Menu
I'm debugging an C application and I'd like to know how much time it spends in a particular function.
I could change the source code and add some more code to do the measurement, but it doesn't seem right to me. I'd rather do it with external application, without recompiling every time.
I found out it's possible to set up a break point in GDB so I thought, it must be possible to track the time using similar tool by simple procedure: - set breakpoint - when stopped, measure actual time and run the function - when leaving function, measure time again However, i haven't found a way how to do this in gdb :(
any ideas? Thanks
503
To get the elapsed time, we can get the time using clock() at the beginning, and at the end of the tasks, then subtract the values to get the differences. After that, we will divide the difference by CLOCK_PER_SEC (Number of clock ticks per second) to get the processor time.
The time() function is defined in time. h (ctime in C++) header file. This function returns the time since 00:00:00 UTC, January 1, 1970 (Unix timestamp) in seconds. If second is not a null pointer, the returned value is also stored in the object pointed to by second.
Returns the decimal number for a particular time. If the cell format was General before the function was entered, the result is formatted as a date. The decimal number returned by TIME is a value ranging from 0 (zero) to 0.99988426, representing the times from 0:00:00 (12:00:00 AM) to 23:59:59 (11:59:59 P.M.).
measure execution time of a program. Using time() function in C & C++. time() : time() function returns the time since the Epoch(jan 1 1970) in seconds. Prototype / Syntax : time_t time(time_t *tloc);
If you're using GCC, you want the compile option "-pg" and the application gprof.
97
gprof is only reliable -- in my experience, only works at all -- if you statically link -pg compiled versions of every library, including the C library. You can try to do this using gcc's -profile option (which does what -pg does plus tries to sub in -pg libraries) but the problem is, GNU libc really does not like being statically linked, and your distro may not provide -pg compiled versions of every library you need.
I suggest you try cachegrind, which is a valgrind mode of operation and only needs debug info for everything. That's much easier to get. The catch is, it has huge overhead costs; so huge that it might invalidate your testing. Expect at least a 2x slowdown.
You can also try perf -- if you can get your hands on a copy. It's very clever, but it is of, by, and for kernel hackers who think people like building stuff from scratch. I have had very mixed luck with it. (Do read http://web.eecs.utk.edu/~vweaver1/projects/perf-events/ which is about the underlying API, not the utility, but might still save you from a great deal of wasted time.)
39
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With
