基于GPU集群的Level Set并行高精度演化

Level Set Parallel Highly Accurate Evolution Based on GPU Cluster

设计实现基于张量积B样条的并行Level Set演化算法,张量积B样条提高了演化精度和并行度;每步演化都需要反算B样条系数.针对对角占优三对角方程组,设计实现基于精确LU分解的高精度并行追赶法,并用它反算B样条系数;采用两步通信方法,消除通信的依赖关系,实现有效的并行通信.实验表明,本文的并行算法可以有效加速演化过程.

We design and implement parallel level set evolution algorithm based on tensor product B-spline which improves accuracy and parallelism of level set evolution. Each step of evolution need compute B-spline coefficients. Furthermore,a parallel high approximation solver for diagonally dominant tridiagonal linear system based on exact LU decomposition is implemented,which is used to compute B-spline coefficients. Two step communications are used to remove communicating dependency,so as to communicate in parallel. As a result,it speeds up level set evolution efficiently.