Max Katz

Note: #998 partially addressed the situation because I believe we observed a case in Castro where building the subcommunicator was unnecessary. However I am not sure if that addressed the...

Yeah, I think we should just get the operation happening on the GPU.

Is it true that your code spends the majority of its time in FillBoundary even within a single node, or is that statement only true at large scale? If so,...

This is one where the multi-node behavior may be significantly different than the intra-node behavior. It is also the case that different MPI implementations are different with respect to CUDA-aware...

Yes.

> is there a reason to do aprox13, but not aprox19 and aprox21? Tabulation hasn't been implemented yet in those networks. > Also, are the tables compatible with #56? Yes.

Another angle here would be to retry an individual burning step with the other Jacobian. It is probably easier to implement the first idea because we have already done something...

The prerequisite for either idea is that the interfaces in the burner going down to dvjac need to decide which Jacobian to use via function arguments rather than reading the...

Should be straightforward because we can just put the Jacobian choice into the `dvode_t`. There may be an even easier way though: to the extent that the comments in `vode_dvnlsd.H`...

One possible version of this is: ``` $ git diff diff --git a/integration/VODE/vode_dvjac.H b/integration/VODE/vode_dvjac.H index 04345a94..89f7a3dc 100644 --- a/integration/VODE/vode_dvjac.H +++ b/integration/VODE/vode_dvjac.H @@ -67,7 +67,7 @@ void dvjac (IArray1D& pivot, int&...