(University of Oregon, 1989-03-24) Meyer, David M.; Conery, John S.
In this paper, we present an architected approach to failure handling for independent
AND parallel logic programs. That is, the architecture presented here
represents its failure handling algorithm as a sequence of simple abstract machine instructions,
rather than as a built-in function. Information about data dependencies
is used by a compiler to generate special purpose fail routines on a clause-by-clause
basis. We also present two simple optimizations that further specialize the handling
of failures for each clause.
(University of Oregon, 1990-12-07) Conery, John S.
The Mayfly, a parallel processor being built at HP Labs in Palo Alto, has architectural
support for several import.ant. aspects of the OM virtual machine for parallel
logic programs. Each node has a coprocessor that is able to relieve the main processor
of a significant amount oft.he "housekeeping" work of memory management ,
task switching, and message handling. This paper describes how the coprocessor
implements kernel level functions in OM, with particular attention to the operations
that support task switching. The paper includes detailed timing data from
a program with interleaved parallel threads to show that while the main processor
is busy in one thread the coprocessor can effectively build the context for the next
thread .
