ICTP / RegCM

ICTP Regional Climate Model
https://www.ictp.it/esp/about
Other
54 stars 42 forks source link

Trouble withe terrain domains crossing the date line when using Intel compiler #32

Open jalder-usgs opened 1 year ago

jalder-usgs commented 1 year ago

I've run into trouble running terrain on domains that span the date line, which seg faults when using the Intel compiler. Fortunately it seems to work with the GNU compiler.

Here are the two software stacks I've tried:

GNU: gcc / gfortran 8.4.0 netcdf_c_ver=4.9.0 netcdf_f_ver=4.5.4 hdf5_ver=1.13.2 zlib_ver=1.2.13 ompi_ver=4.1.4

Intel: icc (ICC) 2021.6.0 20220226 ifort (IFORT) 2021.6.0 20220226 netcdf_c_ver=4.9.0 netcdf_f_ver=4.5.4 hdf5_ver=1.13.2 zlib_ver=1.2.13 ompi_ver=4.1.4

Output when running Intel compiled version of terrain:

This is Terrain part of RegCM package version 5
   GIT Revision: 5.0.0 compiled at: data : Mar 23 2023  time: 16:33:25

 : this run start at  : 2023-03-24 11:41:04-0700
 : it is submitted by : alderj
 : it is running on   : kona
 : in directory       : /head6/Regional/RegCM5_tests/GoA

 Using pre-defined set of sigma levels.

 Doing Horizontal Grid with following parameters
 iy     =           40
 jx     =           81
 kz     =           18
 ds     =    60.0000000000000     
 clat   =    55.0000000000000     
 clon   =   -160.000000000000     
 iproj  = LAMCON
 Grid setup done
 Geo mapping done
 Special case for timeline crossing
 Calculated large extrema:
          MINLAT =    40.0000000000000     
          MAXLAT =    67.0000000000000     
          MINLON =    155.000000000000     
          MAXLON =   -114.000000000000     
 Determined Grid coordinate range
 No resampling used.
forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image              PC                Routine            Line        Source             
terrainNETCDF4_HD  0000000000531839  for__signal_handl     Unknown  Unknown
libpthread-2.31.s  00007F22DB7DE420  Unknown               Unknown  Unknown
libnetcdff.so.7.1  00007F22DC276AB8  netcdf_mp_nf90_ge     Unknown  Unknown
terrainNETCDF4_HD  00000000004C6DA2  Unknown               Unknown  Unknown
terrainNETCDF4_HD  0000000000425429  Unknown               Unknown  Unknown
terrainNETCDF4_HD  000000000040DD22  Unknown               Unknown  Unknown
libc-2.31.so       00007F22DB5FC083  __libc_start_main     Unknown  Unknown
terrainNETCDF4_HD  000000000040DC2E  Unknown               Unknown  Unknown

Output when running the same file when compiled with GNU:

  This is Terrain part of RegCM package version 5
   GIT Revision: 5.0.0 compiled at: data : Mar 24 2023  time: 10:47:07

 : this run start at  : 2023-03-24 11:34:57-0700
 : it is submitted by : alderj
 : it is running on   : hilo
 : in directory       : /head6/Regional/RegCM5_tests/GoA

 Using pre-defined set of sigma levels.

 Doing Horizontal Grid with following parameters
 iy     =           40
 jx     =           81
 kz     =           18
 ds     =    60.000000000000000     
 clat   =    55.000000000000000     
 clon   =   -160.00000000000000     
 iproj  = LAMCON
 Grid setup done
 Geo mapping done
 Special case for timeline crossing
 Calculated large extrema:
          MINLAT =    40.000000000000000     
          MAXLAT =    67.000000000000000     
          MINLON =    155.00000000000000     
          MAXLON =   -114.00000000000000     
 Determined Grid coordinate range
 No resampling used.
 Static DEM data successfully read in
 Interpolated DEM on model GRID
 Static landcover BATS data successfully read in
 Interpolated landcover on model GRID
 Static texture data successfully read in
 Interpolated texture on model GRID
 Fudging data (if requested) succeeded
 Skip moisture creation, no moisture file to read
 Vertical Grid Description (T estimated)

 --------------------------------------------------
 k        sigma       p(mb)           z(m)     T(K)
 --------------------------------------------------
 19       1.000     1013.25          0.00     277.1
 18       0.990     1003.12         81.48     277.0
 17       0.980      992.99        163.73     276.8
 16       0.960      972.72        330.59     276.5
 15       0.930      942.32        587.04     275.9
 14       0.890      901.79        941.19     275.2
 13       0.840      851.13       1404.91     274.0
 12       0.780      790.34       1994.43     271.8
 11       0.710      719.41       2734.57     268.9
 10       0.630      638.35       3662.84     265.3
  9       0.550      557.29       4696.79     260.1
  8       0.470      476.23       5863.50     253.6
  7       0.390      395.17       7205.95     245.8
  6       0.310      314.11       8795.49     236.5
  5       0.230      233.05      10776.24     226.7
  4       0.160      162.12      13127.01     221.3
  3       0.100      101.33      16141.53     219.1
  2       0.050       50.66      20571.28     218.3
  1       0.000        0.00      Infinity     219.3
 Grid data written to output file
 Successfully completed terrain fields generation

I don't know if the compile problems are specific to the Intel compiler or just the default compile flags used with Intel and configure. Hopefully I'll be able to create the domain file under a GNU build and then run ICBC and the model with an Intel build.