Range of `xexpf` to clamp to 0 and inf in scalar vs simd

[aconz2]

I noticed that the scalar and simd xexpf versions seem to use slightly different constants to clamp to 0 or inf. In scalar, it is -104 and 104, but simd uses -104 and 100. Is this intentional?

Scalar

sleef/src/libm/sleefsp.c

Lines 1174 to 1175 in 5533114

	if (d < -104) u = 0;
	if (d > 104) u = SLEEF_INFINITYf;

SIMD

sleef/src/libm/sleefsimdsp.c

Lines 2047 to 2048 in 5533114

	u = vreinterpret_vf_vm(vandnot_vm_vo32_vm(vlt_vo_vf_vf(d, vcast_vf_f(-104)), vreinterpret_vm_vf(u)));
	u = vsel_vf_vo_vf_vf(vlt_vo_vf_vf(vcast_vf_f(100), d), vcast_vf_f(SLEEF_INFINITYf), u);

It isn't really making a difference to me as I can guarantee the inputs will be inside that range, so mainly a curiosity question. Thanks for making this project, it is really nice.

[shibatch]

Thank you for pointing out.
I will fix it.

[shibatch]

This does not need to be fixed, because the value of expf becomes infinity at around 88.8.
This clamping is only for preventing the results of internal computation from overflowing.