Author:
	Ervan Darnell

Title:                                                                          
        Automatic Software Cache Coherence through Vectorization
 
Abstract:
 
In large-scale shared-memory multiprocessors, memory latency is a
potential performance bottleneck since, with respect to each processor,
most memory is on the far side of the interconnection network.  Caches
are an accepted way to reduce the average memory latency that
processors observe; however,  there is no generally accepted hardware
solution to provide cache coherence for multiprocessors with multistage
interconnection networks.
 
In this poster we describe a software strategy for maintaining cache
coherence that relies on dependence analysis and a vectorization
algorithm to insert cache control directives. Software coherence
strategies offer scalability with current hardware.  Experiments on the
BBN TC2000 for a pair of numerical problems show that the run-time cost
of coherence using our strategy is less than that for previously
proposed software methods and suggest that it should compare favorably
with proposed hardware schemes that maintain strong coherence.