Lauer, Hugh C
Wills, Craig E
This project investigated how to implement implicit fine-grain parallelism in graphical dataflow programs at the function-level. We succeeded in developing a runtime model that accomplishes this by changing the content of each function's activation record, and changing its location from the stack to the heap. Simple atomic instructions are used for synchronization purposes, and a runtime library manages a thread pool that it uses to execute functions to minimize operating system calls. In addition, we describe how to implement this runtime model as an optimization pass within a compiler that would work on a well-defined C representation of graphical dataflow programs. Finally, we describe how this optimization pass might be implemented in GCC, which opens the door for future work.
Worcester Polytechnic Institute
Major Qualifying Project
All authors have granted to WPI a nonexclusive royalty-free license to distribute copies of the work, subject to other agreements. Copyright is held by the author or authors, with all rights reserved, unless otherwise noted.