🚀 go-pugleaf

Chae Ho An (pala...@umich.edu) wrote:
: >	Not necessarily... if the Pentium's abilities are utilised fully, 
: > then the 486 will NOT - because the pentium is like 2 chips in one, 
: > ie 2 general processing chips in one, not a beefed up 486 and co-pro.

: 	Don't credit the Pentium's superscalar design too much.  
: ... So far, the coders have practically no incentives to
: utilize the Pentium's abilities fully.

Pentiums ability goes more than just superscalar. The integration of
some RISC methods into the chip makes for a different coding philosophy
and optimization, such as maximizing register use.

Coders have a great incentive to optimize to the Pentium, especially
for CPU bound applications. Modern UNIX SVR4 for the Intel processors,
for example, benefit greatly by having a Pentium-optimized kernel.

: 	 I think tight, efficient coding, excellent gameplay, and good 
: audiovisuals can make a game a classic.  But so far, some makers are 
: concentrating on the last factor over the first two. And it's not 
: coincidental that increasing processing power and emergence of 
: 'multimedia' in games are linearly pathed.  And it's also not 
: surprising to note that the first two factors don't require a big 
: honking hoss of a processor.

The more efficient instruction set and scheduling, and the more
efficient structure of the Pentium as a whole, makes for more efficient
coding in general. That is, from the same high level source code,
a compiler can optimize better and generate more efficient code
for the Pentium. There is an even bigger gain if programmers
purposefully take advantage of these features in the original code.

[How can Pentium optimized code benefit 486]
: 	It can benefit a 486 in that tighter and more efficient code can 
: benefit just about any chips.  Branch prediction should allow Pentium to
: achieve closer to 2 instructions per tick, and just as easily allow faster
: performance from a 486.  

Sorry, but under many (most) circumstances code optimized for the 
Pentium would be detremental to the 486. How adversely depends on
the nature of the program itself. Coders see the biggest advantage
for optimizing for the Pentium if the program were CPU-bound.

For example, a coder decides to optimize for the Pentium either
manually by specifying more work to be done on the register level
or by using a compiler flag (say the -K Pentium on SVR4 CCS compilers).
This typically results in a *bigger* code size, but because the
Pentium capabilities are fully utilized, also results on a *faster*
execution time. (Analogous to CISC vs RISC design; RISC gains in
code size but typically faster in CPU-bound execution.)

Trying to run the same code on a 486, (if it runs at all), will
result in a performance *hit* because now the 486 has to cope
with a larger code and does not have the ability to optimize
on the register level the way Pentium does. This results in 
a *looser* code for the 486, not tighter.

Indeed, the fact that Pentium optimized code is detremental
to the 486 and below is a big reason (from a marketing point
of view) why it is not done often. Many vendors (such as
Microport for its 4.2.1 operating system offering) offer
both Pentium and non-Pentium versions.

The same was true re: optimizing for the 486 would hurt the
386, especially because the availability of the L1 memory
in the 486.

-Ade Barkah
--
Head of Development
Renaissance Knowledge Systems
Englewood, Colorado