by Nalini Vasudevan and Stephen Edwards (Columbia University, USA)
Summary of reviewer comments:
Stated simply, the authors argue that compilers should generate deterministic code from parallel programs. The idea is to transform multi-threaded programs with non-deterministic parts into fully-deterministic programs using barrier-insertions.
Even though this paper is technically thin and may not reflect the level of understanding that the community has about the problem, the presentation may nonetheless stimulate a productive discussion. For example, is is possible to automatically synchronize a program sufficiently to avoid non-determinism while also avoiding-deadlock? Is it likely that determinizing algorithms may not be robust against program evolution? The programmer may slightly extend the program and be surprised that that the determinizer now picked an execution with a different behavior, one that is still correct but different from the one selected by the determinizer in the previous version of the program.
Showing posts with label determinizing compiler. Show all posts
Showing posts with label determinizing compiler. Show all posts
Saturday, June 13, 2009
Subscribe to:
Posts (Atom)