Re: Real Time-ing
On Wednesday, 9 December 2015 at 10:00:46 UTC, Andrea Fontana
wrote:
On Tuesday, 8 December 2015 at 15:35:18 UTC, Taylor Hillegeist
wrote:
So, I mostly do programming that is of run to completion
verity. But I have a dream of calling functions periodically.
So my question is:
What is the best (most time accurate) way to call a function
every n time units?
What is the best way to measure the jitter of these calls?
I'm also interested in waiting vs calling periodically eg.
call
wait(1 ms)
call
is not the same as 1 ms from call to call, due to the time
duration of the function call.
Thanks!
import core.thread;
import std.datetime;
import std.algorithm.comparison;
import std.math;
import std.stdio;
import core.time;
long loopTime = 0;
long CmdTime = 500_000_000; //Time in ns
void main()
{
for(int x = 0; x<20; x++){
auto time = TickDuration.currSystemTick.nsecs;
myPrinter(loopTime);
while(TickDuration.currSystemTick.nsecs - time
< CmdTime){}
loopTime = TickDuration.currSystemTick.nsecs -
time;
}
}
void myPrinter(long time){
writeln(time," nsecs with jitter of :",
abs(CmdTime-time), " nsecs");
}
What about this?
Anyway in order to avoid error accumulation it's probably a good
idea not to reset timer every time if possibile.
If you know the operations you need to perform, probably it works
better to store target time rather than delta.
So instead of:
- move x for 5 seconds
- move y for 3 seconds
- move z for 4 seconds
- ...
I think this works better:
- move x since second 5
- move y since second 8
- move z since second 12
So errors won't accumulate.
Re: Real Time-ing
On Tuesday, 8 December 2015 at 16:40:04 UTC, Taylor Hillegeist wrote: However i seem to get jitter of around 1 ms. Is there anything else i can do to improve? Do you want to get precision better than period of thread switches?
Re: Real Time-ing
On Tuesday, 8 December 2015 at 15:35:18 UTC, Taylor Hillegeist
wrote:
So, I mostly do programming that is of run to completion
verity. But I have a dream of calling functions periodically.
So my question is:
What is the best (most time accurate) way to call a function
every n time units?
What is the best way to measure the jitter of these calls?
I'm also interested in waiting vs calling periodically eg.
call
wait(1 ms)
call
is not the same as 1 ms from call to call, due to the time
duration of the function call.
Thanks!
import core.thread;
import std.datetime;
import std.algorithm.comparison;
import std.math;
import std.stdio;
import core.time;
long loopTime = 0;
long CmdTime = 500_000_000; //Time in ns
void main()
{
for(int x = 0; x<20; x++){
auto time = TickDuration.currSystemTick.nsecs;
myPrinter(loopTime);
while(TickDuration.currSystemTick.nsecs - time <
CmdTime){}
loopTime = TickDuration.currSystemTick.nsecs -
time;
}
}
void myPrinter(long time){
writeln(time," nsecs with jitter of :",
abs(CmdTime-time), " nsecs");
}
What about this?
Re: Real Time-ing
prev=now; call(); wait(prev+dur-now); call();
Re: Real Time-ing
On Tuesday, 8 December 2015 at 15:50:35 UTC, Kagamin wrote: Oops, no. next+=dur; wait(next-now); call(); what calls does this use from the std library? to get the current time? Wait a amount of time?
Real Time-ing
So, I mostly do programming that is of run to completion verity. But I have a dream of calling functions periodically. So my question is: What is the best (most time accurate) way to call a function every n time units? What is the best way to measure the jitter of these calls? I'm also interested in waiting vs calling periodically eg. call wait(1 ms) call is not the same as 1 ms from call to call, due to the time duration of the function call. Thanks!
