Building WRF with optimization schemes ( ie. -O3)

[Deleted member 3607]

My question today is about optimization schemes for the fortran compilers.

What exactly does -O3 for CFLAGS do to WRF and does it help improve it?

 

[kwerner]

The "-O3" option increases optimization by 3x. I *believe* this allows the code to build and run faster, without printing non-vital information. The WRF model builds with optimization by default, and if users are having trouble building specific routines or if they need to debug a simulation, we recommend turning optimization off, or decreasing it. Unfortunately I'm not a software engineer and don't have the background to provide a better answer specific to the WRF model, but there are several resources online about compiler optimization.

 

[Deleted member 3607]

The "-O3" option increases optimization by 3x. I *believe* this allows the code to build and run faster, without printing non-vital information. The WRF model builds with optimization by default, and if users are having trouble building specific routines or if they need to debug a simulation, we recommend turning optimization off, or decreasing it. Unfortunately I'm not a software engineer and don't have the background to provide a better answer specific to the WRF model, but there are several resources online about compiler optimization.

Thank you

 

[alainF]

Hello,
I have also tried -Ofast with gcc10 or higher. It speeds up the simulation, but I don't know the impact on numerical precision. But the simulation runs. I have seen that the binary compiled with gcc >= 10 has a speed closer to the binary compiled with Intel.