Re: cvsup on amd64 just broke today

On Sun, Aug 29, 2004 at 04:07:44PM -0700, Sean McNeil wrote:
> Finally catching up on your email? ;)

Yeah. :-/

> On Sun, 2004-08-29 at 15:53, David O'Brien wrote:
> > Why is it odd?!?
> > The ability to run legacy 32-bit x86 binaries under a 64-bit OS at
> > full-speed is one of the huge capabilities AMD brought with this
> > architecture.  Unless a binary does 64-bit math or addresses >4GB of
> > memory why does something need to be 64-bit???
>
> This is a little misleading.  You are throwing out the fact that the
> amd64 has additional features in 64-bit mode that can significantly
> increase performance.  Such as extra registers.

I'm not throwing out the fact about the extra registers.  In my day-job I
track AMD performance.  Trust me, for a network-based program such as
CVSup isn't going to reap a noticeable benefit from them.  What slows down
CVSup is network BW and latency, same for the disk BW and latency.  CVSup
isn't doing Seti_at_Home calculations.


> > The fact that all Open Source OS's have a 64-bit userland on all their
> > 64-bit platforms that grew up from 32-bit CPU's shows how unsophisticated
> > our build framework is.  "64-bit" Solaris today is really a 64-bit kernel
> > and mostly 32-bit userland.
>
> Except Solaris has identical architectures that were extended to
> 64-bit.

Actually Sparc isn't 100% identical -- performance enhancements were made
in 64-bit mode.  They just aren't as many and to the extent as done in
AMD64.


> amd64 is a slightly different story.

Solaris for AMD64 will also have a 32-bit userland, and its compiler will
default to producing 32-bit binaries.  For things like 'vi' and 'ls'
64-bit + extra registers simply doesn't matter.

--
-- David  (obrien_at_FreeBSD.org)