Dynamic partitioning strategies for automatic parallelization of code on Manycore architectures

• Grant number: 12/19997-2
• Support Opportunities: Scholarships abroad - Research Internship - Doctorate
• Start date: May 01, 2013
• End date: April 30, 2014
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Sandro Rigo
• Grantee: Cristianno Martins Vieira
• Supervisor: José Nelson Amaral
• Host Institution: Instituto de Computação (IC). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil
• Institution abroad: University of Alberta, Canada
• Associated to the scholarship: 11/09603-4 - Optimizations for Automatic Extraction of Parallelism Based on Decoupled Software Pipelining, BP.DR

Abstract

The recent advances on multicore architectures have created significant changes on computer systems. Essentially, the software industry is not ready for a massive development of portable multithreaded applications. Thus, the lack of tools to assist to make the most of multicore processors is a critical issue.Furthermore, it is not easy to perform the manual parallelization of a program. There are several problems that appear only on parallelized codes, such as race conditions and false sharing. These problems exponentially increase the complexity of programming these applications.This project aims to optimize the infrastructure built on LLVM as part of the tied thesis. The main objective of the thesis is achieve an adaptation of the Decoupled Software Pipelining technique (DSWP) applied to architectures with tens of cores. The focus in this project will be to resolve issues related to implementing a dynamic partitioning strategy for our task scheduler. Optimizations in communication between threads will also be an important part of this project. (AU)