Skip to content

Latest commit

 

History

History
89 lines (58 loc) · 3.13 KB

Profiling with GProf.md

File metadata and controls

89 lines (58 loc) · 3.13 KB
Raw

Profiling with GProf

Premature optimization is the root of all evil.

Thus spake Donald Knuth, and he's right, as usual. Programmers spend a lot of time worrying about code that feels slow, but doesn't actually contribute much to the runtime of the program. If you really want to improve your program's runtime, you need to figure out what it spends most of its time doing, and then focus on that.

There's a tool called gprof that can help you do this. The process has three main phases. First, you compile your program with g++ and enable profiling. Then you run the profiled program. Finally, you use gprof to display the output.

gprof works on Linux, and should work on WSL without issues. Non-WSL Windows users and Mac users can find it pre-installed on CSIL.

How to Use GProf

  1. Find a Test Case

    Before you can test your program, you need some data to test it on. Find some data that takes your program a reasonable amount of time to process: at least a few seconds, but less than a minute. Ten or fifteen seconds is perfect.

    If your test case requires input from the command line, save this input into a text file (in this example it's called input.txt). This lets you send data directly to your program, so you don't measure the time it spends waiting for the user.

  2. Compile your Program

    Compile as usual, but add the -pg and -g flags. The -pg flag enables the profiler, and the -g flag adds debug symbols, which will improve the output.

    g++ -o path/to/program ... -g -pg whatever.cpp ...

    The program that gets created will automatically write profiling information as it runs, so it'll run slower than usual.

  3. Run your Program

    Run the program that you just created. Use the shell's < operator to send your input.txt file to the program's standard input.

    path/to/program arg1 arg2 < input.txt

    As it runs, it will create a file named gmon.out in your current directory. This file contains runtime statistics for your program.

  4. Run gprof

    To see the results, run gprof and pass the path the the program that you profiled as the first command line argument. If you want to use a profile other than the gmon.out file in the current directory, you can pass that as the second argument.

    gprof path/to/program

    This command will produce a lot of output. There are two main sections: the Flat Profile and the Call Graph. Each section starts with a header explaining how to read it, but you can also find a more detailed explanation here:

    https://ftp.gnu.org/old-gnu/Manuals/gprof-2.9.1/html_mono/gprof.html#SEC10

  5. Optimize!

    Now that you know what's taking up the most time, you're ready to optimize. Go for it! Work on the slowest functions first. See how much you can speed them up - or try to call them fewer times.

I've never needed to do much more than this, but gprof is a powerful program with a lot more options. If you're curious what else it can do, see the manual:

https://ftp.gnu.org/old-gnu/Manuals/gprof-2.9.1/html_chapter/gprof_toc.html