As was advised long time ago, I always build my release executables without frame pointers (which is the default if you compile with /Ox).

However, now I read in the paper http://research.microsoft.com/apps/pubs/default.aspx?id=81176, that frame pointers don't have much of an effect on performance. So optimizing it fully (using /Ox) or optimizing it fully with frame pointers (using /Ox /Oy-) doesn't really make a difference on peformance.

Microsoft seems to indicate that adding frame pointers (/Oy-) makes debugging easier, but is this really the case?

I did some experiments and noticed that:

  • in a simple 32-bit test executable (compiled using /Ox /Ob0) the omission of frame pointers does increase performance (with about 10%). But this test executable only performs some function calls, nothing else.
  • in my own application the adding/removing of frame pointers don't seem to have a big effect. Adding frame pointers seems to make the application about 5% faster, but that could be within the error margin.

What is the general advice regarding frame pointers?

  • should they be omitted (/Ox) in a release executable because they really have a positive effect on performance?
  • should they be added (/Ox /Oy-) in a release executable because they improve debug-ablity (when debugging with a crash-dump file)?

Using Visual Studio 2010.

performance visual-studio-2010 visual-c++ x86 stack-frame
Edit
Report