32-bit or 64-bit

[David Fox]

On Fri, Jun 5, 2009 at 10:19 AM, Chris Jones<jonesc at hep.phy.cam.ac.uk> wrote:

> For instance SSE instructions - The compiler knows that *all* 64 bit
> machines have these, so it can safely use them to speed up various math

And I've been bitten several times (when I had an Athlon Tbird)
getting binaries that included instructions that the processor
couldn't support (they were available in later revs of the CPU). At
least using amd64, that's not going to happen, or at least it
*shouldn't* happen (there seems to be some instructions in amd64 that
aren't available on all amd64-like processors).

Most recently, it was an update to the nvidia driver, which got
compiled to include instructions my cpu could not support. I was
running Debian at the time, so I had to revert (and pin) to the prior
release of the driver.

And since nearly all math is done using SSE and MMX instructions, so
much the better. The stack-like floating point methods used in 32 bit
code are completely absent in 64-bit code. It's all done in registers,
and the amd64 has more of those (especially in MMX) than in 32-bit
mode.






-- 
thanks for letting me change the magnetic patterns on your hard disk.