Open gmpassos opened 3 years ago
You will need a new benchmark video! ;-P
Super! Will merge your repo to get it earlier :neckbeard: :see_no_evil:
You where performing a random number of shifts, depending of the dirty in the register value (not initialized). This will produce a different number of operations, a not valid benchmark, since you are comparing different scenarios.
You can create a video about benchmarks, and how is difficult to create a scenario that is always reproducible and fair for all tested sides. This issue is a good example of how the same code can produce different scenarios and how is difficult to ensure fairness.
CL register wasn’t initialized before SHR call. Changed to value 1.