Feature/pnorris/rrtmgp v1.7

[dr0cloud]

Upgrade to RRTMGP v1.7 from previous v1.6.
Needs @RRTMGP to be at v1.7.
This should be 0-diff for a default RRTMG run, but please check.
It will be non-0-diff for an RRTMGP run.
v1.7 has various improvements (see @RRTMG) including some numerics improvements.
Under v1.7 RRTMGP can now be run in Single Precision mode with the option below:
./parallel_build.csh -- -DRRTMGP_SINGLE_PRECISION=ON.
Whether this is faster is being investigated ... its likely a trade-off, since it will be less accurate than default double precision.
I have noticed around a -0.15W/m2 difference in global OSR in SP mode, but individual clouds can be -2 W/m2 or more.
This must be matched with a corresponding PR in @GEOS_App, since I change some of the configuration names for RRTMGP in AGCM.rc.tmpl.

[mathomp4]

Here are results from some benchmarking RRTMGP in GEOS in a 1-day run at 72 levels.

---

1.6 DP vs 1.7 DP

Resolution	1.6 DP SW	1.7 DP SW	Speedup	1.6 DP LW	1.7 DP LW	Speedup	Total 1.6 DP	Total 1.7 DP	Speedup
c24	3.16	3.31	-5%	2.07	2.10	-1%	5.23	5.41	-3%
c48	9.47	9.46	0%	6.84	6.79	1%	16.31	16.25	0%
c90	43.50	46.11	-6%	32.99	33.99	-3%	76.49	80.10	-5%
c180	60.27	62.94	-4%	50.40	50.96	-1%	110.67	113.90	-3%
c360	61.55	62.38	-1%	51.92	51.18	1%	113.47	113.56	0%
c720	66.41	65.69	1%	54.63	52.74	3%	121.04	118.43	2%

---

1.6 DP vs 1.7 SP

Resolution	1.6 DP SW	1.7 SP SW	Speedup	1.6 DP LW	1.7 SP LW	Speedup	Total 1.6 DP	Total 1.7 SP	Speedup
c24	3.16	2.87	9%	2.07	1.84	11%	5.23	4.71	10%
c48	9.47	7.76	18%	6.84	6.32	8%	16.31	14.08	14%
c90	43.50	37.82	13%	32.99	32.52	1%	76.49	70.34	8%
c180	60.27	47.82	21%	50.40	42.14	16%	110.67	89.96	19%
c360	61.55	49.24	20%	51.92	42.86	17%	113.47	92.10	19%
c720	66.41	54.88	17%	54.63	46.55	15%	121.04	101.43	16%

[RobertPincus]

@mathomp4 There was a performance regression relatively recently, see earth-system-radiation/rte-rrtmgp#215. This is on my radar of things to fix.

[mathomp4]

@mathomp4 There was a performance regression relatively recently, see earth-system-radiation/rte-rrtmgp#215. This is on my radar of things to fix.

@RobertPincus It was my fault. It turns out my original 1.6 runs used the wrong GEOS driving code. That new code is a bit slower in SW because of more diagnostics, etc. I've updated the tables and now all is good! DP v DP is about a wash, and DP v SP is faster as one would hope!