



Volume 30, Number 2, Page 126 (1986) Vector Processing 

Full article:
PDF
 

Copyright info 






New scalar and vector elementary functions for the IBM System/370 

by R. C. Agarwal, J. W. Cooley, F. G. Gustavson, J. B. Shearer, G. Slishman, B. Tuckerman





Algorithms have been developed to compute short and longprecision elementary functions: SIN, COS, TAN, COTAN, LOG, LOG10, EXP, POWER, SQRT, ATAN, ASIN, ACOS, ATAN2, and CABS, in scalar (28 functions) and vector (22 functions) mode. They have been implemented as part of the new VS FORTRAN library recently announced along with the IBM 3090 Vector Facility. These algorithms are essentially tablebased algorithms. An important feature of these algorithms is that they produce bitwiseidentical results on scalar and vector System/370 machines. Of these, for five functions the computed value result is always the correctly rounded value of the infiniteprecision result. For the rest of the functions, the value returned is one of the two floatingpoint neighbors bordering the infiniteprecision result, which implies exact results if they are machinerepresentable. For the five correctly rounded elementary functions, scalar and vector algorithms are designed independently so as to optimize performance in each case. For other functions, the bitwiseidentical constraint leads to algorithms which compromise between scalar and vector performance. We have been able to design algorithms where this compromise is minimal and thus achieve very good performance on both scalar and vector implementations. For our test measurements on highend System/370 machines, our scalar functions are always faster (sometimes by as much as a factor of 2.5) as compared to the old VS FORTRAN library. Our vector functions are usually two or three times faster than our scalar functions. 


Related Subjects: Algorithms; Computational methods; Largescale computing; Mathematical functions and techniques; Vector processing 


