On Tue, 2005-05-17 at 12:32 +0100, Mike Hearn wrote: > On Mon, 16 May 2005 19:01:59 -0400, Ignacio Vazquez-Abrams wrote: > > Even a 2 clock cycle delay happening hundreds of thousands or even > > millions of times a second builds up. > > I'd really like to see hard statistics about this. If I recall correctly, > Ulrich Drepper did some experiments with using -mregparm for internal > calls inside glibc and found it made no difference, or in some cases made > things worse. There are enough "intuitively" correct optimisations that > slow things down out there already .. actually glibc mostly is lots of function wrappers, where it indeed helps little. For the kernel it actually shrunk code some, and at the time I looked over the assembly generated and it looked a lot better too.
Attachment:
signature.asc
Description: This is a digitally signed message part
-- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx http://www.redhat.com/mailman/listinfo/fedora-devel-list