On Fri, 2008-05-16 at 14:51 -0400, Richard Treumann wrote:

MPI_NO_REDUCTION_DETERMINISM

The MPI standard either requires or strongly urges that an MPI_REDUCE/MPI_ALLREDUCE give exactly the same answer every time. There are applications that do not need a "same answer" promise as long as they can assume they will get a "correct" answer. Maybe they can be provided a faster reduction algorithm.

This makes zero sense as a hint and would make the standard contradictory if adopted. I've already made it clear I would prefer assertions and this option would be useful in that context, if hints turn out the be chosen over assertions then this flag should be dropped. Clarification of the standard over the results of floating point reduction results would probably be a good thing anyway, for example is it OK for the different communicators of the same size but with different process layouts to give different answers given the same input data? I suspect the standard forbids this although without it it's probably impossible to do any shared memory optimisations. Ashley Pittman.