Function matmul_backward_db 

pub fn matmul_backward_db<T: Triton, D: Num, const BLOCK_M: i32, const BLOCK_N: i32, const BLOCK_K: i32, const GROUP_M: i32>(
    dc_ptr: T::Pointer<D>,
    a_ptr: T::Pointer<D>,
    db_ptr: T::Pointer<D>,
    M: i32,
    N: i32,
    K: i32,
)

Backward: dB = A^T @ dC

Grid: one CTA per [BLOCK_K, BLOCK_N] tile of dB. dB[k, n] = sum_m A[m, k] * dC[m, n]