Sundials Solver Options
#######################

Selecting Precision
-------------------
Sundials can be built with different precisions (currently `'single'` which maps
to C `'float'`; `'double'` (the default) which maps to C `'double'`; and
`'extended'` which maps to C `'long double'`), and scikits.odes supports using
whichever precision Sundials is built with. To take advantage of this, you must
first have sundials built and installed with the desired precision setting.

Once you have done this, build scikits.odes against this particular version of
sundials (see the main documentation for how to do this). scikits.odes will
automatically detect the precision, and store this in a variable called
:py:const:`DTYPE`. :py:const:`DTYPE` should be accessed from
:py:mod:`scikits.odes.sundials` (other modules may have :py:const:`DTYPE`
defined, but :py:mod:`scikits.odes.sundials` should be preferred). Additionally
:py:const:`scikits.odes.sundials.precision` contains the precision setting found
by scikits.odes.

To use :py:const:`DTYPE`, treat it as a numpy dtype; use it whenever you need to
create an array::

    np.array([1,2,3], dtype=DTYPE)

or when using scalars::

    DTYPE(0.1) * DTYPE(0.1)