Why are there only i686 and i586 Version of glibc and kernel? -- i386 is still around

[Bryan J. Smith]

Leonard den Ottolander wrote:  
> This issue was discussed in the thread "Making NPTL the default for
> FC3, vanilla i386 support". Both i386s and i486s are already no longer
> supported since FC 1. Try running rpm on either of these CPUs and see
> it fail. See
> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=103078 .

FYI, the original Cyrix 686 (M1?) was only i486 compatible.  You had to go
with the 686"L" or M2 to get full i586 ISA compatibility.

Frankly, I don't think people recognize the _real_ reason for trying to
keep i386/486 ISA compatibility.  It's not to support old systems, but to
support i386/486 ISA embedded microprocessor cores in all sorts of black
box solutions.  These solutions have full desktops/GUIs far more than you
may think.

All it takes is one major black box vendors to adopt Fedora and we're
talking over a 1% marketshare.

GCC still pumps out i386 ISA code, while making other optimizations.  So
anything that has a patch or otherwise that breaks this should be
revisited.  So there is still a good reason to strive for i386/486 ISA.

> So if there is the intention of making the move to i486 as the minimum
> arch we could just as well make the jump to i586.

I would continue to favor i386 with i686 (Pentium Pro) optimizations.

i586 (Pentium) causes a lot of de-optimizations in newer processors, even
Intel's own.  Why?  Understand that about 80% of "Pentium optimizations"
were really "errata workarounds."  The Pentium was the first superscalar
x86 design, and boy did it have some real muck-ups that had to be
compensated in software (like the ALU LOAD being 3x slower, which id
found it far faster to load via integers via the FPU instead).


-- 
Bryan J. Smith, E.I. -- b.j.smith at ieee.org