No vectorization for simple loop

[efocht]

The example examples/vadd.jl should vectorize, but the look at the output of @device_code_native ... shows no vector instruction(s).

There are two reasons for this behavior:

• in our LLVM source tree LoopVectorizer is disabled by means of cost function in order to not interfere with the RegionVectorizer (RV).
• Julia GPUCompiler uses an own set of optimization passes which does not call RV.

Either reenable LoopVectorizer or call RV in the optimization step.

[efocht]

TODO: make vectorization test.