2 ms·

I'm not too sure what the author thinks they did wrong! Perhaps implementing the first attempt before profiling, but... would it have made a difference in this

• Palomides 2y ago
  not using uprof (or vtune for intel users) well seems like a misstep; it can give you the rate of branch prediction success, though that is arguably more a lack of knowledge/experience than a mistake they're cool tools, I encourage anyone to mess with them
  • adgjlsfhk1 2y ago
    branch prediction rates wouldn't have helped here. the problem here is that even though the branch is predicted correctly ~99% of the time, the branch drops you to ipc of 1 because you can't speculate past a 2nd branch
    • Palomides 2y ago
      I think that would be reported as some kind of stall? I can't say I'm an expert either
      • turol 2y ago
        I think vtune and toplev will report that as a frontend stall in their topdown metrics.