Self compile WRF 3.4.1 and use .exe into WRF EMS

All issues/questions about EMS v3.4 package, please ask here.
smartie
Posts: 97
Joined: Sat May 21, 2011 7:34 am

Re: Self compile WRF 3.4.1 and use .exe into WRF EMS

Post by smartie » Tue Apr 08, 2014 8:20 am

I must say that I haven't looked into this in detail up to now. On larger machines I haven't had to worry too much about it. Now I'm thinking of doing regular forecasts on smaller machines I'm looking at getting the most out of the limited resources.
At first sight it's a bit confusing. The Intel articles on compiling WRF with ifort seem to run out in ~2011 and advice is scattered through the forums.

Is there no where advice is gathered together on compiling and optimising recent versions of WRF?

oib
Posts: 118
Joined: Mon Jan 31, 2011 3:29 pm

Re: Self compile WRF 3.4.1 and use .exe into WRF EMS

Post by oib » Thu Apr 10, 2014 3:27 pm

oib wrote: Thanks a lot for your clear explanation!!!
I will try re-compile WRF 3.4.1 with these new knowledges and evaluate the impact on the ems_run performance :)
So I have re-compiled netCDF,mpich3,zlib,libpng,jasper and WRFV3.4.1 ARW + EMS patches with Intel adding to CFLAGS_LOCAL the flag -xCORE-AVX-I (-xHost suitable value for Intel Xeon 26xx CPUs).

Running EMS ARW large benchmark with the Intel compiled real.exe and wrf.exe the simulation completed in 10 minutes 55 seconds

The same but with EMS pre-compiled .exe is 11 minutes 22 seconds.

An improvement in the ems_run computation time by more than 4% :)

It sounds good!

Many thanks to Ivan for the tip on the -xHost flag !!!

Applying the -xHost flag also to the Fortran compiler (entry FCOPTIM in configure.wrf) could be also advantageous?

meteoadriatic
Posts: 1566
Joined: Wed Aug 19, 2009 10:05 am

Re: Self compile WRF 3.4.1 and use .exe into WRF EMS

Post by meteoadriatic » Thu Apr 10, 2014 4:07 pm

Maybe but I don't think you will see difference. You can always try and report to us if you see benefit :)

Those EMS patches slow down WRF :? If you compile original code it is even faster than with EMS patches. Do you actually know what do you get using them?

oib
Posts: 118
Joined: Mon Jan 31, 2011 3:29 pm

Re: Self compile WRF 3.4.1 and use .exe into WRF EMS

Post by oib » Thu Apr 10, 2014 4:17 pm

meteoadriatic wrote:Maybe but I don't think you will see difference. You can always try and report to us if you see benefit :)

Those EMS patches slow down WRF :? If you compile original code it is even faster than with EMS patches. Do you actually know what do you get using them?
Really I don't know exactly :oops:
I think some more output field is calculated, but I haven't ever had time to study this.
Probably I will follow your suggestion and will go without these patches.
In the future could be a good exercise to study what Robert patches do. I'm curious especially about the registry files...

meteoadriatic
Posts: 1566
Joined: Wed Aug 19, 2009 10:05 am

Re: Self compile WRF 3.4.1 and use .exe into WRF EMS

Post by meteoadriatic » Thu Apr 10, 2014 4:20 pm

oib wrote:I think some more output field is calculated, but I haven't ever had time to study this.
That is true, if you need those additional fields you should keep patches. However if you want to use your own unipost, like me, then it won't work OK with files produced with EMS version of WRF (as far as I remember). So I just compiled original versions of everything.

oib
Posts: 118
Joined: Mon Jan 31, 2011 3:29 pm

Re: Self compile WRF 3.4.1 and use .exe into WRF EMS

Post by oib » Fri Apr 11, 2014 1:19 pm

meteoadriatic wrote:
oib wrote:I think some more output field is calculated, but I haven't ever had time to study this.
That is true, if you need those additional fields you should keep patches. However if you want to use your own unipost, like me, then it won't work OK with files produced with EMS version of WRF (as far as I remember). So I just compiled original versions of everything.
I have tried running EMS using real.exe and wrf.exe that was self-compiled WITHOUT replacing original WRF files with the ones (so called "EMS patches") provided into the folder /wrfems/util/wrfems/wrfv3.4.1 .

Good news: post processing made by emsupp (post-processor in EMS version) successfully create the grib2 output files. Run time was ~ 3% faster.

Bad news: a lot of output fields are missing because, as we said, they can be calculated by emsupp only if the netCDF output files contains those fieds defined into the EMS patches, that in this case weren't applied by me.

It would not be so bad news, since many of these fields do not interest me.

BUT unfortunately some missing fields are interesting, such as the relative humidity at 2m above ground and surface accumulated precipitations.

I hope that the missing important fields re-appear in the case in which one would use the original emsupp instead of the EMS emsupp.

... will try ...

meteoadriatic
Posts: 1566
Joined: Wed Aug 19, 2009 10:05 am

Re: Self compile WRF 3.4.1 and use .exe into WRF EMS

Post by meteoadriatic » Fri Apr 11, 2014 3:18 pm

Correct, I had the same conclusion some time ago. However I'm not sure that EMS script will run original upp binary...

Post Reply

Who is online

Users browsing this forum: Google [Bot] and 2 guests