| 1 |
\section[Customizing MITgcm]{Doing it yourself: customizing the code} |
\section[Customizing MITgcm]{Doing it yourself: customizing the code} |
| 2 |
|
\label{sect:customize} |
| 3 |
|
\begin{rawhtml} |
| 4 |
|
<!-- CMIREDIR:customizing_mitgcm: --> |
| 5 |
|
\end{rawhtml} |
| 6 |
|
|
| 7 |
When you are ready to run the model in the configuration you want, the |
When you are ready to run the model in the configuration you want, the |
| 8 |
easiest thing is to use and adapt the setup of the case studies |
easiest thing is to use and adapt the setup of the case studies |
| 13 |
part is covered in the parallel implementation section) and on the |
part is covered in the parallel implementation section) and on the |
| 14 |
variables and parameters that you are likely to change. |
variables and parameters that you are likely to change. |
| 15 |
|
|
| 16 |
|
|
| 17 |
|
\subsection{Building/compiling the code elsewhere} |
| 18 |
|
|
| 19 |
|
In the example above (section \ref{sect:buildingCode}) we built the |
| 20 |
|
executable in the {\em input} directory of the experiment for |
| 21 |
|
convenience. You can also configure and compile the code in other |
| 22 |
|
locations, for example on a scratch disk with out having to copy the |
| 23 |
|
entire source tree. The only requirement to do so is you have {\tt |
| 24 |
|
genmake2} in your path or you know the absolute path to {\tt |
| 25 |
|
genmake2}. |
| 26 |
|
|
| 27 |
|
The following sections outline some possible methods of organizing |
| 28 |
|
your source and data. |
| 29 |
|
|
| 30 |
|
\subsubsection{Building from the {\em ../code directory}} |
| 31 |
|
|
| 32 |
|
This is just as simple as building in the {\em input/} directory: |
| 33 |
|
\begin{verbatim} |
| 34 |
|
% cd verification/exp2/code |
| 35 |
|
% ../../../tools/genmake2 |
| 36 |
|
% make depend |
| 37 |
|
% make |
| 38 |
|
\end{verbatim} |
| 39 |
|
However, to run the model the executable ({\em mitgcmuv}) and input |
| 40 |
|
files must be in the same place. If you only have one calculation to make: |
| 41 |
|
\begin{verbatim} |
| 42 |
|
% cd ../input |
| 43 |
|
% cp ../code/mitgcmuv ./ |
| 44 |
|
% ./mitgcmuv > output.txt |
| 45 |
|
\end{verbatim} |
| 46 |
|
or if you will be making multiple runs with the same executable: |
| 47 |
|
\begin{verbatim} |
| 48 |
|
% cd ../ |
| 49 |
|
% cp -r input run1 |
| 50 |
|
% cp code/mitgcmuv run1 |
| 51 |
|
% cd run1 |
| 52 |
|
% ./mitgcmuv > output.txt |
| 53 |
|
\end{verbatim} |
| 54 |
|
|
| 55 |
|
\subsubsection{Building from a new directory} |
| 56 |
|
|
| 57 |
|
Since the {\em input} directory contains input files it is often more |
| 58 |
|
useful to keep {\em input} pristine and build in a new directory |
| 59 |
|
within {\em verification/exp2/}: |
| 60 |
|
\begin{verbatim} |
| 61 |
|
% cd verification/exp2 |
| 62 |
|
% mkdir build |
| 63 |
|
% cd build |
| 64 |
|
% ../../../tools/genmake2 -mods=../code |
| 65 |
|
% make depend |
| 66 |
|
% make |
| 67 |
|
\end{verbatim} |
| 68 |
|
This builds the code exactly as before but this time you need to copy |
| 69 |
|
either the executable or the input files or both in order to run the |
| 70 |
|
model. For example, |
| 71 |
|
\begin{verbatim} |
| 72 |
|
% cp ../input/* ./ |
| 73 |
|
% ./mitgcmuv > output.txt |
| 74 |
|
\end{verbatim} |
