* 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)
24 lines
513 B
C
24 lines
513 B
C
#include "libm.h"
|
|
|
|
double floor(double x)
|
|
{
|
|
union {double f; uint64_t i;} u = {x};
|
|
int e = u.i >> 52 & 0x7ff;
|
|
double_t y;
|
|
|
|
if (e >= 0x3ff+52 || x == 0)
|
|
return x;
|
|
/* y = int(x) - x, where int(x) is an integer neighbor of x */
|
|
if (u.i >> 63)
|
|
y = (double)(x - 0x1p52) + 0x1p52 - x;
|
|
else
|
|
y = (double)(x + 0x1p52) - 0x1p52 - x;
|
|
/* special case because of non-nearest rounding modes */
|
|
if (e <= 0x3ff-1) {
|
|
FORCE_EVAL(y);
|
|
return u.i >> 63 ? -1 : 0;
|
|
}
|
|
if (y > 0)
|
|
return x + y - 1;
|
|
return x + y;
|
|
}
|