fix for some cases where the extreme memory sizes cause issues

 Fix GL interop on Windows

Fix OverflowError: int too big to convert
Cast unsigned integer Windows handle to signed integer cl_context_properties

ListOfListsBuilder: Ensure count and write kernels get passed the

See merge request !10

correct local work size of 1 when do_not_vectorize() is true.

Closes #175

Scalar types

See merge request !9

Improved documentation for enqueue_fill_buffer and enqueue_copy_buffer

Fixed link to pyopencl-complex.h in docs

I hope its okay to send a PR for such a minor thing.

Fix typo in InclusiveScanKernel arguments

Remove chatty compiler warnings when including pyopencl-complex.h in OSX by inlining all complex valued functions

See: https://github.com/pyopencl/pyopencl/issues/134

Make CommandQueue._get_cl_version device- (rather than platform-) specific (Fixes #168 on github)

See merge request !7

Apple AMD Radeon GPU bug-fix

compiler crash on AMD Radeon crad on Apple platform.
For this structed data pointers, ldata[:][:].value, are
replaced by temporal variables tmp and tmp_aux before passing to SCAN_EXPR.
To avoid leftover-worning tmp_aux was added to_PREFIX_WORDS.

Adapt doctest output to new version of numpy

See merge request !6