Motivation
Griffin is moving forward with the Kokkos/libMesh based GPU implementation. In #31981 I outline a plan for an MFEM based solid mechanics which would be leveraged by BISON (eventually) among others. For app drivers within NEAMS, for instance, we will want to be able to (efficiently) communicate data between these different apps
Design
Don't know yet for sure. We could represent the MFEM data on a libMesh mesh which would enable re-use of our existing General field transfers. I believe this is how some old UKAEA implementations did transfers IIRC but that @alexanderianblair and crew ultimately went a different route with the current set of MFEM transfers (which I believe are only really designed for "between MFEM apps"?). I even think we have an issue discussing some of this but I can't find it at the moment.
I don't know whether it would be possible to keep data on device but it would sure be nice.
Impact
Allow mixing of MFEM and Kokkos/libMesh applications. It wouldn't really make sense within NEAMS to take apps down different GPU backends if we can't do multiphysics. This would be a no-go point
Motivation
Griffin is moving forward with the Kokkos/libMesh based GPU implementation. In #31981 I outline a plan for an MFEM based solid mechanics which would be leveraged by BISON (eventually) among others. For app drivers within NEAMS, for instance, we will want to be able to (efficiently) communicate data between these different apps
Design
Don't know yet for sure. We could represent the MFEM data on a libMesh mesh which would enable re-use of our existing General field transfers. I believe this is how some old UKAEA implementations did transfers IIRC but that @alexanderianblair and crew ultimately went a different route with the current set of MFEM transfers (which I believe are only really designed for "between MFEM apps"?). I even think we have an issue discussing some of this but I can't find it at the moment.
I don't know whether it would be possible to keep data on device but it would sure be nice.
Impact
Allow mixing of MFEM and Kokkos/libMesh applications. It wouldn't really make sense within NEAMS to take apps down different GPU backends if we can't do multiphysics. This would be a no-go point