[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: amd64 and dpkg and so



On Sat, 2003-08-30 at 13:27, Peter wrote:
> Martin Jungowski writes:
> 
> [...]
> 
>  > Home users with 64-bit CPUs are going to look for an 64-bit OS, and
>  > that's where I can see Linux still has the upper hand. I wouldn't expect
>  > a 64-bit Windows for another year, which gives us a one year time frame
>  > to push 64-bit Linux - however in order to achieve that, the user has to
>  > be capable of running 32-bit applications as well.
>  > 
> 
> good point, very good point.  And if this was a democratic decision I
> would certainly vote for the mixed 64/32bit path, in which case we
> would make use of one of the main features of the amd64 architecture,
> isn't it?

Well yeah, the 64-bit GPRs however are just one of the main features of
the AMD64 architecture. One should not forget, that in 64-bit mode (also
called Long Mode), the CPU has eight additional GPRs (which effectively
doubles the number of available GPRs) at its command and also, in Long
Mode, the XMM registers are available. These registers are also known as
additional SSE-Registers, hence the CPU does not support SSE2 in 32-bit
mode but in 64-bit only.

Now I don't know how important this might be to Linux developers but now
that both, the Pentium4 and the Athlon64 have SSE2 support, and SSE2 is
way better than SSE (do I need to proof, that four additional 64-bit
floating point operations every clock cycle are better than four 32-bit
floating point operations every two clock cycles? Or that two additional
128-bit or four additional 64-bit integer opterations every clock cycle
are way better than four additional 32-bit integer operations every
second clock cycle?), I think it would be a good thing to use it for
some performance-critical code.
And since every x86_64 compatible CPU supports SSE2 in Long Mode (we can
certainly assume that if Intel really is about to launch the Yamhill
project, the CPU is going to have SSE2 support as well) we don't need to
care about compatibility issues with i386 compatible CPUs in 64-bit mode
anymore. I think this is a great possibility for Linux developers to
speed things up big time.

I really don't know whether how complicated that would be, I'm not
really a software developer but more kinda like the hardware guy.
Maybe somebody who's more into software development could take my point
from here.

Martin







Reply to: