IOSim performance improvement

[coot]

IOSim

Benchmarking results trying out the improvement suggested here ekmett/kan-extensions#79 (comment)

no changes

```hs
(>>=) = \dm f -> IOSim $ oneShot $ \k -> unIOSim dm (\m -> unIOSim (f m) k)

benchmarking delays/threadDelay
time 6.748 ns (6.531 ns .. 6.906 ns)
0.993 R² (0.989 R² .. 0.996 R²)
mean 6.789 ns (6.570 ns .. 6.963 ns)
std dev 670.1 ps (572.9 ps .. 798.3 ps)
variance introduced by outliers: 92% (severely inflated)

benchmarking delays/registerDelay
time 6.660 ns (6.293 ns .. 6.940 ns)
0.987 R² (0.976 R² .. 0.996 R²)
mean 6.367 ns (6.158 ns .. 6.607 ns)
std dev 751.5 ps (676.8 ps .. 841.8 ps)
variance introduced by outliers: 94% (severely inflated)

benchmarking delays/timeout/fail
time 6.343 ns (6.276 ns .. 6.414 ns)
0.997 R² (0.995 R² .. 0.999 R²)
mean 6.511 ns (6.359 ns .. 6.686 ns)
std dev 565.7 ps (460.3 ps .. 729.9 ps)
variance introduced by outliers: 90% (severely inflated)

benchmarking delays/timeout/succeed
time 7.429 ns (7.391 ns .. 7.475 ns)
1.000 R² (1.000 R² .. 1.000 R²)
mean 7.460 ns (7.421 ns .. 7.532 ns)
std dev 165.1 ps (100.6 ps .. 294.7 ps)
variance introduced by outliers: 36% (moderately inflated)

benchmarking delays/timeout/race
time 6.750 ns (6.680 ns .. 6.848 ns)
0.999 R² (0.999 R² .. 1.000 R²)
mean 6.770 ns (6.746 ns .. 6.806 ns)
std dev 100.0 ps (66.80 ps .. 166.5 ps)
variance introduced by outliers: 20% (moderately inflated)

benchmarking threads/50/async silent
time 6.735 ns (6.715 ns .. 6.760 ns)
1.000 R² (1.000 R² .. 1.000 R²)
mean 6.733 ns (6.713 ns .. 6.760 ns)
std dev 78.88 ps (57.27 ps .. 96.87 ps)
variance introduced by outliers: 14% (moderately inflated)

benchmarking threads/50/forkIO silent
time 6.085 ns (5.995 ns .. 6.155 ns)
0.999 R² (0.998 R² .. 1.000 R²)
mean 6.128 ns (6.087 ns .. 6.197 ns)
std dev 175.4 ps (73.11 ps .. 320.4 ps)
variance introduced by outliers: 49% (moderately inflated)

benchmarking threads/50/threadDelay bottleneck silent
time 7.245 ns (7.069 ns .. 7.407 ns)
0.997 R² (0.995 R² .. 0.999 R²)
mean 7.218 ns (7.095 ns .. 7.289 ns)
std dev 304.3 ps (191.7 ps .. 474.7 ps)
variance introduced by outliers: 67% (severely inflated)

benchmarking threads/50/async say
time 148.2 ns (143.2 ns .. 152.6 ns)
0.990 R² (0.982 R² .. 0.996 R²)
mean 146.2 ns (141.7 ns .. 150.0 ns)
std dev 14.58 ns (9.861 ns .. 19.57 ns)
variance introduced by outliers: 91% (severely inflated)

benchmarking threads/50/forkIO say
time 10.94 ns (10.38 ns .. 11.77 ns)
0.983 R² (0.978 R² .. 0.992 R²)
mean 11.85 ns (11.51 ns .. 12.09 ns)
std dev 971.7 ps (707.9 ps .. 1.308 ns)
variance introduced by outliers: 89% (severely inflated)

benchmarking threads/50/threadDelay bottleneck say
time 2.720 ms (2.619 ms .. 2.820 ms)
0.984 R² (0.973 R² .. 0.993 R²)
mean 2.507 ms (2.388 ms .. 2.584 ms)
std dev 306.2 μs (221.2 μs .. 376.1 μs)
variance introduced by outliers: 76% (severely inflated)

benchmarking threads/250/async
time 5.781 ns (5.600 ns .. 5.927 ns)
0.991 R² (0.985 R² .. 0.996 R²)
mean 5.859 ns (5.638 ns .. 6.089 ns)
std dev 767.3 ps (662.8 ps .. 882.1 ps)
variance introduced by outliers: 96% (severely inflated)

benchmarking threads/250/forkIO
time 6.219 ns (6.204 ns .. 6.236 ns)
1.000 R² (1.000 R² .. 1.000 R²)
mean 6.210 ns (6.198 ns .. 6.236 ns)
std dev 55.88 ps (32.82 ps .. 90.82 ps)

## improved version
```hs
(>>=) = \dm f -> IOSim $ oneShot $ \k -> unIOSim dm (oneShot $ \m -> unIOSim (f m) k)

