frankenstein: fix some valgrind complaints
note that I don't know if these changes are semantically correct (I
doubt it), but this patch fixes the valgrind complaints I saw for
SPE9_CP and on Norne.
Also, this makes the timing of flow_ebos for SPE9 determinisic between
runs: without this, I got some random time steps fail in a given run
and in the next run a completely different set of timesteps
failed. Since this was on the same same machine, without any
recompiles and no changes to the deck or any other input parameters, I
initially attributed the behavior to cosmic rays ;)
@totto82: could you have a thorough look on this?
Okay: I just did a full Norne run with this. The number of convergence failures seems to be reduced significantly and the summary results still seem to match. as a consequence of the better convergence, performance significantly improved. on my machine:
> ./bin/flow_ebos ../opm-data/norne/NORNE_ATW2013.DATA
================ End of simulation ===============
Total time taken (seconds): 565.463
Solver time (seconds): 554
Overall Linearizations: 1253
Overall Newton Iterations: 916
Overall Linear Iterations: 14165
I guess that was the first sequential Norne run of any flow variant which was faster than 10 minutes. \o/
> I don't see why the tempV should give trouble, it seems to have the proper size.
yes, but according to valgrind that array was uninitialized for Norne. if it happened to be not equal to zero, it affected convergence?
I have tested SPE9_CP with maxCoeff() enabled. It seems to give "deterministic convergence results" On the other hand I am not able to run norne with this patch. Has there been some recent changes to master that is not accounted for?
I think this should be merge. But first the maxCoeff() should be enabled again.