Points of interest

• Moving MADNESS functions to/from disk

The source is here.

Background

MADNESS functions are parallel objects distributed across the whole machine. If you try to store a function into a sequential archive, only the part local to the calling process will be stored. I.e., you need to use a parallel archive. The parallel archive has an adjustable number of proceses that actually perform I/O for effiency. If your functions are few and small, the default of one writer is OK. For large (or many) functions and I/O is a bottleneck, then increase the number of I/O nodes.

Implementation

Three different functions are created, written to disk, read back, and finally compared with the originals.