Andrew M. Bradley
Andrew M. Bradley
@pbosler only relative error matters. Your point about condition number is the important one. An absolute "big" or "small" number doesn't mean much.
What Jim writes is true: a compiler bug in a specific Intel version affects the default min op.
We probably need to start doing things like [this](https://github.com/E3SM-Project/E3SM/blob/86baec1d2bf48ded4717ea522e48f613e8c2b4bb/components/homme/src/share/gllfvremap_mod.F90#L342) rather than turn off an optimization completely, i.e., regardless of compiler.
Excellent find, @mwarusz! I'm investigating another issue and have come across this same line as a problem. @bogensch it's possible that the GNU issues you see on PM are in...
@bartgol still relevant?
I'm going to do the modifications as part of the mimc-GPU work so that I can do one overall performance check for both sets of changes.
Luca, not sure about example 1. The output in example 2 appears to be used without qualification elsewhere in SHOC.
Seems unlikely. But we want to remove discontinuities caused by the limiters regardless.
Does this feature have an associated test? If not, I recommend adding one. It's up to you whether to run tests on machines. I personally like to run a good...
I just meant that I recommend that this PR include a test that can run in the nightlies. What form it takes is up to you.