Unified Execution Mode in a GPU-Style Softcore

A GPU-style processor has large amount of processing power on a given die compared to a general purpose processor. However, a Graphic Processing Unit must be executed in lock-step where a group of cores execute the same instruction. This constraint puts a real limitation on programming of a GPU. This work proposed a design of a processor that unifies the execution of Graphic Processing Units and a general purpose processor. The discussion of programming model of vectorised instructions and the extension to allow multi-cores to run independently is presented. The proposed design required only 3% additional resource compared to the original design. This design is suitable for embedded applications.