MODELS FOR CONTROL UNIT SYNCHRONIZATION ON SHARED CONTROL ARCHITECTURES

Shared control is an architecture model that extends the SIMD model to support control parallelism. Under shared control, each processing el ement (PE) supports a local instruction stream and control for each PE is received by synchronizing with a shared control unit corresponding to the PE's current instruction. Because the control is shared and no t supplied on demand, the control units must be tightly synchronized t o allow the PEs to select among them. This synchronization is accompli shed by defining a fundamental cycle length for the control units-allo wing PEs to safely switch to a new control unit at the fundamental cyc le edge. In a naive implementation, the fundamental cycle length is go verned by the longest instruction length; other control units are padd ed with idle cycles to that length. Since the longest instruction cost is suffered by each instruction, the performance of the system in the presence of long instructions deteriorates. This paper presents and e valuates alternative models for synchronization that minimize the effe ct of long instructions on the performance. The implementation of thes e solutions is discussed and the improvement in performance is charact erized. (C) 1998 Academic Press, Inc.