math.pow test failure with an exponent of -25, only on MacOS ARM

[laytan]

$ odin test tests/internal/test_pow.odin -file
[Package: test_internal_math_pow]
[Test: pow_test]
/Users/laytan/third-party/Odin/tests/internal/test_pow.odin(36:4): Expected math.pow2_f16(-25) == math.pow(2, -25) (= 0000), got 0001
[pow_test : FAILURE]
----------------------------------------
0/1 SUCCESSFUL

Useful information to add to a bug report:

Odin: dev-2024-02:01640f54e
OS:   macOS Sonoma 14.2.1 (build: 23D56, kernel: 23.2.0)
CPU:  ARM64
RAM:  8192 MiB

[laytan]

This seems like the llvm intrinsic under math.pow which is llvm.pow.f16 is wrong.

It does not make sense to have a different result across architecture, x86 llvm.pow.f16 is returning 0h0001, and arm llvm.pow.f16 is returning 0h0000.

We are consistently returning 0h0001.

I believe there are plans to write our own pow implementation that would resolve this, for now we are skipping this test case (#3163).

Closing because this is an LLVM issue.