<cmath>: Use intrinsics where possible

[sylveon]

Partial fix for #1234. This uses intrinsics where possible to enhance codegen. Due to the STL relying on the UCRT for all functions here, this is the best we can do. These changes only affect users of the float and long double overloads.

[fsb4000]

Starting since this PR it is better to use cmath and std::meow instead of math.h and meow, isnt it?

[sylveon]

Right, but only for float and long double now.

Can bring the same to double (and std::signbit) by manually writing the overload rather than importing it from the UCRT into the std namespace, and same goes for std::meowf and std::meowl.

I'll look into doing that.

Doing the same for the int overloads might be trickier because of the macro machinery to generate them.

[sylveon]

Can't do that because any solution would break overload resolution for consumers with using namespace std;

I can still do something about the integer overloads however.

[fsb4000]

By the way from C standard:

https://web.archive.org/web/20181230041359if_/http://www.open-std.org/jtc1/sc22/wg14/www/abq/c17_updated_proposed_fdis.pdf

paragraph 7.1.4:

Any function declared in a header may be additionally implemented as a function-like macro defined in the header

and below they ever wrote example: https://imgur.com/a/kqzW2dk

So UCRT team also can do:

#define meow(x) __meow(x)

[StephanTLavavej]

Thanks for this optimization - I will explore/validate changes for Clang and hopefully push something soon.

[StephanTLavavej]

@CaseyCarter FYI, I pushed a /clr:pure fix (verified with an MSVC-internal build) after you approved.

[StephanTLavavej]

@CaseyCarter /clr needed the same fix (this is surprising and may indicate a compiler bug but I haven't reported it).

[StephanTLavavej]

Thanks again for this codegen improvement! 😺