B"You want to train a neural network model for your graduation work. There are n images in the dataset, the i -th image's size is a_i bytes. You don't have any powerful remote servers to train this model so you have to do it on your local machine. But there is a problem: the total size of the dataset is too big for your machine, so you decided to remove some images -- though you don't want to make the dataset too weak so you can remove no more than k images from it. Note that you can only remove images, you can't change their order. You want to remove these images optimally so you came up with a metric (you're a data scientist after all) that allows to measure the result of removals. Consider the array b_1, b_2, ldots, b_m after removing at most k images ( n - k <= m <= n ). The data from this array will be uploaded to the machine in blocks of x consecutive elements each. More precisely: There will be cnt = <= ft lceil frac{m}{x} right rceil blocks in total. Note that if m is not divisible by x then the last block contains less than x elements, and it's okay. Let w(i) be the total size of the i -th block -- that is, the sum of sizes of images inside this block. For example, the size of the first block w(1) is b_1 + b_2 + ldots + b_x , the size of the second block w(2) is b_{x + 1} + b_{x + 2} + ldots + b_{2x} . The value of the metric you came up with is the maximum block size over the blocks of the resulting dataset. In other words, the value of the metric is max limits_{i=1}^{cnt} w(i) . You don't want to overload your machine too much, so you have to remove at most k images in a way that minimizes the value of the metric described above. The first line of the input contains three integers n , k and x ( 1 <= n <= 10^5 ; 1 <= k, x <= n ) -- the number of images in the dataset, the maximum number of images yo"...