An extended multi-access memory system for SIMD architecture

In order to satisfy the performance improvement of computing speed, many researchers have been studying in areas of both processor and memory system architectures. In the area of memory system, the linear skewing scheme has been known as a suitable one for the single instrution multiple data stream (SIMD) architecture. The scheme maps the data element located at coordinates (i, j) in an M ×N data array to memory modules (ai + bj)%m, where a and b are constants and can access data elements within a block, a row, or a column subarray simultaneously without conflicts if the number of memory modules is a prime number greater than the number of data elements of within the subarray. Although these functionalities of the linear skewing scheme, it needs more memory cells than M × N data array an application uses. In this paper, a memory system is proposed to make sufficient use of unused memory cells as a local memory system for processing elements in SIMD architecture.