有限元GPU加速计算的实现方法

研究基于GPU的有限元求解中的总刚矩阵生成和线性方程组求解问题.通过对单元着色和分组完成总刚矩阵的生成,并以行压缩存储(Compressed Sparse Row,CSR)格式存储,用预处理共轭梯度法求解所生成的大规模线性稀疏方程组.在CUDA(Compute Unified Device Architecture)平台上完成程序设计,并用GT430 GPU对弹性力学的平面问题和空间问题进行试验.结果表明,总刚矩阵生成和方程组求解分别得到最高11.7和8的计算加速比.     

GPU加速不完全Cholesky分解预条件共轭梯度法

不完全 Cholesky 分解预条件共轭梯度（incomplete Cholesky factorization preconditioned conjugate gradient ,ICCG）法是求解大规模稀疏对称正定线性方程组的有效方法。然而ICCG法要求在每次迭代中求解2个稀疏三角方程组,稀疏三角方程组求解固有的串行性成为了ICCG法在GPU上并行求解的瓶颈。针对稀疏三角方程组求解,给出了一种利用GPU 加速的有效方法。为了增加稀疏三角方程组求解在GPU上的多线程并行性,提出了对不完全Cholesky分解产生的稀疏三角矩阵进行分层调度（level scheduling ）的方法。为了进一步提高稀疏三角方程组求解的并行性能,提出了在分层调度前通过近似最小度（approximate minimum degree ,AMD）算法对系数矩阵进行重排序、在分层调度后对稀疏三角矩阵进行层排序的方法,降低了分层调度过程中产生的层数,优化了稀疏三角方程组求解的GPU内存访问模式。数值实验表明,与利用NVIDIA CUSPARSE实现的ICCG法相比,采用上述方法性能可以获得平均1倍以上的提升。     

基于GPU的稀疏线性系统的预条件共轭梯度法

研究了基于GPU的稀疏线性方程组的预条件共轭梯度法加速求解问题,并基于统一计算设备架构(CUDA)平台编制了程序,在NVIDIAGT430 GPU平台上进行了程序性能测试和分析。稀疏矩阵采用压缩稀疏行(CSR)格式压缩存储,针对预条件共轭梯度法的算法特性,研究了基于GPU的稀疏矩阵与向量相乘的性能优化、数据从CPU端传到GPU端的加速传输措施。将编制的稀疏矩阵与向量相乘的kernel函数和CUSPARSE函数库中的cusparseDcsrmv函数性能进行了对比,最优得到了2.1倍的加速效果。对于整个预条件共轭梯度法,通过自编kernel函数来实现的算法较之采用CUBLAS库和CUSPARSE库实现的算法稍具优势,与CPU端的预条件共轭梯度法相比,最优可以得到7.4倍的加速效果。     

GRAPES动力框架中大规模稀疏线性系统并行求解及优化

赫姆霍兹方程求解是GRAPES数值天气预报系统动力框架中的核心部分,可转换为大规模稀疏线性系统的求解问题,但受限于硬件资源和数据规模,其求解效率成为限制系统计算性能提升的瓶颈。分别通过MPI、MPI+OpenMP、CUDA三种并行方式实现求解大规模稀疏线性方程组的广义共轭余差法,并利用不完全分解LU预处理子（ILU）优化系数矩阵的条件数,加快迭代法收敛。在CPU并行方案中,MPI负责进程间粗粒度并行和通信,OpenMP结合共享内存实现进程内部的细粒度并行,而在GPU并行方案中,CUDA模型采用数据传输、访存合并及共享存储器方面的优化措施。实验结果表明,通过预处理优化减少迭代次数对计算性能提升明显,MPI+OpenMP混合并行优化较MPI并行优化性能提高约35%,CUDA并行优化较MPI+OpenMP混合并行优化性能提高约50%,优化性能最佳。     

