PERF: replace use of isnan once MSVC bug is fixed

[chris-b1]

ref #23182, #21813

MSVC bug tracker - https://developercommunity.visualstudio.com/content/problem/294290/incorrect-codegen-for-double-equality.html

Once that bug is fixed in a widely available MSVC for python 3 builds we should redefine isnan / notnan to be based on an equality check here.
https://github.com/pandas-dev/pandas/pull/23182/files#diff-467e9a847ab859a085518bc17711fe57R20

Current code works around that MSVC bug by using the libc, isnan, which works, but can't be inlined, so likely has a little perf cost
https://godbolt.org/z/y550lB

[jbrockmendel]

@chris-b1 any idea if this has been addressed upstream?

[chris-b1]

That bug has been fixed in the most recent version of MSVC - I'm a little disconnected form our build process to know if that's what were using.

[jbrockmendel]

most recent version of MSVC

Do you know what that version is? In _libs/src/cmath there is a comment "Place upper bound on this check once a fixed MSVC is released"

[TomAugspurger]

VS2019. It's mentioned in https://developercommunity.visualstudio.com/solutions/477668/view.html.

…

On Fri, Nov 8, 2019 at 1:05 PM jbrockmendel ***@***.***> wrote: most recent version of MSVC Do you know what that version is? In _libs/src/cmath there is a comment "Place upper bound on this check once a fixed MSVC is released" — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#23209?email_source=notifications&email_token=AAKAOIVNVAHY7ADN257USPDQSW2ALA5CNFSM4F4T6LE2YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEDTB36Y#issuecomment-551951867>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAKAOIVTTWXYZRM247FMW5DQSW2ALANCNFSM4F4T6LEQ> .

[jbrockmendel]

do we have a way of checking if this is actionable?