| 75 |
|
or if you tend to make multiple runs with the same executable then |
| 76 |
|
running in a new directory each time might be more appropriate: |
| 77 |
|
\begin{verbatim} |
| 78 |
|
% cd ../ |
| 79 |
|
% mkdir run1 |
| 80 |
|
% cp build/mitgcmuv run1/ |
| 81 |
|
% cp input/* run1/ |
| 82 |
|
% cd run1 |
| 83 |
|
% ./mitgcmuv > output.txt |
| 84 |
|
\end{verbatim} |
| 85 |
|
|
| 86 |
|
\subsubsection{Building on a scratch disk} |
| 87 |
|
|
| 88 |
|
Model object files and output data can use up large amounts of disk |
| 89 |
|
space so it is often the case that you will be operating on a large |
| 90 |
|
scratch disk. Assuming the model source is in {\em ~/MITgcm} then the |
| 91 |
|
following commands will build the model in {\em /scratch/exp2-run1}: |
| 92 |
|
\begin{verbatim} |
| 93 |
|
% cd /scratch/exp2-run1 |
| 94 |
|
% ~/MITgcm/tools/genmake2 -rootdir=~/MITgcm \ |
| 95 |
|
-mods=~/MITgcm/verification/exp2/code |
| 96 |
|
% make depend |
| 97 |
|
% make |
| 98 |
|
\end{verbatim} |
| 99 |
|
To run the model here, you'll need the input files: |
| 100 |
|
\begin{verbatim} |
| 101 |
|
% cp ~/MITgcm/verification/exp2/input/* ./ |
| 102 |
|
% ./mitgcmuv > output.txt |
| 103 |
|
\end{verbatim} |
| 104 |
|
|
| 105 |
|
As before, you could build in one directory and make multiple runs of |
| 106 |
|
the one experiment: |
| 107 |
|
\begin{verbatim} |
| 108 |
|
% cd /scratch/exp2 |
| 109 |
|
% mkdir build |
| 110 |
|
% cd build |
| 111 |
|
% ~/MITgcm/tools/genmake2 -rootdir=~/MITgcm \ |
| 112 |
|
-mods=~/MITgcm/verification/exp2/code |
| 113 |
|
% make depend |
| 114 |
|
% make |
| 115 |
|
% cd ../ |
| 116 |
|
% cp -r ~/MITgcm/verification/exp2/input run2 |
| 117 |
|
% cd run2 |
| 118 |
|
% ./mitgcmuv > output.txt |
| 119 |
|
\end{verbatim} |
| 120 |
|
|
| 121 |
|
|
| 122 |
|
\subsection{Using \texttt{genmake2}} |
| 123 |
|
\label{sect:genmake} |
| 124 |
|
|
| 125 |
|
To compile the code, first use the program \texttt{genmake2} (located |
| 126 |
|
in the \texttt{tools} directory) to generate a Makefile. |
| 127 |
|
\texttt{genmake2} is a shell script written to work with all |
| 128 |
|
``sh''--compatible shells including bash v1, bash v2, and Bourne. |
| 129 |
|
Internally, \texttt{genmake2} determines the locations of needed |
| 130 |
|
files, the compiler, compiler options, libraries, and Unix tools. It |
| 131 |
|
relies upon a number of ``optfiles'' located in the |
| 132 |
|
\texttt{tools/build\_options} directory. |
| 133 |
|
|
| 134 |
|
The purpose of the optfiles is to provide all the compilation options |
| 135 |
|
for particular ``platforms'' (where ``platform'' roughly means the |
| 136 |
|
combination of the hardware and the compiler) and code configurations. |
| 137 |
|
Given the combinations of possible compilers and library dependencies |
| 138 |
|
({\it eg.} MPI and NetCDF) there may be numerous optfiles available |
| 139 |
|
for a single machine. The naming scheme for the majority of the |
| 140 |
|
optfiles shipped with the code is |
| 141 |
|
\begin{center} |
| 142 |
|
{\bf OS\_HARDWARE\_COMPILER } |
| 143 |
|
\end{center} |
| 144 |
|
where |
| 145 |
|
\begin{description} |
| 146 |
|
\item[OS] is the name of the operating system (generally the |
| 147 |
|
lower-case output of the {\tt 'uname'} command) |