GPU稀疏矩阵向量乘的性能模型构造

稀疏矩阵向量乘(Sparse matrix-vector multiplication,SPMV)是广泛应用于大规模线性求解系统和求解矩阵特征值等问题的基本运算,但在迭代处理过程中它也常常成为处理的瓶颈,影响算法的整体性能。对于不同形态的矩阵,选择不同的存储格式 ,对应的算法往往会产生较大的性能影响。通过实验分析,找到各种矩阵形态在不同存储结构下体现的性能变化特征,构建一个有效的性能度量模型,为评估稀疏矩阵运算开销、合理选择存储格式做出有效的指导。在14组CSR,COO,HYB格式和8组ELL格式的测试用例下,性能预测模型和测量之间的差异低于9%。     

基于HYB格式稀疏矩阵与向量乘在CPU+GPU异构系统中的实现与优化

稀疏矩阵与向量相乘SpMV是求解稀疏线性系统中的一个重要问题,但是由于非零元素的稀疏性,计算密度较低,造成计算效率不高。针对稀疏矩阵存在的一些不规则性,利用混合存储格式来进行SpMV计算,能够提高对稀疏矩阵的压缩效率,并扩大其适应范围。HYB是一种广泛使用的混合压缩格式,其性能较为稳定。而随着GPU并行计算得到普遍应用以及CPU日趋多核化,因此利用GPU和多核CPU构建异构并行计算系统得到了普遍的认可。针对稀疏矩阵的HYB存储格式中的ELL和COO存储特征,把两部分数据分别分割到CPU和GPU进行协同并行计算,既能充分利用CPU和GPU的计算资源,又能够发挥CPU和GPU的计算特性,从而提高了计算资源的利用效能。在分析CPU+GPU异构计算模式的特征的基础上,对混合格式的数据分割和共享方面进行优化,能够较好地发挥在异构计算环境的优势,提高计算性能。     

带状稀疏矩阵乘法及高效GPU实现

稀疏-稠密矩阵乘法（SpMM）广泛应用于科学计算和深度学习等领域，提高它的效率具有重要意义。针对具有带状特征的一类稀疏矩阵，提出一种新的存储格式BRCV(Banded Row Column Value)以及基于此格式的SpMM算法和高效图形处理单元（GPU）实现。由于每个稀疏带可以包含多个稀疏块，所提格式可看成块稀疏矩阵格式的推广。相较于常用的CSR(Compressed Sparse Row)格式，BRCV格式通过避免稀疏带中列下标的冗余存储显著降低存储复杂度；同时，基于BRCV格式的SpMM的GPU实现通过同时复用稀疏和稠密矩阵的行更高效地利用GPU的共享内存，提升SpMM算法的计算效率。在两种不同GPU平台上针对随机生成的带状稀疏矩阵的实验结果显示，BRCV的性能不仅优于cuBLAS(CUDA Basic Linear Algebra Subroutines)，也优于基于CSR和块稀疏两种不同格式的cuSPARSE。其中，相较于基于CSR格式的cuSPARSE,BRCV的最高加速比分别为6.20和4.77。此外，将新的实现应用于图神经网络（GNN）中的SpMM算子的加速。在实际应...     

一种准对角矩阵的混合压缩算法及其与向量相乘在GPU上的实现

稀疏矩阵与向量乘(SpMV)属于科学计算和工程应用中的一种基本运算,其高性能实现与优化是计算科学的研究热点之一。在微分方程的求解过程中会产生大规模的稀疏矩阵,而且很大一部分是一种准对角矩阵。针对准对角矩阵存在的一些不规则性,提出一种混合对角存储(DIA)和行压缩存储(CSR)格式来进行SpMV计算,对于分割出来的对角线区域之外的离散非零元素采用CSR存储,这样能够克服DIA在不规则情况下存储矩阵的列迅速增加的缺陷,同时对角线采用DIA存储又能充分利用矩阵的对角特征,以减少CSR的行非零元素数目的不均衡现象,并可以通过调整存储对角线的带宽来适应准对角矩阵的不同的离散形式,以获得比DIA和CSR更高的压缩比,减小计算的数据规模。利用CUDA平台在GPU上进行了实验测试,结果表明该方法比DIA和CSR具有更高的加速比。     

