From: wpp@marie.physik.tu-berlin.de (Kai Petzke) Subject: Re: Pentium Date: 21 Apr 1993 12:02:32 GMT
In <1993Apr20.163316.12290@jupiter.sun.csd.unb.ca> u0xh@jupiter.sun.csd.unb.ca (Just me again...) writes:
>>The pentium chip is 100% compatible, the only differences lie in its
>>improvements (if you call expansion improvement) of the command set.
> The Pentium command set has not been expanded as far as I've heard.
> If it did, than Pentium optimized software would not run on a 486, or
>386. This is not the case however, because Pentium optimized software will
>supposedly run ~5% faster on a 486, becuase the optimization is so high.
>There will be little or no change in 386 execution times. This is just
>stuff I've heard and read in magazines, and the intel newsgroup
>(comp.sys.intel? I dont subscribe anymore)...
The pentium command set has been expanded, for example, there is now a
system call to determine the processor (are you just on a pentium, or
even a sextium, ... - this call is no much use on a 486 or lower). There
are also new bits in the system register, and there is stuff added to
cope with multiprocessor environments and caches (think of a page, which
is used by the kernel for locking, which is in the write-back-cache of
two processors: two processes could easily set a bit to one and get
informed, it was zero before, at the same time. Nice crash, isn't it?)
However, all this changes regard the few kernel hackers, which deal
with paging, memory and booting. There are no changes to the
normal command set.
The pentium, however, processes instructions in two pipes in parallel.
So if one instructions changes a register (memory location, ...), and
the next reads this register, the pentium can't do this in parallel.
Other optimizations are for the pipelines (both integer and floating
point) themselfes.
Kai
wpp@marie.physik.tu-berlin.de