EAMxx: expand horiz remap in IO to allow for pg2 to structured latxlon grids

[mahf708]

as title

[bartgol]

Are you saying you want the remap to be done online, without the need for a map file? Or are you asking for something deeper, namely that the lat-lon dim pair be partitioned "unstructuredly" during IO (i.e., not partitioning just "lat" or just "lon", but the full sed of lat-lon pairs)?

The former is doable with current infrastructure, while the latter requires some work. Currently, we assume that only ONE dimension in the stream is partitioned. For lat-lon output, we'd need to partition 2 dimension, and it would not simply be the cartesian product of two partitioning, but rather the full partitioning of a 2d index set. It's not out of reach, but it does require some rework of our internal IO infrastructure.

[mahf708]

Given a map file that goes from ne30pg2->latxlon (180x360), I want SCREAM IO to get me the output for T_mid not in (time, ncol, lev) but in (time, lat, lon, lev)

[bartgol]

Yeah, that needs either a whole rethink of grids (in terms of decomposition) or a hack to "view" <COL,LEV>(10,8) as <LAT,LON,LEV>(5,2,8) (or viceversa) or both. The core issue is that our IO (and our grid class) is assuming that only ONE dimension is partitioned.

Perhaps we can allow this by hand rolling an ad-hoc routine to create the PIO decomp, which does not rely on this assumption. I need to think about this.

[mahf708]

We [mostly you] did a tacky implementation of this back in the day. We can try to revive it and give it to a bot. Or we can try to hack a new logic like ncremap does?

[bartgol]

I think you may be referring to the ACE atm process? That was different. We only had to do the hacky stuff online, there was no interaction with the eamxx IO system for this. Doing IO to/from (lat,lon) is different.