We have recently completed an experimental compiler for SML which exploits parallelism at sites of higher-order function use through algorithmic skeletons. The compiler is novel in:
+ accepting full pure functional bare and derived SML
+ exploiting arbitrary depth nesting of HOFs as nested skeletons
+ generating native code for 6 MIMD architectures with consistent
cross-platform behaviour
+ using proof planning to synthesise HOFs in programs that lack them
The compiler design and development is presented, and the performance of moderately convincing exemplars is discussed.