
Programming for Performance
The goal of using multiprocessors is to obtain high performance. With a concrete
understanding of how the decomposition, assignment, and orchestration of a paral-
lel program are incorporated in the code that runs on the machine, we are ready to
examine the key factors that limit parallel performance and how they are addressed
in a wide range of problems. We will see how decisions made in different steps of the
programming process affect the run-time characteristics presented to the architec-
ture,
as well as how the characteristics of the architecture influence programming
decisions. Understanding programming technique ...