Re: New Release

[Danial Thom <danial_thom@xxxxxxxxx>]

--- Joerg Sonnenberger <joerg@xxxxxxxxxxxxxxxxx>
wrote:
> On Thu, Apr 14, 2005 at 08:52:10AM -0700,
> Danial Thom wrote:
> > --- "Martin P. Hellwig" <mhellwig@xxxxxxxxx>
> > > As a stupid user I wonder what it takes
> > > generally to support the AMD 64 
> > > bit extension, what are the typical issues
> in
> > > supporting both 32 and 64 
> > > bit code? Links would be very welcome too
> :-)
> 
> It's another compatibility layer, which is a
> bit more difficult
> since the pointer size changes and you
> effectively have to wrap
> all system calls. There are other issues as
> well.
> 
> > 
> > A better question is what does it take to do
> it
> > well, or whether its worthwhile. In FreeBSD,
> the
> > amd64 is so clunky its unusable on a server,
> IMO.
> > I shiver at the thought of trying linux, or
> > windows, but I am curious. Is there a way
> with
> > GCC 3+ to utilize the extra registers without
> > running in full-blown 64bit mode? If you
> don't
> > need the memory that seems like it might be
> an
> > ideal enhancement. 
> 
> No. If you really want to use 64bit, do it
> correctly. It's a
> question of time when we do it, there are more
> pressing needs.
> Also keep in mind that a lot of indirect
> changes are done in the
> tree to get it 64bit clean without actually
> having 64bit architectures
> to test. Think of the whole bus_dma support :)

I guess I'm not sure I want to use 64-bit,
particularly if its going to be slower than
32-bit mode. With larger structures, pointers,
etc, if stuff starts not fitting in the cache
then you'll lose more than you gain. So using the
extra registers would improve performance without
having all of the issues.

Danial

__________________________________________________
Do You Yahoo!?
Tired of spam?  Yahoo! Mail has the best spam protection around 
http://mail.yahoo.com