Steve Rogers mentioned this to Gary Weinfurther:
 SR>   But are we not splitting hairs? Haven't so many compromises already
 SR>   been made in the creation of Win 95 and Delphi that our efforts at
 SR>   such optimizations are lost in the sea of "Just throw more CPU power
 SR>   at it!" attitudes? ;)
Yes, perhaps we are splitting hairs.  The difference in speed between a 
Pentium processing a 32 bit value versus a 16 bit value is indeed miniscule.  
But knowledge of how the CPU works helps you to make better decisions based 
on the situation at hand.
A couple of days ago, I optimized a routine by converting it to inline 
assembler.  After spending several hours on it, I compared its performance to 
the original Object Pascal version.  With optimizations turned on, the 
assembler version was faster, but only by a small percentage.  The gain was 
so small that the time spent optimizing could have been better spent 
elsewhere.
Another time, I did the same thing to a custom bit vector class.  In that 
case, the assembler version took one fourth of the time the Object Pascal 
version did.
The lesson?  Readability and maintainability *in most cases* is probably more 
important than speed, but when you do optimize, do it in the places where it 
will make a difference.
                                ...Gary
--- GoldED 2.41
---------------