Function _mm512_maskz_max_ph

#[target_feature(enable = "avx512fp16")]
pub fn _mm512_maskz_max_ph(k: __mmask32, a: __m512h, b: __m512h) -> __m512h

Compare packed half-precision (16-bit) floating-point elements in a and b, and store packed maximum values in dst using zeromask k (elements are zeroed out when the corresponding mask bit is not set). Does not follow the IEEE Standard for Floating-Point Arithmetic (IEEE 754) maximum value when inputs are NaN or signed-zero values.

Intel's documentation