Reflective-Folding Data Mapping onto a Parallel Mesh of Processors

Existing techniques for mapping image data onto the processors of a SIMD machine are suitable for algorithms which operate on data of spatially regular patterns. However, many algorithms for image analysis deal with data on irregular and/or dynamically varying regions. We present a new data mapping technique called reflective-folding which can cater effectively for these situations. An example to show how the technique is applied to a practical problem is also given. Technical areas: Image analysis, Parallel Computing. 1 Introduction Efficient parallel image processing on SIMD machines greatly relies on an effective mapping of the image data onto the processors. There exists a number of commonly used data mapping techniques which can be used for mapping data onto a 1D or 2D processor array (see for example [1, 2, 4]). However these techniques do not cater adequately for algorithms which operate on irregular and/or dynamically varying regions. Some examples of such algorithms are conto..