topic Our implementation of in Intel® Fortran Compiler

hybrid coarray+MPI supported? any examples?

AShte — Mon, 11 Nov 2013 13:30:28 GMT

I'm trying to use MPI/IO to speed-up coarray IO operations. This does work fine with Cray. Can I get this to work somehow with 14.0.0 20130728? If yes, are there any examples? How much of this tutorial will have to be changed to compile and run a coarray+MPI program: http://software.intel.com/en-us/articles/distributed-memory-coarray-fortran-with-the-intel-fortran-compiler-for-linux-essential ? Thanks Anton

This is not something we have

Steven_L_Intel1 — Mon, 11 Nov 2013 16:52:49 GMT

This is not something we have tested and don't officially support. We don't have examples at this time. There's nothing I know of that would prevent this combination from working, but Cray doesn't use MPI for coarrays so with our implementation you're using MPI for both and there could be some unintended interaction.

On a somewhat related note,

Izaak_Beekman — Mon, 11 Nov 2013 17:46:25 GMT

On a somewhat related note, what about combining asynchronous IO and OpenMP or MPI? I suspect asynchronous IO might spawn new threads in some implementations.... Or it delegates to the OS? I don't really understand the behind the scenes stuff relative to asynchronous IO, and asynchronous features of the language specification when paired with other existing parallel programming models.

Our implementation of

Steven_L_Intel1 — Mon, 11 Nov 2013 20:03:08 GMT

Our implementation of asynchronous I/O spawns threads using the operating system's thread API (pthreads on Linux). It should coexist fine with OpenMP and MPI (MPI doesn't use threads.) The threads are just waiting for I/O completion.