Possible edge case issue with asinpi(half)

I am seeing this test_bruteforce failure on my implementation:

...
math_edge_cases...
float
half
FAIL: asinpi(0x4000) - expected NaN, got 0x0000
FAIL: asinpi(0xc000) - expected NaN, got 0x8000
double
math_edge_cases FAILED
...

But with a different test program I get:
0.00 ulp at (0x1.000p+1): *nan vs nan

I suspect an issue with this new bruteforce edge case test.

Start with the test_bruteforce failure in math_edge_cases for the half-precision asinpi cases 0x4000 and 0xc000, then compare it with the separate test program's NaN result. Determine whether the edge-case test or the implementation produces the inconsistent expectation; done means the test and implementation agree for these cases.

Written by the indexing model from the issue text.