Languages and Compiling for Multithreaded Architectures
Presenter: Gary M. Zoppetti, University of Deleware
Multithreaded architectures offer the ability to tolerate latencies by
spawning new threads during long latency operations. The ability to
mask these latencies makes these machines ideal for running irregular
applications. However, programming these architectures is not an easy
task. Explicitly specifying thread boundaries and synchronization
between threads is tedious and error-prone.
A non-preemptive multithreaded architecture model called EARTH
(Efficient Architecture for Running Threads) has been developed by
Guang Gao's group at the University of Delaware and McGill University.
Two C dialects can be used to program this architecture. It is
desirable to program in the higher-level parallel C dialect which can
then be compiled into efficient, lower-level explicitly threaded code.
The compiler infrastructure used is McCAT (McGill Compiler
Architecture Testbed), developed by Laurie Hendren's group at McGill.
At the University of Delaware, we are currently working on developing
a set of novel optimization techniques for this architecture. In this
poster, background information will be provided on the architecture
model, languages, and the compiler testbed. Some novel optimizing
transformations will also be presented.
Please revisit this page later as poster abstract are begin collected.