>>>>> "Peter" == Peter Jay Salzman <p@dirac.org> writes: Peter> begin: Etay Meiri <cl1@netvision.net.il> quote >> On Mon, Nov 19, 2001 at 01:16:05PM -0800, Peter Jay Salzman wrote: >> > can that be right? i was able to compile both the devfs and non-devfs >> > versions of grumpy with -O3. >> > >> > the only difference between the two is that -O3 implies -finline-functions, >> > and -O2 doesn't. would that really break compiling? >> > >> > pete >> >> I read somewhere that a lot of kernel code depends on the characterstics of optimization >> generated code and would simply not work without -O2. I can't provide a pointer to that info. >> From my own experience I know that without optimization at all your module would fail insmod >> for unresolved symbols. I haven't tried -o3. Perhaps someone else can elaborate on this point. Peter> this is true, but -O3 is exactly equivalent to -O2 with automatic function Peter> inlining wherever gcc can do it. (clearly recursion is an example of where Peter> gcc can't automatically do function inlining). IIRC, inlining in the kernel is explicitly managed by the developers for performaence reasons -- Kernelnewbies: Help each other learn about the Linux kernel. Archive: http://mail.nl.linux.org/kernelnewbies/ IRC Channel: irc.openprojects.net / #kernelnewbies Web Page: http://www.kernelnewbies.org/
