* faster, smaller, cleaner implementation than the bit hacks of fdlibm * use arithmetics like y=(double)(x+0x1p52)-0x1p52, which is an integer neighbor of x in all rounding modes (0<=x<0x1p52) and only use bithacks when that's faster and smaller (for float it usually is) * the code assumes standard excess precision handling for casts * long double code supports both ld80 and ld128 * nearbyint is not changed (it is a wrapper around rint)
28 lines
446 B
C
28 lines
446 B
C
#include "libm.h"
|
|
|
|
double round(double x)
|
|
{
|
|
union {double f; uint64_t i;} u = {x};
|
|
int e = u.i >> 52 & 0x7ff;
|
|
double_t y;
|
|
|
|
if (e >= 0x3ff+52)
|
|
return x;
|
|
if (u.i >> 63)
|
|
x = -x;
|
|
if (e < 0x3ff-1) {
|
|
/* raise inexact if x!=0 */
|
|
FORCE_EVAL(x + 0x1p52);
|
|
return 0*u.f;
|
|
}
|
|
y = (double)(x + 0x1p52) - 0x1p52 - x;
|
|
if (y > 0.5)
|
|
y = y + x - 1;
|
|
else if (y <= -0.5)
|
|
y = y + x + 1;
|
|
else
|
|
y = y + x;
|
|
if (u.i >> 63)
|
|
y = -y;
|
|
return y;
|
|
}
|