        |
 |
|
|||
|
|
Silvius Rus - Research
Why do memory reference pattern analysis?
Several compiler analyses benefit from knowing the memory reference
pattern:
parallelization,
vectorization, checkpointing, locality enhancement.
Analysis of memory references is increasing in
importance with the latest attempts to address the processor-memory bottleneck:
processor-in-memory and multiple cache levels.
Is it a hard problem?
Automatic analysis of memory reference patterns of dynamic applications
is complicated by the fact that the shape and size of the memory used
by the same piece of code varies for different input values.
Optimization based on such analysis cannot be performed until the
needed values become available, i.e. during execution.
What has been done before?
Although several analysis methods based on instrumentation of all memory
load/store operations have been proposed, they usually incur too much overhead.
The overhead of run-time techniques is largely due to redundancy.
Most of this redundancy comes from the fact that partial results of
inconclusive compile-time analysis are discarded. Such symbolic partial
results are reconstructed at run-time for each dynamic instance at a high cost.
What do we do?
We propose a new type of analysis, Hybrid Analysis of Memory Reference
Patterns, in which partial symbolic results are
used to extract lightweight run-time tests
so that the analysis can continue during code execution with
little overhead.
You can click here for a detailed description of Hybrid Analysis, or read the papers listed below.
Silvius Rus, Lawrence Rauchwerger, "Hybrid Dependence Analysis for Automatic Parallelization," Technical Report, TR05-013, Parasol Laboratory, Department of Computer Science, Texas A&M University, Nov 2005.
Technical Report(ps, pdf, ppt, abstract)
Silvius Rus, Lawrence Rauchwerger, Jay Hoeflinger, "Hybrid Analysis: Static & Dynamic Memory Reference Analysis," International Journal of Parallel Programming, 31(4):251-283, Aug 2003. Also, In Proc. ACM Int. Conf.
Supercomputing (ICS), pp. 274-284, New York City, Jun 2002. Also, Technical Report, TR02-002, Parasol Laboratory, Department of Computer Science, Texas A&M University, Jan 2002.
Journal(ppt, abstract) Proceedings(ps, pdf, ppt, abstract) Technical Report(ps, pdf, ppt, abstract)
Parasol Home | Research | People | General info | Seminars | Resources Parasol Lab, 301 Harvey R. Bright Bldg, 3112 TAMU, College Station, TX 77843-3112 Contact Webmaster Phone 979.458.0722 Fax 979.458.0718 
Department of Computer Science | Dwight Look College of Engineering | Texas A&M University Privacy statement: Computer Science Engineering TAMU |