Title :
Why value prediction is limited?
Author :
Markovski, J. ; Gusev, M.
Author_Institution :
Fac. of Natural Sci. & Math., Ss. Cyril & Methodius Univ., Skopje
Abstract :
True data dependences are considered as fundamental limits of extractable instruction level parallelism (ILP). Principles of value locality and value predictability offer a solution to overcome these problems by supporting various value prediction techniques. However, despite the numerous reports on the benefits of value prediction there is still no commercial processor that employs this technique. In this paper we overview the main reasons and discuss why value prediction is not implemented in contemporary processors. After this analysis we discuss its further possible application
Keywords :
data flow computing; instruction sets; multiprocessing systems; parallel architectures; parallel programming; ILP; contemporary processors; extractable instruction level parallelism; true data dependences; value locality; value prediction techniques; Cache memory; Data mining; Delay; Hardware; Information technology; Mathematics; Microprocessors; Pipelines; Predictive models; Registers;
Conference_Titel :
Information Technology Interfaces, 2004. 26th International Conference on
Conference_Location :
Cavtat
Print_ISBN :
953-96769-9-1
