On Extracting Course-Grained Function Parallelism from C Programs

Li, Chien-Wei

Permalink

https://hdl.handle.net/2142/11174 Copy

Description

Title

On Extracting Course-Grained Function Parallelism from C Programs

Author(s)

Li, Chien-Wei

Issue Date

2006-05

Keyword(s)

computer science

Abstract

To efficiently utilize the emerging heterogeneous multi-core architecture, it is essential to exploit the inherent coarse-grained parallelism in applications. In addition to data parallelism, applications like telecommunication, multimedia, and gaming can also benefit from the exploitation of coarse-grained function parallelism. To exploit coarse-grained function parallelism, the common wisdom is to rely on programmers to explicitly express the coarse-grained data-flow between coarse-grained functions using data-flow or streaming languages. This research is set to explore another approach to exploiting coarse-grained function parallelism, that is to rely on compiler to extract coarse-grained data-flow from imperative programs. We believe imperative languages and the von Neumann programming model will still be the dominating programming languages programming model in the future. This dissertation discusses the design and implementation of a memory data-flow analysis system which extracts coarse-grained data-flow from C programs. The memory data-flow analysis system partitions a C program into a hierarchy of program regions. It then traverses the program region hierarchy from bottom up, summarizing the exposed memory access patterns for each program region, meanwhile deriving a conservative producer-consumer relations between program regions. An ensuing top-down traversal of the program region hierarchy will refine the producer-consumer relations by pruning spurious relations. We built an in-lining based prototype of the memory data-flow analysis system on top of the IMPACT compiler infrastructure. We applied the prototype to analyze the memory data-flow of several MediaBench programs. The experiment results showed that while the prototype performed reasonably well for the tested programs, the in-lining based implementation may not efficient for larger programs. Also, there is still room in improving the effectiveness of the memory data-flow analysis system. We did root cause analysis for the inaccuracy in the memory data-flow analysis results, which provided us insights on how to improve the memory data-flow analysis system in the future.

Type of Resource

text

Permalink

http://hdl.handle.net/2142/11174

Copyright and License Information

You are granted permission for the non-commercial reproduction, distribution, display, and performance of this technical report in any format, BUT this permission is only for a period of 45 (forty-five) days from the most recent time that you verified that this technical report is still available from the University of Illinois at Urbana-Champaign Computer Science Department under terms that include this permission. All other rights are reserved by the author(s).

Owning Collections