Automatically Parallelizing Legacy Binary Code for Multicore Architectures
Abstract
Legacy codes must be adapted to multicore in order to sustain the current rate of progress made in science and commerce. The goal of this seedling was to study the suitability of particular automatic parallelization techniques for parallelizing legacy codes. We found that n-gram (i.e., repeated instruction sequences) parallelism does exist, but it is not easily leveraged. We found parallelism in many codes previously thought to be inherently sequential. We believe that efforts that consider input data can help liberate high-order parallelism from almost all such "sequential" codes.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2009
- Accession Number
- ADA514449
Entities
People
- David August
- Michael Locasto
- Sal Stolfo
- Simha Sethumadhavan
Organizations
- Princeton University