Description:

Report of research results

The objective of this cluster is to identify the degree of similarity in

the patterns of computation that occur in the static and dynamic

dependence graph of programs and develop hardware and software

mechanisms that can exploit this similarity to improve performance and

power-efficiency.



This work will establish the limits of similarity in the computational

patterns in programs (similarity for brevity) while considering

several transformations that facilitate similarity. The

transformations will be applied to a program's static and/or dynamic

dependence graph to convert otherwise disimilar patterns to similar.



We then plan to develop efficient hardware and software methods for

implementing the various transformations and detecting similarity in

programs. Such mechanisms will enable the static and dynamic

identification of similarity which in turn may be useful to increase

performance by improving existing or developing new microarchitectural

mechanisms.



This is particularly useful for design space exploration,

where it is important to explore the right trade-off between the

capabilities of the hardware and the capabilities of the software that

will run on it. Therefore, a join development and analysis of both of

them together is mandatory to really reach the optimal performance

of applications.



New methodology allow a quick modification of an

architecture, and they also generate an accurate simulator. However,

having a compiler chain tuned and optimized to the new architecture is

still a long process, that compromises the design space exploration

process. The proposed approach is a step toward a more automatic

generation of a optimized code generation for new architectures.



This project will consider various applications of similarity, for

example compiler optimizatins, conditional branch prediction,

confidence estimation, caching, control-flow sequencing and

specialized-execution units for frequent patterns.

Nature:
Research cluster

Total funding:
Requested: € 4000
Granted: € 4000

Including fellowship funding:
Requested: € 0
Granted: € 0

Description of how the funding will be used:

one trip to involved site.

budget for eight people each with a single overnight accomodation.



each ticket 300 euro and 200 for lodging, food and ground transpo.



8x500 = 4000

Duration of the funding:
Requested: 6 month(s)
Granted: 6 month(s), starting on: Tue, January 1, 1980

Participating members:
DRACH Nathalie (CNRS) (--member--)
DURANTON Marc (NXP) (--member--)
FLAUTNER Krisztian (ARM) (--member--)
SAZEIDES Yiannakis (University of Cyprus) (--member--)
TEMAM Olivier (INRIA) (--member--)
VASSILIADIS Stamatis (Delft University of Technology) (--member--)

Other people collaborating:

zbigniew.chamski@philips.com

sami.yehia@arm.com