Bug#743876: ITP: llvmmath - LLVM math library
>> It is based on NumPy's umath and tries to support all floating point
and complex types.
with my numpy dev hat on, numpys umath non-c99 fallbacks have some issues.
It only receives limited testing by windows users, everybody else has
access to the (at least for real numbers) decent libm functions.
Only recently (v1.8.1) corner case bugs in hypot, expm1 and log1p were
found. Those are pretty fundamental functions. From a quick glance the
bugs are not fixed in llvmmath.
Does this package also fall back to libm if available?
If not can it be adapted to do so?
Reply to: