Constant DENSE_GEMM_M16_BF16_N_TILE 

pub const DENSE_GEMM_M16_BF16_N_TILE: u32 = 32;

N columns one CTA owns on the DEFAULT instantiation. SSOT for the launch geometry AND for the dispatch rule’s CTA-count reasoning, so the two cannot drift: the kernel’s DGM16_N_TILE must equal this.