|Title||:||Unique Worker model for OpenMP|
|Speaker||:||Raghesh A (IITM)|
|Details||:||Tue, 12 May, 2015 2:00 PM @ BSB 361|
|Abstract:||:||In OpenMP, because of the underlying efficient `team of workers' model, each worker is given a chunk of tasks (iterations of a parallel-for-loop, or sections in a parallel-sections block), and a barrier construct is used to synchronize the workers (not the tasks). Naturally, the practitioners are discouraged from invoking barriers in these tasks; as otherwise, depending on the number of workers the behavior of the program can vary. Such a restrictive practice can adversely impact programmability and program readability. To overcome such a restrictive practice, inspired from the more intuitive interaction of tasks and barriers found in newer task parallel languages like X10, HJ, Chapel and so on, we present an extension to OpenMP (called UW-OpenMP).
UW-OpenMP gives the programmer an impression that each parallel task has been assigned a unique worker, and importantly these parallel tasks can be synchronized using a barrier construct. Consequently, the semantics of the programs (using parallel-for-loops, sections and barriers) remains independent of the actual number of worker threads, at runtime. We argue that such a scheme allows the programmer to conveniently think and express his/her logic in parallel. We propose a source to source transformation scheme to translate UW-OpenMP C programs to equivalent OpenMP C programs that are guaranteed to not invoke barriers in any task. We have implemented our proposed translation scheme in the ROSE compiler framework. Our preliminary evaluation shows that the proposed extension leads to programs that are concise and arguably easier to understand. Importantly, the efficiency resulting from the `team of workers' model is not compromised.