For conversion from floats to halfs we need to start a CUDA-kernel, that executes this conversion.
For this we need a Device capable of CUDA version 6 or higher. Currently we just select any device which could be a problem when multiple devices are available.
When #57 is done and we can list Devices, we could also select a Device with CUDA version 6 or higher before executing the Kernel
For conversion from floats to halfs we need to start a CUDA-kernel, that executes this conversion. For this we need a Device capable of CUDA version 6 or higher. Currently we just select any device which could be a problem when multiple devices are available.
When #57 is done and we can list Devices, we could also select a Device with CUDA version 6 or higher before executing the Kernel