基于GPU的重启PGMRES并行算法研究

重启的PGMRES算法是求解稀疏线性方程组高效的迭代方法之一,计算过程也比较稳定。为加快大规模稀疏线性方程组的求解速度,对重启PGMRES算法使用GPU并行方式进行并行算法实现。提出了ELL压缩存储格式的新存取方式,并依据问题规模和SM数目提出了动态分配线程策略。实验结果表明,该算法可有效提高SM资源利用率,获得3~10倍的加速比。     

大型复线性方程组预处理双共轭梯度法

当复线性方程组的规模较大或系数矩阵的条件数很大时,系数矩阵易呈现病态特性,双共轭梯度法存在不收敛和收敛速度慢的潜在问题,采用适当的预处理技术,可以改善矩阵病态特性,加快收敛速度。从实型不完全Cholesky分解预处理方法出发,构造了一种针对复线性方程组的预处理方法,结合双共轭梯度法,给出了一种预处理双共轭梯度法。数值算例表明该算法求解速度快,可靠高效,能够应用于大型复线性方程组的求解。     

A segment‐based sparse matrix–vector multiplication on CUDA

The challenge for Sparse Matrix–Vector multiplication (SpMV) performance is memory bandwidth, which mostly depends on input matrices and underlying computing platforms. To solve this challenge, many researchers have explored a variety of optimization techniques. One of the most promising aspects focuses on designing storage formats to represent sparse matrices. However, lots of prior storage formats cannot fully take advantage of the underlying computing platforms, resulting in unsatisfactory performance and large memory footprint. Therefore, a novel storage format, called Segmented Hybrid ELL + Compressed Sparse Row (CSR) (SHEC for short), is proposed to further improve the throughput and lessen memory footprint on Graphics Processing Unit (GPU). SHEC format employs an interleaved combination pattern, which combines certain amount of compressed rows to form a new SHEC row. Segmentation is brought in to balance load and reduce memory footprint. According to the empirical data, an automatic SHEC‐based SpMV is developed to fit for all the matrices. Experimental results show that SHEC approach outperforms the best results of NVIDIA SpMV library and exhibits a comparable performance with state‐of‐the‐art storage formats on the standard dataset. Copyright © 2012 John Wiley & Sons, Ltd.     

大型稀疏线性方程组符号LU分解法

基于有限元总刚矩阵的大规模稀疏性、对称性等特性,采用全稀疏存储结构以及最小填入元算法,使得计算机的存储容量达到最少。为了节省计算机的运算时间,对总刚矩阵进行符号LU分解方法,大大减少了数值求解过程中的数据查询。这种全稀疏存储结构和符号LU分解相结合的求解方法,使大规模稀疏线性化方程组的求解效率大大提高。数值算例证明该算法在时间和存贮上都较为占优,可靠高效,能够应用于有限元线性方程组的求解。     

CUDA-enabled Sparse Matrix–Vector Multiplication on GPUs using atomic operations

Existing formats for Sparse Matrix–Vector Multiplication (SpMV) on the GPU are outperforming their corresponding implementations on multi-core CPUs. In this paper, we present a new format called Sliced COO (SCOO) and an efficient CUDA implementation to perform SpMV on the GPU using atomic operations. We compare SCOO performance to existing formats of the NVIDIA Cusp library using large sparse matrices. Our results for single-precision floating-point matrices show that SCOO outperforms the COO and CSR format for all tested matrices and the HYB format for all tested unstructured matrices on a single GPU. Furthermore, our dual-GPU implementation achieves an efficiency of 94% on average. Due to the lower performance of existing CUDA-enabled GPUs for atomic operations on double-precision floating-point numbers the SCOO implementation for double-precision does not consistently outperform the other formats for every unstructured matrix. Overall, the average speedup of SCOO for the tested benchmark dataset is 3.33 (1.56) compared to CSR, 5.25 (2.42) compared to COO, 2.39 (1.37) compared to HYB for single (double) precision on a Tesla C2075. Furthermore, comparison to a Sandy-Bridge CPU shows that SCOO on a Fermi GPU outperforms the multi-threaded CSR implementation of the Intel MKL Library on an i7-2700 K by a factor between 5.5 (2.3) and 18 (12.7) for single (double) precision.     

