pub fn elemwise_add_backward<T: Triton, D: Num, const BLOCK_SIZE: i32>(
dy_ptr: T::Pointer<D>,
grad_a_ptr: T::Pointer<D>,
grad_b_ptr: T::Pointer<D>,
n_elements: i32,
)where
T::I32Tensor: Tensor<i32, 1> + Comparison<i32, BoolTensor = T::BoolTensor>,
T::Pointer<D>: AddOffsets<i32, 1, T::I32Tensor, Output = T::Tensor<T::Pointer<D>>>,