Luc Berger

``` -D Tpetra_ENABLE_CUDA:BOOL=ON \ -D Kokkos_ENABLE_CUDA:BOOL=ON \ -D Kokkos_ENABLE_CUDA_LAMBDA:BOOL=ON \ -D Tpetra_INST_CUDA:BOOL=ON \ -D TPL_ENABLE_CUDA:BOOL=ON \ ``` From the scripts posted above it's a cuda build

@sayerhs a quick question regarding the build time, can you see if there are weak symbols in the library and if so how many? These indicate that some function or...

@rcknaus thanks for the explanation. Would it be possible to run all the boundary condition algorithms (for momentum, continuity, etc...) on the "master graph" so that we have all connectivity...

@sayerhs would it then make sense to only use these two graphs and remove the others? It would simplify the code base and probably would not cost much more computationally....

@sayerhs usually I am happy to trade a little more flop if I can save some communication. In this case the overall increase of entries in the matrix is modest...

By the way if you call `buildElemToNodeGraph` and `buildFaceElemToNodeGraph` you do end up with the extra zero entries anyway right?

Sure, I'll look into a reproducer. There is an additional issue that is not addressed by the fix above related to the update of the results which require the size...

okay, fixed my testing issues, I think this might well build and pass now. I need to check where things are implemented, I believe some structs need to move to...

Adding @fnrizzi and @mzuzek for some reason can't add you as reviewers but you should have a look at what is done here.

Not too sure what happened to the gcc930 test, marking this as retest, hopefully that will go smoothly this time.