Using some methods of coding to improve the accuracy of calculations

Lutsky G.M., Blinov T.A., Poryev V.M.

At realization in the computer systems of operations, especially vector operations, on floating point numbers the questions of exactness are very important. The work algorithm of floating point number computing device, which oriented to performance of add sequence of paired products, is described. For providing enhanceable exactness the intermediate arithmetic results represent in system with scaled-up basis and with digits are represented by binary code. It enables to get the calculation results which contain, in general case, only error which arises up through mantissa represent of a finite number of digits.


Full text (pdf)