pub fn compare_m16_tc_block(
actual: &[u8],
reference: &[u8],
n: usize,
k: usize,
) -> M16TcDiffExpand description
Compare an m x n BF16 block against the scalar reference under the tier’s
contract. Both slices are m * n little-endian BF16 elements; k is the
reduction depth the kernel ran.
Two passes, because the absolute floor is expressed against the reference row RMS and no row’s RMS is known until it has been walked once.