This chapter describes the problems met while extending ADIFOR 3.0 to handle programs coming from the MPI datatypes. To give a complete picture of the design space, this chapter first explains why creating a differentiated interface for NIPI is preferable to differentiating NIPI source itself. Next, we identify the two main problems that MPI datatypes present and present two potential solution techniques. In conclusion, we give some preliminary implementation results.
CITATION STYLE
Carle, A., & Fagan, M. (2002). Automatically Differentiating MPI-1 Datatypes: The Complete Story. In Automatic Differentiation of Algorithms (pp. 215–222). Springer New York. https://doi.org/10.1007/978-1-4613-0075-5_25
Mendeley helps you to discover research relevant for your work.