Parallel Implementation of the BiCGStab（2） Method in GPU Using CUDA and Matlab for Solution of Linear Systems

This paper presents a parallel implementation of the hybrid BiCGStab（2） （bi-conjugate gradient stabilized） iterative method in a GPU （graphics processing unit） for solution of large and sparse linear systems. This implementation uses the CUDA-Matlab integration, in which the method operations are performed in a GPU core using Matlab built-in functions. The goal is to show that the exploitation of parallelism by using this new technology can provide a significant computational performance. For the validation of the work, we compared the proposed implementation with a BiCGStab（2） sequential and parallelized implementation in the C and CUDA-C languages. The results showed that the proposed implementation is more efficient and can be viable for simulations being carried out with quality and in a timely manner. The gains in computational efficiency were 76x and 6x compared to the implementation in C and CUDA-C, respectively.     

基于MPI+CUDA异步模型的并行矩阵乘法

矩阵乘法在科学计算领域中起着重要的作用,不同结构模型能够改善并行矩阵乘的性能。现有的MPI+CUDA同步模型中,主机端需要进入等待状态,直到设备端完成任务后才能继续工作,这显然浪费时间。针对上述问题,提出一种基于MPI+CUDA异步模型的并行矩阵乘法。该模型避免了主机端进入等待状态,并采用CUDA流技术解决数据量超过GPU内存问题。通过分析异步模型的加速比和效率,实验结果表明,此方法显著提高了并行效率和大型矩阵乘法的运算速度,充分发挥了节点间分布式存储和节点内共享内存的优势,是一种有效可行的并行策略。     

MR-DBSCAN： a scalable MapReduce-based DBSCAN algorithm for heavily skewed data

DBSCAN （density-based spatial clustering of ap- plications with noise） is an important spatial clustering tech- nique that is widely adopted in numerous applications. As the size of datasets is extremely large nowadays, parallel process- ing of complex data analysis such as DBSCAN becomes in- dispensable. However, there are three major drawbacks in the existing parallel DBSCAN algorithms. First, they fail to prop- erly balance the load among parallel tasks, especially when data are heavily skewed. Second, the scalability of these al- gorithms is limited because not all the critical sub-procedures are parallelized. Third, most of them are not primarily de- signed for shared-nothing environments, which makes them less portable to emerging parallel processing paradigms. In this paper, we present MR-DBSCAN, a scalable DBSCAN algorithm using MapReduce. In our algorithm, all the crit- ical sub-procedures are fully parallelized. As such, there is no performance bottleneck caused by sequential process- ing. Most importantly, we propose a novel data partitioning method based on computation cost estimation. The objective is to achieve desirable load balancing even in the context of heavily skewed data. Besides, We conduct our evaluation us- ing real large datasets with up to 1.2 billion points. The ex- periment results well confirm the efficiency and scalability of MR-DBSCAN.     

Parallel computation of with multivariate determinants of matrices polynomial entries

In this paper we present an extension to the work of Bjorck et al. for computing the determinants of matrices with univariate or bivariate polynomials as entries to multivariate case. The algorithm supports parallel computation and has been implemented on a multi-core cluster computer system. We show how to use our approach to calculate two unsolved problems, which arise from computational geometry optimization and electric power engineering, and analyze the time complexity as well as bits complexity.     

The Methodology of Application Development for Hybrid Architectures

This paper provides an overview of the main recommendations and approaches of the methodology on parallel computation application development for hybrid structures. This methodology was developed within the master＇s thesis project ＂Optimization of complex tasks＇ computation on hybrid distributed computational structures＂ accomplished by Orekhov during which the main research objective was the determination of＂ patterns of the behavior of scaling efficiency and other parameters which define performance of different algorithms＇ implementations executed on hybrid distributed computational structures. Major outcomes and dependencies obtained within the master＇s thesis project were formed into a methodology which covers the problems of applications based on parallel computations and describes the process of its development in details, offering easy ways of avoiding potentially crucial problems. The paper is backed by the real-life examples such as clustering algorithms instead of artificial benchmarks.     

