通用图形处理器(GPGPU)在并行计算方面的巨大优势使其在高性能计算领域的应用日趋广泛，尤其是在面向高吞吐量的通用计算计算领域已经成为了主流的计算加速部件。然而，由于分支转移、访存离散和片上资源访问竞争等因素的存在，当GPGPU处理不规则应用程序时，其性能并未能得以充分发挥。本课题将从线程调度优化的角度出发，基于对GPGPU微体系结构的分析，构建基于线程调度的GPGPU性能优化整体框架，对分支转移、访存离散和片上资源访问竞争等三个影响GPGPU性能的主要因素进行研究，重点解决由于分支转移引起的资源利用率降低、访存离散和片上资源访问竞争导致的访存效率降低等方面的问题。

General purpose graphics processor(GPGPU) has an increasing application in the field of high performance computing based on its parallel computing capability, which makes it become one of the main stream computing speedup components, especially in the field of general-purpose computing feild oriented throughput. However, the performance of GPGPU fails to reach the maximum when some irregular applications run on GPGPU as there exist some factors affecting the performance of GPGPU such as branch divergence, memory divergence and on-chip resources contention, etc. We will study these factors affecting the performance of GPGPU, especially branch divergence, memory divergence and on-chip resources contention based on thread scheduling and microarchitecture of GPGPU. Moreover, we will construct the overall performance optimization framework based on thread scheduling. The performance of GPGPU will be optimized via solving the issues including the reduction of on-chip resources utilization occurred by branch divergence and the drop of memory efficiency occurred by memory divergence and on-chip resources contention.