NDOWN with nest ratio other than 1:3 (Posted 6/15/06)
Problem: If you are using ndown
with a nest ratio other than 1:3, the code would be in error.
Solution: Edit WRFV2/main/ndown_em.F,
and look for string 'rdx' and 'rdy', change the hardwired ratio
3 to the one you would like to use. Recompile.
RSL Advisory (Posted 6/15/06)
Use RSL_LITE for compilation whenever possible,
unless you use y-periodic boundary conditions. Do not use RSL
until further notice.
PGI F90 Compiler version 6.1-1, 6.1-2 (Posted 3/3/06)
Problem: If you have PGI Fortran
90 compiler versions 6.1-1 and 6.1-2, you may experience problem
with running WRF code.
Solution: Upgrade your compiler
to 6.1-3, available from PGI site since March 2.
Note: problems were still reported with real.exe
with 6.1-3 compiler.
Update (3/10/06): PGI has discovered
the error associated with compiling real. A new upgrade will be
released on April 6 (version 6.1-4). Meanwhile, it is suggested
to use -O0 when compiling real.exe.
Fix for ESMF compilation Using some old PGI compilers (Posted
2/15/06)
Problem: Some upgrade of ESMF
time manager code in 2.1.2 caused the compilation to fail when
older versions of the PGI compiler are used.
Solution: Download the patch
tar file, patchpgi.tar.gz,
and untar it in the WRFV2/ directory. This tar file contains 6
modified routines and a makefile for the external/esmf_time_f90/
directory. Do 'clean -a' and recompile.
Bug Fix for RSL_LITE with Double-Precision Computation (Posted
2/15/06)
Problem: The bug in RSL_LITE
communication layer that will cause run-time problems (probably
seg fault or bus errors) when running WRF with double-precision
(64-bit) floating point data; the problem does not affect single-precision
(32-bit) floating point data. It only affects distributed memory
parallel compiles using RSL_LITE.
Solution: In external/RSL_LITE/c_code.c:
change line 305 from
F_PACK_LINT
( buf, p+yp_curs, &js, &je, &ks, &ke, &is,
&ie,
to
F_PACK_LINT
( buf, p+ym_curs, &js, &je, &ks, &ke, &is,
&ie,
and delete line 308. Do 'clean -a', and recompile.
Bug Fix for NEST (Posted 2/15/06)
Problem: An error is found
in the boundary zone of the nest, if nest input file is used (namelist
variable input_from_file set to true for the nest). The
topography (and reference state geopotential and column pressure)
along the nest lateral boundary is comprised of three pieces:
an outer section that is horizontally interpolated from the coarse
grid, and inner region where nest terrain is used, and a blended
region. The bug is to do with weight computed for the blending
zone. The effect of this on a simulation is small.
Solution: Download a new version
of the routine,
nest_init_utils.F.fix, and replace the one in dyn_em/ directory.
Recompile.
Bug Fix for YSU PBL Scheme (Posted 1/30/06)
Problem: The YSU PBL scheme
in V2.1.2 is found to be unstable under certain circumstance.
The most noticeable symptom is the CFL prints occurring at the
lowest model levels.
Solution: Download a new version
of the routine, module_bl_ysu.F.fix,
and replace the one in phys/ directory. Recompile.
Problem with running ndown.exe on some platforms (Posted 1/30/06)
Problem: On some platforms,
noticeably Linux. ndown could seg fault at the end of one time
period.
Solution: Download a new version
of the routine, io_grib1.F.fix,
and replace the one in external/io_grib1/ directory. Recompile.
Long Compile Time on SGI Altix: (Posted 11/8/05)
Problem: The Intel ifort compiler
spends a long time compiling the following files:
module_ra_gfdleta.f
module_advect_em.f
mediation_force_domain.f
solve_em.f
wrf_inputout.f
module_configure.f
ndown_em.f
module_big_step_utilities_em.f
module_diffusion_em.f
Partial Solution: You can reduce
the compile time by adding the following compiler option:
-mP3OPT_ecg_lra_switch=T
This does impact the performance of the module/routines
that have this optimization removed - they slow down. If "speed
of compile" is more important that total run time (such as
for development work), this is not a bad option to include for
all files. If "total runtime" is important, then the
user should accept the slow compiles. A reasonable compromise
isto only modify the above files with the new compiler option.
The Intel group plans to have a fix for the
ifort compiler soon. (Thanks to Gerardo Cisneros of SGI.)
Using RSL for Nesting on Linux (posted on 11/8/05):
Problem: The Linux systems
with the PGI compiler, when using RSL for a nested run, incorrectly
interpolate fields within the SINT subroutine for some variables,
for some decomposition configurations (for example, using 10 processors
to run). The most identifiable symptom is that the topography
field of the nest at the initial time shows characteristic lines
along the decomposition boundaries.
Solution: Use RSL_LITE as the
communication layer instead of RSL.
As long as there is no need for periodic
boundary condition in the y-direction, the use of RSL_LITE is
transparent to the user (with the added benefits of 1) allowing
larger domains, 2) decomposition of more processors, and 3) running
faster).
RSL_LITE:
Problem: Does not work for
periodic boundary condition in y-direction.
