bit-for-bit diagnostics

[apcraig]

CICE currently has 3 different flags that affect bit-for-bit diagnostics. The first "DITTO" is an env variable in cice.settings. That triggers the cpp -DREPRODUCIBLE but only for a few machine/compiler implementations. It probably should be specified in all machine/compiler implementations if any. Finally, there is the bfbflag in namelist. That turns on yet another calculation for the global diagnostics but only on the mpi side.

I see several issues

• multiple independent approaches to address the same issue
• inconsistency in scripts application to various machines and compilers
• inconsistency in implementation between mpi and serial versions (serial does not have a bfbflag implementation)
• lack of reuse in implementation of sums, mpi vs serial, int/float/double, prod sums, etc.
• no recent testing as far as I know
• really does need to be cleaned up

I would like to refactor this with the following requirements

• run time flag only
• implementation in both mpi and serial, recognizing this is not only a pe count issue but also a decomp/block issue.
• code reuse where possible
• leverage new techniques such as the Worley ddpdd approach to reduce cost
• potentially support multiple approaches via a consistent, well documented namelist input
• test and validate

This feature is useful, especially for quick bit-for-bit comparison of log files during testing. If the cost is low enough, it may even be a viable default setting.

It is probably not a lot of work, I have implemented the same thing in other projects and would leverage some of that work. I would like to propose it be added to the CICE6 release project.

Pinging @eclare108213, @mattdturner, @dabail10 for any comments or feedback.

[eclare108213]

This makes sense to me. DITTO and REPRODUCIBLE came from UKMO a long time ago, specifically because they wanted to be able to compare diagnostic output across configurations (these only affect the diagnostic output). bfbflag came from CESM 4 years ago (the "blame" functionality in github is very handy). @apcraig, yes, please clean this up.

[duvivier]

Also get rid of the following in cice.settings:
setenv CAM_ICE no # set to yes for CAM runs (single column)
setenv BARRIERS no # prevent MPI buffer overflow during gather/scatter

[apcraig]

PR #300 complete