Sean Miller

Ah, sounds like a perfect alternative then. I guess this leads into our previous email discussion where the performance of "reuse: type" = "tP" is a bit slow (13-17x slower...

@csiefer2 The Tpetra changes are there because the ProfilingRegion can enable "groups" of profiling regions. If a group is not enabled, then the ProfilingRegion class will still pop the top...

@csiefer2 Ah, take a look at the `timeit` variable as part of the `ProfilingRegion (const char name[], const char group[])` constructor. It will turn off the profiling region if the...

@cgcgcg Will do, I'll remove the MueLu changes.

@cgcgcg MueLu changes have been removed.

This will need to wait until https://github.com/trilinos/Trilinos/pull/11087. Tests pass up through MueLu, but I haven't tested against Panzer.

@lucbv @jczhang07 Hi guys, is there an update on this? I have some overlapping changes that I would like to put in to help support rocsparse in Tpetra, but it...

@lucbv I had problems in my implementation with adding the raw descriptor to the CrsMatrix. In an application like Tpetra, these KokkosSparse::CrsMatrix objects are generated on the fly (i.e. not...

On the Trilinos, these fixes are being incorporated in https://github.com/trilinos/Trilinos/pull/12681 (holdover until a new KK snapshot goes in), though there are a couple other ROCm fixes going in that PR...

@lucbv are you up for reviewing this? It adds some feature support for SpGEMM with rocsparse.