The Experts below are selected from a list of 12 Experts worldwide ranked by ideXlab platform
Przemyslaw Stpiczynski - One of the best experts on this subject based on the ideXlab platform.
-
a parallel non square Tiled Algorithm for solving a kind of bvp for second order odes
Parallel Processing and Applied Mathematics, 2009Co-Authors: Przemyslaw StpiczynskiAbstract:The aim of this paper is to show that a kind of boundary value problem for second-order ordinary differential equations which reduces to the problem of solving tridiagonal system of linear equations with almost Toeplitz structure can be efficiently solved on modern multicore architectures using a parallel Tiled Algorithm based on the divide and conquer approach for solving linear recurrence systems with constant coefficients and novel data formats for dense matrices.
Guangming Tan - One of the best experts on this subject based on the ideXlab platform.
-
tilespmv a Tiled Algorithm for sparse matrix vector multiplication on gpus
International Parallel and Distributed Processing Symposium, 2021Co-Authors: Yuyao Niu, Meichen Dong, Zhou Jin, Weifeng Liu, Guangming TanAbstract:With the extensive use of GPUs in modern supercomputers, accelerating sparse matrix-vector multiplication (SpMV) on GPUs received much attention in the last couple of decades. A number of techniques, such as increasing utilization of wide vector units, reducing load imbalance and selecting the best formats, have been developed. However, the 2D spatial sparsity structure has not been well exploited in the existing work for SpMV on GPUs. In this paper, we propose an efficient Tiled Algorithm called TileSpMV for optimizing SpMV on GPUs through exploiting 2D spatial structure of sparse matrices. We first implement seven warp-level SpMV methods for calculating sparse tiles stored in a variety of formats, and then design a selection method to find the best format and SpMV implementation for each tile. We also adaptively extract nonzeros in the very sparse tiles into a separate matrix to maximize the overall performance. The experimental results show that our method is faster than state-of-the-art SpMV methods such as Merge-SpMV, CSR5 and BSR in most matrices of the full SuiteSparse Matrix Collection and delivers up to 2.61x, 3.96x and 426.59x speedups, respectively.
Yuyao Niu - One of the best experts on this subject based on the ideXlab platform.
-
tilespmv a Tiled Algorithm for sparse matrix vector multiplication on gpus
International Parallel and Distributed Processing Symposium, 2021Co-Authors: Yuyao Niu, Meichen Dong, Zhou Jin, Weifeng Liu, Guangming TanAbstract:With the extensive use of GPUs in modern supercomputers, accelerating sparse matrix-vector multiplication (SpMV) on GPUs received much attention in the last couple of decades. A number of techniques, such as increasing utilization of wide vector units, reducing load imbalance and selecting the best formats, have been developed. However, the 2D spatial sparsity structure has not been well exploited in the existing work for SpMV on GPUs. In this paper, we propose an efficient Tiled Algorithm called TileSpMV for optimizing SpMV on GPUs through exploiting 2D spatial structure of sparse matrices. We first implement seven warp-level SpMV methods for calculating sparse tiles stored in a variety of formats, and then design a selection method to find the best format and SpMV implementation for each tile. We also adaptively extract nonzeros in the very sparse tiles into a separate matrix to maximize the overall performance. The experimental results show that our method is faster than state-of-the-art SpMV methods such as Merge-SpMV, CSR5 and BSR in most matrices of the full SuiteSparse Matrix Collection and delivers up to 2.61x, 3.96x and 426.59x speedups, respectively.
James C. Gibeaut - One of the best experts on this subject based on the ideXlab platform.
-
The index array approach and the dual Tiled similarity Algorithm for UAS hyper-spatial image processing
GeoInformatica, 2016Co-Authors: Yuxia Huang, James C. GibeautAbstract:Unmanned aerial systems (UAS) have been used as a robust tool for agricultural and environmental applications in recent years. Remote sensing systems based on UAS typically acquire massive hyper-spatial images in its short turnaround. This paper takes advantage of graphics processing unit (GPU) massive parallel computation in order to process the huge data timely and efficiently. More specifically, this paper presents an index array approach for lens distortion correction and geo-referencing. They are the two essential components in UAS hyper-spatial image processing. The index array approach is also capable of parallelizing image file I/O and the orthoimage generation. In addition, this paper presents the dual Tiled similarity Algorithm for the image co-registration. The index array approach and the dual Tiled similarity Algorithm were evaluated using two UAS remote sensing datasets of South Padre island shorelines. The results show that this index array approach was able to speed up at least 10 times the lens distortion correction and the geo-referencing relative to the central processing unit (CPU) computation. This dual Tiled Algorithm could provide 12 times speedup compared with the CPU similarity computation.
Meichen Dong - One of the best experts on this subject based on the ideXlab platform.
-
tilespmv a Tiled Algorithm for sparse matrix vector multiplication on gpus
International Parallel and Distributed Processing Symposium, 2021Co-Authors: Yuyao Niu, Meichen Dong, Zhou Jin, Weifeng Liu, Guangming TanAbstract:With the extensive use of GPUs in modern supercomputers, accelerating sparse matrix-vector multiplication (SpMV) on GPUs received much attention in the last couple of decades. A number of techniques, such as increasing utilization of wide vector units, reducing load imbalance and selecting the best formats, have been developed. However, the 2D spatial sparsity structure has not been well exploited in the existing work for SpMV on GPUs. In this paper, we propose an efficient Tiled Algorithm called TileSpMV for optimizing SpMV on GPUs through exploiting 2D spatial structure of sparse matrices. We first implement seven warp-level SpMV methods for calculating sparse tiles stored in a variety of formats, and then design a selection method to find the best format and SpMV implementation for each tile. We also adaptively extract nonzeros in the very sparse tiles into a separate matrix to maximize the overall performance. The experimental results show that our method is faster than state-of-the-art SpMV methods such as Merge-SpMV, CSR5 and BSR in most matrices of the full SuiteSparse Matrix Collection and delivers up to 2.61x, 3.96x and 426.59x speedups, respectively.