Re: confused about performance
On Thu, Jun 14, 2007 at 08:08:39PM +0100, Thomas Steffen wrote:
> I hope so, too, but I am far from convinced. In my tests gcc 4 has been a
> very small improvement for 32bit code on my Athlon64, but it produces slower
> 64bit code than gcc 3.4. And I hear that Intel CPUs are very inefficient
> with 64bit code.
The Pentium 4 seemed generally slower with 64bit, but the Core 2 seems
generally faster with 64bit. The same applies to the Xeons based on the
P4 and Core2 designs respectively.
gcc 4 seems in general to be a bit behind gcc 3.4 in generated code
quality. gcc 4 was a major rewrite and I don't think all the
optimizations have been reimplemented in the new system yet.
> I agree with this statement, but I think the real question is: why is it
> such a pain to run a mixed system? 64bit is intended to be an improvement,
> and as such it should be able to run legacy code!
Because a 32bit program can't call a 64bit library (the pointer sizes
don't match). A 64bit program can't call 32bit libraries. So you would
need 2 copies of everything, which is just a mess. Simpler to just make
everything the same in general.
Sure you can write hacks to translate calls but that costs performance
too, and is also messy. A chroot on the other hand does work and lets
you run both environments at the same time, but keeping them seperated
and simple to manage properly. Hopefully some day the 32bit chroot can
simply go away and die when it no longer has any purpose left.
--
Len Sorensen
Reply to: