and software for increasing of the calculation speed on personal computers
Алгоритмы и программное обеспечение ускорения вычислений на ПК
Article
published earlier
spellingShingle Algorithms and software for increasing of the calculation speed on personal computers
Shapoval, I.N.
Krivorukov, E.V.
Experimental methods and computations
title Algorithms and software for increasing of the calculation speed on personal computers
title_alt Алгоритмы и программное обеспечение ускорения вычислений на ПК
title_full Algorithms and software for increasing of the calculation speed on personal computers
title_fullStr Algorithms and software for increasing of the calculation speed on personal computers
title_full_unstemmed Algorithms and software for increasing of the calculation speed on personal computers
title_short Algorithms and software for increasing of the calculation speed on personal computers
title_sort algorithms and software for increasing of the calculation speed on personal computers
topic Experimental methods and computations
topic_facet Experimental methods and computations
url https://nasplib.isofts.kiev.ua/handle/123456789/78452
work_keys_str_mv AT shapovalin algorithmsandsoftwareforincreasingofthecalculationspeedonpersonalcomputers
AT krivorukovev algorithmsandsoftwareforincreasingofthecalculationspeedonpersonalcomputers
AT shapovalin algoritmyiprogrammnoeobespečenieuskoreniâvyčisleniinapk
AT krivorukovev algoritmyiprogrammnoeobespečenieuskoreniâvyčisleniinapk