| 148 |
|
\item[HARDWARE] is a string that describes the CPU type and |
| 149 |
|
corresponds to output from the {\tt 'uname -m'} command: |
| 150 |
|
\begin{description} |
| 151 |
|
\item[ia32] is for ``x86'' machines such as i386, i486, i586, i686, |
| 152 |
|
and athlon |
| 153 |
|
\item[ia64] is for Intel IA64 systems (eg. Itanium, Itanium2) |
| 154 |
|
\item[amd64] is AMD x86\_64 systems |
| 155 |
|
\item[ppc] is for Mac PowerPC systems |
| 156 |
|
\end{description} |
| 157 |
|
\item[COMPILER] is the compiler name (generally, the name of the |
| 158 |
|
FORTRAN executable) |
| 159 |
|
\end{description} |
| 160 |
|
|
| 161 |
|
In many cases, the default optfiles are sufficient and will result in |
| 162 |
|
usable Makefiles. However, for some machines or code configurations, |
| 163 |
|
new ``optfiles'' must be written. To create a new optfile, it is |
| 164 |
|
generally best to start with one of the defaults and modify it to suit |
| 165 |
|
your needs. Like \texttt{genmake2}, the optfiles are all written |
| 166 |
|
using a simple ``sh''--compatible syntax. While nearly all variables |
| 167 |
|
used within \texttt{genmake2} may be specified in the optfiles, the |
| 168 |
|
critical ones that should be defined are: |
| 169 |
|
|
| 170 |
|
\begin{description} |
| 171 |
|
\item[FC] the FORTRAN compiler (executable) to use |
| 172 |
|
\item[DEFINES] the command-line DEFINE options passed to the compiler |
| 173 |
|
\item[CPP] the C pre-processor to use |
| 174 |
|
\item[NOOPTFLAGS] options flags for special files that should not be |
| 175 |
|
optimized |
| 176 |
|
\end{description} |
| 177 |
|
|
| 178 |
|
For example, the optfile for a typical Red Hat Linux machine (``ia32'' |
| 179 |
|
architecture) using the GCC (g77) compiler is |
| 180 |
|
\begin{verbatim} |
| 181 |
|
FC=g77 |
| 182 |
|
DEFINES='-D_BYTESWAPIO -DWORDLENGTH=4' |
| 183 |
|
CPP='cpp -traditional -P' |
| 184 |
|
NOOPTFLAGS='-O0' |
| 185 |
|
# For IEEE, use the "-ffloat-store" option |
| 186 |
|
if test "x$IEEE" = x ; then |
| 187 |
|
FFLAGS='-Wimplicit -Wunused -Wuninitialized' |
| 188 |
|
FOPTIM='-O3 -malign-double -funroll-loops' |
| 189 |
|
else |
| 190 |
|
FFLAGS='-Wimplicit -Wunused -ffloat-store' |
| 191 |
|
FOPTIM='-O0 -malign-double' |
| 192 |
|
fi |
| 193 |
|
\end{verbatim} |
| 194 |
|
|
| 195 |
|
If you write an optfile for an unrepresented machine or compiler, you |
| 196 |
|
are strongly encouraged to submit the optfile to the MITgcm project |
| 197 |
|
for inclusion. Please send the file to the |
| 198 |
|
\begin{rawhtml} <A href="mail-to:MITgcm-support@mitgcm.org"> \end{rawhtml} |
| 199 |
|
\begin{center} |
| 200 |
|
MITgcm-support@mitgcm.org |
| 201 |
|
\end{center} |
| 202 |
|
\begin{rawhtml} </A> \end{rawhtml} |
| 203 |
|
mailing list. |
| 204 |
|
|
| 205 |
|
In addition to the optfiles, \texttt{genmake2} supports a number of |
| 206 |
|
helpful command-line options. A complete list of these options can be |
| 207 |
|
obtained from: |
| 208 |
|
\begin{verbatim} |
| 209 |
|
% genmake2 -h |
| 210 |
|
\end{verbatim} |
| 211 |
|
|
| 212 |
|
The most important command-line options are: |
| 213 |
|
\begin{description} |
| 214 |
|
|
| 215 |
|
\item[\texttt{--optfile=/PATH/FILENAME}] specifies the optfile that |
| 216 |
|
should be used for a particular build. |
