Compile-time ng in ecCKD and TripleClouds + GPTL timing library

[peterukk]

Currently the official repo is missing two major optimizations found in my ecRad-OTP: compile-time ng and cloudy layer batching. I'm making a pull request for compile-time ng first - TripleClouds might not get much additional benefit from cloudy layer batching since it already collapses the two cloudy regions in the shortwave, but we'll see.

Here the compile-time NG is implementing in TripleClouds-LW, TripleClouds-SW and ecCKD and the #ifdef NG_LW are added to ecrad_config.h, which helps avoid code bloat. To avoid compilation errors I needed to move the #include "ecrad_config.h" in all files to occur after the module statement.

Support for General Purpose Timing Library is also added in order to measure the performance impact. I can remove it if you wish but it adds minimal extra code: a few lines in the main Makefile, the driver (I only added it to the blocked IFS driver since this is what should be used for timing tests) and a few instrumentation calls in radiation_interface.

Compile-time ng has minimal impact on TripleClouds but speeds up ecCKD by a factor of 3 when using the Intel compiler on Atos:

make PROFILE=intel_atos SINGLE_PRECISION=1 GPTL_TIMING=1 
cd test/ifs; make test_ifsdriver_blocked_ecckd_tc 
cat timing.ECCKD_Tripleclouds_block8_nrep10_0038
                                   Called  Recurse     Wall      max      min %_of_radi
 radiation                              1     -       0.114    0.114    0.114   100.00
   radiation_interface:radiation       40     -       0.111 4.76e-03 2.39e-03    97.13
     gas_optics                        40     -       0.030 1.00e-03 7.44e-04    26.61
     cloud_optics                      40     -    4.22e-03 4.44e-04 8.00e-05     3.69
     aerosol_optics                    40     -       0.011 3.75e-04 2.71e-04     9.83
     solver_longwave                   40     -       0.032 1.77e-03 6.86e-04    27.59
     solver_shortwave                  40     -       0.033 1.10e-03 5.17e-04    29.08

make PROFILE=intel_atos SINGLE_PRECISION=1 GPTL_TIMING=1 NG_SW=32 NG_LW=32
cd test/ifs; make test_ifsdriver_blocked_ecckd_tc 
                                   Called  Recurse     Wall      max      min %_of_radi
 radiation                              1     -       0.087    0.087    0.087   100.00
   radiation_interface:radiation       40     -       0.083 2.56e-03 1.79e-03    96.29
     gas_optics                        40     -       0.011 4.27e-04 2.55e-04    12.16
     cloud_optics                      40     -    4.08e-03 3.17e-04 8.10e-05     4.71
     aerosol_optics                    40     -       0.011 3.45e-04 2.71e-04    12.79
     solver_longwave                   40     -       0.028 8.28e-04 6.35e-04    32.40
     solver_shortwave                  40     -       0.029 9.20e-04 4.69e-04    33.85

So a 20-25% speed-up in total.

[FussyDuck]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Peter Ukkonen seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}