Summary: | 碩士 === 國立海洋大學 === 電子工程學系 === 81 === This thesis is about the parallelization of matrix
multiplication on various parallel architectures. We have
configured these parallel architectures from a transputer
network, including a linear array, a mesh, a cylindrical array,
a two-layered mesh, and a cube. The parallelism as- sociated
with each architecture is then studied by decom- posing the
running time of each multiplication method into the computation
and communication parts. The second aspects of this thesis is
the study of para- llel programming using the Occam language.
Major results include the basic program skeletons for each
parallel arch- itecture, programming techniques for input/
output data dis- tribution, assignments of channels,
manipulation of time de- lay, deadlock avoidance, etc.
|