Abstract

General purpose programming on the graphics processing units(GPGPU) has received a lot of attention in the parallel computing community as it promises to offer a large computational power at a very low price. GPGPU is best suited for regular data parallel algorithms. They are not directly amenable for algorithms which have irregular data access patterns such as convex hull, list ranking etc. In this paper, we present a GPU-optimized implementation for finding the convex hull of a two dimensional point set. Our implementation tries to minimize the impact of irregular data access patterns. Our implementation can find the convex hull of 10 million random points in less than 0.2 seconds and achieves a speedup of up to 14 over the standard sequential CPU implementation. We also discuss some of the practical issues relating to the implementation of convex hull algorithms on massively multi-threaded architectures like that of the GPU.

Abstract

ndexing image data for content-based image search is an important area in Computer Vision. The state of theart uses the 128-dimensional SIFT as low level descriptors. Indexing even a moderate collection involves severalmillions of such vectors. The search performance depends on the quality of indexing and there is often a need tointeractively tune the process for better accuracy. In this paper, we propose a a visualization-based tool to tunethe indexing process for images and videos. We use a feature selection approach to improve the clustering of SIFTvectors. Users can visualize the quality of clusters and interactively control the importance of individual or groupsof feature dimensions easily. The results of the process can be visualized quickly and the process can be repeated.The user can use a filter or a wrapper model in our tool. We use input sampling, GPU-based processing, and visualtools to analyze correlations to provide interactivity. We present results of tuning the indexing for a few standarddatasets. A few tuning iterations result in an improvement of over 4% in the final classification performance, whichis significant.

Abstract

Graphics processing units provide a large compu-tational power at a very low price which position them as anubiquitous accelerator. General purpose programming on thegraphics processing units (GPGPU) is best suited for regulardata parallel algorithms. They are not directly amenable foralgorithms which have irregular data access patterns suchas list ranking, and finding the connected components of agraph, and the like. In this work, we present a GPU-optimizedimplementation for finding the connected components of agiven graph. Our implementation tries to minimize the impactof irregularity, both at the data level and functional level.Our implementation achieves a speed up of 9 to 12 timesover the best sequential CPU implementation. For instance, ourimplementation finds connected components of a graph of 10million nodes and 60 million edges in about 500 millisecondson a GPU, given a random edge list. We also draw interestingobservations on why PRAM algorithms, such as the Shiloach-Vishkin algorithm may not be a good fit for the GPU and howthey should be modified.

Abstract

Genetic algorithms are effective in solving many optimiza-tion tasks. However, the long execution time associated withit prevents its use in many domains. In this paper, we pro-pose a new approach for parallel implementation of geneticalgorithm on graphics processing units (GPUs) using CUDAprogramming model. We exploit the parallelism within achromosome in addition to the parallelism across multiplechromosomes. The use of one thread per chromosome byprevious efforts does not utilize the GPU resources effec-tively. Our approach uses multiple threads per chromosome,thereby exploiting the massively multithreaded GPU moreeffectively. This results in good utilization of GPU resourceseven at small population sizes while maintaining impressivespeed up for large population sizes. Our approach is mod-eled after the GAlib library and is adaptable to a varietyof problems. We obtain a speedup of over 1500 over theCPU on problems involving a million chromosomes. Prob-lems of such magnitude are not ordinarily attempted due tothe prohibitive computation times.

Abstract

Large-scale 3D reconstruction has received a lot of attentionrecently. Bundle adjustment is a key component of the reconstructionpipeline and often its slowest and most computational resource intensive.It hasn’t been parallelized effectively so far. In this paper, we present ahybrid implementation of sparse bundle adjustment on the GPU usingCUDA, with the CPU working in parallel. The algorithm is decomposedinto smaller steps, each of which is scheduled on the GPU or the CPU. Wedevelop efficient kernels for the steps and make use of existing libraries forseveral steps. Our implementation outperforms the CPU implementationsignificantly, achieving a speedup of 30-40 times over the standard CPUimplementation for datasets with upto 500 images on an Nvidia TeslaC2050 GPU.

Abstract

Graphics processing units provide a large compu-tational power at a very low price which position them as an ubiquitous accelerator. General purpose programming on the graphics processing units (GPGPU) is best suited for regular data parallel algorithms. They are not directly amenable for algorithms which have irregular data access patterns such as list ranking, and finding the connected components of a graph, and the like. In this work, we present a GPU-optimized implementation for finding the connected components of a given graph. Our implementation tries to minimize the impact of irregularity, both at the data level and functional level. Our implementation achieves a speed up of 9 to 12 times over the best sequential CPU implementation. For instance, our implementation finds connected components of a graph of 10 million nodes and 60 million edges in about 500 milliseconds on a GPU, given a random edge list. We also draw interesting observations on why PRAM algorithms, such as the Shiloach-Vishkin algorithm may not be a good fit for the GPU and how they should be modified.

