Programmatic Function Profiling
-
Yes, plz plz. Or gimme more drugz! :-D
If the Lord God Almighty had consulted me before embarking upon the Creation, I would have recommended something simpler. -- Alfonso the Wise, 13th Century King of Castile.
This is going on my arrogant assumptions. You may have a superb reason why I'm completely wrong. -- Iain Clarke
[My articles] -
:-\
Nobody can give you wiser advice than yourself. - Cicero .·´¯`·->Rajesh<-·´¯`·. Microsoft MVP - Visual C++[^]
-
Hi All, Is there a way to find out the time taken by a function to complete (in terms of nanoseconds) in my c++ code without using "Function Profiling" feature of the linker. I tried GetTickCount(), but it gives the time in milliseconds. It always returns 0 because my function completes its task less than in one millisecond. Thanks in advance
I will suggest that you run this function a large number of times. Take the total time and the compute the average time, this way you will have more realistic measure of time function takes. -Saurabh
-
Hi All, Is there a way to find out the time taken by a function to complete (in terms of nanoseconds) in my c++ code without using "Function Profiling" feature of the linker. I tried GetTickCount(), but it gives the time in milliseconds. It always returns 0 because my function completes its task less than in one millisecond. Thanks in advance
[EDIT] Yeah, what he said. (seems I had the reply window open longer than I thought :-O ) I don't quite understand what the big deal is - am I really missing something really obvious? This just seems like a relatively simple exercise in maths to me, with the added fun of error estimation and minimization. There's 1,000ms in a second right. So, if I can complete 2,000 events in 1,000 ms then each event is taking 1,000 / 2,000 = 0.5ms. So then if I complete something 5,000,000 times in 1,000ms, each event is taking 0.0002 ms, or 1/5,000,000 = 0.0000002s - we've already measured/calculated the time of completion to 1/5000th the resolution of the millisecond timer. Something to consider though is that as the time of each event becomes smaller and smaller, the overhead of the loop becomes a higher and higher margin of the error. The other major source of error is task-switching. This makes timing such an event in a non-realtime os more than a bit of a challenge. Things that can help you minimize these errors are to (i) kill all unnecessary processes first (ii) complete the maximum number of events you can - a longer sample time will help give more consistant results. (iii) start the timing loop on a keypress. This will allow you to start your program, set it's priority to "realtime" and then get timing. Lets say I have a loop that calls my function 14 times per iteration(inlining the code would be much better than calling a function of course). Next, I time 500 million iterations of my loop and find that it completes in 2785 ms. So, I can tell that I'm getting 179,533.2136 loop iterations per millisecond. I then multiply this by 14 and find that I'm getting 2,513,464 function calls per millisecond. This would correspond to 2.5 function calls per nanosecond. (not a result you could hope to get close to, given current clock-rates of cpus and the error-inducing factors I mentioned above, but that's the basic idea)
-
SandipG wrote:
And finally hijacked!!
Whom??
"Opinions are neither right nor wrong. I cannot change your opinion. I can, however, change what influences your opinion." - David Crow
Never mind - my own stupidity is the source of every "problem" - Mixturecheers, Alok Gupta VC Forum Q&A :- I/IV Support CRY- Child Relief and You/xml>
-
CPallini wrote:
I prefer lower temperatures to have cups of tea.
Urgent?
Nobody can give you wiser advice than yourself. - Cicero .·´¯`·->Rajesh<-·´¯`·. Microsoft MVP - Visual C++[^]
cup of coffee also help :)
"Opinions are neither right nor wrong. I cannot change your opinion. I can, however, change what influences your opinion." - David Crow
Never mind - my own stupidity is the source of every "problem" - Mixturecheers, Alok Gupta VC Forum Q&A :- I/IV Support CRY- Child Relief and You/xml>
-
Rajesh R Subramanian wrote:
In the best case, the HAL may conclude that the RDTSC counter runs at a constant frequency, so it uses that instead.
That's
APIrobustness. You should make the (in)famousQueryPerformanceFrequency/QueryPerformanceCounterpair make the best (while still reliable) choiche for you. ;)If the Lord God Almighty had consulted me before embarking upon the Creation, I would have recommended something simpler. -- Alfonso the Wise, 13th Century King of Castile.
This is going on my arrogant assumptions. You may have a superb reason why I'm completely wrong. -- Iain Clarke
[My articles]CPallini wrote:
QueryPerformanceFrequency/QueryPerformanceCounter pair make the best (while still reliable) choiche for you
damm i missed it!
"Opinions are neither right nor wrong. I cannot change your opinion. I can, however, change what influences your opinion." - David Crow
Never mind - my own stupidity is the source of every "problem" - Mixturecheers, Alok Gupta VC Forum Q&A :- I/IV Support CRY- Child Relief and You/xml>
