Re: 64bit status report : biarch toolchain and ppc64 debian kernel.
On Fri, 2005-08-19 at 00:58 -0400, Albert Cahalan wrote:
> On 8/19/05, Benjamin Herrenschmidt <firstname.lastname@example.org> wrote:
> > On Thu, 2005-08-18 at 22:55 -0400, Albert Cahalan wrote:
> > > On 8/18/05, Sven Luther <email@example.com> wrote:
> > >
> > > > That brings us to the next step, what is the best way to get most libraries to
> > > > build 64bit packages ? This would need some extensive change in the packaging
> > > > stuff probably.
> > >
> > > As you know, there are two ways:
> > >
> > > a. Major hacking for each and every package.
> > > b. Pure and sane 64-bit system.
> > At the expense of performances, disk space, etc... of course
> You could apply that logic to the kernel just as well. Even on hardware
> without 32-bit capability, an ILP32 model can be used. If it's so good,
> why not?
Except that you don't expect to spend most of your time in the kernel.
> Looking at the whole system, you can save disk space and get
> better performance by having only 1 copy of each library stored
> on disk, in RAM, and in the CPU's cache. It is thus not at all
> certain that a mixed system will be faster than a pure system,
> even if 64-bit is inherently slower when compared alone.
Except that in a mixed system, I don't expect many binaries to be 64
bits. It's not like x86_64 where you benefit from additional registers &
> I do note that the Itanic-style ABI features look pretty bad. If they
> are as bad as they look, then a new ABI should come first.
Heh, good luck :)