get_partial_tanh_val_sum Subroutine

  pure subroutine get_partial_tanh_val_sum(this, upstream_grad, output)
    !! Fused partial+sum for tanh: output(:,1) = sum_s(upstream * (1 - val^2))
    implicit none
    class(array_type), intent(in) :: this
    real(real32), dimension(:,:), intent(in) :: upstream_grad
    real(real32), dimension(:), intent(out) :: output

    integer :: s, n_samples

    n_samples = size(upstream_grad, 2)
    output(:) = upstream_grad(:,1) * &
         (1._real32 - this%val(:,1) * this%val(:,1))
    do s = 2, n_samples
       output(:) = output(:) + upstream_grad(:,s) * &
            (1._real32 - this%val(:,s) * this%val(:,s))
    end do
  end subroutine get_partial_tanh_val_sum