FP32 log1pf_u10 and FP64 log1p_u10 are infinitely inaccurate for large inputs

[ebavier]

On x86 targets and possibly others, the log1pf_u10 and log1p_u10 functions are infinitely inaccurate for large input values. Currently

• log1pf_u10 (x > 1e38) = inf, and
• log1p_u10 (x > 1e307) = inf

No input value except x = inf should result in an inf result for log1p.

Relevant source:

sleef/src/libm/sleefsp.c

Line 1761 in 85440a5

if (d > 1e+38) r = SLEEF_INFINITYf;

sleef/src/libm/sleefdp.c

Line 2179 in 85440a5

if (d > 1e+307) r = SLEEF_INFINITY;