异构平台下格子Boltzmann方法实现及性能分析

对CPU+GPU异构平台下的多种并行编程模式进行了研究,并针对格子Boltzmann方法实现了CUDA,MPI+CUDA,MPI+OpenMP+CUDA多级并行算法。结果表明,算法具有较好的加速性能;提出的根据计算量比例参数调节CPU和GPU之间负载均衡的方法,对于在异构平台上实现多级并行处理及资源的有效利用具有一定的参考和应用价值。     

Design and aerodynamic investigation of a parallel vehicle on a wide-speed range

In order to design a hypersonic vehicle for a wide-ranged Mach number, a novel parallel vehicle for a wide-speed range has been proposed. In this paper, we employ a numerical method to investigate a parallel vehicle＇s aerodynamic performance and flow field characteristics. The obtained results show that the aerodynamic performance of the novel parallel vehicle is better than that of the waverider designed with a single Mach number for the wide-speed range. With the increase in Mach number, the lift-to-drag ratio of the novel parallel vehicle first increases and then decreases. When the Mach number is 7 and the angle of attack is 3°, the lift-to-drag ratio is the largest, and its value is 3.968. When the angle of attack is 3°, the lift-to-drag ratio is not lower than 3.786 in the range considered in the current study, and the novel parallel vehicle＇s aerodynamic performance is good. The wing changes the drag performance of the parallel vehicle remarkably, and results in the decrease of the lift-to-drag ratio. Meanwhile, the wing can enhance the pitching moment performance.     

广义稠密对称特征问题标准化算法在GPU集群上的有效实现

广义稠密对称特征问题的求解是许多应用科学和工程的主要任务,并且是计算电磁学、电子结构、有限元模型和量子化学等计算中的重要部分。将广义对称特征问题转化为标准对称特征问题是求解广义稠密对称特征问题的关键计算步骤。针对GPU集群,文中给出了广义稠密对称特征问题标准化块算法在GPU集群上基于MPI+CUDA的实现。为了适应GPU集群的架构,广义对称特征问题标准化算法将正定矩阵的Cholesky分解与传统的广义特征问题标准化块算法相结合,降低了标准化算法中不必要的通信开销,并且增强了算法的并行性。在基于MPI+CUDA的标准化算法中,GPU与CPU之间的数据传输操作被用来掩盖GPU内的数据拷贝操作,这消除了拷贝所花费的时间,进而提高了程序的性能。同时,文中还给出了矩阵在二维通信网格中行通信域和列通信域之间完全并行的点对点的转置算法和基于MPI+CUDA的具有多个右端项的三角矩阵方程BX=A求解的并行块算法。在中科院计算机网络信息中心的超级计算机系统“元”上,每个计算节点配置2块Nvidia Tesla K20 GPGPU卡及2颗Intel E5-2680 V2处理器,使用多达32个GPU对不同规模矩阵的基于MPI+CUDA的广义对称特征问题标准化算法进行测试,取得了较好的加速效果与性能,并且具有良好的可扩展性。当使用32个GPU对50000×50000阶的矩阵进行测试时,峰值性能达到了约9.21 Tflops。     

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。     

A CUDA-enabled parallel algorithm for accelerating retinex

Retinex is an image restoration approach used to restore the original appearance of an image. Among various methods, a center/surround retinex algorithm is favorable for parallelization because it uses the convolution operations with large-scale sizes to achieve dynamic range compression and color/lightness rendition. This paper presents a GPURetinex algorithm, which is a data parallel algorithm accelerating a modified center/surround retinex with GPGPU/CUDA. The GPURetinex algorithm exploits the massively parallel threading and heterogeneous memory hierarchy of a GPGPU to improve efficiency. Two challenging problems, irregular memory access and block size for data partition, are analyzed mathematically. The proposed mathematical models help optimally choose memory spaces and block sizes for maximal parallelization performance. The mathematical analyses are applied to three parallelization issues existing in the retinex problem: block-wise, pixel-wise, and serial operations. The experimental results conducted on GT200 GPU and CUDA 3.2 showed that the GPURetinex can gain 74 times acceleration, compared with an SSE-optimized single-threaded implementation on Core2 Duo for the images with 4,096 × 4,096 resolution. The proposed method also outperforms the parallel retinex implemented with the nVidia Performance Primitives library. Our experimental results indicate that careful design of memory access and multithreading patterns for CUDA devices should acquire great performance acceleration for real-time processing of image restoration.     

