2D matrix multiplication on a 3D systolic array
نویسنده
چکیده
The introduction of systolic arrays in the late 1970s had an enormous impact on the area of special purpose computing. However, most of the work so far has been done with onedimensional and two-dimensional (2D) systolic arrays. Recent advances in three-<limensional VLSI (3D VLSI) and 3D packaging of2D VLSI components, has made the idea of 3D systolic arrays feasible in the near future. In this paper we introduce one algorithm for 2D matrix multiplication, using a 3D systolic array. We analyze advantages and disadvantages of 3D systolic arrays in the context of the analysis algorithm. The analytical work is combined with examples and discussions of relevant details.
منابع مشابه
Triple-Matrix Product-Based 2D Systolic Implementation of Discrete Fourier Transform
Realization of N-point Discrete Fourier Transform (DFT) using one-dimensional or two-dimensional systolic array structures have been developed for power of two DFT sizes. DFT algorithm, which can be represented as a triple -matrix product, can be realized by decomposing N into smaller lengths. Triple matrix product form of representation enables to map the Npoint DFT on a 2-D systolic array. In...
متن کاملCommunication-Optimal Parallel 2.5D Matrix Multiplication and LU Factorization Algorithms
Extra memory allows parallel matrix multiplication to be done with asymptotically less communication than Cannon’s algorithm and be faster in practice. “3D” algorithms arrange the p processors in a 3D array, and store redundant copies of the matrices on each of p layers. ‘2D” algorithms such as Cannon’s algorithm store a single copy of the matrices on a 2D array of processors. We generalize the...
متن کاملNew Parallel Matrix Multiplication Algorithms for Wormhole-Routed All-Port 2D/3D Torus Networks
New matrix multiplication algorithms are proposed for massively parallel supercomputers with 2D/3D, all-port torus interconnection networks. The proposed algorithms are based on the traditional row-by-column multiplication matrix product model and employ a special routing pattern for better scalability. They compare favorably to the variants of Cannon’s and DNS algorithms since they allow matri...
متن کاملFilomat 17 (2003), 135–141 Matrix Multiplication on Bidirectional Linear Systolic Arrays
This paper addresses the problem of rectangular matrix multiplication on bidirectional linear systolic arrays (SAs). We analyze all bidirectional linear SAs in terms of efficiency. We conclude that the efficiency depends on the relation between the loop boundaries in the systolic algorithm (i.e. matrix dimensions). We point out which SA is the best choice depending on the relation between matri...
متن کاملTowards a Multi-array Architecture for Accelerating Large-scale Matrix Multiplication on FPGAs
Large-scale floating-point matrix multiplication is a fundamental kernel in many scientific and engineering applications. Most existing work only focus on accelerating matrix multiplication on FPGA by adopting a linear systolic array. This paper towards the extension of this architecture by proposing a scalable and highly configurable multi-array architecture. In addition, we propose a work-ste...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2003