为了满足4 m天文望远镜液晶自适应光学系统的波前处理要求，研究了基于多GPU的波前处理器。介绍了液晶自适应光学波前处理方法。分析了用于匹配4 m望远镜的哈特曼探测器数目、Zernike模式数和液晶校正器驱动单元数。详细论述了多GPU下波前处理方法，包括: 单GPU下计算斜率; 按列分块法拟合Zernike系数; Zernike对称性算法和按行分块法计算液晶校正器灰度图。最后，分析了匹配4 m望远镜的液晶自适应光学系统的残余误差传递函数，并由此模拟了残余误差传递函数的幅频响应。实验结果表明，斜率计算延迟18 μs，Zernike系数拟合延迟39 μs，校正器灰度图计算延迟114 μs。多GPU下总的波前处理延迟171 μs，小于哈特曼探测器采样时间500 μs，液晶自适应光学系统-3 dB残余误差抑制带宽可达53 Hz。满足4 m天文望远镜的应用要求。

In order to meet the wavefront processing requirements of liquid crystal adaptive optics system which match 4 m astronomy telescope, a wavefront processor (LCWP) based on multi-GPU is developed. First, the liquid crystal adaptive optics wavefront processing method is described. The numbers of hartmann detector’s microlenses, Zernike modes and liquid crystal corrector’s drive unit are analyzed, which need to match 4 m telescope. Then the wavefront processing methods based on multi-GPU are discussed in detail: single GPU method is used to calculate slope; block-column distribution method is used to fit Zernike coefficient; Zernike polynomial symmetry algorithm and block-row distribution method are used to compute gray map of liquid crystal corrector. Finally, the transfer function of liquid crystal adaptive optics system for matching 4 m telescope is analyzed and magnitude of the residual error transfer function is simulated. The latency of slope calculation is 18 μs, the latency of Zernike coefficient fitting is 39 μs and the latency of computing gray map is 114 μs. The total computing latency is 109 μs and temporal band width in our liquid crystal adaptive optics system is 53 Hz. It is shown that our LCWP based on multi-GPU has the ability to match 4 m astronomy telescope.