EduardoS (no.delete@this.spam.com) on September 11, 2013 8:24 am wrote:
> someone (someone.delete@this.somewhere.com) on September 11, 2013 7:38 am wrote:
> > timcaffrey (timcaffrey.delete@this.aol.com) on September 11, 2013 7:26 am wrote:
> > > Does anybody remember the Intel 376? A 386sx derivative that
> > > 1) Didn't support segments.
> > > 2) Booted directly into 32 bit mode.
> > > 3) Didn't support 16 applications.
> > >
> > > My guess is Quark is something like this. No FP (x87/SSE/AVX), perhaps removal of
> > > BCD/ASCII math instructions.
> >
> > I doubt they would drop FP completely. One could implement a very low complexity
> > adder, low performance FPU for x87/SSE that would easily speed FP up >10X better
> > than a software based solution. Going with an SSE only basic FPU would be desirable
> > but IIRC is problematic on the software side.
> >
> > IMO it is more likely they would drop the ancient x86 16 bit modes.
>
> Stop please, why in hell Intel is going to release a x86-incompatible x86 embbedded solution?
>
> A full x86 compatible is far more flexible, just drop in whatever software already exists,
> Intel itself would be able to sell as low end CPU for phones, but x86 incompatible means
> a very restrcited market, as if they released a CPU with a completly new ISA.


Any x86 compiler I am aware of is perfectly capable of generating vanilla 32 bit
code that runs on a 486 or P5. That is about as compatible as it gets.

The difference in ISA complexity between a P5 and all that the latest Core or Atom
can execute is a huge factor for a synthesizable CPU for low end embedded control.