Re: Mandatory -dbg packages for libraries?
On 09-May-07, 04:02 (CDT), Wouter Verhelst <firstname.lastname@example.org> wrote:
> On Tue, May 08, 2007 at 02:23:37PM +0200, Josselin Mouette wrote:
> > Le lundi 07 mai 2007 ? 13:02 +0200, Wouter Verhelst a ?crit :
> > > > Dropping most .a libraries is something I agree with. I see no reason
> > > > why we should have them for most of the libraries.
> > >
> > > As a courtesy to our users. Statically linked programs are slightly
> > > faster (since they don't need to do PLT lookups, so they spare a jump on
> > > every function call to a shared object). For people for whom performance
> > > is critical, providing .a libraries is a good idea.
> > Hm, I thought the double-jump issue was specific to i386. On other
> > architectures, there is still the need for an extra register, but AIUI
> > the performance impact should be much smaller.
> I'm not entirely sure about the specifics, and especially not across
> architectures; but regardless, doing a PLT lookup is more expensive than
> doing a function call to something that was statically linked in.
True. Now, does anyone have measurements to show that this has
any actual significance in real world code on modern hardware? In
particular, anything *besides* the BLAS routines, which is about the
only thing I'd expect to be measurable.
The irony is that Bill Gates claims to be making a stable operating
system and Linus Torvalds claims to be trying to take over the
world. -- seen on the net