Automatic Differentiation by program transformation uses static data-flow analysis to produce efficient code. This data-flow analysis must be adapted for parallel programs with Message-Passing communication. Starting from a context-sensitive and flow-sensitive data-flow analysis scheme initially devised for sequential codes, we extend this scheme for parallel codes. This extension is independent of the particular analysis and does not require a modification of the code's internal representation, i.e. the flow graph. This extension relies on an accurate matching of communication points, which can't be found automatically in general, and thus new user directives prove useful. © 2012 Springer-Verlag.
CITATION STYLE
Pascual, V., & Hascoët, L. (2012). Native handling of message-passing communication in data-flow analysis. In Lecture Notes in Computational Science and Engineering (Vol. 87 LNCSE, pp. 83–92). https://doi.org/10.1007/978-3-642-30023-3_8
Mendeley helps you to discover research relevant for your work.