Going 64 bit

[Nicholas Leippe]

On Friday 26 January 2007 19:08, Michael Torrie wrote:
> Sure, but this doesn't matter anymore.  Let's face it.  They used to say
> RISC was the wave of the future.  And they were partially right.  But
> now the gap between RISC and CISC is narrowed significantly.  Most x86
> chips (AMD or Intel) are really RISC cores with a microcode translator
> that converts the more compact CISC instruction sequences into RISC
> microcode where it is pipelined, reordered, etc.  This gives all the
> advantages of RISC without having to actually force RISC ISA on the
> compilers and programmers.  In effect this means the RISC never really
> panned out like everyone thought it would at the higher level.  x86_64
> has the advantage of having about twice the code density of a 64-bit
> instruction word 64-bit RISC processor.  And even though memory and disk
> space is cheap, this higher density pays off in terms of increased cache
> performance.

Agreed. Today's x86 implementations essentially render the CISC isa as a 
compression of the RISC microcode to which the decode units translate 
internally. When I hear the term "RISC-based", I think load-store 
architecture with a fix-width instruction size. I was merely pointing out 
that "true RISC-based" (while also a subjective term) still doesn't apply 
directly to x86_64 as viewed from the outside.

> On the flip side, the x86 ISA, 64-bit or not, is old, bloated, and full
> of strange anachronisms like memory segmentation and "real mode" garbage.
> It's likely you can still boot MS-DOS on an AMD 64-bit machine.  However the
> x86_64 extensions do give us a path forward and perhaps future cpus can drop
> support for older things like real mode, 16 and 32-bit instructions.  Who
> knows.    

Maybe they'll also deprecate the direct support for BCD numbers... do any 
compilers even use it?