-fomit-frame-pointer

[choward]

People commonly say that using -fomit-frame-pointer is a big speed boost, since it frees up one of the few registers on the x86 CPU. Because of that, I've built my system without frame pointers. However, is this assumption true?

I noticed a new thread on lkml recently:
http://marc.theaimsgroup.com/?t=103785428900001&r=1&w=2

I realize this is specific to the kernel. Has anyone done actual comparisons of other systems on their gentoo box?
_________________
Craig Howard
4B Computer Science -- University of Waterloo

[rac]

I would like to make my standard plea that people not use this option, because it impairs the ability to file useful bug reports.
_________________
For every higher wall, there is a taller ladder

[Lovechild]

[rac]

More so than "you can't file a bug report because people will blame it on omit-frame-pointer", I meant that compiling things (applications and libraries) with that option makes stack traces and core dumps virtually unusable, because debuggers will be unable to synchronize the symbolic info.
_________________
For every higher wall, there is a taller ladder

[pilla]

But then you may have libraries compiled with -fomit-frame-pointer and so on.... would you compile the entire KDE again? Thus, I agree with rac

I stopped using -fomit-frame-pointer after some packages gave me some trouble in gcc 2.95 (Arts and kdemultimedia). For gcc 3.2, I think that optimizations are not completely safe yet.

I am using "-march=pentium3 -mmmx -msse -mfpmath=sse -O2 -pipe" and things are quite stable, without loosing aparent performance. I know that mmx and sse optimizations can also give me some trouble, but I couldn't resist

[rac]