static inline v4sf
vfastlog2 (v4sf x)
{
union { v4sf f; v4si i; } vx = { x };
union { v4si i; v4sf f; } mx; mx.i = (vx.i & v4sil (0x007FFFFF)) | v4sil (0x3f000000);
v4sf y = v4si_to_v4sf (vx.i);
y *= v4sfl (1.1920928955078125e-7f);
const v4sf c_124_22551499 = v4sfl (124.22551499f);
const v4sf c_1_498030302 = v4sfl (1.498030302f);
const v4sf c_1_725877999 = v4sfl (1.72587999f);
const v4sf c_0_3520087068 = v4sfl (0.3520887068f);
return y - c_124_22551499
- c_1_498030302 * mx.f
- c_1_725877999 / (c_0_3520087068 + mx.f);
}
int
main (int argc,
char *argv[])
{
char buf[4096];
(void) argc;
float x;
for (x = -50; x > -1000; x -= 100)
{
assert (fastexp (x) >= 0);
assert (fasterexp (x) >= 0);
#ifdef __SSE2__
v4sf vx = v4sfl (x);
assert (v4sf_index (vfastexp (vx), 0) >= 0);
assert (v4sf_index (vfasterexp (vx), 0) >= 0);
#endif
}
srand48 (69);
strncpy (buf, argv[0], sizeof (buf) - 5);
strncat (buf, ".out", 5);
fclose (stderr);
stderr = fopen (buf, "w");
test_fastexp ();
test_fasterexp ();
test_vfastexp ();
test_vfasterexp ();
time_fastexp ();
time_fasterexp ();
time_vfastexp ();
time_vfasterexp ();
return 0;
}
static inline v4sf
vfastlog (v4sf x)
{
const v4sf c_0_69314718 = v4sfl (0.69314718f);
return c_0_69314718 * vfastlog2 (x);
}
