基于GPU的快速夏克-哈特曼波前重构方法研究

在夏克-哈特曼波前重构方法中，为了提高波前重构的速度，解决实时性差的问题，提出了一种基于GPU的快速夏克-哈特曼波前重构方法：将Zernike模式波前重构算法中Zernike多项式系数和波前数据的矩阵求解部分按照棋盘式和带状式分解并行化，根据不同的分解并行模式调度线程块及线程块中的线程数，转存数据到共享内存提高数据访问效率，优化数据存储结构提高全局存储器访问带宽，多个线程并发执行，快速重构波前。实验表明：基于GPU的快速算法相比传统CPU方式具有明显加速效果，当波前插值分辨率为2048*2048时，速度甚至提高了206倍；在加速的同时仍可保持原有算法的高准确度，为高精度、高分辨率的波前重构提供了快速、实时的计算方法。

Fast Wavefront Reconstructions for Hartmann-Shack Wavefront Sensor Based on GPU

In the Hartmann-Shack wavefront reconstruction method in order to increase the speed of the wavefront re-construction,to solve the problem of poor real-time,we propose a fast GPU-based Shack - Hartmann wavefront re-construction method:the matrix solver part of the polynomial coefficients in accordance with checkerboard is decom-posed parallel, and part of wavefront data decomposed according to the ribbon-style parallelization in Zernike mode wave reconstruction algorithm. Schedule thread block and the number of threads in the block depending on the parallel decomposition mode,dump data to a shared memory to improve data access efficiency,optimize the data storage struc-ture to improve global memory access bandwidth, execute concurrently multiple threads and rapid reconstruct wavefront. Experimental results show that:compared to traditional CPU mode,fast algorithm based on GPU has significantly ac-celerated effects. The speed increase even 206 times in the wavefront interpolated resolution of 2048*2048;This fast al-gorithm can still maintain high accuracy,as a fast and real-time calculation method for the high-precision,high-resolu-tion wave front reconstruction.