一种基于Hadoop+CUDA实现相关器的方法

根据21CMA相关器的算法特点,在对比基于CPU并行的MPI集群、MPI+CUDA异构并行集群和Hadoop+CUDA异构并行集群的架构特点的基础上,提出了一种基于Hadoop+CUDA平台实现软相关器的方法。本方法利用GPU在计算FFT、向量乘和向量加等密集型计算模型的优势,设计相关器的并行模型,使其性能较前期在CPU并行的MPI集群实现的相关器有了大幅提升。同时,本文选择广泛应用于大数据处理平台的Hadoop软件架构,利用Hadoop Streaming工具实现非Java编写的程序在分布式系统中并行执行,非常便捷地获得了集群系统的线性加速比。Hadoop HDFS并行文件系统管理结果数据和过程日志更加灵活可靠,为后续的大数据分析提供了支撑环境。     

基于CUDA架构的FFT并行计算研究

FFT（快速傅里叶变换）是基于提高DFT（离散傅里叶变换）计算的高效算法,它在众多科学和工程领域都得到了广泛的应用。自FFT算法出现以后,从早期的以降低复杂度到近年以来的大规模并行FFT计算,各种优化算法得到广泛的研究。在并行运算领域中,随着可编程的、并行化GPU的不断推广,特别是通用并行统一计算架构CUDA的出现,极大增强了GPU的计算能力,在编程和优化等方面都有显著地提升。鉴于此,本文在分析FFT算法实现的基础上,研究了一种适合GPU运算的FFT并行计算方法,并通过CUDA架构实现了FFT算法在GPU上的运算。该方法的引入在理论不计算数据传输的情况下,使一维FFT运算时间的复杂度由O（N logN2）可以降到O（N/rlogN2）。通过验证,本文提出的CUDA的并行FFT方法得到较好的加速效果,在精度计算上也符合实际的要求,从而证明了该方法的正确性和有效性。     

CUDA并行计算技术在情报信息研判中的应用

文章在研究公安情报信息研判技术的基础上,提出了一种基于CUDA并行计算技术的方法,实现对公安情报信息中文本信息快速分类的方法,实现将CUDA技术的快速计算能力应用到公安情报研判工作中。该文从介绍CUDA技术的概况出发,阐述了基于CUDA并行计算技术的文本分类方法,以及该方法的详细实现过程,解决了高效处理海量文本信息的问题。实验结果证明,CUDA并行计算技术在公安情报信息研判工作中卓有成效。     

Parallel data mining techniques on Graphics Processing Unit with Compute Unified Device Architecture (CUDA)

Recent development in Graphics Processing Units (GPUs) has enabled inexpensive high performance computing for general-purpose applications. Compute Unified Device Architecture (CUDA) programming model provides the programmers adequate C language like APIs to better exploit the parallel power of the GPU. Data mining is widely used and has significant applications in various domains. However, current data mining toolkits cannot meet the requirement of applications with large-scale databases in terms of speed. In this paper, we propose three techniques to speedup fundamental problems in data mining algorithms on the CUDA platform: scalable thread scheduling scheme for irregular pattern, parallel distributed top-k scheme, and parallel high dimension reduction scheme. They play a key role in our CUDA-based implementation of three representative data mining algorithms, CU-Apriori, CU-KNN, and CU-K-means. These parallel implementations outperform the other state-of-the-art implementations significantly on a HP xw8600 workstation with a Tesla C1060 GPU and a Core-quad Intel Xeon CPU. Our results have shown that GPU + CUDA parallel architecture is feasible and promising for data mining applications.     

