🚀 go-pugleaf

bbrown@infonode.ingr.com (Bailey Brown) writes:

>stuartw@ccu1.aukuni.ac.nz (Stuart Woolford) writes:


>>perhaps this would have more to do with the key-repeat rate on the 486 ??
>>also: a 486 running 32bit OS/2 2.0 programs is about 4 time faster on
>>integer code ( let alone floating point ) than the same machine under dos.
>>(ie: think about what you are actually measuring !!!)

>I can't see how this is possible.  I can see how recompiling stuff for
>32-bit *might* make it up to twice as fast, but not 4 times as fast. Please
>justify this.  I, myself, have ported DFLAT (the text mode user
>interface lib from Dr. Dobb's Journal) and it's sample application,
>to 32-bit extended dos and have noticed no speed increase at all.
>And this code is full of far pointers.  I have seen some small, tight
>programs (gif decoder code) that does not benefit at all from
>32-bit.  The pointers are all near and the words used are all 16bit, and
>there is no reason to make them 32-bit.  From my experience, the 
>advantages of having 32 bits are more in the form of having a large
>flat address space, and much less in the form of more speed.

no.. but with a poor compiler the speedup will be 1-2x..
a 486 is not just a 386 with a numeric copro..It has a complex pipeline, and
gains significantly from instruction reordering on even 386 code...
( note: this requires the use of a compiler with 486 support, like Watcom C9 )
so you get:
	1) 1-2x from moving to 32bit for many app's.
	2) 0.5-1x from moving to native-mode
	3) 3-5x from using '486 code reorganisation.

so the total moving from 286 code on a 486 to 486 code is about 3-10x with a
	GOOD COMPILER!!!

note: 486 mode does not stop it running on a 386. it just reorders the code to
	stop pipeline stalls..also, pipeline may not be technically correct,
	but the effect is the same as pipeline stalls..

-------------------------------------------------------------------------------
stuartw@ccu1.aukuni.ac.nz


			>>>>In VI Where Available<<<<
-------------------------------------------------------------------------------