Bugfix for regional runs when dycore is compiled in double precision

[climbfuji]

This PR fixes issue #24.

It also adds one additional line to the initial printout, which we found useful to make sure that all constants and tracers are set up correctly.

[climbfuji]

@DusanJovic-NOAA @junwang-noaa @ligiabernardet @bensonr @ChunxiZhang-NOAA @BinLiu-NOAA @JeffBeck-NOAA @binli2337

[climbfuji]

@mzhangw

[bensonr]

buf1-4 are now allocatable variables, but the count argument for irecv is still using the ibufexch parameter. I recommend changing the count in the irecv be the size of buf1 and buf2 respectively.

[climbfuji]

buf1-4 are now allocatable variables, but the count argument for irecv is still using the ibufexch parameter. I recommend changing the count in the irecv be the size of buf1 and buf2 respectively.

Good catch. Thanks very much. Will cleanup!

[climbfuji]

buf1-4 are now allocatable variables, but the count argument for irecv is still using the ibufexch parameter. I recommend changing the count in the irecv be the size of buf1 and buf2 respectively.

Good catch. Thanks very much. Will cleanup!

I fixed this - removed ibufexch entirely and used size(bufX) in the messages (since the count isn't known at the time the first MPI commands occur). I reran the tests on Cheyenne, it compiles and runs in DEBUG mode for both 32bit and 64bit without crashing, and the PROD runs in 64-bit give b4b identical results from run to run (which was the issue that triggered this PR in the first place).

[bensonr]

Changes look consistent.

I would recommend some inline comments or even protection logic to ensure buffering/unbuffering goes as planned. Any future modification to the logic needs to understand index changes must be thoroughly tested - especially when packing multiple variables into the same buffer.

[climbfuji]

Changes look consistent.

I would recommend some inline comments or even protection logic to ensure buffering/unbuffering goes as planned. Any future modification to the logic needs to understand index changes must be thoroughly tested - especially when packing multiple variables into the same buffer.

@bensonr does this look better? It still works on Cheyenne as expected.

[bensonr]

Looks good

[climbfuji]

Associated PRs:
#25
NOAA-EMC/fv3atm#133
ufs-community/ufs-weather-model#155

For regression testing, see ufs-community/ufs-weather-model#155.