kflops_util.c File Reference

#include "kflops.h"
Include dependency graph for kflops_util.c:

Go to the source code of this file.


REAL kflops_epslon (REAL x)

Detailed Description

This file contains a routine for a kflops estimator inherited from earlier ICL projects. This function epsilon is needed by kflops(), but is separated since, under advice, it should not be compiled with optimization

Definition in file kflops_util.c.

Function Documentation

REAL kflops_epslon ( REAL  x  ) 

Definition at line 19 of file kflops_util.c.

  REAL a,b,c,eps;
     this program should function properly on all systems
     satisfying the following two assumptions,
        1.  the base used in representing dfloating point
            numbers is not a power of three.
        2.  the quantity  a  in statement 10 is represented to 
            the accuracy used in dfloating point variables
            that are stored in memory.
     the statement number 10 and the go to 10 are intended to
     force optimizing compilers to generate code satisfying 
     assumption 2.
     under these assumptions, it should be true that,
            a  is not exactly equal to four-thirds,
            b  has a zero for its last bit or digit,
            c  is not exactly equal to one,
            eps  measures the separation of 1.0 from
                 the next larger dfloating point number.
     the developers of eispack would appreciate being informed
     about any systems where these assumptions do not hold.

     this routine is one of the auxiliary routines used by eispack iii
     to avoid machine dependencies.

     this version dated 4/6/83.

  a = 4.0e0/3.0e0;
  eps = ZERO;
  while (eps == ZERO) {
    b = a - ONE;
    c = b + b + b;
    eps = fabs((double)(c-ONE));

Here is the caller graph for this function: