Bob Cain wrote:
Scott Dorsey wrote:

Again, address space and word length are independant.... you can get all
the address space you want with the expense of nastiness like segmentation.

To what, then, do attribute the move to 64 bits?

The fact that being able to move data in 64 bit chunks is an enormous win
if you are doing 64 bit arithmetic, and that can be a big deal for some
stuff. Code that ran in double precision mode now runs in single, with
a considerable speed improvement.

But the 32-64 jump isn't going to be anywhere near as big as the 16-32
jump was, just because there is less of a need to do that for typical
applications.

The wider data word also allows a big improvement in the speed of block
operations, where you're moving arrays from place to place in memory,
since you can now do it in 64 bit chunks.

And if you are really good, or your compiler is really good, you can do
things like store adjacent 32 bit values into a single 64 bit word and
then use 64 bit operations to load and store them both at the same time.
This is less of a win than it might seem at first.

But address space is the least of your worries.
--scott
--
"C'est un Nagra. C'est suisse, et tres, tres precis."