受功耗、面积的限制，高性能众核处理器倾向于将片上SRAM组织成SPM这种非Cache形式，与片外主存构成多级存储架构。这种存储架构需要软件显式管理应用程序中的数据存储和传输。为此，本文提出了一种简便的栈式片上内存动态管理方法。该方法首先选择应用程序中可进行访存优化的数组变量，分析这些数组变量的生存周期，根据生存周期相干情况提出一种栈式的动态片上内存管理方法，将更多的数组变量动态存储在片上内存中，同时结合数组变量的优化收益评估将那些访存密度高的变量有限布局在片上内存中。实验结果验证了该方法的有效性。

Manycore CPUs prefer selecting SPM, and a noncached onchip SRAM acts  as the fast memory in these memory hierarchies owing to the area and power consumption. These memory hierarchies need software to  explicitly manage data placement and data transfer. This article proposes an easy stackanalogy onchip memory dynamic deallocation approach applied to program memoryaccesssequence optimizations with manycore memory hierarchies. It introduces an ILP solver to develop a method when the sum of the array variable sizes in each program phase is larger than the onchip memory size. The experimental results show that this approach can improve significantly the performance.