Automating Run-Time Reordering Transformations with the Sparse Polyhedral Model
Michelle Strout (Colorado State University)
COMPUTER SYSTEMS SEMINARDATE: 2012-07-12
TIME: 16:00:00 - 17:00:00
LOCATION: CSIT Seminar Room, N101
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
In the age of multicore, turning data reuse into data locality is a necessity to improve performance in memory bandwidth bound computations such as those that iterate over data structures such as sparse matrices and unstructured meshes. To exploit parallelism and improve locality in such applications, prior work has developed a number of run-time reorderings to transform the iterations and/or data in a specific loop. A Run-Time Reordering Transformation (RTRTs) is implemented with inspector code, which traverses the index arrays (i.e. array B in ABi), generates the appropriate reordering transformation, and reorders the data and/or index arrays with the goal of improving data locality or exploiting parallelism. In addition, a transformed version of the original loop, the executor, is also generated to use the reordered data or computation that results from the inspector. Unlike loop and data transformations for loops with affine bounds and array references, RTRTs do not have frameworks for specifying a sequence of transformations and the automatic code generation from such a specification.
In this talk, I show how we can represent computations such as molecular dynamics simulations and computations over meshes in the Sparse Polyhedral Framework and how to leverage the SPF framework to generate inspectors and executors to perform run-time reordering transformations. Additionally I show how an individual inspector algorithm can be proven correct and pose the question of automating such proofs.
BIO:
Michelle Strout is an associate professor in the computer science
department at Colorado State University. From August 2003 through June
2005, she was an Enrico Fermi Postdoctoral Scholar at Argonne National
Laboratory and a Research Associate at the University of Chicago. Her
main research area is high performance computing and her research
interests include compilers and run-time systems, scientific computing,
and software engineering. In 2008, Michelle received a CAREER Award
from the National Science Foundation for her research in parallelization
techniques for irregular applications, such as molecular dynamics
simulations. In 2010, she received a DOE Early Career award to fund her
research in separating the specification of scientific computing
applications from the specification of implementation details such as
how to parallelize such computations.
