Original from  Eddy Thilleman  to Denis Tonn on 12-27-1998
Original Subject: more power

                         ---------------------------------------

 DT> Thunking can be either expensive (in CPU overhead) or cheap
 DT> depending on how it is done.
 
ET> So starting with OS/2 v2 the 'cheap' conversion (=little 
ET> CPU overhead) is choosen, so all 16-bit OS/2 programs run 
ET> at about the same speed as on 16-bit OS/2-versions.

 Yep. 
 BTW: It is a fallacy to assume that 32 bit programs are always 
"faster" than 16 bit programs. The only Intel CPU this is true for is 
the Pentium Pro. As long as the program code fits into a 64K segment 
and the data fits into 2 (or 4 on a 386) 64K data segments the "speed"
of the 16 bit program is the same as a 32 bit version. There is a 
significant overhead when a selector gets reloaded, so 16 bit code 
that is larger than 64K and/or needs more than 256K (128K on a 286) 
data is slower than a 32 bit program. The 32 bit program does not need
to reload a selector to access more than 256K. 

 DT> 32 bit allocations are all done on a 64K boundries and the
 
ET> In order to keep the pointer conversion to the minimum? 
ET> Otherwise the conversion takes a lot more work?

 Yes. The tiling of the LDT depends on this, and the "minimum" 
conversion work is based on the tiling. 


   Denis       

 All opinions are my very own, IBM has no claim upon them
. 
. 
.
 

 



--- Maximus/2 3.01