On Fri, Mar 17, 2006 at 11:53:30PM +0100, Roland Mainz wrote: > > Note that when building ON with gcc, we use -fno-builtin. > > Why ? I understand this is usefull for debugging - but it kills some > interesting opportunities for optimisation... ;-(
While I'm not the one who implemented this, I'm all but certain it's to prevent gcc from sticking big fat functions like memcpy inline, and/or using less-efficient generic functions when the platform-optimized libc function should be used instead. Have you actually benchmarked with builtin/no-builtin? It may be that a handful of specific functions can be replaced with their builtins for a win, but on the whole, inlining large functions turns out to be a loss. The wasted I-cache far exceeds the gain from call overhead. -- Keith M Wesolowski "Sir, we're surrounded!" Solaris Kernel Team "Excellent; we can attack in any direction!" _______________________________________________ opensolaris-discuss mailing list opensolaris-discuss@opensolaris.org