| 217 |
|
|
| 218 |
|
If no "optfile" is specified (either through the command line or the |
| 219 |
|
MITGCM\_OPTFILE environment variable), genmake2 will try to make a |
| 220 |
|
reasonable guess from the list provided in {\em |
| 221 |
|
tools/build\_options}. The method used for making this guess is |
| 222 |
|
to first determine the combination of operating system and hardware |
| 223 |
|
(eg. "linux\_ia32") and then find a working FORTRAN compiler within |
| 224 |
|
the user's path. When these three items have been identified, |
| 225 |
|
genmake2 will try to find an optfile that has a matching name. |
| 226 |
|
|
| 227 |
|
\item[\texttt{--pdefault='PKG1 PKG2 PKG3 ...'}] specifies the default |
| 228 |
|
set of packages to be used. The normal order of precedence for |
| 229 |
|
packages is as follows: |
| 230 |
|
\begin{enumerate} |
| 231 |
|
\item If available, the command line (\texttt{--pdefault}) settings |
| 232 |
|
over-rule any others. |
| 233 |
|
|
| 234 |
|
\item Next, \texttt{genmake2} will look for a file named |
| 235 |
|
``\texttt{packages.conf}'' in the local directory or in any of the |
| 236 |
|
directories specified with the \texttt{--mods} option. |
| 237 |
|
|
| 238 |
|
\item Finally, if neither of the above are available, |
| 239 |
|
\texttt{genmake2} will use the \texttt{/pkg/pkg\_default} file. |
| 240 |
|
\end{enumerate} |
| 241 |
|
|
| 242 |
|
\item[\texttt{--pdepend=/PATH/FILENAME}] specifies the dependency file |
| 243 |
|
used for packages. |
| 244 |
|
|
| 245 |
|
If not specified, the default dependency file {\em pkg/pkg\_depend} |
| 246 |
|
is used. The syntax for this file is parsed on a line-by-line basis |
| 247 |
|
where each line containes either a comment ("\#") or a simple |
| 248 |
|
"PKGNAME1 (+|-)PKGNAME2" pairwise rule where the "+" or "-" symbol |
| 249 |
|
specifies a "must be used with" or a "must not be used with" |
| 250 |
|
relationship, respectively. If no rule is specified, then it is |
| 251 |
|
assumed that the two packages are compatible and will function |
| 252 |
|
either with or without each other. |
| 253 |
|
|
| 254 |
|
\item[\texttt{--adof=/path/to/file}] specifies the "adjoint" or |
| 255 |
|
automatic differentiation options file to be used. The file is |
| 256 |
|
analogous to the ``optfile'' defined above but it specifies |
| 257 |
|
information for the AD build process. |
| 258 |
|
|
| 259 |
|
The default file is located in {\em |
| 260 |
|
tools/adjoint\_options/adjoint\_default} and it defines the "TAF" |
| 261 |
|
and "TAMC" compilers. An alternate version is also available at |
| 262 |
|
{\em tools/adjoint\_options/adjoint\_staf} that selects the newer |
| 263 |
|
"STAF" compiler. As with any compilers, it is helpful to have their |
| 264 |
|
directories listed in your {\tt \$PATH} environment variable. |
| 265 |
|
|
| 266 |
|
\item[\texttt{--mods='DIR1 DIR2 DIR3 ...'}] specifies a list of |
| 267 |
|
directories containing ``modifications''. These directories contain |
| 268 |
|
files with names that may (or may not) exist in the main MITgcm |
| 269 |
|
source tree but will be overridden by any identically-named sources |
| 270 |
|
within the ``MODS'' directories. |
| 271 |
|
|
| 272 |
|
The order of precedence for this "name-hiding" is as follows: |
