MQP Topics for 2011-2012
Professor Hugh C. Lauer
Department of Computer Science
For the academic year 2011-2012, I am offering MQP topics in the following area:–
Fine-Grain Parallelism
The general problem is how to exploit the current trend in microprocessors of many cores per chip and multiple threads per core. I am pursuing a particular approach based on parallelism at the level of a function call or basic block. A general overview of this approach can be found at
Research on Programming for a
Multi-Core Environment (July 2009)
.doc
format
.pdf format
A sketch of a graphical dataflow language to support this approach can be found at
Sketch of Graphical Dataflow
Programming Environment (July 2009)
.doc
format
.pdf format
The report of the 2009-2010 MQP team
on this topic can be found here.
The report of the 2010-2011 MQP project will be posted here when it is ready.
In 2011-2012, the following three avenues of investigation are possible (pptx
, pdf):–· Implement an interactive front-end for a graphical dataflow language. This will allow a programmer to “write” a program in the graphical language and to compile it into the stylized form of C defined in the 2010-2011 project.
· Modify the backend of the LLVM compiler system to implement the runtime model for this language. A previous attempt to implement a backend in gcc did not work because the gcc plugin system was not robust enough to be modified this way.
· Create a runtime library that executes in user-space and supports the runtime model generated by the backend of the compiler.
The ultimate goal is to port an unusual application to the graphical dataflow environment and study how much fine-grained parallelism can be found in the application. A robotics application is suggested.
If you are interested, please contact me to brainstorm about possibilities.