Abstract

Graphics Processing Units (GPU) are application specific accelerators which provide high performance to cost ratio and are widely available and used, hence places them as a ubiquitous accelerator. A computing paradigm based on the same is the general purpose computing on the GPU (GPGPU) model. The GPU due to its graphics lineage is better suited for the data-parallel, data-regular algorithms. The hardware architecture of the GPU is not suitable for the data parallel but data irregular algorithms such as graph connected components and list ranking. In this paper, we present results that show how to use GPUs efficiently for graph algorithms which are known to have irregular data access patterns. We consider two fundamental graph problems: finding the connected components and finding a spanning tree. These two problems find applications in several graph theoretical problems. In this paper we arrive at efficient GPU implementations for the above two problems. The algorithms focus on minimising irregularity at both algorithmic and implementation level. Our implementation achieves a speedup of 11-16 times over a corresponding best sequential implementation.

Abstract

Graphics processing units provide a large computational power at a very low price which position them as an ubiquitous accelerator. Efficient primitives that can expand the r ange of operations performed on the GPU are thus important. Discrete Range Searching(DRS) is one such primitive with direct applications to string processing, document and text retrieval systems, and least common ancestor queries. In this work, we present a GPU specific implementation of DRS with an optimal space-time trade off. Toward this end, we also present GPU amenable succinct representations and discuss limitations on the GPU. Our method uses 7.5 bits of additional space per element. The speedup achieved by our method is in the range of 20-25 for preprocessing, and 25-35 for batch querying over a sequential implementation. Compared to an 8-threaded implementation, our methods obtain a speedup of 6-8. We study applications of the DRS on the GPU. Also, we suggest that most graph algorithms which focus on using least common ancestor, can easily be enabled on the GPU based on range minima primitive. Beyond this, we show applications of DRS in string querying and tree queries, and suggest how DRS can be helpful in implementing tree based graph algorithms on the GPU.

Abstract

The Graphics Processing Units (GPUs) provide highcomputation power at a low cost and is an important computeaccelerator with a massively multithreaded architecture. Inthis paper, we present fast implementations of common graphoperations like breadth-first search, st-connectivity, single-sourceshortest path, all-pairs shortest path, minimum spanning tree,and maximum flow for undirected graphs on the GPU using theCUDA programming model. Our implementations exhibit highperformance, especially on large graphs. We use two data-parallelprogramming methodologies for these algorithms. One is aniterative, mask-based approach that processes valid data elementslike vertices and edges using independent threads for each.The other is a divide-and-conquer approach that reduces theproblem into smaller problems that are handled later using thesame approach. Parallel algorithms for such problems have beenreported in the literature before, especially on supercomputers.The massively multithreaded model of the GPU makes it possibleto adopt the data-parallel approach even to irregular algorithmslike graph algorithms, usingO(V)orO(E)simultaneous threads.The algorithms and the underlying techniques presented in thispaper are likely to be applicable to many irregular algorithms.We show the impact of our implementations on random, scale-free, and real-life graphs of up to millions of vertices on high-end and low-end GPUs. The availability and spread of GPUs todesktops and laptops make them ideal candidates to accelerategraph operations over the CPU-only implementations. Practicalimplementations of basic operations go a long way in realizingtheir potential.

Abstract

We present efficient implementations of two primitives for datamapping and distribution on the massively multithreaded architec-ture of the GPUs in this paper. Thesplitprimitive distributes el-ements of a list according to their category. Split is an importantoperation for data mapping and is used to build data structures,distribute work load, etc., in a massively parallel environment.Thegather/scatterprimitive performs fast, distributed data move-ment. Efficient data movement is critical to high performance onthe GPUs as suboptimal memory accesses can pay heavy penalties.The split we implement is a generalization of the binary split [Blel-loch 1990] and is implemented using the shared memory and theatomic operations available on them. The split performance scaleslogarithmically with the number of categories, linearly with the listlength, and linearly with the number of cores on the GPU. Thismakes it useful for applications that deal with large data sets. Wealso present a variant of split that partitions the indexes of records.This facilitates the use of the GPU as a coprocessor for split or sort,with the actual data movement handled separately. We can computethe split indexes for a list of 32 million records in 180 millisecondsfor a 32-bit key and in 800 ms for a 96-bit key. The instantaneouslocality of memory references play a critical role in data movementon the current GPU memory architectures. For scatter and gatherinvolving large records, we use collective data movement in whichmultiple threads cooperate on individual records to improve the in-stantaneous locality. The split, gather, and their combinations findmany applications and expect our primitives will be used by fu-ture GPU programmers. We show sorting of 16 million 128-byterecords in 379 milliseconds with 4-byte keys and in 556 ms with8-byte keys