摘要 |
An efficient technique for concurrently executing the iterations of an iterative construct is described. A parallel-processing computer is provided which is capable of successfully processing computationally intensive applications typically found in engineering and scientific applications. A technique is also described for interleaving the memory elements of a parallel-processing computer, and in particular one adapted for use in processing computationally intensive applications involving memory accesses at fixed strides. The memory elements (for example, cache sections) are highly accessible to the processors. There is also described a backplane-resident switch for selectively connecting any selected plurality of first sub-system buses to any selected plurality of second sub-system buses. |