Suggestion / feature request
Please add --benchmark parameter.
E.g. for X16r it would benchmark each of the 16 sub-algorithms n times and then report the average hash rate.
In theory this will produce the most accurate hash rate average because over time each sub-algorithm would be run equally often.
Any thoughts?
Most kernels don't have the same timings if they are first (or last). There is also some kernel sequences that reduce timings ("imported" from ccminer X17).
All those optimisation tricks make benchmark inaccurate for X16x