benchmarking delays/threadDelay
time                 7.477 ns   (7.434 ns .. 7.538 ns)
                     0.999 R²   (0.999 R² .. 1.000 R²)
mean                 7.494 ns   (7.448 ns .. 7.615 ns)
std dev              239.9 ps   (104.7 ps .. 459.3 ps)
variance introduced by outliers: 54% (severely inflated)

benchmarking delays/registerDelay
time                 6.705 ns   (6.326 ns .. 7.134 ns)
                     0.985 R²   (0.978 R² .. 0.997 R²)
mean                 6.976 ns   (6.765 ns .. 7.187 ns)
std dev              687.1 ps   (524.6 ps .. 997.8 ps)
variance introduced by outliers: 92% (severely inflated)

benchmarking delays/timeout/fail
time                 6.868 ns   (6.615 ns .. 7.019 ns)
                     0.996 R²   (0.996 R² .. 0.997 R²)
mean                 6.558 ns   (6.474 ns .. 6.666 ns)
std dev              320.9 ps   (229.8 ps .. 394.7 ps)
variance introduced by outliers: 74% (severely inflated)

benchmarking delays/timeout/succeed
time                 7.548 ns   (7.485 ns .. 7.629 ns)
                     0.999 R²   (0.999 R² .. 1.000 R²)
mean                 7.526 ns   (7.491 ns .. 7.589 ns)
std dev              153.0 ps   (109.5 ps .. 220.9 ps)
variance introduced by outliers: 32% (moderately inflated)

benchmarking delays/timeout/race
time                 7.480 ns   (7.459 ns .. 7.506 ns)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 7.472 ns   (7.464 ns .. 7.484 ns)
std dev              31.95 ps   (19.80 ps .. 49.13 ps)

benchmarking threads/50/async silent
time                 5.955 ns   (5.948 ns .. 5.963 ns)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 5.955 ns   (5.951 ns .. 5.961 ns)
std dev              15.28 ps   (10.90 ps .. 20.70 ps)

benchmarking threads/50/forkIO silent
time                 5.959 ns   (5.905 ns .. 5.990 ns)
                     1.000 R²   (0.999 R² .. 1.000 R²)
mean                 5.954 ns   (5.921 ns .. 5.965 ns)
std dev              62.33 ps   (24.32 ps .. 124.2 ps)
variance introduced by outliers: 11% (moderately inflated)

benchmarking threads/50/threadDelay bottleneck silent
time                 6.915 ns   (6.706 ns .. 7.139 ns)
                     0.995 R²   (0.993 R² .. 0.998 R²)
mean                 7.121 ns   (7.000 ns .. 7.197 ns)
std dev              310.3 ps   (185.5 ps .. 445.8 ps)
variance introduced by outliers: 69% (severely inflated)

benchmarking threads/50/async say
time                 144.8 ns   (141.7 ns .. 147.2 ns)
                     0.995 R²   (0.992 R² .. 0.997 R²)
mean                 142.4 ns   (138.8 ns .. 144.2 ns)
std dev              7.526 ns   (4.104 ns .. 13.53 ns)
variance introduced by outliers: 73% (severely inflated)

benchmarking threads/50/forkIO say
time                 10.02 ns   (9.748 ns .. 10.41 ns)
                     0.988 R²   (0.984 R² .. 0.993 R²)
mean                 10.90 ns   (10.58 ns .. 11.22 ns)
std dev              1.145 ns   (1.018 ns .. 1.310 ns)
variance introduced by outliers: 93% (severely inflated)

benchmarking threads/50/threadDelay bottleneck say
time                 2.694 ms   (2.608 ms .. 2.760 ms)
                     0.982 R²   (0.964 R² .. 0.993 R²)
mean                 2.547 ms   (2.445 ms .. 2.617 ms)
std dev              279.7 μs   (193.0 μs .. 362.2 μs)
variance introduced by outliers: 70% (severely inflated)

benchmarking threads/250/async
time                 5.420 ns   (5.190 ns .. 5.747 ns)
                     0.976 R²   (0.968 R² .. 0.985 R²)
mean                 6.050 ns   (5.767 ns .. 6.288 ns)
std dev              877.1 ps   (800.2 ps .. 967.3 ps)
variance introduced by outliers: 96% (severely inflated)

benchmarking threads/250/forkIO
time                 6.570 ns   (6.360 ns .. 6.758 ns)
                     0.994 R²   (0.993 R² .. 0.996 R²)
mean                 6.207 ns   (6.053 ns .. 6.354 ns)
std dev              482.9 ps   (369.3 ps .. 600.6 ps)
variance introduced by outliers: 88% (severely inflated)

Remakrs

The benchmarking threads/50/async silent improved from 6.7ns to 5.955ns, but benchmarking threads/250/async silent only improved from 5.781ns to 5.420ns.