Parallel netcdf: colstat, virtualmeasurement and warmstart fixes - #302
Parallel netcdf: colstat, virtualmeasurement and warmstart fixes#302Xetalim wants to merge 22 commits into
Conversation
plus chunking options.. sorry. I checked modcolstat by taking a column profile at every single point, and comparing the average of all columns with modgenstat. differences were ~epsilon for all variables
if dt_lim is tnext, it currently works at a coldstart, but at a warmstart, dt_lim would be (t_warmstart + dt_lim) instead of dt_lim. also make sure dt_lim is correctly set at start for new netcdf interface
…cs (fresh)" This reverts commit 5d5afa5.
|
To be continued after the weekend... |
|
I've fixed the SP compilation, but I don't really like the solution. If any of you have a better idea of how to do this, let me know! |
|
Last commit should have fixed all SP issues, so should be ready for review :) |
| crossplane(2:100) = -999 | ||
| crossortho(1) = 2 | ||
| crossortho(2:100) = -999 | ||
| crossheight(1)=2 |
There was a problem hiding this comment.
Why remove spaces???
There was a problem hiding this comment.
Good point, I had originally removed this initialization towards cross*(1)=2, but that broke the default behaviour, so I brought it back (wrongly). Fixed now!
| ! Monin-obukhov / surface layer: | ||
| real, allocatable :: obuk(:,:), ustar(:,:), ra(:,:) | ||
| real(field_r), allocatable :: obuk(:,:) | ||
| real, allocatable :: ustar(:,:), ra(:,:) |
There was a problem hiding this comment.
Can't these be field_r as well?
There was a problem hiding this comment.
I agree, but we might want to separate that out in a different PR? Cleaning up some of the extra field_r stuff
| !> Loop over the files and write those that are due. | ||
| subroutine write_output_files() | ||
|
|
||
| use modtimer, only: timer_tic, timer_toc |
There was a problem hiding this comment.
Please put this at the top of the module with the other use statements :-)
|
|
||
| call timer_tic("write_output_files") | ||
| do ifile = 1, nfiles | ||
| call timer_tic("write_output_files_file_"//trim(file_list(ifile)%file%filename)) |
There was a problem hiding this comment.
Ahhh, timers per file. I like it
There was a problem hiding this comment.
I like them as well but I unfortunately had to remove them, because when you use crosssections, it might be that 1 core has more timers than another, which leads to a hang in the timer writes at the end :/
| if(zeta <= 0) then | ||
| x = (1. - 16. * zeta) ** (0.25) | ||
| psim = 3.14159265 / 2. - 2. * atan(x) + log( (1.+x) ** 2. * (1. + x ** 2.) / 8.) | ||
| x = (1._field_r - 16._field_r * zeta) ** (0.25_field_r) |
There was a problem hiding this comment.
Integers don't have to be casted to field_r explicitly, the compiler will take care of this. It's not wrong, but not doint it can keep lines like these a bit shorter and more readable
There was a problem hiding this comment.
I changed them back to normal integers, seems like a good idea
| u0av_patch = patchsum_1level_field_r(u0(2:i1,2:j1,k)) * (xpatches*ypatches/ijtot) | ||
| v0av_patch = patchsum_1level_field_r(v0(2:i1,2:j1,k)) * (xpatches*ypatches/ijtot) | ||
| w0av_patch = patchsum_1level_field_r(w0(2:i1,2:j1,k)) * (xpatches*ypatches/ijtot) | ||
| u0av_patch = patchsum_1level(u0(2:i1,2:j1,k)) * (xpatches*ypatches/ijtot) |
There was a problem hiding this comment.
The lhetero stuff is already gone in dev, so this might lead to merge conflicts.
There was a problem hiding this comment.
I think it's better to have the merge conflict later, as it's nice to know that it compiles correctly on this branch, as not doing anything there will lead to SP comp errors
CasparJungbacker
left a comment
CasparJungbacker
left a comment
There was a problem hiding this comment.
I have one general question about the new statistics modules: why do you setup (allocatable) variables for all quantities, instead of using pointers and the get_pointer method from the various NetCDF file type wrappers?
|
I moved towards using the pointer manager instead. The only possibly nasty thing is that the scalar value calculations have to get a new pointer each time we move to a different column index. |
2 participants
To be merged in parallel netcdf. Contains some cleanup for modstat_nc, and (optional) chunking behavior.