Re: Real Time-ing
Oops, no. next+=dur; wait(next-now); call();
Re: Real Time-ing
On Tuesday, 8 December 2015 at 15:35:18 UTC, Taylor Hillegeist wrote: So, I mostly do programming that is of run to completion I took a stab at the problem: http://dpaste.dzfl.pl/2eef530d00fc 0 nsecs with jitter of :5 nsecs 498937256 nsecs with jitter of :1062744 nsecs 499036173 nsecs with jitter of :963827 nsecs 500025650 nsecs with jitter of :25650 nsecs 500020735 nsecs with jitter of :20735 nsecs 499057062 nsecs with jitter of :942938 nsecs 498932033 nsecs with jitter of :1067967 nsecs 591037317 nsecs with jitter of :91037317 nsecs 499032794 nsecs with jitter of :967206 nsecs 499034637 nsecs with jitter of :965363 nsecs 499022963 nsecs with jitter of :977037 nsecs 498976577 nsecs with jitter of :1023423 nsecs 499076723 nsecs with jitter of :923277 nsecs 499023885 nsecs with jitter of :976115 nsecs 499018048 nsecs with jitter of :981952 nsecs 499004224 nsecs with jitter of :995776 nsecs 499048461 nsecs with jitter of :951539 nsecs 499013747 nsecs with jitter of :986253 nsecs 499018048 nsecs with jitter of :981952 nsecs 499007604 nsecs with jitter of :992396 nsecs However i seem to get jitter of around 1 ms. Is there anything else i can do to improve?
Re: Real Time-ing
On Tue, 08 Dec 2015 15:35:18 +, Taylor Hillegeist wrote: > So, I mostly do programming that is of run to completion verity. > But I have a dream of calling functions periodically. So my question is: > > What is the best (most time accurate) way to call a function every n > time units? Busy-wait and access the CPU's high precision clock. If you have an extended wait and don't want to busy-wait the whole time, sleep up to, say, 5ms before the time you want to call the function, then busy-wait.
Re: Real Time-ing
On Tuesday, 8 December 2015 at 16:40:04 UTC, Taylor Hillegeist
wrote:
On Tuesday, 8 December 2015 at 15:35:18 UTC, Taylor Hillegeist
wrote:
So, I mostly do programming that is of run to completion
I took a stab at the problem:
http://dpaste.dzfl.pl/2eef530d00fc
0 nsecs with jitter of :5 nsecs
498937256 nsecs with jitter of :1062744 nsecs
499036173 nsecs with jitter of :963827 nsecs
500025650 nsecs with jitter of :25650 nsecs
500020735 nsecs with jitter of :20735 nsecs
499057062 nsecs with jitter of :942938 nsecs
498932033 nsecs with jitter of :1067967 nsecs
591037317 nsecs with jitter of :91037317 nsecs
499032794 nsecs with jitter of :967206 nsecs
499034637 nsecs with jitter of :965363 nsecs
499022963 nsecs with jitter of :977037 nsecs
498976577 nsecs with jitter of :1023423 nsecs
499076723 nsecs with jitter of :923277 nsecs
499023885 nsecs with jitter of :976115 nsecs
499018048 nsecs with jitter of :981952 nsecs
499004224 nsecs with jitter of :995776 nsecs
499048461 nsecs with jitter of :951539 nsecs
499013747 nsecs with jitter of :986253 nsecs
499018048 nsecs with jitter of :981952 nsecs
499007604 nsecs with jitter of :992396 nsecs
However i seem to get jitter of around 1 ms. Is there anything
else i can do to improve?
1) a computer is not real-time, for example audio is always
buffered (e.g 512 samples)
2) thread are not good for timing. In a simple test you could get
satisfying results while in a stressed IRL environment it won't
work as well. so you should rather use the OS API to make a timer
(or the one proposed by a framework if it applies, for example
gui app with SDL: SDL timer).
However if you want to have better times you can use core.time in
a thread callback:
~
// to, t1 and interval are uint;
if (!t0) t0 = TickDuration.currSystemTick.msecs;
t1 = TickDuration.currSystemTick.msecs;
if ((t1 - t0) > interval)
{
t0 = 0;
onTimer();
}
~
