HARDBACK SOLUTION TO ACCELERATE MULTIMEDIA COMPUTATION THROUGH MGP IN CMP

Jimcy Babu, Kavitha .V, K.V.Ramakrishnan

Abstract: Current multimedia applications are becoming more and more complex; thereby it increases workload for General purpose processors (GPPs). This resulted in the advent of Chip multiprocessors (CMPs). In most of the cases CMPs were not fully utilized. Hence, how to achieve the contemporary multimedia requirement like speed i.e. completion time, by using CMP became a question. In this project work a solution had been put forward to tackle the crisis. The idea behind the solution is the proper exploitation of the parallelism thereby accelerating the multimedia computation in CMPs. To achieve the less completion time, four levels of parallelism has been considered i.e. Data Level, Thread Level, Instruction Level and Memory Level called as Multi-Grain Parallelism (MGP). POSIX thread concept is used to implement Data Level, Thread Level, Instruction Level and for Memory Level, pre-fetch concept. To run, Linux based platform is required. Here UBUNTU 12.02 version is used. This project also performs a comparison between serial computation and parallel computation at different levels, based on completion time. The experimental results show that parallel computation consumes less time as compared to that of serial computation. Thereby, making it viable for multimedia applications

Keywords: Chip Multiprocessor (CMP), Data Level Parallelism (DLP), Thread Level Parallelism (TLP), Instruction Level Parallelism (ILP), Memory Level Parallelism (MLP), POSIX thread (p-thread), Pre-fetch.

DOI: https://doi.org/10.15623/ijret.2014.0307079