UIImage resizableImageWithCapInsets official document description are below.
During scaling or resizing of the image, areas covered by a cap are not scaled or resized. Instead, the pixel area not covered by the cap in each direction is tiled, left-to-right and top-to-bottom, to resize the image. This technique is often used to create variable-width buttons, which retain the same rounded corners but whose center region grows or shrinks as needed. For best performance, use a tiled area that is a 1x1 pixel area in size.
I don't understand why use 1x1 pixel tiled area is the best performance. I think tiled block by block, the performance is better than 1x1 area. In theory, block by block is fast than point by point, is that right? who can told me the implementation of this in machine?