Using openmp

The AMRClaw and GeoClaw codes now have openmp directives so that some of the work can be done in parallel on multicore machines.

To compile with openmp

To use openmp, first go into each of the directories $CLAW/amrclaw/2d/lib and $CLAW/geoclaw/2d/lib and modify the Makefile to add the flag -fopenmp to the FFLAGS = line. To also turn on optimization, you might want to use:

FFLAGS ?=  -O2 -fopenmp

Then do the following:

make clean
make .objs

Now go into your application directory and make the same modification to the Makefile as above. Then:

make clean
make .exe

should make a new version of the executable that uses openmp.

Note that if you want to later turn off openmp, you must change the Makefiles to remove -fopenmp from the FFLAGS line and then recompile as above. In particular, if you want to run an example that does not have -fopenmp in the local Makefile then you will have to recompile the appropriate library routines first or you will get error messages when you compile such as:

Undefined symbols:
  "_omp_get_max_threads_", referenced from:
      _MAIN__ in amr2ez.o
      _MAIN__ in amr2ez.o

To run the code

To specify the number of threads that openmp should use, first set the environment variable OMP_NUM_THREADS. For example, in bash:

$ export OMP_NUM_THREADS=2

to use 2 threads. You should set this to at most the number of cores in your machine. If this environment variable is not set, then by default it will probably use the number of cores available, but this may be compiler dependent.

You can put this export statement in your .bashrc file if you want it to be set everytime you open a new shell.

Once the code compiles properly with the -fopenmp flags and you have set the environment variable OMP_NUM_THREADS, you should be able to run it as usual, e.g.

$ make .output