Uses of sizeof(T) for local memory sizes is dangerous with integers, hence promoting type to OpenCL types via metafunction.

Causes compilation on GCC 4.4 to fail.

Integration of random number generation is postponed until we know how to deal with the 'other' backends.
The removal of the auxiliary/ folder now allows for a direct parallel builds (e.g. `make -j4`)
and makes the developer repository much more accessible for Windows users. Hurray! :-)

Execution of tutorial/spai fails on AMD APP SDK when using an APU. Check with GPU required.

Matrix-vector product as well as host->device copy for alignment > 1 was broken. This is now fixed as well.

Found a bug when running blas3_solve-tests on CPUs via OpenCL.

Includes minor code cleaning

This is necessary since pow() is only defined for floating point types in the OpenCL standard.

This is particularly handy for storing 'off-processor' matrices in an MPI setting.

Contains the add of TCLAP (Header-Only, MIT Licensed) command line parsing library

Other changes in the template seem to have resolved the issue with NVidia

From 1.2 ~ 2.6TFLOP/s to stable 2.2~2.3 TFLOP/s

Adding support for ell_matrix * dense matrix multiplication