INSTALL

Caveat: Swan can be compiled correctly ONLY with GCC 4.6 at this stage.

Installation instructions:
$ git clone or unpack tar-ball into directory swan
$ cd swan
$ mkdir obj
$ cd obj
$ ../configure CC=gcc-mp-4.6 CXX=g++-mp-4.6

To build in debugging mode:
$ OPT="-O0 -DMODE=0" make
To build in performance mode:
$ OPT="-O4 -DMODE=1" make

Always do a full cleanup and recompile when changing the value of -DMODE={0,1}

Don't use the -jN option of make at the moment.

There are a number of tests in the the swan/tests directory.
You can check that things work correctly. E.g. tests/explain is
parallel fibonacci:
$ cd swan/obj/tests
$ NUM_THREADS=1 time ./explain 35
$ NUM_THREADS=2 time ./explain 35
$ NUM_THREADS=4 time ./explain 35
Should not dump core, should print the correct number (you know the
35th Fibonacci number, don't you?) and should show good linear scaling
(provided that you have the cores available).

Alternatives:
When you observe segmentation faults inside the longjmp() function, this is
most likely due to GCC's FORTIFY_SOURCE option. You can disable it by adding
the --disable-fortify-source option to configure:
$ cd to/swan/obj directory
$ ../configure CC=gcc-4.6 CXX=g++-4.6 --disable-fortify-source
then cleanup
$ make clean
and then repeat all other steps