Bad random numbers, or TIMER limitation...?

[Statsman1]

@Richard ...

Thank you very much for the code.  I modified it to behave somewhat in the way I am trying to do my thing...

Code: QB64: [Select]

1. version$ = "Precision_timer-999"
2. Declare Dynamic Library "kernel32" ' Acquiring high-resolution time stamps  ' https://docs.microsoft.com/en-us/windows/win32/sysinfo/acquiring-high-resolution-time-stamps
3.     Sub QueryPerformanceCounter (ByVal lpPerformanceCount As _Unsigned _Offset)
4.     Sub QueryPerformanceFrequency (ByVal lpFrequency As _Unsigned _Offset)
5. End Declare
6. Dim Shared PerformanceCount As _Integer64
7. Dim Shared Frequency As _Integer64
8. Dim Shared StartingCount As _Integer64
9. Dim Shared dElapsedTime As Double
10. Dim Shared dElapsedTimeOLD As Double
11.  
12. For reps = 1 To 20
13.     QueryPerformanceFrequency (_Offset(Frequency))
14.     QueryPerformanceCounter (_Offset(StartingCount))
15.  
16.     TIMERe: dElapsedTimeOLD = dElapsedTime
17.  
18.     dist = 1440
19.     draws = 0
20.     Do
21.  
22.         steps = Int(Rnd(1) * 3) + 1
23.         dist = dist - steps
24.         draws = draws + 1
25.  
26.     Loop Until dist < 0
27.  
28.     TIMERe
29.  
30.     Print "Time to take 1440 steps =";: Print Using "####.###"; (dElapsedTime - dElapsedTimeOLD) * 1000000;: Print "    Loops =";: Print draws
31.  
32. Next reps
33.  
34. _Delay 10: System: End
35.  
36. Sub TIMERe
37.     QueryPerformanceCounter (_Offset(PerformanceCount))
38.     dElapsedTime = (CDbl(PerformanceCount - StartingCount)) / Frequency
39. End Sub

...and attached the results in the first attachment.  I then did the same for 14,400 steps, if you will, and the results are in the second attachment.

Those outliers that you talk about - I guess that can be lived with, as that would impact every horse in a race, not just one.  I am encouraged by the results of the second run - the more loops, the more varied the times seem to be, even for the same number of loops to complete the test.

[Richard]

@Statsman1

Thanks for letting me know about your success with the program.

Just for your info, the two lines immediately following the FOR reps  =  1 to 20 statement can be moved to just outside the loop (i.e. place line before FOR...)

The "frequency" relates to a conversion factor to convert (eventually) computer crystal cycles to 1 second counts - as it is highly unlikely that you would change the clock speed of the computer (or that the processor + hardware is even capable to achieve this on "older computers") - it only needs to be "determined once" (I think the value for all MS compatible computers is 1000000 (or 10000000) - it is actually not determined by windows in any manner I think).

The "StartingCount" I believe relates to clock counts (as 100 nano-second counts) since the computer was restarted (or is it since JAN 1 1972 or something)? In either case for your computer session (i.e. since you did a restart) - likewise it only has to be established once per run.

In fact because you are measuring a time interval of an event (race time) - you are only working with the DIFFERENCE (from start to finish) - so the subtraction of StartingCount (in both TIMERe instances cancel out mathematically). Similarly with Frequency (which can  be taken as being CONSTANT) - that too can be taken out (i.e. program calculate using number of 100 nanosecond counts (= Performance count) and when you finally report (print/display the results then divide Difference in PerformanceCounts by 10000000 to convert to seconds. You may gain a timing improvement of 1 to 10 nanoseconds!. :)

As far as the outliers go - and I am still trying to find a work-around to the problem - in my case (referring to reply # 180 mentioned above) - the outlier WAS usually NOT the first timing in any of my loops (which had a number of code line between timing points) - which is a "killer" in trying to determine which code module was "faster".

So if the outlier occurred within your group of events (rather than start) - it might have the effect (say) of a horse "jumping the rails" to take a short cut to the finish line (or to go back to the starting gate). :)