Research on the Multiscalar execution model was introduced in the early
1990's, recognizing the inadequacies of relying entirely on Instruction
Level Parallelism (ILP). The goal was to augment the parallelism
bridgehead established by ILP with the ground forces of Thread-Level
Parallelism (TLP), a coarser form of parallelism. Many studies on
parallelism indeed confirm the significant performance potential of
executing multiple threads of a program in parallel. The difficulties
that have been plaguing the parallelization of ordinary, non-numeric
programs for decades have been complex flows of control and ambiguous
data dependences through memory. The breakthrough provided by the
Multiscalar execution model was the use of speculative threads along
with data speculation. Speculative threads allow sequential programs to
be parallelized, and data speculation allow threads to be executed in
parallel even if they are likely to have data dependences.
Multiscalar Processors presents a comprehensive treatment of the
basic principles of Multiscalar execution, and advanced techniques for
implementing the Multiscalar concepts. Special emphasis is placed on
highlighting the major challenges involved in Multiscalar processing.
This book is organized into nine chapters, and provides an excellent
synopsis of a large body of research carried out on multiscalar
processors in the last decade. It starts with technology trends that
provide an impetus to the development of multiscalar processors and
shape the development of future processors. The work ends with a review
of the recent developments related to multiscalar processors.
Multiscalar Processors is designed to meet the needs of a
professional audience composed of designers and programmers of
next-generation processors and researchers in computer architecture and
engineering. This book is also suitable as a secondary text for graduate
level students in computer science and engineering.
