ENH: Bitwise reproducible `np.log`

[zmbc]

Proposed new feature or change:

I was quite surprised to not find a way to make the following result bitwise reproducible:

>> import np
>> np.log(0.04048147466152369)
-3.206910825280994

I got the above result on a Intel(R) Xeon(R) Gold 6130 CPU @ 2.10GHz; on a Intel(R) Xeon(R) CPU E5-2660 v4 @ 2.00GHz I get:

>> import np
>> np.log(0.04048147466152369)
-3.2069108252809944

I understand that this is due to floating-point imprecision; I am not asking for the result to be more precise, but just to have a way to make it imprecise in the same way across CPUs. For even very complicated BLAS operations I can do this with MKL's CNR, but I have yet to find a way to do so in this case.

The simple workaround of manually defining an approximation of e seems to work, so it seems like this should be possible to support in Numpy.

[zmbc]

Ah, this appears to be a duplicate of #23523.