An optimal storage format for sparse matrices

The irregular nature of sparse matrix-vector multiplication, Ax=y, has led to the development of a variety of compressed storage formats, which are widely used because they do not store any unnecessary elements. One of these methods, the Jagged Diagonal Storage format (JDS) is, in addition, considered appropriate for the implementation of iterative methods on parallel and vector processors. In this work we present the Transpose Jagged Diagonal Storage format (TJDS) which drew inspiration from the Jagged Diagonal Storage scheme but requires less storage space than JDS. We propose an alternative storage scheme which makes no assumptions about the sparsity pattern of the matrix and only needs three linear arrays instead of the four linear arrays required by JDS. Specifically, the data is aligned in such a way that the permutation array used in JDS, to permute the solution vector back to the original ordering, is unnecessary. This allow us to save the memory space required to store an integer vector of length n, where n stands for the number of columns in the sparse matrix A. This storage saving reaches, for the selection of matrices used in this work, from 14% up to 45% of the number of non-zero values of the sparse matrices. We present a case study of a 6×6 sparse matrix to show the data structures and the algorithm to compute Ax=y using the TJDS format.     

GPU accelerated sparse matrix‐vector multiplication and sparse matrix‐transpose vector multiplication

Many high performance computing applications require computing both sparse matrix‐vector product (SMVP) and sparse matrix‐transpose vector product (SMTVP) for better overall performance. Under such a circumstance, it is critical to maintain a similarly high throughput for these two computing patterns with the underlying sparse matrix encoded in a single storage format. The compressed sparse block (CSB) format proposed by Buluç et al. allows computing both problems on multi‐core CPUs with nearly identical throughputs. On the other hand, a direct porting of CSB to graphics processing units (GPUs), which have been recently recognized as a powerful general purpose computing platform, turns out to be inefficient. In this work, we propose a new data structure, designated as expanded CSB (eCSB), to minimize the throughput gap between SMVP and SMTVP computations on GPUs, while at the same time enable a high computing throughput. We also use a hybrid storage format to store elements in each block, which can be selected dynamically at runtime. Experimental results show that the proposed techniques implemented on a Kepler GPU delivers similar throughput on both SMVP and SMTVP and the throughput is up to 13 times faster than that of the CPU‐based CSB implementation. In addition, our eCSB procedure outperforms the previous GPU results by up to 188% and 914% in computing SMVP and SMTVP, and we validate the effectiveness of eCSB by means of wall‐clock time of bi‐conjugate gradient algorithm; our eCSB is 25% faster than Compressed Sparse Rows (CSR) and 6% faster than HYB, respectively. Copyright © 2014 John Wiley & Sons, Ltd.     

Parallel Solutions for Large-Scale General Sparse Nonlinear Systems of Equations

In solving application problems,many large-scale nonlinear systems of equaions result in sparse Jacobian matrices.Such nonlinear systems are called sparse nonlinear systems.The irregularity of the locations of nonzrero elements of a general sparse matrix makes it very difficult to generally map sparse matrix computations to multiprocessors for parallel processing in a well balanced manner.To overcome this difficulty,we define a new storage scheme for general sparse matrices in this paper,With the new storage scheme,we develop parallel algorithms to solve large-scale general sparse systems of equations by interval Newton/Generalized bisection methods which reliably find all numerical solutions within a given domain.I n Section 1,we provide an introduction to the addressed problem and the interval Newton‘s methods.In Section 2,some currently used storage schemes for sparse systems are reviewed.In Section 3,new index schemes to store general sparse matrices are reported.In Section 4,we present a parallel algorithm to evaluate a general sparse Jacobian matrix.In Section 5,we present a parallel algorithm to solve the corresponding interval linear system by the all-row preconditioned scheme.Conclusions and future work are discussed in Section 6.