Update dtc/develop from EMC develop and ufs_public_release 2019/12/16

[climbfuji]

This PR and NCAR/fv3atm#21 update the dtc/develop branch with changes that were originally made for the EMC ufs_public_release branch and with changes from ECM's develop branch:

• new cmake build system for ufs-weather-model
• add support for Cheyenne with Intel and GNU to new cmake build
• add regression tests for suites FV3_GFS_v15p2 and FV3_GFS_v16beta
• tests/fv3_conf/ccpp_*_run.IN: remove INPUT and RESTART as first step, because CCPP regression test directories are reused when the same job is run twice in CCPP dynamic and static mode

For regression testing, see below.

[climbfuji]

Regression testing on hera/intel against DTC baseline (trunk-20191127; this is identical to the EMC baseline with the same date tag, except that updates of the NoahMP and GSD tests are already included): all tests pass, except the newly added CCPP PROD tests rt_136_fv3_ccpp_gfs_v15p2_prod.log and rt_137_fv3_ccpp_gfs_v16beta_prod.log ("missing baseline").

rt_full.log

[climbfuji]

Regression testing on Cheyenne/GNU using rt_gnu.conf: all tests pass, except the newly added CCPP PROD tests rt_011_fv3_ccpp_gfs_v15p2_prod.log and rt_012_fv3_ccpp_gfs_v16beta_prod.log ("missing baseline").

rt_gnu.log

[climbfuji]

Regression tests on hera/intel using rt_ccpp_hafs.conf (first create, then verify): all tests pass.

rt_ccpp_hafs_create.log
rt_ccpp_hafs_verify.log

[climbfuji]

NEMSCompsetRun test on hera/intel against EMC baseline (trunk-20191127): tests pass except fv3_gfdlmprad_noahmp. This is expected, because the EMC baseline was created before the NoahMP bugfix was made.

NEMSCompsetRun.log

[climbfuji]

Regression tests on hera/intel using rt_ccpp_gsd.conf (first create, then verify): all tests pass.

rt_ccpp_gsd_create.log
rt_ccpp_gsd_verify.log

[climbfuji]

Regression tests on hera/intel using rt_ccpp_caps.conf (first create, then verify): all tests pass.

rt_ccpp_caps_create.log
rt_ccpp_caps_verify.log

[climbfuji]

Regression testing on Cheyenne/Intel using rt_intel.conf: all tests pass, except the newly added CCPP PROD tests rt_011_fv3_ccpp_gfs_v15p2_prod.log and rt_012_fv3_ccpp_gfs_v16beta_prod.log ("missing baseline").

In the first pass, the test fv3_ccpp_gfdlmprad_gwd_repro also failed because of a PBS error. I reran the test and it passed.

rt_intel.log
rt_intel_rerun_fv3_ccpp_gfdlmprad_gwd_repro.log

[climbfuji]

Regression testing on hera/intel using rt.conf (first create, then verify against own baseline): all pass

rt_verify.log
rt_create.log

[llpcarson]

Looks good to me, approved

[grantfirl]

OK by me. I hope changes to building/running get documented appropriately!

[climbfuji]

OK by me. I hope changes to building/running get documented appropriately!

Thanks. There are no immediate changes to running and building the model. The old compile.sh script still works, and rt.sh can use either of them (this is transparent to the user).