Confusing numbers in histogram

[giordano]

Trying to reproduce the benchmarks in this discourse post, I've got

ulia> @benchmark $a[$w] .= $b[$w]
BenchmarkTools.Trial: 10000 samples with 1 evaluation.
 Range (min … max):  10.176 μs …  1.150 ms  ┊ GC (min … max): 0.00% … 98.96%
 Time  (median):     10.987 μs              ┊ GC (median):    0.00%
 Time  (mean ± σ):   13.660 μs ± 17.326 μs  ┊ GC (mean ± σ):  3.08% ±  2.71%

  █                                                            
  █▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁▁ ▁
  53.4 μs         Histogram: frequency by time        11.1 μs <

 Memory estimate: 39.33 KiB, allocs estimate: 2.

The mean time is 13.660 μs ± 17.326 μs, I haven't looked at the code but the standard deviation larger than the average time itself looks suspiciously wrong (or maybe large one-sided outliers are screwing it up?). 👈 Yes, I believe the problem is that the distribution isn't normal, there are large outliers only on one side, so the standard deviation is messed up. The issue below still stands though 👇

Also, note that the histogram has 53.4 μs on the left and 11.1 μs on the right: I have no idea how to read this histogram, is it reversed?

[giordano]

Other examples of funny numbers in the histogram: https://discourse.julialang.org/t/why-is-minimum-so-much-faster-than-argmin/66814/8

[tecosaur]

Yea, that's funky. I wonder if it has anything to do with my assumption that times were sorted? In #245 this assumption looks like it could be a bit dodgy.

[vchuravy]

#240 times are not sorted