Abstract: Channelization processing is the first task in modern electronic warfare digital systems. CPU-based channelization processing is based on polyphase filtering and segmented convolution to ensure phase continuity of the channelized results. However, as the increasing of data amount, it cannot meet the requirements of real-time processing. The study on how to implement high-performance channelization processing based on GPU is currently an urgent issue that needs to be addressed. The inefficiency of traditional segmented convolution methods on GPU architecture is first analyzed, then a buffer shift strategy is proposed to ensure the phase continuity of channelization results with lower buffer space, reduced computational overhead and simplified logic control. Additionally, the analysis of two multistage filtering approaches, i.e., polyphase filtering and direct filtering, demonstrates the superiority of direct filtering on GPU architecture. Simulation experiments show that the proposed method correctly ensures the phase continuity of the channelization results and confirms that direct filtering is faster than polyphase filtering on GPU architecture. The acceleration ratios of direct filtering using GPU and CPU are provided to illustrate the substantial improvement in computational efficiency by the proposed strategy. In a word, the proposed buffer shift strategy-based GPU-efficient channelization method has significant advantages in processing speed and data phase continuity and is particularly suitable for real-time processing of large-scale data.