【摘要】 为了进一步提高组件程序KYLIN V2.0的计算速度，本文开展了图形处理器加速方法研究，分析了硬件架构。将特征线输运计算中的大量简单重复计算从中央处理器转移至图形处理器，建立了异构系统特征线扫描算法，并利用异构通信机制，对于组件基准题加速比最高可达到50倍。同时，为了进一步完善KYLIN V2.0程序验证确认矩阵，本文采用LCT临界实验装置开展了验证确认研究，验证算例包括了LCT003系列、LCT011系列共14个临界实验装置，最大特征值偏差8.05×10^-3,与同类型程序具有相近的计算精度。本文研究验证了图形处理器具有良好的加速效果，实现了KYLIN V2.0程序的大幅加速，同时进一步补充了程序验证矩阵，支持了程序的工程应用。更多还原

【Abstract】 The GPU acceleration method was examined to enhance the calculation speed of the developed KYLIN V2.0 modular program. This method offloads a large number of simple, repetitive MOC transport calculations from CPUs to GPUs. In addition, a characteristic line scanning algorithm for heterogeneous systems is established, and the heterogeneous communication mechanism is fully utilized. The acceleration ratio can reach up to 50 times with the VERA-2 modular benchmark problems. Further verification was conducted with the LCT critical experiment, including 14 cases from the LCT003 and LCT011 series. The maximum deviation in eigenvalue results is 8.05×10^-3, with calculation accuracy comparable to that of similar programs. This research verifies the GPU′s excellent acceleration effect, realizing the acceleration of the KYLIN V2.0 program greatly, and further supports its engineering application through an expanded program verification matrix.更多还原