Chrono library for high resolution timer

BlitzMax Forums/BlitzMax Programming/Chrono library for high resolution timer

Chapman7

(Posted 2016) [#1]

C++11 has a library called Chrono that deals with time/timers. It has a high resolution one that can process microseconds and nanoseconds. Would it be possible to use that with BlitzMax?

TomToad

(Posted 2016) [#2]

I posted a hi-res timer module in the code archives. High Resolution Timer Module
It is Windows only. Here is another one High Resolution Timers. You might find others with a search.

Chapman7

(Posted 2016) [#3]

I need it to be cross platform :/ i guess ill just play around with it and see if I can come up with anything

Chapman7

(Posted 2016) [#4]

Okay I think I have a working example but I have to add "-std=c++11" or "-std=gnu++11" to the compiler and I can't figure out how... Do I have to build with command line? I tried linking with import inside bmax but it didn't work

Chapman7

(Posted 2016) [#5]

Okay if anyone needs a cross platform high resolution timer, here is the code. I used the latest TDM64 (had to build BlitzMax from source). TDM64 comes with the <chrono> library. I had to use 'ModuleInfo "CC_OPTS: -std=c++11"' for gcc to access the library since its experimental.

Mod name: Pub.Chrono

chrono.bmx:

Strict

Rem
bbdoc: High Resolution Timer
End Rem
Module Pub.Chrono

ModuleInfo "Version 1.0"
ModuleInfo "Author: Cole Chapman"

ModuleInfo "CC_OPTS: -std=c++11"

Import "chronotimer.cpp"

Extern
	Function getmicrosecs:Double()
	Function getnanosecs:Double()
EndExtern

Rem
bbdoc: Get microsecond counter
about: Returns microseconds (as a double) since application startup.
EndRem
Function MicroSecs:Double()
	Return getmicrosecs()
EndFunction

Rem
bbdoc: Get nanosecond counter
about: Returns nanoseconds (as a double) since application startup.
EndRem
Function NanoSecs:Double()
	Return getnanosecs()
EndFunction

chronotimer.cpp:

#include <chrono>

std::chrono::steady_clock::time_point start_time = std::chrono::steady_clock::now();

extern "C" double getmicrosecs()
{
	std::chrono::steady_clock::time_point get_time = std::chrono::steady_clock::now();
    return std::chrono::duration_cast<std::chrono::microseconds>(get_time - start_time).count();
}

extern "C" double getnanosecs()
{
	std::chrono::steady_clock::time_point get_time = std::chrono::steady_clock::now();
    return std::chrono::duration_cast<std::chrono::nanoseconds>(get_time - start_time).count();
}

Example code:

Import Pub.Chrono

Print NanoSecs()
Print NanoSecs()
Print NanoSecs()
Print NanoSecs()
Print NanoSecs()

Print "-------"

Print MicroSecs()
Print MicroSecs()
Print MicroSecs()
Print MicroSecs()
Print MicroSecs()

Output:

Unlike MilliSecs(), MicroSecs() and NanoSecs() get the time since the application has started, not since the system has started.

skidracer

(Posted 2016) [#6]

So from your results, TDM64's implementation is a fail and you will get better accuracy with the Window's Performance Counter which typically has a frequency of 16mHz or better. It would be more clear if your test called Print NanoSecs() three times in a row to illustrate.

On Linux clock_gettime(CLOCK_PROCESS_CPUTIME_ID, &now) will give you a more optimal method of sampling without the C++11 hassle.

Chapman7

(Posted 2016) [#7]

How do you figure it's a fail?
Edit: Seems like you might be correct, ill have to look at it in a little bit.

skidracer

(Posted 2016) [#8]

This suggests steady_clock option in your code may work better on Windows.

Chapman7

(Posted 2016) [#9]

Haha that did it. I was playing around with the c++ code in linux and couldn't for the life of me figure out why it was good there but not when I moved it over and ran it in BMax. I updated the code and the example code/output. I appreciate you helping me out

Chapman7

(Posted 2016) [#10]

Updated the code so the functions produce a double (they were producing and int and it would continually rollover)

I think you can use NanoSecs() for about 38 months before you start losing accuracy... Don't quote me on that.