Efficient Tiling for an ODE Discrete Integration Program: Redundant Tasks Instead of Trapezoidal Shaped-Tiles
نویسندگان
چکیده
In this paper, we present an efficient and simple solution to the parallelization of discrete integration programs of ordinary differential equations (ODE). The main technique used is known as loop tiling. To avoid the overhead due to code complexity and border effects, we introduce redundant tasks and we use non parallelepiped tiles. Thanks both to cache reuse ( 4:3) and coarse granularity ( 24:5) , the speedup using 25 processors over the non-tiled sequential implementation is larger than 106. We also present the draft of a fuzzy methodology to optimize the tile size and we illustrate it using real measurements for the communication cost and the execution time. In particular, we observe that the model of communication latencies over a Myrinet network is not as simple as is usually reported.
منابع مشابه
Wang Tiles for Image and Texture Generation
This paper discusses the use of Wang Tiles for image and texture generation as originally presented in Cohen et al. , 2003. Creating large expanses of nonperiodic texture, point distributions or geometry can be efficiently done at runtime by constructing a tiling of the plane using a small set of Wang tiles. Wang Tiles are squares whose edges are each assigned a color. The method involves filli...
متن کاملTask Ordering in Linear Tiles
In this report we address the issue of loop tiling to minimize the completion time of the loop when executed on multicomputers We remove the restriction of atomicity of tiles and internal parallelism within tiles is exploited by overlapping computation with communication The e ectiveness of tiling is then critically dependent on the execution order of tasks within a tile In this paper we presen...
متن کاملEvery Tiling of the First Quadrant by Ribbon L n-Ominoes Follows the Rectangular Pattern
Let n 4 ≥ and let n be the set of four ribbon L-shaped n-ominoes. We study tiling problems for regions in a square lattice by n . Our main result shows a remarkable property of this set of tiles: any tiling of the first quadrant by n , n even, reduces to a tiling by n 2× and n 2 × rectangles, each rectangle being covered by two ribbon L-shaped n-ominoes. An application of our result is th...
متن کاملClass Notes Tile-Based Methods for Interactive Applications
Many complex signals, such as point distributions and textures, cannot efficiently be synthesized and stored. In this work we present tile-based methods to solve this problem. Instead of synthesizing a complex signal when needed, the signal is synthesized on forehand over a small set of tiles. Arbitrary large amounts of that signal can then efficiently be generated when needed by generating a s...
متن کاملOptimal Task Scheduling to Minimize Inter-Tile Latencies
This work addresses the issue of exploiting intra-tile parallelism by overlapping communication with computation removing the restriction of atomicity of tiles. The effectiveness of tiling is then critically dependent on the execution order of tasks within a tile. In this paper we present a theoretical framework based on equivalence classes that provides an optimal task ordering under assumptio...
متن کامل