Currently, detect_ocl_gpus implements a dummy sorting based on device vendor.
The sorting must be improved and should rely on performance.
Check if evaluating the performance from detect_ocl_gpus can include launching and comparing the performance for the nbnxn kernels that are going to be used for the requested simulation.
Currently, detect_ocl_gpus implements a dummy sorting based on device vendor. The sorting must be improved and should rely on performance. Check if evaluating the performance from detect_ocl_gpus can include launching and comparing the performance for the nbnxn kernels that are going to be used for the requested simulation.