| 273 |
|
\begin{itemize} |
| 274 |
|
\item ``MODS'' directories (in the order given) |
| 275 |
|
\item Packages either explicitly specified or provided by default |
| 276 |
|
(in the order given) |
| 277 |
|
\item Packages included due to package dependencies (in the order |
| 278 |
|
that that package dependencies are parsed) |
| 279 |
|
\item The "standard dirs" (which may have been specified by the |
| 280 |
|
``-standarddirs'' option) |
| 281 |
|
\end{itemize} |
| 282 |
|
|
| 283 |
|
\item[\texttt{--mpi}] This option enables certain MPI features (using |
| 284 |
|
CPP \texttt{\#define}s) within the code and is necessary for MPI |
| 285 |
|
builds (see Section \ref{sect:mpi-build}). |
| 286 |
|
|
| 287 |
|
\item[\texttt{--make=/path/to/gmake}] Due to the poor handling of |
| 288 |
|
soft-links and other bugs common with the \texttt{make} versions |
| 289 |
|
provided by commercial Unix vendors, GNU \texttt{make} (sometimes |
| 290 |
|
called \texttt{gmake}) should be preferred. This option provides a |
| 291 |
|
means for specifying the make executable to be used. |
| 292 |
|
|
| 293 |
|
\item[\texttt{--bash=/path/to/sh}] On some (usually older UNIX) |
| 294 |
|
machines, the ``bash'' shell is unavailable. To run on these |
| 295 |
|
systems, \texttt{genmake2} can be invoked using an ``sh'' (that is, |
| 296 |
|
a Bourne, POSIX, or compatible) shell. The syntax in these |
| 297 |
|
circumstances is: |
| 298 |
|
\begin{center} |
| 299 |
|
\texttt{\% /bin/sh genmake2 -bash=/bin/sh [...options...]} |
| 300 |
|
\end{center} |
| 301 |
|
where \texttt{/bin/sh} can be replaced with the full path and name |
| 302 |
|
of the desired shell. |
| 303 |
|
|
| 304 |
|
\end{description} |
| 305 |
|
|
| 306 |
|
|
| 307 |
|
\subsection{Building with MPI} |
| 308 |
|
\label{sect:mpi-build} |
| 309 |
|
|
| 310 |
|
Building MITgcm to use MPI libraries can be complicated due to the |
| 311 |
|
variety of different MPI implementations available, their dependencies |
| 312 |
|
or interactions with different compilers, and their often ad-hoc |
| 313 |
|
locations within file systems. For these reasons, its generally a |
| 314 |
|
good idea to start by finding and reading the documentation for your |
| 315 |
|
machine(s) and, if necessary, seeking help from your local systems |
| 316 |
|
administrator. |
| 317 |
|
|
| 318 |
|
The steps for building MITgcm with MPI support are: |
| 319 |
|
\begin{enumerate} |
| 320 |
|
|
| 321 |
|
\item Determine the locations of your MPI-enabled compiler and/or MPI |
| 322 |
|
libraries and put them into an options file as described in Section |
| 323 |
|
\ref{sect:genmake}. One can start with one of the examples in: |
| 324 |
|
\begin{rawhtml} <A |
| 325 |
|
href="http://mitgcm.org/cgi-bin/viewcvs.cgi/MITgcm/tools/build_options/"> |
| 326 |
|
\end{rawhtml} |
| 327 |
|
\begin{center} |
| 328 |
|
\texttt{MITgcm/tools/build\_options/} |
| 329 |
|
\end{center} |
| 330 |
|
\begin{rawhtml} </A> \end{rawhtml} |
| 331 |
|
such as \texttt{linux\_ia32\_g77+mpi\_cg01} or |
| 332 |
|
\texttt{linux\_ia64\_efc+mpi} and then edit it to suit the machine at |
| 333 |
|
hand. You may need help from your user guide or local systems |
| 334 |
|
administrator to determine the exact location of the MPI libraries. |
| 335 |
|
