I have an application written in c++ using visual studio 2005. The application has certain performance problems. I would like to explore where. I need to drill down in which classes/methods/lines the application spends most of the time. Can this be done with the WPA? If yes, can you, please give me a pointer to documentation?
A:
Try the YourKit .net profiler. I tried this and few others recently, but this was the only one I managed to get to work with my C++/CLI project (with a mixture of managed and unmanaged code).
[Edit] Oops; I read WPA as WPF and assumed you wanted to profile some .net code.
Jon Cage
2010-02-01 14:02:03
+1
A:
No, WPT leverages windows events, it would only help you diagnose a problem when Windows is the cause of your slow-down. You certainly won't get any diagnostics for your code. What you need is a real profiler. Good ones cost money. Check this thread for more advice.
Hans Passant
2010-02-01 15:02:53
Thanks. I was desperate trying to force WPT to read my symbol tables.
danatel
2010-02-01 20:20:38
+1
A:
The simplest way to find out what is spending time unnecessarily is stackshots, which you can easily get with the "pause" button in Visual Studio.
You don't need to measure; all you have to do is catch it in the act. If you find it doing something on two or more stackshots, and it's something you could avoid doing, you will save significant time. And, you can repeat the process, because problems that are small at first will "bubble up" as you remove other problems.
Mike Dunlavey
2010-02-01 15:40:21
Good advice but unsuitable for my situation. The application performs well on windows (99% of time waiting for user input), the performance problems are visible only on an embedded system with much slower processor and for which I have no debugger/profiler at all. Good profiler could help as it can distingusih between 0.5% and 0.05% of time.
danatel
2010-02-01 20:25:23
What I do in that case is wrap a loop (10x or 100x) around each piece of UI-driven processing, so as to "amplify" what's going on. This has the same effect as slowing down the processor. Tune the code down to minimum time, then remove the loop.
Mike Dunlavey
2010-02-02 20:23:11