| 主权项 |
1. An image processing method of a graphic processing unit (GPU) including a plurality of fragment processors, the method comprising:
reading geometrically processed tile data from an external memory disposed outside of the GPU, and connected to the GPU via a system bus; determining a sequence of a plurality of tiles to be processed by at least one of the plurality of fragment processors, based on a locality of each tile among the plurality of tiles; selecting a tile to be processed, based on the determined sequence; checking a plurality of fragment processors inside the GPU, in response to determining any fragment processor being in an idle state, wherein the checking continues in response to a fragment processor being determined not to be in an idle state, and wherein the checking is delayed for a predetermined time before being performed again in response to the fragment processor being determined to be in the idle state; allocating the tile to be processed to an idle fragment processor among the plurality of fragment processors, based on the determined sequence; and rendering, by the fragment processor to which the tile is allocated, the allocated tile, wherein each of the plurality of fragment processors is configured to perform rendering of a tile by using a respective internal cache, and wherein each of the plurality of fragment processors is configured to access, by using a shared cache, data for performing rendering with respect to allocated tiles. |
