64-bit CPU

[Andrej Marjan]

Lennart Sorensen wrote:

>As far as I remember, MMX was mostly about doing the same operation (one
>instruction) on a chunk of data at a time, so doing a bitshift on each 8
>bit pieece in a 64bit register, or each 16bit piece in a 64bit register,
>or even 32bit in a 64bit register.  so that you could process more than
>one audio sample or pixel in one instruction.  I do not believe it did
>anything to improve 64bit integer operations, which I am not actually
>sure if are available in hardware on x86.  I would be surprised if the
>386 had added those, and I don't know how many new instructions have
>really been added (outside MMX, SSE, etc) since.  Of course an
>application using MMX can't use floating point.  It has to switch back
>and forth between the mode to do that.  Probably no big deal since they
>are probably mostly used in different applications.
>  
>

I know that was the intent of MMX. I had just (incorrectly) assumed that 
since they had gone to the trouble of overloading the floating point 
registers to allow for integer arithmetic, they would have gone the 
extra step of allowing arithmetic on the full 64 bits as a single entity.

>Well x86-64 does add quite a few new instructions when in 64bit mode.
>Being a new mode it also makes sense to add more registers (why SSE and
>such didn't do that while they were at it, I can't imagine.)
>  
>

Well... considering that MMX was just grafted onto a ppro core, it would 
be less invasive (and cheaper) to just reuse the only registers that 
were big enough for the new purpose...
--
The Toronto Linux Users Group.      Meetings: http://tlug.ss.org
TLUG requests: Linux topics, No HTML, wrap text below 80 columns
How to UNSUBSCRIBE: http://tlug.ss.org/subscribe.shtml