Re: POSIX Layer / Emulation

[Jonathan S. Shapiro]

On Wed, 2005-10-26 at 21:40 +0200, Alfred M. Szmidt wrote:
>    This is not silly/weird/strange/etc. Many systems do that when they
>    do big changes on their design, why we cannot do that?
> 
> because you end up with crap like IA-32.

Crap like IA-32 is what you get when you try to do legacy support in the
core. Building emulators that run outside the core isn't so bad.

This is way off topic, but in the early days of HaL Computer Systems
(circa 1990) we told Sun that they should abandon register windows as
they moved to a 64-bit platform. We had a long list of good technical
reasons, and we recommended that they should use binary translation for
compatibility with existing 32-bit code. Register windows were a really
big problem for multiple instruction issue, and we gave them a schedule
for the death of the SPARC part if this change was not made.

Dave Ditzel, who was then head of processor architecture at Sun,
rejected this suggestion on the grounds that it was insufficiently
compatible. In direct consequence, the SPARC chip has suffered from
increasingly non-competative performance over the last decade, on pretty
much the schedule we predicted in 1990.

Ditzel went on to use most of the ideas we advocated in the design of
the Transmeta processor.


shap