I looked through old threads but could not find the answer to my question:
How can I time the body of my function inside a C program?
1
- 3Don't worry, SO can be a hard site to search. It's usually better to just have google filter the results to the site instead: google.com/search?q=site%3Astackoverflow.com+profile+c+functionGManNickG– GManNickG2009-07-31 01:38:20 +00:00Commented Jul 31, 2009 at 1:38
Add a comment |
5 Answers
A simple method is to use the 'clock' function:
#include <time.h> clock_t start, end; double cpu_time_used; start = clock(); ... /* Do whatever you want to time */ end = clock(); cpu_time_used = ((double) (end - start)) / CLOCKS_PER_SEC; Or if you're on linux you can use the 'time' command to time how long it takes your app to execute; this doesn't allow you to time a specific code section though, and includes time taken to initiate the process etc.
time ./myapp Edit: This is very much a basic 'hack-in-a-quick-timer' approach. For true performance profiling you want to look at a proper profiler, as suggested by Larry Watanabe.
6 Comments
jkeys
Very insignificant, but how much time do you think it adds to call end = clock()? In the online competitions where you have 1-2 seconds for your program to execute, that could actually be significant.
Edmund
It's easy to see the overhead: just call 'start = clock(); end = clock();' and see the difference (if it's measurable).
CapBBeard
I cant imagine it would add much, at least with current compilers and optimisation etc. I'd expect anything you would be timing in this manner would be orders of magnitude slower than calling clock().
foobarfuzzbizz
@Larry interesting thought, as I'm spawning 10,000 threads between the clock calls. Would a call to pthread_join be what I'm looking for?
Larry Watanabe
See my answer below - the answer is just to use a profiling tool.
|
It depends on your compiler and OS. On Sun workstations, I used to use "prof" or "gprof". There is surely a profiling tool for your compiler, OS, and machine - just google "C profile yourOS yourcompiler" (substitute the name of your OS and your compiler )
2 Comments
Wayne
+1 as this truly a better way of profiling your code as the clock method does not account for threads, interupts and other vectors that the kernel controls.
Larry Watanabe
I agree - and also, what's the point of measuring the time of just 1 specific function? If you profile the whole thing then you will find out what the bottlenecks are, which is more important -- unless it's just a case of CYA for YOUR part of the project
The basic method is using the clock() function, that is where we all started.
Ex.:
clock_t start = clock(); /* Your code */ printf("Time elapsed: %f\n", ((double)clock() - start) / CLOCKS_PER_SEC); However, when you start learning about Operating systems, hardware, schedulers, multi-threading, etc. You realized that execution time is something very subjective. When you want to measure performance (which does not necessarily mean execution time) then you need more robust tools.
Gprof is a really easy to use C profiler, which could help you understand better the concept of performance.
Comments
time ./a.out run above one, and output will be:
real 0m0.017s user 0m0.017s sys 0m0.000s real: Total end to end time taken by program/command
user: Time taken in user mode.
sys: Time taken in kernel mode
Comments
Not sure about specific functions names in C, but a common practice is to store a microsecond timestamp before the body starts, then grab another one at the end and subtract.