GPGPU编程技术初探

伴随着GPGPU计算技术的不断发展,HPC高性能计算系统体系结构正在悄然发生着一场变革,这场变革为高性能计算发展提供了一个新的方向、CUDA是NIVIDIA公司提供的利用GPGPU进行并行运算应用开发的一套C语言编程平台,通过它可以利用特定显卡的高性能运算能力进行一些大规模高性能计算,有效提升计算机系统的使用效率,本文主要介绍GPU发展现状以及如何利用CUDA编程技术进行并行运算软件开发．     

CUDA并行技术与数字图像几何变换

CUDA是GPU通过并发执行多个线程以实现大规模快速并行计算能力的技术,它能使对GPU编程变得更容易。介绍了CUDA基本特性及主要编程模型,在此基础上,提出并实现了基于NVIDIA CUDA技术的图像快速几何变换。采用位置偏移增量代替原变换算法中大量乘法运算,并把CUDA技术的快速并行计算能力应用到数字图像几何变换中,解决了基于CPU的传统图像几何变换运算效率低下的问题。实验结果证明使用CUDA技术,随着处理图像尺寸的增加,对数字图像几何变换处理效率最高能够提高到近100倍。     

Hybrid CUDA, OpenMP, and MPI parallel programming on multicore GPU clusters

Nowadays, NVIDIA's CUDA is a general purpose scalable parallel programming model for writing highly parallel applications. It provides several key abstractions – a hierarchy of thread blocks, shared memory, and barrier synchronization. This model has proven quite successful at programming multithreaded many core GPUs and scales transparently to hundreds of cores: scientists throughout industry and academia are already using CUDA to achieve dramatic speedups on production and research codes. In this paper, we propose a parallel programming approach using hybrid CUDA OpenMP, and MPI programming, which partition loop iterations according to the number of C1060 GPU nodes in a GPU cluster which consists of one C1060 and one S1070. Loop iterations assigned to one MPI process are processed in parallel by CUDA run by the processor cores in the same computational node.     

基于GPU的H.264并行解码算法

针对并行处理H.264标准视频流解码问题,提出基于CPU/GPU的协同运算算法。以统一设备计算架构(CUDA)语言作为GPU编程模型,实现DCT逆变换与帧内预测在GPU中的加速运算。在保持较高计算精度的前提下,结合CUDA混合编程,提高系统的计算性能。利用NIVIDIA提供的CUDA语言,在解码过程中使DCT逆变换和帧内预测在GPU上并行实现,将并行算法与CPU单机实现进行比较,并用不同数量的视频流验证并行解码算法的加速效果。实验结果表明,该算法可大幅提高视频流的编解码效率,比CPU单机的平均计算加速比提高10倍。     

CUDA体系结构上的一层浅水系统的模拟

对于使用支持NVIDACUDA程序设计模型的GPU的二维一层浅水系统,给出了如何加速平衡性良好的有限体积模式的数值解,同时给出并实现了在单双浮点精度下使用CUDA模型利用潜在数据并行的算法。数值实验表明,CUDA体系结构的求解程序比CPU并行实现求解程序高效。     

CUDA架构下大规模稠密线性方程组的并行求解

在Gauss-Jordan消去法的基础上,给出了一种适应于CUDA架构的改进Gauss-Jordan消去并行算法。通过分析该方法的处理过程以及CUDA架构的相应限制,在CUDA的grid-block-thread三层组织结构的基础上,从算法构造的角度提出了grid-strip-group-block-thread五层结构,给出了基础行以及全局基础行等概念,并构建了适应于CUDA架构的Gauss-Jordan消去法的并行版本,在最高维数为4 000维的大规模稠密线性方程组的算例求解上与串行Gauss-Jordan消去法进行了比较,实验结果表明,该算法能够充分利用GPU的硬件特性,有效地降低了大规模稠密线性方程组的求解时间。