Using type suffix

Fixing CompilerWarning: <program source>: warning: double precision constant requires cl_khr_fp64, casting to single precision
An application that wants to use double will need to include the directive before any double precision data type is declared in the kernel code.

Better checking for stride equality

Ignore dimensions where the shape is 1, since the stride in these
dimensions does not matter. This is necessary when the Numpy flag
``NPY_RELAXED_STRIDES_CHECKING`` is set.

fix issue #130

Fix error when decoding a NULL string

If some device info is NULL  then this line would cause a RuntimeError: cannot use string() on <cdata 'char *' NULL>.

For example, in my PC, the Intel CPU runtime under Linux has BUILT_IN_KERNELS = NULL.

SVM map: Only use __array_interface__.descr if typestr can't do the job (PyPy doesn't yet like descr)