tmLQCD Report inconsistent input for MPI dimension

Currently, when the code is compiled with e.g. 1-dimensional parallelization, the code will still accept NrXProcs, NrYProcs and NrZProcs, but just silently ignore them. Of course this will usually result in an error and the user will figure it out soon enough, but it might be good to actively check if the input makes sense given the way the code has been compiled and spit out an error as soon as things are being mixed up.

Mar 25 '13 10:03 deuzeman

Yes, I know it is like this, and it has always been. Of course, we could catch this as an input error, but...

Mar 25 '13 11:03 urbach

Well, it's a wishlist thing. But it seems like it wouldn't be all that much work to implement and it would remove one of these annoying 'wait, why isn't this just working' moments. Let's say I can have a look at this when I'm bored. :)

Mar 25 '13 11:03 deuzeman

A related issue. If the starting condition is defined as continue, one should also set InitialStoreCounter or the code will attempt to read 'conf' and probably crash. This could also be intercepted easily at the level of the input file reading with a more descriptive error message.

Mar 25 '13 18:03 deuzeman

As an extension of this problem it would be nice if the code reported that:

g_nprocs != nrtprocs * nrxprocs * nryprocs * nrzprocs

I just wasted an hour looking for a problem due to a stupid oversight... (I adjusted the parallelization parameters as if going from one rack to one nodeboard instead of just leaving them alone)

Mar 26 '13 10:03 kostrzewa