Anthony DeRobertis wrote: > On Wed, May 03, 2006 at 11:02:57AM -0300, Henrique de Moraes Holschuh wrote: > > > For Etch and Sid, it is probably a good idea to use -Os instead of -O2 at > > least on the bigger arches (ia32, ia64, amd64, etc), as we can probably > > trust gcc not to screw up. > > If gcc generally generates faster code with -Os than -O2, then isn't > that a gcc bug,
It doesn't do so generally. >in that the optimizations enabled by -O2 are incorrectly picked? The compiler can only make some broad assumptions about the CPU's memory/cache system, and none about the size of the working set. If the working set with -O2 spills out of the cache but doesn't with -Os then you get a performance improvement from -Os, for the next CPU with bigger caches it can swing back. > [Also, are there that man AMD64 machines with limited memory? Or IA64?] Swap is now mostly irrelevant for performance discussions, if you hit swap you won't have performance anyway. :-) Thiemo -- To UNSUBSCRIBE, email to [EMAIL PROTECTED] with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