If libraries are not installed, MPI implementations and related |
| 336 |
|
tools are available including: |
| 337 |
|
\begin{itemize} |
| 338 |
|
\item \begin{rawhtml} <A |
| 339 |
|
href="http://www-unix.mcs.anl.gov/mpi/mpich/"> |
| 340 |
|
\end{rawhtml} |
| 341 |
|
MPICH |
| 342 |
|
\begin{rawhtml} </A> \end{rawhtml} |
| 343 |
|
|
| 344 |
|
\item \begin{rawhtml} <A |
| 345 |
|
href="http://www.lam-mpi.org/"> |
| 346 |
|
\end{rawhtml} |
| 347 |
|
LAM/MPI |
| 348 |
|
\begin{rawhtml} </A> \end{rawhtml} |
| 349 |
|
|
| 350 |
|
\item \begin{rawhtml} <A |
| 351 |
|
href="http://www.osc.edu/~pw/mpiexec/"> |
| 352 |
|
\end{rawhtml} |
| 353 |
|
MPIexec |
| 354 |
|
\begin{rawhtml} </A> \end{rawhtml} |
| 355 |
|
\end{itemize} |
| 356 |
|
|
| 357 |
|
\item Build the code with the \texttt{genmake2} \texttt{-mpi} option |
| 358 |
|
(see Section \ref{sect:genmake}) using commands such as: |
| 359 |
|
{\footnotesize \begin{verbatim} |
| 360 |
|
% ../../../tools/genmake2 -mods=../code -mpi -of=YOUR_OPTFILE |
| 361 |
|
% make depend |
| 362 |
|
% make |
| 363 |
|
\end{verbatim} } |
| 364 |
|
|
| 365 |
|
\item Run the code with the appropriate MPI ``run'' or ``exec'' |
| 366 |
|
program provided with your particular implementation of MPI. |
| 367 |
|
Typical MPI packages such as MPICH will use something like: |
| 368 |
|
\begin{verbatim} |
| 369 |
|
% mpirun -np 4 -machinefile mf ./mitgcmuv |
| 370 |
|
\end{verbatim} |
| 371 |
|
Sightly more complicated scripts may be needed for many machines |
| 372 |
|
since execution of the code may be controlled by both the MPI |
| 373 |
|
library and a job scheduling and queueing system such as PBS, |
| 374 |
|
LoadLeveller, Condor, or any of a number of similar tools. A few |
| 375 |
|
example scripts (those used for our \begin{rawhtml} <A |
| 376 |
|
href="http://mitgcm.org/testing.html"> \end{rawhtml}regular |
| 377 |
|
verification runs\begin{rawhtml} </A> \end{rawhtml}) are available |
| 378 |
|
at: |
| 379 |
|
\begin{rawhtml} <A |
| 380 |
|
href="http://mitgcm.org/cgi-bin/viewcvs.cgi/MITgcm_contrib/test_scripts/"> |
| 381 |
|
\end{rawhtml} |
| 382 |
|
{\footnotesize \tt |
| 383 |
|
http://mitgcm.org/cgi-bin/viewcvs.cgi/MITgcm\_contrib/test\_scripts/ } |
| 384 |
|
\begin{rawhtml} </A> \end{rawhtml} |
| 385 |
|
|
| 386 |
|
\end{enumerate} |
| 387 |
|
|
| 388 |
|
An example of the above process on the MITgcm cluster (``cg01'') using |
| 389 |
|
the GNU g77 compiler and the mpich MPI library is: |
| 390 |
|
|
| 391 |
|
{\footnotesize \begin{verbatim} |
| 392 |
|
% cd MITgcm/verification/exp5 |
| 393 |
|
% mkdir build |
| 394 |
|
% cd build |
| 395 |
|
% ../../../tools/genmake2 -mpi -mods=../code \ |
| 396 |
|
-of=../../../tools/build_options/linux_ia32_g77+mpi_cg01 |
| 397 |
|
% make depend |
| 398 |
|
% make |
| 399 |
|
% cd ../input |
| 400 |
|
% /usr/local/pkg/mpi/mpi-1.2.4..8a-gm-1.5/g77/bin/mpirun.ch_gm \ |
| 401 |
|
-machinefile mf --gm-kill 5 -v -np 2 ../build/mitgcmuv |
| 402 |
|
\end{verbatim} } |
| 403 |
|
|
| 404 |
\subsection{Configuration and setup} |
\subsection{Configuration and setup} |
| 405 |
|
|
| 406 |
The CPP keys relative to the ``numerical model'' part of the code are |
The CPP keys relative to the ``numerical model'' part of the code are |
Parent Directory
| 
Revision Log
| 
Revision Graph
| 
Patch