This cluster aims to provide a breakthrough technology that will
provide the flexibility of processors, the efficiency of specialized
circuits, and minimal development effort. This novel technology
is based on the notion that each program contains, potentially large,
patterns of interdependent instructions that frequently recur. If
these patterns are executed as atomic instructions using a specialized
circuitry, they can consume far less energy and execute much faster
than traditional sequences of instructions. If these patterns can be
automatically identified, the specialized hardware accelerators
automatically derived for each pattern class, and the program easily
tuned to accommodate these hardware accelerators, then the
corresponding integrated system would be just as easy to use as a
standard processor.

This cluster extension aims to continue investigating the above and build on the work that started during a HiPEAC internship at ARM by Marios Kleanthous.

During the three month internship a technique to reduce pressure on
large data structures was developed. This technique enables
small structures to perform - as good - as a big structure. A small table is combined with a larger profile table. Entries in the two tables contain the same information, but, a profile table entry contain less (approximate) information. This inexact information is used to determine the impact of each profile entry on the performance. Knowing the importance of such entries allows smaller structures to be updated with the best possible subset of entries. This idea was submitted for patent by ARM and a paper is in preparation.

A HiPEAC cluster extension will be very conducive
towards fullfilling the project objectives.

Funding will be used to cover travel expenses between involved cites Cyprus and/or to attend HiPEAC cluster meetings.
6 trips @ 1500 = 9000

please note that the above amounts do not include overheads.

KLEANTHOUS Marios, University of Cyprus *PhD Student