Matrix multiply host code is too complicated

[tomdeakin]

Provide a matrix multiply host code to just run a single kernel, rather than multiple kernels along with the serial version.

[Smiril]

how to run multiple kernels and blocks so on threads
mean: 256 * 256 * 256 once start