optimize cn-heavy AMD

- explicit loop unrolling

based on changes in @imperdin fork https://github.com/imperdin/xmr-stak/blob/master/xmrstak/backend/amd/amd_gpu/opencl/cryptonight.cl